Two novel BRM insertion promoter sequence variants are associated with loss of BRM expression and lung cancer risk

Abstract

SWI/SNF (SWItch/sucrose non-fermentable) complexes are ATP-dependent chromatin remodeling enzymes critically involved in the regulation of multiple functions, including gene expression, differentiation, development, DNA repair, cell adhesion and cell cycle control. BRM, a key SWI/SNF complex subunit, is silenced in 15–20% of many solid tumors. As BRM-deficient mice develop 10-fold more tumors when exposed to carcinogens, BRM is a strong candidate for a cancer susceptibility gene. In this paper, we show that BRM is regulated by transcription, thus demonstrating that the promoter region is important for BRM expression. We sequenced the BRM promoter region, finding two novel promoter indel polymorphisms, BRM −741 and BRM −1321, that are in linkage disequilibrium (D′ ≥0.83). The variant insertion alleles of both polymorphisms produce sequence variants that are highly homologous to myocyte enhancer factor-2 (MEF2) transcription factor-binding sites; MEF2 is known to recruit histone deacetylases that silence BRM expression. Each polymorphic BRM insertion variant is found in ~20% of Caucasians, and each correlates strongly with the loss of protein expression of BRM, both in cancer cell lines (P=0.009) and in primary human lung tumor specimens (P=0.015). With such strong functional evidence, we conducted a case–control study of 1199 smokers. We found an increased risk of lung cancer when both BRM homozygous promoter insertion variants were present: adjusted odds ratio of 2.19 (95% confidence interval, 1.40–3.43). Thus, we here demonstrate a strong functional association between these polymorphisms and loss of BRM expression. These polymorphisms thus have the potential to identify a sub-population of smokers at greater lung cancer risk, wherein this risk could be driven by an aberrant SWI/SNF chromatin-remodeling pathway.

Introduction

Though smoking significantly contributes to lung cancer development, the genetic contribution to this disease remains elusive. Only 10% of smokers develop cancer, whereas 10–20% of lung cancer patients are lifetime never smokers. Clearly, other factors, such as genetics, are contributing to lung cancer susceptibility (Liu et al., 2005). In the past, molecular epidemiological studies of genetic variation have yielded contradictory results, in part because of the lack of knowledge about the biology of the experimental system (Liu et al., 2005; Gazdar and Boffetta, 2010). Recently, genome-wide association studies (GWASs) have finally yielded consistent results (Gazdar and Boffetta, 2010; Truong et al., 2010), in which cancer risk can be localized to specific chromosomal regions containing promising candidate genes. Further studies of these promising candidate genes may improve the knowledge base of etiological factors, but they still may not translate into clinically useful knowledge that could be used to improve lung cancer morbidity or mortality. GWAS also suffer in that true functional polymorphisms have either not been identified or fully characterized, the currently identified risk associations are extremely modest (odds ratios (ORs) of 1.1–1.3), and GWAS platforms still miss a proportion of human genetic variation.

In addition, in the case of complex heterogeneous cancers such as lung cancer, the molecular pathways driving the development and progression of these cancers can be quite diverse. Already there are specific molecular subsets with differing epidemiological risk factors and varied natural histories (Gazdar and Boffetta, 2010), as evidenced by subsets of cancer driven by epidermal growth factor receptor-activating mutations and ALK–EML4 translocations (John et al., 2009; Solomon et al., 2009). These two examples represent somatic molecular changes that arise during tumor development and have resulted in both treatment changes and improved clinical outcome in molecularly defined subsets of lung cancer patients. We can surmise that future examples may include other patient subsets that are defined not only by other somatic genetic changes (that arise as cancers develop), but also by germline sequence variants as well.

