Role of MXD3 in Proliferation of DAOY Human Medulloblastoma Cells

A subset of medulloblastomas, the most common brain tumor in children, is hypothesized to originate from granule neuron precursors (GNPs) in which the sonic hedgehog (SHH) pathway is over-activated. MXD3, a basic helix-look-helix zipper transcription factor of the MAD family, has been reported to be upregulated during postnatal cerebellar development and to promote GNP proliferation and MYCN expression. Mxd3 is upregulated in mouse models of medulloblastoma as well as in human medulloblastomas. Therefore, we hypothesize that MXD3 plays a role in the cellular events that lead to medulloblastoma biogenesis. In agreement with its proliferative role in GNPs, MXD3 knock-down in DAOY cells resulted in decreased proliferation. Sustained overexpression of MXD3 resulted in decreased cell numbers due to increased apoptosis and cell cycle arrest. Structure-function analysis revealed that the Sin3 interacting domain, the basic domain, and binding to E-boxes are essential for this activity. Microarray-based expression analysis indicated up-regulation of 84 genes and down-regulation of 47 genes. Potential direct MXD3 target genes were identified by ChIP-chip. Our results suggest that MXD3 is necessary for DAOY medulloblastoma cell proliferation. However, increased level and/or duration of MXD3 expression ultimately reduces cell numbers via increased cell death and cell cycle arrest.


Introduction
Tumors of the central nervous system (CNS) comprise nearly one quarter of all pediatric cancers. Among them, medulloblastomas, embryonic neuroepithelial tumors of the cerebellum, are the most common [1]. Subsets of medulloblastomas are thought to have different developmental origins. ''SHH-type'' medulloblastomas originate from a distinct population of cells within the cerebellum -granule neuron precursors (GNPs) -in which the SHH pathway is persistently activated [2]. During normal development, GNPs proliferate in response to SHH [3], and later differentiate and migrate to the internal granule layer. Two of the most dramatically SHH-induced genes are Dcyclin and Mycn [4]. MYCN is a member of the MYC/MAX/MAD network of the basic helix-loop-helix leucine zipper (bHLHZ) DNA binding proteins. MYC and MAD form heterodimers with the cofactor MAX to achieve binding to the E-box sequence CANNTG [5]. It is accepted that MYC/MAX heterodimers promote proliferation by activating transcription of cell cycle-specific genes through mechanisms that include recruitment of histone acetyl-transferase (HAT) activity [6]. In contrast, MAD/MAX heterodimers repress transcription by recruiting histone deacetylase (HDAC) activity [7], thereby antagonizing MYC/MAX heterodimers, leading to proliferation arrest and differentiation. Hence, current models suggest that the MYC/MAX/MAD network plays an important role in the switch from proliferation (MYC/MAX) to differentiation (MAD/MAX) [5].
The Mad family of transcription factors consists of four paralogous genes: MXD1, MXI1, MXD3 and MXD4. Consistent with the proliferation/differentiation switch paradigm, genetic ablation of mxd1 in mice results in defects in cell cycle exit during myeloid differentiation [8], and disruption of the mxi1 locus results in hyperplasia in several tissues [9]. Several reports, however, suggest that MXD3 is an atypical member of the MAD family. Mice with a targeted deletion of Mxd3 showed a mild phenotype consisting of increased sensitivity to apoptosis in response to DNA damage [10]. In the developing mouse embryo, Mxd1, Mxi1 and Mxd4 are expressed in postmitotic cells while Mxd3, in contrast, is only present in mitotic cells during the S-phase of the cell cycle [11][12][13]. Mxd3 was identified as being upregulated during cerebellar GNP development [14], and in a previous study from our lab, MXD3 was shown to be expressed in response to SHH stimulation, and to be necessary and sufficient for cerebellar GNP proliferation [15], challenging the current paradigm that Mad proteins arrest proliferation and promote differentiation by antagonizing Myc function. In agreement with this challenge, Mxd3 has recently been reported to be upregulated in immature B cells in mouse spleen, where it negatively regulates B cell differentiation [16]. Furthermore, MXD3 is expressed in tumors derived from patched heterozygous (ptc +/2 ) mice (a model for medulloblastoma, [17]) but not in surrounding normal mature cerebellar tissue [15]. Finally, MXD3 appears to be expressed in human medulloblastomas [18], suggesting a role in medulloblastoma biogenesis, maintenance and/or proliferation.
