Structure-Function Analysis of the Estrogen Receptor α Corepressor Scaffold Attachment Factor-B1

Scaffold attachment factor-B1 (SAFB1) is a nuclear matrix protein that has been proposed to couple chromatin structure, transcription, and RNA processing. We have previously shown that SAFB1 can repress estrogen receptor (ERα)-mediated transactivation. Here we present a structure-function study showing that transactivation is mediated via an intrinsic and transferable C-terminal repression domain (RD). A similar C-terminal RD was found in the family member SAFB2. Removal of the RD from SAFB1 resulted in a dominant-negative SAFB1 protein that increased ligand-dependent and -independent ERα activity. SAFB1RD-mediated repression was partly blocked by histone deacetylase inhibitors; however, no histone deacetylase inhibitors were identified in a yeast two-hybrid screen using the RD as bait. Instead, SAFB1RD was found to interact with TAFII68, a member of the basal transcription machinery. We propose a model in which SAFB1 represses ERα activity via indirect association with histone deacetylation and interaction with the basal transcription machinery.

Scaffold attachment factor-B1 (SAFB1) is a nuclear matrix protein that has been proposed to couple chromatin structure, transcription, and RNA processing. We have previously shown that SAFB1 can repress estrogen receptor (ER␣)-mediated transactivation. Here we present a structure-function study showing that transactivation is mediated via an intrinsic and transferable Cterminal repression domain (RD). A similar C-terminal RD was found in the family member SAFB2. Removal of the RD from SAFB1 resulted in a dominant-negative SAFB1 protein that increased ligand-dependent and -independent ER␣ activity. SAFB1RD-mediated repression was partly blocked by histone deacetylase inhibitors; however, no histone deacetylase inhibitors were identified in a yeast two-hybrid screen using the RD as bait. Instead, SAFB1RD was found to interact with TAFII68, a member of the basal transcription machinery. We propose a model in which SAFB1 represses ER␣ activity via indirect association with histone deacetylation and interaction with the basal transcription machinery. SAFB1 1 belongs to a family of nuclear proteins that are localized in the nuclear matrix. The matrix is a proteinaceous structure consisting of a network of ribonucleoproteins and non-histone proteins, such as transcription factors, that serve as a scaffold for the higher organization of chromatin into loop structures (1). A characteristic of SAFB family members is the presence of an N-terminal SAF box (2)(3)(4) that binds to DNA regulatory regions termed scaffold/matrix attachment regions (S/MARs). S/MARs are bound to the nuclear matrix, define the loop structure of higher order chromatin (5)(6)(7)(8), divide the genome into structural and functional domains, and are impli-cated in the regulation of gene expression (1).
SAFB proteins contain a central RNA recognition motive (RRM), which suggests a role in mRNA processing. Because SAFB1 also has been shown to interact with members of the RNA processing machinery and with RNA polymerase II, it has been suggested that this protein is part of a "transcriptosome" complex, coupling chromatin structure to transcription and RNA processing (9 -12).
We have previously reported that SAFB1 plays an important role in breast cancer because its overexpression results in growth inhibition (13). SAFB1 maps to a chromosomal locus that displays unusually high rates of loss of heterozygosity (14), and mutations have been identified in breast tumors (14). We have also shown that SAFB1 can bind to and repress transcriptional activity of the estrogen receptor ␣ (ER␣), thereby functioning as an ER␣ corepressor (15).
ER␣ is a steroid receptor that regulates transcription of genes involved in proliferation, apoptosis, migration, and other cellular processes (16). ER␣ corepressors and coactivators are components of large protein complexes that tightly control the activity of ER␣ (17,18). Although the role of coactivators in the activity of ER␣ is well established, the role of ER␣ corepressors is less clear. Based on recent studies, it is believed that ER␣ corepressors play a role in: (i) modulating the estrogen response and thereby provide tissue specificity; (ii) conferring anti-estrogen-mediated repression of ER␣; (iii) mediating estrogen-induced repression of genes; and (iv) controlling the activity of ER␣ bound to DNA in the absence of ligand (for a recent review see Ref. 19). Emerging studies show that the repression of ER␣ activity is essential in preventing the cell from responding inappropriately to estradiol, and it is hypothesized that the failure of these control mechanisms is a characteristic of estrogen-responsive breast cancer (14,20,21).
