AML3/CBFα1 Is Required for Androgen-specific Activation of the Enhancer of the Mouse Sex-limited Protein (Slp) Gene*

A complex 120-base pair enhancer, derived from the mouse sex-limited protein (Slp) gene, is activated solely by the androgen receptor (AR) in specific tissues, although it contains a hormone response element recognized by several steroid receptors. The generation of this transcriptional specificity has been ascribed to the interactions of the receptor with tissue-specific nonreceptor factors bound to accessory sites within the enhancer. Protein-DNA interaction assays revealed two factors binding the 5′ part of the enhancer that differ widely in abundance between cells showing AR-specific activation of the Slp element compared with those that also permit activation by glucocorticoid receptor (GR). The factor designated B formed a complex centered on the sequence TGTGGT, a core motif recognized by members of the AML/CBFα transcription factor family. This complex was competed by a high affinity binding site specific for AML/CBFα and was specifically supershifted by an antibody to AML3/CBFα1, placing factor B within the AML3/CBFα1 subclass. Interestingly, this factor was shown to bind to a second site in the 3′ part of the enhancer, positioned between the two critical AR binding sites. Transfection studies revealed that AML1-ETO, a dominant-negative AML/CBFα construct, abrogated AR induction of the enhancer, but not of simple hormone response elements. Furthermore, overexpression of AML3/CBFα1 could rescue the AML1-ETO repression. Finally, glutathione S-transferase-AML/CBFα fusion proteins demonstrated direct interaction between AML/CBFα and steroid receptors. Although this interaction was equivalent between AML1/CBFα2 and AR or GR, AML3/CBFα1 showed stronger interaction with AR than with GR. These data demonstrate that AML3/CBFα1 is functionally required for hormonal induction of the Slpenhancer and that direct, preferential protein-protein interactions may contribute to AR-specific activation. These results demonstrate an intriguing role of AML3/CBFα1 in steroid- as well as tissue-specific activation of target genes.

Steroid receptors mediate a wide variety of hormonal effects by their direct modulation of gene expression. Transcriptional activation stems primarily from receptor recognition of specific DNA hormone response elements (HREs) 1 (1)(2)(3). However, it remains a paradox how the signaling pathways of receptors are distinguished at the DNA level in vivo, since in vitro several receptors recognize the same consensus HRE (4 -8). DNA elements that vary from the consensus and provide preferential recognition by a single receptor exist but have not yet proven broadly applicable (6,9). A more versatile mechanism to generate specificity invokes the participation of nonreceptor factors bound to DNA sequences neighboring HREs of hormonally responsive genes (5,10,11). Differential interactions between such factors and steroid receptors may organize an enhancerspecific activation complex that confers transcriptional specificity dependent on information intrinsic to both the gene and the receptor.
To study transcriptional specificity imposed by the androgen receptor (AR), we have characterized the enhancer of the mouse sex-limited protein gene (Slp) (4,12,13). Slp expression is androgen-dependent and tissue-specific, occurring primarily in liver and kidney (14 -16). The 120-bp enhancer residing 2 kilobases upstream of the gene contains a consensus HRE that is necessary but not sufficient for hormonal induction in transfection. In monkey kidney CV-1 cells, the enhancer is activated only by AR, but in mammary carcinoma T47D cells, glucocorticoid (GR) and other steroid receptors can drive gene expression as well (4,13). This demonstrates that non-HRE enhancer sequences and cell-type restricted factors are required for AR specificity.
Distinct sites within the enhancer that are crucial for efficient androgen induction, in addition to the consensus HRE, include a perfect half-site HRE and a sequence similar to that recognized by the ubiquitous octamer transcription factor, Oct-1 (17)(18)(19)(20). These two sites may be involved in hormonal specificity, as they have differential effects on induction by AR compared with GR in contexts where both receptors can activate (17). Linker-scanning (LS) mutations and deletions of the 5Ј part of the enhancer define a 44-bp region critical for AR (or GR) induction (13), suggesting that this part of the enhancer harbors factors for efficient activation. Moreover, LS mutations within the 5Ј region differ in effect in CV-1 versus T47D cells, suggesting tissue-specific variation (12). In this study, we characterized factors interacting with this part of the enhancer, detected in protein-DNA gel shift complexes. One factor, initially designated B, has been identified as the transcription factor AML3/CBF␣1, a member of the AML/CBF␣/PEBP2 transcription factor family (21)(22)(23). These proteins have broad significance in gene regulation, as reflected by their isolation for roles in acute myeloid leukemia (AML) and as viral enhancerbinding proteins (retroviral core binding factor (CBF) and polyoma enhancer-binding protein (PEBP)) but have not previously been demonstrated to be associated with androgen action.

