Analysis of the DNA-binding site for Xenopus glucocorticoid receptor accessory factor. Critical nucleotides for binding specificity in vitro and for amplification of steroid-induced fibrinogen gene transcription.

In addition to the glucocorticoid receptor, DNA-binding proteins called accessory factors play a role in hormone activation of many glucocorticoid-responsive genes. Hormonal regulation of the gamma-fibrinogen subunit gene from the frog Xenopus laevis requires a novel DNA sequence that binds a liver nuclear protein called Xenopus glucocorticoid receptor accessory factor (XGRAF). Here we demonstrate that the recognition site for XGRAF encompasses GAGTTAA at positions -175 to -169 relative to the start site of transcription. This sequence is not closely related to the binding sites for known transcription factors. The two guanosines make close contact with XGRAF, as shown by the methylation interference assay. Single-point mutagenesis of every nucleotide in the 9-base pair region from positions -177 to -169 showed an excellent correlation between ability to bind XGRAF in vitro and ability to amplify hormone-induced transcription from DNA transfected into Xenopus primary hepatocytes. Conversely, XGRAF had little or no effect on basal transcription of the gamma-fibrinogen gene. Maximal hormonal induction also requires three half-glucocorticoid response elements (half-GREs) homologous to the downstream half of the consensus GRE. Interestingly, the XGRAF-binding site is immediately adjacent to the most important half-GRE. This close proximity suggests a new mechanism for activation of a gene lacking a conventional full GRE.

Steroid hormones, which include glucocorticoids and mineralocorticoids from the adrenal cortex and estrogens, progestins, and androgens from the gonads, regulate a vast array of physiological processes that are essential for development, differentiation, growth, metabolism, homeostasis, behavior, and reproduction in vertebrate organisms. In the classical model of steroid hormone action (1), the steroid ligands bind to specific intracellular protein receptors in target cells. The hormonereceptor complexes interact with particular short nucleotide sequences in the chromosomal DNA and modulate transcription of nearby genes. This model, however, cannot account fully for the complex tissue-specific and gene-specific actions of hormones. Transcriptional induction by steroids is influenced by many factors, such as local chromatin structure (2), stages of the cell cycle (3), cellular morphology or differentiation state (4,5), specific hormone ligand (6), and physiological state (7). Differential hormone responsiveness depends in part on the availability of other transcriptional regulatory proteins including coactivators and corepressors, which do not themselves bind to DNA (8), and accessory factors, which are DNA-binding proteins (9). For glucocorticoid-regulated genes, several accessory factors have been identified (2, 10 -17), but the mechanisms by which these proteins potentiate hormonal activation of transcription are not known.
To understand the role of accessory DNA-binding proteins in determining responsiveness to a steroid hormone signal, we are investigating glucocorticoid induction of fibrinogen gene expression in the liver. Fibrinogen is the precursor of fibrin, the major structural protein of a blood clot, and its synthesis is regulated by adrenal steroids both in basal homeostasis and following physiological stresses such as infections, inflammation, surgery, burns, etc. (18,19). Using primary liver cells from the frog Xenopus laevis as a model system (20), we have demonstrated that glucocorticoids stimulate transcription of the three separate genes coding for the fibrinogen subunits, termed A␣, B␤, and ␥ (21).
