Mutation and Methylation Analysis of the Chromodomain-Helicase-DNA Binding 5 Gene in Ovarian Cancer^,

Abstract

Chromodomain, helicase, DNA binding 5 (CHD5) is a member of a subclass of the chromatin remodeling Swi/Snf proteins and has recently been proposed as a tumor suppressor in a diverse range of human cancers. We analyzed all 41 coding exons of CHD5 for somatic mutations in 123 primary ovarian cancers as well as 60 primary breast cancers using high-resolution melt analysis. We also examined methylation of the CHD5 promoter in 48 ovarian cancer samples by methylation-specific single-stranded conformation polymorphism and bisulfite sequencing. In contrast to previous studies, no mutations were identified in the breast cancers, but somatic heterozygous missense mutations were identified in 3 of 123 ovarian cancers. We identified promoter methylation in 3 of 45 samples with normal CHD5 and in 2 of 3 samples with CHD5 mutation, suggesting these tumors may have biallelic inactivation of CHD5. Hemizygous copy number loss at CHD5 occurred in 6 of 85 samples as assessed by single nucleotide polymorphism array. Tumors with CHD5 mutation or methylation were more likely to have mutation of KRAS or BRAF (P = .04). The aggregate frequency of CHD5 haploinsufficiency or inactivation is 16.2% in ovarian cancer. Thus, CHD5 may play a role as a tumor suppressor gene in ovarian cancer; however, it is likely that there is another target of the frequent copy number neutral loss of heterozygosity observed at 1p36.

Introduction

Chromodomain helicase DNA-binding 5, CHD5, is a member of a subclass of the chromatin remodeling Swi/Snf proteins [1]. Proteins within this subclass contain a Swi-Snf-like helicase and two chromodomain motifs. Members of this protein class have been shown to be part of complexes that mediate chromatin remodeling and affect gene transcription. Recently, Bagchi et al. [2] identified CHD5 as a putative tumor suppressor gene through functional analysis in a mouse model. The model suggested that partial CHD5 deficiency compromises p53 signaling and therefore abrogation of CHD5 function might represent a generic mechanism for cancer development. Evidence that CHD5 functions as a tumor suppressor in primary human cancers has come principally from studies of neuroblastoma where loss of the CHD5 locus on chromosome 1p36.3 is very common [3]. CHD5 expression is consistently down-regulated in primary neuroblastomas and cell lines [4] and may be affected by methylation in neuroblastoma cell lines based on reexpression after treatment with 5-azacytidine [5].

To date, the only evidence for a broader role of CHD5 in human cancer has come from a genome-wide breast and colon cancer genome sequencing study where CHD5 was proposed as a “CAN-gene” [6]. Heterozygous missense mutations were identified in 2 of 24 primary breast cancers and 1 of 11 cell lines. Loss of heterozygosity (LOH) at 1p is a common event in breast and ovarian cancers and has been shown to correlate with poor survival [7,8], and therefore, CHD5 is a logical candidate for the target of this LOH in these cancer types. In this study, we have extended the range of tumors where CHD5 plays a tumor suppressor role by demonstrating the existence of somatic mutations and methylation in primary epithelial ovarian cancers.

Materials and Methods

DNA Samples

One hundred and twenty-three primary ovarian cancers (56 serous, 20 mucinous, 34 endometrioid, and 13 other) were obtained from patients presenting to the hospitals in the south of England, UK. DNA for mutation and methylation analyses was extracted from whole fresh frozen specimens. Representative sections were hematoxylin and eosin-stained, and all tumors were assessed to contain >60% tumor epithelium. Normal DNA was extracted from matching peripheral blood samples. Matching tumor and normal DNA from 60 primary breast cancers was provided by the Peter MacCallum Cancer Centre tissue bank or by Dr Nick Hayward (Queensland Institute for Medical Research, Brisbane, Australia). This study was approved by institutional ethics committees. Before mutation screening, all stock DNA underwent whole genome amplification (WGA) using the Repli-G Phi-mediated amplification system (Qiagen, Hilden, Germany). To minimize the potential for generation of artifacts,WGA was carried out in triplicate, using 25 ng of primary DNA, and the products were pooled.

