GAGA Factor-dependent Transcription and Establishment of DNase Hypersensitivity Are Independent and Unrelated Events in Vivo *

Using a Drosophila transgenic system we investigated the ability of GAGA factor, a putative anti-repressor, to modulate transcription-related events in the absence or presence of a bona fide activator, the Adf-1 transcription factor. In contrast to previous in vitro and in vivo data linking the binding of GAGA factor to the acquisition of DNase hypersensitivity at heat shock promoters, we observed that inserting multiple GAGA binding motifs adjacent to a minimal alcohol dehydrogenase (Adh) promoter led to strongly elevated embryonic transcription without creation of a promoter-associated DNase-hypersensitive (DH) site. Establishment of DNase hypersensitivity required the presence of both GAGA and Adf-1 binding sites and was accompanied by a further, synergistic increase in transcription. Because Adf-1 is capable neither of establishing a DH site nor of promoting efficient transcription by itself in embryos, it is likely that DH site formation depends on a GAGA factor-mediated binding of Adf-1 to chromatin, perhaps facilitated by a locally remodeled downstream promoter region. More generally we suggest that GAGA factor-binding sequences may operate in a promoter-specific context, with transcriptional activation, polymerase pausing, and/or DH site formation critically dependent on the nature of the sequences (and their binding partners) linked in cis.

Using a Drosophila transgenic system we investigated the ability of GAGA factor, a putative anti-repressor, to modulate transcription-related events in the absence or presence of a bona fide activator, the Adf-1 transcription factor. In contrast to previous in vitro and in vivo data linking the binding of GAGA factor to the acquisition of DNase hypersensitivity at heat shock promoters, we observed that inserting multiple GAGA binding motifs adjacent to a minimal alcohol dehydrogenase (Adh) promoter led to strongly elevated embryonic transcription without creation of a promoter-associated DNase-hypersensitive (DH) site. Establishment of DNase hypersensitivity required the presence of both GAGA and Adf-1 binding sites and was accompanied by a further, synergistic increase in transcription. Because Adf-1 is capable neither of establishing a DH site nor of promoting efficient transcription by itself in embryos, it is likely that DH site formation depends on a GAGA factor-mediated binding of Adf-1 to chromatin, perhaps facilitated by a locally remodeled downstream promoter region. More generally we suggest that GAGA factor-binding sequences may operate in a promoter-specific context, with transcriptional activation, polymerase pausing, and/or DH site formation critically dependent on the nature of the sequences (and their binding partners) linked in cis.
Control of eukaryotic transcription is a complex, tightly regulated process requiring the action of many distinct proteins, including chromatin-interacting/modifying proteins, transcriptional activators, and general transcription factors (GTFs) 1 (for review see Refs. [1][2][3]. In addition to recent progress in determining the role of chromatin-interacting proteins, unraveling the mechanisms by which sequence-specific, DNA-binding activators promote transcription has traditionally occupied a good deal of effort in the field. The consensus emerging from studies on many activators is that a significant activity of these proteins is to recruit components of the general transcription machinery and/or chromatin-interacting factors to promoters through direct protein-protein interactions (4).
The Drosophila GAGA factor is unusual in that its activity in transcription appears to be neither that of a purely sequencespecific transactivator nor that of an ATP-dependent chromatin-interacting factor. GAGA protein was first identified as a transcription factor that bound a repetitive GA element upstream of the engrailed (5) and Ultrabithorax (6) promoters. So-called GAGA elements have subsequently been identified in numerous promoters, including those controlling housekeeping, developmentally regulated, and inducible genes (7). GA repeats appear to be important in the acquisition of constitutive DNase I-hypersensitive (DH) sites found at the transcriptionally inactive but inducible hsp26 (8) and hsp70 (9) promoters in vivo. Moreover, in concert with the ATP-dependent nucleosome remodeling factor (NURF), GAGA factor aids in the local remodeling of GAGA element-containing chromatin templates in vitro (10 -12), including the establishment of DH sites at an hsp70 promoter (10). Other in vitro studies have revealed that although GAGA factor stimulates transcription of repressed promoters, it does not activate transcription above basal (i.e. nonrepressed) levels (12)(13)(14). Null mutations of the Trithorax-like (Trl) gene, which encodes GAGA factor, result in larval lethality (15), whereas less severe alleles show enhancement of position effect variegation as well as deficiencies in chromosome condensation, segregation, and nuclear cleavage (15,16). Taken together, the data suggest a role for GAGA factor in establishing or maintaining particular chromatin structures. A model has emerged where, when located in the vicinity of a promoter, GAGA factor facilitates promoter activation not by actively recruiting GTFs but rather by functioning as an anti-repressor molecule. GAGA factor binding is proposed to promote local displacement or disruption of nucleosomes in creating a DNase I-hypersensitive site, thereby allowing bona fide activators and/or the general transcription machinery to gain access to DNA at promoter regions (Refs. 10 and 17; for review see Ref. 18).