There is accumulating evidence of the importance of impairment and/or inactivation of the SWI/SNF (SWItch/sucrose non-fermentable) chromatin-remodeling complex in lung cancer development and progression. SWI/SNF mediates gene expression by shifting the position of histones, thereby making the DNA more accessible to transcription factors and key cellular proteins. Thus, many signal transduction pathways and anticancer proteins, such as retinoblastoma (Rb), p53 and BRCA1, are functionally dependent on this complex (Bochar et al., 2000; Muchardt and Yaniv, 2001; Klochendler-Yeivin et al., 2002; Reisman et al., 2009). Although the loss of BRM in mouse models was not found to be tumorigenic, exposure to lung carcinogens yielded 10-fold more tumors (Glaros et al., 2007), suggesting that BRM may be a tumor susceptibility gene. Hence, BRM loss may promote carcinogenesis in the presence of other causal factors such as smoking. As loss of BRM can impact a diverse array of pathways, it is not surprising to find this protein is suppressed or lost in many tumor types. Loss of heterozygosity (LOH) at the BRM locus, 9p23–24, is commonly observed across multiple tumors, including lung cancer (An et al., 1999; Girard et al., 2000; Sarkar et al., 2002; Tripathi et al., 2003; Sabah et al., 2005; Gunduz et al., 2009). Moreover, BRM expression is lost in 30–40% of lung cancer cell lines (Reisman et al., 2003) and is lost consistently in 15–18% of all major human lung cancer histological subtypes (Glaros et al., 2007). Molecular studies have shown that BRM binds to, and is required for, the function of Rb and Rb family proteins p107 and p130, and BRM re-expression is thought to reconnect growth inhibitory pathways, because re-expressing BRM halts cellular growth and induces a flattened morphology (Dunaief et al., 1994; Strober et al., 1996; Muchardt et al., 1998). Analyses of lung tumors show that loss of BRM protein expression leads to worse non-small cell lung cancer (NSCLC) prognosis (Reisman et al., 2003; Fukuoka et al., 2004). A key feature of BRM is that the loss of BRM function and expression in cancer cells is because of the reversible epigenetic silencing (Glaros et al., 2007). To this end, we have shown that pharmacologic reversal of BRM loss of expression is possible (in a parallel article in this edition of Oncogene; Gramling et al., 2011), thereby opening the chance for novel interventional approaches, a key feature that has been lacking in the current slate of other promising risk candidates for lung cancer.

In this paper, we describe two novel BRM promoter polymorphisms that are strongly associated with the loss of BRM expression both in lung cancer cell lines and in primary lung cancers. The two polymorphic sites are in the promoter region of BRM at −741 bp (rs34480940; 7 bp indel (insertion–deletion) polymorphism) and −1321 bp (rs3832613 or rs59259177; 6 bp indel polymorphism) upstream of the BRM transcription start site. We have observed that these insertion variants create sequences that are highly homologous to myocyte enhancer factor-2 (MEF2) transcription factor binding sites (92% similarity), providing a potential mechanism for how the variant polymorphic alleles might facilitate the silencing of this gene. These two polymorphisms are further found to be important germline risk factors in the development of smoking-related lung cancer.

Results

BRM is regulated by transcription

To understand how BRM is silenced, we investigated whether the BRM promoter is essential for gene expression. It has been shown that BRM is regulated by histone deacetylase (HDAC) inhibitors (Bourachot et al., 2003; Yamamichi et al., 2005; Glaros et al., 2007), but it was not known whether these reagents impact BRM transcription or post-transcription mechanisms. To address this question, we examined whether or not the BRM gene could be regulated by transcription. As a baseline, we first conducted quantitative PCR to see if and how the BRM mRNA levels differed in BRM-positive cell lines compared with those that lack BRM (Figure 1a). As expected, BRM-negative cell lines differed (that is, were lower) by an average of at least 6–8 delta threshold cycles (Cts), or 32- to 64-fold different. However, a few cell lines had low levels of BRM rather than being completely devoid of it; these had a difference of about 3–4 delta Cts, an 8- to 16-fold difference. These low levels of BRM mRNA suggest that this gene is regulated by transcription or mRNA turn-over (for example, microRNA). To begin to distinguish between these two possibilities, we next measured the level of BRM mRNA after applying two HDAC inhibitors (either tricostatin or CI-994) over time (30 min to 12 h) in both six BRM-negative (A427, SW13, H522, C33A, H1299, H23) and six BRM-positive cell lines (Calu 3, A549, H441, H460, H2450 and Calu 6). These data show a sharp induction of BRM mRNA in BRM-deficient cell lines compared with the BRM-positive cell lines after the application of either HDAC inhibitor (Figures 1b and c). Because heterogeneous mRNA is spliced to form mature mRNA, its rapid induction after HDAC application would be indirect evidence that BRM restoration is caused by increases in transcription. Using quantitative PCR and BRM intron primers that straddle exon 23, we observed that BRM heterogeneous mRNA was rapidly induced >75- to 100-fold following HDAC inhibitor application over time in BRM-negative cells compared with BRM-positive cell lines (Figures 1d and e). Repeating this experiment using BRM intron primers that straddle either BRM exons 3 or 34, we observed essentially the same results (data not shown), indicating that elongation block is not the cause of BRM suppression. Next, to test whether BRM is regulated by transcription, we conducted nuclear run-on experiments as described (Patrone et al., 2000) on two BRM-deficient cell lines after treatment with either tricostatin or CI-994. After a 3-h treatment period with these HDAC inhibitors, we measured transcription by quantitative PCR and observed a large, >400-fold induction of BRM transcription in the C33A or SW13 cell lines (Figure 1f). We repeated this experiment in two other cell lines (H522 and A427), observing similar results—a >300-fold induction within hours (data not shown). Therefore, transcription appears to be the major mode of gene regulation for the BRM gene and indicates that the BRM promoter is central to the regulation of BRM transcription.

