BRG1/SMARCA4 inactivation promotes non-small cell lung cancer aggressiveness by altering chromatin organization

Cell culture

The H358, H441, SK-MES, H2170, H727, SW900, H2228, H520 & H1395, H1703, H522, A427, H23, H1299 and A549 human NSCLC carcinoma cell lines were utilized in this study. All cell lines were obtained from the ATCC and always used from frozen stocks within 3 months of acquisition. All cell lines were cultured at 37°C/5%CO2 in RPMI 1640 supplemented with 10% fetal bovine serum (Gibco, Grand Island, NY, USA) (standard growth medium). H358 and H441 cell lines containing stable BRG1 knockdowns, Brg1i.1, Brg1i.2 and Brg1i.3, were maintained in standard growth medium supplemented with 1 μg/ml puromycin or 0.4mg/ml neomycin.

Generation of BRG1-deficient cell lines

We utilized 2 methods to establish stable knockdown cell lines, lentiviral infection (H358 Brg1i.2 and Brg1i.3) and DNA transfection (H358 Brg1i.1). Lentivirus targeting BRG1 with small hairpin RNA (shRNA) was purified from 293FT cells (17). 293FT cells were transfected one of the MISSION shRNA lentiviral transduction particles [TRCN0000015548 (Brg1i.3) or TRCN0000015549 (Brg1i.2] using Lipofectamine 2000 (Invitrogen, Carlsbad, CA, USA) to generate infectious viral particles. The H358 cell line was then infected in the presence of polybrene for 30 min. Twenty-four hours post-infection, cells were selected with puromycin at 1 μg/ml. Stably infected clones were harvested, expanded and screened for BRG1 expression by Western blotting. Clones with the most complete BRG1 silencing were combined to generate pooled cell lines for each respective BRG1 knockdown. Similarly treated control cell lines were also generated (Supp. Figure 1A).

The H358 Brg1i.1 and the H358 control cell line, designated Control, were generated using PHTP RNAi expression vectors that targets BRG1, or the PHTP empty vector, respectively (18, 19). The H358 cell line was transfected with each vector using FuGene (Roche). After 24 hours, stable clones were selected in puromycin at 1 μg/ml. Over 30 Brg1i.1 and 20 Control puromycin resistant clones were expanded and isolated for further characterization. Clones were screened for reduced BRG1 expression by Western blotting.

To generate stable knockdown H441 cell lines, we infected cells with either pLKO.1, a non-targeting short hairpin RNA (shRNA) control vector (SHC002; Sigma), or the MISSION BRG1 shRNA lentiviral transduction particle (TRCN0000015549) as and selected with 1mg/ml puromycin as described above for the H358 cells. We designated the BRG1 knockdown cell lines as H441 Brg1i.2 (TRCN0000015549) and the control non-targeting shRNA cell line as H441 Control.

In vivo growth assay

Five × 10⁷ cells for each cell line were harvested by trypsinization, rinsed with PBS and resuspended one-to-one in ice-cold Matrigel (BD Biosciences, San Jose, CA). Six-week old Nu/Nu female mice were inoculated into the left lung with 40 μl (5×10⁶ cells/lung) of cell/Matrigel suspension. Mice were monitored daily for signs of distress and were sacrificed when they exhibited weight loss or difficulty in breathing.

Western blotting

Western blotting was performed as previously described (20). Protein expression was analyzed with the following antibodies: anti-BRG1 G7 (sc-17796; Santa Cruz) and anti-ß-ACTIN (A2066; Sigma). Secondary mouse IgG and rabbit IgG antibodies (GE Healthcare) were detected with enhanced chemiluminescence (GE Healthcare).

Quantitative real-time reverse transcription-PCR analysis

RNA expression was examined by QT-PCR as previously described (20). All genes were normalized to β-actin and quantified using 2^−ΔΔCt statistical method (21). The TaqMan primer and probes sets obtained from Applied Biosystems included BRG1 (Hs00946396_m1), SEM3B (Hs00190328_m1), EHF (Hs00171917_m1), DUSP6 (Hs04329643_s1, IL8 (Hs99999034_m1), SYK (Hs00895377_m1), IFI16 (Hs00194261_m1), BATF (Hs00232390_m1) and β-ACTIN (Hs00357333_g1).