MATERIALS AND METHODS
Cell Culture and Preparation of Nuclear Extracts-CV-1 cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (Hyclone), and T47D cells were grown with 5% serum plus 1 g of insulin/ml. CV-1 or T47D nuclear extracts were prepared by high salt extraction (47). Briefly, cells at about 90% confluence were collected, washed with phosphate-buffered saline solution, and homogenized by Dounce in hypotonic buffer (20 mM HEPES, pH 7.9, 1 mM EDTA, 1 mM dithiothreitol, 0.5 mM phenylmethylsulfonyl fluoride, 0.1 M aprotinin, 1 M pepstatin, 1 M leupeptin). After centrifugation at 1,000 ϫ g, the nuclear pellets were washed and resuspended in the hypotonic buffer in a volume equal to that of the pellets. The suspensions were extracted with an equal volume of 0.9 M KCl in the hypotonic buffer containing 20% glycerol. After a 1-h incubation on ice with occasional vortexing, the extracts were centrifuged, and the supernatants were dialyzed against binding buffer (20 mM HEPES, pH 7.9, 1 mM EDTA, 75 mM KCl, 10% glycerol). Extracts were quantified by Bio-Rad protein assay.
DNA-Protein Interaction Analysis-Gel shift assays were performed as described previously (13,17) with some modifications. 5 g of nuclear extracts were preincubated at room temperature for 10 min with 2 g of poly(dI-dC) in binding buffer containing 1 mM dithiothreitol, 0.05% Nonidet P-40, 0.5 mM phenylmethylsulfonyl fluoride, 20 M ATP, and 0.2 g/ml BSA. 32 P end-labeled probes as indicated in the figure legends were added at 0.1-0.5 ng with or without unlabeled competitors to the preincubated mixtures and incubated for an additional 20 min in a final reaction volume of 15 l. The reaction mixtures were electrophoresed on pre-run 8% native acrylamide gels (37.5:1, acrylamide:bis) in 0.5 ϫ Tris borate-EDTA. Gels were dried and autoradiographed. The oligonucleotides used to identify factor B as a member of the AML/CBF␣ family included a high affinity core binding site (23), 5Ј-GGATATTTGCGGT-TAGCA-3Ј, and its mutant, 5Ј-GGATATTGCCATTAGCA-3Ј. For antibody supershift experiments, 0.2 g of anti-AML3/CBF␣1 or anti-AML1/CBF␣2 antibodies (33) (Calbiochem), both having high specificity to their respective subtypes, were used in gel shift assays.
For DNase I footprinting, a PvuII-BamHI fragment containing the 120-bp enhancer was excised from CЈ⌬9tkCAT (13), 32 P-labeled at the BamHI sites for either top or bottom strand by Klenow fill-in or polynucleotide kinase reaction, and gel-purified. The DNase I reaction was performed as described previously (13,17). The sequence ladder of the fragment for comparison to the protected region was achieved by cleaving naked probe with piperidine (A/G).
For transfection, CV-1 or T47D cells were plated at 50% confluence in 6-well plates in Dulbecco's modified Eagle's medium containing 5% charcoal-treated NuSerum IV (Collaborative Research) and cotransfected by the calcium phosphate precipitation method, with plasmids as indicated in figures. Differing plasmid DNA amounts between plates were adjusted with plasmid pGem3. Following overnight incubation, CV-1 cells were washed twice with phosphate-buffered saline, and T47D cells were glycerol-shocked for 3 min before phosphate-buffered saline washing. New medium was then added, with or without 30 nM dihydrotestosterone for AR or 30 nM dexamethasone for GR, for 28 -30 h. CAT activity was measured as described previously (13) and was expressed as fold induction (reporter activity in the presence of hormone relative to uninduced activity).