Because much of the aforementioned work on GAGA factor has employed complex, multi-factor binding promoters or in vitro conditions, we sought to clarify the activity of GAGA factor in vivo in the context of a defined promoter at the level of both transcription and chromatin perturbation. In addition, it was of interest to ask the same question of a bona fide activator for direct comparison. For this latter purpose we chose Adf-1, first identified as a regulator of the alcohol dehydrogenase (Adh) distal promoter (19) and subsequently shown to bind the antennapedia P1 and dopa decarboxylase promoters (20) and the ftz-proximal enhancer (21). In our experiments, therefore, we have analyzed Drosophila carrying transgenes consisting of GAGA factor or Adf-1-binding elements, either alone or in combination, upstream of a minimal Adh distal promoter. Un-expectedly, we find that GAGA factor alone is able to stimulate embryonic transcription markedly in vivo, whereas Adf-1 is a relatively poor activator when operating unaided in these cells. As might be anticipated, the highest levels of transcription are achieved through a synergistic contribution of these two independent activities. A second, unforeseen result is that a promoter-proximal DH site is established only in the presence of both activities rather than via the action of GAGA factor alone. Therefore, in the context of the Adh distal promoter, GAGA factor and Adf-1 synergize both at the level of activating transcription and of modulating chromatin structure.

EXPERIMENTAL PROCEDURES
Construction of the P Element Constructs and Transgenic Lines-All recombinant DNA procedures were performed using standard techniques (22); full details can be provided upon request. In outline, a minimal derivative of the Drosophila distal Adh promoter (Ϫ41 to ϩ20) driving a luciferase reporter gene was isolated from p-41ADHLUC (kind gift from A. Brasier) and subcloned into the EcoRV site of pBlue-scriptSKII(Ϫ) (Stratagene) to generate pSK:L. The su(Hw)-binding element of the gyspy retrotransposon was isolated by SalI digestion of pGBaBx (23) and inserted at the XhoI site of pSK:L to generate pSK:LS. A bacterial lacZ gene flanked by promoter and terminator sequences for T7 RNA polymerase was cloned into the BamHI site of pSK:LS, in a divergent orientation relative to the Adh-luciferase transcription unit, to generate pSK:ZLS. For the GAGA construct, single-stranded synthetic oligonucleotides comprising the proximal GAGA element of Drosophila hsp26 (Ϫ125 to Ϫ80) (along with appropriate flanking restriction sites) were annealed and inserted between the PstI and SmaI sites of pSK:ZLS, i.e. in the region between the T7 RNA polymerase and Adh promoters, to generate pSK:ZGLS. For the Adf-1 and GAGAϩAdf-1 constructs, we performed polymerase chain reaction using p-89ADHLUC (kind gift from A. Brasier) to generate an approximately 180-bp product spanning regions Ϫ89 to ϩ20 of the Adh distal promoter fused to the luciferase gene and used this product to replace the Ϫ41Adh promoter derivative in both pSK:ZLS and pSK:ZGLS to generate pSK:ZALS and pSK:ZGALS. The SpeI-KpnI fragment from each of pSK:ZLS, pSK:ZGLS, pSK:ZALS, and pSK:ZGALS, which included the lacZ, luciferase, and su(Hw) regions, was subcloned into pC4:S linearized with KpnI and SpeI. pC4:S is a derivative of the pCaSpeR4 P element vector (24) containing a su(Hw)-binding element at its XhoI site; thus in the final constructs, both transcription units were now flanked by insulator sequences. Multiple homozygous transgenic lines carrying single site insertions of the transgenes were generated by the standard P element-mediated transformation protocol (25).
