Genome-wide Impact of the BRG1 SWI/SNF Chromatin Remodeler on the Transforming Growth Factor β Transcriptional Program*

The transcription factors Smad2 and Smad3 mediate a large set of gene responses induced by the cytokine transforming growth factor β (TGFβ), but the extent to which their function depends on chromatin remodeling remains to be defined. We observed interactions between these two Smads and BRG1, BAF250b, BAF170, and BAF155, which are core components of the SWI/SNF chromatin-remodeling complex. Smad2 and Smad3 have similar affinity for these components in vitro, and their interactions are primarily mediated by BRG1. In vivo, however, BRG1 predominantly interacts with Smad3, and this interaction is enhanced by TGFβ stimulation. Our results suggest that BRG1 is incorporated into transcriptional complexes that are formed by activated Smads in the nucleus, on target promoters. Using BRG1-deficient cell systems, we defined the BRG1 dependence of the TGFβ transcriptional program genome-wide. Most TGFβ gene responses in human epithelial cells are dependent on BRG1 function. Remarkably, BRG1 is not required for the TGFβ-mediated induction of SMAD7 and SNON, which encode key mediators of negative feedback in this pathway. Our results provide a genome-wide scope of the participation of BRG1 in TGFβ action and suggest a widespread yet differential involvement of BRG1 SWI/SNF remodeler in the transcriptional response of many genes to this cytokine.

Genomic DNA is packed into chromatin units or nucleosomes whose accessibility determines the ability of particular genes to respond to regulatory inputs (1)(2)(3). Various classes of ATP-dependent chromatin remodeling factors (remodelers) have been identified that control the accessibility of DNA. Remodelers act on histone-DNA contacts to induce nucleosome sliding, DNA looping, or complete dissociation of histones from DNA (4 -9). Although the role of remodelers in developmental stages and during the cell cycle has been characterized in some detail, little is known about their role in the genome-wide response to extracellular cytokine signals. The recent identification of interactions between Smad transcription factors and a SWI/SNF remodeler complex (10,11) provides an opportunity to investigate this question. SW1/SNF refers to a set of remodeler complexes consisting of up to 12 evolutionarily conserved protein subunits (12)(13)(14)(15). Mammalian SWI/SNF complexes are nucleated by either BRG1 or BRM, which are mutually exclusive ATPase subunits that recognize acetylated lysines in histone tails. Other SWI/SNF core subunits include either BAF250 or BAF180 and the shared subunits BAF170, BAF155, BAF60, BAF57, BAF53, actin, and SNF5/INI1 (12,16). SWI/SNF complexes have been shown to function in association with developmental factors such as MyoD, NeuroD, and GATA-1; cell cycle checkpoint components such as p53, pRB and BRCA1; and nuclear receptors for androgen, estrogen, or glucocorticoid (17)(18)(19)(20). Brm-null mice are viable and show only a subtle phenotype (21), whereas brg1null mice die early during embryogenesis, and hemizygotes are tumor-prone (22). These phenotypes imply important roles for the BRG1 SWI/SNF complex in development and cancer.
Smad transcription factors are central mediators of many actions of transforming growth factor ␤ (TGF␤) 2 and other members of this morphogen and cytokine family (23). TGF␤ transduces signals from the plasma membrane by interacting with two receptor subunits or TGF␤ receptors type I and type II, which are serine/threonine kinases. On ligand-induced phosphorylation and activation by the type II receptor, the type I receptor phosphorylates Smad2 and Smad3 at C-terminal serine residues, inducing these proteins to accumulate in the nucleus to regulate gene expression (24 -26). Activated Smad factors target different genes by forming complexes with other transcription factors. The DNA binding activity of each particular Smad cofactor combination directs such complex to a particular subset of target genes. Specific histone-modifying enzymes such as p300/CBP and others are recruited to these complexes for either transcriptional activation or repression (23,24). Up to several hundred genes are directly regulated in this manner by TGF␤ in any given cell type. The choice of target genes and the ultimate biological consequences of TGF␤ stimulation depend on the repertoire of Smad transcriptional partners that are present in the cell, as a function of the context and the cell type (27,28). This signaling system controls many aspects of cell division and behavior and is essential for embryonic developmental and tissue homeostasis, and its malfunctions underly many inherited and somatic disorders from congenital malformations to fibrosis, inflammation, and cancer.