Protein-Protein Interaction Analysis-To test direct interaction of AML3/CBF␣1 with AR or GR, a glutathione S-transferase (GST) fusion to AML3/CBF␣1 was constructed by inserting the murine cDNA into pGEX-3X using the BamHI-SmaI sites. The coding sequence was generated by polymerase chain reaction amplification of pEF-BOS␣A1 with the primers 5Ј-GAAGATCTCCATGCGTATTCCTGTAGATCC-3Ј and 5Ј-ACCAGCTGATATGGCCGCCAAACAGACTC-3Ј. The polymerase chain reaction product was digested with BglII and PvuII and inserted into pGEX-3X. Similarly, GST-AML1/CBF␣2 was constructed by inserting the BglII-and NaeI-digested polymerase chain reaction product generated from pKS-AML1/CBF␣2 with primers 5Ј-GTAGATCTCCAT-GCGTATCCCCGTAGATGC-3Ј and 5Ј-ATAGCCGGCGTAGGGCCGCC-ACACGGC-3Ј. The resulting constructs were confirmed by restriction enzyme digestion. The expression of GST and fusion proteins was induced with 0.1 mM isopropyl-␤-D-thiogalactopyranoside, and equal amounts of induced preparations were immobilized on glutathione Sepharose 4B (Amersham Pharmacia Biotech) according to the manufacturer's instructions. pGEM4-AR and pC7G-GR have been described previously (4,12); their encoded proteins were produced by in vitro transcription and translation with the TnT kit (Promega) according to the manufacturer's instructions with [ 35 S]methionine (Amersham Pharmacia Biotech). The DNA binding form of the receptors was achieved by a 15-min incubation of translates at 30°C in the presence of 50 nM respective hormone (49). The interaction was performed by incubating equivalent amounts of translated AR or GR, judged by incorporated radioactivity, with the GST fusion proteins or GST alone in binding buffer (1 mM dithiothreitol, 0.05% Nonidet P-40, 0.5 mM phenylmethylsulfonyl fluoride, 0.2 mg/l bovine serum albumin). After a 2-h incubation on ice, the Sepharose pellets were washed 4 ϫ 1 ml with the reaction buffer, and the retained receptors were resolved in 8% polyacrylamide SDS gels (SDS-polyacrylamide gel electrophoresis) and visualized by fluorography.

Two Proteins Bind the 5Ј Part of the Enhancer and Differ in
Abundance in Different Cell Types-As noted previously, two linker scan mutants, LS3 and LS4, and deletions of the 5Ј part of the enhancer (CЈ⌬9) abrogated AR inducibility of CЈ⌬9tkCAT in CV-1 cells (13). DNA footprints confirm that LS3 and LS4 disrupted protein-DNA interactions observed at these sites with the wild type enhancer (13), suggesting the necessity of these factors for efficient AR activation. To characterize these factors, gel shift assays were performed with a 44-bp oligonucleotide, Oligo I, corresponding to the 5Ј part of the enhancer, and two overlapping probes, Oligo II and Oligo III (Fig. 1A). With Oligo I and CV-1 nuclear extract, two major protein-DNA complexes of similar intensity were observed (Fig. 1B), complex A and B; a third complex, C, was less reproducible. Complex formation was specific in that it was competed by unlabeled Oligo I itself but not by unrelated oligonucleotides containing sites for NF-B, HNF4, and HNF-3␤. The FPIV probe, containing the Oct-1 motif, caused a noticeable reduction in complex A at 100-fold molar excess, but this was insignificant compared with competition by Oligo I.
Competition with unlabeled Oligo II and Oligo III assigned complex A to Oligo II and B to Oligo III (Fig. 2, upper left panel). This separation of complexes was further confirmed with labeled Oligo II or III as probe (Fig. 2, lower panels), suggesting that the region of overlap between Oligo II and III was not sufficient for formation of either complex. Gel shift assays with these Oligos and nuclear extracts from T47D cells revealed complexes similar in position to those observed with CV-1 extracts but with dramatic differences in intensity (Fig. 2, upper right panel). Complex A was much more abundant in T47D than in CV-1 cells, whereas the B complex could only be faintly detected when unlabeled Oligo II was added as a competitor. Similar to CV-1 proteins, the T47D Complex A associated independently with Oligo II, and complex B associated independently with Oligo III (Fig. 2, lower panels). These data demonstrate that two factors interact with the 5Ј part of the enhancer and vary in abundance in different cells.
