BRD9 EXPRESSION, ALTERATION, SURVIVAL AND PATHWAY ANALYSIS IN 11 INDEPENDENT PROSTATE CANCER COHORTS.

Background and aims: Despite recent advances in advanced prostate cancer treatments, there are no clinically useful biomarkers or treatments for men with such cancers. Targeted therapies have shown promise, but there remain fewer actionable targets in prostate cancer than in other cancers. This work aims to characterize BRD9, currently understudied in prostate cancer, and investigate its co-expression with other genes to assess its potential as a biomarker and therapeutic target in human prostate cancer. Materials and methods: Omics data from a total of 2053 prostate cancer patients across 11 independent datasets were accessed via Cancertool and cBioPortal. mRNA expression and co-expression, mutations, amplifications, and deletions were assessed with respect to key clinical parameters including survival, Gleason grade, stage, progression and treatment. Network and pathway analysis was carried out using Genemania, and heatmaps were constructed using Morpheus. Results: BRD9 is overexpressed in prostate cancer patients, especially those with metastatic disease. BRD9 expression did not differ in patients treated with second generation antiandrogens versus those who were not. BRD9 is co-expressed with many genes in the SWI/SNF and BET complexes, as well as those in common signaling pathways in prostate cancer. Summary and conclusions: BRD9 has potential as a diagnostic and prognostic biomarker in prostate cancer. BRD9 also shows promise as a therapeutic target, particularly in advanced prostate cancer, and as a co-target alongside other genes in the SWI/SNF and BET complexes, and those in common prostate cancer signalling pathways. These promising results highlight the need for wider experimental inhibition and co-targeted inhibition of BRD9 in vitro and in vivo, to build on the limited inhibition data available.


INTRODUCTION Prostate cancer and its biomarkers
Prostate Cancer (PCa) is the second most common cancer in men and the most commonly diagnosed in the UK, representing 7.1% of new cancers and 3.8% of cancer deaths worldwide [1][2][3][4].
Prostate-Specific Antigen (PSA) is a protein typically produced in the prostate and used as a biomarker to prompt further investigation (e.g. biopsy) for PCa and its recurrence) [5]. PSA does not provide any information about the aggressivity and stage of any potential cancer and testing is often associated with overdiagnosis and overtreatment, with a recent meta-analysis finding that screening did not impact overall mortality despite men who were screened having a higher incidence of PCa [6]. There are efforts to improve the diagnosis and discovery of dx biomarkers including the United States Food and Drug Administration (FDA)-approved Progensa assay (PCA3 (prostate cancer gene 3) and PSA) and STHLM3 (Stockholm-3) which is currently being validated [7][8][9]. However, these are imperfect and more sensitive and specific biomarkers are needed both for detecting PCa and driving treatment [10].

Current therapies for PCa
Though PCa survival is relatively high (85% five-year survival), this is much reduced (50%) for stage four PCa [11]. Treatments are often invasive with side effects for radical prostatectomy (surgical removal of the prostate) and radiotherapy, including urinary incontinence and erectile dysfunction [12,13]. Therapies for PCa range from hormonal therapies (including androgen deprivation therapy (ADT)) to chemotherapy and recentlyapproved targeted therapies. Targeted therapies aim to inhibit specific proteins and are a promising future area of therapeutics for PCa, with the recent FDA approval of the poly-ADP ribose polymerase inhibitors (PARPi) Olaparib and Rucaparib for metastatic castrationresistant prostate cancer (mCRPC) [14].
With poor outcomes for patients with advanced stage cancer and the development of resistance to existing therapies, BRD9 may have utility along the PCa clinical pathway as a biomarker, and a therapeutic target. New targeted therapies with less toxic and life-altering effects would be of great benefit in treating patients with PCa.