The Smad pathway is subject to an intricate set of functional and regulatory inputs through diverse protein-protein interactions. In a recent search for Smad-binding proteins, we noticed an interaction between Smad2 and several SWI/SNF components (10). In vitro this interaction was independent of the activation state of Smad2, sharing this property with the interaction of Smad2 with the histone acetyl-transferase CBP (10). An interaction between Smad2 and BRG1 in cell culture and a requirement for BRG1 in the activation of two TGF␤ target genes (lefty and nodal) were recently reported by others (11). These observations suggested that Smad proteins may bring to DNA not only histone acetylating enzymes to relax the chromatin structure but also SWI/SNF complexes to provide chromatin remodeling activity.
The identification of protein interactions between Smad2 and SWI/SNF provided an opportunity to investigate the role of this chromatin remodeling complex in the cellular response to a cytokine. The present work is focused on this question at a genome-wide scale to define the requirement for BRG1 in TGF␤ gene responses in human epithelial cells. The results provide a global view of the dependence of the TGF␤ transcriptional program on BRG1 SWI/SNF function.
Transfection, Immunoprecipitation, and Immunoblotting-Transient transfections for coimmunoprecipitation experiments were carried out using Lipofectamine (Invitrogen), according to the manufacturer's instructions. The cells were harvested 36 h post-transfection and were lysed by sonication in HNG lysis buffer (0.2% Nonidet P-40, 25 mM HEPES, pH 7.9, 100 mM NaCl, 10% glycerol, 5 mM MgCl 2 , 10 mM NaF, 1 mM sodium orthovanadate, 20 mM ␤-glycerophosphate, 1 tablet of protease inhibitor (Roche Applied Sciences)/10 ml) at 4°C. For the detection of endogenous protein complexes, one 15-cm plate of cells was treated with or without 100 pM TGF␤ for 2 h and then lysed by sonication in 1 ml of HNG lysis buffer. The lysates were precleared with normal mouse or rabbit IgG (Upstate Biotechnology Inc.). Cleared supernatants were incubated with the indicated antibodies and protein A beads (Zymed Laboratories Inc.) for 4 h at 4°C. Normal rabbit IgG or mouse IgG were used as negative controls. Immunoprecipitates were resolved by SDS-PAGE gels and were electroblotted onto Immunobilon-P membrane (Millipore). Immunoblot analysis was carried out as described before (30).
Retroviral Infection-H29 packaging cells or amphotrophic Phoenix packaging cells were transfected with pRetrosuper or pQCXIP or pQCXIH (BD Bioscience) retroviral vectors using Lipofectamine 2000 (H29) or Lipofectamine (Phoenix). Retroviral supernatants were collected 48 h post-transfection, passed through a 0.45-mm filter, concentrated at 18000 rpm for 2 h, and used for overnight infection of cells in the presence of 5 g/ml of polybrene. The cells were allowed to recover for 24 h in fresh medium and selected with 5 g/ml puromycin for 48 h or 300 g/ml hygromycin B for 72 h.
Microarray Analysis-RNA was collected in duplicate and labeled as described previously (27). Each sample was hybridized to an Affymetrix human genome U133A plus 2.0 microarray at the Memorial Sloan-Kettering Cancer Center Genomics Core Laboratory. The hybridization images were preprocessed by the Robust multiarray analysis (33), provided by the Affy package of R programming software. Differentially expressed genes were identified using a Bayes model Implemented by the limma package of R (34).
Quantificative Reverse Transcription-PCR Analysis-qRT-PCR was carried out as described previously (27). Primer sequences are available upon request.

