SWI/SNF Alterations in Squamous Bladder Cancers

Dysfunction of the SWI/SNF complex has been observed in various cancers including urothelial carcinomas. However, the clinical impact of the SWI/SNF complex in squamous-differentiated bladder cancers (sq-BLCA) remains unclear. Therefore, we aimed to analyze potential expression loss and genetic alterations of (putative) key components of the SWI/SNF complex considering the co-occurrence of genetic driver mutations and PD-L1 expression as indicators for therapeutic implications. Assessment of ARID1A, SMARCA2, SMARCA4, SMARCB1/INI1, SMARCC1, SMARCC2 and PBRM1 mutations in a TCGA data set of sq-BLCA (n = 45) revealed that ARID1A was the most frequently altered SWI/SNF gene (15%) while being associated with protein downregulation. Genetic alterations and loss of ARID1A were confirmed by Targeted Next Generation Sequencing (NGS) (3/6) and immunohistochemistry (6/116). Correlation with further mutational data and PD-L1 expression revealed co-occurrence of ARID1A loss and TP53 mutations, while positive correlations with other driver mutations such as PIK3CA were not observed. Finally, a rare number of sq-BLCA samples were characterized by both ARID1A protein loss and strong PD-L1 expression suggesting a putative benefit upon immune checkpoint inhibitor therapy. Hence, for the first time, our data revealed expression loss of SWI/SNF subunits in sq-BLCA, highlighting ARID1A as a putative target of a small subgroup of patients eligible for novel therapeutic strategies.


Introduction
In 2018, bladder cancer was the 10th most common cancer worldwide, with estimated 549,000 new cases and 200,000 deaths. It is more common in men than in women, with respective incidence and mortality rates of 9.6 and 3.2 per 100,000 men, about four times those of women globally [1]. Furthermore, bladder cancer is the second most common genitourinary malignancy [2]. The most significant risk factor is age and the median age at diagnosis is 70 years [3]. Because of the demographic change, it can be assumed that the number of new cases will increase in the near future. Over 90% of bladder cancers are urothelial carcinomas with distinct molecular characteristics for muscle-invasive bladder cancers (MIBCs) such as TP53 mutations or non-muscle-invasive bladder cancers (NMIBCs) including activating FGFR3 mutations or PIK3CA alterations [3]. Only 5% of all bladder cancers are

Patient Samples and Tissue Microarrays
Squamous differentiated bladder cancers were retrospectively collected from pathology archives of the German Study Group of bladder cancer (n = 68 pure, n = 48 Mixed) over 17 years (1998-2015). For cohort characteristics see Table 1. Tissue microarrays of formalin-fixed paraffin-embedded (FFPE) surgical specimens were used as previously described [32][33][34]. The RWTH University Hospital Aachen local ethics committee approved the retrospective, pseudonymized study of archival tissues (RWTH EK 009/12).

DNA Extraction
DNA extraction of FFPE tissue samples (n = 69 samples) was performed using a QIAamp DNA Mini Kit (Qiagen, Hilden, Germany) as previously described [36]. For Targeted Next Generation Sequencing (NGS) analyses, DNA was isolated from FFPE tissue (n = 6) using the automated Maxwell 16 system and corresponding FFPE Tissue LEV DNA Purification Kit (Promega, Mannheim, Germany) according to the manufacturer's instructions.

SNaPshot and Sanger Sequencing
PIK3CA and FGFR3 mutational analyses were performed using the SNaPshot method for the simultaneous detection of hotspot mutations according to Hurst et al. [37] and as described previously [38]. PCR-amplification and Sanger sequencing was performed for TP53 and CDKN2A as specified in 2016 [6]. For details of primer sequences and PCR conditions see Supplementary Table S1.