Identification of the specific DNA sequences that mediate steroid regulation of the Xenopus fibrinogen genes was accomplished by linking the 5Ј-flanking DNA of these genes to the firefly luciferase reporter gene and transfecting the DNA into primary Xenopus hepatocytes. Hormonal activation of the B␤ gene occurs through a single element (22) with a close match to the consensus glucocorticoid response element (GRE), 1 GGTA-CAnnnTGTTCT (23). Full stimulation of the ␥ gene, on the other hand, requires three closely spaced weak binding sites for the glucocorticoid receptor (GR) between nucleotides Ϫ168 and Ϫ135 relative to the start site of transcription (24). These sites have homology only to the downstream portion of the consensus GRE and therefore are referred to as half-GREs. In addition, steroid responsiveness is affected by bases within the sequence AAGAGTTAA at positions Ϫ177 to Ϫ169, immediately adjacent to the 5Ј-most half-GRE (24). This tract is unrelated to the conventional GRE and does not match other known transcription factor-binding sites. A protein present in Xenopus liver nuclei binds to this DNA sequence in vitro. Experiments with DNA containing blocks of mutations within and around bases Ϫ177 to Ϫ169 showed that the DNA-protein interaction correlates with hormonal activation of transcription. Thus, we named the nuclear protein Xenopus glucocorticoid receptor accessory factor (XGRAF). Compared with other known GR accessory factors, the location of the XGRAF-binding site is unusual since it replaces the sequence that would normally constitute the upstream half of a GRE.
In this work, we describe the detailed mutational analysis of the 9-bp region to which XGRAF binds. This investigation located the 5Ј-and 3Ј-boundaries of the recognition sequence and revealed which nucleotides are most critical for DNA binding in vitro and for glucocorticoid-stimulated transcription of transfected DNA in Xenopus primary hepatocytes.

EXPERIMENTAL PROCEDURES
Introduction of Point Mutations into Reporter Gene Constructs-All possible single-point mutations in the potential XGRAF-binding site in the ␥ gene upstream region were obtained by the following general strategy. 1) The ␥ gene sequence from positions Ϫ187 to ϩ41 was synthesized by the polymerase chain reaction (PCR) using a downstream primer with wild-type sequence and an upstream primer consisting of a mixed population of oligonucleotides with mutations in the XGRAF-binding region from positions Ϫ177 to Ϫ169. 2) The PCR products were inserted into a luciferase reporter vector, and the DNA was cloned by transformation into bacteria. 3) The nucleotide sequence of the ␥ DNA in individual clones was determined to ascertain which nucleotide(s) of the XGRAF region had been mutated.
The upstream primer, 5Ј-GGGGTACCAGACAGAAAAGAGTTAAT-GTTCCCTCTTATGTTC-3Ј, was synthesized (Genemed Biotechnologies, Inc.) in a single reaction, with the reservoir for each nucleotide in the potential binding site (underlined) deliberately contaminated to a concentration of 2.5% with each of the three nondesignated bases. PCR products from this primer yielded 15 of the 27 possible single-point mutants. A second population of mixed primers was synthesized (Genosys Biotechnologies), with contamination at the desired positions to optimal levels based on the formula described by Derbyshire et al. (25), yielding several additional mutants. The few remaining mutants were synthesized in PCRs with specific individual primers.
The PCR amplification was carried out with the pLL␥Ϫ187 construct (24) as the template DNA (which contains ␥ gene DNA from positions Ϫ187 to ϩ41), the upstream primers described above, a downstream primer within the vector, and Pfu polymerase (Stratagene) following the protocol from the manufacturer. The PCR products were digested upstream with KpnI and adjacent to position ϩ41 downstream with HindIII, purified through 2% low-melting-temperature agarose gels (26), and cloned into KpnI-and HindIII-digested pLuc-Link 2.0 (27). Transformation into Escherichia coli DH5␣ was as described (24). The nucleotide sequences of the ␥ DNA and of the junctions with vector DNA were confirmed for each mutant. Plasmid DNA was purified over anionexchange resin and, for transfection, over a cesium chloride gradient (24).
Gel Shift Assays-Both the probe and the competitors for the gel shift assays contained the ␥ sequence extending from positions Ϫ187 to Ϫ115 and were generated by PCR using Pfu polymerase. In addition, the molecules included 19 bases of vector sequence upstream of position Ϫ187 and a MfeI restriction enzyme site downstream of position Ϫ115. Wild-type pLL␥Ϫ187 was used as the template for making the probe, and the point mutation constructs described above were the templates for the competitors. The PCR products were purified through 2% lowmelting-temperature agarose gels (26), and the probe was 5Ј-end-labeled (28).
