RAD51AP1 promotes progression of ovarian cancer via TGF‐β/Smad signalling pathway

Abstract Ovarian cancer (OC) is one of the leading causes of female deaths. However, the molecular pathogenesis of OC has still remained elusive. This study aimed to explore the potential genes associated with the progression of OC. In the current study, 3 data sets of OC were downloaded from the GEO database to identify hub gene. Somatic mutation data obtained from TCGA were used to analyse the mutation. Immune cells were used to estimate effect of the hub gene to the tumour microenvironment. RNA‐seq and clinical data of OC patients retrieved from TCGA were used to investigate the diagnostic and prognostic values of hub gene. A series of in vitro assays were performed to indicate the function of hub gene and its possible mechanisms in OC. As a result, RAD51AP1 was found as a hub gene, which expression higher was mainly associated with poor survival in OC patients. Up‐regulation of RAD51AP1 was closely associated with mutations. RAD51AP1 up‐regulation accompanied by accumulated Th2 cells, but reduced CD4 + T cells and CD8 + T cells. Nomogram demonstrated RAD51AP1 increased the accuracy of the model. Down‐regulation of RAD51AP1 suppressed proliferation, migration and invasion capabilities of OC cells in vitro. Additionally, scatter plots showed that RAD51AP1 was positively correlated with genes in TGF‐β/Smad pathway. The above‐mentioned results were validated by RT‐qPCR and Western blotting. In conclusion, up‐regulation of RAD51AP1 was closely associated with mutations in OC. RAD51AP1 might represent an indicator for predicting OS of OC patients. Besides, RAD51AP1 might accelerate progression of OC by TGF‐β/Smad signalling pathway.


| INTRODUC TI ON
Ovarian cancer (OC), one of the most fatal and aggressive tumours of the female reproductive system, has emerged with an increased incidence in recent years. 1 Due to confusing symptoms and no screening for early detection, the 5-year overall survival (OS) was reported as high as about 45%. Early diagnosis was correlated with improved OS, while poor OS was associated with late diagnosis and high rate of recurrence. 2,3 However, the evolutionary mechanisms of OC have still remained elusive. Thus, it is crucial to understand the precise molecular mechanisms involved in the carcinogenesis, proliferation, invasion of OC and develop further effective diagnostic and therapeutic methods for managing OC. [4][5][6] In recent years, the microarray and high-throughput sequencing technologies have rapidly advanced to screen the genetic alterations in genome level and explore further potential biomarkers. 7,8 A number of databases are available from free public gene expression data repositories, such as The Cancer Genome Atlas (TCGA) project and the Gene Expression Omnibus (GEO) database. Numerous studies employed these databases and different bioinformatics methods to seek novel markers and molecular mechanisms for OC. 9,10 RAD51-dependent homologous recombination (RAD51AP1) is a RAD51 accessory protein that specifically stimulates joint molecule formation through the combination of structure-specific DNA binding and physical contact with RAD51, and knockdown of RAD51AP1 results in increase of sensitivity to DNA damaging agents and impaired HR. 11 In the present study, 3 data sets of OC were downloaded from the GEO database to identify RAD51AP1 as a hub gene in OC. A variety of bioinformatics methods and experimental assays were utilized to understand the biological functions and relative molecular mechanism underlying carcinogenesis. To our acknowledge, this study initially explored the relationship of RAD51AP1 and gene mutations and tumour microenvironment (TME), then established a RAD51AP1-related nomogram to investigate patients' survival, and demonstrated that RAD51AP1 might accelerate progression of OC by TGF-β/Smad signalling pathway.

| OC and normal controls data sets
Data retrieved from multiple researches were used for integrated analysis in this study, including data from TCGA project and GEO database. RNA-seq data and corresponding clinicopathological data of OC patients in TCGA were obtained from UCSC Xena (https:// xenab rowser.net/datap ages/). The clinicopathological characteristics included age, pharmaceutical therapy, stage, grade, outcome, follow-up, etc The somatic mutation status for OC (workflow type: VarScan2 Variant Aggregation and Masking) was obtained from TCGA (https://portal.gdc.cancer.gov/repos itory). The microarray data sets (GSE14001, GSE40595 and GSE54388) were downloaded from GEO database (https://www.ncbi.nlm.nih.gov/) and were annotated according to the platform of Affymetrix HG-U133 Plus 2.0 (GPL570).
Raw microarray data were downloaded and normalized using a robust multi-array average (RAM) method using 'affy' package in R software for estimation of missing values, background correction, log2 transformation, quantile normalization and data summarization. The GSE14001 data set included 3 normal samples and 20 OC samples; GSE40595 contained 6 normal samples and 32 OC samples; GSE54388 composed of 6 normal samples and 16 OC samples.