Targeted Next Generation Sequencing (NGS) Analysis
Targeted NGS was conducted for n = 6 samples to evaluate the ARID1A, SMARCA4 and SMARCB1 mutation status. . Changes with an allele frequency above 10% were taken into account if not already classified as known artifacts for the panel. Further variant filtering was conducted as follows: Missense variants with an allele frequency >2% in the normal population (according to 1000 Genomes (http://www.internationalgenome.org, last accessed 2nd September 2019) or dbSNP v153 (https://www.ncbi.nlm.nih.gov/snp, last accessed 27 September 2019)) and non-splicing-relevant silent, untranslated region (UTR) and intronic variants not affecting the canonical splice-site were considered benign. Additionally, we excluded missense variants classified as benign or likely benign in the ClinVar database (https://www.ncbi.nlm.nih.gov/clinvar, last accessed 27 September 2019). A possible effect of mutations on splicing was determined using Alamut software (Alamut VISUAL v2.14.0, SOPHiA interactive biosoftware, Rouen, France) (included splicing predictions: SpliceSiteFinder-like, MaxEntScan, NNSPLICE 0.9, GeneSplicer; for detailed information see software documentation). For ARID1A, SMARCA4 and SMARCB1 copy number variation (CNV) analysis, an in-house algorithm (validated using three NGS panels; >150 samples) was used. The ACopy tool is based on an exponential growth model for amplification of PCR products [39].

Analysis of the TCGA Sq-BLCA Data Set
To identify mutations for seven genes of the SWI/SNF-complex of SCC in the TCGA-BLCA cohort [40] we extracted the patient IDs for samples classified as "NOS with squamous differentiation" (n = 42) and "Squamous cell carcinoma" (n = 3) based on the pathologic classification described by Robertson et al. [40]. No further publicly available platforms providing additional squamous bladder cancer data sets exist. Not without reason, in particular pure, non-schistosomiasis-associated SCC is a rare disease in the western world, with very low incidence rates (e.g., 0.6-1.2 per 100,000 person-years) [41]. Assessment of genetic alterations and the mutation spectrum was performed using cBioPortal (https://www. cbioportal.org/, [42,43]) filtering for the extracted patient IDs.

Statistical Analysis
Statistical analysis was performed using Statistical Package for the Social Sciences (SPSS) software version 26.0 (SPSS inc., Chicago, IL, USA). p-values < 0.05 were considered significant. Statistical associations between clinico-pathological and molecular factors were determined by Fisher's exact test. Survival curves for recurrence-free (RFS) and overall survival (OS) were calculated using the Kaplan-Meier method with log-rank statistics. RFS/OS was measured from surgery until relapse/death and was censored for patients alive without evidence of relapse/death at the last follow-up date. Correlation analysis was performed by calculating a Spearman's rank correlation coefficient. The Expression correlation network was plotted using the R package "Rgraphviz" [44], and significant correlations (Spearman correlation coefficient, p < 0.05) between subunits were plotted as edges.

Analysis of Frequently Altered Subunits of the BAF and PBAF SWI/SNF Complexes in TCGA Sq-BLCA
In order to give first insights into the mutational status of (putative) key components of the SWI/SNF complex in squamous bladder cancers (for study design see Supplementary Figure S1), carcinomas with histologically squamous differentiation (n = 3 pure SCC and n = 42 MIX) of The Cancer Genome Atlas (TCGA) were analyzed for genetic alterations of seven frequently affected subunits of the SWI/SNF complexes BAF and PBAF [11,12]. Alterations of the BAF-specific component ARID1A were the most frequent events (15.2%, 7/46), comprising one deep deletion, four truncating mutations and two missense mutations indicating impaired protein function ( Figure 1A,B).
Genes 2020, 11, x FOR PEER REVIEW 5 of 15 Statistical analysis was performed using Statistical Package for the Social Sciences (SPSS) software version 26.0 (SPSS inc., Chicago, IL, USA). p-values < 0.05 were considered significant. Statistical associations between clinico-pathological and molecular factors were determined by Fisher's exact test. Survival curves for recurrence-free (RFS) and overall survival (OS) were calculated using the Kaplan-Meier method with log-rank statistics. RFS/OS was measured from surgery until relapse/death and was censored for patients alive without evidence of relapse/death at the last followup date. Correlation analysis was performed by calculating a Spearman's rank correlation coefficient. The Expression correlation network was plotted using the R package "Rgraphviz" [44], and significant correlations (Spearman correlation coefficient, p < 0.05) between subunits were plotted as edges.