A number of ER␣ corepressors contain domains that can repress the transcriptional activity of a reporter construct when they are transferred to heterologous proteins. These independent repression domains recruit proteins that modulate activity by remodeling local chromatin or by interacting directly with members of the basal transcription machinery. For example, SMRT, NCoR (22), and metastasis-associated protein 1 (MTA1) (23,24) recruit histone deacetylases (HDAC) or HDAC-containing protein complexes to target promoters. Additionally, NCoR can also function by locking the central components of the transcription initiation machinery (TFIIB and TATA-binding protein-associated factors) into a nonfunctional complex or confirmation that is not favorable for transcription (25). In contrast to repressors containing independent and transferable repression domains, the repressor of estrogen receptor activity functions by competing with coactivators for ER␣ binding sites (26), and the testicular orphan nuclear re-ceptor 2 prevents ER␣ homodimerization and its subsequent interaction with DNA (27). Other proteins such as repressor of tamoxifen transcriptional activity may modulate ER␣ by regulating facets of mRNA processing or stability (28).
Here we report that the ER␣ corepressor SAFB1 functions similarly to SMRT and NCoR in that it has an independent repression domain that maps to the C terminus, a characteristic also found in SAFB2, a protein that is very similar to SAFB1. Repression through this domain can be partially released with HDAC inhibitors. Detailed structure-function studies of SAFB1 showed that the repression domain is separable from its main ER␣ interaction domain and other highly conserved regions. Yeast two-hybrid studies revealed an interaction with TAFII68, a member of the basal transcription machinery. Based on these and other previously published data, we suggest that SAFB1 provides a scaffold for recruiting members of both the chromatin remodeling and the basal transcription machinery to estrogen-regulated promoters.
Immunoblotting-Immunoblotting was performed as described previously by us (4). SAFB and Gal4-DNA-binding domain (DBD) antibodies were purchased from Upstate Biotechnology (Waltham, MA) and used at dilutions of 1:1000. The generation of the TAFII68 antibody (kindly provided by Dr. L. Tora, INSERM, Illkirch, France) has been described recently (29).
Transient Transfection and Luciferase Assays-For reporter assays to measure transcriptional repression, cells were plated at a density of 6 -8 ϫ 10 4 cells/well in 6-well plates. Twenty-four hours later, cells were cotransfected with 1 g of a luciferase reporter containing 17ϫ4-Gal4-UAS, 25 ng of ␤-galactosidase, and Gal4 fusion constructs (see Figs. 4 -8 legends). Cells were transfected using FuGENE 6 reagent (Roche Applied Science) for 8 -10 h in 1 ml of complete IMEM. The medium was then removed, and 3 ml/well fresh medium was added. Twenty-four hours later, cells were washed twice with 4°C phosphate-buffered saline and lysed, and luciferase activity was measured using the luciferase assay kit from Promega (Madison, WI). As described previously (3), luciferase activity was normalized to ␤-galactosidase activity to give relative luciferase units.
In preparation for assays that measure the ability of proteins to repress ER activity, MCF-7 cells were plated in IMEM without phenol red at a density of 1.5 ϫ 10 5 cells/well of a 6-well plate. Before transfection, cells were placed in 1 ml of serum-free medium. Cotransfections were performed 24 h later using FuGENE 6 reagent with 1 g of ERE-TK-luciferase reporter construct, 25 ng of ␤-galactosidase, 50 ng of ER␣ expression plasmid, and expression plasmids for other proteins as indicated in the figure legends. Ten hours after transfection, the medium was replaced with 3 ml of fresh serum-free medium. Fourteen hours later, estradiol (10 -8 M) was added to the appropriate wells, and 24 h later, cells were lysed, and the number of relative luciferase units was determined as described above.
