SWI/SNF regulates a transcriptional program that induces senescence to prevent liver cancer

Here, Tordella et. al identified senescence regulators relevant to cancer by screening an shRNA library targeting genes deleted in hepatocellular carcinoma (HCC). They show that knockdown of the SWI/SNF component ARID1B prevents oncogene-induced senescence and cooperates with RAS to induce liver tumors, and their results provide new insights into the mechanisms by which epigenetic regulators can affect tumor progression.

sections from mice injected with the indicated constructs and co-stained with antibodies against Nras and Ki67 are shown. DAPI was used to visualize the nuclei. Inset, higher magnification views. Scale bars, 10 μm.   (vec) or shRNAs targeting p53, p21 CIP1 or p16 INK4a . Graphs represent mean ± SD from n=4 (A) and n=2 (C). *** p < 0.001 by two-tailed Student t-test. Figure S5. A focused shRNA screen identifies novel ARID1B effectors controlling senescence.
ARID1B regulates senescence to prevent cancer Tordella et.al. 4 (A) Genes selected to generate the focused screen shRNA library. The x axis shows the log 2 FC between IMR90 ER:RAS + 4OHT and -4OHT (Ras induction), while the y axis the log 2 FC between IMR90 ER:RAS vector and shARID1B, both 4OHT treated (ARID1B dependency). Genes targeted in the shRNA library are marked in colours other than grey [(x > 1; y < 0) U (x > 0; y < 0.4)]. Blue; genes with a ratio value between 2 and 0.5 (induced by RAS, dependent on ARID1B). Orange; genes with a ratio value > 2 (induced by Ras, independently of ARID1B). Yellow; genes with a ratio value < 0.5 (induced by ARID1B only).
(B) Timeline of the screen is shown. The screen was performed in duplicate. (C) Enrichment for Ras-induced ARID1B-dependent genes in the screen. Comparison between the distribution of genes belonging to the three groups in the library (left, 43%, 44% and 13%, respectively) and in the top hits of the screen (right, Fisher's combined p value < 0.001, 88% of genes induced by Ras and dependent on ARID1B; 12% of genes induced by Ras independently of ARID1B). (D) IPA upstream regulator analysis of the screen top hits (p < 0.05) reveals activation (z-score) of SMARCA4, TP53 and ROS pathways. (E) shRNA enrichment analysis. The x axis shows the log 2 median FC of the multiple shRNAs per gene in both replicates. The y axis shows the -log 10 of the Fisher's combined p value calculated for each gene. Significantly enriched shRNAs (at p < 0.001) are marked in blue. Among those, genes selected for further analysis are marked in red, with CDKN1A used as internal control. (F) qRT-PCR analysis of the expression of the indicated genes in IMR90 transduced with either an empty vector or cDNA of ARID1A, ARID1B or BRG1. (G) ENCODE database analysis of the co-localization of marks for transcription initiation (H3K4me3, H3K27Ac and Dnase hypersensitive sites) with SMARCC1 (coding for SWI/SNF core subunit BAF155) ChIP-seq at the promoter region of the indicated genes in HeLa cells.

Figure S9. Nucleotide metabolism is a liability of ARID1B-deficient cells.
(A) Scheme indicating the role of ARID1B signalling in senescence and a suggested therapeutic approach for ARID1B-mutated tumours. ARID1B regulates a complex senescence response upon oncogene activation, leading to induction of cell cycle regulators p16 and p12, but also of a set of newly discovered targets (ENTPD7, SLC31A2 and NDST2), which contribute to p53 activation via generation of DNA damage and ROS. The blue arrows indicate a pro-senescence therapeutic approach for ARID1B mutated tumours using drugs (DON, Gemcitabine) which mimic the function of ARID1B-downstream mediators, such as ENTPD7. (DDR, DNA-damage response; ROS, reactive oxygen species). (B) Cell growth assay (crystal violet) and quantification of IMR90 ER:RAS cells cotransduced with an empty vector or ENTPD7 cDNA and shRNAs targeting ARID1B, p53 or control (+ 4OHT treatment). (C) Cell growth assay (crystal violet) and quantification of stable ARID1B-knockdown, p53-knockdown or control IMR90 ER:RAS cells treated with 4OHT, followed by Gemcitabine (5 µM) or DMSO. The graphs represent mean ± SD from n=2. *** p < 0.001 by two-tailed Student t-test. Table S1. shRNA target sequences * h, human; m, mouse.

Construct
Target sequence  Table S3. Primers used for RT-qPCR.  Retroviral and lentiviral infection were performed as previously described ; Barradas et al. 2009).
Plasmids pLNC-ER:RAS has been described previously (Acosta et al. 2008 their pRRL backbone and subcloned into the transposon-based Nras expression plasmid CaNIGmirE-5'. The pBabe ENTPD7 plasmid was obtained from GenScript. ENTPD7 cDNA was subsequently sub-cloned (BamHI, SalI) into an N-terminal HA-tagged pBabe empty vector. All expression plasmids used in the study are listed in Table S2.

shRNA libraries
To screen genes deleted in HCC with a role in regulating senescence, we took advantage of a previously described shRNA library. This shRNA library consists of 631 miR30-based shRNAs, targeting 301 mouse orthologs of genes deleted in HCC cloned in a murine stem cell virus (MSCV)-SV40-GFP vector (Zender et al. 2008).
To identify ARID1B-targets controlling senescence, we designed a library consisting of 1,504 miRE-based shRNAs, targeting 255 human genes (most genes were targeted by 6 shRNAs each). To select the targeting sequences, we used sensor-based shRNA predictions (Fellmann et al. 2011;Fellmann et al. 2013). Oligonucleotides (136 nt-long oligonucleotides containing 97 nt coding for the respective shRNAs, an EcoRI cloning site and a 20 nt adaptor site) were synthesized on an oligonucleotide array (MYcroarray) (Cleary et al. 2004) and pooled for cloning (Cleary et al. 2004;Fellmann et al. 2011). The pool of oligonucleotides was PCR-amplified and cloned through XhoI/EcoRI sites into the pRRL-SFFV-GFP-miRE-PGK-Puro vector (Fellmann et al. 2013).