Chromatin Analyses-Nuclei were isolated from 2.5 g of frozen embryos as described previously (26). A minimum of two independently transformed lines was tested in each assay. For DNase I treatment, nuclei were resuspended in 1 ml of DNase I digestion buffer (60 mM KCl, 15 mM NaCl, 15 mM Tris-HCl, pH 7.4, 3 mM MgCl 2 , 0.5 mM dithiothreitol, 0.25 M sucrose) and dispensed as 250-l aliquots. Increasing amounts of DNase I (Worthington) were added, samples were incubated on ice for 3 min, and reactions were stopped with 5 l of 0.4 M EDTA, pH 8.0. 8 -10 g of purified genomic DNA was digested with either SstI and EcoO109I or BamHI (as indicated in Fig. 3), subjected to gel electrophoresis in a 1.2% agarose gel, transferred to Gene Screen Plus membrane (Amersham Pharmacia Biotech), and probed with the appropriate 32 P-labeled fragment (see figure legend). For the micrococcal nuclease (MNase) digestions, nuclei were resuspended in 1 ml of MNase digestion buffer (60 mM KCl, 15 mM NaCl, 15 mM Tris-HCl, pH 7.4, 2 mM CaCl 2 , 0.5 mM dithiothreitol, 0.25 M sucrose) and dispensed as 250-l aliquots. Increasing amounts of MNase (Worthington) were added, tubes were incubated at 25°C for 5 min, and reactions were stopped with 5 l of 0.4 M EDTA, pH 8.0. 10 g of isolated genomic DNA was subjected to Southern blot analysis as above. For the restriction enzyme analyses, nuclei were resuspended in 750 l of the appropriate enzyme digestion buffer containing the following protease inhibitors, phenylmethylsulfonyl fluoride (0.5 mM), leupeptin (5 mg/ml), and aprotinin (0.1 unit/ml) and dispensed as 250-l aliquots. Various concentrations of either PstI or BamHI were added, as indicated in Fig. 5, tubes were incubated at 37°C for 45 min, and reactions were stopped with 5 l of 0.4 M EDTA, pH 8.0. 8 -10 g of isolated genomic DNA was digested with SstI and EcoO109I and subjected to Southern blot analysis as above.
Transcription Analyses-To prepare embryonic extracts, approximately 25-50 mg of 0 -12-h embryos were dechorionated by incubation in 50% bleach for 3 min, washed thoroughly, placed into a microcentrifuge tube containing 100 l of lysis buffer (1% Triton X-100, 25 mM glycylglycine, 15 mM MgSO 4 , 4 mM EGTA, 1 mM dithiothreitol, pH 7.8), and homogenized using a Kontes pestle. For adult extracts, 10 anesthetized flies were homogenized in 100 l of lysis buffer. Both extracts were cleared by 5 min of centrifugation at 4°C. 10 -20 l of extract was analyzed for luciferase activity using the luciferase reporter assay system (Promega). Luciferase activity for each sample was normalized for protein concentration and adjusted for background luminescence by subtracting the signal obtained from samples derived from the nontransgenic parental line.

Strategy for Investigating the Role of GAGA Factor in Gene
Activation in Vivo-To systematically investigate the in vivo activity of GAGA repeats and presumably GAGA factor binding, our approach was to add an extensive GAGA element upstream of either a minimal promoter or a promoter containing the binding site for an activator, linked to a reporter gene. We assayed homozygous transgenic lines generated by P element-mediated transformation (25) carrying one of four possible constructs for their effects on luciferase reporter gene activity and perturbations in chromatin structure.