Microarray analyses

Total RNA was extracted from H358 (parental), H358 Control (empty vector) and H358 Brg1i.2 and H358 Brg1i.1 (BRG1 shRNA knockdown) cell lines and submitted to the UNC Lineberger Genomics Core for Agilent microarray analysis. RNA was labeled with Cy5 (experimental- H358 Control, H358 Brg1i.1 and H358 Brg1i.2) and Cy3 (control-H358) and hybridized to 4×44 whole human genome microarrays (Agilent Technologies, Wilmington, Delaware, USA). The normexp background correction and loess normalization procedures were applied to the probe-level data (22). Expression measurements for each gene were calculated by computing the mean of the normalized intensity values for all probes mapping to that gene, as specified in a gene annotation database. This produced expression values for 19749 genes. These data are available at GEO (http://www.ncbi.nlm.nih.gov/geo/), under accession number GSE58542.

The SAMR package was used to detect differentially expressed genes by comparing the expression values in each of the two BRG1-deficient cell lines to the expression values in the Control cell line (23). Gene expression values were first standardized within each array. Differential expression was then assessed by finding the smallest value of the tuning parameter delta that produced a median FDR smaller than .001. A total of 1000 permutation were used in each analysis.

NSCLC Cell Line RNA-seq Generation and Analysis

Nine NSCLC cell lines with wild-type BRG1 (wtBRG1) expression (H358, SK-MES, H2170, H441, H727, SW900, H2228, H520 & H1395) and 6 cell lines with BRG1 nonsense/truncating mutations (mtBRG1) (H1703, H522, A427, H23, H1299 & A549) were grown in standard growth medium. Isolated RNA samples were prepared and libraries created using TruSeq RNA Sample Preparation Kit v2 (Illumina), which included a poly A selection step. Libraries were pooled at 2 nm concentration and the samples were then subject to cBot cluster generation using TruSeq Rapid PE Cluster Kit (Illumina). The amplified libraries were sequenced using the TruSeq Rapid SBS Kit on the HiSeq 2500 (Illumina). mRNA-seq data were aligned with Mapsplice (24) and genes were quantified with RSEM (25). Gene expression estimates were upper quartile normalized (26). For comparison of target gene expression in wtBRG1 vs. mtBRG1, gene expression measurements were computed by replacing all RSEM values identically equal to zero with the smallest non-zero RSEM value and then applying a log2 transformation. The SAMR package was used to detect genes that were differentially expressed when the BRG1 mutant cell lines were compared to the BRG1 wild-type cell lines (23). Using an FDR threshold of .025, a total of 135 genes were down-regulated in the BRG1 mutant cell lines (Supplementary Table 2).

To identify an expression matched but unregulated gene set for comparison for each cell line, the mean expression of all 275 down-regulated genes was calculated. We then used ranked values of the Pearson correlation coefficient to identify a set of 300 genes that had expression patterns most similar to mean expression values across all cell lines. This set of 300 genes was chosen from among those genes that did not belong to the set of 591 differentially expressed genes. Using a similar procedure, for each of the 275 down-regulated genes we identified a set of 100 non-differentially expressed genes that had expression patterns most similar to each down-regulated gene of interest across all cell lines. This produced a list of 1484 unique genes that contained all of the 300 genes described above. For sampling purposes, RefSeq genes were selected from this group.

Expression Analyses of Primary NSCLCs

To determine whether reduced BRG1 protein levels correlated with lower mRNA expression in NSCLC, we obtained RNA-seq based gene expression data from the TCGA lung adenocarcinoma project. RNA-seq data were analyzed by first replacing all RSEM values identically equal to zero with the smallest non-zero RSEM value, then a log₂ transformation was applied (25). Samples with BRG1 mutations were identified, and the expression values from samples with nonsense or frame shift mutations were used to identify samples that have reduced BRG1 expression. The threshold for reduced BRG1 expression was defined to be the highest BRG1 expression value observed in any of the samples with BRG1 nonsense or frameshift mutations. The one-sided Wilcoxon rank sum test was applied to assess the statistical significance of differences in the expression values for specific BRG1 target genes, where the null hypothesis corresponds to no difference in expression and the alternative hypothesis is that expression is lower in samples that have reduced BRG1 expression, as defined above.