To demonstrate the expression of Gal4 fusion proteins, 5 ϫ 10 5 CV-1 or 1 ϫ 10 6 MCF-7 cells were plated in 6-cm dishes. Twenty-four hours later, cells were transfected in complete IMEM with 1-2 g of expression construct using FuGENE 6. Eight to ten hours after transfection, 3 ml of fresh medium was added to the plates. Cells were lysed 24 h later in 5% SDS, and immunoblotting was performed.
Plasmid Constructs-GST fusion constructs were produced using either pGEX-2TK (Amersham Biosciences) or the universal cloning system of Elledge and co-workers (30). SAFB1 fragments in pGEX-2TK were generated by 25-35 cycles of PCR using platinum Taq Pfx DNA polymerase (Invitrogen). The 50-l reactions contained 50 pmol of primer, 1ϫ enhancer solution, 1.5-3 mM MgCl 2 , and 200 M dNTPs in a PCR reaction buffer (Invitrogen). Annealing temperatures and extension times for each primer pair were determined for use on either an MJ Research PTC 200 thermal cycler (MJ Research Inc., Waltham, MA) or a PerkinElmer 9600 (PerkinElmer Life Sciences). PCR products and vector were digested with BamHI and EcoRI (pGEX2T) or BamHI and NdeI (pUNI10) and gel-purified using the Qiagen Gel Extraction Kit (Qiagen Inc., Valencia, CA).
Wild-type SAFB1 was subcloned from pCDNA1.1 into the EcoRI site of pUNI10. To produce the SAFB1⌬ domain mutants, SAFB1 fragments were PCR-amplified using platinum Taq Pfx as described above and cloned first into pBluescript KS(ϩ) (Stratagene). The DNA for the mutant protein was then cloned into pUNI10. GST fusion constructs were produced using Cre recombinase reactions with pHB2-GST as the recipient host vector. The 20-l reactions contained 100 -200 ng of construct in pUNI10, 200 ng of pHB2-GST, and a range of 0.25 to 1 unit of Cre recombinase (Invitrogen). Reactions were allowed to proceed for 20 min at 37°C and were then inactivated by heat at 65°C for 5 min. 2-5 l were transformed into BW2374 (30) or Pir1 (Invitrogen) cells. Restriction digests and their ability to generate GST fusion proteins of the appropriate size confirmed all constructs.
Gal4-DBD fusion constructs were generated using the pCMX-Gal4N expression vector. To clone full-length SAFB1 into pCMX-Gal4N, SAFB1 cloned into the EcoRI site of pCDNA1.1 was digested with BamHI and SapI to remove the first 675 5Ј bp. A PCR fragment was generated using the primers 5Ј-GGA AGG ATC CGA ATT CAT GGC GGA GAC TCT GTC AGG-3Ј and 5Ј-GGA AGG AAC ATA TGT TTG CTG AAA AGA TTC TTC AAA TCT GTA CTG G-3Ј and digested with BamHI and SapI. The fragment was then used to replace the 5Ј 675 bp of SAFB1 in pCDNA1.1. Full-length SAFB1 was then excised from pcDNA1.1 using EcoRI and ligated in-frame into the EcoRI site of the pCMX-Gal4N vector. Other Gal4 fusion constructs in pCMX-Gal4N were generated by PCR with appropriate primers and 5Ј restriction sites or by restriction digests of full-length SAFB1 to produce SAFB1 fragments that were then cloned in-frame to pCMX-Gal4N.
Analysis of Protein-Protein Interactions-Proteins translated in vitro were produced using the TNT rapid in vitro translation kit (Promega) and used in GST pull-down experiments as described previously (4). For co-immunoprecipitation, MCF-7 cells were lysed in a high stringency buffer, and 500 g of protein were precleared with protein G-agarose and incubated with 4 g of the appropriate antibody, as described previously by us (4). For the yeast two-hybrid interactions, the Cterminal SAFB1 fragment (aa 599 -915) was cloned into pGBK-T7 vector (BD Biosciences). Yeast two-hybrid assays using a normal breast tissue cDNA library were undertaken using the Matchmaker3 system (BD Biosciences) according to the manufacturer's instructions and as described previously (4).