GAGA Factor and Adf-1 Synergize to Activate Transcription-Both GAGA factor and Adf-1 are known to be players in the transcriptional activation of specific RNA polymerase IIregulated genes in Drosophila. Adf-1 is believed to recruit specific GTFs to the promoter (27). GAGA repeats have been shown to be important for transcription in vivo of hsp26 (28), hsp70 (29), and Ultrabithorax (30). For our study we utilized the hsp26-proximal GAGA element. This element is one of the longest GAGA repeats characterized, with 35 nucleotides protected in footprinting assays, most likely reflecting the binding of multiple GAGA proteins along its length (31,32). We created the following transgenic lines (Fig. 1): control, consisting of a TATA box-containing promoter derived from the distal tran-FIG. 1. Schematic of the transgenes used to study GAGA factor activity. A, sequences common to each of the transgenes are shown, with the promoter region expanded for clarity. In the control construct a luciferase reporter gene is driven by a minimal RNA polymerase II promoter (shaded) derived from the Drosophila Adh distal promoter (Ϫ41 to ϩ20); the position of the TATA element within the promoter is also depicted. Approximately 20 bases upstream of the Adh promoter the bacteriophage T7RP promoter (Ϫ23 to ϩ26) was cloned in divergent orientation. This was included in the constructs to allow for additional measurements of DNA accessibility in chromatin. All constructs are flanked by su(Hw)-binding elements derived from the gypsy retrotransposon. B, various insertions made between the two promoters. The Adf-1 construct contains the binding element for Adf-1 in its natural context upstream of the Adh promoter (Ϫ89 to Ϫ42). The GAGA construct contains the proximal GAGA element (Ϫ125 to Ϫ80) of the Drosophila hsp26 regulatory region. The GAGAϩAdf-1 construct consists of both elements upstream of the promoter. OE, pertinent restriction sites; B, BamHI; P, PstI; A, AvaII. The angled arrows above each construct indicate the approximate positions of transcription start sites, whereas open arrows below each construct indicate the additional GAGA sites present in the Adh promoter (see "Discussion"). scription unit of Adh; Adf-1, containing the binding site for this transactivator in its natural position at the Adh promoter; GAGA, containing the hsp26-proximal GAGA element 5Ј of the promoter; and GAGAϩAdf-1, containing both elements. Inspection of the Adh promoter used shows there are additional short GAGA elements present, located at Ϫ41 to Ϫ37 and at ϩ8 to ϩ11 (33). Reiterated su(Hw)-binding sites derived from the gypsy retrotransposon flank the transgene. These "insulator" elements were anticipated to eliminate potential position effects in the multiple lines analyzed (34).
In whole cell extracts of 0 -12-h embryos from the various transgenic lines, luciferase transcription is increased slightly (3-fold) in the Adf-1 and significantly (15-fold) in the GAGA lines, relative to those carrying the control transgene ( Fig. 2 and Table I). Combining these two elements in a single construct (GAGAϩAdf-1) has a synergistic effect, with an average 29-fold activation. The relative patterns of transcription activation in adults are altered somewhat from those seen in embryos, with an even greater degree of synergy observed in the GAGAϩAdf-1 lines (Table I). It is important to note, however, that the absolute levels of transcription are substantially lower in adults, possibly because of low concentrations of GAGA factor resulting from a significantly decreased amount of its mRNA in adults compared with embryos (35). In addition, the several GAGA mRNA species present in adults produce a variety of protein isoforms (35,36) that may have distinct functions. Furthermore, there could be differential adult stage-or tissue-specific variations in Adf-1 and GAGA factor expression patterns relative to their ubiquitous expression in embryos (35,37). Overall, these results indicate that GAGA factor acting alone at a minimal promoter can lead to significantly elevated levels of transcription in vivo and, additionally, that GAGA factor and Adf-1 synergize in transcription activation.