Smad2 and Smad3 Bind a BRG1 SWI/SNF Remodeler
Complex-We used affinity purification followed by mass spectrometry to identify proteins that bind to Smad2 and Smad3. Recombinant glutathione S-transferase fusion proteins con-taining the transcriptional regulatory regions (linker region and MH2 domain, LϩMH2) of Smad2 or Smad3 were bound to glutathione S-transferase beads and incubated with human HeLa cell extracts. Coomassie staining of electrophoresis gels containing the eluates from these beads reproducibly revealed a similar pattern of proteins using the Smad2 and Smad3 baits (Fig. 1A). Mass spectrometry analysis of the excised bands identified these proteins as the SWI/SNF complex subunits BRG1, BAF155, BAF170, and BAF250b (also known as OSA2) in both eluates (Fig. 1A). In addition, both eluates contained the histone acetyl transferase CBP, the ErbB2-interacting protein Erbin, and the transcription factors NCoA3 and TIF1␥/TRIM33. All of these proteins were previously observed in affinity purification of Smad2-binding proteins (10). The interactions of Smad2/3 with CBP, Erbin, and TIF1␥ have been characterized (10,23,35). The Smad3 eluate additionally contained NCoR1 and NCoR2/SMRT, which were previously identified as nuclear receptor corepressors (36,37) (Fig. 1A). Of these proteins, only TIF1␥ was increased in abundance when the experiments were repeated using activated (C-terminal acidic mutant) versions of Smad as baits (data not shown and Ref. 10). These results demonstrate a similar, intrinsic affinity of Smad2 and Smad3 for these SWI/SNF components and additionally identify NCoA3 and NCoR as potentially important partners of Smad transcriptional complexes.
Differential Interaction of BRG1 with Smad2 and Smad3 in Vivo-Focusing on SWI/SNF, we verified its interaction with Smad2/3 in vivo by means of coimmunoprecipitation of the endogenous proteins from HaCaT human keratinocytes (Fig.  1B). We chose this cell line because it is a well established model for the analysis of TGF␤ action. The interaction between endogenous Smad2/3 and BRG1 in the cells was increased by TGF␤ stimulation. In response to TGF␤, Smad proteins accumulate in the nucleus (25), which is where SWI/SNF components reside (4,6,8,9,12). Similar results were obtained by immunoprecipitation from cells overexpressing FLAG epitope-tagged BRG1 and HA-tagged Smad proteins (Fig. 1C). Despite the similar affinity of SWI/SNF for Smad2 and Smad3 in vitro (Fig. 1A), in vivo BRG1 interacted more robustly with Smad3 than with Smad2 (Fig. 1C). In contrast to Smad3, the main isoform of Smad2 lacks DNA binding activity (38). The stronger interaction of BRG1 with Smad3 may therefore reflect the recruitment of BRG1 to Smad transcriptional complexes that are competent to bind to DNA. HA-tagged Smad4 showed a very weak interaction with BRG1, which could be detected only on long exposures of immunoblots ( Fig. 1C and data not shown). Smad4 coexpression did not increase the interaction of BRG1 with Smad 2 or 3 (data not shown).
BRG1 and BRM are mutually exclusive ATPases in SWI/SNF complexes, whereas BAF250, BAF170, and BAF155 are core subunits that bind to both ATPases. Because BRG1 but not BRM was identified among the Smad2/3-binding proteins (Fig.  1A), we postulated that BRG1 might be the main Smad-binding component in SWI/SNF complexes. Consistent with this possibility, BRG1 interacted with Smad3 more strongly than did BAF170 (Fig. 1D). BRG1 has four conserved domains, including a proline-rich domain (domain I), a conserved domain II, a DNA-dependent ATPase domain, and a bromodomain (39,40). Using a series of vectors that encode different regions of BRG1 and Smad3, we mapped the interacting regions to the MH2 domain on Smad and to domain II and the ATPase domain on BRG1 (supplemental Fig. S1; summarized in Fig. 1E).