To test this hypothesis, we studied the role of MXD3 in medulloblastoma using the human medulloblastoma cell line DAOY. Here, we present evidence that MXD3 promotes proliferation; however, sustained high levels of the protein negatively influence proliferation of medulloblastoma cells by the induction of programmed cell death. We report patterns of gene expression in MXD3-overexpressing cells, as well as candidate direct target genes, as a first effort to elucidate mechanisms of MXD3 function and its role in tumor cell proliferation and tumorigenesis pathways.

MXD3 is Expressed in Human Medulloblastomas
Since MXD3 promotes GNP proliferation during normal cerebellar development and it is abnormally expressed in cerebellar tumors in ptc +/2 mice [15], we reasoned that it might play an important role in the pathways that lead to uncontrolled proliferation in human medulloblastoma. Indeed, analysis of expression databases suggested that MXD3 is expressed in many human neoplasias, and in particular in tumors of the CNS, most significantly in glioblastomas and medulloblastomas [18], while it is absent in most human adult tissues. MXD3 is expressed in normal cerebellum during the GNP expansion. GNPs cease to proliferate shortly after birth and, during the first 2 years of life in humans, they differentiate as they migrate to form the internal granular layer (IGL). Accordingly, we observed very low levels of MXD3 in mature cerebellum. As shown in Fig. 1, MXD3 levels in mature cerebellum (where granular neurons are not proliferating) is 2 orders of magnitude lower than in developing cerebellum (where GNPs are proliferating). Interestingly, 8 out of 10 human medulloblastoma samples analyzed showed MXD3 levels significantly higher than normal mature cerebellum (p,0.05). Matched normal tissue was not available for analysis; nonetheless, since patient ages ranged from infants to adolescents ( Fig. 1), MXD3 levels in normal tissue is expected to be comparable to a mature cerebellum sample. Even more, 4 of the tumors showed levels of MXD3 significantly higher than those observed in fetal developing cerebellum. Taken together, these results indicate that abnormally high MXD3 expression is a characteristic of at least a subset of medulloblastomas.
To validate our experimental model used for the rest of this study, MXD3 expression in the DAOY cell line was analyzed. DAOY was chosen as it is the only medulloblastoma-derived cell line that can be grown in a monolayer, passed as single-cell suspension and easily transfected; these represent obvious technical advantages for cell counting, immunocytochemistry, overexpression and knock-down experiments. MXD3 cDNA was cloned from DAOY total RNA extracts; the full coding sequence obtained was 100% identical to the wild-type sequence (NM_031300). Immunoblot analysis failed to show any specific band using three different commercial anti-MXD3 antibodies. These results suggest that although the wild-type message is present, MXD3 protein may be expressed at very low levels, below the limit of detection for the antibodies available. Indeed, analysis by qRT-PCR showed that MXD3 transcript is present in DAOY cells at low levels (Fig. S1).

MXD3 in DAOY Cell Proliferation
To assess MXD3's role in the proliferation of DAOY cells, we performed knock-down and overexpression experiments. Knocking down the endogenous protein with two different specific siRNAs in DAOY cells resulted in a significant decrease in proliferation: total cell numbers were reduced to 75-80% after 48 hrs and 65-55% after 72 hrs when compared to untransfected cells ( Fig. 2A). In comparison, transfection of control siRNA (siRNAc) resulted in only 5-10% reduction, which is attributed to toxicity of the procedure. Thus, reduced levels of MXD3 result in reduced proliferation in this cell culture system. Knockdown efficiency was validated in DAOY cells cotransfected with HA-tagged MXD3 (HA-MXD3), since the endogenous protein cannot be detected in immunoblots with available antibodies. Both siRNAs reduced HA-MXD3 protein expression by at least 60% ( Fig. 2B; siRNA1, 6865%; siRNA2, 90610%) when compared to control siRNAc transfected cells. These data imply that MXD3 is necessary for medulloblastoma cell proliferation, in agreement with the previously reported effect during proliferation of non-tumor GNPs [15]. Furthermore, although MXD3 protein in DAOY cells could not be detected due to technical limitations of the available antibodies, the confirmation of a wild type mRNA and the observation of a phenotype upon MXD3 knock-down consistent with previous observations in cerebellar GNPs suggests not only the presence but also the biological relevance of this protein in DAOY cell proliferation.