| Cell culture and siRNA transfection
In the present research, OC cell lines (HEY, SKOV3) were ob- HEY and SKOV3 cells were cultured in 6-well plates. Cells were transfected with 100 nmol/L RAD51AP1 siRNA or control siRNA.
The cells transfected siRNA were incubated for 24 hours for subsequent assays, including quantitative reverse transcription polymerase chain reaction (RT-qPCR), proliferation, wound healing and transwell migration. In addition, the OC cells are transfected with siRAD51AP1 or siControl for 4-6 hours, and then, the cells are cultured with RPMI-1640 medium supplemented without FBS for 24 hours; subsequently, cells transfected with siRAD51AP1 are incubated with TGFβ for 48 hours for subsequent assays.

| RNA extraction and RT-qPCR
Total RNA was isolated using TRIzol reagent according to the manufacturer's instructions (Invitrogen, Carlsbad, CA). Synthesis of cDNA was performed by using the ReverTra Ace qPCR RT kit (Toyobo, Shanghai, China). The RT-qPCR was carried out through an ABI 7500 Real-Time PCR system (Applied Biosystems, Foster City, USA) using the SYBR Premix EX Taq™ (Takala, Dalian, China).
The primer sequences are listed in Table S1. The expression of glyceraldehyde-3-phosphate dehydrogenase (GAPDH) served as internal standard. Relative gene expression was determined by using the 2 −ΔΔCT comparative method.

| Cell wound healing assay
Cell migration was evaluated with wound healing assay. 1 × 10 6 SKOV3 cells were seeded on 6-well plates. Wound was produced by a 100 μL pipette tube. Images were then captured after three times washing with phosphate-buffered saline (PBS), and the exchanging medium (RPMI-1640 medium) was added into the plate. The spread of wound was observed after 0 and 96 hours. Cell migration was photographed in several pre-marked spots with a microscope and quantified by ImageJ software.

| Transwell migration and invasion assays
The transwell migration and invasion assays were conducted in 24well plates (Corning, Inc, NY, USA). For the transwell migration assay, After 48 hours, the cells on the lower surface were fixed, stained and photographed microscopically.

| Identification and functional annotation of differentially expressed genes (DEGs)
We merged the three data sets to one data set, and then, key DEGs were identified by using a normalization method via 'sva' package in

| Construction of protein-protein interaction (PPI) network
The STRING database (ver. 10.5, https://strin g-db.org/) was used to construct PPI network in DEGs. A combined score >0.4 was taken as inclusion criterion into account. The visualization of PPI network was realized by Cytoscape 3.6.0 software. Furthermore, the key genes were identified with molecular complex detection (MCODE) in Cytoscape for finding modules in PPI network.

| Gene Set Enrichment Analysis (GSEA) and Gene Set Variation Analysis (GSVA)
To investigate RAD51AP1-mediated biological parameters in OC, GSEA was conducted by clusterProfiler package. 12 We downloaded h.all.v6.2.symbols.gmt from the Molecular Signatures Database (http://www.broad.mit.edu/gsea/msigd b/). False-discovery rate (FDR) < 0.05 and P < 0.05 were utilized as the enriched terms. In addition, 308 OC patients in the TCGA data set were divided into high-expression group and low-expression group according to the median value of RAD51AP1. GSVA conducted by 'gsva' package was applied to further validate different biological procedures between the two groups.