Nuclear Extraction and MNase Digestion

We initially determined the optimal conditions of MNase digestion to generate a distribution of ~35% mononucleosomes for the parental H358 cell line. We then confirmed a similar distribution of nucleosomes after MNase digestion of chromatin for each cell line to preferentially analyze nucleosomes with similar sensitivity to MNase and avoid overdigestion (27). H358 Control, Brg1i.2 and Brg1i.1 pooled cell lines were harvested at 60-80% confluence. Cells were then removed by treatment with trypsin-EDTA and resuspended in 1ml RSB buffer (10 mM Tris-HCl [pH 7.4], 10 mM NaCl, 3 mM MgCl₂) and placed on ice for 10 minutes, followed by addition of 0.1ml 10% NP-40 detergent for 30min. After two washes with RSB buffer, nuclei were stored at −80°C. For MNase digestion, extracts were resuspended in 0.2ml 1x MNase reaction buffer (10 mM Tris-HCl [pH 7.5], 5 mM MgCl₂, 5 mM CaCl₂, 0.1 mM PMSF, 0.5 mM DTT), aliquoted into 0.5 ml samples with an OD₂₆₀ of 0.2 and treated with 10 U of MNase (Affymetrix) for 10 minutes. MNase was inactivated with the addition of 10 mM EDTA and EGTA. All samples were then treated with RNase A and Proteinase K followed by DNA isolation by Phenol-Chloroform extraction and ethanol precipitation. DNA was resuspended in 10μl 0.1xTE and run on a 2% agarose gel to separate nucleosomes. The mononucleosome band was cut from the gel and purified with the Qiagen Gel Extraction Kit (Qiagen, Valencia, CA). The manufactures protocol was followed with the exception that 6 volumes of QG were used to dissolve the gel, 2 gel volumes of isopropanol were used and DNA was eluted with 30μl EB.

Solexa Library Preparation and Illumina Sequencing

Libraries were created following manufacturer's specifications (Illumina). Library preparation included blunt ending of DNA, addition of a polyA-overhang, adapter ligation (Illumina), two times SPRI beads clean up, PCR amplification with PfuUltra II Fusion HS DNA Polymerase (Stratagene), and size selection from a 2% agarose gel. After library generation paired-end sequencing was performed (Illumina HiSeq2000, UNC Chapel Hill High Throughput Sequencing Facility).

Reduced BRG1 expression in NSCLC cells induces in vitro and in vivo growth changes

In order to determine the effects of BRG1 loss on NSCLC development, we generated BRG1 knockdown clonal cell lines derived from the human NSCLC H358 cell line (hereafter referred to as “NSCLC cells”) that expresses a wild-type BRG1 protein (5). We used two different shRNAs (“Brg1i.1” and “Brg1i.2”) to control for possible off-target effects. In general, the Brg1i.2 shRNA proved more efficient at reducing BRG1 mRNA and protein levels than the Brg1i.1 shRNA. Therefore, to control for clonal variation, we generated pools of at least 4 clonally derived cell lines from each BRG1 shRNA or the vector control, referred to as “Control” (Supp. Figure 1). Cells with reduced BRG1 expression exhibited a consistent change in cellular morphology from a tightly packed and cuboidal appearance to an elongated morphology with distinct cellular borders (Figure 1A).

We next examined whether BRG1 loss altered the tumorigenic potential of NSCLC cells in vivo. Parental, Control and Brg1i.1 cell lines were orthotopically inoculated into the lungs of athymic nude mice. Reduced expression of BRG1 led to decreased survival (mean survival=37.5 days) compared to either the parental cell line (mean survival= 110 days, p<0.05) or the Control (mean survival= 100 days, p<0.018) (Figure 1B). The reduced survival correlated with the appearance of larger tumors within the lung and mediastinum in mice with the Brg1i.1 cells compared to small tumors limited to the lungs of the animals with Control cells (Figure 1C). Therefore, knockdown of BRG1 expression resulted in dramatic changes in the tumorigenic potential of NSCLC cells.

BRG1 reduction causes specific changes in gene expression

We next asked how BRG1 loss might lead to the observed in vitro and in vivo changes. Because of the SWI/SNF complex's role in regulating gene transcription, we determined the effects of reduced BRG1 protein levels on gene expression. We identified genes that were differentially expressed between either the Brg1i.1 or Brg1i.2 cell lines relative to parental cells (median FDR < 0.001, Supp. Table 1- genes altered by BRG1 KD). Consistent with more efficient BRG1 silencing, a greater number of genes demonstrated differential expression in the Brg1i.2 cells when compared to the parental cells (2050 downregulated, 2251 upregulated) than the H358 Brg1i.1 cells (363 downregulated, 454 upregulated).

We identified 591 genes that were differentially expressed in both BRG1 knockdown cell lines-275 downregulated genes (Table 1) and 316 upregulated genes (Supplementary Table 1-consensus genes). Hierarchical clustering showed highly concordant gene expression patterns among the replicates as well as consistent alterations in gene expression between the each of the two BRG1 knockdown cell lines relative to control cells (Figure 2). Our previous study had correlated BRG1 loss with reduced expression of genes frequently downregulated during NSCLC development (15), so we therefore focused on downregulated genes after BRG1 knockdown in this current study.

