The entity known as the mammalian SWI/SNF complex actually consists of a small series of compositionally distinct assemblies distinguished by the presence of alternative subunits. The choice of ARID family subunit (ARID1A or ARID1B) is a determinant of complexes with generally opposing roles in cell cycle control (Ref. 6 and reviewed in Ref. 7). The complexes also contain either of two closely related alternative ATPases: Brahma (BRM)³ or Brahma-related gene 1 (BRG1). Although BRM and BRG1 share a high degree of amino acid sequence identity, they are not equally important for development. Brg1-null mice die at a pre- or peri-implantation stage (8), indicating a critical developmental role for BRG1. In contrast, Brm-null mice are viable and fertile, exhibiting only mild abnormalities that include a larger animal size and deregulated cell growth control in derived fibroblasts (9). This study also showed an increased level of BRG1 in the animal tissues in the absence of BRM, and several studies indicate that BRG1- and BRM-containing SWI/SNF complexes play largely compensatory roles in cell cycle control (e.g. Ref. 10 and reviewed in Refs. 11 and 12). Due to these phenotypes, it has been generally thought that BRM plays a similar but mostly auxiliary role to BRG1 in regulation of tissue-specific gene expression (reviewed in Ref. 5). However, few studies have compared the roles of BRM and BRG1 directly in differentiation models, and where considered (e.g. Ref. 13), BRM was generally confirmed as non-essential with relatively little other detail.

Given the ARID family evidence that alternative subunits can be important functional determinants (6, 7), we decided to probe more closely the question of the roles of BRG1 and BRM in differentiation. The ATPases were depleted individually by shRNA approaches in a differentiation model chosen for its well ordered multistep nature, with the expectation that such a model might reveal subtle differences in the requirement for each of the respective complexes. Osteoblast precursors including the mouse calvaria-derived MC3T3-E1 line undergo a tightly regulated differentiation process when induced with appropriate agents such as ascorbic acid and a source of organic phosphate (14-18). An important advantage of this model is that differentiation proceeds through discreet stages with predictable timing, providing a window for observing subtle changes in the rate of differentiation in addition to overall inhibition of differentiation. The expectation was that BRG1 depletion would block differentiation, whereas BRM depletion might cause modest delay. However, the studies described here unexpectedly revealed a programmatic role for BRM-containing complexes in repression of BRG1-dependent differentiation. Deficiency of BRM does not correlate with impaired differentiation; in contrast, it results in an accelerated rate of mineralization with constitutively higher levels of expression of osteogenic markers. These results reveal a new aspect of the alternative ATPases, identifying them as determinants of SWI/SNF complexes with opposing roles across a whole program of tissue-specific gene expression.

shRNA and Isolation of Stable BRM and BRG1 Knockdown Lines—The shRNA sequences were tested in a pSUPER vector (19). For BRM, two targeted constructs were generated using the RNAi Designer at the Clontech site. The 64-bp forward sequence for the first (BRM:seq1rnai) sequence is: 5′-GATCCCCGATCCAGAAGCTCTCCAAATTCAAGAGATTTGGAGAGCTTCTGGATCTTTTTGGAAA-3′ (the 19-bp target sequence is underlined). The respective 19-bp sequence for the alternative BRM sequence (BRM:seq5rnai) is 5′-GTCATAAGCCTGAGGCAAA-3′. The respective 19-bp BRG1 target sequence is 5′-GCCTATGGAGTCCATGCAC-3′ (adapted from Ref. 20). The pSUPER-derived vectors containing the respective knockdown sequences were introduced into MC3T3-E1 cells by lipofection together with a selectable neo marker. G418-resistant clones were amplified and screened by Western blot for BRM or BRG1 expression. Aliquots of low passage depleted lines were frozen as stocks. A control line transfected with a scrambled non-targeting sequence has been described previously (21).

Alkaline Phosphatase Staining—Cell monolayers were rinsed in PBS, fixed in 100% methanol, rinsed with PBS, and then overlaid with 1.5 ml of 0.15 mg/ml BCIP plus 0.3 mg/ml NBT (Promega, Madison, WI) for 30 min and rinsed again with PBS three times.

Mineralization Assay—Cells were induced and plated as described above. The monolayers were washed with PBS, covered with 0.1% alizarin red S for 10 min, and then rinsed with PBS three times and dried.