Mutation Analysis

The CHD5 gene was analyzed by high-resolution melt (HRM). Exons 9, 13, 30, and 31 were amplified using previously reported primer sequences [6]. For the remaining exons, primers were designed to amplify each of the 41 exons and intron/exon boundaries of the coding sequence in 156- to 477-bp fragments (median, 202 bp). Primers were designed using the software packages ExonPrimer and Primer3 [9]. Primer sequences and amplification conditions are listed in Table W1. Owing to their larger size, exons 5, 10, 11, 15, 23, and 38 were amplified in two overlapping fragments. TP53 (exons 5–8), KRAS (amino acids 1–36), and BRAF (V600E) were analyzed previously [10] or using HRM (Table W1). High-resolution melt was carried out in duplicate using polymerase chain reaction (PCR) products amplified from 10 ng of WGA template DNA. Gene scanning analyses were carried out for each exon using the LightCycler 480 (Roche Diagnostics, Mannheim, Germany). Samples with replicated shifts in the DNA melt curves were reamplified, and the PCR product was directly sequenced using BigDye Terminator v3.1 (Applied Biosystems, Foster City, CA). Somatic alterations were confirmed by resequencing from the unamplified stock tumor DNA and the matching normal DNA. Polymerase chain reaction products for sequencing were purified by nucleotide removal columns or agarose gel extraction (Qiagen).

Analysis of CpG Island Methylation by Methylation-Specific Single-Stranded Conformation Polymorphism and Bisulfite Sequencing

The CHD5 CpG island was identified in University of California Santa Cruz genome browser (genome.ucsc.edu) and methylation-specific single-stranded conformation polymorphism (MS-SSCP) PCR primers designed from genomic DNA sequence using MethPrimer (www.urogene.org/methprimer) [11]. Both forward and reverse oligonucleotide primers were fluorescently labeled with either FAM or HEX. Primer sequences are listed in Table W1.

DNA samples were treated with bisulfite using the MethylEasy Kit (Human Genetic Signatures, Sydney, Australia) following the manufacturer's instructions. After PCR amplification, products were analyzed by SSCP using the ABI 3130 Genetic Analyzer (Applied Biosystems) as described previously [12]. Samples showing a shift in mobility were sequenced, whereas the remainder were considered normal. Twenty-four samples were sequenced without performing SSCP. SssI methylase-treated normal DNA was used as a positive control for CpG island methylation. This enzyme methylates all CpGs before bisulfite treatment. Polymerase chain reaction products for sequencing were reamplified using unlabeled primers and purified by nucleotide removal columns (Qiagen). Purified PCR products were sequenced in both forward and reverse directions using BigDye Terminator v3.1 (Applied Biosystems). Full methylation at a particular CpG was defined as follows: >60% of the average bisulfite sequencing signal was “C,” whereas partial methylation was 40% to 60%. “Methylation-positive” was defined as at least 18 of the 35 CpGs within the PCR product showing full or partial methylation. “Partial methylation” was defined as at least 6 of 35 CpGs showing full or partial methylation.

Real-time Reverse Transcription Quantitative PCR

RNA was extracted from cell lines or from microdissected tumors using the miRVana RNA isolation kit according to the manufacturer's instructions (Ambion, Austin, TX). Samples were reverse transcribed and amplified according to the Affymetrix Gene 1.0ST array protocol (Affymetrix, Santa Clara, CA) and 30 ng of the ssDNA was used per 10-µl PCR. This RNA amplification step was performed to ensure sufficient template in the reaction for reproducible quantitation. Primers were designed to CHD5 and a control gene PGK1 (Table W1), and PCR was performed with the SYBRgreen QPCR mix (ThermoScientific, Waltham, MA) and the LightCycler 480 (Roche Diagnostics, Mannheim, Germany). Cp (“crossing point”) was calculated using the second derivative maximum method. The relative levels of expression of CHD5 and PGK1 were calculated using a normal ovarian surface epithelial cell line (HOSE) RNA as a control standard curve.

