Dysregulation of X Chromosome Inactivation in High Grade Ovarian Serous Adenocarcinoma

Background One of the two copies of the X chromosome is randomly inactivated in females as a means of dosage compensation. Loss of X chromosome inactivation (XCI) is observed in breast and ovarian cancers, and is frequent in basal-like subtype and BRCA1 mutation-associated breast cancers. We investigated the clinical implications of the loss of XCI in ovarian cancer and the association between the loss of XCI and BRCA1 dysfunction. Materials and Methods We used open source data generated by The Cancer Genome Atlas (TCGA) Genome Data Analysis Centers. Ward’s hierarchical clustering method was used to classify the methylation status of the X chromosome. Results We grouped 584 high grade serous ovarian adenocarcinomas (HG-SOA) according to methylation status, loss of heterozygosity and deletion or gain of X chromosome into the following five groups: preserved inactivated X chromosome (Xi) group (n = 175), partial reactivation of Xi group (n = 100), p arm deletion of Xi group (n = 35), q arm deletion of Xi group (n = 44), and two copies of active X group (n = 230). We found four genes (XAGE3, ZNF711, MAGEA4, and ZDHHC15) that were up-regulated by loss of XCI. HG-SOA with loss of XCI showed aggressive behavior (overall survival of partial reactivation of Xi group: HR 1.7, 95% CI 1.1–2.5, two copies of active X group: HR 1.4, 95% CI 1.0–1.9). Mutation and hypermethylation of BRCA1 were not frequent in HG-SOA with loss of XCI. Conclusions Loss of XCI is common in HG-SOA and is associated with poor clinical outcome. The role of BRCA1 in loss of XCI might be limited. XCI induced aberrant expression of cancer-testis antigens, which may have a role in tumor aggressiveness.


Materials and Methods
We used open source data generated by The Cancer Genome Atlas (TCGA) Genome Data Analysis Centers. Ward's hierarchical clustering method was used to classify the methylation status of the X chromosome.

Results
We grouped 584 high grade serous ovarian adenocarcinomas (HG-SOA) according to methylation status, loss of heterozygosity and deletion or gain of X chromosome into the following five groups: preserved inactivated X chromosome (Xi) group (n = 175), partial reactivation of Xi group (n = 100), p arm deletion of Xi group (n = 35), q arm deletion of Xi group (n = 44), and two copies of active X group (n = 230). We found four genes (XAGE3, ZNF711, MAGEA4, and ZDHHC15) that were up-regulated by loss of XCI. HG-SOA with loss of XCI showed aggressive behavior (overall survival of partial reactivation of Xi group: HR 1.7, 95% CI 1.1-2.5, two copies of active X group: HR 1.4, 95% CI 1.0-1.9). Mutation and hypermethylation of BRCA1 were not frequent in HG-SOA with loss of XCI.

Introduction
One of the two copies of the X chromosome is randomly inactivated in females as a means of dosage compensation. Random X chromosome inactivation (XCI) is initiated by an X-inactive specific transcript (XIST), a 17 kilobase noncoding RNA, and processed by histone modification and promoter methylation, resulting in heterochromatin formation [1]. Heterochromatin is tightly packed DNA and prevents gene expression.
Loss of XCI is thought to have an oncogenic effect in hematologic malignancy in mice [2]. It is also observed in human breast and ovarian cancers [3][4][5]. Cancer cells with loss of XCI have two active copies of the X chromosome. It was suggested that the presence of the two active copies increases the expression of the oncogenes located on the X chromosome. However, it is unknown which genes are affected by loss of XCI.
Chromatin segregation errors that move both active X chromosomes (Xa) to the same daughter cell during mitosis are thought to be a main mechanism of loss of XCI in cancer cells. Breast cncer 1, early onset (BRCA1) inactivation is another mechanism of the loss of XCI. However, there is a debate whether BRCA1 dysfunction directly induces the loss of XCI or if does so indirectly due to increasing the probability of a chromosome segregation error [1,[6][7][8][9][10].
Loss of XCI is frequent in basal-like subtype and BRCA1 mutation-associated breast cancers, which show aggressive behavior [11][12][13]. However, the clinical implications of the loss of XCI in ovarian cancer are largely unknown. We investigated the clinical implications of the loss of XCI in ovarian cancer and the association between the loss of XCI and BRCA1 dysfunction.