In Vitro RNA-binding Assay-A modified version of an in vitro RNAbinding assay (31) was used to identify RNA molecules bound by the RRM of SAFB1. SAFB1 fragments containing the RRM (GST-RRM, amino acids 354 -538) and ⌬RRM (GST-⌬RRM, amino acids 354 -428 and 483-538) were cloned into the host vector, pUNI10, using PCR. A Cre recombinase reaction with the vector pHB2-GST produced the GST fusion protein constructs. All steps were performed at 4°C, unless otherwise stated. Total RNA was isolated from MCF-7 cells using the RNaeasy miniprep kit from Qiagen. Bacterially expressed fusion proteins were produced in BL21-Gold cells as described above and bound to 200 l of GST-agarose beads. The beads were washed in 10-ml volumes of phosphate-buffered saline. The GST-Sepharose beads with bound GST fusion proteins were resuspended in 500 l of phosphate-buffered saline with 1 unit/l RNasin (Promega) and 100 g of Escherichia coli tRNA. The beads were blocked with the tRNA by mixing for 10 min followed by the addition of 60 g of MCF-7 RNA. The RNA was bound for 30 min at room temperature, and the beads were washed by centrifugation at 1000 ϫ g for 1 min. The beads were then resuspended in 10 ml of phosphate-buffered saline and washed twice more with vortexing. The RNA was extracted using TRI reagent from Sigma. Extracted RNA was precipitated and resuspended in 20 l of RNA-free water for the reverse transcription reaction, which was performed using 2-4 l of purified RNA and the SMART RACE cDNA amplification kit from Clontech. A modified primer, 5Ј-TAA GTA GCG GCG GGT GAA GTG T(20) AGC AGCT-3Ј, was used for the reverse transcriptase reaction. PCR amplification of the cDNA was performed using 1 unit of platinum Taq (Invitrogen) and 2-4 l of the reverse transcription reaction with the primer 5Ј-GGA AGG ATC CTA AGT AGC GGC GGG TGA AGT G-3Ј and a modified universal primer mix short primer, 5Ј-GGA AGA ATT CCT AAT ACG ACT CAC TAT AGG GC-3Ј. The resulting reverse transcriptase PCR products were cloned into the BamHI/EcoRI sites of pBluescript KS(ϩ) (Stratagene) and then sequenced.

SAFB1 Contains Two Domains That Interact with Estrogen
Receptor ␣-The SAFB1 protein contains a number of conserved domains (indicated in Fig. 1, top panel) that may function as protein-protein interaction domains, and we asked, first, if any of these motifs were important for the SAFB1-ER␣ interaction. A series of GST-SAFB1 fusion proteins containing a deletion in one or another of these motifs were constructed and then used to assay in vitro protein-protein interactions with in vitro translated ER␣. All of the SAFB1 mutants bound ER␣ (Fig. 1, bottom panel), suggesting that none of these domains were required for ER␣ binding.
Next, we generated a number of SAFB1 GST deletion constructs as depicted in Fig. 2A, left panels, and confirmed the correct protein expression ( Fig. 2A, right panels). Performing in vitro protein-protein interactions with the N-terminal (aa 1-260), central (aa 260 -600), and C-terminal (aa 600 -915) regions, we identified the central region of SAFB1 as the major ER␣ interaction domain (Fig. 2B, EID). We repeatedly observed weak binding of ER␣ to the SAFB1 C terminus (aa 600 -915) (Fig. 2B, EID2) that could not be conclusively defined using smaller GST fusion proteins because of the relatively weak binding. To further refine EID1, a series of iterations using consecutively smaller fusion proteins from aa 240 -600 were generated (Fig 2A). The results from the GST pull-down assays (Fig. 2C) suggest that the SAFB1 region encompassing aa 426 -600 contains at least three interaction domains, which may bind in a cooperative manner. The existence of multiple interaction domains may explain how a mutant missing the central RRM is still able to bind ER␣ (Fig. 1). The EIDs in SAFB1 are illustrated in Fig. 2D.