SWI/SNF and BET complexes
BRD9 is a genetic subunit of the non-canonical barrier-to-autointegration factor (ncBAF), also termed GBAF. This is one of the subunits composing the Switch/Sucrose Non-Fermentable (SWI/SNF) complex along with canonical barrier-to-autointegration factor (cBAF) and Polybromo-barrier-to-autointegration factor (PBAF). These complexes are involved in chromatin remodelling and cancer, containing both oncogenes and tumour suppressor genes [15]. Similarly, the Bromodomain and extra-terminal domain (BET) complex is involved in regulating transcription by RNA polymerase II [16]. This represents a possible therapeutic target in PCa with the selective small-molecule BET inhibitors JQ1 and I-BET preventing growth of androgen receptor (AR) positive castrate-resistant prostate cancer (CRPC) cell lines [17].

Common signaling pathways in PCa
Pathways involved in driving PCa proliferation involve the JAK-STAT (janus kinase; signal transducer and activator of transcription), MAPK (mitogen-activated protein kinase) and PI3K-AKT-mTOR (phosphoinositide 3-kinase; Protein Kinase B; mammalian target of rapamycin) pathways.
The JAK-STAT pathway is regulated by SOCS (suppressor of cytokine signaling) genes and is involved in regulation of "cell growth, differentiation, proliferation, invasion, survival, and inflammation" [18]. Its persistent activation in PCa can result in progression, rendering it a worthwhile therapeutic target, with one study finding STAT5A/B gene amplification and increased protein expression in PCa, and to a greater extent in higher grade, castrationresistant and metastatic disease [18,19]. The MAPK pathway is also involved in cell regulation and represents a possible therapeutic target in PCa, with one study finding patients with increased nuclear MAPK protein expression more likely to develop CRPC, earlier biochemical relapse and reduced survival [20,21]. Another study found 32% of patients with mCRPC have frequent amplifications in MAPK pathway-involved genes, and that this pathway is a promising target for mCRPC, with extracellular receptor kinase 1 (ERK1) inhibitor trametinib (currently approved for melanoma) being investigated for mCRPC in a phase II trial [22]. PI3K-AKT-mTOR is another often-dysregulated signaling pathway, especially in CRPC [23]. Though no drugs targeting this pathway have been approved in PCa, the AKT-inhibitor ipatasertib performed well in combination with abiraterone in a phase Ib/II study, and the IPATential150 phase III trial is underway [24,25]. Initial data shows a conclusions to be drawn [26]. Transmembrane protease, serine 2-erythroblast transformationspecific (TMPRSS2-ETS) genetic fusion is a molecular subtype (ETS-related gene (ERG) positive) of PCa associated with cancer invasiveness found in 50% of PCa tumours [27,28].
These pathways may represent opportunities for co-targeting with BRD9 as well as being targets themselves.

BRD9 and PCa
BRD9 has not been well studied in prostate cancer to date with one paper published in December 2020 [30]. This paper showed that BRD9 inhibition and knockdown have overlapping effects, reducing AR-positive cell line growth both in vivo and in vitro. The authors also found that BRD9 interacts with the AR in PCa cell lines, even those resistant to androgen deprivation and inhibition. They also provided cell-line evidence that BRD9 and the BET complex associate with each other and interact given they have overlapping transcriptional targets.
Despite being the first paper on BRD9 in PCa, the authors did not begin with an unbiased characterisation of this gene in human PCa using extensive existing publicly available data.
Instead they focussed almost exclusively on CRPC (developed by 10-20% PCa patients within 5 years), and worked with cell lines [31]: Much remains to be answered regarding BRD9's behaviour across the clinical pathway. However, the authors' findings were encouraging, warranting further investigation which could lead to clinical trials if future results continue to have promise.
Here we set out to determine whether BRD9 represents a potential biomarker, monotherapeutic target and co-target alongside other targeted therapies in PCa. 6

Cancertool, cBioPortal and Genemania
Cancertool was used to access BRD9 messenger ribonucleic acid (mRNA) expression in PCa cohorts as well as gene mRNA correlations in PCa. It is a publicly available interface that generates graphical visualisations of data, performs some statistical analyses and allows users to download raw data for processing and analysis [32].
The cBioPortal for Cancer Genomics was also used to access PCa cohort data. This platform integrates various genomic data, allowing users to investigate, view and download data relating to gene expression, mutation data (for some cohorts) and clinical attributes [33,34].
Genemania performs network analyses between inputted genes and presents relationships as a colour-coded network [35].

