Comprehensive assessment of the expression of the SWI/SNF complex defines two distinct prognostic subtypes of ovarian clear cell carcinoma

Somatic mutations in the ARID1A tumor-suppressor gene have been frequently identified in ovarian clear cell carcinoma (CCC) cases. BAF250a encoded by ARID1A is a member of the SWI/SNF complex, but the expression and mutation status of other SWI/SNF subunits have not been explored. The current study aimed to elucidate the biological and clinical significance of the SWI/SNF complex subunits, by assessing the expression and mutation status of SWI/SNF subunits, and distinct genomic aberrations associated with their expression. Of 82 CCC specimens, 38 samples presented no BAF250a expression, and 50 samples exhibited the loss of at least one subunit of the SWI/SNF complex. Cases which lack at least one SWI/SNF complex component exhibited significantly more advanced stages, faster growth and stronger nuclear atypia compared with SWI/SNF-positive samples (p<0.05). Although BAF250a expression is not related to poor prognosis, the group presenting the loss of at least one SWI/SNF complex subunit exhibited significantly shorter overall and progression-free survivals (p<0.05). A multivariate analysis suggested that the expression status of the SWI/SNF complex serves as an independent prognostic factor (p<0.005). The cases positive for all SWI/SNF subunits demonstrated significantly greater DNA copy number alterations, such as amplification at chromosomes 8q.24.3 and 20q.13.2-20q.13.33 (including ZNF217) and deletion at chromosomes 13q12.11-13q14.3 (including RB1), 17p13.2-17p13.1 (including TP53) and 19p13.2-19p13.12. In conclusion, the CCCs exhibiting the loss of one or multiple SWI/SNF complex subunits demonstrated aggressive behaviors and poor prognosis, whereas the CCCs with positive expression for all SWI/SNF components presented more copy number alterations and a favorable prognosis.

Because CCC cells frequently exhibit nuclear atypia, the cancer cell nuclei were assessed based on their SWI/ SNF status. The total area of the nucleus was smaller in cases lacking at least one SWI/SNF component compared with the nuclear area in the positive cases (p<0.05, Supplementary Figure S7-a); however, the perimeter of the nucleus was not significantly different (p=0.5712, Supplementary Figure S7-b). This difference may be due to the marked irregularity in the nuclear shape of the cells in the cases lacking at least one SWI/SNF subunit. The roundness, denoted by the shape factor, is markedly reduced in negative cells compared with immunoreactive cells (p<0.0001, Supplementary Figure S7

Pathological review
All tumor samples were independently reviewed by a gynecologic pathologist who was blinded to the patient data. Cases in which the review diagnosis differed from the original diagnosis were further reviewed by a second gynecologic pathologist. The tumors were histologically classified according to World Health Organization (WHO) criteria [1].
Clear cell adenocarcinomas exhibit three main architectural patterns: papillary, tubulocystic and solid patterns [2]. The papillary pattern is characterized by papillae that contain fibrous tissue or hyaline material, whereas the tubulocystic pattern is characterized by multiple tubules and cysts that are variable in size and lined by cuboidal to flattened epithelium. The solid pattern consists of groups of polyhedral cells with abundant clear cytoplasm separated by delicate fibrovascular stroma. The tumor pattern is assigned based on the predominance of a certain architecture in more than 50% of all examined sections of the tumor.

Immunohistochemistry
Nine proteins that compose the SWI/SNF complex and six proteins related to tumor development were stained to allow comparisons of their expression profiles in CCC with their expression profiles in other histological subtypes of ovarian cancer.
Sections from formaldehyde-fixed, paraffin-embedded tissue were deparaffinized and rehydrated. Antigen retrieval was performed by placing sections in citrate buffer (pH 6.0) and then in a decloaking chamber at 120°C for 10 minutes. Heat-induced epitope retrieval (HIER) of BAF180, SNF5, HNF1B and pAKT was performed with EDTA buffer (pH 8.0). After endogenous peroxidase activity was blocked, the sections were sequentially incubated with 5% normal blocking serum for 30 minutes, the appropriate antibodies (Supplementary Table S7), the corresponding biotinylated anti-mouse, anti-goat or anti-rabbit IgG (Nichirei Histofine ® SAB-PO staining kit, Nichirei Biosciences, Tokyo, Japan) and an avidin-biotin peroxidase complex from the same kit for 30 minutes. The samples were then exposed to diaminobenzidine tetrahydrochloride solution (DAKO ® DAB Chromogen Tablets in D 2 W, Dako, Glostrup, Denmark).

