FAM83D promotes ovarian cancer progression and its potential application in diagnosis of invasive ovarian cancer

Abstract Although invasive epithelial ovarian cancer (IOC) and low malignant potential ovarian tumour (LMP) are similar, they are associated with different outcomes and treatment strategies. The current accuracy in distinguishing these diseases is unsatisfactory, leading to delays or unnecessary treatments. We compared the molecular signature of IOC and LMP cases by analysing their transcriptomic data and re‐clustered them according to these data rather than the pathological dissection. We identified that FAM83D was highly expressed in IOC. To verify the role of FAM83D in the progression and metastasis, we used the isogenic ovarian cancer metastatic models, highly metastatic cells (HM) and non‐metastatic cells (NM). Overexpression of FAM83D significantly promoted cell proliferation, migration and spheroid formation. This was consistent with previous data showing that high FAM83D expression is associated with poor prognosis in cancer patients. Moreover, similar to the HM cells, the FAM83D‐overexpressing NM cells demonstrated stronger phosphorylation of the epidermal growth factor receptor (EGFR) and c‐Raf. This indicates that the action of FAM83D is mediated by the activation of the EGFR pathway. Taken together, this report suggested that FAM83D might be an excellent molecular marker to discriminate between IOC and LMP.


| INTRODUC TI ON
Low malignant potential tumour (LMP) is a semi-malignant ovarian tumour, which was classified by the Federation of Gynecology and Obstetrics in 1961. 1 LMP accounts for 15%-20% of epithelial ovarian tumours. 2 LMP is defined as a tumour with abnormal nuclear division and cell proliferation, lacking observable invasion into the stroma or invasion-like implants. 3 In contrast, invasive epithelial ovarian cancer (IOC), which represents approximately 70% of epithelial-originated ovarian tumours, exhibits strong invasive properties. Based on the different invasiveness, the outcome of LMP and IOC differ considerably. The 5-year survival rate of LMP patients is >90%, whereas that of IOC patients is <30%. 4,5 Therefore, the clinical management of patients with LMP and IOC is different. Considering the malignant status of the tumour and the desire for fertility-sparing in patients, different operative procedures may be employed for LMP. In particular, preservation of fertility should be considered in younger patients.
Regarding the management of IOC, the gynecologist may recommend total hysterectomy and bilateral salpingo-oophorectomy even in patients with Stage I ovarian cancer. In the extended resection, chemotherapy will be administered to eliminate invisible cancer cells, aiming to prevent relapse of ovarian cancer. 6 Abdominal hysterectomy is the standard treatment for LMP.
However, considering that the average age of LMP occurrence is 40 years, preservation of fertility may be important in these patients. In such cases, a more conservative surgical managementunilateral oophorectomy (ie removal of only one ovary)-may be considered. Since the managements of LMP and IOC are significantly different, accurate diagnosis of IOC and LMP is essential for the selection of the most appropriate treatment and will be beneficial to the patients. Indeed, approximately 20%-30% of cases initially diagnosed with LMP are eventually confirmed to be IOC. 1 The diagnosis is based on histopathological observation without the use of molecular markers, leading to inaccuracy in the diagnosis of LMP. 1 Hence, the pathologist may often use terms such as 'rule out LMP' or 'at least LMP' in diagnostic reports. 7,8 The gene expression profile determines the phenotype of the tumour. 9 Therefore, revealing the molecular differences between LMP and IOC and identifying useful molecular markers may increase the accuracy of diagnosis.

| Cell culture
The immortalized ovary surface epithelial cells (IOSE8) are cultured in M199/MCDB105 medium supplemented with 10% FBS and in 1% penicillin and streptomycin at 37°C in a humidified atmosphere of 5% CO2. The highly metastatic (HM) and non-metastatic (NM) cells used in this study were isogenic cells lines derived from SKOV3.ip1 cells. 10 The HM cells exhibited a strong metastatic signature, unlike NM cells which were shown to be non-metastatic and failed to form detectable metastasis. Therefore, the HM/NM model offered a wellcontrolled experimental system to study the metastasis of ovarian cancer. The cells are maintained in RPMI 1640 supplemented with 5% foetal bovine serum (Gibco, NY) and 1% penicillin and streptomycin at 37°C in a humidified atmosphere of 5% CO 2 . These two types of cells were kindly provided by Professor Alice ST Wong.

| FAM83D-overexpressing stable cell line
The FAM83D-expressing plasmid was constructed by inserting the coding region sequence of FAM83D into the pcDNA3.1+ vector (Invitrogen, Burlington, Canada). To generate the FAM83D-overexpressing stable cell line NM-FAM83D, the FAM83D/pcDNA3.1 plasmid was transfected into NM cells using Lipofectamine 3000 (Invitrogen, Burlington, Canada). The NM cells transfected with an empty pcDNA3.1+ vector served as the control (NM-Vector).
Twenty-four hours after transfection, G418 (150 μg/mL) was added for FAM83D stable expression cell line selection for 1 month. The expression of FAM83D was confirmed by real-time polymerase chain reaction (PCR) and western blotting analysis.

