The Brm-HDAC3-Erm repressor complex suppresses dedifferentiation in Drosophila type II neuroblast lineages

The control of self-renewal and differentiation of neural stem and progenitor cells is a crucial issue in stem cell and cancer biology. Drosophila type II neuroblast lineages are prone to developing impaired neuroblast homeostasis if the limited self-renewing potential of intermediate neural progenitors (INPs) is unrestrained. Here, we demonstrate that Drosophila SWI/SNF chromatin remodeling Brahma (Brm) complex functions cooperatively with another chromatin remodeling factor, Histone deacetylase 3 (HDAC3) to suppress the formation of ectopic type II neuroblasts. We show that multiple components of the Brm complex and HDAC3 physically associate with Earmuff (Erm), a type II-specific transcription factor that prevents dedifferentiation of INPs into neuroblasts. Consistently, the predicted Erm-binding motif is present in most of known binding loci of Brm. Furthermore, brm and hdac3 genetically interact with erm to prevent type II neuroblast overgrowth. Thus, the Brm-HDAC3-Erm repressor complex suppresses dedifferentiation of INPs back into type II neuroblasts. DOI: http://dx.doi.org/10.7554/eLife.01906.001


Introduction
The mechanism by which self-renewal and differentiation are balanced is a crucial issue in stem cell and cancer biology. The neural stem cells, or neuroblasts, of the Drosophila larval brain have emerged as a new model for studying stem cell self-renewal and tumorigenesis. In Drosophila larval central brains, there are at least two classes of neuroblast lineages (Bello et al., 2008;Boone and Doe, 2008;Bowman et al., eLife digest Stem cells show great promise for repairing damaged tissue, and maybe even generating new organs, but stem cell therapies will only be successful if researchers can understand and control the behaviour of stem cells in the lab. Neural stem cells or 'neuroblasts' from the brains of larval fruit flies have become a popular model for studying these processes, and one type of neuroblast-known as a 'type II' neuroblast-is similar to mammalian neural stem cells in many ways. When type II neuroblasts divide, they generate another neuroblast and a second cell called an intermediate neural progenitor (INP) cell. This progenitor cell then matures and undergoes a limited number of divisions to generate more INP cells and cells called ganglion mother cells. The process by which stem cells and INP cells become specific types of cells is known as differentiation. However, under certain circumstances, the INP cells can undergo the opposite process, which is called dedifferentiation, and become 'ectopic neuroblasts'. This can give rise to tumors, so cells must employ a mechanism to prevent dedifferentiation. Researchers have known that a protein specifically expressed in INP cells called Earmuff is involved in this process, but many of the details have remained hidden. Now, Koe et al. have discovered that a multi-protein complex containing Earmuff and a number of other proteins-Brahma and HDAC3-have important roles in preventing dedifferentiation. All three proteins are involved in different aspects of gene expression: Earmuff is a transcription factor that controls the process by which the genes in DNA are transcribed to make molecules of messenger RNA; Brahma and HDAC3 are both involved in a process called chromatin remodeling. The DNA inside cells is packaged into a compact structure known as chromatin, and chromatin remodeling involves partially unpacking this structure so that transcription factors and other proteins can have access to the DNA.
Koe et al. also showed that Earmuff, Brahma and HDAC3 combine to form a complex that prevents dedifferentiation. An immediate priority is to identify those genes whose expression is regulated by this complex in order to prevent dedifferentiation. absent from type II neuroblasts, triggers cell cycle exit and GMC differentiation (Bello et al., 2006;Betschinger et al., 2006;Choksi et al., 2006;Lee et al., 2006c). However, the underlying mechanism by which Erm prevents dedifferentiation is poorly understood.