Analysis of Frequently Altered Subunits of the BAF and PBAF SWI/SNF Complexes in TCGA Sq-BLCA
In order to give first insights into the mutational status of (putative) key components of the SWI/SNF complex in squamous bladder cancers (for study design see Supplementary Figure S1), carcinomas with histologically squamous differentiation (n = 3 pure SCC and n = 42 MIX) of The Cancer Genome Atlas (TCGA) were analyzed for genetic alterations of seven frequently affected subunits of the SWI/SNF complexes BAF and PBAF [11,12]. Alterations of the BAF-specific component ARID1A were the most frequent events (15.2%, 7/46), comprising one deep deletion, four truncating mutations and two missense mutations indicating impaired protein function ( Figure 1A,B).  Only low mutation frequencies were observed in sq-BLCA for components potentially involved in assembly of both complexes-i.e., BAF and PBAF: SMARCA4 6.5% (3/46), SMARCC2 4.3% (2/46), SMARCA2 2.2% (1/46), SMARCC1 2.2% (1/46) and SMARCB1 0% (0/46). The gene encoding the PBAF-specific subunit PBRM1 was mutated in 4.3% of samples (2/46) ( Figure 1A). Determining the ARID1A protein level in dependency of its mutational status, we confirmed a significantly lower expression in tumors with genetic alterations of the ARID1A gene including missense mutations ( Figure 1C)-i.e., frequent ARID1A mutations correlate with loss of ARID1A protein in TCGA Sq-BLCA.
As genetic alterations of SWI/SNF components are known to be associated with patients' outcome, we correlated SWI/SNF mutations with clinico-pathological parameters and analyzed recurrence-free (RFS) and overall survival (OS) as an indicator of potential prognostic impact. We focused on patients with at least one genetic alteration in one or more of the analyzed components as well as on those harboring ARID1A mutations (missense vs. nonsense). Using a Fisher's exact test, no associations of SWI/SNF mutations or ARID1A mutations with clinico-pathological characteristics were observed (Supplementary Tables S2-S4). Kaplan-Meier analysis did not show any association of mutated SWI/SNF components and/or ARID1A mutations with RFS and OS (Supplementary Figure S2).

Immunohistochemical Analysis of Frequently Altered Subunits of the SWI/SNF-Complex in an Independent Squamous Bladder Cancer Cohort
Next, n = 116 samples of patients with pure squamous cell carcinoma (n = 68) and mixed urothelial carcinoma with substantial squamous differentiation (n = 48) were analyzed for seven SWI/SNF complex proteins (ARID1A, SMARCA4, SMARCB1, SMARCC1, SMARCC2, SMARCA2 and PBRM1) by immunohistochemistry (Figure 2A). In total, 68.1% of the carcinomas presented as high-grade cancers, while except for one sample, all bladder tumors showed muscle-invasion (for cohort characteristics see Table 1).

Correlation of ARID1A Expression Loss with Clinico-Pathological Parameters and Known Genetic Drivers
Next, we tested associations between expression loss of analyzed subunits with both clinico-pathological characteristics and driver mutations as indicators for prognostic risk and therapeutic implication ( Figure 3A,B). Novel PIK3CA mutational analyses were performed by SNaPshot according to Hurst et al. [37], while PCR-amplification and Sanger sequencing was performed for CDKN2A as specified in 2016 [6]. All findings were correlated including previously published mutational data for TP53 and FGFR3 [5]. With the exception of ARID1A, neither associations between expression loss and clinico-pathological parameters-such as age at diagnosis, tumor size and histological tumor type -nor driver mutations were found. ARID1A expression did not significantly correlate with clinico-pathological parameters (Table 3). Please note that the sample number showing ARID1A expression loss is limited, which could affect statistical accuracy. However, all six tumors that did not express ARID1A were diagnosed in advanced stages-i.e., n = 5 pT3, n = 1 pT4, n = 2 with positive lymph node status, and n = 4 with high-grade differentiation. ARID1A expression was further significantly associated with TP53 mutations (p < 0.05) ( Table 3). Interestingly, ARID1A expression loss was not observed in tumors with genetic alterations of FGFR3 or PIK3CA.

Correlation of ARID1A Expression Loss with Clinico-Pathological Parameters and Known Genetic Drivers
Next, we tested associations between expression loss of analyzed subunits with both clinicopathological characteristics and driver mutations as indicators for prognostic risk and therapeutic implication ( Figure 3A,B). Novel PIK3CA mutational analyses were performed by SNaPshot according to Hurst et al. [37], while PCR-amplification and Sanger sequencing was performed for CDKN2A as specified in 2016 [6]. All findings were correlated including previously published mutational data for TP53 and FGFR3 [5]. With the exception of ARID1A, neither associations between expression loss and clinico-pathological parameters-such as age at diagnosis, tumor size and histological tumor type -nor driver mutations were found. ARID1A expression did not significantly correlate with clinico-pathological parameters (Table 3). Please note that the sample number showing ARID1A expression loss is limited, which could affect statistical accuracy. However, all six tumors that did not express ARID1A were diagnosed in advanced stages-i.e., n = 5 pT3, n = 1 pT4, n = 2 with positive lymph node status, and n = 4 with high-grade differentiation. ARID1A expression was further significantly associated with TP53 mutations (p < 0.05) ( Table 3). Interestingly, ARID1A expression loss was not observed in tumors with genetic alterations of FGFR3 or PIK3CA.