Data acquisition
We used open source data generated by The Cancer Genome Atlas (TCGA) Genome Data Analysis Centers. High grade serous ovarian adenocarcinoma (HG-SOA) data on DNA methylation, XIST expression levels, copy number variation (CNV), loss of heterozygosity (LOH), mRNA expression, and clinical data, including age, race, and ethnicity, were downloaded from the Broad genome data analysis center (GDAC) Firehose website (http://gdac.broadinstitute. org/). Names of files downloaded from firehose website are described in S1 Text. Data on BRCA1 mutations were obtained from the c-bio portal for cancer genomics website (http:// www.cbioportal.org/public-portal/). Data on survival, tumor stage, and grade were gathered using the CGDS-R package which is a package of R for querying the Cancer Genomics Data Server and is hosted by the Computational Biology Center at Memorial-Sloan-Kettering Cancer Center [14]. Data on XIST expression levels in normal endometrium and clinical data were downloaded from the TCGA Data Portal (https://tcga-data.nci.nih.gov/tcga/tcgaHome2.jsp).

Brief description of the data
The beta values for DNA methylation status were estimated using the Illumina Infinium HumanMethylation27 array. The beta value is an estimate of the ratio of intensities between methylated and unmethylated alleles. Segmented copy number was estimated by log2 of the ratio of total intensity of the tumor and the normal tissue using Agilent 1M array. Segmented LOH was estimated by difference in allelic ratio (delta B) between tumor and normal tissue at loci where the normal sample is genotyped as heterozygous using 1MDuo SNP arrays. Large delta B values mean LOH. Normalized RNA-Seq by Expectation Maximization (RSEM) was used for estimating XIST expression [15]. Data on RSEM were generated using IlluminGA_RNASeqV2 or IlluminaHiSeq_RNASeqV2 platforms [16,17]. DNA sequencing of the exome was done with Illumina GAIIx or ABI SOLiD platforms. AgilentG4502A was used to estimate z-score of mRNA expression. Detailed information about participants, specimen processing, and analysis of each molecular profiling platform have been described elsewhere [18].

Ethics statement
All data used in this study were obtained from TCGA. The National Cancer Institute and National Human Genome Research Institute work with physicians who collect tissue for TCGA to gain approval with local institutional review boards (http://cancergenome.nih.gov/ abouttcga/policies/informedconsent).

Clustering analysis of X chromosome methylation
We selected 147 methylation array probes that showed large variations in beta values (larger than the third quartile) among probes targeting CpG islands in the X chromosome. CpG islands are major sites of methylation during XCI [19]. Ward's hierarchical clustering method was used to classify the methylation status of X chromosomes.

Segmental deletion or gain and LOH of X chromosome
The status of segmental deletion or gain and LOH of X chromosome were visualized using Integrative Genomics Viewer (IGV) [20,21].

Genes regulated by loss of X chromosome inactivation
Up-regulated genes by loss of XCI were investigated. To find hypomethylated genes in the HG-SOA with loss of XCI, we selected the genes which the difference of the median beta values were larger than 0.2. mRNA expression of the selected differently-methylated genes was compared using the Wilcox rank sum test. We selected genes for which the difference of the mRNA expression levels were larger than 0.4 and the p-value was less than 0.001. Gene set enrichment analysis (GSEA) was done to identify the cytogenetic locations where were regulated by the loss of XCI [22,23]. The positional gene sets were used under default setting.

Global methylation
We selected 4879 methylation array probes that showed large variations in beta values (larger than the third quartile) among probes targeting somatic chromosomes. The median beta value of each sample was compared among X chromosome methylation clusters using ANOVA and Tukey's test.

BRCA1 methylation classification
The DNA methylation status of BRCA1 was classified as hypermethylation or hypomethylation by k-means clustering methods. This method has been described in detail elsewhere [24].

Clinical data and survival
Patient and tumor characteristics were compared among the groups of X chromosomes via methylation clustering analysis. The significance was estimated using Fisher's exact test for categorical variables and analysis of variance for continuous variables. Univariate and multivariate survival analyses were performed with Log-rank and Cox regression tests, respectively.

XIST expression in normal female tissues
To determine the baseline level of XIST expression in normal tissue, we analyzed XIST expression data from normal tissues. XIST expression was relatively constant across different female normal tissues (S1 Fig.). Because normal ovarian tissue was not available in TCGA data, we used XIST expression data of normal endometrial tissue (median log-transformed normalized RSEM: 9.3, range: 8.8-10.0). We considered XIST expression to be low when its levels in HG-SOA were lower than the lowest value observed among normal endometrial tissue.