Figure 1.

(a) The baseline mRNA levels in both BRM-positive and BRM-negative cell lines using quantitative PCR of complementary DNA made from total mRNA each cell lines. BRM mRNA levels were measured by quantitative PCR using BRM complementary DNA primers and standardized to POLR2A (RNA polymerase 2A) levels. The POLR2A sense and POLR2A antisense primers that were used are 5′-ATGTCTGTGACGGAGGGTGGCA-3′ and 5′-GCCAGGACACTCTGTCATGTTCCT-3′, respectively, and the BRM complementary DNA primers were 5′-GATTGTAGAA GACATCCATTGTGG-3′ and 3′-GACATATAACCTTGGCTGT GTTGA-5′. Panels b (tricostatin (TSA)) and c (CI-994) illustrate the changes in levels of BRM mRNA in both BRM-deficient and BRM-positive cell lines over time after applying the two HDAC inhibitors, either trichostatin A or CI-994 as measured by quantitative PCR using the above primers sets. Induction of BRM mRNA levels is seen as early as 15–30 min after application of these HDAC inhibitors. Panel d (TSA) and e (CI-994) shows induction in levels of unspliced BRM mRNA (heterogeneous BRM mRNA) after application of either TSA or CI994 using quantitative PCR with intron primers flanking exon 23: 5′-TTCAGGGCTGAACTGT ATCG-3′ and 3′-TATGGCTTCCGAGCCTAATG-5′. Panel f illustrates the induction of BRM transcriptional activity in C33A or SW13 cell lines after the application of either TSA or CI-994 at 3 h, compared with the BRM transcription activity before HDAC application using the exon/intron primer set spanning the boundary of exon 23.

BRM promoter sequence analysis identified two insertion polymorphisms

In our search to understand how BRM is silenced, in our previous studies we sequenced both BRM mRNA and genomic DNA, but we did not find any mutations in the coding region of the BRM gene in multiple cell lines or in primary lung tumors (Glaros et al., 2007). We next sought to analyze the BRM promoter region for alterations that could explain how BRM might be silenced. We identified the transcriptional start site of the BRM gene by conducting 5′-race using mRNA from several cell lines. This start site was then confirmed by using a number of web-based promoter search programs and examining known capped BRM complementary DNAs from the National Center for Biotechnology Information database. Using this information, we then conducted expand long PCR (Roche, Indianapolis, IN, USA) and amplified both a long (2796 bp) and short (944 bp) DNA fragment containing the BRM transcription start site (data not shown). We also cloned both these (944 and 2796 bp) BRM promoter fragments into a luciferase Promega vector pGL3 (Promega, Madison, WI, USA) in both the reverse and forward directions. After transfecting these plasmids into both BRM-negative and BRM-positive cell lines, we observed that both short and long BRM promoter fragments induced luciferase expression when oriented in forward direction, but not in the reverse direction, in each cell line examined (Supplementary data 1). Hence, we confirmed the location of the BRM promoter.

Because the promoter region is important in the regulation and expression of BRM, we then sequenced the BRM promoter region (genomic DNA) for possible alterations that might explain why BRM is silenced in cancer cells. Although no mutations were found in the promoter region, after sequencing ten BRM-deficient cell lines and several primary lung cancers (Glaros et al., 2007) using Sanger sequencing, we found two promoter indel sequence variants (Figure 2a). We also found that imputation of these polymorphisms from existing GWAS data would not have been feasible, given that these two BRM promoter polymorphisms were not in linkage disequilibrium with the polymorphisms found on GWAS platforms at that time (Bailey-Wilson et al., 2004; Hung et al., 2008; Landi et al., 2008; Landi et al., 2009). To determine the frequency of these BRM polymorphisms, we sequenced these regions in 161 healthy Caucasian-predominant volunteers. The minor allele frequencies and genotype frequencies are shown in Figure 2d. Both polymorphisms were in Hardy–Weinberg equilibrium (P>0.10) and in linkage disequilibrium (D′=0.86). Sequence homology analysis revealed that both the BRM −741 and BRM −1321 insertion alleles created a sequence that had 92% homology to consensus sequences for MEF2-binding sites (Fickett, 1996; Figures 2b and c), whereas the wild-type deletion alleles contained no such MEF2 consensus sequence.

Figure 2.