| Construction of a prognostic nomogram
It should be noted that OC patients have typically a poor prognosis; thus, we established a nomogram for evaluating the prognostic risk of each patient. TCGA data set was used to construct a nomogram using Cox proportional hazards model. The discrimination of the model was assessed by using the Harrell's concordance index (C-index).
Calibration curves were utilized to validate the accuracy of prediction.
Diagnostic prediction models were analysed by receiver operating characteristic (ROC) curve and evaluated by the area under ROC curve (AUROC). The nomogram function was used in 'rms' package.
All the statistical analyses were performed using R 3.5.0 and GraphPad Prism 7.0 software. In at least three independent experiments, the data were presented as mean ± standard deviation (SD).
Two-tailed Student's t test was used to assess the differences between the two groups. P < 0.05 was considered statistically significant. All the data sets used in the present research are summarized in Table S2.

| Identification and enrichment analysis of the DEGs
A total of 347 DEGs were identified, including 75 up-regulated genes and 272 down-regulated genes in OC ( Figure 1A and B).
Subsequently, we constructed PPI network of DEGs, and one module was obtained using the MCODE in Cytoscape. Finally, 10 candidate hub genes were identified as follows ( Figure 1C): RAD51AP1, FAM83D, KIF4A, TTK, BUB1B, TRIP13, NUF2, CEP55, TPX2 and DLGAP5. Their expression levels were all higher in OC group than those in control group in 3 GEO data sets ( Figure S1).
To further analyse the biological functions and signalling pathways of the 347 DEGs, GO and KEGG enrichment analyses were undertaken. GO analysis mainly concentrated on cell growth, mesenchyme development, extracellular matrix, cell-cell junction, protein binding, etc KEGG pathway indicated proteoglycans in OC, basal cell carcinoma, etc ( Figure 1D).

| Up-regulation of RAD51AP1 was associated with poor survival in OC
RAD51AP1 is a critical protein for DNA repair by homologous recombination (HR). We analysed the expression level of RAD51AP1 in OC group and control group in Oncomine database, which revealed that RAD51AP1 was higher in OC (Figure 2A

| Immune phenotype landscape of the tumour cells in the TME of OC samples
We evaluated the abundance of 28 tumour-infiltrating immune cells in OC using TCGA data sets. In the radar chart, RAD51AP1 was statistically significant correlation with TME cells ( Figure 3B). Then, the OC were divided into 2 clusters based on the median expression of RAD51AP1. The anti-tumour cells including central memory CD4 T cell, central memory CD8 T cell, Th1 cell, Th17 cell, etc were higher in low-level RAD51AP1 group; however, Th2 cell and the pro-tumour cell were enriched in high-level RAD51AP1 group ( Figure 3C).

| RAD51AP1-related prognostic nomogram
A prognostic nomogram that integrated variables of TCGA data set was constructed to predict the OS of OC patients ( Figure 4A), including RAD51AP1 and several significant clinical factors, such as age, pharmaceutical therapy, atomic neoplasm, stage, grade, intermediate dimension and outcome. Calibration curves depicted that the predictive capability of 8-year nomogram was remarkably superior than 5-year one ( Figure 4B). In our nomogram, RAD51AP1 was  Figure 4C).

| RAD51AP1 was positively correlated with tumour mutational burden (TMB)
It is noteworthy that TMB was recently deemed as an important marker to assess the reaction of patients to the target medicine (eg PD-1/PD-L1). It has been recommended as a useful marker for non-small cell lung cancer (NSCLC) patients who underwent immunotherapy as described previously. 15 The somatic mutation data of OC patients were processed to acquire 436 TMB data sets of OC patients. Besides, TMB and RAD51AP1 expression were simultaneously existed in 210 OC patients. Spearman's rank-order correlation was used to explore the correlation of TMB with RAD51AP1, in which the scatter plots showed that TMB was correlated with increased RAD51AP1 (P < 0.05, Figure S3). Furthermore, in different four subgroups, TMB was significantly positively correlated with RAD51AP1 in immunoreactive group ( Figure 4D). This might be a new outcome related to RAD51AP1 that the deserving gene assist