| Cell proliferation assay
The cell proliferation assay was performed using the IncuCyte ZOOMTM Live-Cell Imaging and Analysis System according to the manufacturer's protocol. In brief, the NM-Vector and NM-FAM83D cells were seeded at a density of 3000 cells/well into 96-well plate.
The plate was maintained in the IncuCyte system for consecutive monitoring of cell proliferation. Images were recorded every 3 hours and cell confluency was analysed using the IncuCyte software (Essen BioScience; version 2018A).

| Colony formation assay
The cells (500 cells) were seeded into a 100-mm petri dish and incubated in a CO 2 incubator at 37°C for 10 days or until cells in control plates formed colonies with substantially good size. Subsequently, the medium was removed and the colonies were stained with 0.5% crystal violate for 5 minutes and washed twice with phosphate buffered saline (PBS). The dishes were air-dried at room temperature. Count images were captured and the number of colonies was counted using a stereomicroscope (Olympus, SZ61, Tokyo, Japan).

| Wound healing assay
NM-Vector and NM-FAM83D cells were collected and washed once using Hank's buffer. Cells (1 × 10 5 ) were seeded into a 12-well plate and incubated until they reached >90% confluency. The samples were subsequently manually scratched using a P200 pipette. Images were acquired on days 0, 1, 2 and 3 using a light microscope (EVOS, ThermoFisher, MA, USA). The migration distance was measured using the ImageJ software.

| Migration assay
To test the migration ability of cancer cells, NM-Vector and NM- to the bottom of the plates were fixed using 4% paraformaldehyde and stained with 0.5% crystal violate. Images were captured under a light microscope (EVOS, ThermoFisher, MA) with 40× magnification and the areas of staining were calculated using the ImageJ software.
GSE12172 raw data were normalized in MultiExperiment Viewer (MeV). The differential gene expressions (fold changes >2, P < 0.001) were identified using the 'Linear Models for Microarray' method (LIMMA, http://mev.tm4.org). 11 The gene copy number variation of FAM83D and FAM81B was obtained from the Oncomine online database. 12 The correlation of gene expression with tumour stage and tumour grade were evaluated using the Ovarian cancer database of the Cancer Science Institute of Singapore and statistical significance was calculated using the Mann-Whitney test. For the analysis of the percentage of overall survival (OS) and disease-free survival (DFS), the log-rank test was used to compare the survival expectation of a group with different gene expression. 13

| Immunohistochemistry
All clinical samples were analysed by standard immunohistochemical staining at the same time. Briefly, 5-μm sections were deparaffnized,

| Statistical analysis
Data are presented as means ± standard error of mean (SEM) from at least three independent experiments. Student's t test (with Welch's correction) was performed for comparison between two groups and P < 0.05 denoted statistical significance. A receiver operating characteristic curve (ROC) was generated to estimate the diagnostic ability of a parameter. Principal component analysis (PCA) was employed to distinguish the IOC from the LMP samples (Gastinel, 2012).
PCA based on transcriptomic data was used to distinguish patients.

| Re-clustering of LMP and IOC cases based on the transcriptomic data
To understand the differences between the molecular profiles of IOC and LMP, we initially classified the patients with LMP or IOC under a definitive pathological diagnosis. The GSE12172 dataset including 30 LMP and 60 malignant IOC cases was used to establish a training methodology for the accurate differentiation of IOC and LMP. We re-clustered the cases using whole transcriptomic data from both pathologically defined LMP and IOC cases. The results of the PCA analysis using whole transcriptomic data suggested that the LMP and IOC could be accurately separated except for two cases (IOC36 and IOC58). These two cases were originally classified as IOC according to the pathological diagnosis. However, after re-clustering, they were clustered into the LMP group ( Figure 1A). In addition, hierarchical clustering data analysis was employed to re-cluster the cases ( Figure 1B). Similar to the first analysis, IOC36 and IOC58 were identified as LMP rather than IOC. Therefore, these two cases were probably rare misdiagnosed cases. To further identify differences between the molecular expression of LMP and IOC, we excluded these two diagnostically inconsistent cases. Comparison of the newly defined groups showed that 409 and 593 genes had higher and lower expression, respectively in the IOC group versus the LMP group. The eight genes with the most significant differential expression were further analysed in the survival analysis to evaluate the association between expression and overall patient survival. Among the highly expressed genes in the IOC group, FAM83D and SH2D3C showed the strongest association with poor survival (hazard ratio [ Figure S1 and Table 1. The molecular detection of genes with higher expression is easier. Therefore, we propose FAM83D as a potential molecular marker to distinguish IOC from LMP.

