Mechanisms of CHD5 Inactivation in Neuroblastomas

Abstract

Purpose

Neuroblastomas (NBs) have genomic, biological and clinical heterogeneity. High-risk NBs are characterized by several genomic changes, including MYCN amplification and 1p36 deletion. We identified the chromatin-remodeling gene CHD5 as a tumor suppressor gene that maps to 1p36.31. Low or absent CHD5 expression is associated with a 1p36 deletion and an unfavorable outcome, but the mechanisms of CHD5 inactivation in NBs are unknown.

Experimental Design

We examined 1) the CHD5 sequence in 188 high-risk NBs investigated through the TARGET initiative; 2) the methylation status of the CHD5 promoter in 108 NBs with or without 1p36 deletion and/or MYCN amplification; and 3) mRNA expression of CHD5 and MYCN in 814 representative NBs using TaqMan low-density array microfluidic cards.

Results

We found no examples of somatically acquired CHD5 mutations, even in cases with 1p36 deletion, indicating that homozygous genomic inactivation is rare. Methylation of the CHD5 promoter was common in the high-risk tumors, and it was generally associated with both 1p deletion and MYCN amplification. High CHD5 expression was a powerful predictor of favorable outcome, and it showed prognostic value even in multivariable analysis after adjusting for MYCN amplification, 1p36 deletion, and/or 11q deletion.

Conclusions

We conclude that 1) somatically acquired CHD5 mutations are rare in primary NBs, so inactivation probably occurs by deletion and epigenetic silencing; 2) CHD5 expression and promoter methylation are associated with MYCN amplification, suggesting a possible interaction between these two genes; and 3) high CHD5 expression is strongly correlated with favorable clinical/biological features and outcome.

Introduction

Neuroblastoma (NB) is the most common extracranial solid tumor of childhood, and it accounts disproportionately for childhood cancer deaths (1). NBs demonstrate clinical heterogeneity, from spontaneous regression to relentless progression. We and others have identified different patterns of genomic change that underlie these disparate clinical behaviors (2-7). Deletion of the short arm of chromosome 1 (1p) occurs in 35% of primary tumors and 80% of tumor-derived cell lines, representing one of the most characteristic genomic changes in NBs (8-11). Presumably, 1p deletion reflects loss of a tumor suppressor gene (TSG) from this region. We analyzed over 1,200 NBs and mapped the smallest region of consistent deletion (SRD) to a ∼2 Mb region on 1p36.31 (12, 13). Indeed, the SRD identified by most other groups mapping 1p deletions in NBs overlaps our region (14-17). We analyzed 23 genes mapping to the maximal SRD we defined on 1p36.31 and identified CHD5 as the most likely TSG within this region (12, 18, 19).

The CHD5 gene encodes a novel member of the chromodomain-helicase-DNA binding (CHD) family (19). This gene contains 42 exons spanning over 78 kb, and it encodes a transcript of 9.6 kb. The encoded protein is predicted to contain four distinct functional regions that are shared with other CHD proteins. These include two CH3 type PHD zinc finger domains; two chromodomains with motifs characteristic of the CHD family; a DEAD-like helicase domain and a SNF2-like helicase/ATPase domain that regulate chromatin conformation; and a conserved motif in the C-terminal third of the protein possibly related to DNA binding. All CHD proteins have nuclear localization signals (20). CHD5 has greater homology with CHD3 and CHD4 than with other CHD family members. There is almost exclusive expression in the nervous system and in testis, and expression is virtually undetectable in a panel of NB cell lines compared to fetal brain (12, 19, 21).

We transfected CHD5 into four NB cell lines, and clonogenicity and tumorigenicity were suppressed only in lines with 1p deletion (18). Although mutations were rare, we found epigenetic silencing of the remaining allele in lines with 1p deletion. High CHD5 expression was associated with favorable clinical and biological risk factors in 101 NBs retrospectively analyzed by microarray expression profiling (22). Because these prior studies had been conducted primarily on NB cell lines and we used semiquantitative CHD5 expression data for the cohort of primary tumors, we wanted to assess a larger number of representative primary NBs using quantitative real-time RT-PCR to definitively assess the prognostic value of CHD5 expression. We also wanted to determine if mutation or promoter methylation contributed to the decrease or loss of CHD5 expression in these tumors. We assessed 188 primary NBs for coding sequence or splice site mutations. We also examined the methylation status of the CHD5 promoter in 108 primary NBs to determine if promoter methylation correlated with 1p deletion, MYCN amplification or CHD5 expression. Finally, we examined the level of CHD5 expression in 814 NBs using quantitative real-time RT-PCR to determine its association with clinical and biological variables as well as outcome.