ATP-dependent chromatin-remodeling factors are critical for the expression of the eukaryotic genome. Four major classes of ATP-dependent chromatin remodeling complexes have been identified, including the extensively studied SWI/SNF complexes (Narlikar et al., 2002;Reisman et al., 2009). The mammalian SWI/SNF complex termed the Brahma (Brm or Brg1) complex regulates critical cellular processes such as differentiation and cell cycle arrest (Klochendler-Yeivin et al., 2002). Drosophila Brm complex acts similarly to control cell proliferation (Brumby et al., 2002) and differentiation (Marenda et al., 2003). A genome-wide RNAi study in Drosophila neuroblasts showed that the knockdown of genes encoding several core subunits of the SWI/SNF Brahma (Brm) remodeling complex may lead to neuroblast overproliferation (Neumuller et al., 2011). However, the precise role of the Brm remodeling complex during neuroblast self-renewal and the mechanism that underlying underlies this effect mechanism remain to be elucidated. Besides ATP-dependent chromatin remodeling complexes, the other major class of chromatin remodelers is histone modifiers. Histone deacetylases (HDACs) remove acetyl groups from the tails of core histones in the nucleosome and are often associated with transcriptional co-repressors (Dokmanovic et al., 2007). However, despite the critical role for histone modifiers in transcriptional regulation, it is unknown whether histone modifications play any role in Drosophila larval brain neuroblasts.
In this study, we report the critical role of a central chromatin remodeler, the Brm complex in preventing the formation of ectopic neuroblasts in type II lineages. We show that another chromatin remodeling factor, HDAC3 functions cooperatively with the Brm complex to suppress the formation of ectopic type II neuroblasts. Interestingly, multiple components of the Brm complex and HDAC3 physically associate with Erm. brm and hdac3 interact genetically with erm to prevent type II neuroblast overgrowth. Thus, the Brm-HDAC3-Erm complex is a novel repressor complex that suppresses dedifferentiation of INPs back into type II neuroblasts.

Results
The Brm complex suppresses the formation of ectopic neuroblasts in type II neuroblast lineages We independently identified brm from a RNA interference (RNAi) screen in which brm RNAi knockdown in larval brains resulted in an increase of Miranda-positive neuroblast-like cells in larval central brains ( Figure 1-figure supplement 1A), showing a phenotype similar to one previously reported (Neumuller et al., 2011). Brm is a DNA-dependent ATPase and a major component of a multi-protein SWI/SNF chromatin remodeling complex, which controls gene expression by altering chromatin structure (Klochendler-Yeivin et al., 2002). The number of cells expressing the proto-oncogene dMyc was significantly increased upon brm RNAi knockdown ( Figure 1-figure supplement 1B), consistent with the neuroblast overgrowth phenotype. To determine the function of Brm in different neuroblast lineages, we generated MARCM clones in two brm loss-of-function alleles. Type I wild-type (wt) clones always contained one neuroblast that is positive for both Dpn and Ase (data not shown). Similarly, only one neuroblast was present in both amorphic brm 2 and hypomorphic brm T362 type I clones (data not shown), indicating that Brm has no significant effect on type I neuroblast numbers. Each wt type II MARCM clone also possessed only one neuroblast that was positive for Dpn, but negative for Ase ( Figure 1A; n = 25). Unlike the wt control, 6.4 ± 3.3 and 4.5 ± 2.6 ectopic neuroblasts were observed in brm 2 ( Figure 1B,B′,D; 88.6%, n = 34) brm T362 ( Figure 1C,C′,D; 75.9%, n = 58) type II clones, respectively. These phenotypes in brm alleles could be rescued by expressing a wild-type brm transgene (Figure 1figure supplement 1C). Consistent with phenotypes in brm clones, knockdown of brm by RNAi using a type II neuroblast-specific neuroblast driver worniu (wor)-Gal4, ase-Gal80 (henceforth referred to as 'type II driver'; 'Materials and methods') was sufficient to produce ectopic neuroblasts in 97.4% of type II neuroblast lineages ( Figure 1F,K; 7 ± 3.4 neuroblasts/lineage, n = 38), while each control clone always has one neuroblast ( Figure 1E,K; n = 80). We therefore conclude that Brm suppresses the formation of ectopic neuroblasts in type II neuroblast lineages.