Clinical correlations and co-expression analysis
Clinical correlations were investigated by downloading raw data from Cancertool and cBioPortal and processing the data using Microsoft Excel. Following this, data were graphed, and statistical analyses carried out using GraphPad Prism version 9.1.0.
Co-expression between BRD9 and genes involved in signalling pathways and complexes of interest were investigated using Genemania and Cancertool. Spearman's rank-order correlation coefficient (R) and its statistical significance were calculated in Cancertool.
Correlation coefficient data was then processed using Microsoft Excel and heatmaps were created using Morpheus to visually represent the correlation between BRD9 expression and expression of other related genes of interest [47]. Once generated, the heatmaps were examined and p values were manually indicated by adding '*' to any pixel whose correlation was significant (α=0.05).

Comparing continuous data
Before comparisons were performed, data were graphed and tested for normality using Shapiro-Wilk and Kolmogorov-Smirnov tests. All data were unpaired. When comparing two groups, normally distributed data were analysed using Welch's t-test, as group standard deviations differed. Non-normally distributed data were analysed using a Mann-Whitney U test. Where more than two groups were to be compared, normally distributed data were analysed using the Brown-Forsythe and Welch's analysis of variance (ANOVA) test as group standard deviations differed. Dunnett's T3 test was then performed as a post-hoc multiple comparisons test. Non-normally distributed data were analysed using the Kruskal-Wallis test.
Dunn's test was then performed as a post-hoc multiple comparisons test.
Following statistical analysis, data were graphed using violin plots, as these better illustrate data distribution and density than bar graphs and box and whisker plots [48].

Correlations
Spearman's R correlations could be classified very weak, weak, moderate, strong or very strong (

Survival analysis
Disease-free Survival (DFS) was illustrated using Kaplan-Meier survival curves. Patients were grouped by quartile expression of BRD9. A Mantel-Cox test was performed to compare the differences between curves and a cox regression model was used to calculate hazard ratio (HR) between the first and fourth quartiles.

Comparing mutation distribution
Where mutation data was available, this was graphed using stacked bar charts. As categorical variables investigated had small sample sizes and expected frequencies<1 for some cells, Fisher's exact test was required to analyse mutation distribution. This test was implemented in R. In some instances, modified estimates of the p-value were necessary as certain tables were too computationally intensive to calculate. The approximation was calculated via a Monte Carlo simulation (number of samples = 1,000,000). In every instance, the p-values indicated that the data were considerably below the threshold for significance (p<10 -5 ).

RESULTS
BRD9 has potential as a diagnostic and prognostic biomarker in PCa.
BRD9 is overexpressed in human prostate tumour tissue, playing a role as a diagnostic biomarker. It may also be overexpressed in more aggressive PCa and this overexpression may correlate with reduced survival in patients. Comparison of BRD9 expression in the Grasso cohort revealed that BRD9 is overexpressed in PCa patients (p=0.0462) (Figure 1a).
In the Taylor 1j). In the Glinsky cohort, patients with higher BRD9 expression had significantly higher DFS (p=0.0190) (Figure 1k). In the Taylor cohort, there was no significant difference (p=0.8532) in DFS between patients with the lowest and highest BRD9 expression ( Figure   1l). In the TCGA cohort, patients with higher BRD9 levels had significantly lower DFS than those with lower levels (p=0.0015) (Figure 1m).
prevalent in patients with more advanced disease (p<0.0001) (Supplementary figure 1b), however this did not appear to be the case in the Ren cohort (p=0.7062) (Supplementary   figure 1d).
BRD9 does not appear to play a role as a predictive biomarker in PCa.
BRD9 expression levels were assessed in patient responses to therapy and in patients undergoing various therapeutic regimens. Data regarding this was lacking in most cohorts.
There was no significant difference in BRD9 expression between primary therapy outcomes in the TCGA cohort (α=0.05) (Figure 2a). BRD9 mutations were less prevalent in patients with a complete response to primary therapy in this cohort (p<0.0001) (Figure 2b).  BRD9 has shared protein domains with genes found in the BET complex, physically interacted with BRD4 and was co-expressed with BRD2 ( Supplementary Figure 3b).