Differential BRG1 Sensitivity of TGF␤ Gene Responses-In various cell types that have been studied, the acute stimulation with TGF␤ typically leads to the rapid induction of over 100 genes and the repression of a smaller number of genes (23,24). To determine the role of BRG1 in the genome-wide transcriptional response to TGF␤, we first determined the sensitivity of individual TGF␤ gene responses to a partial depletion of BRG1. HaCaT cells were transfected with a combination of two siRNA oligonucleotides that decreased the level of BRG1 to ϳ20% of the level in cells transfected with siRNA control ( Fig. 2A, inset). The knockdown of BRG1 did not affect the levels of Smad2/3 ( Fig. 2A, inset) or the C-terminal phosphorylation of these proteins by TGF␤ (data not shown). These cells were incubated with TGF␤ for 2 h and subjected to transcriptomic analysis using Affymetrix HG-U133 plus 2.0 microarrays. Several comparisons were conducted to identify the effect of partial BRG1 depletion on the transcriptomic response to TGF␤. We first identified 128 genes (represented by 174 probe sets on the microarrays) whose expression in control cells was increased or decreased at least 2-fold (p Ͻ 0.05) upon TGF␤ addition (data not shown). Among these genes, 106 were up-regulated and 22 were down-regulated. Genes whose basal expression in the absence of TGF␤ was significantly affected by the BRG1 siRNA treatment were excluded from further analysis, because they would confound our interpretation of the results. This step left 84 up-regulated and 13 down-regulated genes ( Fig. 2A). The response of these 97 genes to TGF␤ was diminished to varying degrees, from 60% to no effect, in the BRG1-depleted cells compared with control cells. Fig. 2B shows a rank order of these genes according to the extent of this reduction. BRG1 depletion blunted many but not all the gene induction and repression responses to TGF␤. The affected responses included genes that are strongly responsive to TGF␤ as well as genes that are minimally responsive ( Fig. 2A). Thus, the BRG1 sensitivity of these gene responses was independent of the sign or magnitude of TGF␤ response.
To validate the microarray results, we selected a functionally diverse set of TGF␤ target genes and determine their mRNA level by quantitative real time qRT-PCR. The chosen genes include several that scored as BRG1-sensitive in the microarray assays and others that scored as BRG1-insensitive. These genes included the secretory factors PAI1 (plasminogen-activator-inhibitor 1) and CTGF (connective tissue growth factor), which are implicated in pericellular proteolysis control, extracellular matrix regulation, and fibrogenesis (41, 42); the BMP and activin membranebound inhibitor BAMBI, the inhibitory Smad SMAD7, and the Ski-like transcriptional regulator SNON, all of which participate in negative feedback control of TGF␤ signaling (43)(44)(45)(46)(47); the transcriptional activator of cell growth and proliferation MYC (48,49), which is repressed by TGF␤ action; the MYC antagonist MXD1 (50); the phosphatidic acid phosphatase PPAP2B (7); the homeobox transcription factor DLX2, which is critical in bone formation (51); the protooncogene MN1 (meningioma 1), whose fusion with TEL causes myloid leukemia (52); and the jumonji domain-containing nuclear factor JMJD3, which is a H3K27 histone demethylase and has been linked to polycomb-group-protein-mediated suppression of Hox genes and animal body patterning, X-chromosome inactivation, and possibly maintenance of embryonic stem cell identity (53)(54)(55)(56).
qRT-PCR analysis of the mRNA levels of all these genes confirmed that BRG1 depletion by either of two siRNA probes strongly inhibited the induction of BAMBI, CTGF, MXD1, and PPAP2B (Fig. 3A) and partially blunted the repression of MYC (Fig. 3B) and the induction of DLX2, JMJD3, MN1, and SMAD7 (Fig. 3A). In contrast, this level of BRG1 depletion did not significantly alter the induction of PAI1 or SNON (Fig. 3, A and C).