Validation of novel targets associated with BRG1 loss

We examined the 275 genes that showed decreased expression in both BRG1-knockdown cell lines for candidates previously associated with cancer development (Table 1). These putative targets included many genes in signaling pathways associated with tumor development including inflammation (IL1B, IL23A, IFI16 and FAS), cell invasion (ICAM1, ICAM2 and MMP7) and cell proliferation (AREG, BATF, and SYK). We additionally identified three genes with reduced expression that had been previously associated with the development of lung and other cancers: DUSP6, EHF and SEMA3B. DUSP6 (MKP3) is a dual specificity phosphatase that regulates MAPK signaling through inactivation of ERK2 (34, 35), and recent studies have demonstrated decreased DUSP6 expression in adenocarcinomas of the lung (36, 37). EHF (ESE-3) is a member of the ETS transcription factor family that has been implicated as a tumor suppressor for prostate cancer as well as a regulator of inflammation in airway epithelium (38, 39). SEMA3B codes for a secreted member of the semaphorin/collapsin family (40). The protein plays a critical role in the guidance of growth cones during neuronal development and has been implicated as a tumor suppressor gene for NSCLC (40-42).

We next validated the gene expression differences of 6 representative genes by quantitative PCR (qPCR) in the BRG1-deficient cell lines as well as a third BRG1 silenced cell line (Brg1i.3) (Figure 3). Reduced levels of mRNA for these genes generally correlated with the degree of BRG1 loss (Figure 3). We then determined if BRG1 knockdown in a second NSCLC cell line would cause similar changes. We generated a BRG1-deficient H441 cell line using the Brg1i.2 shRNA as well as a non-targeting shRNA Control cell line (Supplementary Figure 1B). Similar to the H358 Brg1i cell lines, the H441 Brg1i.2 cells show a consistent change in cellular morphology (Supplementary Figure 1C). A gene expression analysis of our 6 putative target genes showed reduced expression of 3 genes with no change or increased expression of the remaining genes (Figure 3). To further validate the remaining target genes, we compared RNA-seq expression data from 9 NSCLC cell lines with wild-type BRG1 genes to 6 cell lines with documented BRG1 truncating/nonsense mutations (5). Using a FDR of 0.025, we confirmed that lower EHF, IFI16 and SYK expression correlated with BRG1 loss (Supplementary Table 2). Therefore, we identified 3 candidate genes whose expression decreases in the presence of reduced BRG1 expression.

Target gene expression correlates with BRG1 expression in NSCLC primary tumors

We then asked whether these putative target genes demonstrate similar expression changes in primary human tumors based on BRG1 status. Using the recently released TCGA data for human adenocarcinomas of the lung, we assessed whether BRG1 gene expression correlated with BRG1 mutational status. BRG1 nonsense and frameshift mutations that should lead to truncated or absent protein showed reduced expression of BRG1 mRNA (Figure 4A). In contrast, tumors with missense mutations in BRG1 did not exhibit an obvious change in mRNA expression compared with wild-type BRG1 tumors (Figure 4A). Based upon these observations, we divided the dataset into tumors with either high or low BRG1 expression. We then asked whether expression of our BRG1 target genes correlated with BRG1 expression. EHF, IFI16 and SYK showed a statistically significant association with BRG1 expression (p ≤ 0.01, one-sided Wilcoxon rank sum test, Figure 4B-D). In contrast, we did not observe a statistically significant association for ACTB expression (p= 0.121, Figure 4E). Therefore, for these putative downstream targets, expression changes resulting from BRG1 loss in cell lines was similar to that seen in primary tumors.

BRG1 loss alters nucleosome positioning and occupancy

Given the role of SWI/SNF in active chromatin remodeling, we examined nucleosomal differences associated with BRG1 loss. By virtue of its ability to cleave internucleosomal linker DNA, MNase analyzed by high throughput sequencing (MNase-seq) offers the ability to compare multiple features of nucleosomes. MNase-seq data from control cells and the two BRG1-silenced cell lines were processed to generate normalized signal tracks (DANPOS) (30). Local peaks in MNase signal indicate the presence of nucleosomes and their position whereas the signal magnitude indicates relative occupancy. Occupancy in this context refers to the fraction of chromatin that is nucleosomal at a specific position among the population of cells being assayed.