SAFB1 Binds RNA but This Function Is Not Required for Corepressor Activity-A number of nuclear receptor cofactors contain RNA recognition motifs, and in some cases (e.g. repressor of tamoxifen transcriptional activity (see Ref. 28)) these domains have been shown to be essential for corepressor activity. SAFB1 also has a central RRM, and we asked, first, if this RRM could in fact bind RNA. Based on secondary structure predictions and molecular modeling of the RRM in SAFB1, we chose to remove amino acids 429 -483, which constitute ␤2␤3␣2␤4 of the canonical RRM ␤1␣1␤2␤3␣2␤4 motif, to produce a ⌬RRM-SAFB1 construct. GST-SAFB1 (aa 354 -538) and GST-SAFB1-⌬RRM (aa 354 -428 and 483-538) were expressed in bacteria, bound to GST-agarose beads, and combined with total RNA isolated from MCF-7 cells. The beads were exten-sively washed, and the RNA was purified to make cDNA, which was then amplified by PCR and separated on agarose gels. The presence of PCR product in the GST-SAFB1-RRM lane indicated that this domain is able to bind RNA in vitro, whereas the GST-SAFB1-⌬RRM could not (Fig. 3A).
Second, we asked whether RNA binding and, therefore, the RRM domain were essential for SAFB1 corepressor activity as measured in the transient transfections assays. Full-length SAFB1 (aa 1-915) or SAFB1-⌬RRM (aa 1-428 and 483-915) were transiently transfected into MCF-7 cells along with an ER␣-responsive reporter construct, ERE-TK-Luc. Both SAFB1 and SAFB1-⌬RRM repressed ER␣ activity, indicating that RNA binding is not an essential component for the repression of ER␣ activity (Fig. 3B), at least as measured in these assays. Further experiments included a series of ERE reporter assays in which we cotransfected expression plasmids that encoded SAFB1-interacting proteins, which function in mRNA splicing and stability, such as heterogeneous nuclear ribonucleopro-teinA1, tra2, and AUF1/heterogeneous nuclear ribonucleopro-teinD, along with SAFB1. These experiments, however, failed to show any synergistic effects (data not shown), suggesting that there might be a delineation between SAFB1 activities in RNA processing and transcriptional regulation.
SAFB1 Has an ER␣-independent Transferable Repression Function When Tethered to DNA-Next, we tested whether SAFB1 has an intrinsic repression activity that is independent of an interaction with ER␣ as demonstrated for other corepressors such as SMRT (32). Conventional repression assays using Gal4-DBD fusion proteins and a Gal4-responsive reporter construct were performed with CV-1 cells, and the N-terminal repression domain (RD) of SMRT (SMRTRD) (aa 1-1230) was used as a positive control (22). The Gal4-DBD-SMRTRD and -SAFB1 fusion proteins were expressed at comparable levels ( Fig. 4, inset). Like SMRTRD, full-length SAFB1 repressed the reporter activity in a dose-dependent manner (Fig. 4). Transfection of SAFB1 that was not fused to Gal4-DBD (pcDNA1-SAFB1) failed to reduce the basal reporter activity (data not shown) verifying that SAFB1 must be bound to DNA to repress transcription. Additionally, we tested an SAFB1⌬RRM-Gal4-DBD mutant using this assay, and, as expected, we did not see a loss of repression (data not shown). The Gal4-DBD-SAFB1 repression assay was also performed with NIH3T3, MCF-7, HEK 293, Saos-2E, and HeLa cells, and, although we detected minor differences in the level of repression, SAFB1 was able to repress transcription in those cell lines (data not shown).
The Repression Domain Maps to the C-terminal Region in SAFB1-Having shown that SAFB1 harbors intrinsic repression activity, we next mapped the region in SAFB1 responsible for this activity. We generated a series of truncated Gal4-DBD-SAFB1 expression constructs (Fig 5A, top panel) that, when expressed in CV-1 cells, led to the expression of peptides of the expected sizes (Fig. 5A, bottom panel) and were expressed at levels comparable with SMRTRD. Transient transfections using the Gal4-DBD-SAFB1 deletions showed that the repression domain resides in the C terminus (aa 599 -915) (Fig. 5B). The repressive activity of this C-terminal region is comparable with that of SMRT (Fig. 5B, compare SMRTRD and aa 599 -915).