As an experimental tool for genomic approaches to identify MXD3 target genes (see next section), DAOY cells were engineered to stably express HA-MXD3. A number of stable transfectants carrying either the HA-MXD3 construct or the empty vector were isolated under G418 selection. Several MXD3 (''M'') and control (''C'') clones were characterized in terms of MXD3 expression. MXD3 mRNA levels as measured by qRT-PCR were increased 30 to 1,700 times in MXD3 lines (M1-M5, Fig. S3A). Protein levels reflected the magnitude of message levels. Cellular fractionation followed by immunoblot analysis confirmed that, as expected, HA-MXD3 was present exclusively in the nuclear fraction (Fig. S3B). Furthermore, immunofluorescence on fixed, permeabilized cells showed HA-MXD3 localized to nuclei (Fig. S3C). Therefore, the ''M'' cell lines express HA-MXD3 in the appropriate subcellular compartment. ''C'' cell lines showed MXD3 mRNA levels indistinguishable from the parental cell line.
Surprisingly, when proliferation rates of the cell lines were analyzed, we observed decreased cell numbers in ''M'' lines when compared to parental or control lines in proliferation assays. Seven individual ''M'' clones and four individual ''C'' clones were analyzed to avoid non-specific phenotypes associated with random DNA insertion events inherent in the generation of stable cell lines (Fig. 3A). For statistical analysis, all data were pooled into an ''MXD3'' group and a ''Control'' group. As shown in Fig. 3B, the MXD3 group had a significantly lower proliferation rate when compared to the control group or the parental (p,0.001). This difference was evident as early as the beginning of the logarithmic proliferation phase (day 7, p,0.05) and reached a maximum towards the end of such phase, before the proliferation plateau. Indeed, on day 15, while ''C'' lines had a mean fold increase in cell number (relative to day 1) of 392.6, ''M'' lines showed a mean fold increase of only 225.4, representing a MXD3-mediated decrease of 42% (p,0.003) in cell proliferation.
To dissect the mechanism underlying this effect, cell cycle progression and apoptosis were studied in ''M'' and ''C'' cell lines. Again, to account for individual cell line variations, four of each cell lines were analyzed. As shown in Fig. 4A, ''M'' cell lines presented on average a small but significant increase in the G2 population, with no significant changes in the G1 or S populations. Therefore, persistent MXD3 overexpression seems to result in a longer G2 phase. Staining of the same eight cell lines for apoptosis indicated a significant increase in the apoptotic fraction in ''M'' lines ( Fig. 4B). Taken together, these results suggest that sustained high levels of MXD3 ultimately result in cell cycle arrest and trigger apoptosis, thus accounting for the observed reduced total number of cells in our experiments. Indeed, when these two factors were input into a derivative of the exponential growth model (N (t) = N 0 (1+ r/100) t/d ), the predicted growth curves showed a very good fit to the experimental data (Fig. 4C). Hence, G2 arrest and increased apoptosis seem to account for the experimentally determined difference between the proliferation rates of our ''M'' and ''C'' lines.

MXD3 Structure-function Analysis
MXD proteins consist of two functional modules. The bHLHLZ domain mediates heterodimerization with other factors (most importantly, Max) via the second helix and the leucine zipper, and DNA binding through the basic region [19]. The Sin-Interaction Domain (SID [7]) mediates recruitment of a repressor complex via interaction with mSin3 proteins. To dissect the domains required for MXD3 overexpression activity, deletion constructs were generated in which the SID domain (DSID) or the basic region of the bHLHZ domain (Dbasic) were removed. As shown in Figure 5, overexpression of HA-MXD3 by transient transfection resulted in decreased cell numbers after 48 hrs. Both deletions, however, completely abrogated this phenotype, suggesting that binding to DNA and interaction with, presumably, mSin3 cofactors are essential. Three amino acid residues within the basic region are responsible for direct, specific contact with DNA bases: H62, E66 and R70 [19][20][21] (MXD3 numbering, according to RefSeq accession NP_112590). To investigate further the requirement for DNA binding, the glutamic acid at position 66 was substituted with an aspartic acid residue. Although E to D is a conservative substitution, it has been shown that this substitution completely abolishes DNA binding of the bHLH protein PHO4 [22,23]. Accordingly, E66D substitution resulted in complete disruption of MXD3 activity in our experimental model (Fig. 5A). It has been suggested that the length of the side chain rather than the nature of the functional group is important for the interaction [22]. Accordingly, the E66Q mutation showed no difference in activity (p,0.05) compared to the wild type, while E66N resulted in a complete loss of activity. Importantly, these mutations have been shown in other bHLH proteins to result in binding or no binding, respectively, to DNA target sequences by gel-shift assays [22]. Finally, deletion of the C-terminal 14 amino acid residues of MXD3 (DC), a region highly conserved in all MXD proteins, did not have any significant effect (Fig. 5A). A schematic representation of the constructs used is presented in Fig. S2. Protein expression levels of the constructs used were found to be similar in all cases (Fig. 5B), and localization to the nucleus was confirmed (Fig. S3).