Probing the Physical and Functional Interactions of BRG1 and Smad2/3 in Stable Knockdown Cells-To further probe the interaction of BRG1 with Smad2/3 and to avert possible confounding effects of acute, transient expression of siRNAs in functional assays cells, we repeated these experiments using HaCaT cells that stably expressed two different shRNA vectors targeting BRG1. Each shRNA achieved a reduction of ϳ80% in BRG1 protein level (Fig. 4A). Neither shRNA altered the level of Smad2 or Smad3 or their phosphorylation in response to TGF␤, as confirmed by Western immunoblotting of cell lysates with anti-BRG1 or anti-(phospho) Smad antibodies (Fig. 4A). The level of BRG1 remained constant in both the control and the knockdown cells for at least 7 h after TGF␤ addition (Fig. 4B).
The knockdown of BRG1 strongly diminished the level of endogenous Smad2/3-BRG1 complexes that were formed in TGF␤-treated cells (Fig. 4C). The level of BAF170 protein in the BRG1 knockdown cells was not decreased compared with the level in control cells. However, the interaction of Smad2/3 and BAF170 was decreased (Fig. 4C). This result is in agreement with the possibility that BRG1 is a direct and principal Smad-binding component in BRG1-dependent SWI/SNF complexes (refer to Fig. 1). This result also sug-gests that BRM-dependent SWI/SNF complexes, which also share BAF170 (12, 16), do not take over when BRG1 levels become limiting in HaCaT cells.
Under these stable knockdown conditions we confirmed the results obtained with the siRNA-transfected HaCaT cells, with the exception of the effect on PPAP2B, which was less pronounced in the shRNA-transfected cells. The effect of BRG1 depletion was sustained for most of the duration of the TGF␤ response over a 12-h period (Fig. 4D).
BRG1 Recruitment to TGF␤/Smad Target Gene Promoters-Collectively our results with BRG1 knockdown cells suggest a widespread sensitivity of TGF␤ gene responses to BRG1 depletion and a general dependence of these responses on BRG1 function. However, this sensitivity was absent in certain TGF␤ gene responses, either because these responses can be supported by the residual levels of BRG1 that remain in knockdown cells or because the responses are truly independent from BRG1 function. To distinguish between these two possibilities, we investigated the recruitment of Smad proteins and BRG1 to TGF␤ target promoters. We focused on CTGF as an example of a TGF␤-responsive gene that is highly sensitive to BRG1 depletion, SMAD7 as a partially sensitive gene, and PAI1 and SNON as BRG1 depletion-insensitive genes.  A and B, BRG1 was depleted in HaCaT cells using a combination of two siRNA oligonucleotides targeting BRG1, as shown by immunoblot analysis (inset). Control and BRG1 knockdown HaCaT cells were incubated with 100 pM TGF␤ for 2 h, and then the total RNA was analyzed using Affymetrix HG-U133-plus 2.0 GeneChip. Blue bars depict the fold change in the mRNA signal of 97 genes whose expression in control cells was increased or decreased by at least 2-fold (p Ͻ 0.05) in response to TGF␤. The fold change of the same genes in BRG1-depleted cells is indicated with red bars. C, all 97 genes were ranked according to the percentage of reduction of their response to TGF␤ in BRG1-depleted cells. The genes whose response was verified by qRT-PCR are labeled in red. IB, immunoblotting.