Sequences Required for Complex B-DNA sequences involved in the formation of complexes A and B were determined by DNase I footprinting. A PvuII-BamHI DNA fragment containing the 120-bp enhancer was used to bind nuclear extracts. In addition to the previously observed FPIV region (13,17), two major regions were protected following incubation with CV-1 nuclear extracts (Fig. 3A). By comparison to the sequence ladder of the same fragment, the upper protected region was within Oligo III, indicating that the sequence labeled b was protected by the factor for complex B. This was further confirmed by the loss of protection upon addition of excess Oligo III but not Oligo II. Interestingly, a lower protected region, 5Ј of the consensus HRE, also disappeared with excess Oligo III but not Oligo II. This result could indicate that the same factor was involved in both protections or that factor B interacted with a factor required for the lower protection. Examination of the lower protected region revealed a centrally located TGTGGT sequence, as in the region bound by factor B.
To investigate whether the second protected region resulted from factor B occupancy, Oligo IV, including sequences from the 3Ј part of FPIV to the 5Ј edge of the HRE (see Fig. 1A), was used in gel shift assays. When the minimal sequence b demarcated from the footprint was used as a probe, a complex comigrating with complex B formed with CV-1 extracts (Fig. 3B) as well as with T47D extracts (data not shown). The complex disappeared with excess unlabeled Oligo III, b, and Oligo IV but not with Oligo II, indicating that the minimal sequence was sufficient for formation of complex B and suggesting that this factor also was responsible for the lower protected footprint. When Oligo IV was used as a probe (Fig. 3B, right), the major DNA-protein complex observed was competed by unlabeled Oligo III and b but not by Oligo II, demonstrating that the Oligo IV complex is similar to complex B and that the factor is likely common to both. Thus a second site within the enhancer interacts with factor B. The two sites correspond, respectively, to the LS4 and LS8A mutations (13,17). Both of these mutations decrease androgen inducibility of the CЈ⌬9 enhancer to less than 20% of wild type (17). In the context of a larger enhancer fragment that, unlike CЈ⌬9, has some activity in the absence of hormone, these mutations also reduce uninduced expression. Thus, factor B appears to play a critical role at both sites in AR activation and in enhancer activation in general.
The minimal sequence for complex A could not be discerned from the narrow protected region above that for factor B, although excess Oligo II reduced protection at that site (Fig. 3A). Mutated Oligo II probes used in gel shift assays led to definition of a palindromic sequence with two crucial central T residues: AGATTTTAATCT. This sequence was sufficient for the formation of complex A with both CV-1 and T47D extracts (data not shown). No match to the sequence was found in the data base of DNA-interacting proteins, suggesting A may be a novel factor.
Factor B Is AML3/CBF␣1-The ability of factor B to interact with two sites within the enhancer suggested that the common sequence TGTGGT was critical for recognition. A data base search of transcription factors revealed this to be the core motif shared by the AML/CBF␣ family (21,22). The DNA binding domain of this family is highly conserved and homologous to that of the Drosophila runt protein (26). Binding sites for these factors have been found in many viral and cellular genes, particularly those involved in hematopoiesis and osteogenesis (22,23,40). To determine whether factor B belonged to this family of transcription factors, an 18-bp oligonucleotide containing a high affinity (HA) site for the family was used in gel shift assays (23). The complex formed with HA and CV-1 extract could be competed by HA itself, by Oligo III, and by sequence b but not by Oligo II (Fig. 4). Furthermore, HA, like Oligo III, antagonized the formation of complex B with the minimal sequence b (Fig. 4), whereas a mutant HA probe did not compete (not shown). HA also antagonized the complex formed with Oligo IV (not shown). Similar results were observed with T47D extracts (data not shown). These data suggest that factor B is a member of the AML/CBF␣ transcription factor family.