MXD3 Direct Transcriptional Targets and Downstream Genes
The results presented thus far suggest a role for MXD3 in medulloblastoma proliferation. The next goal was to shed light into the genetic programs that govern the phenotypes observed. According to the current paradigm, MXD3 is expected to work as a transcriptional repressor by binding promoter sequences as a heterodimer with MAX. However, its role as a transcriptional regulator is mostly inferred from experimental data available for other MXD family members. Furthermore, it is important to note that specific MXD3 gene targets have not yet been identified. We decided to address this question through a genomic approach by performing ChIP-chip on our stable cell lines with high expression of HA-MXD3. Taking advantage of the HA tag fused to the MXD3 coding sequence, commercial anti-HA antibodies previously used successfully in ChIP experiments [24] were employed. Since no MXD3 target genes have been identified to serve as controls, two types of positive controls were used. First, samples were immunoprecipitated in parallel with anti-RNA polymerase II antibodies. Enrichment of specific targets as confirmed by PCR indicated that the technique worked consistently under our experimental conditions (not shown). Secondly, MC7-E2F1 cells that stably express HA-tagged E2F1 [25] were immunoprecipitated in parallel with the same anti-HA antibody. PCR confirmed enrichment of E2F1 targets in the immunoprecipitate as previously reported [25] (data not shown), thereby validating the use of the anti-HA antibody in our experimental conditions.
To identify MXD3 binding sites in our ''M'' cell lines, we performed ChIP for hybridization to DNA microarrays. Differ-entially labeled immunoprecipitated and input DNA was cohybridized to Nimblegen HG18 RefSeq human promoter arrays. Data were analyzed using NimbleScan software and binding sites were ranked using Maxfour as described [26]. Using permissive criteria (Table S1), 788 genes whose promoter sequences were enriched in the immunoprecipitated material compared to the input sample were identified (Table S1). A subset of the highest ranking candidates was confirmed by PCR on an independent ChIP sample (12 ChIP hits, Fig. 6). Candidates were considered to be confirmed when PCR-amplification showed enrichment in the immunoprecipitated material and were absent in the control (IgG) immunoprecipitate. The list of validated MXD3 targets is presented in Table 1. Since members of the MXD family are known to bind E-boxes, we reasoned that these DNA motifs should be present in the promoters of these genes. Indeed, using PROMO [27,28], E-boxes were identified in all of the genes in Table 1 (Fig. 6). All E-box binding proteins bind the common consensus CANNTG, while target specificity for each transcription factor is thought to depend on the bases flanking that consensus sequence; based on this, algorithms have been defined to identify E-box ''subtypes'' [27,28]. Interestingly, MXD3 binding regions (as determined by ChIP-chip) in the promoters of the above genes overlap with predicted MYC binding sites (as suggested by the presence of MYC-type E-boxes, Fig. 6) in 11 out of 12 genes analyzed. MAX-type and MYCN-type E-boxes are found in MXD3 binding regions in 6 and 4 of the above genes, respectively. Notably, 6 genes (MYCL1, BMP3, IRF4, RUNX1T1, ENHO and C10orf120) only contain MYC-type Eboxes in MXD3 binding regions.