BRG1 SWI/SNF in TGF␤ Signaling
TGF␤/Smad-responsive regions have been defined in the promoters of PAI1 (57), SNON (58), and CTGF and SMAD7. 3 ChIP assays using primers encompassing these regions demonstrated a specific, TGF␤-dependent recruitment of BRG1 to all four genes (Fig. 5A). Given that the interaction between BRG1 and Smad2/3 is stimulated by TGF␤ treatment (refer to Fig. 1), these results suggest that TGF␤-activated Smad transcriptional complexes drive the recruitment of BRG1 to these promoter regions. The knockdown of BRG1 had little or no effect on the TGF␤-dependent recruitment of Smad2/3 to these regions, as determined by Smad2/3 ChIP (Fig. 5B), suggesting that Smad2/3 transcriptional complexes can gain access to these target promoters under limiting BRG1 conditions. The TGF␤dependent recruitment of BRG1 to the CTGF and SMAD7 promoters was inhibited in the knockdown cells (Fig. 5B). This result is in keeping with the partial sensitivity of these TGF␤ responses to BRG1 depletion. The recruitment of BRG1 to the SNON promoter was also strongly inhibited. The fact that the SNON response to TGF␤ in these cells remained intact suggests that this gene response is independent of BRG1.
Surprisingly, the recruitment of BRG1 to the PAI1 promoter was not decreased in BRG1 knockdown cells. This result may explain the unaltered response of PAI1 to TGF␤ under these conditions. This also indicates that the TGF␤-dependent Smad complex targeting PAI1 recruits BRG1 more avidly than do the Smad complexes targeting CTGF, SMAD7, or SNON. In agreement with the extent of the TGF␤ response in each of these genes, the knockdown of BRG1 decreased the recruitment of RNA polymerase II almost completely in the case of CTGF, partially in the case of SMAD7, and not at all in the cases of PAI1 and SNON (Fig.  5B). Collectively these results show four distinct classes of TGF␤ target genes that are distinguished by the avidity of the corresponding Smad complexes for BRG1 or the degree of their dependence on BRG1.
Restoration of TGF␤ Responses in BRG1-null Lung Cancer Cells-As an independent approach to determine the dependence of the TGF␤ transcriptional program on BRG1, we investigated the TGF␤ response in BRG1-null cells. Unlike normal lung epithelial cells, which are sensitive to growth inhibition by TGF␤, a majority of human lung cancer cell lines examined lack this response (59,60). The lung adenocarcinoma cell line NCI-H522 has been reported to lack expression of BRG1 (61,62). Additionally, our analysis of this cell line revealed that it also lacks expression of the TGF␤ type II receptor (T␤R-II) (data not shown). To use this cell line for the analysis of the BRG1dependence of various TGF␤ gene responses, we employed a retroviral vector encoding the human TGFBRII cDNA. NCI-H522 cells that were transduced with this vector were able to respond to TGF␤ with Smad phosphorylation, whereas control cells transduced with empty vector lacked this response (Fig. 6A). TGF␤ addition did not alter the expression of the exogenous BRG1 in these transfected cells (Fig. 6B). Expression of BRG1 decreased the proliferation rate of NCI-H522 cells (Fig. 6C), suggesting that the loss of endogenous BRG1 expression conferred a selective advantage in this tumor cell line. TGF␤ addition fully inhibited the proliferation of NCI-H522 cells expressing T␤R-II and BRG1 in combination (Fig. 6C).
NCI-H522 cells that were transduced with BRG1, T␤R-II, or both vectors combined were incubated with TGF␤ and subjected to transcriptomic analysis. Genes whose basal expression was significantly affected by expression of BRG1 alone were excluded from the analysis. We identified 77 genes whose expression was increased or decreased by at least 2-fold upon TGF␤ addition in cells expressing both T␤R-II and BRG1 (Fig. 6D). Of these genes, 55 were up-regulated, and 22 were down-regulated. To determine the degree of dependence of these gene responses on BRG1, we ran this analysis in NCI-H522 cells expressing T␤R-II but not BRG1. A comparison between the two data sets revealed that 43 of these 77 TGF␤ gene responses in NCI-H522 are fully dependent on BRG1 function, because they were completely absent in cells expressing T␤R-II but lacking BRG1 (Fig. 6D). The remaining 34 gene responses were at least partly independent of BRG1 function, 3 Q. Wang, C. Alarcó n, and J. Massagué , unpublished observations.