Panel a illustrates the location and sequence of the two BRM promoter insertion–deletion polymorphisms. The polymorphic sites are located at 1321 bp (6 bp insertion) and 741 bp (7 bp insertion) upstream of the transcriptional start site. (b, c) Each polymorphism creates a MEF2-like binding motif (motif is shown in bold). The MEF2-like binding sites for the BRM −741 and BRM −1321 variants are shown based on the sequence algorithm described in Fickett, 1996. This 12-base-pair-defined binding site (in bold) is under the BRM promoter sequence for each site. (d) Institutional review board approval was obtained from University of Michigan before this analysis. In all, 161 self-described healthy male and female volunteers, with over 95% describing themselves as Caucasian, were recruited from the University of Michigan hospital system. Whole-blood samples were obtained from each participant. DNA was extracted (Qiagen, Valencia, CA, USA) and genotyped for both BRM promoter polymorphisms using Sanger sequencing. Blood DNA was amplified using the following PCR conditions were 94 °C for 4 min initial melt; 94 °C for 30 s, 59 °C for 30 s annealing and 72 °C extension for 30 s. This was continued for 42 cycles followed by a final extension of 7min at 72 °C. The PCR primers were: poly1: 5′ BRMPROM-6955: 5′-TTTGGAAGCTTGCAGTCCTT-3′; 3′ BRM PROM-7089: 3′-CCGGCTGAAACTTTTTCTCC-5′; poly2: 5′ BRMPROM-6269: 5′-CCCAGTTGCTCAAATGGAGT-3′; 3′ BRM PROM-6573: 3′-AGGTCGGTGTTTGGTGAGAC-5′.

Homozygous variants of these polymorphisms were associated with BRM-deficient cell lines and primary lung tumors

We have previously reported that a number of cell lines and lung cancer tumors were either BRM positive or negative according to immunohistochemistry and/or western blotting analyses (DeCristofaro et al., 2001; Reisman et al., 2003; Glaros et al., 2007). On inspection of the BRM-negative cell lines, we noted that both BRM polymorphisms occurred in BRM-negative cell lines at a significantly higher-than-expected frequency: in 12 BRM-negative and 12 BRM-positive cell lines (Table 1a; Muchardt and Yaniv, 2001; Reisman et al., 2003), all BRM-deficient cell lines contained at least one homozygous variant genotype of BRM −741 or BRM −1321; and 5 of 12 contained both homozygous variant insertion genotypes. In contrast, BRM-positive cell lines yielded a mix of genotypes for both polymorphisms, with only four BRM-positive cell lines containing at least one homozygous variant BRM genotype (P=0.009 for presence of at least one homozygous variant polymorphism genotype, Fisher’s exact test). In contrast to the BRM-positive cell lines, LOH around 9q24 was demonstrated in 8 of 11 BRM-negative cell lines (data not shown). Thus, these data demonstrate strong, significant associations between the homozygous variants of these promoter polymorphisms and loss of BRM expression.

Table 1.

(a) Cell lines and (b) human NSCLC tumors

BRM-positive cell lines			BRM-negative cell lines
Cell line	BRM-741	BRM-1321	Cell line	BRM-741	BRM-1321
(a)
A549	Wt	Wt	A427	Homo	Homo
ES2	Hetero	Hetero	C33A	Wt	Homo
H2052	Wt	Wt	H125	Wt	Homo
H28	Wt	Homo	H1299	Homo	Homo
H792	Wt	Wt	H1573	Wt	Wt
HeLa	Wt	Homo	H23	Homo	Homo
PA-1	Hetero	Wt	H513	Wt	Homo
Calu-6	Hetero	Homo	H522	Homo	Homo
HCC95	Wt	Wt	Panc-1	Homo	Wt
H441	Hetero	Homo	SW13	Homo	Wt
H460	Wt	Wt	SSC-9	Homo	Homo
HCC2450	Wt	Wt	SSC17B	Wt	Homo

BRM-positive NSCLC tumors					BRM-negative NSCLC tumors
Tumor ID	BRM-741		BRM-1321		Tumor ID	BRM-741		BRM-1321
	Tumor	Normal tissue	Tumor	Normal tissue		Tumor	Normal tissue	Tumor	Normal tissue
(b)
1	Wt	Wt	Wt	Hetero	A	Homo	Homo	Homo	Homo
2	Wt	Wt	Wt	Wt	B	Homo	Homo	Homo	Homo
3	Wt	Wt	Homo	Homo	C	Homo	Homo	Homo	Homo
4	Wt	Hetero	Wt	Hetero
5	Homo	Hetero	Wt	Hetero	D	Wt	Wt	Homo	Homo
6	Homo	Homo	Wt	Wt	E	Homo	Homo	Homo	Homo
7	Homo	Homo	Wt	Hetero	F	Homo	Homo	Homo	Homo
8	Wt	Wt	Hetero	Hetero	G	Homo	Homo	Homo	Homo
9	Hetero	Hetero	Hetero	Hetero	H	Wt	WT	Homo	Homo
10	Homo	Hetero	Wt	Wt	I	Homo	Homo	Homo	Homo
11	Hetero	Hetero	Hetero	Hetero	J	Homo	Hetero	Homo	Homo
12	Hetero	Hetero	Homo	Homo