Northern Blots—Total cell RNA was prepared using TRIzol reagent (Invitrogen) or Tri Reagent (Sigma) according to manufacturer's recommendations. 20 μg of RNA were loaded per lane and separated by electrophoresis through a 1% formaldehyde-agarose gel. The RNA was transferred to a Hybond-N nylon membrane (Amersham Biosciences) and cross-linked by UV irradiation. ³²P-labeled probes were prepared using a random primer labeling kit (Roche Applied Science). 500 μCi of [α-³²P]ATP. was used per labeling reaction. Between successive probes, blots were stripped by treatment with boiling 0.1% SDS. The osteocalcin probe and plasmid pGB.GAPDH were described previously (22, 23).

Immunoblotting—Cells were washed and harvested in PBS and lysed in p300 lysis buffer (24). Proteins were separated by polyacrylamide gel electrophoresis, transferred to Immobilon-P membrane (Millipore), and visualized using either Western Lighting chemiluminescence reagent Plus (PerkinElmer Life Sciences) or BCIP/NBT (Promega).

Real-time PCR Assay—Real-time assays were performed with the RT² Profiler™ PCR array: mouse osteogenesis (Super-Array, Frederick, MD, catalogue number PAMM-026), according to the manufacturer's directions. The array contains primer sets for 84 osteogenesis-related genes and five housekeeping genes. The starting amount of RNA used was 1 μg. PCR was carried out on an ABI7500 cycler using the following parameters: 1 cycle for 10 min at 95 °C and 40 cycles for 15 s at 95 °C, 1 min at 60 °C. Data were analyzed using the PCR Array Data Analysis Web Portal.

All quality control parameters (genomic DNA control, reverse transcription control, and positive PCR control) were within manufacturer's recommended limits in each assay. Results for each assay were normalized to the average of all five housekeeping genes. The parental cell population and the scrambled sequence line were each analyzed in duplicate, and the averages were compared. The 10 genes that differed more than 4-fold between these two controls were excluded from further consideration. Each of the three independent BRM knockdown lines (GG5, ZD1, and ZD17) was analyzed in duplicate, and the average of the six runs was compared with the average of the parental and scrambled cells to obtain the -fold change. Likewise, the two BRG1 knockdown lines (D16 and B19) were analyzed in duplicate and again compared with the parental and scrambled cells. In accordance with recommended thresholds, a gene expression change is reported in Table 1 if it is greater than 4-fold and the t test p value is less than 0.05.

ChIP Assays—Chromatin immunoprecipitation (ChIP) assays were performed with the EZ ChIP™ system (Upstate Cell Signaling Solutions, Lake Placid, NY), according to the manufacturer's directions, modified to include preclearing of lysates with 60 μl of a 50% slurry of protein G/salmon sperm DNA for 1 h at 4 °C, and again performed overnight. Negative controls consisted of either IgG or the viral-specific monoclonal antibody 419. Primer sequences are listed in Fig. 3B. PCR conditions were 40 cycles at for 30 s at 95 °C, 30 s at 72 °C, and 30 s at 60 °C.

Re-ChIP Assays—Re-ChIP assays were performed with the Active Motif Re-ChIP-IT™ system (Active Motif, Carlsbad, CA), according to the manufacturer's directions. The assay was modified to include overnight incubations of the antibodies at 4 °C, for both the first and the second chromatin IP. Primer sequences for osteocalcin are listed in Fig. 3B. PCR conditions are same as above.

The same patterns were seen when the cells were tested for mineralization activity. Formation of calcium-containing mineralization products in the cell matrix can be detected by staining with Alizarin Red S (Fig. 1C). In this assay, the BRG1-depleted line again behaved as expected, showing virtually no mineralization (row 3 as compared with row 1). In addition, unexpectedly again, but consistent with the alkaline phosphatase induction pattern, the BRM-depleted cells showed accelerated progression to the mineralization phenotype (row 4 as compared with row 1). A control cell line (J6) derived from transfection with a non-targeting sequence (row 2) behaved like the parental cells.

Because this pattern was so contrary to expectations, additional BRM-depleted lines were constructed using a second, independent, shRNA sequence. Two separate clones (ZD1 and ZH17) were isolated. The pattern of all three BRM-depleted lines is analyzed in Fig. 1D, and all show the same phenotype. In each line, there is a constitutively enhanced level of alkaline phosphatase activity (day 0), accelerated induction of alkaline phosphatase activity, and accelerated progression to the mineralization phenotype. Rather than augmenting BRG1-dependent progression to terminal differentiation, the BRM-containing subset of SWI/SNF complexes participates in an opposing pathway, restraining differentiation.