Patients and Methods

CHD5 mutation studies

We examined 188 high-risk NB cases that were chosen for study by the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) initiative of the National Cancer Institute (http://target.cancer.gov/). This initiative aims to uncover the genomic factors that distinguish groups of children with favorable prognoses from those that do not respond to treatment, and to accelerate research in novel markers and drug development for NB and other childhood cancers. All tumors in this analysis were from patients with high-risk NB: disease stage 4 (23), age over 18 months (24, 25), with or without MYCN amplification (3, 26, 27). Sequence analysis examined tumor DNA (and the corresponding constitutional DNA, if available) for coding domain sequence (CDS) or splice site mutations.

Paired tumor and normal DNAs were obtained from the TARGET project for validation and determination of whether or not the variant was somatically acquired. A total of 19 sequence variations were identified in 17 cases (out of 188 total high-risk cases) by the TARGET gene sequencing project. We designed primers around the areas of suspected mutation (primer sequences available on request), and the region of suspected mutation was re-sequenced from the tumor DNA and constitutional DNA (if available) in both directions using the ABI 3730 DNA Analyzer. The sequences obtained were compared to the canonical CHD5 mRNA (accession no. NM_015557) (19) and genomic sequence (chr1:6,161,853-6,240,183) in GenBank.

CHD5 promoter methylation studies

We chose 108 cases for which we had determined CHD5 expression to analyze the methylation status of the CHD5 promoter. We chose 42 cases with neither 1p deletion nor MYCN amplification, 20 cases with 1p deletion only, 21 cases with MYCN amplification only, and 25 cases with both 1p deletion and MYCN amplification. These groups would allow us to determine whether promoter methylation was associated with 1p deletion, MYCN amplification, or both.

Genomic DNA was isolated from primary NBs. Genomic DNA (1.5 μg) was treated with sodium bisulfite using an Epitect Bisulfite Kit (Qiagen, Valencia, CA) to modify unmethylated cytosine residues to uracil. Modified DNA was amplified by PCR with seven sets of primers in the CHD5 promoter region (-1093 to +168) to survey for methylation status. PCR products were purified using agarose gel electrophoresis and the Gel Purification Kit (Qiagen, Valencia, CA), and the products were sequenced. To compare CHD5 promoter methylation to CHD5 expression, we counted the number of methylated CpGs in the target region between -780 and -480 in a given case, and this was used as a methylation score. Previously, we determined that this region showed differential methylation in two NB cell lines with 1p deletion (and MYCN amplification) compared to two lines with neither genomic abnormality (18). We then compared the methylation score to the level of CHD5 expression.

To validate that sequencing of promoter PCR products was representative, we identified 5 representative cases from each of the four groups and cloned the modified DNA products using pGEM-T Easy Vector System (Promega, Madison, WI). Ten colonies from each case were isolated and sequenced to determine the percent methylation for each CpG, i.e., the number of clones with methylation of the given CpG divided by the total sample number of each group. The percentage methylation at each CpG site was then averaged for each of the four genomic groups (data not shown). Comparison of these data to the group averages determined from the PCR products showed that the latter were representative.

CHD5 expression studies

For CHD5 expression studies, we selected 814 NB samples from patients diagnosed between 1995 and 2008 that were representative based on patient age, International NB Staging System (INSS) disease stage (23), and the prevalence of MYCN amplification. All patients were enrolled on Children's Oncology Group (COG) Biology Protocol ANBL00B1. Clinical and outcome data were stored and analyzed in the COG Statistics and Data Center (WBL, Gainesville, FL). Tumor samples were snap frozen, and samples of tumor RNA were obtained from the COG Nucleic Acid Bank.