To investigate whether the SAFB1 C terminus contained more than one intrinsic repression domain and whether we could finely map the repression domain(s), a series of overlapping Gal4-DBD fusion proteins from the C-terminal region were constructed (Fig. 5C, top panel), and correct protein expression was confirmed (Fig. 5C, bottom panel). The data in Fig. 5D indicate that all deletion fragments were able to repress, at least to some extent, and that there may be multiple elements in this region that are able to direct repression. We previously demonstrated that HDAC inhibitors can reduce SAFB1 corepressor activity on ER␣ (33), and here we investigated whether HDAC inhibitors would affect SAFB1-mediated repression in an ER␣-independent assay. As shown in Fig. 6, the addition of TSA led to a release of repression of both SAFB1RD and SMRTRD; however, TSA did not fully reverse SAFB1-mediated repression of the reporter. Similar results were obtained with sodium butyrate, the HDAC inhibitor (data not shown), suggesting that SAFB1, like other ER␣ corepressors (34, 35), might have multiple mechanisms of repression, including HDAC- We previously demonstrated that SAFB2, a protein with 74% homology to SAFB1, also interacts with ER␣ and functions as an ER␣ corepressor. The C termini of SAFB1 and SAFB2 are highly conserved, and we asked whether the SAFB2 C-terminal region also harbors an independent repression domain. We generated an SAFB2 C-terminal (aa 600 -953) Gal4-DBD expression construct, and, using transient transfection assays, we were able to show that, like SAFB1, SAFB2 also has intrinsic repression activity (data not shown). These data suggest that both SAFB family members function via a common mechanism to repress nuclear ER activity.
The SAFB1 C-terminal Repression Domain Is Essential for Corepressor Activity-Having identified the C terminus as the location of the repression domain, we asked whether an SAFB1 expression construct that contained the main ER␣-interacting domain but lacked the repression domain would fail to repress ER␣ as predicted from our data. Therefore, MCF-7 cells were transiently transfected with ERE-TK-Luc reporter, full-length SAFB1, and SAFB1⌬RD (aa 1-600) expression constructs. As shown in Fig. 7, the SAFB1⌬RD mutant failed to repress ER␣, indicating that the RD is essential for the ER␣ corepressor activity of SAFB1. Interestingly, expression of SAFB1⌬RD resulted in increased ER␣ activity in the presence and absence of ligand, suggesting a role for SAFB1 in repressing ER␣, even in the absence of ligand. We repeated these experiments using an estrogen-responsive reporter system that lacks the TK promoter and instead harbors a minimal TATA box (ERE-TATA-Luc). Using this reporter system in which transcription depends only on ER␣ and the basal transcription machinery, we again detected an increase in ER␣ activity after transfection of SAFB1⌬RD (data not shown).
These data suggested either that SAFB1⌬RD functions as a classical dominant-negative protein (⌬RD is able to bind ER␣ as shown in the GST pulldown in Fig. 7) or that the first 600 amino acids of SAFB1 contain inherent transactivation activity. The latter could be excluded because we failed to detect any measurable transactivation activity of SAFB1⌬RD (data not shown) (see aa 1-600 in Fig. 5B). We therefore concluded that SAFB1⌬RD functions as a dominant-negative protein, which results in increased ER␣ activity in the presence and in the absence of ligand.
SAFB1 Repression Domain Interacts with the TATA Element-binding Protein-associated Factor TAFII68 -To identify proteins that interact with the SAFB1 repression domain, we screened a normal breast cDNA library by yeast two-hybrid assay using the C terminus of SAFB1 as bait. We identified a number of previously known SAFB1-interacting proteins, including SF2/ASF (SFRS1), SRp30c (SFRS9), and Sam68, confirming the validity of our screen. To our surprise, we did not identify any HDACs, suggesting that there might be an indirect interaction between SAFB1 and HDAC proteins.