Next, changes in patterns of gene expression in DAOY cells upon expression of MXD3 were identified. RNA was extracted from four ''M'' lines, three ''C'' lines, and the parental DAOY cells, labeled and hybridized to Illumina human expression arrays. Data were pooled into groups for identification of genes that were up-regulated or down-regulated (.2-fold change) in MXD3 versus control lines. Genes that were differentially expressed in control lines when compared to the parental were excluded from further analysis regardless of the level of expression in MXD3 lines. Hence, only genes differentially expressed in all ''M'' lines but in none of the ''C'' lines compared to the parental were considered. This analysis identified 131 genes that were differentially expressed (Table S1). Of those, 47 were down-regulated and 84 were upregulated.
Genes that were directly bound by MXD3 and showed different levels of expression were submitted to gene ontology analysis using DAVID [29,30]. The most relevant functional categories included neurogenesis and nervous system development, reflecting previously reported roles of MXD3; interestingly, several categories were related to cancer biology, including apoptosis, cell differentiation and regulation of cell proliferation, suggesting a role for MXD3 in tumorigenesis in support of the hypothesis in this study. By far, most of the genes fell into the category of regulation of transcription, suggesting that MXD3 may be part of a transcriptional cascade. The complete results are presented in Table S2.

Discussion
Several lines of evidence suggest that MXD3 is an atypical member of the MAD family of transcription factors. MXD3 is transiently upregulated during GNP proliferation in the cerebellum in a pattern similar to that of Cyclin D2 [14]. MXD3 has been shown to be necessary and sufficient for SHH-mediated GNP proliferation [15], suggesting a role in the regulation of cell proliferation rather than differentiation or quiescence as demon- strated for MAD proteins. Recently, MXD3 has been shown to negatively regulate B cell differentiation in the spleen [16]. Besides MXD3 proliferative activity in normal development [14,15], several pieces of evidence have emerged from disease models that point in the same direction. In weaver mice, where cerebellar GNPs fail to exit the cell cycle and differentiate [31], MXD3 fails to undergo the physiological downregulation observed in wild type littermates [14]. Ptc +/2 mice, a model for medulloblastoma, show high expression of MXD3 in the tumor tissue but not in the adjacent normal postmitotic cerebellar tissue [15]. Database analysis suggests that MXD3 is overexpressed in human cancers, predominantly in medulloblastomas and glioblastomas [18]. Taken together, these findings prompted us to explore the role of MXD3 in human medulloblastomas. Probing a small cohort of tumor samples by qRT-PCR validated the database analysis, supporting our initial hypothesis that MXD3 might be relevant to medulloblastoma proliferation. The results presented in this report support the conclusion that MXD3 is required for proliferation of DAOY medulloblastoma cells; however, persistent and/or high expression of MXD3 ultimately leads to apoptosis and cell death.
Two main aims in this study were to (i) analyze the role of MXD3 in proliferation of tumor cells and (ii) identify potential target genes and patterns of gene expression induced by this transcription factor as a first approach to understand the molecular mechanisms by which MXD3 exerts its function. Our experimental model consisted of knock-down or overexpression of MXD3 in DAOY cells. Two different siRNAs against MXD3 resulted in decreased proliferation, in agreement with its role in GNP proliferation during normal cerebellar development [15].
Surprisingly, overexpression of MXD3 resulted in reduced cell numbers, in apparent disagreement with the results discussed above and with previous reports by our group and others [15,16]. These seemingly contradictory results likely represent distinct phases of MXD3 activity, namely proliferation followed by apoptosis. Furthermore, this might also reflect why MXD3 levels are so low in the medulloblastoma-derived DAOY cell line. Several possibilities might explain these findings. First, MXD3 might trigger a proliferative response which in turn triggers cellular homeostatic mechanisms to control proliferation. This possibility is in agreement with previous results from our laboratory which, in a normal cerebellar development model, reported MXD3 proliferative activity was followed by induction of apoptosis [15]. Remarkably, although most cells died after an initial burst of proliferation in MXD3-transfected GNPs, those that survived and continued to express MXD3 were found to be mitotic [15]. These data indicate that MXD3 direct transcriptional regulation could account for both phenotypes described in this paper, namely proliferation (upregulation of putative direct targets IRF4, FBL, HTR1B) and growth arrest and apoptosis (upregulation of RUNX1T1 and BMP3; downregulation of USP2 and MYCL1) as shown in Table 1. In this regard, MYC proliferative and apoptotic activities are well documented (for review, see [32]). Recently, ARF has been reported as a switch for c-MYC, such that during physiological low-ARF conditions c-MYC drives expres-sion of canonical targets that result in proliferation, while high-ARF (induced by oncogenic stress) directly inhibits canonical c-MYC target expression and induces non-canonical targets leading to apoptosis [33]. Thus, it is conceivable that such a molecular switch exists for MXD3, representing a fail-safe mechanism to prevent MXD3 driven hyper-proliferation. In this sense, yet uncharacterized binding partners might act as switches that define the MXD3 mediated response.