Three subtypes of AML/CBF␣ factors have been described: AML1/CBF␣2, AML2/CBF␣3, and AML3/CBF␣1 (21,22,24). To distinguish which subtype includes factor B, subtype-specific antibodies were used in gel super-shift assays. As shown in Fig. 4 (middle panel), anti-AML3/CBF␣1, but not anti-AML1/ CBF␣2, antibody shifted the major complex formed with the minimal sequence b and CV-1 extract. Rabbit pre-serum and other antibodies tested failed to do so (data not shown). The Oligo IV complex was also supershifted specifically by anti-AML3/CBF␣1 antibody (Fig. 4, right). To confirm these observations, the anti-AML/CBF␣ antibodies were evaluated with nuclear extracts from NIH 3T3 cells, which express high levels of AML3/CBF␣1 (48). Again, anti-AML3/CBF␣1, but not anti-AML1/CBF␣2 antibody, supershifted the complex formed with the HA probe (data not shown). The lower minor band in Fig. 4 (right) was not affected by either antibody, suggesting it is not due to an AML/CBF␣-related protein. Similar results were also obtained with T47D cells (data not shown). These data indicate that AML3/CBF␣1 is the major AML/CBF␣ family member in both CV-1 and T47D cells, although it is much less abundant in the latter. The difference in AML3/CBF␣1 levels may contribute to differences in activation of the Slp enhancer by AR or GR in the two cell lines.
AML3/CBF␣1 Is Required for Efficient AR Induction of the Enhancer-To test a role in hormonal induction, an AML3/ CBF␣1 expression vector (48, 50) and a dominant-negative AML1-ETO plasmid (27,35) were used in transfection. The AML3/CBF␣1 used was the cDNA originally cloned from NIH 3T3 cells as PEBP2␣A (48). AML1-ETO is the t(8; 21) translocation product frequently found in acute myeloid leukemia (36). This protein functions as a dominant-negative for the AML/ CBF␣ family by interfering with their ability to transactivate (27,(33)(34)(35). In preliminary tests, the vectors were assessed in CV-1 cells with a tkCAT reporter driven by four copies of the minimal sequence b. Overexpression of AML3/CBF␣1 not only increased reporter activity but also reversed the inhibition caused by AML1-ETO (data not shown). For the Slp enhancer in CV-1 cells, overexpression of AML3/CBF␣1 did not further increase AR-specific induction but efficiently rescued the total repression of activation provoked by AML1-ETO (Fig. 5, upper  panel). In contrast, similar treatments had no significant effect on AR-mediated activity of 3xHREtkCAT (Fig. 5, lower panel). This indicates that the effects of the two proteins are specific for the complex enhancer and that AML3/CBF␣1 is required for CЈ⌬9 activation by AR in CV-1 cells. Neither AML3/CBF␣1 nor AML-ETO could reverse the inability of GR to activate the enhancer nor significantly influence GR-mediated activation of simple HREs (Fig. 5). This suggests that the failure of GR to activate CЈ⌬9 in CV-1 cells is not (solely) related to AML3/ CBF␣1. In T47D cells, where AML3/CBF␣1 is much less abun- dant and where both receptors can activate the enhancer, overexpression of AML3/CBF␣1 did not show notable or differential effects on enhancer activation by AR or GR (Fig. 6). This suggests that AML3/CBF␣1 is not an exclusive determinant of enhancer induction and cannot override factors in T47D cells that interact with either receptor.
Distinct AML/CBF␣ subtypes are able in vitro to transactivate the target genes of each other, such as osteocalcin (41) or T-cell receptor ␤ (33), with apparent specificity in vivo due to their largely nonoverlapping, cell-restricted expression patterns. To evaluate whether the function of AML3/CBF␣1 in AR-specific activation could be replaced by another family member, the related AML1/CBF␣2 (also termed PEBP2␣B (27)) was tested in transfection in CV-1 cells. Interestingly, overexpression of this factor decreased the AR-mediated activity of the enhancer by 50% and only rescued the AML1-ETOprovoked repression to that level (Fig. 7). That AML1/CBF␣2 has to compete with AML1-ETO for enhancer binding sites in order to rescue repression (see "Discussion) suggests a relatively stringent requirement of AML3/CBF␣1 for cooperation with AR. AML3/CBF␣1 has two major domains not present in AML1/CBF␣2 (24,51). Thus, the observed inhibition may be due not only to competition for DNA binding but also to differences between AML/CBF␣ family members in their peptide domains that interact with steroid receptors or other enhancerbound factors.