Western blot analysis with antibodies specific for BRM and BRG1, respectively, indicate that depletion of one ATPase subunit does not have a major effect on expression of the other (Fig. 1E). It is also known from established tumor cell lines that deficiency of BRG1 or BRM, or both, does not otherwise disrupt assembly of the SWI/SNF complex (e.g. Ref. 26).

Osteocalcin Is Constitutively Expressed in BRM-depleted Cells—The best studied marker of late stage differentiation in osteoblasts is the mineralized matrix component osteocalcin. The osteocalcin gene (Bglap2) is a well established model for induction of tissue-specific gene expression whose activation has been shown to be dependent on SWI/SNF complex activity in a differentiating rat osteosarcoma cell line (27, 28). To probe the molecular events underlying the phenotypes of the knockdown lines, osteocalcin expression was assessed quantitatively by Northern blot analysis. Normally, osteocalcin expression is barely detectable in non-induced cells. After induction of differentiation, expression increases dramatically in parallel with mineral deposition. The Northern blot in panel A of Fig. 2 shows the typical pattern of osteocalcin (OSC) induction in parental cells as compared with expression in BRM and BRG1 knockdown lines. The BRG1-depleted cells show greatly impaired induction of osteocalcin (lanes 11-15 as compared with lanes 1-5), correlating with the severe defect in mineralization phenotype. In contrast, BRM-depleted cells show strikingly high constitutive expression of osteocalcin (lane 6 as compared with lane 1) and rapid induction to higher levels (lanes 6-10 as compared with lanes 1-5), concordant with the accelerated mineralization phenotype. Results averaged from three independent experiments are shown quantitatively in panel B.

Real-time PCR Array Analysis of Osteogenesis-associated Gene Expression—The deregulation of the osteocalcin gene indicates that BRM-depleted cells have lost a major promoter repression function. To gauge the extent of genes affected by BRM depletion, an array of 84 osteogenesis-associated genes was analyzed in the osteoblast precursors by quantitative real-time reverse transcription-PCR (QPCR). Each of the three independent BRM knockdown lines (GG5, ZD1, ZD17) was analyzed in duplicate, and the average of the six runs was compared with the average of duplicate runs performed on both parental cells and the J6 control line to obtain the -fold change. Taking the recommended 4-fold difference as the cut-off point, 12 genes scored as affected by BRM depletion in this assay; 10 were up-regulated, and two were down-regulated (Fig. 3A). (Osteocalcin was not present on the array.)

Among the genes constitutively up-regulated in BRM-depleted cells is Akp2, encoding alkaline phosphatase, a major osteoblast differentiation marker whose increase was also apparent at the level of enzyme activity (Fig. 1). The remainder of the list encompasses genes from multiple classes, including those encoding hematopoietic cell-associated antigen CD11b (Itgam) and the widely expressed CD36 antigen, both of which have been observed in differentiating osteoblasts (29, 30). Genes encoding various extracellular matrix components (DMP1, collagen 11, and tuftelin) are activated. In addition, expression of the enzyme-encoding genes Phex and Mmp10 is increased. These enzymes, like alkaline phosphatase, participate in phosphate and matrix metabolism. The list also includes genes encoding one of the receptors for the bone morphogenic proteins (Bmpr1b), as well as the osteogenic transcription factor MSX1. (A list of common names of the genes is shown in Table 1.)

Not every osteoblast marker on the array is constitutively activated in BRM-depleted cells, nor would they be predicted to be as the BRM-depleted cells do not mineralize spontaneously, and only a minority of promoters are thought to be targets of regulation mediated by the SWI/SNF complexes. The two down-regulated genes in the BRM-deficient cells both encode additional members of the large collagen gene family. Down-regulation of these two genes diverges from the general pattern, but the overall profile is clearly consistent with a central role for BRM complexes in restraining precocious osteoblast differentiation.

The effect of BRG1 depletion was analyzed as well. As expected, multiple genes whose expression is characteristic of osteoblast commitment were down-regulated in BRG1-depleted cells. A comparison of the effects of BRG1 versus BRM depletion (Table 1) highlights the largely antagonistic nature of their effects, consistent with the concept that the two ATPases are specificity determinants of complexes with generally opposing roles in osteogenesis.