Clustering of X chromosome methylation
A total of 584 HG-SOAs were classified into six clusters according to X chromosome methylation (Fig. 1A). Cluster 1 showed heterozygous X chromosomes (Fig. 1C) with hypermethylation on most of the included loci on the X chromosome (Fig. 1A), as well as high XIST expression (Table 1). These results meant that the HG-SOAs of cluster 1 had both Xa and Xi. Clusters 3 and 4 showed selective p arm and q arm hypomethylation, respectively (Fig. 1A). This selective hypomethylation corresponded to deletion and LOH of the X chromosome in the p and q arms ( Fig. 1B and 1C). The hypomethylation in clusters 3 and 4 suggested that the deletion occurred in the inactive X chromosome (Xi), but not Xa. Clusters 3 and 4 also showed q and p arm gain, respectively (Fig. 1B). However, it is not clear if the gain occurred in Xi or Xa. Cluster 2 showed heterozygous X chromosomes (Fig. 1C) and high XIST expression (Table 1) like cluster 1. These results meant the HG-SOAs of cluster 2 had both Xa and Xi. But cluster 2 showed partial hypomethylation on both the p and q arms of the X chromosome unlike cluster 1 (Fig. 1A). The partial hypomethylation represented partial reactivation of Xi, thus resulting in an increased dosage of Xa. Clusters 5 and 6 showed LOH at X chromosomes ( Fig. 1C) but not deletion of a whole X chromosome (Fig. 1B). These results meant that clusters 5 and 6 had copy-neutral LOH. Cluster 5 and 6 also showed hypomethylation at most loci on the X chromosome (Fig. 1A) and low XIST expression (Table 1). These results suggested that HG-SOAs of cluster 5 and 6 had uniparental two copies of an active X chromosome. Each cluster was designated as the preserved Xi group in the case of cluster 1 (n = 175), partial reactivation of the Xi (Xa + ) group for cluster2 (n = 100), p arm deletion of the Xi (Xi-p -) group for cluster 3 (n = 35), q arm deletion of the Xi (Xi-q -) group for cluster 4 (n = 44), and two copies of the Xa (two Xa) group for clusters 5 and 6 (n = 230). The result of the post-hoc analysis of XIST expression is summarized in S1 Table. Genes regulated by loss of X chromosome inactivation We found four genes including X antigen family, member 3 (XAGE3), zinc finger protein 711 (ZNF711), melanoma antigen family A, 4 (MAGEA4), and zinc finger, DHHC-type containing  15 (ZDHHC15), which showed significantly higher expression in the two Xa group than in the preserved Xi group (S2 Fig.). GSEA did not identify significantly over-expressed cytogenetic locations of X chromosome in the two Xa group than the preserved Xi group (S2 Table). Contrary to the expectation, the preserved Xi group was more enriched higher expression of genes located at Xp22 and Chr9q22 than the two Xa group (S3 Table).

Global methylation
Clusters 2 and 6 showed lower median beta values of methylation on somatic chromosomes (overall P-value <0.001, Fig. 2). The results of the post-hoc test are summarized in S4 Table. BRCA1 mutation and hypermethylation BRCA1 mutation information was available for 316 HG-SOAs. Mutation and promoter lesion hypermethylation of BRCA1 were not frequent in HG-SOAs with a loss of XCI (Xa + and two Xa groups). The frequency of germline mutations in BRCA1 was significantly higher in the Xiqgroup (Table 1).

Clinical and tumor characteristics
Patients in the Xa + and two Xa groups were significantly older than those in the other groups ( Table 2). There were no significant differences in race, ethnicity, tumor stage, grade, or residual tumor size among the groups. The results of the post-hoc test for difference of patient age among the groups are summarized in S5 Table.

Survival analysis and chemotherapy response
Of the 584 patients, 99 and 188 were excluded from the overall and progression free survival analysis, respectively, due to the absence of available data. Patients in the Xa + and two Xa groups showed the worse overall survival (Fig. 3A). Patients in the Xa + group showed worse progression free survival (Fig. 3B). The prognostic significance remained after adjusting for age, tumor stage, and residual tumor size using multivariate Cox regression test (Table 3). Patients in the Xi-por Xi-qgroups showed no survival differences when compared to the preserved Xi group. HG-SOA with Xi-p -, Xa + , or two Xa showed more frequently chemotherapy resistance, but this was not statistically significant (Table 2).