Nuclear extracts were made from X. laevis primary hepatocytes after 4 days in culture exactly as described (24), except that the concentration of HEPES-KOH in buffer C ϩ was 20 mM. The gel shift assays were carried out as described (24) in a final volume of 15 l with 0.5 ng of radioactive probe (7000 -51,000 cpm) and either no specific competitor or a 0.5-500-fold molar excess of specific competitor. Native 5% polyacrylamide gels (24) were run at 350 V for 6 -7 h at 4°C, dried at 80°C for 2 h, exposed to XAR film (Eastman Kodak Co.) at Ϫ80°C with one Lightning Plus intensifying screen (Dupont), and exposed to a Phos-phorImager screen (Molecular Dynamics, Inc.). The data from the phosphorimaging scan were analyzed with ImageQuant 3.3 software (Molecular Dynamics, Inc.).
Data Analysis for Gel Shift Assays-The ability of XGRAF to bind ␥ DNA with mutations in the putative XGRAF-binding region was determined by competition gel shift assays, which contained a constant amount of radioactively labeled wild-type ␥ DNA and either no competitor or various concentrations of DNA with a single-point mutation. An example of the gel shift assay with mutant A Ϫ171 as the competitor is shown in Fig. 1A. Radioactivity in the shifted DNA⅐XGRAF complex in each lane was quantitated from the phosphorimaging scan. The total amount of XGRAF in the complex with radioactive DNA in the absence of any competitor (Fig. 1A, lane 2) was defined as 1.0, and the amount of XGRAF in the complex with radioactive DNA in the presence of a 100-fold molar excess of wild-type competitor was defined as zero ( . Bound XGRAF is expressed as a fraction of total XGRAF, and competitor DNA is expressed as the -fold excess over the radioactive probe.
1A, lane 1). The quantity of XGRAF remaining in the complex with radioactive DNA in the presence of mutated competitor (Fig. 1A, lanes [3][4][5][6][7][8][9][10][11][12][13][14][15][16][17] was expressed as a fraction of 1.0. The difference between the total and the fraction still bound to the radioactive probe equaled the fraction of XGRAF bound to the competitor DNA. The equilibrium binding equation describes the interaction between XGRAF and DNA (Equation 1), where f is the ratio of bound XGRAF to total XGRAF, [DNA⅐XGRAF]/ [XGRAF] t . As explained above, [XGRAF] t has been defined as 1.0. Since we are not calculating an absolute value for the equilibrium constant, K d is substituted with the term C 50 , which represents the -fold excess of competitor required to displace 50% of XGRAF from the probe DNA, and Equation 2 is simplified to Equation 3.
[DNA⅐XGRAF] is expressed as a fraction of [XGRAF] t .
[DNA] t is expressed in units of -fold excess of unlabeled competitor DNA over radioactively labeled DNA. In Fig. 1B, the data from Fig. 1A were plotted according to Equation 3, and C 50 was calculated as the reciprocal of the y intercept. For each of the 27 single-point mutations, C 50 was determined in this way in three independent experiments. Values for [DNA⅐XGRAF] below 0.1 or above 0.9 were not included in the Scatchard plots. Essentially the same results were obtained when C 50 was calculated as the negative reciprocal of the slope. Methylation Interference Footprinting-DNA probes containing ␥ gene upstream sequence from positions Ϫ232 to Ϫ6, 32 P-labeled on the sense strand, and from positions Ϫ232 to Ϫ115, 32 P-labeled on the antisense strand, were produced by PCR with 0.1 M primers and Taq enzyme (24). The PCR products were purified through a native 6% polyacrylamide gel (26) followed by organic extraction. The end-labeled DNA fragments (ϳ1 ϫ 10 6 cpm/pmol) were partially methylated at the guanine moieties by incubation for 2-3 min with dimethyl sulfate without carrier (30); in some cases underwent organic extraction; were precipitated twice with ethanol; and were dissolved in 10 mM Tris-HCl and 0.1 mM EDTA (pH 8.0). For preparative binding, the reaction volumes described above for the gel shift assay were scaled up 15-30fold, with ϳ300,000 cpm of DNA probe, and electrophoresis was carried out at 250 V for 2.5-3.5 h at 4°C. Wet gels were exposed to XAR film at 4°C overnight; portions of the gel containing bound and unbound DNA were excised; and DNA was eluted by shaking overnight once or twice at 37°C in 0.2 M NaCl, 10 mM Tris-HCl (pH 7.5), and 0.1 mM EDTA. The eluted DNA was purified over a NACS column (Life Technologies, Inc.) according to the manufacturer's instructions, and 8 -16 g of carrier yeast RNA was added. After ethanol precipitation, piperidine cleavage was performed (31), and piperidine was removed (24). The entire recovered bound DNA fraction and an approximately equal amount of radioactivity of the unbound DNA were analyzed by gel electrophoresis as described previously for methylation protection footprinting (24). The gels were exposed to XAR film at Ϫ80°C with one Lightning Plus intensifying screen.