| RAD51AP1 accelerated proliferation of OC cells
GSEA shown that high-RAD51AP1 group were mainly associated with G2M Checkpoint, E2F Targets and DNA Repair ( Figure 5A), which were validated by GSVA ( Figure S4). Additionally, in the correlation heatmap, cell cycle-related genes were clustered and their expressions were positively correlated with RAD51AP1; besides, apoptosis-associated genes were agminated and negatively correlated with RAD51AP1 ( Figure 5B). These results demonstrated that RAD51AP1 was associated with cell cycle progression and accelerated cell proliferation.
To detect the effects of RAD51AP1 knockdown on cell viability, three distinct RAD51AP1 target-specific siRNAs were used to transfect SKOV3. After 24 hours, the knockdown efficiency was validated by RT-qPCR ( Figure S5). Additionally, it was disclosed that siRAD51AP-2 and siRAD51AP-3 were more effective; consequently, the cells transfected with siRNA-2 and siRNA-3 were used for subsequent assays.
We further investigated the effects of siRAD51AP1 on cell proliferation. We found that down-regulation of RAD51AP1 notably decreased the proliferation of SKOV3 and HEY ( Figure 5C).

| RAD51AP1 promoted the invasion and migration of OC cells
As shown in Figure 5D-F, the capabilities of cell invasive and migration were lower in siRAD51AP1 cells than those in the control cells (P < 0.05). The wound healing assay showed that migratory distances of the siRAD51AP1 group were significantly wider ( Figure 5D), reflecting that siRAD51AP1 inhibited SKOV3 migration.

| RAD51AP1 accelerated progression of OC by TGF-β/Smad signalling pathway
Bioinformatics analysis and in vitro assays demonstrated that RAD51AP1 accelerated progression of OC. To explore the mechanism, we studied the relationship between RAD51AP1 and genes related to TGF-β/Smad signalling pathway in GEPIA (http://gepia.cance r-pku.cn). Scatter plots showed that SMAD2, SMAD3, SMAD4 and TGFBR1 were significantly positively correlated with RAD51AP1 ( Figure S6). Moreover, the results of RT-qPCR and Western blotting were consistent with the scatter plots. Furthermore, mRNA levels of SMAD2, SMAD3, SMAD4 and TGFBR1 were lower in silent RAD51AP1 group ( Figure 5G and H); besides, the protein levels of p-SMAD2, p-SMAD3 and total-SMAD2/3 were lower in silent RAD51AP1 group ( Figure 5I). In addition, when the cells were transfected with siRNA then incubated with TGFβ, p-smad2 was lower in siRAD51AP1 group, but higher in siRAD51AP1 + TGFβ group compared to siControl group, indicating that TGFβ reversed the effect of siRAD51AP1 inhibiting p-smad2 ( Figure 5J). In the present research, we explored the relationship between RAD51AP1 and markers in TGF-β/Smad signalling pathway. The transforming growth factor β (TGF-β) could regulate a fascinating array of cellular processes, including cell proliferation, apoptosis, differentiation, migration, invasion and adhesion. A number of scholars demonstrated that different types of cancer progressed along with activating TGF-β/Smad signalling pathway. 34,35 Scatter plots showed that RAD51AP1 was positively correlated with genes in TGF-β/ Smad signalling pathway. Our experiments uncovered that the levels of p-SMAD2, p-SMAD3 and total-SMAD2/3 were lower in siR-AD51AP1 group; when the cells were transfected with siRNA then incubated with TGFβ, p-smad2 was lower in siRAD51AP1 group, but higher insiRAD51AP1 + TGFβ group compared to siControl group, indicating that TGFβ reversed the effect of siRAD51AP1 inhibiting p-smad2 ( Figure 5J). This suggested that RAD51AP1 might promote cell proliferation, invasion and migration by TGF-β/Smad signalling pathway.

| CON CLUS IONS
In summary, we performed a comprehensive analysis on RAD51AP1. We found that up-regulation of RAD51AP1 was closely associated with poor outcome and more mutations in OC patients. RAD51AP1 might have effect on TME, thus expedited progression of OC. We presented a further accurate nomogram for predicting prognosis of OC. Additionally, we demonstrated a potential mechanism of RAD51AP1, involving in the carcinogenesis, proliferation and invasion of OC might via TGF-β/Smad signalling pathway. No.Z181100001718193

CO N FLI C T O F I NTE R E S T
The authors declare that they have no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
All the data sets used in the present research are summarized in Table S2.