We used a quantitative, real-time reverse transcription polymerase chain reaction (qRT-PCR) technique to measure the expression of CHD5 (as well as MYCN) mRNA. We reverse transcribed 1-2 μg RNA using the Applied Biosystems Inc. (ABI, Foster City, CA) High Capacity cDNA Archive Kit and standard protocol. Reactions were generally performed at 20 μl total volume. Quantitative gene expression analysis was performed via TaqMan RT-PCR with ABI TaqMan Low Density Array (LDA) microfluidic cards. Custom-designed arrays contained 8 sample-loading ports, each with 16 detectors in triplicate, for 48 reaction chambers per port, and a total of 384 reactions per card. Twelve primer/probe sets (detectors) for genes of interest and 4 endogenous controls were included on each card (details available on request). qRT-PCR was performed using the ABI PRISM 7900HT Sequence Detection System (ABI). Gene expression for each sample-detector pair was measured as a fold change in amplification relative to a calibrator sample. First-pass analysis was performed using the comparative Ct method with ABI PRISM SDS 2.2.2 and RQ Manager software. Amplification efficiencies of all genes were assumed to be approximately equal to the efficiencies of the endogenous controls (within 5%).

Prior to assay selection, 2 endogenous control LDA's (ABI) were run with 16 representative samples to determine which 4 endogenous control detectors to include in the study. GeNorm (28) was used to select the 3 endogenous control primer/probes with the most stable expression among the 16 samples. ABI custom arrays must include either 18S RNA or GAPDH; of the two, GAPDH was more stably expressed and was included. IPO8, UBC, and HPRT1 were included as additional endogenous controls because they were the most stably expressed in our samples, and there was precedent for use of these genes for NB (28). In addition to internal control normalization, we performed further analysis prior to statistical analysis, using Integromics' RealTime StatMiner software. With this software, the most stable 2 endogenous controls for the entire sample set were selected via the software's internal GeNorm application. The geometric mean of the expression of these 2 endogenous controls was used to normalize samples, using human fetal brain total RNA converted to cDNA as a calibrator sample (Clontech, Cat. No. 636526). The expression levels of CHD5 and MYCN were dichotomized into low and high using the median expression values for this cohort.

Statistical analyses

CHD5 mutation analysis (188 high-risk cases)

To estimate predictive power, we calculated the probability of finding a CHD5 mutation in at least 2 out of 200 primary NBs using two models that assumed mutation rates of 0.05 and 0.01, assuming that the number of NBs with CHD5 mutation followed a binomial distribution. The predicted probability for each model is 0.99 and 0.63 for mutation rates of 0.05 and 0.01, respectively, when using the equation N=log(1-r)/log(1-p), where N=# samples, r=confidence, and p=frequency. To find a 5% mutation frequency with 90% confidence, we would need to sequence 45 samples. The power to detect a 5% mutation rate goes from 90% to 99% with an increase in samples from 45 to 95. Or, one can maintain 95% power to detect mutations with a frequency of less than 3% by analyzing 188 cases.

CHD5 promoter methylation (108 cases)

To assess the prevalence of CHD5 promoter methylation and its association with genomic changes in primary NBs. We analyzed sequence encompassing 100 CpG dinucleotides of the CHD5 promoter from −1200 bp to − 105 bp (chr1:6,160,653-chr1:6,161,748) relative to the start site of the first exon of CHD5 (chr1:6,161,853). We analyzed the tumors in four groups: 1) normal 1p and MYCN (N=42); 2) 1p deletion only (N=20), 3) MYCN amplification only (N=21); and 4) both 1p deletion and MYCN amplification (N=25). We counted the total number of CpG methylations in the CHD5 promoter between positions -780 and -480, and we compared this score to the level of CHD5 mRNA expression. Most of the samples in all groups showed extensive CpG methylation at base position -475 and at -105, so these sites were ignored for the purposes of this analysis. We also dichotomized CHD5 expression as high or low based on the median value in the sample set. The results were analyzed using a chi-Square test. We estimated that by using a sample size of 50 NBs per group (high vs. low CHD5 expression), we could detect the difference of proportions 60% vs. 30% with 90% power and significance level of 0.05 for a two-sided test. We used a Student's T-test to determine whether or not CHD5 promoter methylation was significantly associated with low CHD5 expression.