Abbreviations: hetero, heterozygous variant; homo, homozygous variant; NSCLC, non-small cell lung cancer; Wt, wild type. Homozygous variants of BRM-741 and BRM-1321 polymorphisms are associated with loss of BRM protein expression in (Table 1a) cell lines and (Table 1b) human NSCLC tumors and their adjacent normal tissue. Cell lines were chosen based on published western blotting data (Reisman et al., 2002; Strobeck et al., 2002; Reisman et al., 2003). To find BRM-negative and robustly BRM-positive tumors cases, we stained tissue microarrays as described in (Reisman et al., 2005; Glaros et al., 2007). Adjacent normal lung tissue was histologically confirmed and chosen from an area distant from the tumor. DNA was extracted and BRM genotyping was performed for these cell lines and human normal/tumor lung tissue, blinded to BRM immunohistochemistry status. Genotyping results were categorized as wild-type, heterozygous variant or homozygous variant for BRM-741 and BRM-1321 separately. The loss of BRM expression was strongly associated with the presence of at least one homozygous variant in cell lines (P=0.009), in the NSCLC tumors (P=0.015), and in their adjacent normal lung tissue (P=0.002). The table also empirically shows the much higher rates of homozygous variants of both polymorphisms in the BRM-negative cell lines and BRM-negative human NSCLCs. For the DNA obtained from paraffin blocks, we laser captured the tumors, isolated the DNA and genotyped these tumors using PCR and nested PCR for the two BRM polymorphisms. We also collected and genotyped DNA from normal paraffin-embedded lung tissue from the same patients. The PCR primers were: 3′-POLY1-7042: 3′-CTGCCCCCTATTCCAGGTAA-5′; 3′-POLY1-7089: 3′-CCGGCTGAAACTTTTTCTCC-5′; 5′-POLY1-6955: 5′-GCAACAGTAAAATGGTCTTA-3′; 5′-POLY2-6296: 5′-CCCAGTTGCTCAAATGGAGT-3′; 3′-POLY2-6573: 3′-AGGTCGGTGT TTGGTGAGAC-5′; 3′-POLY2-6547: 3′-ATTTTTAGTTTTATGAAGTG-5′. The magnesium concentrations are as follows for each primer pair: 7042/6955 Mg=4 μM; 7089/6955 Mg=3 μM; 6296/6573 Mg=6 μM; 6296/6547 Mg=6 μM. The PCR conditions are as follows for all reactions: 94 °C for 3 min initially, then 94 °C for 30 s, annealing at 58 °C for 30 s and extension for 72 °C for 30 s for 40 cycles and then a final extension of 5 min. Promega Taq (2 μl, Promega) with buffer and containing no Mg was used for all reactions. Final primer concentrations were 0.1 μM.

However, as cell lines can produce artifactual results, we also examined the relationship between BRM loss and these polymorphisms in human NSCLCs. We examined 22 primary NSCLC tumors, chosen such that 12 tumors had robust BRM staining (BRM positive) and 10 tumors were completely devoid of BRM staining (BRM negative). The majority of both tumor and normal samples from the BRM-negative cases were homozygous for both BRM variants, whereas the BRM-positive cases followed closely to a normal population distribution, with minor allele frequencies of 42–46% for each polymorphism (Table 1b). The loss of BRM expression in the tumor was strongly correlated to the presence of both homozygous variants identified from DNA derived from lung tumors (P=0.015) as well as DNA derived from the adjacent normal lung tissue (P=0.002). We also compared the relationship between genotypes from tumor DNA and normal adjacent DNA. The quadratic-weighted κ-statistics comparing genotype results from tumor and normal tissue was 0.79 for BRM −741, and 0.70 for BRM −1321, suggesting good correlation between tumor and normal tissue genotyping results. Because the BRM-negative tumors did not demonstrate any heterozygous alleles, we could not infer sites of LOH, but in positive tumors, the change from heterozygous in the normal to wild type in the tumor was observed in four cases, indicating that LOH does indeed occur in this locus. In our analysis, LOH affected the analyses materially in only 1 of 22 samples evaluated (5%), as heterozygous and wild-type variants were grouped together for our primary analyses and compared with the homozygous insertion variants.