Single Nucleotide Polymorphism Array Analysis

Affymetrix single nucleotide polymorphism (SNP) 500K and SNP 6.0 Mapping arrays were performed on unamplified DNA obtained from microdissected fresh frozen tissue sections, and the data were analyzed using Partek Genomics Suite (Partek, St Louis, MO), GTYPE (Affymetrix), and CNAG software as described previously [13]. Loss of heterozygosity was determined by examination of the allele-specific copy number (CN) ratios in CNAG [14,15].

Results and Discussion

Mutation Analysis of CHD5 Using HRM

High-resolution melt analysis covering all of the CHD5 protein coding sequence and intron/exon boundaries (a total of 47 PCR products) was carried out on DNA from 123 primary ovarian cancers and 60 primary breast cancers. No mutations were identified in any of the breast tumors, but three ovarian tumors were shown to harbor somatic heterozygous missense alterations: C4992T (Ser1631Phe), C4412T (Arg1438Cys) and G4386A (Arg1429Gln) (Figure 1 and Table 1). The somatic nature of the mutations was confirmed by sequencing matching normal lymphocyte DNA. The Ser1631Phe and Arg1429Gln mutations were identified in grade 3, stage III serous type tumors, and the Arg1438Cys mutation was identified in a grade 2 stage IA endometrioid tumor. Analysis of the effect of the missense mutations on protein structure and function was performed using three different prediction algorithms: PolyPhen [16], SIFT [17], and PMUT [18]. All three mutations were predicted to affect protein function by at least two of the prediction algorithms (Table W2).

Figure 1.

Somatic alterations in CHD5. (A) Somatic mutation in ovarian tumor IC318. The shift in exon 33 melting profile for IC318 is indicated in red. (B) Somatic mutations in exon 29. The melting profiles are shown for tumor samples IC114 (purple) and IC139 (red). (C) Normal and tumor sequence traces for exon 33 (left, arrow indicates C4992T heterozygous mutation) and exon 29 (center, arrow indicates C4412T; and right, arrow indicates G4386A).

Table 1.

CHD5 Coding Sequence Alterations.

Exon	Sequence Alteration	Codon	Amino Acid	Heterozygote Frequency
				Breast Cancer	Ovarian Cancer
Somatic mutations
29	G4386A	CGG to CAG	Arg1429Gln	0/60	1/121
29	C4412T	CGC to TGC	Arg1438Cys	0/60	1/121
33	C4992T	TCC to TTC	Ser1631Phe	0/60	1/123
Polymorphisms
4*	G529C	CTG to CTC	Leu143Leu	7/60	13/123
5	G679C	CGG to CGC	Arg193Arg	0/60	1/122
6	C876G	TCC to TGC	Ser259Cys	1/60	0/123
7*	C1003T	TTC to TTT	Phe301Phe	27/60	41/123
7	G1014A	AGC to AAC	Ser305Asn	1/60	0/123
8*	A1204G	GTA to GTG	Val368Val	22/60	39/123
9	C1378A	GGC to GGA	Gly426Gly	2/60	0/123
10	T1666C	CAT to CAC	His521His	0/60	1/123
11	C1768T	TAC to TAT	Tyr556Tyr	1/60	0/123
12*	C1957T	TAC to TAT	Tyr619Tyr	9/60	15/123
14	G2200A	CTG to CTA	Leu700Leu	1/60	0/123
15*	C2479T	AAC to AAT	Asn793Asn	13/60	21/123
16*	T2593C	ATT to ATC	Ile831Ile	24/60	51/123
18	G2878A	CCG to CCA	Pro926Pro	0/60	1/123
22*	G3436A	GCG to GCA	Ala1112Ala	9/60	13/123
28	C4336T	CTC to CTT	Leu1412Leu	3/60	4/123
31*	T4715C	TCG to CCG	Ser1539Pro	30/60	53/121
32	G4828T	ATG to ATT	Met1576Ile	1/60	0/123
34	C5089T	TCC to TCT	Ser1663Ser	0/60	1/123
35	C5170T	GAC to GAT	Asp1690Asp	0/65	1/123
36	C5344T	ATC to ATT	Ile1748Ile	0/60	1/123
36	C5349T	ACG to ATG	Thr1750Met	1/60	0/123

Asterisks indicate previously identified polymorphisms.