BRD9 correlates with genes involved in common PCa proliferation-driving pathways
The correlation of BRD9 with known genes involved in the JAK-STAT, MAPK and PI3K-AKT-mTOR pathways was investigated ( Figure 5). We found that BRD9 is negatively correlated with JAK2 and positively with TYK2. BRD9 was also positively correlated with STAT2, STAT4, STAT5A, STAT5B and STAT6 as well as PIM1, PIM2 and PIM3 at a significant level (α=0.05). The cohorts disagreed on its correlation with BCL2 and MYC; Taylor demonstrated a significant negative correlation between BRD9 and BCL2 however

DISCUSSION
Though BRD9 is understudied in PCa, growing interest in the community has recently pointed to a potential role in other cancers. Here, we set out to characterise BRD9 expression and mutations in depth within 11 independent PCa cohorts, to identify whether BRD9 has potential as a biomarker and therapeutic target or co-target in PCa.

BRD9 as a biomarker
SWI/SNF subunits are the most commonly mutated chromatin-regulatory complexes in human cancer, mutated in 19.6% cancers, with subunits ARID1A, PBRM1, SMARCA4, and ARID2 already part of routine cancer diagnostics [15,51]. The literature suggests BRD9 plays an oncogenic role in many cancers with promise both as a diagnostic and prognostic biomarker [52][53][54]. The only paper on BRD9 and PCa used the TCGA evidence to show BRD9 worsened DFS, however did not investigate BRD9 as a biomarker in PCa [30].
Our findings show BRD9 is overexpressed in PCa and therefore may play a role as a diagnostic biomarker in this disease. Currently, researchers are investigating panels of genes rather than individual genes in PCa diagnostics [55]. BRD9 could be a useful panel component given most genes composing a biomarker panel would not exhibit as strong correlations with disease status as BRD9 exhibits. Though BRD9 expression does not change with increases in Gleason grade and tumour stage, its expression was higher in metastatic PCa when compared to cancer-free patients and patients with local disease, suggesting BRD9 could play a role as a prognostic biomarker and therapeutic target, especially in metastatic PCa. As with any study of historic data, it should be noted that analysis methods have changed over time here. For Gleason scoring, both pathological methods and WHO classifications have changed during the collection of these data, which could affect the conclusions here, and is an unavoidable limitation in the use of publicly available retrospective cohorts. It has been shown that BRD9 is required for cell growth and its degradation prevents synovial sarcoma tumour progression [52,53]. Similar findings have been reported in hepatocellular carcinoma (HCC) with BRD9 overexpressed in HCC patients as well as promoting cell growth and metastasis, and its depletion and inhibition reducing these effects in HCC cells [54]. Data on DFS and BRD9 expression was conflicting, however the widely-studied TCGA cohort with the largest sample size found patients with higher  [57,58,59]. We look forward to clinical outcome data becoming available, allowing us to further investigate these findings in more relevant settings.
AR expression is moderately negatively correlated with AR expression in the TCGA cohort.
Though this cohort was treatment-naïve, the correlation suggests that BRD9 could be a viable target in patients with AR-negative disease. After PCa differentiates into CRPC (stops responding to ADT) and is treated with second generation antiandrogens, it may stop responding to these also and then de-differentiate into AR-negative disease [60]. Though not in line with existing literature, if successful in AR-negative PCa, targeting BRD9 could be clinically very useful: following the introduction and approval of second-generation antiandrogens the percentage of patients with AR-negative tumours has increased from 11.7% (1996-2011) to 36.6% (2012-16) [61]. Prognosis for AR-negative CRPC is poor with a median survival of 12 months for AR-negative small cell carcinoma [62]. A recent randomised-control trial found that the targeted PARPi Olaparib (recently FDA approved for PCa in May 2020) increased progression-free survival in patients with mCRPC whose cancer progressed despite abiraterone or enzalutamide treatment [63]. This suggests, like Olaparib, BRD9 may be a good therapeutic target in patients who stop responding to second generation antiandrogens. Another study found BRD9 depletion increased cancer cell chemotherapy sensitivity as well as sensitivity to Olaparib in Ovarian cancer, suggesting a drug targeting BRD9 may work synergistically with Olaparib [56].