Transfection and Assays for Reporter Gene Activity-Isolation of hepatocytes from three adult female X. laevis frogs and transfection were as described previously (24), except that 10 g of control plasmid pCMV-␤gal (32) was used, and electroporation was conducted at a setting of 130 V. The cells were divided equally and incubated with or without 10 Ϫ7 M dexamethasone and 10 Ϫ9 M triiodothyronine in 12-well Primaria plates (Falcon) with 1 ϫ 10 6 cells in 4 ml of medium. After 44 -48 h, cell extracts were prepared, and the activities of ␤-galactosidase (using the Galactolight method) and of luciferase were assayed as described (24).
Data Analysis for Transfection Assays-Luciferase values were normalized to the ␤-galactosidase activity in each transfection. -Fold hormonal induction was calculated by dividing the amount of luciferase activity/␤-galactosidase activity in hormone-treated cells by the amount of luciferase activity/␤-galactosidase activity in untreated cells. The data for each mutant are expressed relative to wild-type DNA as follows. XGRAF activity with wild-type DNA is defined as F WT Ϫ F D , where F WT represents the -fold increase in transcription in response to hormone treatment for the wild-type construct pLL␥Ϫ187 and F D represents the -fold increase in transcription in response to hormone treatment for the mutation D construct, which has multiple mutations in the XGRAF site. Mutation D ( Ϫ175 ACT Ϫ173 ) abolished the XGRAF-binding site, so the remaining -fold response is due solely to the half-GREs (24). For each single-point mutant, the percent of wild-type XGRAF activity was calculated as ((F M Ϫ F D )/(F WT Ϫ F D )) ϫ 100, where F M is the -fold increase in transcription in response to hormone treatment for each mutant. In a total of eight experiments, the average F WT was 5.3 Ϯ 0.6 (mean Ϯ S.E.), and the average F D was 2.2 Ϯ 0.2 (mean Ϯ S.E.). Each mutant was tested in three independent experiments.

Identification of Specific Guanosines Involved in XGRAF
Binding-Previously, we mapped the DNA elements between 177 and 135 bp upstream of the transcription start site that are important for glucocorticoid regulation of the Xenopus ␥-fibrinogen subunit gene (24). Three sites designated half-GRE1 (positions Ϫ168 to Ϫ163), half-GRE2 (positions Ϫ156 to Ϫ151), and half-GRE3 (positions Ϫ140 to Ϫ135) in Fig. 2C  GRE1 is the most critical of the three GREs for hormonal induction. In addition, the tract from positions Ϫ177 to Ϫ169 is necessary for full hormone responsiveness even though it does not match the consensus GRE. The nuclear protein XGRAF binds to this region of the DNA (Fig. 2C).