DNase I-hypersensitive Site Generation Requires the Cooperative Action of Both GAGA Factor and Adf-1-Traditionally, DH sites have been thought to demarcate regions of chromatin that are either nucleosome-free or markedly disrupted (38). In vivo and in vitro analyses from several laboratories have revealed that GAGA sequences are one important component in the formation of constitutive DH sites at both the hsp26 and hsp70 promoters. However, multiple sequence elements in these promoters can quantitatively contribute to DH site generation and the loading of the transcriptional apparatus (8,17,18). We tested for the presence of DH sites in 0 -16-h embryos of several transgenic lines carrying each of the constructs outlined in Fig. 1. Unexpectedly, introduction of the extensive hsp26-proximal GAGA element does not lead to the creation of a DH site in the vicinity of the minimal Adh promoter (Fig. 3A,  lanes 7-9). Similarly, no DH site was observed in the Adf-1 lines (Fig. 3A, lanes 4 -6). Only those lines carrying the GAGAϩAdf-1 transgene exhibit a DH site at the promoter (Fig.  3A, lanes 10 -12; note its very close correspondence with the location of the AvaII site at Ϫ27 bp). As a control for adequate DNase I treatment, prominent DH sites can be easily detected upstream of the endogenous hsp26 promoter (39) in all transgenic lines tested using the identical DNase I-treated samples (Fig. 3B, DH sites denoted by asterisks). This result indicates that, at the transgene, GAGA factor works in concert with Adf-1 to generate DNase I hypersensitivity. Neither GAGA factor nor Adf-1 independently has the ability to induce a DH site in the context of the minimal Adh promoter tested here.
In Vivo Effects of GAGA Factor on Nucleosome Integrity-Because the GAGA transgenes exhibited a strong stimulation of transcription but no detectable DH site formation, we sought to determine what kind of more subtle changes, if any, in chromatin structure might be associated with the presence of GAGA factor adjacent to the Adh promoter. Brief treatment of chromatin with MNase followed by gel analysis of purified DNA reveals a pattern of DNA fragments increasing in size by approximately 200-base pair increments derived from digestion within linker DNA at mono-, di-, tri-nucleosomal, etc., intervals. MNase analysis of the hsp70 (10), hsp26 (11), and ftz (12) promoters reconstituted into chromatin in vitro showed that addition of GAGA factor, in the presence of both embryonic nuclear extract and ATP, led to a localized disruption of the nucleosomal array at the GAGA sequences. Because both cofactors were essential, it was suggested that the observed remodeling depended critically on the presence of NURF in the extract. To assess the ability of GAGA factor to locally remodel nucleosomal arrays in vivo on a simple model promoter, we conducted MNase digestion analysis on embryonic nuclei derived from each of the transgenic lines.
Short probes located immediately adjacent to (and on either side of) the insertion site can reveal the local effects of individual inserts across the various lines. Using a sequence probe derived from directly downstream of the insertion (encompassing both the Adh promoter and the immediately 5Ј portion of the luciferase gene), those lines that contain the GAGA element reveal an altered, much less distinct, local nucleosomal pattern relative to the control and Adf-1 lines (Fig. 4A, compare  lanes 5-8 with lanes 1-4; densitometric tracing of lanes 2, 4, 6, and 8 is shown in Fig. 4D). Probing immediately upstream of the insertion site reveals rather little in the way of obvious perturbation in comparing any of the lines (Fig. 4B; see Fig. 4E for a densitometric trace). Moreover, when the same blot is probed with a DNA sequence derived from the endogenous   Fig. 2) was averaged. Fold activation was then determined by dividing the transgene-specific average activity by the average activity of the control transgene. RLU, relative light units. hsp26 gene (Fig. 4, C and F), a regular and distinct nucleosomal ladder is present in all four of the transgenic lines, showing that the perturbations observed in the GAGA and GAGAϩAdf-1 lines in Fig. 4A were genuine and not an experimental artifact. Because the same blot was used to produce the clearly differing data of Fig. 4 (compare A with B), it appears that