Evaluation of staining
Protein expression was evaluated by determining the Immuno-Reactive Score (IRS) [3], which is obtained by multiplying the percentage of positively stained cells by the intensity of the reaction. The percentage of positive cells ranged from 0% to 100%, and the intensity of the reaction was given a score ranging from 0 to 3 as follows: no reaction, 0; weak, 1; moderate, 2; or strong, 3. Therefore, the scores ranged from 0 to 300. Scores between 0 and 49 were classified as negative, whereas scores ≥50 were considered to indicate positive expression.
In the analysis of P53 staining, cases with high immunoreactivity scores (≥150) are considered to exhibit a potential gain-of-function mutation of TP53, whereas cases with lower scores are considered to express wildtype P53 or null mutations based on the results from a previous study [4]. The IRSs were reviewed by a gynecologic pathologist who was blinded to the patient data.
The Ki-67 labeling index was analyzed using image analysis software (ImageJ) [5]. For each case, 10 highpower field areas (×400) of maximal tumor positivity were selected, and each field contained 50-250 nuclei. The percentages of positive nuclei were determined [6]. For the clinicopathologic correlations, Ki-67 positivity was classified into two groups: the low-proliferation group contained less than 25% Ki-67-positive tumor cells, and the high-proliferation group contained at least 25% Ki-67positive tumor cells [7].

Analysis of nuclear shape
Several images were captured for CCC hematoxylin and eosin staining at high magnification (100× oilimmersion lens using a Nikon DS-Fi1 digital camera coupled to a Nikon Eclipse Ti microscope [Nikon, Tokyo, Japan] and a 40× lens using a Bravio 9000 [KEYENCE, Osaka, Japan]). The nuclear shape was analyzed using the integrated morphometry package of the MetaMorph ® Image analyzer (version 6.1, Universal Imaging Corp., West Chester, PA, USA). The images were converted to 16-bit monochrome images, and the tumor cell nuclei were selected as regions of interest (ROIs) to measure different parameters of nuclear morphology (Supplementary Figure  S11). At least four areas were selected for each case.
The total area, perimeter (P), shape factor, inner radius, outer radius, and other factors were measured using the imaging software for each thresholded nucleus in the image. The measurements were converted to micrometers after image calibration. The perimeter of the nucleus was measured from the mid-point of each pixel composing the thresholded border or margin. The shape factor has a value from 0 to 1 that corresponds to how closely the region resembles a circle. A value near 0 indicates a flattened object, whereas a value of 1.0 indicates a perfectly circular nucleus. The inner radius is the distance from the centroid to the nearest point along the object's edge. The outer radius is the distance from the centroid to the farthest point along the object's edge. The remaining parameters were defined according to the manufacturer's user guide.

Microarray analysis
A total of 17 ovarian CCC samples, which were frozen immediately after surgery and stored at -80 ºC, were used for mRNA expression microarray analysis. Total RNA was extracted using the Qiagen RNeasy Mini Kit (Qiagen K.K.). Microarray analysis was performed using Affymetrix Human Genome U133A 2.0 Arrays following standard Affymetrix protocols (Affymetrix, Santa Clara, CA, USA). The expression intensities across the samples were normalized using the Robust Multichip Average (RMA) algorithm [8]. Fifty-eight genes were identified as differentially expressed between cases lacking at least one SWI/SNF component and positive cases if the expression change was greater than twofold based on a t-test with a p-value less than 0.05. The biological roles of the 58 genes were analyzed by categorical analysis (gene ontology terms and TRANSFAC, which represents activation of transcription) using the web-based Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.7 (http://david.abcc.ncifcrf.gov/).

Whole-exome sequencing
The frozen tissue samples of 39 CCC cases and 16 lymphocyte samples isolated from whole blood of the corresponding CCC patients were used for whole-exome sequencing analyses. To identify somatic mutations (single nucleotide variants), 16 samples were compared with their normal counterparts. Thirty-nine CCC cases were assessed for copy number variations between SWI/SNF-positive cases and cases lacking at least one SWI/SNF subunit. The detailed methodology is described in a paper describing a genomewide assessment in ovarian CCC (submitted). GISTIC2.0 [9] and the Bioconductor package "copynumber" in R [10] were used for the analysis and visualization of the copy number variations (CNVs). A samroc analysis [11] was performed to evaluate the copy numbers between SWI/SNF-positive CCC cases and CCC cases lacking one or more subunits. This analysis was performed using transformed log 2 ratios of tumor/normal coverages (the details of which have been described in a secondary supporting manuscript, currently in submission). The samroc q-values, generated by analyzing the distribution of the false-discovery proportion (using the Benjamini-Hochberg procedure with 100 sampling iterations), were less than 0.05 by intent.