The absence of somatic mutations in 60 primary breast cancers is at variance with the 9% mutation frequency (2/24 primary breast cancers and 1/11 breast cancer cell lines) reported by Sjöblom et al. [6] in a genome-wide sequencing screen. It is unlikely that the absence of somatic mutations in the breast cancers is caused by lack of sensitivity of HRM, which has a growing reputation as a highly sensitive mutation detection technique [19,20]. We were able to detect a large number of polymorphisms located in the coding sequence (22 variants) or within the intron sequences (24 variants; Table 1, Table W3, and Figure W1). In our hands, sequence variants were readily detectable even in samples where there was normal DNA contamination or where the variant was present at a low abundance due to LOH (Figure W1). In addition, we were able to detect compound polymorphisms, in which a sample with a common polymorphism also had a less frequent polymorphism present (Figure W1). The distribution of histologic subtypes, grade, and stage of tumors examined in our study and by Sjöblom et al. [6] was similar, suggesting that the discrepancy in tumor mutation frequency might be due to chance given that it was not statistically significant (Fisher's exact test, P = .079). Our study does suggest that the frequency of CHD5 mutations in breast cancers might be lower than the 9% reported previously and highlights the importance of following up leads from genome-wide sequencing screens with independent sample sets. This is the first study of somatic mutation and methylation of CHD5 in ovarian cancer, and the data indicate that CHD5 has a tumor suppressor role in a subset of cases. Interestingly, all three mutations were detected in tumors that were heterozygous across the CHD5 locus.

The CHD5 Promoter Is Sometimes Methylated in Primary Ovarian Cancer

Because the expression of tumor suppressor genes is sometimes reduced as a consequence of promoter hypermethylation, we examined the promoter of CHD5 in ovarian tumors. CHD5 has a predicted CpG island spanning 1577 bp, beginning 631 bp upstream of the transcription start site and comprising 180 CpGs (Figure 2). Methylation was detected in three of six ovarian cancer cell lines, with at least 80% of CpGs fully methylated as determined by bisulfite sequencing (Figure 2). Primary ovarian cancers showed less frequent methylation, with 5 of 48 methylated and 2 of 48 partially methylated. We verified that methylation was not the result of contamination of tumor by fibroblasts or lymphocytes by bisulfite sequencing the CHD5 promoter in one cancer-associated fibroblast cell line, one microdissected stromal DNA sample, and one normal lymphocyte DNA sample. None of these samples showed any methylation. Promoter methylation was more frequent and extensive in the cell lines than the primary tumors, suggesting that methylation of the CHD5 promoter may be common in the transition from primary tumor to cell line. Notably, two methylated samples (IC114T and IC139T) also carried somatic mutations (Table 2). We carried out real-time reverse transcription PCR for CHD5 on samples for which sufficient RNA could be extracted. Samples with methylated promoters showed uniformly low levels of expression (Figure 2). Unmethylated samples showed variable expression levels, suggesting that there may be other mechanisms by which CHD5 expression is regulated. The three CHD5 wild type cancers with a high level of promoter methylation comprised a mixed mullerian tumor and two mucinous tumors.

Figure 2.

CHD5 promoter methylation in ovarian cancer. (A) University of California Santa Cruz genome browser view of the CHD5 gene, which is located on the reverse strand (genome.ucsc.edu). The CHD5 promoter contains a strong CpG island as demonstrated in the MethPrimer (www.urogene.org/methprimer) output below. The location of the primers used for SSCP and sequencing is shown (F1 and R1). (B) Sequence electropherogram traces from primary ovarian tumors showing methylated and unmethylated samples. (C) Summary of bisulfite sequencing from cell lines and primary tumors showing methylation. CpG dinucleotide number within the PCR product listed across the top from distal to proximal relative to transcription start site. Black, fully methylated (>60%); gray, partial methylation (40–60%); white, <40% methylation. An additional 15 tumors were sequenced that showed no methylation. Samples with a CHD5 mutation are shown in bold. Sample IC139T data are based on cloning the PCR product and sequencing five clones as the direct sequencing was poor. (D) Quantitative PCR of CHD5. The expression level of CHD5 is shown as a ratio relative to the control gene. An asterisk indicates a CHD5-mutated sample. SEs are shown.

Table 2.

Summary of CHD5 and Pathway Interactions.