Concomitantly with the formation of supernumerary neuroblasts, the number of mature INPs that are positive for Dpn and Ase were dramatically reduced in both brm 2 ( Figure 2B,B′,D; 5.3 ± 4.3/clone, n = 39) and brm T362 ( Figure 2C,C′,D; 9.6 ± 4.6/clone, n = 56), compared with the control type II clones (Figure 2A,A′,D;18.9 ± 3.7,n = 22). The number of Dpn − PntP1 + immature INPs and early mature INPs appeared to be normal or slightly increased in brm 2 ( Figure 2F; 8.4 ± 2.5, n = 26) and brm T362 ( Figure 2G; 7.1 ± 2.1, n = 33) clones compared with the control ( Figure 2E; 5.9 ± 1.1, n = 26). These results suggested that ectopic neuroblasts in brm − clones likely originate from INPs that fail to undergo maturation. To further    0 neuroblast/INP clone, n = 53). The relatively weak phenotype is likely due to incomplete knockdown of Brm, as shown by the reduced Brm staining in the INP clones ( Figure 2-figure supplement 1). Thus, our data suggest that Brm functions in INPs to prevent INP dedifferentiation back into neuroblasts.
To assess whether Brm was affecting apico-basal cell polarity, we examined the localization of aPKC, Numb and Brat, which are asymmetrically localized in wild-type neuroblasts in prometaphase/ metaphase. brm 2 MARCM clones did not show any defects in the localization of these markers ( Figure 3-figure supplement 1A-C), suggesting that Brm is not important for the apical-basal polarity regulation in neuroblasts.
We then assessed the function of other core subunits of the Brm remodeling complex using RNAi knockdown in larval brains. Knock down either of the three core components snr1 ( Figure 1G,K, 9.5 ± 5.5 neuroblasts/lineage, 64.5%, n = 34), bap60 ( Figure 1H,K, 5.1 ± 2.8 neuroblasts/lineage, 87.0%, n = 52), and moira (data not shown; also reported in Neumuller et al. (2011)) resulted in prominent phenotypes with ectopic Dpn + Ase − neuroblasts in type II neuroblast lineages. Two additional snr1 RNAi lines under the control of the type II driver also displayed excess type II neuroblasts ( Figure 1figure supplement 1D). Furthermore, bap55 LL5955 MARCM clones also showed ectopic neuroblasts in type II ( Figure 1J,J′, 2.3 ± 1.7 neuroblasts/clone; 30.5%, n = 57), but not type I neuroblast lineages ( Figure 1-figure supplement 1F; n = 20). The number of Dpn + Ase + mature INPs in bap55 LL5955 was significantly reduced (13.8 ± 4.1; n = 49) compared with the control (18.9 ± 3.7, n = 22), while the number of Dpn − PntP1 + immature and early mature INPs (4.2 ± 1.9, n = 26) was similar to the control (5.9 ± 1.1, n = 26). In various mutants and RNAi lines described above, we also observed an increased number of Dpn + PntP1 + cells (data not shown), which serves as an independent set of marker for type II neuroblasts. This data further supports that loss-of-function of the Brm complex caused the phenotype of ectopic type II neuroblasts. We conclude that core components of the Brm remodeling complex are required to suppress ectopic neuroblast formation in type II neuroblast lineages.
Next, we ascertained whether other chromatin remodeling complexes such as Nucleosome remodeling factor (NURF) and ACF complex (ATP-utilizing chromatin assembly and remodeling factor), play any role during neuroblast self-renewal. RNAi knockdown of nurf301 or ACF complex components iswi and acf1 (Figure 1-figure supplement 1G) in the type II neuroblast lineages did not result in any obvious neuroblast overgrowth. These data suggest that they may not be important for type II neuroblast lineages or their RNAi targeting was insufficient to induce an effect.