| FAM83D participates in the migration of ovarian cancer
Since the expression FAM83D was shown to be strongly associated with the overall patient survival, we hypothesized that FAM83D may promote the progression and metastasis of ovarian cancer. Here, we used the isogenic ovarian cancer metastatic models-highly metastatic (HM) cells and non-metastatic (NM) cells-to test this hypothesis. It is worth noting that the use of the HM and NM cells in this project is just the tools to verify the role of FAM83D in cancer metastasis, but not to mimic the different between IOC and LMP.
Western blotting and RT-PCR analysis showed that the expression of FAM83D was higher in HM cells compared with that observed in NM cells (Figure 2A). In addition, we tested the expression of meta-

| FAM83D is up-regulated in ovarian carcinoma and correlated with ovarian malignant characteristics
To verify the expression of FAM83D in patients with high-grade serous ovarian cancer, we measured the expression of FAM83D using RT-PCR. Our data showed that all six samples investigated demon-

| Up-regulation FAM83D triggers the EGFR signalling pathway and promotes migration and proliferation of cancer cells
A previous study reported that the FAM83 protein family has a highly conserved N-terminal domain of unknown function (DUF1699) that able to prevent the interaction between cRaf and the regulatory

| FAM83D was effective in the diagnosis of LMP and IOC
The ROC is a methodology to illustrate the ability of a parameter to accurately diagnose a disease. The area under the curve (AUC) of the ROC curve is an index evaluating the performance of a parameter in the discrimination of certain diseases. 15 To test the potential of FAM83D in the diagnosis of LMP and IOC in clinical practice, the dataset GSE9891 was selected as a testing model. The results showed that FAM83D could distinguish IOC from LMP with an AUC of 0.978 ( Figure 6A). We collected other GEO datasets and analysed the capacity of FAM83D in the diagnosis of LMP and IOC using a ROC curve. The results are summarized in Table 2. Within the tested datasets, the lowest AUC was 0.742 and the highest up to 1.0. In most cases, the AUC was close to or higher than 0.9, which strongly implies that FAM83D could serve as a marker to distinguish IOC from LMP. In this analysis, we also found that the expression of FAM81B was mutually exclusive with that of FAM83D in both IOC and LMP ( Figure 6B). We suggest that the use of these two genes as markers may further increase diagnostic accuracy for LMP and IOC by replacing transcriptome-wide PCA analysis. The results showed that FAM83D and FAM81B could identify IOC36 and IOC58 as LMP ( Figure 6C), which is consistent with the analysis using whole transcriptome data ( Figure 1A). Therefore, detecting the expression of these two genes may provide a molecular diagnosis approach to identifying LMP and IOC with significantly improved accuracy compared with that offered by histopathological observation.
IOC is a general term for ovarian cancer including many histotypes, such as serous, endometrioid, clear and mucinous carcinoma. Serous was significantly lower in LGSOC compared with that observed in HGSOC ( Figure 6D). The AUC of FAM83D in the diagnosis of HGSOC and LGSOC was >0.8 ( Figure 6E and Table 2), suggesting that FAM83D may also be a potential marker to distinguish HGSOC from LGSOC. To verify this hypothesis, we compared the gene expression profiles from GSE27651 dataset, which contains data for HGSOC, LGSOC, LMP and human ovarian surface epithelial (HOSE) cancers. The clustering clearly indicated that the gene expression profile of HGSOC was significantly different to that of others. Moreover, the LGSOC pattern was similar to that of LMP ( Figure 6F). Therefore, utilization of FAM83D may effectively discriminate HGSOC from LGSOC (Table 2). LGSOC into type I tumour and its precursor is LMP cell. In contrast, the HGSOC was classified as type II tumour. 19 In this study, we propose that FAM83D is a good method for the accurate diagnosis of IOC and LMP, and differentiation of HGSOC from LGSOC.

| D ISCUSS I ON
There are several studies showing FAM83D gain of function in several types of cancer, 22,23 especially breast cancer. 22,24 Interestingly, up-regulation of the FAM83D predicts cancer patients (breast, lung, liver) with a high risk of mortality. 22  Interestingly, we have demonstrated FAM83D overexpression not only activate the MEK/ERK pathway, but also activation of EGFR and PI3K/AKT pathway ( Figure 5A). We speculate that the FAM83D initiates MEK/ERK1/2 could up-regulate the expression of MMP2 expression, which active EGFR in a ligand-dependent mechanism.
Indeed, previous reports also suggested the knockdown of FAM83D with shRNA will increase the activation of EGFR/MAPK pathway. 25 The overall signalling pathway is summarized in Figure 5C.
In conclusion, we demonstrated that FAM83D is a highly expressed gene in IOC and correlated with tumour stage and grade.
Functional analyses further suggested the role of FAM83D in promoting the proliferation, migration and metastasis of ovarian cancer cells. Therefore, FAM83D may be an excellent marker for distinguishing IOC from LMP and may contribute to differentiating HGSOC from LGSOC.

CO N FLI C T O F I NTE R E S T
The authors have no competing interests to disclose.