BRG1 SWI/SNF in TGF␤ Signaling
because they still occurred in cells transduced with T␤R-II in the absence of BRG1. These results further support the findings made with BRG1 knockdown HaCaT cells that a majority but not all of the TGF␤ gene responses require BRG1 function.
BRG1-dependent and -independent Gene Responses in the TGF␤ Program-To validate these findings with TGF␤ gene responses of interest, we focused on gene responses that were shared between NCI-H522 and a least two of four other human epithelial cell lines. These lines include HPL1 lung epithelial cells, HaCaT keratinocytes, MCF10A breast epithelial cells, and MDA-MB-231 breast carcinoma cells, all of which were previously subjected to transcriptomic analysis on the same microarray platform (63,64). A set of 14 TGF␤ gene responses fulfilled these criteria (Fig. 7A). Within this set, six representative genes that scored with different degrees on BRG1 dependence in the microarray analysis of the TGF␤ response were chosen for qRT-PCR analysis. These genes included PAI1, SMAD7, SNON, DLX2, the stress response factor GADD45B, and the Notch target gene HEY1 (hairy/enhancer of split 1) (Fig. 7B). None of these genes was induced by TGF␤ in the absence of T␤R-II, whereas TGF␤ increased their expression severalfold in NCI-H522 cells with restored receptor and BRG1. In cells containing T␤R-II but not BRG1, the response of these genes to TGF␤ ranged from essentially none in the cases of PAI1, DLX2, and HEY1 to a partial response in the cases of GADD45B and SMAD7, and an essen-  tially full response in the case of SNON (Fig. 7B). PAI1, whose response was unaffected by an 80% depletion of BRG1 in HaCaT cells, showed a complete dependence on BRG1 in the BRG1-null NCI-H522. Thus, the TGF␤ transcriptional program includes a majority of responses that are highly dependent on BRG1 function and the rest that are partially or fully independent. Interestingly, two of the most prominent mediators of negative feedback in the Smad pathway, SMAD7 and SNON, are in this latter group.

DISCUSSION
Prompted by our previous finding that BRG1 and other core components of the SWI/SNF complex bind to recombinant Smad2 (10) and findings by others that BRG1 is required for the activation of two genes by TGF␤ in a mouse cell line (11), we have undertaken to define the requirement of BRG1 function for the genome-wide response of human cells to TGF␤. This is, to our knowledge, the first analysis on this scale of the role of a SWI/SNF core component in the transcriptional response to a cytokine-activated pathway.
In vitro Smad2 and Smad3 show a similar ability to bind a BRG1 SWI/SNF complex minimally consisting of BRG1, BAF155, BAF170, and BAF250b/OSA2. In vivo, the Smad-BRG1 interaction is enhanced upon TGF␤ addition. TGF␤ receptors act directly on Smads, enabling their accumulation in the nucleus and their association with other DNA-binding factors to specifically recognize target gene promoters. Smad2 is inferior to Smad3 at binding BRG1 in cell, which mirrors the restricted ability of the main Smad2 splice form to bind DNA (38). Thus, the properties of the BRG1-Smad interaction in vivo suggest that this interaction is consolidated upon the accumulation of Smad in the nucleus and the assembly of Smad transcriptional complexes on target gene promoters. Under conditions of RNA interference-mediated BRG1 depletion, Smad2/3 complexes still bind to target promoters in response to TGF␤, even though their ability to regulate transcription is inhibited. Therefore, BRG1 is not required for Smad binding to target promoters in vivo. Our results suggest that BRG1 is a direct and principal mediator of Smad-SWI/SNF interactions. The SWI/SNF complex component BAF170, which is also shared in BRM SWI/ SNF complexes (12,16), fails to be recruited to Smad2/3 complexes in BRG1 knockdown cells. This suggests that BRG1 plays a nonredundant role as a mediator of Smad-SWI/SNF interactions.