Direct Interaction between AML3/CBF␣1 and Steroid Receptors-
The observed functional interaction of AML3/CBF␣1 with AR or GR implied that this factor might also interact physically with the receptors. To test this, GST pull-down assays were performed with full-length receptors that were transcribed and translated in vitro (Fig. 8). Compared with GST alone, GST-AML3/CBF␣1 specifically retained both AR and GR. Although GST-AML1/CBF␣2 displayed similar retention of AR and GR, GST-AML3/CBF␣1 bound significantly more to AR than GR, suggesting that AML3/CBF␣1 interacted with AR preferentially to GR. Analysis of the autoradiogram of Fig. 8 by densitometric scanning using NIH Image version 1.61 software indicated that 10% of the input AR was retained by GST-AML3/CBF␣1 but only 2% of the input GR. This differential interaction may enter into the ability of AR, but not GR, to activate the enhancer in CV-1 cells. These data provide compelling evidence of direct interaction between AML/CBF␣ family members and steroid receptors and suggest that there is differential recognition of individual members within both transcription factor families. DISCUSSION This study reveals two factors interacting with the 5Ј part of the androgen-specific Slp enhancer and identifies one of them as the transcription factor AML3/CBF␣1. The functional requirement of AML3/CBF␣1 for AR-specific activation in CV-1 cells was demonstrated by the repression imposed by the dominant-negative fusion protein AML1-ETO and rescue from that repression by overexpression of AML3/CBF␣1. However, AML3/CBF␣1 is not restricted to interaction with AR since it also affected GR activation of the enhancer in T47D cells. The lack of stringent specificity is also shown by the direct physical interaction of AML3/CBF␣1 with both AR and GR, although interaction with AR was 5-fold stronger than with GR. These data demonstrate that nonreceptor factors play an indispensable role in tissue-dependent AR specificity, which may stem from preferential, rather than stringently specific, interactions.
The function of Isoforms II and III is critical for osteoblast differentiation, and deletion or mutation of the AML3/CBF␣1 gene leads to failure of bone ossification and/or cleidocranial dysplasia syndrome (42)(43)(44). However, the function of Isoform I is unclear, despite its ability to activate AML/CBF␣ elementcontaining enhancers similarly to Isoforms II and III (51,52). The present study suggests that the function of Isoform I, unlike the osteoblast-specific Isoforms II and III, may include cooperation with other factors for gene expression in nonosteogenic cells, similar to isoforms of AML1/CBF␣2, which cooperate with Ets (46) or C/EBP (45,53).
Functional requirement of AML3/CBF␣1 for AR activation of the enhancer is clearly demonstrated by its ability to rescue repression caused by the dominant-negative AML1-ETO pro-tein. Both AML1-ETO and AML3/CBF␣1 act specifically on the enhancer, since there is no significant effect of either on the activation of simple HREs by either AR or GR (Fig. 5). AML1-ETO, created by the t(8; 21) translocation, has the N-terminal DNA binding portion of AML1 fused to nearly all of ETO, a protein of largely unknown function (36). The fusion product retains ability to bind specific DNA sites and exerts active repression via ETO on all tested AML/CBF␣-dependent genes (27,(33)(34)(35)39). The mechanism underlying the repression has been recently shown to involve the interaction of AML1-ETO with the nuclear receptor corepressor (N-CoR), through the C-terminal zinc fingers of ETO (37)(38)(39). N-CoR is increasingly recognized as critical for repression of numerous targets via its associated histone deacetylase activity. Recruitment of N-CoR to the Slp enhancer by binding of AML1-ETO most likely accounts for the repression of androgen induction. When AML3/CBF␣1 is overexpressed, it competes with AML1-ETO for the two AML/CBF␣ sites within the enhancer, restoring AR-induced transcription.