ARID1A Protein Loss Overlaps with Genetic ARID1A Alterations and PD-L1 Expression in the Independent Squamous Bladder Cancer Cohort
Since ARID1A loss seems to be associated with advanced tumor stages, we focused on this important SWI/SNF component with therapeutic potential to confirm that ARID1A protein loss results from genetic ARID1A gene alterations. Six tumors (n = 4 MIX, n = 2 SCC) with loss of ARID1A expression ( Figure 4)

ARID1A Protein Loss Overlaps with Genetic ARID1A Alterations and PD-L1 Expression in the Independent Squamous Bladder Cancer Cohort
Since ARID1A loss seems to be associated with advanced tumor stages, we focused on this important SWI/SNF component with therapeutic potential to confirm that ARID1A protein loss results from genetic ARID1A gene alterations. Six tumors (n = 4 MIX, n = 2 SCC) with loss of ARID1A expression ( Figure 4)    . The mutation c.4005-2A>G affects the canonical splice site and leads to loss of the acceptor splice site according to distinct prediction tools. All three mutations are, therefore, most likely deleterious and probably lead to a nonsense-mediated decay of the truncated protein and, therefore, to a protein loss.
It is thought that ARID1A-mutated cancers may cooperate with immune checkpoint blockade therapy [19], thus providing novel therapeutic strategies for cancer management. As we recently showed that PD-L1 was frequently expressed in squamous bladder cancer [45], ARID1A alterations were correlated with expression of PD-L1 using the 28-8 antibody clone as an indicator of immune checkpoint inhibitor (ICI) treatment access. According to current European Medicines Agency (EMA)-approved guidelines for first line therapy of bladder cancer with pembrolizumab (CPS ≥ 10) and atezolizumab (IC-score ≥ 2/IC ≥ 5%), a single ARID1A-mutated cancer (1/3) was identified to be potentially eligible for atezolizumab first line therapy (Supplementary Table S5). Considering protein loss of ARID1A, another case was revealed-i.e., overall two SCC/MIX specimens of the urinary bladder were characterized by ARID1A protein loss and strong PD-L1 expression (IC-score ≥ 2/IC ≥ 5%) suggesting a putative synergistic impact and improved ICI therapy success similar to recent studies in urothelial cancers [24].