To identify bases that have direct contacts with XGRAF, we used the methylation interference assay, which disrupts protein binding by methylation of critical guanines within a recognition site. A radioactively labeled DNA fragment including nucleotides Ϫ177 to Ϫ169 of the ␥ gene upstream region was partially methylated in vitro. The DNA was incubated with Xenopus liver nuclear extract, and both protein-bound DNA and free DNA were isolated by preparative native gel electrophoresis. The DNA was cleaved at all methylated guanosines, and the two populations of DNA were compared for relative abundance of fragments ending at particular positions (Fig. 2). When the sense strand was radioactively labeled (sequence shown in Fig. 2C), the two DNA fragments ending at positions Ϫ175 and Ϫ173 were significantly reduced in intensity in the protein-bound DNA fraction (Fig. 2A, lane B) as compared with the free DNA fraction (Fig. 2A, lane F), indicating that methylation of these guanines interfered with binding of a protein in the nuclear extract. Thus, bases G Ϫ175 and G Ϫ173 are important contact points for XGRAF. The intensity of the fragment ending at G Ϫ184 was also reduced. This nucleotide is not considered to be part of the core XGRAF-binding site since the intervening bases from positions Ϫ178 to Ϫ181 are not required for XGRAF binding or function (24). When the antisense strand was radioactively labeled, no differences in intensities of bands were observed (Fig. 2B, compare lanes F and B). Although no guanosines are present on the antisense strand within the putative XGRAF-binding site between positions Ϫ177 and Ϫ169, this experiment confirmed that the binding site does not extend upstream to position Ϫ182 or downstream to position Ϫ164, where the nearest guanosines are located.
Effect of Point Mutations on XGRAF Binding Ability-The relative contribution of each nucleotide to the specific interaction between XGRAF and the ␥ DNA was assessed by saturation mutagenesis of the site from positions Ϫ177 to Ϫ169, generating all 27 single-point mutants in the 9-bp sequence. The effect of the mutations on DNA binding in vitro was analyzed by the gel shift assay, using the mutated DNA sequences as competitors for binding of XGRAF to radioactively labeled wild-type DNA, as described under "Experimental Procedures." Binding ability is expressed as C 50 , the -fold excess of competitor required to displace half of XGRAF from the wild-type probe. For each mutant, the C 50 value was calculated in three independent experiments, and the results are presented as the mean Ϯ S.E. of the three determinations ( Fig. 3 and Table I). The wild-type DNA had a C 50 value of 1.7-fold excess.
At positions Ϫ177, Ϫ176, and Ϫ174, all of the mutants bound strongly to XGRAF, with low C 50 values of 1.7-3.8-fold excess. Therefore, these positions are not critical determinants for binding specificity since any nucleotide can be substituted without significantly affecting the interaction with XGRAF. Similarly, changing G Ϫ175 to either C or T was not deleterious to binding. However, when this position was mutated to A, the binding ability was reduced, with a C 50 of 36-fold excess. It is interesting that the introduction of A at position Ϫ175 creates a stretch of six adenosines, which may interfere with binding due to structural alterations rather than the specific nucleotide substitution (33).
In contrast, nearly all changes in the bases from positions Ϫ173 to Ϫ169 substantially impaired XGRAF binding ability ( Fig. 3 and Table I), with C 50 values from 20-fold excess for mutant C Ϫ171 to 323-fold excess for mutant C Ϫ173 . The only exception was mutant C Ϫ169 , which retained relatively strong XGRAF binding (C 50 ϭ 6.3-fold excess).