Co-targeting BRD9
The literature suggests that suggest BRD9 interacts with other genes in the SWI/SNF complex and cooperates with BET proteins [30,64].
We provide evidence showing that BRD9 correlates and associates with many genes in the SWI/SNF complex as well as most genes in the BET complex. BRD9 could therefore be a viable co-target alongside other genes in these complexes. Though their positive correlation is very weak, GLTSCR1 and its paralog GLTSCR1-like (GLTSCR1-L), the other unique GBAF complex components, interact with BRD9 [64]. Interestingly, BRD9 appears to be weakly positively correlated with SMARCB1 despite these genes being intrinsic competitors during SWI/SNF complex formation [65]. Alpsoy [68]. Interestingly BRD9 is negatively correlated with PIK3CA, its correlation with KRAS is non-uniform though it has been shown that BRD9 may mediate a PI3KCA-KRAS mutant oncogenic cancer phenotype [67]. BRD9 positively correlates with most genes in the MAPK pathway with a few exceptions and has some associations with this signaling cascade. Given its correlations with genes in this pathway and trametinib being trialled here in PCa, BRD9 has most promise as a co-target with this pathway.
We provide some evidence (in two treatment naïve cohorts) that BRD9 is overexpressed in ERG fusion positive cancers and that ERG expression is very weakly correlated with BRD9 expression. This suggests BRD9 may be a worthwhile co-target alongside ERG. Our data shows BRD9 mutations are less prevalent in ERG fusion positive cancers, in line with the literature suggesting these cancers are less likely to have mutated SWI/SNF complexes [69,70]. BRD9 warrants further investigation in ERG fusion positive cancers and possibly cotargeted with ERG inhibition, as ERG depends on BAF complexes for its epithelial to mesenchymal transition (EMT) [71].

Limitations
This work investigated publicly available cohorts which have limitations including that not all cohorts had mutation data. Some cohorts had small sample sizes and most lacked data on patients with more advanced stage cancers. Most cohorts also did not include information on treatment response, meaning it is difficult conclude whether BRD9 has potential as a predictive biomarker using publicly available data. When comparing BRD9 expression in patients, available data, normal patients are not actually healthy patients, rather they have benign prostates. Some of the less recent data also uses grading classifications that we would not now e.g. Gleason Grade 5 is now not considered cancer.
In addition, it is not possible to determine causation -we can only study correlation in publicly available datasets; In future, overexpression, and knockdown studies must be performed to establish causation.
CONCLUSION results and alongside the one published paper, forms a basis for future study of BRD9 in PCa.
BRD9 has most potential as a diagnostic biomarker and drug target in metastatic disease.
This work warrants further experimental work, in vitro, in vivo and ex vivo, to continue determining where and how targeting BRD9 could be successful in PCa.