Further evidence that AML3/CBF␣1 is stringently required for optimal AR induction is seen in interference with activation by AML1/CBF␣2. Exogenous AML1/CBF␣2 may compete with AML3/CBF␣1 and interact differently with the enhancer. The major differences between these related factors is the occurrence of two transactivation domains, one at the N terminus and one at the C terminus, in AML3/CBF␣1 that are not present in AML/CBF␣2 (33). The lack of these transactivation domains in AML1/CBF␣2 may contribute to its weaker participation in AR induction of the enhancer, resulting in the apparent inhibition when this factor is overexpressed relative to AML3/CBF␣1 in CV-1 cells. This also demonstrates that occupancy of the AML/CBF␣ sites by any family member due to the highly conserved DNA binding domain is not sufficient for optimal activation. Interactions with other domains and with other enhancer-bound factors, as well as AR, thus are implicated.
Both AML/CBF␣ binding sites within the enhancer are necessary for AR induction. Although the site proximal to the HRE would seem more likely to interact directly with receptor, the 5Ј site was found critical for induction in previous studies, where individual mutation of either site (LS4 or LS8A) abrogated hormonal response nearly equivalently (13). These mutations also abolish AML/CBF␣ interaction in gel shift assays (data not shown). The function of each site, however, may vary due to context-dependent interactions with distinct neighboring factors. This is supported by the observation that the 5Ј site (LS4) is more critical for uninduced enhancer expression than is the 3Ј site (LS8A) (17). The 3Ј site, sandwiched between the halfsite and consensus HREs, may be more dependent on receptor interactions.
AML3/CBF␣1 interaction, however, is not specific to AR but occurs also with GR, as seen by the rescue of GR induction in T47D cells (Fig. 6) and interaction with GR in vitro (Fig. 8). As shown previously (4,13), the inability of GR to activate the enhancer is cell-dependent and is a major component of AR specificity in CV-1 cells. GR can acquire activity when the position of the HRE is altered, suggesting that GR is repressed by proximity to other enhancer factors (13). Since one of the AML3/CBF␣1 sites is juxtaposed to the HRE, it is tempting to speculate that AML3/CBF␣1 may mediate the repression of GR. All AML/CBF␣ family members have a C-terminal amino acid motif, VWRPY, that represses transcription by interaction with homologs of the Drosophila protein Groucho (54). It may be that the interaction of AML3/CBF␣1 with GR alters its conformation to expose the VWRPY motif and recruit a Groucho homolog, such as TLE2 (55), in CV-1 cells, thereby repress-FIG. 8. AML3/CBF␣1 interacts with both AR and GR in vitro but with different strength. Immobilized GST and GST fusion proteins, GST-AML3/CBF␣1, and GST-AML1/CBF␣2, (12.5 l each in batch) were incubated for 2 h with equal radioactivity of the 35 S-labeled AR or GR, as described under "Materials and Methods." After extensive washing, the retained receptors were analyzed by 8% SDS-polyacrylamide gel electrophoresis. 5% of the amount of each receptor used for the incubations is shown. The result is representative of three independent experiments.
ing GR activity. In T47D cells, a repressive interaction may not be evident due to the low abundance of AML3/CBF␣1 and the presence of other enhancer factors that interact equivalently with AR and GR. The quantitative difference between GR and AR in physical interaction with AML3/CBF␣1 (Fig. 8) may reflect a conformational difference that allows AR to escape repression and instead interact synergistically with AML3/ CBF␣1. AML3/CBF␣1-mediated GR repression is not observable in the presence of AML1-ETO (Fig. 5), because corepressors are recruited. It would be intriguing to test whether a mutated AML3/CBF␣1 lacking the VWRPY motif would allow GR to activate the enhancer.
Nonreceptor factors that contribute to transcriptional specificity of steroid receptors have been noted in several genes. For example, GR-dependent activation of liver-specific tyrosine aminotransferase requires both Ets and HNF3 factors bound to elements adjacent to an HRE (56). The accessory factors by themselves cannot initiate efficient induction of their targets (10,11,57), suggesting that steroid receptors may act to organize the assembly of a transcriptionally active complex of enhancer-bound factors as well as recruited non-DNA-binding proteins. Thus, specificity for a particular receptor may vary with the arrangement of accessory factors within a complex. Inclusion or exclusion of a particular factor can produce significant changes in specificity (17,58). This further suggests that major determinants of specificity for steroid receptor action include the precise architecture of enhancer response elements and the array and ratio of accessory factors within target tissues, no single one of which need be stringently specific.