Effect of Point Mutations on Glucocorticoid Responsiveness-We also examined the effects of the point mutations in the XGRAF-binding site on glucocorticoid induction of transcription. The mutated ␥ gene upstream region was inserted into a luciferase reporter vector, and the constructs were transfected into primary Xenopus hepatocytes. Transfected cells were divided for plus or minus glucocorticoid treatment for 2 days, and lysates were analyzed for luciferase activity. The total -fold increase in luciferase levels in hormone-treated cells reflects the role of not only the XGRAF site, but also the three half-GREs in the upstream regulatory region of the ␥ gene (Fig.  2C). As described under "Experimental Procedures," the effect of the mutations only on the XGRAF contribution to the hormonal stimulation was assessed by comparing each singlepoint mutant with wild-type DNA, representing 100% induction, and with a triple mutant in the XGRAF-binding site, which eliminated XGRAF binding and therefore represented 0% activity of XGRAF in the induction. The triple mutant, which changed Ϫ175 GAG Ϫ173 to Ϫ175 ACT Ϫ173 , has been shown previously to reduce glucocorticoid responsiveness through ef- FIG. 3. Relative ability of single-point mutants of the ␥ DNA from positions ؊177 to ؊169 to bind XGRAF. Using the gel shift assay and Scatchard analysis described under "Experimental Procedures" and exemplified in Fig. 1, the ability of each mutant to bind to XGRAF was quantitated. The wild-type (WT) nucleotides at positions Ϫ177 to Ϫ169 and their corresponding single-point mutations (mut) are indicated along the y axis. C 50 on the x axis represents the -fold excess of mutant DNA required as competitor to displace half of XGRAF from the radioactively labeled wild-type DNA probe. The data are shown as the mean Ϯ S.E. of values determined in three independent experiments. For each mutant, three independently produced preparations of competitor DNA and at least two different batches of nuclear extract were used (except for mutant A Ϫ172 , for which the data were derived from two experiments with two different preparations of competitor and one nuclear extract). fects on the XGRAF-binding site rather than the GRE (24). Retention of at least 60% of XGRAF activity was considered normal function, whereas activity below 47% was defined as impaired.
As shown in Fig. 4 and Table I, full XGRAF activity ranging from 79 to 135% of the value obtained with wild-type DNA was achieved for all single-nucleotide substitutions at positions Ϫ177, Ϫ176, and Ϫ174. Hence, these positions are not essential for conferring XGRAF function on the ␥ gene. At position Ϫ175, XGRAF activity was reduced to 33% when the site was mutated to A, but 85 and 107% of wild-type function were attained with the T and C substitutions, respectively. Thus, the only functionally deleterious mutation from positions Ϫ177 to Ϫ174 is the G to A transition at position Ϫ175.
Conversely, almost all mutations at positions Ϫ173 to Ϫ169 significantly decreased XGRAF activity to between 11 and 46% of function with wild-type DNA (Fig. 4 and Table I). The most dramatic departure from this general pattern is the G to C transversion at position Ϫ173, which improved the glucocorticoid induction, apparently to 453% of normal XGRAF activity. The other two exceptions are mutants G Ϫ172 and C Ϫ171 , which allowed 60 and 71% of wild-type activity, respectively.
Effect of Point Mutations on Basal Transcription-The level of transcription of the single-point mutants in the absence of hormone treatment ranged from 48 to 142% of that of wild-type DNA (Table I). Each data point for basal expression was obtained from an independent transfection event, whereas changes due to hormone treatment were assessed on a single cell population that was divided after the transfection. Table I, the ability of DNA with single-point mutations at positions Ϫ177 to Ϫ169 to bind to XGRAF in the gel shift assay is classified as strong (ϩ) if half-maximal competition was achieved with Ͻ7-fold molar excess of mutated competitor over wild-type probe or as weak (Ϫ) if 20-fold or greater excess competitor was needed. Similarly, in Table I, the ability of DNA with each single-point mutation to enhance glucocorticoid-induced transcription in transfected primary hepatocytes is labeled as high (ϩ) if at least 60% of wild-type XGRAF activity was retained or low (Ϫ) if 46% or less of wild-type XGRAF activity was observed. With only four exceptions, which will be discussed below, high activity in the transfection assay correlated with strong binding ability in the gel shift assay, whereas low functional activity corresponded with weak binding of XGRAF to the mutated DNA in vitro. This correlation can be seen clearly in a plot of XGRAF activity as a function of C 50 (Fig. 5). The excellent agreement between binding and function firmly supports the conclusion that XGRAF, defined