The lack of detectable alteration in DNA repair caused by SNF5 loss together with the resulting genomically stable cancers suggest a central role for epigenetically driven changes in gene expression in the etiology of RT.^2,22 To search for targets essential for the tumor suppressor activity of SNF5, we compared genome-wide expression data that we had previously generated from primary RT of the brain (also known as atypical teratoid/rhabdoid tumor or AT/RT) to that from primary medulloblastomas that we had generated. We also included other data sets from the Gene Expression Omnibus representing cancer types not known to be driven by SNF5 inactivation, including Wilms tumor, neuroblastoma, Ewing sarcoma, clear cell sarcoma of the kidney, prostate cancer, ovarian cancer and lung cancer. As an initial test of this approach, we utilized comparative marker selection, an algorithm that identifies the genes whose differential regulation is most closely correlated with a particular class of samples, to query expression of SNF5 (http://genepattern.broadinstitute.org/).⁴¹ As expected, low levels of SNF5 expression were strongly associated with RT, and SNF5 ranked as one of the 10 (out of 4498) most downregulated genes in RT compared with the other pediatric and adult cancers used in the analysis (Table 1).

In order to identify potential targets of SNF5, we next used the GeneNeighbor algorithm (GenePattern) to identify genes whose expression levels were most similar to those of SNF5 across both SNF5-deficient and SNF5-expressing human cancers.⁴¹ We reasoned that such genes would represent candidates for novel tumor suppressor targets that are aberrantly silenced following SNF5 inactivation. The nearest gene neighbor of SNF5 was BIN1, a gene with known tumor suppressor properties (cosine distance score = 0.008, Fig. 1A). Comparative marker selection also identified BIN1 as the ninth most downregulated gene in RT, whereas downregulation of BIN1 was not strongly correlated with any other tumor type analyzed (Table 1). These analyses indicate that, like SNF5, BIN1 is specifically downregulated in RT (Table 1 and Fig. 1B). Furthermore, the expression data suggest downregulation of all BIN1 isoforms, as the BIN1 probes on the array platforms recognize all isoforms (NetAffx Analysis Center).

We next sought to validate these findings in an independent data set by evaluating BIN1 expression in SNF5-deficient tumors originating in a different organ. We utilized comparative marker selection to analyze recently published gene expression data from pediatric kidney tumors that included RT of the kidney, cellular mesoblastic nephroma and samples of Wilms tumor and clear cell sarcoma independent from those analyzed in Figures 1A and B.⁴² BIN1 was in the top 1% of genes downregulated in RT of the kidney compared with clear cell sarcoma, Wilms tumor and cellular mesoblastic nephroma (Table 1). Consequently, downregulation of BIN1 in RT is independent of tissue of origin.

To evaluate the relationship between SNF5 and BIN1 in a third cancer system, we examined the expression of Bin1 in murine CD8⁺ lymphomas driven by deletion of Snf5.¹⁸ Bin1 expression was strongly decreased in the lymphomas compared with wild-type CD8⁺ cells (Fig. 1D), indicating that loss of BIN1 expression is a conserved event in SNF5-deficient tumors.

We next investigated the mechanism underlying altered expression of BIN1 to determine whether the downregulation was directly due to SNF5 loss. We retrovirally transduced the SNF5-deficient RT cell line BT16 with either FLAG-tagged SNF5 or vector control. BIN1 was reproducibly upregulated upon SNF5 reintroduction (Fig. 2A). As a positive control, we analyzed expression of p16^INK4A, a gene that has been previously shown to be upregulated upon SNF5 reintroduction into RT cell lines.²⁸ To further evaluate the relationship between SNF5 and BIN1, we next tested whether expression of Bin1 is dependent upon SNF5 in normal mouse cells. Deletion of Snf5 via a retrovirus expressing Cre recombinase in Snf5^flox/flox primary murine embryonic fibroblasts resulted in a 1.4-fold decrease (p = 0.0134) in Bin1 mRNA levels (Fig. 2B), indicating that Bin1 expression is also dependent on SNF5 in primary cells.