As the major mechanisms of BRM silencing are not due to mutation, but through epigenetic changes, reversibility of such silencing is possible. We and others found that BRM can be upregulated by HDAC inhibitors (Reisman et al., 2009), and HDACs are known to be recruited by MEF2 transcription factors (Gregoire et al., 2007) leading to silencing of target genes. Functionally, our data can explain that the insertion alleles of these two BRM promoter polymorphisms might lead to MEF2 binding, which subsequently causes recruitment of HDACs, finally resulting in the silencing of BRM expression. This, in turn, leads to an increased chance of cancer development. Ongoing studies in our laboratory are aimed at investigating this hypothesis.

The homozygous variants of both polymorphisms are associated with lung cancer risk

BRM appears to be a tumor-susceptibility gene (based on published BRM-null mice studies) and BRM polymorphisms are tightly correlated with loss of BRM protein expression. For these two reasons, we hypothesized that the presence of BRM promoter polymorphic variants defines a sub-population of individuals that has a higher risk of developing lung cancer. To test this hypothesis, we conducted a case–control study, whereby we genotyped 484 smoking lung cancer and 715 smoking matched healthy controls. Table 2a presents the clinical and demographic data for cases and controls. For each separately, both polymorphisms were in Hardy–Weinberg equilibrium (P>0.05). The two polymorphisms were in linkage disequilibrium (D′=0.83). For each polymorphism, crude and adjusted models found significance with BRM −741 (global Wald test, P=0.02 for crude and adjusted models) and with BRM −1321 (global Wald test, P=0.006 crude and P=0.004 adjusted models). A discrete genetic model revealed that the main driver of these associations came from the homozygous variants of both promoter polymorphisms (Table 2b). Additive genetic models (global Wald P-value of 0.008 across both polymorphisms, with reference category of ‘no variants’) confirmed these findings: comparing four versus no variant alleles (adjusted OR (aOR), 2.21 (95% confidence interval, 1.4–3.4) was highly significant, whereas three (aOR, 1.31; 95% confidence interval, 0.9–2.0), two (aOR, 1.41; 95% confidence interval, 1.0–2.0) and one (aOR 1.13; 95% confidence interval, 0.8–1.7) variant alleles had only trends toward significance. In the analysis comparing number of homozygous variants, the combination of having both homozygous variants carried the greatest risk, with adjusted OR of 2.19 (95% confidence interval, 1.40–3.43), P=0.0006 (global Wald test, P=0.008). No associations were identified between the number of variants and clinical characteristics such as age, sex, disease stage, smoking status or histology (P>0.15 for each comparison). Sensitivity analysis revealed that the results were virtually identical when only NSCLCs were included in the analysis. In exploratory analyses (data not presented), late-stage cancers and lung adenocarcinomas had the strongest lung cancer risk associations when carrying homozygous variants of BRM promoter polymorphisms. Results of an exploratory haplotype analysis were weaker than the results of the homozygosity combination analysis shown above: the aOR of carrying one copy of the −1321 variant/−741 variant allele was 1.08 (95% confidence interval, 0.7–1.6), whereas the aOR of carrying two alleles was 1.65 (95% confidence interval, 0.9–2.2; global omnibus test, P=0.12), suggesting that the main association relates to having homozygosity in both loci, rather than to a haplotype effect. As in previous separate analyses, we had identified several additional promising compounds that promoted the re-expression of BRM (Gramling and Reisman, 2011) pharmacological methods of the future may modulate or reduce lung cancer risk in high-risk, homozygous variant-carrying smokers, through the restoration of BRM function.

Table 2.

(a) Case–control demographics and (b) case–control analysis of BRM polymorphisms

Characteristic	Cases	Controls	P-value
(a)
N	484	715	NA
Age
Mean (s.d.)	65 (10)	65 (7)	0.65 (t-test)
Gender, n (%)
Males	273 (60%)	409 (59%)	0.78 (χ2)
Females	211 (40%)	306 (41%)
Pack-years
Mean (s.d.)	42 (30)	35 (21)	<0.0001 (t-test)
Years quit for ex-smokers
Mean (s.d.)	16 (11)	20 (11)	<0.0001 (t-test)
Smoking status, n (%)
Current smokers	242 (50%)	363 (51%)	0.79 (χ2)
Ex-smokers	242 (50%)	352 (49%)
Histology, n (%)
Adenocarcinoma	280 (58%)	NA	NA
Squamous cell	106 (22%)
Large cell	28 (6%)
NSCLC NOS	49 (10%)
Adenosquamous	3 (1%)
Small cell	18 (4%)
Stage, n (%)
1	138 (30%)	NA	NA
2	46 (10%)
3	162 (35%)
4	112 (24%)