Correlation of CHD5 promoter methylation with CHD5 expression (87 cases)

CHD5 methylation was analyzed as both a continuous and a binary variable. The binary variable was created by dichotomizing the values using the median CHD5 methylation of the patient cohort. A Wilcoxon test was also used to test association of CHD5 methylation (continuous) with CHD5 expression (binary), MYCN status (amplified, not amplified), and 1p (LOH, normal). A Wilcoxon test was also used to test association of CHD5 expression (continuous) with CHD5 methylation (binary). A Fisher's exact test was used to test the association of CHD5 methylation (binary) with MYCN status (amplified, not amplified) and 1p (LOH, normal). P-values <0.05 were considered statistically significant.

CHD5 expression (814 cases)

Univariate analyses

Fisher's exact test was used to test for association of the expression level of a given gene (CHD5, MYCN) versus each risk factor. Gene expression was dichotomized by the median RQ value into “high” and “low” expression. We also tested other cut points for “high” expression, but the 50% cut point gave the best discrimination. Risk factors included: patient age at diagnosis (<18 mo vs. ≥ 18 mo); INSS stage (1, 2, 3, 4S vs. 4) (23); MYCN amplification status (nonamplified vs. amplified) (3, 26, 27, 29); cell ploidy (DNA index: hyperdiploid vs. diploid) (3, 29-31); Shimada histopathology (favorable versus unfavorable) (32-34); 1p status (normal vs. deleted) (8, 10, 35); 11q status (normal vs. deleted) (8, 36); and risk group (low/intermediate vs. high) according to COG criteria (22). To adjust for multiple comparisons with Fisher's exact test, p-values less than 0.01 were considered statistically significant. EFS time was calculated from the time of diagnosis until the time of the first relapse, progressive disease, secondary malignancy, or death, or until last contact. OS time was calculated until the time of death, or until last contact. We used the Kaplan-Meier method to generate survival curves and determine 5-year estimates ± standard error of EFS and OS, with standard errors according to Peto. Survival curves were compared using a log rank test (p<0.05 for statistical significance).

Multivariable analyses

We used a Cox proportional hazards model of EFS to test the independent prognostic value of CHD5 expression after adjusting for each of the other currently accepted prognostic genomic markers (MYCN amplification, 1p deletion, 11q deletion). We used stepwise backwards selection to build the most parsimonious Cox model of factors prognostic for EFS, testing both clinical and genomic factors: age, stage, MYCN amplification, ploidy, Shimada histopathology, 1p deletion, 11q deletion.

Results

CHD5 mutation analysis

We obtained CHD5 sequence information on 188 primary tumor samples from the TARGET initiative, which detected a total of 19 CHD5 sequence variations in 17 NB samples (Table 1). Re-sequencing of the region in both directions did not confirm the suspected sequence variations in nine of the cases, so these were attributed to technical artifacts. We did confirm a suspected missense mutation in six cases, but all six had the same mutation in the germline DNA, so these were considered likely to be polymorphisms that had not been described previously in dbSNP (http://www.ncbi.nlm.nih.gov/projects/SNP/). The remaining four sequence variations were predicted to alter splicing (Table 1). No constitutional DNA was available for two splice site variations, but these did not involve highly conserved sequences, indicating that these are likely non-pathogenic changes. For the remaining two cases, the splice site sequence variations were found in the germline DNA. Interestingly, case COG-1266 had two sequence variations (one missense variant and one splice site variant) and only the abnormal alleles were found in the tumor. This was most likely the result of a hemizygous deletion involving the CHD5 gene at 1p36.31, with loss of the normal allele. The missense variant was not in a conserved domain, but the splice site variant is predicted to alter splicing and alternative CAG usage (37). Nevertheless, no examples of nonsense mutation, insertion/deletion or frameshift mutation were identified in any of the 188 cases, so the overall prevalence of “biologically significant” sequence variation is less than 1%, consistent with our previous findings in 30 NB cell lines (18).

Table 1. Mutations Identified in 188 Neuroblastomas from TARGET Set.