HDAC3 acts cooperatively with the Brm complex to suppress the formation of ectopic type II neuroblasts
To assess the involvement of histone modifications in type II neuroblast lineages, we screened a collection of 43 histone modifiers ( Table 1; Kirilly et al., 2011) by RNAi under the control of a type II-specific driver but failed to identify any RNAi lines with ectopic neuroblasts. We reasoned that histone modifiers may act cooperatively with the Brm complex in type II neuroblast lineages and the phenotype may be masked due to the presence of the functional Brm complex. We therefore re-screened the same collection of potential histone modifiers in a brm RNAi background and showed that the simultaneous knockdown of both brm and hdac3 under the control of the type II driver resulted in a more severe phenotype of ectopic neuroblasts ( Figure 3D,E; 22.1 ± 7.4 neuroblasts/clone, n = 21) compared with brm RNAi ( Figure 3B,E; 10.3 ± 5.6 neuroblasts/clone, n = 30) or hdac3 RNAi knockdown alone ( Figure 3C,E; 1.1 ± 0.5 neuroblasts/clone, n = 85). This finding suggests that HDAC3 functions cooperatively with Brm to regulate type II neuroblast lineages. Next, we took advantage of an existing deletion mutant snr1 6c hdac3 6c , which removes the entire snr1 coding region and the C-terminal region of hdac3. Type II neuroblast MARCM clones from snr1 6c hdac3 6c homozygotes possessed a large number of ectopic neuroblasts ( Figure 3G,H, 22 ± 11.5 neuroblasts/clone, 83.0%, n = 22). The number of mature Dpn + Ase + INPs in each snr1 6c hdac3 6c type II neuroblast clone was modestly reduced to 15.6 ± 4.4 (n = 22) compared with 20.3 ± 3/clone (n = 20) in control, while the number of Dpn − PntP1 + immature and early mature INPs (6.3 ± 3/clone, n = 21) are slightly greater compared with the control clones (4.1 ± 0.9, n = 21). In contrast, type I neuroblast clones of this double mutant appeared normal, as there was only one neuroblast per clone (  Because Snr1 protein was speculated to have extended perdurance in somatic clones of its null allele (Marenda et al., 2004), the lack of phenotype in snr1 R3 is likely due to protein perdurance in the neuroblast clones. Our data suggest that HDAC3 acts cooperatively with the Brm complex to suppress the formation of ectopic type II neuroblasts. To ascertain whether snr1 6c hdac3 6c causes tumorigenesis, larval brain tissues carrying snr1 6c hdac3 6c MARCM clones were transplanted into the abdomen of wildtype hosts. A significant portion of the mutant tissue ( Figure 3J; 21%, n = 14) proliferated massively and formed malignant tumors, whilst control clones did not proliferate after the implantation ( Figure 3I; n = 25). In subsequent rounds of transplantation, 70% (T1, n = 10) and 80% (T2, n = 5) of the snr1 6c hdac3 6c mutant brain tissues developed tumors, suggesting that snr1 6c hdac3 6c can induce malignant tumor-like growth after allograft culture.

The Brm remodeling complex physically associates with Erm and HDAC3
Given that Brm is ubiquitously expressed in various cell types of larval brains, including neuroblasts, INPs, GMCs and neurons (Figure 2-figure supplement 1 and data not shown), it is conceivable that the Brm complex may associate with regulatory protein(s) or co-factor(s) that is/are specifically expressed in type II neuroblast lineages to suppress the formation of ectopic neuroblasts. We therefore assessed whether Brm could associate with two such type II-specific transcription factors, Erm and PntP1. Flag-tagged Brm was co-transfected with Myc-tagged Erm in S2 cells. Following immunoprecipitation (IP) of Flag-Brm, Erm can be specifically detected in the immune complex ( Figure 4A). Consistently, Flag-Brm was detected in the immune complex following the IP of Myc-Erm ( Figure 4A). In contrast, Myc-Brm did not associate with Flag-PntP1 in similar co-IP experiments (Figure 4-figure supplement 1A), suggesting that Brm specifically associates in a protein complex with Erm, but not with PntP1.