Sample	Subtype	CHD5	KRAS Mut	KRAS CN	BRAF Mut	TP53 Mut
IC114	Endometrioid	Arg1438Cys, methylation	wt	—	wt	wt
IC139	Serous	Arg1429Gln, methylation	wt	—	V600E	wt
IC318	Serous	Ser1631Phe	wt	Gain	wt	wt
IC197	MMT	Methylation	wt	—	wt	wt
IC50T	Mucinous	Methylation	wt	n	V600E	wt
IC80T	Mucinous	Methylation	G12V	n	wt	wt
IC281	Serous	CN loss	wt	n	wt	—
IC288	Serous	CN loss	wt	n	wt	—
IC382	Serous	CN loss	wt	n	wt	—
IC594	Endometrioid	CN loss	wt	n	wt	wt
P0566	Mixed	CN loss	wt	n	wt	wt
P5338	Serous	CN loss	—	Gain	—	—
IC022	Serous	CN gain	wt	Gain	wt	Y220C
IC135	Serous	CN gain	wt	n	wt	del156–159
IC434	Endometrioid	CN gain	wt	n	wt	wt

(—) indicates not done; n, normal CN; wt, wild type.

Copy Number Loss Is an Alternate Mechanism of CHD5 in Ovarian Cancer

A mouse model of CHD5 deficiency suggested that haploinsufficiency of CHD5 may contribute to cancer progression rather than a “two-hit” mechanism expected of a classic tumor suppressor [2]. In light of this, we considered the possibility that CHD5 might be the target of CN loss, which would be consistent with (but not proof of) targeted haploinsufficiency of CHD5. We evaluated 85 primary ovarian cancers (56 with known CHD5 mutation status) using Affymetrix 500K or 6.0 SNP Mapping arrays [13]. These arrays are able to detect both CN losses and CN neutral LOH. We detected CHD5 CN loss (defined as a log₂ ratio of <-0.3) [13] in 6 (7%) and gain (log₂ ratio of >0.3) in 3 (3.5%) of 85 ovarian tumors (Figure W2). The CHD5 region showed LOH (both CN neutral and CN loss) in 30 samples (35%); however, LOH at any locus on chromosome 1p was detected in 39 samples (46%) and suggests that another gene(s) is the target of LOH on chromosome 1p (Figure W3).

The mouse and in vitro models of CHD5 gene dosage suggested that CHD5 acts within the p53 pathway, with loss of CHD5 resulting in reduced expression of p53 target genes. CHD5 loss also interacted with KRAS to promote transformation. We therefore looked to see whether CHD5 mutation or methylation coincided with alterations in the KRAS pathway or TP53 mutation. Interestingly, three of six of the CHD5 mutation- or methylation-affected samples had alteration in the KRAS pathway, one by KRAS mutation and two by BRAF mutation (P = .04, Fisher's exact test), whereas none had mutation of TP53 (Table 2). However, this association was not evident among the samples showing CHD5 CN loss only with none from five samples having either KRAS or BRAF mutation (P = 1, Fisher's exact test).

Altogether, when mutation (2.4%) and methylation (without mutation, 6.7%) are combined, CHD5 is affected in 9.1% of ovarian cancers, which is extended to 16.2% if cases with CN loss (7.1%) are included. However, the consequence of heterozygous CN loss at CHD5 in ovarian cancer is not clear, given that two of three samples with mutation also showed methylation, suggesting biallelic inactivation rather than haploinsufficiency. Our study supports the contention that CHD5 is a tumor suppressor gene in a subset of ovarian tumors. The disparity between the frequency of CHD5 alteration (16%) and LOH at 1p (46%) strongly suggests another tumor suppressor gene in the region. CHD5 mutation and/or methylation, but not CN loss, may co-operate with the KRAS pathway in tumorigenesis; however, the number of samples is small and this result will require future validation. The lack of CHD5 mutations in the breast cancer samples screened in this study is in contrast to the 9% frequency reported previously [6]. Because the difference in mutation frequencies is not statistically significant, it is likely to be due to chance, although a contribution of the screening methodologies used cannot be excluded.

Abbreviations

CN

copy number

HRM

high-resolution melt

LOH

loss of heterozygosity