by its interaction with DNA in vitro, is the same protein that plays a role in hormonal stimulation of gene transcription in vivo. Ϫ169 of the ␥ gene to bind XGRAF, to enhance glucocorticoid induction of transcription, and to regulate basal transcription a C 50 , -fold excess of mutant competitor required to displace half of XGRAF from wild-type DNA. b C 50 values Ͻ7-fold excess were defined as strong (ϩ) binding and Ն20-fold excess as weak (Ϫ) binding. c XGRAF activity, hormonal stimulation attributable to XGRAF for mutated DNA as a percentage of that for wild-type DNA. d XGRAF activity Ն60% was defined as strong (ϩ) and Յ46% as weak (Ϫ). Gray shading indicates mutants with a discrepancy between binding and function. The most striking discrepancy between physical association with XGRAF and ability to amplify glucocorticoid responsiveness was seen with mutant C Ϫ173 , for which the hormonal induction was much higher than for any other construct, while ability to bind XGRAF was the weakest (Table I). This mutant was not included in Fig. 5 because its functional activity was much greater than that of the other mutants. Our method of computation attributed the stimulation of transcription to XGRAF, but we believe that the effect in this case was actually due to GR. The C Ϫ173 mutation generated the following sequence: Ϫ177 AAGACTnnnTGTTCC Ϫ163 , with two matches to the upstream half of the consensus GRE (GGTACAnnnTGT-TCT) in addition to the five out of six matches to the downstream half. We have shown previously that the C at position Ϫ173 is essential for GR to interact with this site as a dimer and for hormonal stimulation greater than 10-fold (24). Therefore, the strong glucocorticoid response seen with the singlepoint mutation to C Ϫ173 could be explained by strong GR binding that eliminated the role of XGRAF in the induction. A comparable effect was not expected with any of the other nucleotides because even when all the bases except Ϫ173 were changed to match the consensus GRE, no increase in GR dimer binding or hormonal induction was observed (24).

Correlation between XGRAF Binding to DNA in Vitro and Stimulation of Transcription-In
Two other mutants, G Ϫ172 and C Ϫ171 , were also classified in Table I as having weak binding while retaining the capacity to stimulate transcription. The levels of XGRAF transcriptional activity were, however, moderate at 60 and 71%, respectively, the lowest values that were still considered positive. The binding abilities (C 50 values of 24-and 20-fold excess), although defined as weak, were intermediate between the strongest and weakest. In Fig. 5, the data for these mutants are represented by the two points in the center of the plot, which lie close to the line and therefore show good correlation between binding and function.
Mutant C Ϫ169 had a striking disjunction between strength of binding to DNA and ability to amplify glucocorticoid action. XGRAF bound quite well to the C Ϫ169 mutant since only a 6.3-fold molar excess of this DNA was required to displace half of XGRAF from the wild-type probe in the gel shift assay (Table  I). Nonetheless, hormonal stimulation was very poor, with only 25% of wild-type XGRAF activity. The distinction between the C Ϫ169 mutant and all other constructs is evident from the anomalous position of the C Ϫ169 data in the lower left portion of Fig. 5. These results cannot be explained by changes in the interaction of GR with the DNA since the mutation lies within the 3-bp region between the two half-sites of the GRE, which is not critical for GR binding and function (23). Although the C Ϫ169 mutant binds fairly tightly to XGRAF in vitro, we hypothesize that it is incapable of conferring a functionally important conformational change on the protein that would occur upon binding to the natural recognition sequence.
Another important question is whether XGRAF had general effects on transcription independent of its amplification of GR action. The level of basal transcription for each of the singlepoint mutants is shown in Table I. Basal transcription was inherently more variable than -fold hormonal induction in the transfection assay because each data point was derived from an independently transfected sample. When transcriptional activity in the absence of hormone treatment was plotted versus binding ability (Fig. 6), only a slight correlation was observed. Therefore, we conclude that XGRAF may have a small effect on general transcription of the ␥ subunit gene, but that the major function of XGRAF is to enhance glucocorticoid induction in response to GR. This specificity is in contrast to many other glucocorticoid receptor accessory factors (such as nuclear factor-1; activator protein-1; cAMP response element-binding protein; hepatocyte nuclear factor-1, -3, and -4; and chicken ovalbumin upstream promoter transcription factor), which also stimulate basal transcription (2, 10 -17).