Since full-length Erm had very low expression levels in S2 cells, we expressed two truncated proteins, Erm N-terminal 1-441aa (Erm-N containing N-terminal region and four of six zinc-finger domains) and Erm C-terminal 332-611aa (Erm-C containing the last four zinc-finger domains and its C-terminus), and used them for the subsequent co-IP analysis ( Figure 4B). In co-IP experiments, Myc-Brm associated strongly with Flag-Erm-N and weakly with Flag-Erm-C ( Figure 4C). Furthermore, we ascertained whether Brm could associate with Erm in a protein pull-down assay. Full-length Erm could not be efficiently expressed when fused with Maltose-Binding Protein (MBP) in bacteria; we therefore expressed truncated MBP-Erm-N or MBP-Erm-C. These fusion proteins were then bound to amylose resin, and subsequently incubated with protein extracts from S2 cells transfected with Myc-Brm. Following the pull-down of amylose resin, Myc-Brm associated intensely with MBP-Erm-N and weakly with Erm-C, but not with the MBP control ( Figure 4H). These data suggest that Brm physically associates with Erm and Erm N-terminus appears to be more important for this association. We next ascertained whether Erm associates with other components of the Brm remodeling complex, such as BAP60 and Snr1. In co-IP experiments, Flag-BAP60 was detected in the immune complex following IP of Myc-Erm-N but not Myc-Erm-C ( Figure 4D). Likewise, Myc-Erm-N was detected in the immune complex when IP was performed using Flag-BAP60 ( Figure 4D). Moreover, Myc-Snr1 associated with Flag-Erm-N or Flag-Erm-C in co-IP experiments ( Figure 4E). Consistently, Myc-Snr1 associated with MBP-Erm-N and weakly with MBP-Erm-C in protein pull-down assays ( Figure 4H). Therefore, we conclude that the several components of the Brm remodeling complex specifically associates with Erm.
Given that HDAC3 functions cooperatively with the Brm complex to suppress the generation of ectopic neuroblasts, we ascertained whether HDAC3 can physically associate with both Brm and Erm. Flag-HDAC3 and Myc-Brm were co-transfected in S2 cells. Following immunoprecipitation of Flag-HDAC3, Myc-Brm was clearly detected in the immune complex ( Figure 4F). Similarly, Flag-HDAC3  Figure 4F). Interestingly, HDAC3 also physically associates with Erm-N, but not Erm-C, in both co-IP ( Figure 4G) and protein pull-down assay ( Figure 4H). Taken together, these results show that Brm physically associates with Erm and HDAC3 in the protein complex.
brm and hdac3 genetically interact with erm to prevent dedifferentiation of INPs to neuroblasts Given that Brm and Erm associate in a protein complex, and both of them suppress the formation of ectopic neuroblasts, we assessed whether brm genetically interacts with erm to prevent dedifferentiation of INPs to neuroblasts. First, we ascertained whether erm knockdown can exacerbate the brm RNAi phenotype in type II neuroblast lineages. The simultaneous knockdown of brm and erm resulted in a much more severe phenotype with a large number of ectopic neuroblasts in each lineage ( Figure 5D,E; 37.1 ± 7.6 neuroblasts/lineage, n = 20), in contrast to brm knockdown alone ( Figure 5B,E; 9.5 ± 2.8, n = 32). It was reported that in ermmutants, dedifferentiated neuroblasts can establish ectopic type II neuroblast lineages and form ectopic glial chambers (Weng et al., 2010). Presumably due to incomplete knockdown of erm that only led to weak ectopic type II neuroblasts phenotypes, erm RNAi under the type II driver resulted in ectopic type II lineages with each lineage containing one type II neuroblast ( Figure 5C,E; 1 neuroblast/lineage, n = 60). This dramatic enhancement suggests that brm and erm genetically interact to prevent the dedifferentiation of INPs back to neuroblasts. Furthermore, knock down of erm by RNAi in the brm T362 MARCM clones ( Figure 5-figure supplement 1A,B; 9.9 ± 5.5 neuroblasts/clone, n = 32) also significantly enhanced neuroblast overgrowth compared with brm T362 clones ( Figure 5-figure supplement 1A,B; 4.1 ± 2.4 neuroblasts/clone, n = 30). However, the size of the brm T362 clones with erm knockdown remained smaller than the control clones, probably due to the reduced number of INPs that are required to expand the clonal size. Erm overexpression has previously been shown to result in premature differentiation of type II neuroblasts (Weng et al., 