To more specifically evaluate the mechanism of BIN1 regulation, we evaluated whether loss of SNF5 resulted in aberrant silencing of BIN1 due to altered recruitment of SWI/SNF to the BIN1 promoter. We reintroduced SNF5 into BT16 cells as above and performed chromatin immunoprecipitation analysis of the SWI/SNF ATPase BRG1 at the BIN1 promoter. We observed specific SNF5-dependent enrichment of BRG1 (Fig. 3), suggesting that SNF5 facilitates recruitment of the SWI/SNF complex to the BIN1 promoter to drive BIN1 expression.

9 and isoform 10, into BT16 cells (Fig. 4A and B). Given the potential for negative selection against BIN1 expression in RT cells, we expressed the isoforms as fusions with AcGFP and DsRed to both label infected cells and ensure that fluorescence levels correlated with BIN1 expression. Transduction of the BIN1 isoforms into U2OS, an osteosarcoma cell line that expresses endogenous BIN1 (Fig. 4C), had no effect on proliferation (Fig. 4D). In contrast, transduction of BIN1 into BT16 cells blocked cell proliferation (Fig. 4E).

The RT and medulloblastoma microarray data sets were previously described in reference 22. Briefly, snap-frozen primary tumor samples were collected from brain tumor resections collected at Children’s Hospital Boston, Children’s Hospital of Philadelphia and St. Jude’s Research Hospital for Children under approval from their respective Institutional Review Boards. To identify the genes with expression patterns closest to SNF5 across a panel of human tumors, we compared expression data from 9 RT, 4 RT cell lines and 19 medulloblastomas with samples from the following data sets downloaded from the Gene Expression Omnibus: 5 Ewing sarcomas and 5 neuroblastomas (GSE1825), 12 Wilms Tumors and 14 clear cell sarcomas of the kidney (GSE2712), 10 prostate cancers (GSE6919), 9 ovarian cancers (GSE7463), 10 lung cancers (GSE7670) and 12 breast cancers (GSE1456). For each individual data set, the CollapseData set tool from the Gene Set Enrichment Analysis desktop application (Broad Institute) was used to create a new data set using single gene symbols as the identifier for the maximum probe value. The data sets were then merged using an R programming script that only retained genes that were common among the different Affymetrix platforms (U133A 2.0, U133A and U95Av2) and normalized the samples by replacing the raw expression value for each gene with its rank in each individual microarray sample, from lowest to highest. Comparative marker selection was performed using a two-sided t-test to assess differential expression with zero permutations (because several tumor classes had fewer than 10 samples) of the phenotype test one vs. all. The nearest neighbors of SNF5 (genes whose expression values followed similar patterns across the microarray samples) were identified in the merged, normalized data set using the GeneNeighbor algorithm on the GenePattern website (genepattern.broadinstitute.org). The raw expression data from Gadd et al.⁴² was downloaded from GEO (GSE11482), normalized using robust multichip average on the GenePattern website and processed using comparative marker selection (two-sided t-test with 1,000 permutations).⁴¹

Whole-cell lysates were prepared by incubating cells on ice in RIPA lysis buffer for 20 min followed by centrifugation at 15,000x g for 10 min. Supernatant was collected and protein concentration determined using Bradford reagent (Bio‑Rad). Proteins were separated using SDS-PAGE and transferred to PVDF membranes (Millipore). The following antibodies were used: SNF5 (Bethyl; A301-087A), BIN1/Amphiphysin II (Santa Cruz Biotechnology; 2F11 clone, sc-23918), actin-HRP (Abcam; 20272-200), DsRed-monomer (Clontech; 632496), AcGFP (Clontech, 632375) and HRP conjugated secondary antibodies (Jackson Immunoresearch).

Murine CD8⁺ T cells were isolated from Snf5 wild-type spleens (n = 3) or from Snf5-deficient CD8⁺ lymphomas (n = 3) from the same line of 129-BL6 mice using the MACS CD8⁺ T-cell isolation kit (Miltenyi Biotec). Total RNA was extracted using Trizol reagent (Invitrogen) and reverse-transcribed by the Reverse Transcription System (Promega). Quantitative real-time PCR was performed using the iQ SYBR Green Supermix and iCycler (Bio-Rad). For analysis, gene expression was normalized to Rps8. mBin1 Fwd: TTC GGA CCT ATC TGG CTT CTG, Rev: CCT CCT GAA GAC ACT CAC TCA; mSnf5 Fwd: ACA GTG GAG ATT GCC ATC CG, Rev: ACG CCT CAT TCG CCT TGT GT; mRps8 Fwd: GCG AAA ACC CTA CCA CAA GA, Rev: GAA AAG TTC CCC ACA TCC AA.