Consensus Sequence for the XGRAF Recognition Site-Based on the physical and functional data presented here, a consensus sequence can be derived for the XGRAF-binding site that reflects the most favorable nucleotide at each position. Bases Ϫ177 and Ϫ176 are no longer considered part of the recognition sequence since no substitutions at these positions affected XGRAF binding or activity. For nucleotides Ϫ175 to Ϫ169, the consensus sequence is BNGTTAA (B ϭ C, G, or T; N ϭ A, C, G, or T). Even with this more well defined site, we found no striking matches to recognition sequences in the transcription factor site data base TRANSFAC 3.2 (34) using the TESS FIG. 5. Correlation between the ability of the single-point mutants in region ؊177 to ؊169 to amplify hormone-induced transcription and to bind XGRAF. The portion of the glucocorticoid response attributable to XGRAF, expressed as XGRAF activity with mutant DNA as a percentage of that with wild-type DNA, is plotted as a function of the ability of the DNA to bind XGRAF in vitro, expressed as C 50 . See "Experimental Procedures" and "Discussion" for details.
FIG. 6. Correlation between the ability of the single-point mutants in region ؊177 to ؊169 to stimulate basal transcription and to bind XGRAF. Basal transcription, the luciferase activity/␤galactosidase activity in untreated cells transfected with mutant DNA expressed as a percentage of that with wild-type DNA, is plotted as a function of the ability of the DNA to bind XGRAF in vitro, expressed as C 50 . See "Experimental Procedures" and "Discussion" for details. searching program (36). 2 Thus, XGRAF appears to have a novel sequence specificity for binding to DNA.
Models for Interaction of XGRAF and GR with Contiguous or Overlapping Binding Sites-Previously, we presented four possible models for the interaction of XGRAF and GR with DNA at closely juxtaposed sites (24). Model 1 depicted simultaneous binding of XGRAF to its site at nucleotides Ϫ175 to Ϫ169 and GR binding as a dimer at nucleotides Ϫ177 to Ϫ163, which would constitute a full-length GRE. To bind these sites concurrently, XGRAF must contact the DNA in the minor groove since GR is known to occupy the major groove (35). However, the methylation interference experiment (Fig. 2) established that XGRAF also binds in the major groove since modification of the N-7 positions of guanines, which are accessible only in the major groove, interfered with binding. Therefore, Model 1 is not a likely mechanism for interaction of GR and XGRAF with their respective sites.
Model 2 proposed an interaction between XGRAF and a monomer of GR. Model 3 depicted a trimeric complex consisting of one molecule of XGRAF and a dimer of GR, with GR contacting only the downstream half of the GRE. Both of these scenarios are possible but must take into account that the XGRAF-and GR-binding sites are directly contiguous. In Model 4, binding of XGRAF and GR was sequential rather than simultaneous, which would obviate problems of steric hindrance for two protein molecules binding to abutted recognition sites. Experiments are in progress to distinguish between these mechanisms.
Classically, the presence of a receptor defined a tissue as being a target for a steroid hormone, and the presence of a high affinity receptor-binding site on the DNA was a prerequisite for a responsive gene. It is becoming increasingly clear that steroid hormone action is dramatically influenced by many other aspects of the local cellular environment and the structure of the gene regulatory region. Accessory DNA-binding proteins such as XGRAF can play as important a role as the receptor in determining the extent of hormonal induction. The fact that diverse cellular responses rely on unique combinations of accessory DNA-binding proteins, coactivators, corepressors, and other factors makes it possible for multiple control mechanisms to regulate genes differentially in the same cell in response to a single hormonal stimulus.