Sample Name	WT Allele	Mutated Allele	AA Change	In Tumor	In Blood
CHOP-1337	C	T	T157M	yes (C+T)	yes (C+T)
CHOP-1359	G	C	E194D	yes (G+C)	yes (G+C)
CHOP-1622	G	C	G563R	yes (G+C)	yes (G+C)
CHOP-0283	G	T	G643C	yes (G+T)	yes (G+T)
CHOP-1446	G	T	K867N	yes (G+T)	yes (G+T)
CHOP-1266	G	A	R1621Q	yes (A+0)	yes (G+A)
CHOP-1266	CAG/EXON33	TAG/EXON33	E33_splice	yes (T+0)	yes (C+T)
CHOP-2601	CAG/EXON13	TAG/EXON13	E13_splice	yes (T+0)	NA
CHOP-1111	CAG/EXON33	TAG/EXON33	E33_splice	yes (C+T)	NA
CHOP-1199	CAG/EXON33	TAG/EXON33	E33_splice	yes (C+T)	yes (C+T)
CHOP-1823	T	C	H1608R	no mutation	–––
CHOP-1613	C	A	Q1572H	no mutation	–––
CHOP-1613	T	C	P1632L	no mutation	–––
CHOP-1902	T	C	E38_splice	no mutation	–––
CHOP-1403	C	A	N1813H	no mutation	–––
CHOP-1840	G	A	E1845K	no mutation	–––
CHOP-1403	G	T	S1846A	no mutation	–––
CHOP-1998	G	A	N1905D	no mutation	–––
CHOP-1488	T	T	E40_splice	no mutation	–––

NA = Not available

CHD5 promoter methylation

We previously examined the methylation status of the CHD5 promoter in four NB cell lines (18). In the current study, we analyzed 108 primary NB tumor samples for the methylation status of the CHD5 promoter region to determine if there is transcriptional silencing and correlation with expression, as seen in NB cell lines. We divided these 108 tumor samples into four groups by combinations of 1p deletion and MYCN amplification status. There were strong sites of methylation in all groups at a site at -475 base pairs and a secondary site at -105 base pairs (Figure 1). Interestingly, Groups 3 and 4 (MYCN amplification positive group) had more strongly methylated sites between base pairs -780 to -480, whereas Groups 1 and 2 (MYCN amplification negative group) had no strong peaks around this site. However, there was no significant difference between the 1p deleted groups (Group 2 and Group 4) and 1p intact groups (Group 1 and Group 3). Indeed, we saw the strongest methylation in cases with both 1p loss and MYCN amplification. Taken together, these data suggest a possible association between MYCN amplification and the methylation status of the distal CHD5 promoter region.

Figure 1. Methylation of the CHD5 promoter region in neuroblastoma primary tumor samples.

A) CHD5 promoter methylation in Group 1, which have hemizygous 1p deletions and MYCN amplification. B) CHD5 promoter methylation in Group 2, which have hemizygous 1p deletion and no MYCN amplification. C) CHD5 promoter methylation in Group 3, which lack 1p deletion and MYCN amplification. D) CHD5 promoter methylation in Group 4, which lack 1p deletion and no MYCN amplification. We counted the number of methylated CpGs in the target region between -780 and -480 in a given case, and this was used as a methylation score for each case. Then the percent methylation for each group at each site was calculated

We also wanted to determine if CHD5 expression level was correlated with CHD5 promoter methylation status. Although CHD5 expression was generally lower in cases with promoter methylation, no statistically significant association was found between CHD5 promoter methylation and either CHD5 expression (p=0.16) or 1p deletion status (p=0.8118). However, there was a trend for an association of higher CHD5 promoter methylation with MYCN amplification (p=0.0735).

CHD5 mRNA expression

We analyzed the association of CHD5 expression by qRT-PCR in 814 cases using TaqMan LDA microfluidic cards, and we compared expression (high versus low) with clinical and biological variables as well as outcome. High expression (above the median, based on RQ value) was highly associated with younger age (<18 mo) at diagnosis, favorable disease stage (INSS stages 1-3 or 4S), the absence of MYCN amplification, hyperdiploidy (DNA index >1.0), and favorable Shimada histopathology (p<0.0001 for all) (Table 2). However, it was not significantly associated with 11q deletion status. High CHD5 expression was also strongly associated with favorable EFS and OS (p<0.0001 for both) (Table 3, Figure 2). In multivariable analysis, low CHD5 expression remained significantly predictive of poor EFS after adjusting for MYCN amplification, 1p deletion, and/or 11q deletion (Table 4).