Primary WT and Snf5-conditional MEFs were cultured and harvested as described previously and infected two times with pBabe-puro^r-Cre retroviral supernatant at 4 h intervals (with 4 μg hexadimethrine bromide per mL retrovirus).^22,29 Cells were stably selected in medium containing puromycin (2.5 μg/mL) 48 h after infection. Total RNA was isolated using Trizol reagent (Invitrogen) and applied to Affymetrix 430A 2.0 arrays. CEL files were preprocessed using ExpressionFileCreator and the robust multichip average algorithm on GenePattern and scatterplots generated using GraphPad Prism software.

BT16 cells were cultured in DMEM supplemented with 10% Fetalplex (Gemini Bio-Products) and 1% penicillin/streptomycin. RT cell lines were infected two times with either pBABE-puro^r or pBABE-puro^r-FLAG-hSNF5 (a kind gift of R. Kingston) retroviral supernatant at 3 h intervals with 4 μg hexadimethrine bromide per mL retrovirus. Cells were stably selected in medium containing puromycin (1.0–1 .5 μg/ mL) 48 h after infection. Western blots and quantitative real-time PCR were performed as described above. For analysis, gene expression was normalized to GAPDH (Fwd: AAT CCC ATC ACC ATC TTC, Rev: ATG AGT CCT TCC ACG ATA CC). The BIN1 primers were designed to recognize all BIN1 isoforms. BIN1 Fwd: GCA GTG CGT CCA GAA TTT CAA, BIN1 Rev: CCA ATC GGG CTC ATA CAC CT. p16^INK4A Fwd: GAA GGT CCC TCA GAC ATC CCC, p16^INK4A Rev: CCC TGT AGG ACC TTC GGT GAC.

Chromatin immunoprecipitation was performed using the SimpleChIP Enzymatic Chromatin IP Kit (Cell Signaling) supplemented with Protein G Sepharose Fast Flow (GE Healthcare). Approximately 1.5 x 10⁷ cells were used per IP. Chromatin was digested using 3.5 μL MNase per 4.5 x 10⁷ cells and sonicated 3x for 10 sec pulses with the Fisher Sonic Dismembrator Model 100 at power setting 3. BRG1 was immunoprecipitated with 10 μL α-SNF2β/BRG1 (Upstate/Millipore, 07–478, lot numbers DAM1647877, DAM1718075, R0706B0022). Quantitative real-time PCR was performed as described above. IGX1A negative control primers (SABiosciences) were used to demonstrate specific target binding enrichment. BIN1 ChIP primers: Fwd-CTT CGC CGC ACT TTC TCT TTG AT, Rev-AGT AAT CTC CTG CAA CCC AGG AA.

BIN1 isoform 9 and 10 cDNA was cut from the EcoRI site of pcDNA3.1 (provided by G. Prendergast) and, respectively, cloned into the EcoRI site of pRetroQ-AcGFP‑C1 and pRetroQ-DsRed-monomer-C1 (Clontech), which both contain a hybrid CMV/MSV promoter to drive gene expression. Retroviral supernatant was harvested after co-transfection of 293T cells with a pRetroQ-BIN1 construct and the pCL-10A1 retroviral packaging vector using FuGeneHD transfection reagent (Roche). To reintroduce both BIN1 isoforms, cells were first infected as above with pRetroQ-DsRed-BIN1 isoform10, stably selected in medium containing puromycin (1.0–1.5 μg/mL) then infected with pRetroQ-AcGFP-BIN1 isoform9. BT16 and U2OS cells were cultured as above. The top 50% of cells expressing both DsRed and AcGFP were sorted using the FACSAria II cell sorting system (BD Biosciences) and plated 2–5 x 10⁴ per well. Cells were counted every 3 d after sorting for 15 d total with the Countess Automated Cell Counter (Invitrogen).