The data presented here highlights several aspects of MXD3 atypical function, consistent with an emerging body of evidence regarding the MYC/MAD proteins. For example, MNT (a Mad family protein) has been found to bind DNA targets that are not shared by either MYC or MAX [34], suggesting it participates in alternative mechanisms of gene regulation. Likewise, MAXindependent functions of MYC have been reported [35]. Interestingly, sequence analysis indicates that most MXD3 binding sites at promoter regions lie within or very close to MYC-type Eboxes, suggesting the intriguing possibility that it may bind target genes as a heterodimer with MYC. Indeed, MXD3 is predicted to interact with MYC proteins to a similar extent as the known binding partner MAX [18]. As another example, MXD3 has  Figure S2. MXD3-transfected cells showed 50% reduction of proliferation compared with vector-only (pHM6). Transfection of DC and E66Q constructs resulted in similar reduction in proliferation as full-length MXD3. Transfection of the DSID, Dbasic, E66D and E66N constructs had no effect on proliferation compared to control (ns, not significant; asterisks, p,0.001; n = 4, one-way ANOVA with Bonferroni's multiple comparison test). (B) Immunoblot analysis of total protein extracts from DAOY cells transfected with the MXD3 constructs in (B). All constructs used expressed the expected proteins at similar levels. Bottom panel, anti-HA; top panel, anti-b-tubulin used as a load control. Molecular mass markers are indicated on the right. doi:10.1371/journal.pone.0038508.g005 recently been shown to bind directly to the Id2 promoter and induce (rather than repress) transcription of the gene by histone acetylation (rather than deacetylation) via recruitment of HATs (rather than HDACs) [16]. MXD3 binds the Id2 promoter at an E-box, which is typical for other bHLH proteins. However, it is not clear whether the atypical recruitment of HATs is mediated by the SID domain. In line with this report, the majority of MXD3 targets that we identified are upregulated, in contrast with the classical transcriptional repression activity of other family members.
To our knowledge, this is the first report on a genome-wide approach to identify direct MXD3 targets. Based on ChIP-chip experiments, we present a list of candidate genes that seem to be directly bound by MXD3 to modulate transcription in DAOY medulloblastoma cells. Our results provide a significant contribution for future experimentation aimed at elucidating the role of MXD3 in tumorigenesis as well as in normal development.
The classical switch model of the MYC/MAX/MXD network, in which competition between MYC-MAX and MXD-MAX heterodimers for shared DNA binding sites controls proliferation or differentiation, needs to be expanded to accommodate these emerging challenges. MXD3's dual effect on medulloblastoma cell proliferation may be another example of this increasingly complex regulation.

Proliferation Assays
Proliferation of DAOY and MXD3 stable cell lines was assessed by plating DAOY cells (American Tissue Culture Collection) in 6well plates at 1,500-3,000 cells/well. Medium was changed daily and cells counted for 10-15 days. Proliferation assays with MXD3 constructs or siRNAs were performed by transiently transfecting 10 5 cells/well in 6-well plates; cells were counted 48 hrs after transfection. In all cases, attached cells were trypsinized and counted using a Coulter Counter Z1 instrument (Beckman Coulter). Two replicates were counted for each data point, in at least three independent experiments.

RNA Isolation and Expression Profiling
Total RNA was extracted from DAOY or MXD3 stable cell lines (0.5-1610 6 cells) using the RNeasy kit (Qiagen). For extraction of RNA from tumors, tissue was first homogenized in Trizol (Invitrogen) and after chloroform extraction, RNA was purified with RNeasy kits. RNA quality was assessed with a Bioanalyzer instrument (Agilent).