One of the intriguing interacting proteins was TAFII68 (also known as TAF15, RBP56, and TAF2N), a protein that that has been shown to be associated with both TFIID and RNA polymerase II. Using directed yeast two-hybrid assays, we were able to show that this interaction was specific for the C-terminal end of the repression domain (Fig. 8A, aa 720-915). The interaction between SAFB1 and TAFII68 was confirmed in GST pull-down assays using in vitro translated TAFII68 and the SAFB1 C terminus fused to GST (Fig. 8B). We also were able to detect an interaction of endogenous SAFB and TAFII68 proteins in co-immunoprecipitation as shown in Fig. 8C. Finally, we generated a Gal4-DBD-SAFB1 construct in which the interaction domain was deleted (Fig. 8D, aa ⌬720 -915), which was used in repression assays. As shown in Fig. 8D, deletion of the TAFII68 interaction domain resulted in the loss of repression. This result suggests that, within the context of the entire protein, the interaction with TAFII68 is necessary for the repressive activity of SAFB1.

DISCUSSION
The transcriptional activator ER␣ is regulated by a number of coactivators and corepressors. The studies presented here deal with the further characterization of SAFB1, a recently identified ER␣ corepressor. Here we have shown that SAFB1 harbors an intrinsic RD that is indispensable for its corepressor activity and is separate from its main ER␣-interacting domain. An SAFB1 mutant, which lacks the RD, functions as a dominant negative in both the presence and absence of ligand. The RD interacts with TAFII68, which is a member of the basal transcription machinery.
To further characterize SAFB1 as an ER␣ corepressor, we first analyzed the EID. We determined that the major EID in SAFB1 maps to aa 426 -600, and this domain can be further divided into smaller but weaker interaction domains. This region does not contain any previously described ER interaction motifs such as the "CoRNR box." The CoRNR box, which contains the (L/I)XX(I/V)I motif, can be found in NCoR and SMRT (36). It is related to the LXXLL motif ("NR box") commonly found in the EID in ER␣ coactivators (37,38) as well as in a few corepressors, including LCoR (34) and RIP140 (39). Other corepressors such as repressor of estrogen receptor activity (26) harbor novel ER␣ interaction domains. Likewise, SAFB1 mediates ER␣ interaction via a novel domain, and therefore our results confirm the existence of additional binding motifs, other than NR and CoRNR boxes, used by corepressors.
SAFB1 is a large protein containing a number of functional domains. Like many other proteins, such as polypyrimidine tract-binding protein-associated splicing factor (PSF/p100) (40) and heterogeneous nuclear ribonucleoprotein units/SAF-A (41)(42)(43)(44)(45), SAFB1 is involved in transcription, mRNA stability and/or processing, and chromatin regulation. SAFB1 has been shown previously to bind DNA (2), and here we show that it is also able to bind RNA. Although there is evidence that RNA binding can be important for ER␣ corepressor activity (28), a SAFB1 mutant deficient in RNA binding was able to repress ER␣ activity, comparable with wild-type SAFB1. Although additional relevant assays are needed to draw a final, solid conclusion, these data suggest that the role of SAFB1 in transcriptional repression might be separate from its role in mRNA processing and/or stability. A similar separation of functions has been demonstrated previously for Sam68 (46). This separate role in mRNA processing and other undiscovered activities We have shown previously that SAFB1 interacts with the DBD/Hinge region in ER␣ but that SAFB1-mediated repression is not a result of inhibition of the DNA binding of ER␣ (15). We have also performed a number of experiments that exclude simple competition with coactivators. 2 Consistent with these studies, we show here that SAFB1 has an independent and transferable repression domain similar to that described for the ER␣ corepressors NCoR and SMRT (47,48). Removal of this repression domain resulted in a dominant-negative SAFB1 protein that enhanced ER␣-driven transcription. Intriguingly, SAFB1⌬RD functioned as a dominant negative not only in the presence but also in the absence of ligand, suggesting a role for SAFB1 in repression of ligand-independent ER␣. Similarly, mouse embryo fibroblasts that are deficient in BRCA1 (BRCA1 Ϫ/Ϫ ), which functions as an ER␣ corepressor (49), showed increased ligand-independent ER␣ activity when compared with wild-type mouse embryo fibroblasts (50).