the perturbations seen are extremely localized and directed downstream of the insertion site over the promoter and/or transcribed region. Lastly, because these downstream perturbations appear to be unequivocally associated with the presence of the GAGA element (none seen in the control or Adf-1 lines), they exist independently of the presence of both the Adf-1-binding site in DNA and the DH site in chromatin. In a further attempt to measure the extent and location of GAGA factor-associated chromatin structure perturbation, we investigated restriction enzyme accessibility in and around the  (lanes 3, 4, 7, and 10), 30 units (lanes 2, 5, 8, and 11), or 60 units (lanes 1, 6, 9, and 12). Lanes 13 and 14 contain restriction-digested control plasmid DNA as location markers (M); BamHI and AvaII sites adjacent to the promoter are indicated by arrows at the side. B, DNA samples identical to those in A were digested to completion with BamHI and analyzed by Southern blot with a probe consisting of an ϳ2-kilobase BamHI-EcoRI fragment derived from the endogenous hsp26 gene. Amount of DNase per sample: 0 units (lanes 1, 5, 9, and 13), 10 units (lanes 2, 6, 10, and 14), 30 units (lanes 3, 7, 11, and 15), and 60 units (lanes 4, 8, 12, and 16). 4. Effect of the hsp26-proximal GAGA element on the local nucleosomal array. Following MNase treatment of embryonic nuclei derived from various transgenic lines (as in Fig. 3), genomic DNA was isolated and analyzed by Southern blot. A single filter was probed with three different DNA fragments: a 220-bp fragment encompassing the Adh promoter and 5Ј region of the luciferase gene (A), a 140-bp fragment from the immediately upstream region (B), and a fragment derived from the coding region of the endogenous hsp26 gene (as in Fig.  3B) (C). Units of MNase: 7.5 units (lanes 1, 3, 5, and 7) and 15 units (lanes 2, 4, 6, and 8). D-F, a densitometric trace was made of the nucleosomal array for each of the various transgenic lines shown in A-C, respectively. Lanes 2, 4, 6, and 8 from each autoradiogram were analyzed using NIH Image software. A schematic of the control transgene with probe size and location indicated by the heavy black bars is shown below; the location of the various insertions is indicated by the open triangle.
promoter region of the transgenes. A PstI site lies directly 3Ј of the GAGA element, whereas a BamHI site is approximately 60 base pairs upstream of its 5Ј end (Fig. 1). In embryonic nuclei, PstI accessibility within the transgene increases moderately in the GAGA lines and is highest in the GAGAϩAdf-1 lines (Fig.  5A). In contrast, the level of BamHI (Fig. 5B) accessibility is, if anything, moderately reduced in the GAGA element-containing lines. These results signify a fairly localized and modest downstream perturbation effect and are in general agreement with the MNase data. In addition, we monitored the ability of T7 RNA polymerase (T7RP) to transcribe from its promoter located around 30 bp upstream of the insertion site (Fig. 1). T7RP is a single polypeptide enzyme that has been used previously as a probe for monitoring DNA accessibility in chromatin (for examples, see Refs. 40 and 41). The addition of T7RP to nuclei isolated from a GAGAϩAdf-1 line produced no increase in RNA transcripts relative to the amount generated in nuclei from the control line (data not shown), confirming that the chromatin perturbation in the GAGAϩAdf-1 line is localized to the Adh promoter and/or the luciferase transcribed region and does not extend into a closely adjacent upstream region.

Role of GAGA Factor in Trans-activation versus Chromatin
Perturbation-GAGA factor has been classified as a chromatin anti-repressing factor rather than a conventional activator (for review see Ref. 18). Carefully controlled in vitro experiments showed that GAGA factor had no overt transcription activating ability on naked DNA templates (13,14) but could relieve the repression of basal transcription at a linked promoter assembled into chromatin when the remodeling factor NURF was present (12). The observed stimulation of transcription by GAGA factor in crude extracts or transient assays (35,36,42) might then reflect its ability to partake in the suppression or destabilization of the many nonspecific DNA-protein interactions that are likely to occur on DNA templates in such assays.