Identification of Direct Targets of BRM-specific Complexes—The gene array results identify a minimum of 10 osteogenesis markers in addition to osteocalcin that are coordinately derepressed as a consequence of BRM deficiency. This does not, however, indicate whether the promoters of these genes are direct targets of BRM complexes. To address this question, a panel of BRM-affected genes whose promoter sequences were readily identifiable was probed by ChIP analysis in non-induced cells (Fig. 3, B and C). At least six genes were revealed in this manner to be direct targets of BRM complexes: Akp2, Col11a1, Mmp10, Msx1, Phex, and osteocalcin. In addition, BRM was weakly detectable on the Itgam promoter. In each of these cases, the promoter occupation pattern is consistent with a role for BRM complexes in promoter repression. BRM was not detected on the Col5a1 or Tuft1 promoters, implying that BRM affects these genes only indirectly. This is of particular note for the divergently regulated Col5a1, but negative results remain inconclusive. In most cases where one of the ATPases was identified in association with the promoter, the occupation was either/or with respect to BRM versus BRG1. However, BRG1 was readily detected along with BRM on the Phex and osteocalcin promoters. Phex is one of the minority cases seen in Table 1 in which BRG1 also appears to contribute a repressor role, so this is consistent with the ChIP results. In contrast, BRG1 is a required activator of osteocalcin, and the apparent presence of both ATPases simultaneously on this promoter implies a more complex mechanism of regulation of this key gene product. The osteocalcin promoter is by far the best characterized of the identified BRM-targeted promoters and was subjected to further detailed analysis. First, a serial ChIP assay was performed to determine whether the two ATPases actually do associate with this promoter simultaneously (Fig. 3D). The antibodies used in ChIP 1/ChIP 2 are indicated above the lanes. The re-ChIP (ChIP 2) confirms that BRG1 is present on the BRM-precipitated promoter DNA (lane 5), and conversely, that BRM is present on the BRG1-precipitated promoter DNA (lane 7).

BRM Complexes Override BRG1-dependent Activation of the Osteocalcin Promoter—The simplest mechanisms by which BRM-specific complexes might repress expression from a particular promoter would be by preventing association of a required activator or co-activator or by facilitating association of a required repressor or co-repressor. Interestingly, the results in Fig. 3B demonstrate that BRM complexes do not necessarily simply compete with BRG1 complexes for promoter association. Prior analysis of the osteocalcin promoter has identified certain other key factors that were considered here. RUNX2/CBFA1 is a major tissue-specific transcriptional activator controlling lineage commitment in osteoblasts (reviewed in Ref. 31) and is known to be associated with the osteocalcin promoter prior to activation (e.g. Refs. 27, 32, 33). In contrast to activation, transcriptional repression typically involves associated histone deacetylase (HDAC) activity. HDAC1 appears to be a key regulator for osteoblast differentiation and has been identified in association with the osteocalcin promoter specifically in the predifferentiation (i.e. repressed) state in primary bone marrow cells (34).

Runx2 was included in the QPCR array, and notably, its expression in the non-induced cells is unaffected by depletion of either ATPase (Table 1, Footnote 2). Prior to differentiation, BRM, BRG1, RUNX2, and HDAC1 can all be seen in association with the promoter in parental cells (Fig. 4B, upper panel, lanes 3-6). Association of RUNX2 is unaffected by BRM depletion in the BRM.GG5 cell line, whereas association of HDAC1 is lost in BRM.GG5 cells. Analysis of the BRG1.D16 line shows that BRG1 complexes, although present on the repressed promoter, are not linked with association of HDAC1. (The association patterns are represented schematically in Fig. 5.) These results combined with the biological phenotype indicate that the promoter is poised for expression in non-induced cells but that expression functions are overridden by the presence of BRM-containing complexes and their HDAC1 affiliate. Depletion of BRM essentially converts the association profile of the key indicators (Fig. 4B, lanes 3-6) from the pattern characteristic of the repressed promoter in parental cells (upper panel) to the pattern characteristic of the active promoter (lower panel). The promoter in the BRM.GG5 line at day 0 is almost as active as the parental promoter at day 21 of differentiation (Fig. 2).