For expression analysis, RNA was amplified and biotinylated using Illumina TotalPrep RNA amplification kit (Ambion).

Chromatin Immunoprecipitation and DNA Promoter Arrays
Chromatin immunoprecipitation (ChIP) assays were performed following published protocols [36]. Briefly, 10 7 cells per assay were fixed with 1% formaldehyde for 10 min at room temperature (RT), followed by addition of glycine (125 mM final concentration) for 5 min. Fixed cells were collected with a cell-scraper, washed once with phosphate buffered saline (PBS) and either used immediately or snap-frozen and stored at 280uC. Fixed cells were homogenized in cell lysis buffer (5 mM Pipes pH 8.0, 85 mM KCl, 1% NP40) and pelleted nuclei were lysed in nuclear lysis buffer (50 mM Tris pH 8.0, 10 mM EDTA, 1% SDS). DNA was fragmented to ,200 bp and immunoprecipitated with mouse anti-HA (Covance, cat# MMS-101R) and secondary rabbit antimouse IgG (MP Biomedicals, cat# 55436) antibodies. Nonspecific rabbit IgG (Alpha Diagnostics, cat # 20009-5) was used as a negative control and mouse anti-RNA polymerase II 8WG16 IgG2a (Covance, cat # MMS-126R) as a positive control. Crosslinks were reversed in eluted immunoprecipitates and DNA was purified using QIAquick PCR purification kit (Qiagen). Precipitated DNA from ChIP with smaller starting material (4-30 mg/ assay) was measured by Quant-iT PicoGreen assay (Invitrogen). Amplicons were prepared as previously described using the GenomePlex WGA4 kit (Sigma). DNA fragments were labeled by Klenow extension using 9-mer Cy3-primers for total (unenriched) input and Cy5-primers for immunoprecipitated material. Briefly, 1 mg of DNA and primers (1 O.D./42 ml) were denatured at 98uC for 10 min and incubated with Klenow (50 U/ml) and 1 mM dNTP at 37uC for 2 hrs. After stopping the reaction with 50 mM EDTA, DNA was precipitated with isopropanol and resuspended in H 2 O. Thirteen mg of the test and reference samples were combined, vacuum-dried, resuspended in array hybridization buffer (Nimblegen) and hybridized to Human HG18 RefSeq 385K promoter arrays (Nimblegen) for 16-20 hrs according to manufacturer's instructions in a MAUI station (BioMicro) at 42uC.

Cell Cycle and Apoptosis
Cell subpopulations at the different phases of the cell cycle were quantified by DNA content measured by flow cytometry. Briefly, 10 6 -10 7 cells per sample were trypsinized, washed with PBS, resuspended in 0.5 ml of PBS and added to 4.5 ml 70% ethanol on ice. After 2 hrs, the fixative was removed, the cells were washed with PBS and resuspended in 1 ml of 20 mg/ml propidium iodide, 0.1% Triton X-100, 200 mg/ml RNase A in PBS. After 30 min at RT, data (50,000 events/sample) were acquired in a Becton-Dickinson FacScan instrument and fitted with FlowJo software (Tree Star, Inc.) using the Watson Pragmatic model.
For apoptosis analysis, cells were gently detached with 0.5 mM EDTA/PBS for 30 min at 37uC, washed with Hank's Buffered Salt Solution (HBSS; Invitrogen) and resuspended in the same medium at 1.5610 6 cells/ml. One ml of the cell suspension was stained with Vybrant DyeCycle/Sytox AADvance kit (Invitrogen) according to the manufacturer's instructions. Data (10,000 events/ sample) were acquired in a Becton-Dickinson LSRII instrument and analyzed with FlowJo software.

DNA Constructs and siRNA
Full-length human MXD3 coding sequence (including stop codon) was amplified by RT-PCR from total RNA extracted from DAOY cells and ligated into pHM6 (Roche) to create an Nterminal-HA tagged expression vector. Single amino acid substitutions (E66K and E66D) were generated in the same vector backbone using a QuikChangeII Site-Directed Mutagenesis Kit  [49] and medulloblastomas [50] C10orf120 NM_001010912 3.1 3.53 n/a unknown *The complete list is presented in Table S1.