With GAGA factor characterized purely as an anti-repressor, therefore, we were intrigued to find that the placement of high affinity GAGA-binding sites adjacent to a minimal Adh TATA promoter was sufficient for robust stimulation of transcription in an in vivo chromatin context ( Fig. 2 and Table I). Two alternative, although not mutually exclusive, explanations could account for this activity. First, localized chromatin destabilization or nucleosome sliding in the vicinity of the TATA promoter, modulated in part by GAGA factor binding to its adjacent recognition site (presumably aided by the remodeling activity of the NURF complex Refs. 10 and 43), might allow access of the general transcription machinery to the promoter via binding of TATA-binding protein and the GTFs. In other words the stimulation of in vivo transcription purely by GAGA factor, as seen here, would be analogous to its in vitro ability to establish basal transcription. Considered in this light, the presence of promoter-proximal GAGA elements might be a way of achieving a measure of constitutive transcription in vivo against the normally repressive chromatin background. The data presented in Figs. 4A and 5A show a perturbation of the promoter region and/or downstream nucleosomal array in the presence of the GAGA element, but because the transcription machinery is clearly also present and functional in these cases, it is not possible to unambiguously ascribe the cause of this downstream chromatin disruption to GAGA factor binding alone (though neither can it be ruled out on this evidence). The reduction in accessibility of the upstream BamHI site when GAGA elements are present (Fig. 5B) would be consistent with a GAGA factor-induced sliding of a nucleosome into this location, an effect of GAGA factor binding to an in vitro reconstituted chromatin template that was noted by others (11). However, when we attempted to map nucleosomal positions in the promoter region by a relatively low resolution indirect end labeling analysis of MNase digested samples, we found no clear difference in the cleavage patterns of the control lines compared with the GAGA lines (data not shown), suggesting the effects of GAGA factor on transcription were not related to a major repositioning of the nucleosomal array at the promoter. Only in the GAGAϩAdf-1 line did we observe an obvious change in the promoter-proximal MNase cleavage pattern, consistent with the unique establishment of a DH site in these lines (shown in Fig. 3). A second interpretation of the GAGA activity seen here is that GAGA factor acts as a conventional activator that (directly or indirectly) leads to active recruitment of the general transcription machinery. GAGA factor does have a glutamine-rich region, a domain found in one class of activator proteins (44), and whereas this domain can clearly lead to self-aggregation of GAGA factor (45,46), the only transcription factor so far known to interact with it is heat shock factor (47). More interestingly perhaps, this GAGA domain has recently been found to promote distortion of downstream DNA (45), a perturbation that could facilitate the binding of TATA- FIG. 5. Accessibility of chromatin to restriction enzymes in the transgenic lines. Isolated embryonic nuclei were incubated with increasing amounts of the indicated restriction enzyme. Purified DNA was subjected to indirect end labeling analysis as described in Fig. 3A; the location of the relevant restriction sites is depicted in Fig. 1. A, PstI digestion: 0 units (lanes 1, 4, and 7), 150 units (lanes 2, 5, and 8), and 350 units (lanes 3, 6, and 9). B, BamHI digestion: 0 units (lanes 1, 4, and  7), 150 units (lanes 2, 5, and 8), and 350 units (lanes 3, 6, and 9). The percentage site accessibility was calculated as the density ratio of the digested fragment relative to the sum of the parent band (P) plus the digested fragment using a Molecular Dynamics PhosphorImager and ImageQuant software. Each individual line was analyzed a minimum of three times, and the ratio given (indicated below the bar under the appropriate samples) represents the average site accessibility across a minimum of two independent lines for each transgenic construct. (The control samples in lanes 1-3 of A are relatively overloaded, but the density ratio shown is directly comparable with those of the other samples in A.) binding protein and/or GTFs in accord with the general mechanism envisioned by the first model.
It is particularly important to consider our in vivo transcription data in the light of extensive transcription studies performed on GAGA element-containing Drosophila heat shock promoters. A distinctive property of these promoters is that, in the absence of an activated transcription factor (i.e. the heat shock factor protein), RNA polymerase II is present on the promoter in a transcriptionally initiated, but "paused" condition, a feature that is critically influenced by the presence of upstream GAGA elements (9,17,29). To a limited extent then, our data are in agreement, inasmuch as we observe GAGA-dependent loading of polymerase (measured by strongly increased transcription in the GAGA lines) in the absence of an overt transactivator known to function via recruitment. However, a crucial difference is that although the minimal Adh promoter used here clearly allows passage of RNA polymerase II into productive elongation under the influence of adjacent GAGA elements, the initiated polymerase at heat shock promoters falls into arrest immediately downstream and can only be released into elongation upon binding of the activated heat shock factor protein to sites upstream from the promoter. Recent in vitro reconstitution of a paused polymerase on the Drosophila hsp70 gene strongly suggests that an important element of pausing is inherent to the sequence contained within the heat shock promoter itself (48,49). The absence of such sequence elements (or, conversely, presence of different elements) in the minimal Adh promoter, provides a plausible (and testable) explanation for why productive elongation is able to take place in the presence of GAGA factor alone in the experiments reported here. We are currently conducting promoter swap experiments to assess this potential promoterspecific contribution to the in vivo activity of GAGA factor.