BRM polymorphism or combination	Cases N (%)	Controls N (%)	Crude OR (95%CI); P-value	Adjusted OR (95%CI); P-value
(b)
BRM-741 analysis
Wild type (reference)	122 (25%)	211 (30%)	1	1
Heterozygote	233 (48%)	362 (51%)	1.11 (0.8–1.5); 0.45	1.12 (0.9–1.5); 0.41
Homozygous variant	127 (27%)	42 (20%)	1.57 (1.1–2.2); 0.007	1.55 (1.1–2.2); 0.009
BRM-1321 analysis
Wild type (reference)	128 (26%)	245 (34%)	1	1
Heterozygote	244 (50%)	343 (48%)	1.36 (1.0–1.8); 0.02	1.41 (1.2–2.4); 0.01
Homozygous variant	112 (23%)	127 (18%)	1.69 (1.2–2.4); 0.002	1.74 (1.2–2.4); 0.002
Combined analysis
Wild type (reference)	74 (15%)	145 (20%)	1	1
No homozygous genotypes	239 (49%)	363 (51%)	1.29 (0.9–1.8); 0.12	1.32 (1.0–1.8); 0.10
One homozygous variant	101 (21%)	145 (20%)	1.32 (0.9–2.0); 0.11	1.40 (1.0–2.1); 0.09
Both homozygous variants	70 (14%)	62 (9%)	2.21 (1.4–3.4); 0.0004	2.19 (1.4–3.4); 0.0006

Abbreviations: 95% CI, 95% confidence interval; n, number; NA, not applicable; NSCLC, non-small cell lung cancer; OR, odds ratio. Consecutive cases of smokers with lung cancer at University Health Network (research ethics board approved; Toronto, Canada) were recruited for a molecular epidemiology study of risk and prognosis (Table 2). Eligible patients provided informed consent and a blood specimen, and completed an epidemiological questionnaire. Recruitment rate was 86%. analysis was restricted to Caucasians (88%) to avoid potential population stratification. Of 499 cases, 484 (97%) had complete clinical and genotyping data. Age, sex and smoking status (former versus current) frequency-matched healthy controls were obtained from self-referred participants of the local University Health Network lung cancer early detection program; all were ≥ 50 years and had >10 pack-years of smoking history. The 3% of younger healthy smoking controls and healthy smokers with <10 pack-years required to match the cases were recruited from visitors accompanying outpatients. Thus, all controls were self-referred. We restricted the analyses to self-identified Caucasian controls. A total of 715 Caucasian smoker controls with complete clinical and genotyping were analyzed. All the analyses were performed with SAS 9.3 (SAS Institute, Cary, NC, USA). In a pre-specified power calculation, assuming two-sided α =0.05, power=0.80, and a control prevalence of BRM homozygous insertion variant of 20%, the minimally detectable odds ratio was 1.48 for each BRM homozygous variant. In determining the association between BRM homozygous polymorphic variants and lung cancer risk, Table 2b presents the demographic variables. Table 2b shows the results of the analysis. In multivariate analyses, adjusted ORs included variables for age (continuous), gender, smoking status (current versus former smoker) and cumulative pack-years (continuous). Associations between BRM polymorphisms and lung cancer status were determined using logistic regression, with and without adjusting for covariates. As the variables—smoking status, years since quitting smoking and pack-years of smoking—were highly correlated, to avoid collinearity, adjusted models included smoking status (current versus former smokers) and cumulative pack-years, along with sex and age (continuous variable). OR and 95% CIs were generated. The discrete genetic model and a global Wald test were used to screen for significance, and exploratory additive, dominant and recessive models were used as appropriate. Subgroup and exploratory analyses were performed in specific clinical subgroups.

Discussion

As BRM seems to be an important protein for a number of critical cellular processes, it is likely to be tightly regulated through a number of different mechanisms. Although our data indicate that BRM can be upregulated by transcription after the application of HDAC inhibitors, our data cannot preclude additional regulatory mechanisms activated via other stimuli. To this end, Sakurai et al. (2011) have recently shown that the microRNA, miR-199a, is likely also involved in the downregulation of BRM expression thereby attesting to the complexity of the BRM regulatory mechanism. It is generally accepted that premicroRNAs are transported to the cytoplasm, where they are transformed into their active form and act to target specific mature mRNA in the cytoplasm (Chen and Meister, 2005; Yue and Tigyi, 2006). In comparison, mRNA splicing occurs primarily, if not exclusively, in the nucleus. This means that it is unlikely that microRNA can target BRM heterogeneous mRNA. As such, the findings of the rapid induction of heterogeneous BRM mRNA, together with our nuclear run-on data, indicate a role for transcription in BRM regulation. Interestingly, the microRNA miR-199a has also been found to be suppressed by BRM expression (Sakurai et al., 2011), suggesting the downregulation of BRM (for example, decreased transcription), causes further induction of miR199A, which in turn would cause further suppression of BRM. Thus, both transcriptional and post-transcriptional mechanisms seem to control the expression of BRM. Further studies will be needed to determine if this is a synergistic effect or if one mechanism predominates in specific situations.