Table 2. P-values from the Wilcoxon Tests of Association for Gene Expression (RQ value) with Risk Factors.

Gene LDA	Age at Dx. (n=814)	INSS Stage (n=814)	MYCN Status (n=808)	Ploidy Status (n=809)	Shimada Histopath. (n=779)	1p Status (n=443)	11q Status (n=417)	Risk Group (n=814)
CHD5	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	0.6339	<0.0001
MYCN	0.0141	<0.0001	<0.0001	0.0579	<0.0001	<0.0001	0.1839	<0.0001
MYC	0.0181	0.3405	<0.0001	0.4727	0.8597	0.361	0.2222	0.6053

Table 3. Log rank p-values and 5-year EFS and OS for Gene Expression (RQ value).

Gene RQ Dichotomized by the Median	N	5-year EFS ± SE (%)	EFS p-value	5-year OS ±SE (%)	OS p-value
Overall cohort	814	69 ± 2	N/A	76 ± 2	N/A
CHD5
< median	407	57 ± 3	<0.0001	64 ± 3	<0.0001
≥ median	407	81 ± 3		88 ± 3
MYCN
< median	407	74 ± 3	0.0024	79 ± 3	0.0054
≥ median	407	64 ± 3		72 ± 3
MYC
< median	407	68 ± 3	0.8904	75 ± 3	0.637
≥ median	407	69 ± 3		77 ± 3

Figure 2. Association of CHD5 or MYCN expression with EFS and OS (N=814 for both).

CHD5 and MYCN expression were dichotomized by the median RQ value, and expression level was correlated with EFS and with OS. High CHD5 expression was strongly and significantly associated with favorable EFS (A; p<0.0001) and OS (B; p<0.0001). High MYCN expression was significantly associated with unfavorable EFS (C; p=0.0024) and OS (D; p=0.0054).

Table 4. Multivariable analysis of EFS.

Significant factors	N	p-value	Hazard ratio (95% confidence interval)
Model A*
CHD5 low expression	808	<0.0001	1.9 (1.4, 2.5)
MYCN amplification		<0.0001	2.8 (2.1, 3.7)
Model B*
CHD5 low expression	443	0.0003	2.1 (1.4, 3.1)
1p LOH		0.0005	2.0 (1.4, 3.0)
Model C*
CHD5 low expression	417	<0.0001	2.4 (1.6, 3.6)
11q LOH		0.0095	1.7 (1.1, 2.5)
Model D*
CHD5 low expression		0.007	1.8 (1.2, 2.7)
MYCN amplification	441	0.0004	2.4 (1.5, 3.9)
1p LOH		0.14	NA
Model E*
CHD5 low expression		0.0056	1.8 (1.2, 2.9)
MYCN amplification	415	<0.0001	3.1 (1.9, 4.9)
11q LOH		0.0008	2.0 (1.3, 3.0)
Model F*
CHD5 low expression		0.0057	1.9 (1.2, 2.9)
MYCN amplification	379	0.0002	2.7 (1.6, 4.6)
1p LOH		0.2211	NA
11q LOH		0.0037	1.9 (1.2, 2.9)
Model G – most parsimonious**
Stage 4	808	<0.0001	4.3 (3.2, 5.8)
MYCN amplification		<0.0001	2.0 (1.5, 2.7)

NA – not applicable due to non-significance of factor

^*Models A-F compare the prognostic ability of CHD5 expression to currently accepted prognostic genomic factors.

^**tested in the model and found not significant were age, ploidy, Shimada histopathology, 1p, and 11q.

MYCN mRNA expression

Because of the established association of CHD5 expression with MYCN amplification, we also analyzed the association of MYCN expression with clinical and biological variables in this same cohort of 814 NBs. High MYCN expression was significantly associated with advanced tumor stage, unfavorable Shimada histopathology, 1p deletion and unfavorable risk group (p<0.0001 for all). However, expression was not significantly associated with patient age, ploidy or 11q status in this cohort (Table 2). High MYCN expression was associated with worse EFS and OS (p=0.0024 and p=0.0054, respectively) (Table 3, Figure 3).