ARID Family Subunits Associate Differentially with the Osteocalcin Promoter—We recently reported that the SWI/SNF subunits, ARID1A and ARID1B, are specificity determinants of complexes that play repressing versus activating roles, respectively, on pro-proliferative genes (6). These are the only subunits other than the ATPases known to exist as a mutually exclusive pair in most cells. Examination of the association pattern of the ARID family subunits with the osteocalcin promoter (Fig. 4B, lanes 7 and 8) shows ARID1A present only on the repressed promoter and ARID1B present only when the promoter is active. ARID1A association is not dependent on either ATPase individually, so this subunit is likely associated with both complexes on the repressed promoter. Because only ARID1B is on the active promoter, it appears that the BRG1 complex changes from an ARID1A-containing configuration on the repressed promoter to an ARID1B-containing configuration on the active promoter (Fig. 5, schematic). As a further control, the presence of the INI1/SNF5 subunit was also probed. INI1/SNF5 is present in all known subsets of the complex. Its association profile (lane 9) is consistent with this and with a general finding that the presence of an ATPase subunit is required for promoter association of the complex as a whole.

Although the ARID family subunits help to distinguish activator versus repressor complexes, their role is apparently not essential on the osteocalcin promoter as the ARID1A and ARID1B knockdown lines do not show major differences in mineralization phenotypes (data not shown). This is consistent with the ChIP results indicating that the BRG1 complex does not need to switch to an ARID1B subunit to effect constitutive activation of osteocalcin in BRM-depleted cells. However, a complex specifically containing BRG1 itself is clearly required for activation (Fig. 2B). The four different combinations of ARID subunit and ATPase (25, 35) help explain how each of the subunits can be identified on both active and repressed promoters, although their respective roles are not random. The significance of the alternative subunits is only beginning to be addressed, but evidence so far suggests a general pattern in which BRM+ARID1A complexes are linked most closely with repression and BRG1+ARID1B complexes are linked most closely with activation, whereas the other possible combinations, BRM+ARID1B and BRG1+ARID1A, are more variable in their activities.

Disassociation of BRM from the Osteocalcin Promoter Correlates Temporally with Up-regulation of Osteocalcin Expression—To obtain a more dynamic picture of SWI/SNF-mediated regulation of the osteocalcin promoter, the association of key factors was probed at major intermediate time points (Fig. 4C). The results show that BRM is still present at day 7 but dissociates by day 14. This correlates well with the pattern of osteocalcin induction, which rises rapidly between days 14 and 21. The results reveal further that dissociation of HDAC1 (lane 3) precedes dissociation of BRM and that association of ARID1B (lane 6) precedes complete dissociation of ARID1A. The promoter association dynamics imply the existence of a transition point at about day seven when the promoter is undergoing initial stages of activation (Fig. 5, schematic), which correlates with the initial level of activation seen at day 7 in Northern blots. Presumably, following this transition period, other chromatin events occur that lead to full activation. One important event characteristic of activated promoters and associated with increased histone acetylation is trimethylation at histone H3 lysine 4 (H3K4) (36). This modification increases on the osteocalcin promoter at later times (days 14 and 21) (Fig. 4C, lane 9) coordinately with dissociation of the BRM complexes.

These results do not imply that BRM-containing complexes are linked exclusively with repression; indeed tissue-specific gene activation functions have been identified in other contexts (e.g. Refs. 20, 37-39). In particular, a positive role for BRM in retinal cell differentiation was identified recently (40). Conversely, the results also do not imply that BRG1 is linked only with activation of differentiation. Conditional loss of BRG1 in neuronal stem cells results in precocious neuronal differentiation, implying a role for BRG1 complexes in maintaining the neuronal stem cell phenotype (41); this effect may be analogous to what we see with depletion of BRM in osteoblast precursors. Interestingly, ARID1A also appears to play a role in maintaining a stem cell phenotype (42). What is essential in the current context is that the roles of the two ATPases were not compared directly in these other systems, so the distinct question of balanced opposing functions was not addressed.

The results described here suggest that at least one mechanism by which BRM-containing complexes effect repression is by mediating promoter association of the histone deacetylase, HDAC1. This interpretation is supported by analysis elsewhere of the effects of HDAC1 depletion in a rat osteosarcoma cell model. Depletion of HDAC1 via small interfering RNA-mediated knockdown was accompanied by increased alkaline phosphatase activity and heightened expression of osteoblast differentiation markers (34), a phenotype very similar to the phenotype reported here for BRM-depleted cells. These factors are not the only ones active at the promoter, of course, but the present results establish that BRM-containing complexes are essential for repression and that the default condition of the osteocalcin promoter in the absence of a BRM complex is active expression, even without a signal for differentiation.