Our data point to loss of BRM as a possible early event in cancer development, but just how this might impact cancer evolution is not yet exactly known. However, two complementary mechanisms could explain how the down-regulation of BRM can disrupt normal cellular growth and facilitate the emergence of cancer cells. First, growth control is known to rest with the function of the Rb protein. Although the loss of both BRG1 and BRM have been known for ~10 years to disrupt Rb growth control (Strober et al., 1996; Strobeck et al., 2000; Reisman et al., 2002), recently the selective loss of BRM has been shown to be sufficient to significantly impair Rb growth control (Bartlett et al., 2010). Besides, we have seen an inverse relationship with epidermal growth factor receptor and BRM in unpublished microarray experiments in the non-transformed cell line BEAS-2B (data not shown). Moreover, the immediate-early gene product EGR1, which can stimulate growth, has been shown to be inversely regulated by BRM (Sakurai et al., 2011). Hence, BRM loss can abrogate growth control as well as impart proliferation signaling. Again, studies will be needed to pinpoint just where BRM loss occurs and the major pathways affected by its loss that, in turn, could lead to the emergence of cancer. These data further strengthen, however, a central role for BRM loss in cancer development.

In lung cancer, biomarkers that can identify high-risk subsets of individuals are urgently needed. Surgical interventions are very effective in treating patients if lung cancer is caught early in its course. However, surgery typically only impacts a minor fraction of lung cancer patients because nearly two-thirds of patients present with advanced stage lung cancer. Screening procedures such as computerised tomography scanning of heavy smokers can potentially identify cancers when they are still curable early-stage tumors. Results of large-scale clinical trials of computerised tomography screening are promising. Yet, as only a tiny fraction of smokers develop lung cancer, the identification of additional risk biomarkers may help refine and improve lung cancer risk stratification, rendering radiological screening more efficient and effective. The novel BRM polymorphisms evaluated in this study may enhance our ability to determine patients who are at higher risk of developing a subset of BRM-driven lung cancer, allowing us to better target screening, prevention, and treatment strategies aimed at this subset of patients.

That the risk associations with BRM polymorphisms were first found in smokers is consistent with data showing that BRM-null mice develop higher rates of tumor formation (compared to BRM-positive mice) when exposed to carcinogens. BRM loss may impair a number of anticancer pathways: Rb-mediated growth inhibition, the function of the other Rb family members (p107 and p130), and p53 are each known to be functionally tied to BRM/BRG1 (Strober et al., 1996; Reisman et al., 2002; Strobeck et al., 2002; Wang et al., 2007; Xu et al., 2007; Oh et al., 2008; Naidu et al., 2009). Hence, loss of BRM function could weaken or even abrogate the growth controlling properties of these and possibly other anticancer proteins. Furthermore, a number of DNA repair proteins, such as p53, BRCA1, GADD45A, p21 and Fanconi anemia protein, are functionally tied to BRM (Bochar et al., 2000; Otsuki et al., 2001; Hill et al., 2004; Morrison and Shen, 2006) and the SWI/SNF complex. Additional studies have shown that SWI/SNF is essential for DNA repair (Gaillard et al., 2003; Park et al., 2006; Park et al., 2009), such that the loss of BRM would be expected to block DNA repair mechanisms. Because loss of DNA repair capacity has been repeatedly shown to facilitate cancer development, loss of BRM function could further potentiate cancer development.

To date, the major germline polymorphic risk factors for lung cancer that have been validated in multiple large datasets (Bailey-Wilson et al., 2004; Hung et al., 2008; Landi et al., 2009) have not been translated into clinical practice. Two major limitations to translation into the clinics are (i) compelling functional or clear-cut plausible explanations for these polymorphism associations have been absent or weak; and (ii) while smoking cessation is a general intervention for all at-risk individuals, no specific interventions have been identified that modulate any of the identified genetic risks. Thus, our results may be an improvement over previous results.

In summary, we have shown with multiple lines of evidence that BRM is regulated at the transcriptional level, in contrast to previous publications suggesting that BRM is regulated post transcriptionally (Yamamichi et al., 2005). In addition, we have shown that the homozygous variant insertion genotypes of two BRM promoter polymorphisms are tightly associated with BRM loss in both cell lines and NSCLC tumors. Further, these homozygous insertion variants of the two BRM polymorphisms are strongly associated with the development of lung cancer risk in smokers. These are the first findings of cancer genetic susceptibility within the chromatin-remodeling pathway, specifically involving the SWI/SNF complex and lung cancer.