Cell Culture and Transfection
DAOY cells were grown in MEM medium supplemented with 10% fetal bovine serum (FBS), 1 mM sodium pyruvate and penicillin/streptomycin. For transient expression, cells were transfected at 90-95% confluence using FuGene HD (Roche) at a 5:2 ratio (ml reagent:mg DNA) or Lipofectamine 2000 (Invitrogen) at a 3:1 ratio, according to manufacturer's instructions. Two mg of DNA were used per well in 6-well plates. siRNA transfections were carried out using 2.5 ml Lipofectamine-2000 (Invitrogen) and 100 pmol siRNA per well in 6-well plates with cells at 80-90% confluency.
For the generation of stable lines, cells were transfected as described above in 100 mm dishes. After 48 hrs, G418 (Cellgro) was added to a final concentration of 800 mg/ml and colonies were picked and expanded after 2-3 weeks. All stable cell lines used in the experiments described here were maintained in culture for more than 40 passages with no obvious change in phenotype.

Quantitative Real Time PCR (qRT-PCR)
Reverse transcription reactions were performed using the High Capacity RNA to cDNA Kit (Applied Biosystems). RNA from tumors and stable cell lines was obtained as described above. Frozen tumor specimens were obtained from the Human Cooperative Tissue Network. Normal human developing and mature cerebellum RNA was purchased from BioChain. Quantification of MXD3 mRNA was performed using the StepOne Plus Real-Time PCR System (Applied Biosystems). Validated assays for human MXD3 and endogenous control 18S ribosomal RNA were used (Applied Biosystems). Experiments were performed in triplicate for each data point. Relative expression of MXD3 mRNA level was calculated by the comparative C T method.
Immunofluorescence was performed essentially as described [15]. Paraformaldehyde-fixed cells were permeabilized, blocked and incubated with anti-HA (1:1000) at 4uC overnight. Cellular fractionation was performed by incubating cells in cell lysis buffer for 20 min at 4uC followed by homogenization in a Potter-Elvehjem homogenizer. After centrifugation, the supernatant was reserved as the cytoplasmic fraction; the pellet was washed and pelleted nuclei were lysed in nuclear lysis buffer (50 mM Tris pH = 8.0, 10 mM EDTA, 1% SDS).
Protein concentration was determined using a micro BCA protein assay kit (Pierce).
Statistical analysis was performed with Prism 5 (Graphpad Prism Software, Inc). . HA-MXD3 was absent in the representative control line and was only detected in nuclear fractions of the M lines, but was absent from cytoplasmic fractions. Cytoplasmic and nuclear extracts were confirmed by detection of GAPDH and lamin respectively; these markers were also used as load controls. Anti-lamin antibody detects a double band; anti-GAPDH antibody detects a nonspecific band of lower molecular weight in the nuclear fraction in addition to a specific band in the cytoplasmic fraction; the purity of the cytoplasmic fraction was further confirmed with anti-tubulin antibodies (not shown). (C) Nuclear localization of HA-MXD3 was also confirmed by immunostaining with anti-HA antibody (red) and counterstaining with DAPI for nuclei (blue). Images correspond to M1 and C1 cell lines and are representative of all other lines within each group. Scale bar, 100 mm.  Figure S3 Mutagenized MXD3 protein localizes to the nucleus. All MXD3 mutant proteins localized to the nucleus. Confocal images of DAOY cells transfected as indicated on the left; immunodetection of the expressed protein was performed with anti-HA (first column, blue) and anti-MXD3 (green). As evidenced in merged images, HA-MXD3 and mutated forms localized exclusively to the nucleus. Scale bar, 10 mm. (TIF) Table S1 MXD3 target genes and expression profiles. *Genes in this list were selected from ChIP-chip results based on the following criteria: peak values .3 in at least one biological replicate and/or peak values .2 in both biological replicates. Mxd3 targets that were validated by PCR with an independent ChIP sample are highlighted in yellow. "ChIP 1 and ChIP 2 represent the peak values of two biologically independent experiments and are calculated from promoter array signals using the Max4 algorithm. Significant peak values are highlighted in blue (.3) or light blue (.2). &Expression profile values represent fold increase in mRNA abundance calculated as in Table 1. Based on the criteria described in the Results section, upregulated genes are highlighted in red and downregulated genes are highlighted in green. N/A indicates genes not present on the array platform used. (XLS)