Preparation of libraries for determining shRNA enrichment in the screens
DNA was extracted from infected pools using DNeasy Blood and Tissue Kit (Qiagen) as described by the manufacturer. 1µg of extracted DNA was used for PCR with forward and reverse primers ligated to Solexa adaptors. The forward primer used for the PCR also contains a 3-nucleotide barcode, which is unique and corresponds to the time point at which the infected pool was collected. PCR-products were extracted from gels; each barcoded sample was extracted as an individual product using a MinElute Gel Extraction Kit (Qiagen) and quantified using the Qubit 2.0 Fluorometer. The individual barcoded PCR-products 6pM of the Solexa library was used to create clusters and sequenced on a Genome Analyser IIx.

Statistical analysis of the shRNA screens
Fasta files produced from the sequencing runs were processed and sequences were demultiplexed with CASAVA 1.7. The reverse complement of each read was aligned to the custom shRNA libraries using ELAND. Candidates were ranked using Fisher's combined pvalue and edgeR. The Fisher's combined test allows p-values across independent data sets to be combined, bearing upon the same overall hypothesis (Fisher 1925).

Microarray and RNA-Sequencing
For microarray experiments, cRNA was hybridized to Human Gene 1.0 ST arrays (Affymetrix) following manufacturer's instructions. Three biological replicates were performed for each condition. Microarray data processing and analysis was carried out at EMBL-GeneCore (Heidelberg, Germany). Microarray data was normalized using Robust Multichip Average (RMA) method available in "oligo" Bioconductor package and significant differentially expressed probesets were identified using Limma (Ritchie et al. 2015) with Benjamini-Hochberg corrected p-value < 0.05. "--library-type fr-firststrand --b2-very-sensitive --b2-L 25" and using known transcripts annotation from ensembl gene v72. Number of reads counts on exons were summarised using HTSeq v0.5.3p9 with "--stranded=reverse" option and differentially expressed genes were identified using DESeq2 (Love et al. 2014). Genes were ranked by fold change and Gene Set Enrichment Analysis was performed using GSEA v2.07 (Broad Institute) preranked module. The activity of signalling pathways was identified using Ingenuity Pathways

RNA-
Analysis software (IPA, QIAGEN). Data was analysed using IPA's Upstream Regulator analysis (P < 0.05) to predict activation or inhibition (z-score) of transcriptional regulators.
This analysis is based on prior knowledge of expected effects between transcriptional regulators and their targets stored in the Ingenuity® Knowledge Base.

Immunoblot
Protein extracts from cell lines were processed and analysed as previously described (Barradas et al. 2009). The antibodies used are listed in Table S4.

ARID1B and ENTPD7 expression correlation analysis
TCGA Pan-cancer (PANCAN) expression values were selected from the "gene expression" dataset, measured using the IlluminaHiSeq technology. Correlation was calculated per cohort using Pearson test.  ARID1B mut, n=11; SWI/SNF wt, n=132; SWI/SNF mut, n=63. Differential expression analysis was done using Student's t-test.

Immunofluorescence for high content analysis
IF was performed as previously described ) using the antibodies listed in Investigator software (v 3.2; GE Healthcare), as described elsewhere Barradas et al. 2009;Bishop et al. 2010;Acosta et al. 2013). Briefly, DAPI staining of the nuclei was used to identify cells. The nuclei were segmented using top-hat segmentation, specifying a minimum nucleus area of 100 μm 2 . To define the cell area, a collar segmentation approach was used with a border of 1 μm around DAPI staining or alternatively, multiscale top-hat was used to detect cytoplasmic intensity for a given staining.
Each cell was assigned a nuclear intensity value (and cell intensity value when applicable) for the specific protein being studied. A histogram of the intensity values of all the cells in a sample was produced and this was used to set a threshold filter to determine positive and negative expressing cells. Alternatively, normalized intensity values for a given staining were obtained by subtracting the intensity of the secondary antibody alone (background intensity) to raw intensity values. Normalized intensity values were used to calculate fold changes among different conditions. The antibodies used for the analysis were validated with appropriate controls (shRNAs and/or overexpression vectors) to assess their specificity. Statistical analysis of IF data was conducted using GraphPad software Prism® (version 6.0b). Unpaired Student's t-test was used to calculate p-values.

Staining of tissue sections
Rehydrated paraffin-embedded liver sections (4μm) were subjected to antigen retrieval treatment for 20 minutes in a steamer in 10 mM sodium citrate buffer, pH 6. Then, sections were incubated in 3% (vol/vol) hydrogen peroxide, washed in PBS solution, blocked with 5% Investigator software. Frozen sections were used for SA-β-Gal staining. The staining was performed as in cultured cells (after thawing the sections). After the staining, the slides were counterstained with eosin, dehydrated, mounted and analysed by phase-contrast microscopy. SA-β-Gal tissue staining was quantified using ImageJ Software (NIH) by measuring the percentage of stained area in each section and multiplying it by its mean intensity value.

Measurement of dNTP concentrations in cells
Samples were harvested and dNTP levels were measured as previously described (Wilson et al. 2011).

Accession numbers
Microarray (shARID1B experiment) and RNASeq (ARID1B overexpression experiment) data have been deposited at the Gene Expression Omnibus under the accession numbers GSE75207 and GSE75291 respectively.