Discussion
To date, dysfunction of the components of the SWI/SNF complex has been shown for various cancer entities [17] including urothelial cancer [13]. TCGA data demonstrate that ARID1A is among the most frequently mutated genes among different types of cancer, such as stomach adenocarcinomas (18-31%) or uterine corpus endometrioid carcinomas (34%) [46]. Most ARID1A mutations are inactivating truncating mutations [46]-e.g., 63% of ARID1A gene alterations in urothelial carcinomas [30] or over 90% of ARID1A mutations in ovarian clear cell carcinoma [47]. ARID1A mutated carcinomas are associated with poor prognosis, and for instance, in breast cancer patients, inactivated ARID1A suggests a tumor suppressive function [48,49].
Recently, we revealed frequent genetic alterations of genes encoding for SWI/SNF subunits including ARID1A with a frequency of 26% in urothelial bladder cancer [30]. Most of these mutations, in particular truncating alterations, are likely associated with a functional loss of proteins. In line with this, we identified ARID1A mutations in 15% of sq-BLCA of the TCGA data set. The pathological/functional significance of identified missense mutations remains elusive; however, by stratifying the mutations we significantly observed reduced ARID1A protein levels for both-i.e., for nonsense mutations as well as for the combined group of nonsense and missense mutations. Further mechanisms potentially involved in gene silencing such as epigenetic silencing or mutations in non-coding regions as well as post-transcriptional or translational modifications [50] might be likely. Wu and colleagues showed, for instance, that heterozygous ARID1A mutations correlated with loss of protein expression-i.e., 73% of tumors with heterozygous ARID1A mutations lacked protein expression [48]-suggesting a second hit on the remaining allele. Considering that, we confirmed missense and nonsense mutations of ARID1A in both pure and mixed SCC samples which were characterized by ARID1A protein loss. Interestingly, we found that protein loss of the BAF-specific subunit ARID1A was closely associated with expression loss of the commonly shared and central subunits SMARCA4 and SMARCC1, as well as with the BAF/PBAF-associated factor SMARCC2. In turn, ARID1A expression correlates with all analyzed components potentially involved in assembly of the BAF complex [11,12] suggesting a predominant role of this canonical SWI/SNF complex in SCC. However, we are aware that a statistical correlation does not provide the exact protein interaction in single cells. In addition, residual subunits could be of importance to compensate missing factors, thus maintaining the SWI/SNF activity as, for instance, already demonstrated for the catalytic subunits SMARCA4 and SMARCA2. Both subunits have been shown to be mutually exclusive subunits in SWI/SNF complexes, and survival of SMARCA2-mutated cells depends on the residual SMARCA4-containing complex activity in specific tumor entities [51]. Thus, future studies addressing the role and function of the different SWI/SNF complexes and their corresponding subunits are required to decipher the mechanisms behind this in sq-BLCA.
Besides involvement of ARID1A in SW/SNF-mediated chromatin remodeling, ARID1A is thought to contribute to DNA damage repair, especially DNA double strand break (DSB) repair [50]. It has been shown that suppression of ARID1A led to a higher cellular sensitivity to cisplatin due to higher rates of DSB, triggered by deficient DNA repair [52]. In endometrial carcinomas, ARID1A mutations are associated with mismatch repair deficiency and normal p53 expression [53]. Bosse and colleagues showed a nearly mutual exclusivity of ARID1A loss and mutant-like TP53 expression, while alterations of the PI3K-AKT pathway were more frequent when ARID1A expression was lost [54]. Coexistence of PIK3CA and ARID1A mutations has been shown before [55], whereas association of both events was not observed in our SCC/MIX samples of the urinary bladder. We are aware of the potential bias due to the low number of PIK3CA mutations; however, none of the tumors lacking ARID1A expression showed evidence for any of the analyzed driver mutations (i.e., PIK3CA and CDKN2A). Thus, a hypothesized causal and functional link between both events (e.g., PIK3CA and ARID1A mutations/expression) seems unlikely in squamous bladder cancer. In turn, ARID1A loss occurred in a TP53-deficient genetic background suggesting a regulation of potentially different biological processes to those described to date [56], but affecting cell cycle control which should be further studied in more detail in the future.
However, accumulating studies propose the involvement of functional ARID1A loss in synthetic lethality, which contributes to the response to various classical [20,21] and novel therapeutic options, including immune checkpoint inhibitors (ICI) [19,22,23]. Goswami and colleagues have recently shown that ARID1A mutation in combination with immune cytokine CXCL13 expression predicts response to immune checkpoint inhibitors in metastasized bladder cancers [24]. As we already provided a rationale for ICI treatment of SCC of the urinary bladder [45], ARID1A protein loss may predict increased efficiency of ICI therapy. However, a correlation between ARID1A mutations and increased PD-L1 expression as previously reported [19] could not be confirmed in sq-BLCA. The co-occurrence was rare, and only a subgroup of patients with ARID1A mutations may benefit from ICI treatment. In addition, Fukumoto and colleagues showed that inhibition of histone deacetylase 6 (HDAC6) contributes to growth suppression of ARID1A-mutated tumors, while synergistic effects were shown in combination with anti-PDL1 therapy [25]. Although the clinical results of HDAC-inhibitors have generally been disappointing in the past [26], current studies indicate a specific targeted benefit applying HDAC-inhibitors in tumors with ARID1A loss [25,27]. Thus, ARID1A might be used as an additional biomarker for clinical response to both HDAC inhibition and anti-PD-L1 therapy, albeit its function as a biomarker has only been described for patients with advanced urothelial carcinoma yet [27]. Further clinical trials may be necessary to prove the possible synergistic effect of both HDACand PD-L1-inhibitors on squamous bladder cancer cells with ARID1A mutation.
In conclusion, we provide, for the first time, data describing expression loss of components of the SWI/SNF-complex in sq-BLCA including pure SCC, highlighting ARID1A as an interesting target for a small subgroup of patients which may benefit from novel therapeutics in an ARID1A mutated background.