2010). Similarly, we found that overexpression of Erm in type II MARCM clones caused 100% of the neuroblasts to undergo premature differentiation; 29.3% of the clones contained a neuroblast that gained Ase expression and 41.5% of the clones contained a neuroblast that had gained Ase expression with strongly reduced Dpn expression, while the rest of the clones showed no obvious neuroblasts ( Figure 5G and data not shown; n = 41). To assess whether this effect is dependent on its association with Brm, we overexpressed Erm in brm 2 type II neuroblast clones. The premature differentiation of type II neuroblasts was dramatically suppressed in a brm loss-of-function mutant background, as there were still 37.8% of clones contained ectopic neuroblasts ( Figure 5I), while 62.2% of type II neuroblasts underwent premature differentiation ( Figure 5H; n = 37). Similarly, the simultaneous knockdown of both erm and snr1 resulted in a more severe phenotype of ectopic neuroblasts ( Figure 5M-N; 862 ± 106.8 neuroblasts/brain lobe, n = 20) compared with either erm knockdown ( Figure 5K,N; 76.6 ± 14.2 neuroblasts/brain lobe, n = 20) or snr1 knockdown ( Figure 5L,N; 219.5 ± 52.2 neuroblasts/brain lobe, n = 20). Thus, we conclude that brm and snr1 genetically interact with erm to prevent dedifferentiation of INPs to neuroblasts. Furthermore, the simultaneous knockdown of hdac3 and erm under the control of type II driver also resulted in a more dramatic increase of ectopic type II neuroblasts ( Figure 5Q,R; 565.4 ± 68.1 type II neuroblasts/ brain lobe, n = 20) compared with either erm knockdown ( Figure 5P,R; 76.0 ± 7.7 type II neuroblasts/ brain lobe, n = 20) or hdac3 knockdown ( Figure 5R; 8 type II neuroblasts/brain lobe, n = 20), suggesting that hdac3 and erm genetically interact in type II neuroblast lineages. Taken together, these results indicate that Brm, HDAC3, and Erm function as a repressor complex to prevent INP dedifferentiation into type II neuroblasts.

Discussion
Here, we report a critical function of the  (Reisman et al., 2009). Mutations in epigenetic regulators are found in approximately half of hepatocellular carcinoma and bladder cancers, and represent a significant portion of mutated genes in medulloblastoma (Gui et al., 2011;Fujimoto et al., 2012;Pugh et al., 2012). Drosophila Brm complex is essential for intestinal stem cell proliferation and commitment in the adult intestine (Jin et al., 2013;Zeng et al., 2013). Two other chromatin remodeling factors, Iswi and Domino control germline stem cell and somatic stem cell self-renewal in the ovary (Xi and Xie, 2005).
We have demonstrated that Brm physically associates with Erm, a type II-specific transcription factor that prevents the dedifferentiation of INPs back into neuroblasts. Furthermore, Bap60 and Snr1, two other components of the Brm complex, also physically associate with Erm in a protein complex. Therefore, we have provided the first molecular link during the regulation of type II neuroblast lineages. We speculate that the association with Erm may provide functional specificity of the Brm remodeling complex in type II neuroblast lineages. We have also shown that brm genetically interacts with the type II-specific transcription factor erm. Ectopic neuroblast phenotype resulting from brm knockdown was dramatically enhanced  (Weng et al., 2010), can be partially suppressed by loss of notch ( Figure 5-figure supplement 1C,D). These functional data suggest that Erm is a co-factor of the Brm remodeling complex in type II neuroblast lineages. However, it is uncertain how the Brm-Erm protein complex functions to prevent dedifferentiation in type II neuroblast lineages. Our bioinformatic analysis has identified a 14 bp-long motif as the de novo Erm DNA-binding motif (Figure 4-figure supplement 1B-D; and Supplementary methods) and 202 sites out of the 270 known genomic loci harboring Brm (Negre et al., 2011) also contain the de novo Erm DNAbinding motif (Table 2, Gene list). As there are many genes that are potentially co-occupied by Brm and Erm, it is possible that Brm-Erm complex results in a unique configuration of the chromatin 'landscape' in INPs to prevent INP dedifferentiation into neuroblasts. Therefore, disruption of chromatin remodelers may cause widespread changes to the transcriptome, thus amplifying the effect of the single genetic mutation.