Discussion

Allelic loss of 1p36 is found in about 35% of all primary NBs (70-80% of high-risk tumors and cell lines), and it is one of the most characteristic genetic changes in this tumor type. Indeed, 1p deletion is strongly associated with adverse clinical and biological features, and it is associated with a poor outcome independent of MYCN amplification, at least in subsets of NBs (8, 11, 35, 38). Nevertheless, 1p deletion is strongly associated with MYCN amplification (3, 9, 11), so the two genetic events may be related. We have identified CHD5 as a bona fide TSG deleted from 1p36.31 in human NBs (12, 18, 19). Here we addressed the mechanisms of CHD5 inactivation in a large series of primary NBs.

Somatically acquired sequence variations of CHD5 were rare, and we found no examples of homozygous genomic inactivation of CHD5 in any of the 188 high-risk NBs examined. We did confirm 10 sequence variations in 9 of the 188 cases studied (Table 1), but most were also present in the germline DNA, so we presume these represent rare, previously unreported polymorphisms. Two CHD5 sequence variations present in one tumor (COG-1266) were also present in the germline DNA of this patient. The “normal” allele was missing from this tumor in both cases, presumably reflecting loss of heterozygosity, so both sequence variations were presumably present in the remaining CHD5 allele. Nevertheless, homozygous genomic inactivation of CHD5 is extremely rare (<1%) in primary NBs (data presented here) and tumor-derived cell lines (18), suggesting that tumor cells may not tolerate a complete absence of CHD5 expression. Thus, functional silencing of the remaining CHD5 allele in NBs may occur by epigenetic mechanisms, such as methylation.

We studied the methylation status of the CHD5 promoter in 108 primary NBs, separated into four groups, based on the presence (or absence) of 1p deletion and/or MYCN amplification (Figure 1). There was a unique cluster of methylated CpGs between base pairs -780 to -480 in both groups with MYCN amplification, and to a lesser extent in the group with 1p deletion only, consistent with our findings in four NB cell lines (18). Because the four NB lines either had both 1p deletion and MYCN amplification or neither, we could not distinguish whether methylation was associated with one genomic lesion or the other. However, this analysis of primary tumors suggests a stronger association with MYCN amplification. Interestingly, a recent study by Murphy and colleagues suggested preferential E-box utilization by MYCN, as well as an association of MYCN protein with hypemethylated DNA (39). Thus, MYCN may be playing a role in regulating CHD5 expression, especially in tumors with 1p deletion (3, 9, 11, 13). Even low expression of CHD5 may prevent tumor cells from taking full proliferative advantage of MYCN amplification, perhaps by regulating MYCN expression or facilitating the apoptosis normally associated with MYC family gene overexpression (40).

We did not find a statistically significant correlation between CHD5 promoter methylation and low CHD5 expression. However, there may be particular CpGs or a region remote from the one we analyzed that are more important for regulation of CHD5 expression. Also, CHD5 expression may be downregulated by other epigenetic mechanisms such as histone modification, or the association of transcription factors and/or chromatin remodeling complexes that modulate expression positively or negatively. However, our current results and our previous findings in NB cell lines (18) suggest that CHD5 expression is regulated, at least in part, by promoter methylation. Indeed, there is evidence for epigenomic alterations contributing to NB pathogenesis (41). Furthermore, there have been several reports of other tumor types of CHD5 silencing by promoter methylation (42-46), so there is clear precedent for CHD5 transcriptional regulation by this mechanism.

High CHD5 expression was strongly associated with favorable clinical and biological variables as well as outcome in this representative cohort of 814 NBs (Tables 2-3. Figure 2). Indeed, CHD5 expression remained prognostic after adjusting for MYCN status, 1p deletion status, and 11q deletion status (Table 4). This suggests that CHD5 may be playing a role in the pathogenesis of NBs, or in maintenance of the malignant state. Given its possible function as part of a chromatin-remodeling complex, CHD5 may be contributing to the regulation of genes involved in neuronal growth and/or differentiation. Loss of this gene and its encoded protein could contribute to loss of growth control or failure of immature neuroblasts to differentiate.