Our results reveal a requirement for BRG1 in the majority of TGF␤ gene responses. This suggests that Smad binding to target promoters directs BRG1 SWI/SNF complexes to remodel nearby nucleosomal structures to enable transcriptional regu- lation. These results are consistent with the recently reported ability of Smads to mediate transcription from chromatin templates in vitro (11). However, our results also show that the requirement for BRG1 is not universal among Smad target genes. Different TGF␤ gene responses show a differential sensitivity to depletion of BRG1. This is exemplified by our comparative analysis of the response of two important TGF␤ target genes, CTGF and PAI1. CTGF is implicated in extracellular matrix regulation, wound healing, fibrogenesis, and bone metastasis (42,65), whereas PAI1 is a regulator of pericellular proteolysis (41). The induction of CTGF expression by TGF␤ is highly dependent on BRG1, as shown by the high sensitivity of this response to BRG1 depletion in HaCaT cells. The PAI1 response is strictly dependent on BRG1, as evidenced by the absence of PAI1 response in BRG1-null cell line. However, this response was essentially unaffected under RNA interference conditions that cause an extensive (80%) depletion of BRG1 in HaCaT. Our ChIP results reveal that, under these conditions, the specific TGF␤-activated Smad complex that targets the PAI1 promoter still managed to recruit to this promoter enough BRG1 from the residual pool remaining in the knockdown cells. The more efficient recruitment of BRG1 to the PAI1 promoter compared with other promoters (e.g. the CTGF promoter) argues that Smad cofactors in different transcriptional complexes influence the strength of the Smad-BRG1-promoter interaction.
Although BRG1 is required for many TGF␤ gene responses, our results show that this requirement is not universal. We identified several genes whose response to TGF␤ is at least partly independent of BRG1. As determined by restoration of BRG1 function in the BRG1-null NCI-H522 lung adenocarcinoma cells, approximately one-quarter of the TGF␤ responses fall in this category (refer to Fig. 6D). This suggests that the nucleosomal packing of these genes under basal conditions is favorable for the RNA polymerase II complex to proceed with transcription in the absence of nucleosomal remodeling. Of interest, this class of genes includes two of the most prominent mediators of negative feedback in the TGF␤ pathway, namely SMAD7 and SNON. Smad7 is an inhibitory Smad that negatively regulates the TGF␤ pathway by binding and targeting T␤RII for degradation (46,47,66) and by interfering with Smad-DNA complex formation (67). SnoN is a Ski-like transcription regulator that interacts with Smad2 and Smad4 and represses their ability to activate transcription (45, 68 -70). The increased expression of these genes in response to TGF␤ in NCI-H522 and HaCaT cells is partially independent of BRG1 in the case of SMAD7 and almost fully independent in the case of SNON. The recruitment of Pol II to the SNON promoter and the transcriptional activation of this gene by TGF␤ were essentially undiminished in BRG1-depleted cells, despite the absence of BRG1 binding to this promoter under such conditions. Of note, in contrast to the context-dependent nature of many TGF␤ responses, SMAD7 and SNON are induced by TGF␤ in diverse cell types and conditions. The low requirement for BRG1 suggests a basal readiness of these genes to exert their crucial feedback function in this pathway.
TGF␤-activated Smad-BRG1 complexes bind to target promoters regardless of whether BRG1 is required for the transcriptional response of a particular gene. This is suggested by the case of SNON, whose promoter binds BRG1 along with Smads in response to TGF␤ even though BRG1 is not required for the transcriptional response of SNON. Recruitment of Smad-BRG1 complexes to target promoters may be a general event in Smad pathways. Smads as mediators of the cellular response to the TGF␤ cytokine therefore join other classes of transcriptional regulators, including developmental regulators, cell cycle checkpoint components, and nuclear hormone receptors, as factors that engage SWI/SNF remodeler complexes to effect transcriptional responses (9,12,71). Our results provide a genome-wide scope of the participation of BRG1 in TGF␤ action, suggesting a widespread yet differential involvement of SWI/SNF components in the immediate transcriptional response of many genes to this cytokine.