Most class I HDACs are recruited into large multi-subunit co-repressor complexes for maximal activity (Wen et al., 2000). HDAC1 and 2 are found in multiple co-repressor complexes, while to date HDAC3 appears to be uniquely recruited to the Silencing mediator of retinoic and thyroid receptors (SMRT)/Nuclear receptor co-repressor (N-CoR) complex (Guenther et al., 2000;Li et al., 2000). Here, we report that Drosophila HDAC3 is recruited to a novel multi-subunit complex containing Brm and Erm and that this co-repressor complex prevents dedifferentiation of INPs into type II neuroblasts. The SMRT complex appears not to be important for type II neuroblasts, as knockdown of smrter that encodes a core component of the SMRT complex (Heck et al., 2012) neither resulted in any ectopic type II neuroblasts nor enhanced the phenotype of ectopic neuroblasts by brm knockdown (data not shown). We also showed that HDAC3 dramatically enhanced the phenotype of ectopic neuroblast upon loss of brm or snr1, two core components of the Brm complex. By identifying this novel repressor complex, we have provided a mechanistic link between transcriptional repression and histone deacetylation during the suppression of dedifferentiation. HDACs are typically recruited by oncogenic protein complexes in lymphoma and leukemia and HDAC3 inhibitors are synergistic or additive with anticancer agents for therapeutics (Dokmanovic et al., 2007)  complex in suppressing suppressing dedifferentiation of INPs into neuroblasts and induces tumors in the allograph transplantation revealed an unexpected potential involvement of HDAC3 in tumor suppression in brain tissue. It will be of interest to determine whether this effect is conserved in the mammalian central nervous system and whether it occurs in tissues other than the brain.

Immunohistochemistry and immunoblotting
Third instar larval brains were dissected and fixed with 3.7% formaldehyde in PBS. Fixed brains were blocked with 3% BSA for one hour and then incubated with primary antibody in 3% BSA (in 0.3% PBS-T) over night at 4°C. Following three times washing (10 min each), larval brains were incubated with secondary antibody diluted in 0.3% PBS-T for 1.5 hr. After two times washing (10 min each), DNA was labeled by ToPro-3 (1:5000; Invitrogen, Carlsbad, CA) in 0.3% PBS-T for 20 min. Larval brains were mounted in vector shield (Vector Laboratory, Burlingame, CA) for

Clonal analysis
MARCM clones were generated as previously described (Lee and Luo, 1999

Transplantation
Allograft culture of larval brain tissue was carried out as previously described (Castellanos et

Bioinformatics
From the previously reported ChIP-chip data, we obtained a list of 270 Brm binding sites. To determine if Erm also binds to these binding sites, we first analyzed the DNA binding domains of Erm, which contain 6 zinc fingers. Each zinc-finger domain was assigned a DNA binding preference (position weighted matrix) based on published methods (Kaplan et al., 2005). We then scanned these six DNA binding preferences of approximately +/−200 bp around the 270 Brm binding sites and found the DNA binding preferences of the 1st zinc-finger 'GTAG' and the 4th zinc-finger 'RAAA' are enriched in the 270 Brm binding sites. The sites enriched with these two binding preferences were subjected to further analysis using the de novo motif-finding program SEME , and a 14 bp-long motif was identified as the de novo Erm DNA-binding motif. We scanned +/−200 bp around the 270 Brm binding sites with the de novo Erm DNA-binding motif. Among them, 202 sites (FDR<0.0001) were identified as putative Erm-binding sites with an AUC score of 0.73, which is significantly higher than the AUC score computed for random motifs (0.5). As a negative control, the same approach was also applied to predict the motif of Zinc-finger protein (Zif), which regulates asymmetric division of neuroblasts and therefore is unlikely to be a co-factor with Brm. The predicted Zif DNA-binding motif differs dramatically with the predicted Erm DNA-binding motif in sequence. Furthermore, it was not significantly enriched in the 270 Brm binding sites and had an AUC score of 0.54, similar to the AUC score (0.5) for random motifs. Thus, our data suggests that Brm and Erm can potentially regulate a set of common downstream targets.