Although tissue specificity is a widely known trait of the SWI/SNF complex, no studies have analyzed SWI/SNF composition in a lung epithelial cell model [8]. For this reason, first we aimed to identify which subunits constitute the SWI/SNF complex in lung epithelial cells. We performed an endogenous immunoprecipitation of SMARCA4 followed by liquid chromatography-tandem mass spectrometry/mass spectrometry (LC–MS/MS) in NL20, a non-tumorigenic bronchial epithelial cell line. Twenty SWI/SNF subunits were pulled down along with SMARCA4 (Additional file 1: Fig. S1, Additional file 2: Tables S4–S5). From now on, we will refer to the SWI/SNF subunits that were identified in the immunoprecipitation, plus SMARCA4 and SMARCA2, as “lung SWI/SNF subunits.”
To examine the mutational status of the lung SWI/SNF subunits, we performed targeted DNA sequencing in seventy LUAD primary tumors and twenty-seven of the matched normal adjacent samples. We used the information from a paired analysis on the twenty-seven matched tumor-normal pairs to optimize a pipeline for unpaired mutation calling in the seventy primary LUAD tumors (Additional file 1: Supplementary Methods and Supplementary Note). We analyzed the twenty lung SWI/SNF subunits that had good quality sequencing. In these lung SWI/SNF subunits, we found 38 point mutations and small indels in our LUAD patient cohort (N = 70). Twenty-nine (41.4%) of the primary tumors harbored at least one mutation in a lung SWI/SNF subunit (Fig. 1A). SMARCA4 was the most commonly mutated SWI/SNF gene (11.4% of samples), followed by ARID1A (8.6%), ARID2 (7.1%), ARID1B (4.3%), and PBRM1 (4.3%).
Next, to investigate the mutation frequencies in external LUAD cohorts, we examined publicly available data from TCGA-LUAD (last updated on October 1, 2019. N = 567). The distributions of clinical parameters were comparable between the two cohorts (Additional file 2: Table S1). Our cohort showed similar but slightly higher mutation frequencies in the lung SWI/SNF genes (Additional file 1: Fig. S2A). Overall, the total mutation frequency of the SWI/SNF complex was 41.4% in our cohort and 30.0% in TCGA-LUAD, possibly due to a greater coverage in our protocol or to differences in data analysis protocols. Furthermore, regardless of the cohort, SMARCA4, ARID1A, and ARID2 were the SWI/SNF subunits that accumulated the highest number of truncating mutations.
To predict the functional impact of missense mutations, we used the SIFT algorithm [9] (Additional file 1: Supplementary Methods and Fig. S2B). Based on SIFT predictions, more than half of the missense mutations in our cohort (64%, 16/25) and in the external data (65%, 103/159) were “deleterious.” Overall, considering the truncating mutations and the predicted deleterious missense mutations, more than 70% of the SWI/SNF mutations may have a functional impact.
To complement our mutational study, we analyzed the mRNA levels of the lung SWI/SNF subunits in our cohort. To measure expression accurately, we used RT-qPCR. We found that all lung SWI/SNF subunits were significantly downregulated in LUAD primary tumors compared to their matched normal adjacent samples (FDR-adjusted p < 0.05, Additional file 1: Fig. S3). We set a fold change threshold of + 2/− 2 between the tumor and the paired normal sample to consider a subunit to be up- or downregulated, respectively. Remarkably, most lung SWI/SNF subunits consistently showed lower expression in most tumors when compared to their paired normal tissues (Fig. 1B, Additional file 1: Fig. S3). We found 42 tumors (60%) that had more than 10 downregulated subunits. On average, each lung SWI/SNF subunit was downregulated in 57% of LUAD patients. The top downregulated SWI/SNF subunit was SMARCA2 (82% of the cases). Similar results have been observed in other tumors where SMARCA2 was found to be epigenetically repressed [10,11,12]. Moreover, none of the top 5 downregulated subunits (SMARCA2, DPF2, SMARCD3, PHF10 and SMARCD1) were among the top 5 most frequently mutated subunits. More generally, only 5/11 (45.5%) truncating mutations and 13/23 (56.5%) missense mutations were associated with more than a twofold decrease in expression. Overall, these findings suggest a profound silencing in the expression of the whole SWI/SNF machinery in LUAD and that genetic alterations are not the only cause of SWI/SNF inactivation.
In our mutational analysis, we also observed that SWI/SNF-mutant tumors from TCGA-LUAD showed a significantly higher Tumor Mutation Burden (TMB) than SWI/SNF-wild type tumors (p < 0.05) (Fig. 2A). Furthermore, we evaluated whether the mutational status of the lung SWI/SNF subunits was associated with LUAD overall survival in the TCGA-LUAD cohort. To select variables for a multivariate Cox analysis, we first performed univariate Cox analyses on each of the variables under study and we selected those with p < 0.2. We considered mutations in SWI/SNF and LUAD driver genes, TMB, and other clinically relevant covariates (Additional file 1: Supplementary Methods). In the univariate analysis, none of the individual SWI/SNF subunits were significantly associated with overall survival (OS), but SWI/SNF mutations altogether were significantly associated with poorer OS (HR = 1.42; 95% CI: 1.04–1.93; p = 2.5·10–2) (Fig. 2B and C). These observations led us to consider the SWI/SNF complex as a single functional unit. Mutations in none of the top 10 LUAD driver genes from Bailey et al. [7] were significantly associated with OS (Additional file 1: Fig. S4A–J). Next, all variables with p < 0.2 in the univariate analysis were used for a multivariate analysis. According to this analysis, the SWI/SNF mutational status is an independent prognostic factor associated to shorter OS in LUAD patients (HR = 1.45; 95% CI: 1.05—2.01; p = 2.56·10–2) (Fig. 2D). Therefore, the lung SWI/SNF mutational status distinguishes between two clinically different subgroups.