We first examined averaged signal around all transcription start sites (TSS). We observed a decrease in overall signal in the −1 to +1 kb region surrounding the TSS (Figure 5A). This difference was associated with the degree of BRG1 knockdown and indicates that, in aggregate, fewer nucleosomes occupy these regions of chromatin. In all samples, we observed three well-positioned nucleosomes preceding and five nucleosomes following the TSS (designated as −3 to +5). BRG1 knockdown did not alter the position of nucleosomes around the TSS; rather it resulted in signal attenuation that correlated with BRG1 expression levels. An even broader region of well-defined nucleosomes was also observed around CTCF sites (Supp. Figure 2A). In contrast to the TSS, relative peak signal attenuation was not observed between samples although overall signal was diminished, similar to the TSS. We then examined signal further from the TSS and noted that aggregate signal was greater within the gene body compared with upstream regions (Figure 5B). This polarity was lost when BRG1 was knocked down (Figure 5B). To correlate BRG1 chromatin interaction with variation in MNase signal after BRG1 knockdown, we overlaid differential MNase signal with publically available BRG1 ChIP-seq signal around the TSS (43). BRG1 ChIP signal is greatest at the region of maximal MNase signal change (Supp. Figure 2B). Together these data suggest that BRG1 plays an influential role in maintenance of nucleosomes at the TSS.

Individual nucleosomes were then computationally predicted and assigned occupancy and positional ambiguity scores. These scores were then compared between the control and BRG1 knockdown cells. Occupancy scores for individual nucleosomes were diminished genome-wide when BRG1 was knocked down (Figure 5C, Supp. Figure 2C). Variation in occupancy was also observed at TSS-proximal (+/− 3 kb) nucleosomes. As anticipated, nucleosomes near the TSS demonstrated higher occupancy scores compared with nucleosomes genome-wide. However, BRG1 silencing resulted in TSS-proximal nucleosomes with occupancy lower than nucleosomes genome-wide. These data indicate that nucleosomes with the highest occupancy scores demonstrate the greatest change in the absence of BRG1 with the TSS-proximal nucleosomes being most affected. We also scored nucleosomes based on predicted positional ambiguity. Ambiguity score measures signal characteristics indicative of variable positioning among the cell population. A higher score indicates increased ambiguity. The loss of BRG1 was associated with increased positional ambiguity overall (Figure 5D, Supp. Figure 2B). However, in contrast to occupancy, TSS-proximal nucleosomes demonstrated less change in positional ambiguity when compared with nucleosomes genome-wide. These data suggest that relative to nucleosomes genome-wide, BRG1 plays a greater role in maintaining occupancy at the TSS but relatively less of a role in maintaining the positioning of these nucleosomes. This is consistent with the role of factors such as RNA polymerase in the maintenance of nucleosome positioning around the TSS (44).

We then examined MNase signal around the TSS of those genes differentially expressed after BRG1 knockdown. We also compared signals around the TSS for a set of genes that was expression level-matched to the downregulated genes but without altered expression after BRG1 knockdown (designated as unregulated genes, Supp. Table 1). Because, as a group, RNA levels were higher for the downregulated gene set, expression-level matching was performed to control for the effect of gene expression on MNase signal. We observed decreased signal around TSS in BRG1 silenced cells (Figure 6A, top and Supp. Fig 3). This difference was most clearly observed downstream of the TSS. To better characterize differences between samples, MNase signal was de-noised to remove the effect of general signal depletion around the TSS (Figure 6A, bottom and Supp. Figure 4). Signal peaks corresponding to nucleosomes were diminished, most notably around the downregulated genes, affecting the −2 to +2 nucleosomes. In the absence of BRG1, signal at the NDR was increased. This effect was also evident, albeit to a lesser degree, when all genes were examined (Supp. Figure 4A). This effect was observed when we similarly analyzed a previously published dataset from BRG1 knockout murine embryonic fibroblasts (Supp. Figure 4B) (45). Signal at the TSS of a subset of downregulated genes exemplify this effect and suggest a correlation with the level of BRG1 expression (Figure 6B and Supp. Figure 5).

To further explore this effect, we calculated the ratio of differences in MNase signal between the −1 nucleosome and the NDR for Control and Brg1i.2 samples at each gene in the downregulated and unregulated gene sets. This analysis was performed in the absence of kernel smoothing to eliminate potential effects associated with this correction technique. Loss of BRG1 resulted in a greater MNase signal change for the downregulated genes compared with the unregulated genes. The predominant difference for the downregulated genes (compared with the matched unregulated genes) reflected a decreased signal variation between the −1 nucleosome position and NDR (p < 0.01 based on permutation of a 1300 gene class of matched unregulated genes) (Figure 6C). Together these data support that loss of BRG1 is associated with increased relative nucleosome occupancy at the NDR suggesting that BRG1 may cooperate with RNA Pol II in the eviction and maintenance of nucleosomes around the TSS.