The repression domains in NCoR and SMRT recruit different HDAC-containing complexes (22). Because the SAFB1RD-mediated repression was partially released by treatment with HDAC inhibitors, we expected that a yeast two-hybrid assay using SAFB1RD as bait would reveal an interaction with known HDACs or members of the HDAC complexes. Because we failed to detect such interaction, we concluded that there is 2 S. M. Townson, unpublished data.
FIG. 5. The SAFB1RD maps to the C terminus. A and C, top, schematic presentation of SAFB1 Gal4-DBD fusion proteins used for RD mapping. Bottom, CV-1 cells were transfected with Gal4 fusion constructs and SDS lysates were prepared and immunoblotted using anti-Gal4-DBD antibodies. The arrows in A indicate the bands corresponding to the Gal4-DBD-SAFB1 fusion proteins. B, the repression activity in SAFB1 resides in its C terminus. CV-1 cells were transfected with 1 g of 17ϫ4-UAS-TK-Luc, 25 ng of ␤-galactosidase, and 250 ng of Gal4-DBD expression plasmids as indicated. In the negative control wells (Ϫve), only UAS-TK-Luc and ␤-galactosidase were transfected. The repression domain is delineated in bold (599 -915). D, the SAFB1 C terminus harbors one independent RD that can not be further delineated into independent repression domains. CV-1 cells were transiently transfected with 1 g of 17ϫ4-UAS-TK-Luc, 25 ng of ␤-galactosidase, and 250 ng of Gal4-DBD expression plasmids as indicated. NLS, nuclear localization signal; RLU, relative luciferase units. an indirect interaction between SAFB1 and HDACs. Supporting this conclusion is a recent finding by Tai et al. (51) who discovered an interaction between SAFB and the chromodomain protein CHD1. CHD1 has been proposed to play a role in chromatin architecture and transcriptional regulation through its interactions with HDACs and NCoR. Like SAFB proteins, CHD1 binds to stretches of AT-rich DNA in so-called S/MARs. We therefore suggest that SAFB1-mediated repression is in part mediated through an indirect interaction with HDACs via binding to CHD1. Ongoing experiments in our laboratory will decipher the role of S/MARs in this repression.
The yeast two-hybrid screen identified TAFII68 as a SAFB1RD-interacting protein. TAFII68 belongs to the TET (TLS/FUS, EWSR1, and TAFII68/TAF15/RBP56) family of pro-teins that are best known through their role in transformation (52). Following chromosomal translocation, the N termini of TET proteins are fused to a variety of C-terminal DNA binding domains of transcription factors (29), resulting in dominant oncogenes, presumably by functioning as transactivators (reviewed in Ref. 54). The TET family member TAFII68 was shown to be associated with both TFIID and RNA polymerase II (29,55), and we therefore suggest that the interaction between SAFB1 and TAFII68 bridges the SAFB1-ER␣ complex to the basal transcription machinery. Previous studies have shown that SAFB1 can also directly interact with RNA polymerase II (10). Deletion of the TAFII68 interaction domain in SAFB1 resulted in the loss of SAFB1RD-mediated repression, suggesting that this interaction is necessary for the effect of SAFB1. Interestingly, the GST pull-down experiments indicated that the C-terminal domain has multiple repression domains that, when used independently, can all result in some degree of repression (see Fig. 5D). It is feasible, however, that within the context of the entire protein, the small fragments can not recruit other repressors efficiently enough to transfer repression, and thus deletion of the TAFII68 interaction domain results in a loss of repression.
In summary, we propose a model whereby SAFB1 serves as a scaffold to provide a structural framework for chromatin and the transcriptional machinery that allows for regulation of ER␣ at several steps during the transcriptional process. In this model, the SAFB1 complex defines a repressed chromatin structure through the indirect recruitment of HDACs, possibly via CHD1, and/or the inhibition of RNA polymerase II, either directly or through interaction with TAFII68. Ongoing studies in our laboratory that test this model will enable us to further understand how SAFB1 modifies the structure of chromatin for the regulation of ER␣ activity and will show how these mechanisms are disrupted in the development of breast cancer. The inset shows the result from a GST pull-down assay confirming that SAFB1⌬RD can indeed bind to ER␣. RLU, relative luciferase units.