Discussion
We found that dysregulation of XCI is common in HG-SOA and includes various different types of alterations, such as two copies of Xa, partial reactivation of Xi, and p or q arm deletion of Xi. Among those with dysregulation of XCI, two copies of Xa and partial reactivation of Xi increased the dose of Xa (loss of XCI). Loss of XCI was associated with poor prognosis in HG-SOA. Mutation or hypermethylation of BRCA1 were infrequent in HG-SOAs with a loss of XCI. This result suggests that BRCA1 dysfunction may not be related to the loss of XCI in HG-SOA. Loss of XCI induced up-regulation of oncogenes located on the X chromosome was suggested as a mechanism of tumorigenesis in HG-SOA. However, it was unknown which gene was affected by the loss of XCI. We found four genes (XAGE3, ZNF711, MAGEA4, and ZDHHC15) that were up-regulated by loss of XCI. XAGE3 and MAGEA4 belong to a family of cancer-testis antigens. MAGEA4 belongs to type 1 MAGEs, which cluster in the X chromosome and are regulated by methylation. MAGEA4 was considered as a possible oncogene and a target for immunotherapy [25]. This result suggests that up-regulation of cancer-testis antigens has a role in the aggressiveness of HG-SOA with loss of XCI and explains the mechanism of aberrant expression of cancer-testis antigens in some HG-SOAs.
In previous study, most sporadic and BRCA1-associated, basal-like breast cancer showed two active X chromosomes and loss of XCI (61%), reactivation of XCI (22%), and gain of Xa (16%) [13]. This study also reported that genes located at Xp22 and Xq26-28 were overexpressed in basal-like breast cancer and suggested that over-expression of genes located at Xp22 was an important feature of basal-like breast cancer [13]. In our HG-SOA study, the preserved Xi group was more enriched higher expression of genes located at Xp22 than the two Xa group, contrary to the breast cancer study. The over-expression of genes located at Xp22 is a feature of basal-like breast cancer, but not HG-SOA with loss of XCI.
The Xa + group showed more hypomethylation than the preserved Xi group, but less hypomethylation than the two Xa group. This represents partial reactivation of Xi. Global hypomethylation in cancer cells is known to occur. A previous report suggested that global hypomethylation is age dependent [26]. In HG-SOA, the Xa + group showed more global hypomethylation and occurred in older patients. These results suggest that the partial reactivation of the X chromosome observed in the Xa + group might be induced by global hypomethylation.
A previous study suggested that loss of XCI was associated with chemotherapy resistance in HG-SOA [27]. In our study, HG-SOA with a loss of XCI (Xa + and two Xa) showed more chemotherapy resistance, but it was not statistically significant. The association between loss of XCI and chemotherapy response might be weak or absent. However, the small sample size might preclude a definite conclusion.
In our study, some HG-SOAs showed q arm deletions of Xi, which were associated with BRCA1 germline mutations. The same q arm deletion of Xi has been reported in ovarian tumors with low malignant potential [4]. Two different pathways have recently been suggested for ovarian cancer and were designated 'type I' and 'type II' [28]. Type I ovarian serous cancer develops from benign serous adenoma and serous neoplasms of low malignant potential to low grade serous carcinoma. Type II develops from intraepithelial carcinomas of the fallopian tube or ovarian surface epithelium. HG-SOA is considered to be type II. HG-SOA with a q arm deletion of Xi suggests a different pathogenesis of HG-SOA associated with BRCA1 germline mutation.
In conclusion, our observations provide evidence that dysregulation of XCI is common in HG-SOA and is associated with poor outcome. Loss of XCI induced up-regulation of cancertestis antigens and this may have a role in tumor aggressiveness. The role of BRCA1 in the loss of XCI might be limited. BRCA1 germline mutations may have a role in the differential pathogenesis of HG-SOA associated with a q arm deletion of Xi. A) The points represent the more hypomethylated gene in the two Xa group than the preserved Xa group. X-axis is difference in mRNA expression z-scores between two Xa group and preserved Xi group. The difference of z-scores of mRNA and P-values were estimated using the Wilcox rank sum test. Four labeled genes are significantly more highly expressed in the two Xa group than in the preserved Xi group. B) Box plots represent the distribution of mRNA expression. The top and bottom of the box represent the 75 th and 25 th percentile, the line in the box represents median. The points represent the z-scores of mRNA expression. (TIF) S1