Previously, we had analyzed CHD5 expression in a panel of 101 primary NBs that had undergone microarray expression profiling (18), and we showed a significant association between high CHD5 expression and favorable clinical/biological features as well as outcome. However, microarray expression profiling is semiquantitative at best, and not all subsets of patients were adequately represented. A recent study by Garcia and colleagues examined the expression of CHD5 protein by immunohistochemistry in 90 primary neuroblastic tumors (63 NBs, 14 ganglioneuroblastomas, 13 ganglioneuromas), and they found that high CHD5 expression was associated with lower stage, more differentiated tumors and better outcome. The study described here analyzed a large number of representative NBs using a quantitative, controlled and reproducible technique. Indeed, we showed that CHD5 mRNA expression provides highly significant prognostic information, and it is one of the most powerful biological prognostic markers for this disease. Nevertheless, patient age and INSS stage remain the most potent predictors of overall outcome for NB patients (Table 4).

CHD5 was independently identified as a TSG on the orthologous region of mouse chromosome 4 using a chromosome engineering approach (47). This study, combined with our functional data on the inhibition of NB clonogenicity and tumorigenicity by CHD5, provides compelling support for its role as a TSG. Furthermore, CHD5 has been suggested as a TSG in other tumor systems, including colorectal (48), gastric (45), and ovarian cancers (42, 49), lung cancer (46), laryngeal squamous cell carcinomas (44), and cutaneous melanoma (50), based on the association of high expression with favorable outcome. Interestingly, one copy of CHD5 is frequently deleted in these tumors, and promoter methylation is commonly found in tumors with low expression, similar to our findings in NBs.

Our data support a role for CHD5 in the pathogenesis of NBs. It appears that loss of CHD5 (and possibly other genes) via 1p36 deletion reduces its expression sufficiently to favor continued growth and tumor evolution. However, unlike canonical TSGs such as RB1, the remaining allele is rarely if ever inactivated by somatic mutation, deletion or rearrangement. Rather, CHD5 expression may be regulated by epigenetic modifications, such as promoter methylation. This epigenetic mechanism could inactivate expression, or lower it sufficiently to functionally silence the gene. Indeed, there is precedent for transcriptional silencing by methylation to inactivate the second allele of other TSGs, such as RASSF1A (51, 52) and OPCML (53). Also, we showed that CHD5 could be re-expressed by growing these cells in 5-deoxyazacytidine (18), so pharmacological interventions to re-express the remaining normal allele may be a useful approach to treat NBs with low or absent CHD5 expression.

Further characterization of the structure, expression and function of CHD5 and its encoded protein should provide substantial insights into mechanisms of malignant transformation or progression of NBs. We have shown that high CHD5 expression is a strong and independent predictor of favorable outcome in NB patients. Thus, assessment of CHD5 expression may allow more precise molecular profiling of NBs, which may in turn permit more appropriate risk stratification and treatment selection. Furthermore, clarification of the biological function of the encoded protein and the pathways it affects may suggest novel therapeutic strategies for NBs, as well as other tumor types associated with 1p deletion and decreased expression of CHD5.

Translational Relevance.

Our data strongly suggest that CHD5 plays an important role in the pathogenesis of neuroblastomas. Indeed, assessment of CHD5 expression provides important prognostic information that should permit more accurate risk assessment and therapy selection for neuroblastoma patients. On the other hand, testing for CHD5 inactivating mutations does not seem worthwhile. Because the remaining CHD5 allele in tumors with 1p deletions is seldom mutated, efforts aimed at the targeted re-expression of the remaining CHD5 allele in tumors (such as with demethylating agents) may have therapeutic benefit. Our data also suggest a possible interaction between CHD5 and MYCN, another important contributor to neuroblastoma behavior. Moreover, clarification of the biological function of the CHD5-encoded protein and the pathways it effects may suggest novel therapeutic strategies for neuroblastomas, as well as other tumor types associated with 1p deletion and decreased expression of CHD5.

Abbreviations

COG

Children's Oncology Group

EFS

event-free survival

INSS

International Neuroblastoma Staging System

LDA

low-density array

OS

overall survival

NB

neuroblastoma

qRT-PCR

quantitative reverse transcriptase polymerase chain reaction

SRD

smallest region of consistent deletion

TARGET

Therapeutically Applicable Research to Generate Effective Targets

TSG

tumor suppressor gene