Multiple GAGA Elements Are Insufficient for DH Site Establishment-In considering previous studies on the structural consequences of GAGA factor binding conducted on in vitro reconstituted chromatin templates (particularly at heat shock promoters), the absence of a DH site in vivo at the promoter of the transcriptionally active transgenic GAGA lines (Fig. 3,  lanes 8 and 9) is a further unexpected result. Its absence is unlikely to be related to the demonstrated requirement for three or more independent GAGA elements to generate a DH site in vitro (10). In our constructs, there are two short GAGA sites in the minimal Adh promoter itself that are known to be binding targets both in vitro (33) and in vivo (50), as well as the 42-bp GAGA element derived from hsp26, a sequence that appears capable of binding multiple copies of the GAGA factor (33). Although GAGA elements contribute to the full manifestation of DNase I hypersensitivity at heat shock promoters, it appears that other promoter-proximal sequence elements and nonhistone trans-acting factors play some role in its establishment (see Refs. 8 -11, 43, and 51). Again we suggest that one or more such sequence elements are absent from the minimal Adh promoter studied. Furthermore, it is important to reiterate that our data reflect the results of a true in vivo situation, where the levels of GAGA factor are natural and not achieved by in vitro manipulation or use of a purified or recombinant protein. Our results show that gene "activation," mediated here by GAGA factor, is not necessarily accompanied by DH site formation at the promoter either as a necessary prerequisite for or consequence of transcription. In fact, we feel strongly that the differential manifestations of GAGA activity seen here, in comparison with studies on heat shock promoters in particular, are rather convincing indicators of a promoter-specific context to the function of GAGA factor, with the nature of cis-linked sequences playing a critical role in the process.
The promoter-proximal DH site seen in the GAGAϩAdf-1 lines (Fig. 3) may reflect a DNA structural perturbation created by the efficient binding of Adf-1 in the presence of GAGA factor. Gao and Benyajati (33) have noted that Adf-1 binding to DNA in vitro creates a strong structural perturbation 3Ј of the Adf-1-binding site, but neither the binding of Adf-1 nor the DNA distortion are influenced by adjacent GAGA factor binding; apparently there is no direct interaction between the proteins. We propose that in the context of the multiple GAGA elementcontaining chromatin template studied here, both transcriptional synergy and DH site establishment are readily explained by cooperative binding of Adf-1 and GAGA factor to chromatin. This occurs independently of protein-protein contacts between the two species (33) but may be facilitated by altered histone-DNA interactions or downstream DNA distortions mediated either by the GAGA factor itself or in conjunction with NURF (Fig. 4). A similar cooperative mechanism dependent on structural perturbation has been advanced to account for high levels of activation in model yeast extrachromosomal plasmid templates (52).
GAGA Factor Activities in Context-These observations highlight the point that gene activation may be achieved by a variety of complementary, often synergistic, mechanisms, and it will usually be the case that a hierarchy of different activities are assembled at individual promoters to allow fine tuning of the level of gene expression. It is certainly intriguing to note that GAGA elements are found linked to a number of different types of promoter in Drosophila, some inducible, some developmentally regulated, and some constitutively active (7,18). Moreover, GAGA factor plays other roles in chromosomal metabolism that are not obviously related to transcription (53,54). Our results, together with those of others, show that the transcriptional effects of linking GAGA elements to a promoter and the corresponding influence on the local structure of chromatin may depend rather critically on the promoter context from which the data are obtained. The biochemical activities and potential interacting partners of the various GAGA factor isoforms are at present either ill-defined or undefined. Others have described transcription factors, e.g. GAL4-VP16 or NF-B, that are most likely able to combine both an anti-repressing and a transactivation function in a single entity (55,56). It may be that GAGA factor, given a particular promoter or other cis-linked sequence context, can display one or more such capabilities depending on the combinatorial opportunities presented by that promoter sequence.