Introduction

Aberrant DNA methylation is inherited over cell divisions and is deeply involved in carcinogenesis. Increasing numbers of cancer-specific aberrant methylation have been identified, and their usefulness in cancer diagnostics is becoming clear. DNA-demethylating drugs have been developed, and their usefulness in cancer therapeutics is also becoming clear. In cancer diagnostics, methylation of the Septin-9 gene in plasma is now being clinically evaluated as a detection marker for colorectal cancers.1 Methylation of O6-methylguanine-DNA methyltransferase in glioblastoma tissues is also used to predict the response of tumors to alkylating drugs.2 Methylation of glutathione S-transferase pi1 specific to prostate cancer cells is used to detect the presence of prostate cancer in prostatectomy or biopsy tissue.3 In addition to methylation of specific genes, the CpG island methylator phenotype (CIMP), methylation of multiple genes, has been reported to be associated with prognosis in various cancers.4

The therapeutic application of DNA methylation has already been brought into practice. Two demethylating drugs, 5-aza-2′-deoxycytidine (5-aza-CdR) and 5-azacytidine (5-aza-CR), have been approved by the US Food and Drug Administration for treatment of myelodysplastic syndrome. Application of these demethylating drugs to treatment of solid tumors is being tested in clinical trials.5 In addition to these clinically used drugs, a new generation of DNA-demethylating drugs, such as SGI-110 and CP-4200, are also being developed.6 Combinations of DNA-demethylating drugs with other drugs, such as histone deacetylase inhibitors,7 cellular differentiating drugs8 and cytotoxic drugs,9 have produced some promising results.

In this review, we will introduce CIMP in various cancers and then focus on the high usefulness of CIMP in prognostic diagnosis of neuroblastomas (NBLs) and the potential in their treatment.

CIMP in various cancers other than NBLs

CIMP was originally established in colorectal cancers by the pioneering work of Toyota et al10. Some colorectal cancers had frequent DNA methylation of specific CpG islands and were associated with microsatellite instability. Colorectal cancers are now classified into three groups, CIMP-high (CIMP1, HME), CIMP-low (CIMP2, IME) and CIMP-0 (CIMP-negative, LME), using tumor-specific methylation markers.11, 12, 13 CIMP-high cases show low mortality compared with CIMP-0 (hazard ratio (HR)=0.44; 95% confidence interval (CI)=0.22–0.88; n=649)11 (Table 1). CIMP-high, -low and -0 are strongly associated with mutations of BRAF, KRAS and TP53, respectively.11, 12 Although the cause–consequence relationship between oncogene mutations and CIMP has not been established, some investigators propose that methylation silencing of senescence pathways is necessary to suppress oncogene-induced senescence and for a cell with an oncogene mutation to survive.14, 15

Table 1 CIMP in various cancers and association with prognosis
Full size table

The presence of CIMP has been reported in several other cancers, such as gastric,16, 17 lung,18 liver,19 ovarian cancers20 and leukemias21 (Table 1). The definition of CIMP is different in each cancer and the relationship between CIMP and prognosis is also different. Poor survival is associated with CIMP in bladder cancers (P=0.01, n=98),22 esophageal adenocarcinomas (HR=2.7; 95% CI=1.1–6.5; P=0.02; n=41),23 ovarian tumors (P<0.001, n=19)24 and leukemias (P=0.04, n=54).21 On the other hand, better survival is associated with CIMP in gastric cancers (CIMP-high vs -negative, P=0.004, CIMP-low vs -negative, P=0.012, n=78).25

CIMP in glioblastomas is exceptionally well-characterized from a molecular viewpoint, and designated as G-CIMP. G-CIMP is associated with good prognosis (P=0.017, n=253),26 and is known to be caused by isocitrate dehydrogenase 1 (IDH1) mutation.27 Functionally, IDH1 mutation has been shown as a gain-of-function mutation and to catalyze the NADPH-dependent reduction of α-ketoglutarate to 2-hydroxyglutarate.28 2-Hydroxyglutarate inhibits TET methylcytosine dioxygenase 2 that catalyzes the conversion of 5-methylcytosine to 5-hydroxymethylcytosine,29 and the inhibition of TET is considered to lead to methylation of multiple genes. Similar to the association of IDH1 mutation and G-CIMP, IDH1/2 mutation is also associated with methylation patterns in AML.30

Identification of CIMP in NBLs as a strong prognostic marker

NBL, derived from primitive cells of the sympathetic nervous system, is the most frequent extracranial cancer in childhood,31 and is characterized by two extreme outcomes, spontaneous regression and rapid progression. Accurate risk prediction is important for NBL in order to implement a necessary and sufficient level of treatment, and the International Neuroblastoma Risk Group classification system has been established for this purpose.32 In this system, seven prognostic factors (stage, age, histologic category, grade of tumor differentiation, DNA ploidy and copy-number status at MYCN and at chromosome 11q) are used as the most clinically relevant factors. However, especially in the NBLs without MYCN amplification, it is still difficult to predict accurate prognosis and to decide on the therapeutic strategy.

Considering the major involvement of epigenetic machinery in embryonic development,33, 34 we searched for CGIs specifically methylated in NBLs with poor prognosis, not in those with good prognosis, using methylation-sensitive representational difference analysis.35 Five CGI (or CGI groups), namely the PCDHB family, the PCDHA family, HLP, DKFZp451I127 and CYP26C1, were found to be specifically methylated in NBLs with poor prognosis. Methylation of these five CGI (groups) was dependent upon each other, and conformed to the concept of CIMP. Methylation levels of the PCDHB family showed a clear bimodal distribution (Figure 1a), and NBL cases with high methylation levels of the PCDHB family showed poor overall survival with a HR of 22 (95% CI=5.3–93) in 140 Japanese NBL cases.35 Therefore, we decided to use the methylation levels of the PCDHB family as a marker of CIMP in NBLs.

Figure 1
figure 1

Bimodal distribution of methylation levels of the PCDHB family and its association with survival. (a) In 140 Japanese neuroblastoma (NBL) cases, the methylation level of the PCDHB family showed bimodal distribution (modified from Abe et al.35). The cutoff value for PCDHB family was set at 40%. NBLs with CpG island methylator phenotype (CIMP) had significantly and markedly worse overall survival, analyzed by the Kaplan–Meier method. (b) In 152 German NBL cases, a similar bimodal distribution of the PCDHB family methylation level was observed. As cutoff values between 40 and 60% gave high HRs in Japanese NBL cases, before the analysis, cutoff values of 40 and 60% were set for cases without and with CIMP, respectively, and cases with intermediate values were classified as intermediate. It was reproduced that NBLs with CIMP had significantly and markedly worse overall survival (modified from Abe et al.36).

Full size image

To avoid false ‘too good’ results likely to be obtained by a genome-wide screening, we analyzed the prognostic power of CIMP in 152 German NBL cases in collaboration with Dr Schwab and Dr Westermann36 (Figure 1b). DNA of German NBL cases was sent to our laboratory in Tokyo without clinical information, and methylation levels of the PCDHB family were analyzed. Regarding the cutoff value, as values between 40 and 60% gave high HRs in Japanese NBL cases, cases with methylation levels lower than 40% and higher than 60% were classified as NBLs without and with CIMP, respectively. The strong association between CIMP and poor overall survival was faithfully reproduced in German NBL cases with a HR of 9.5 (95% CI=3.2–28). In addition, German NBL cases had information on disease-free survival, which was not available in Japanese NBL cases, and CIMP was shown to have prognostic significance also for disease-free survival with a HR of 5.4 (95% CI=2.9–10). In addition to our studies, the strong prognostic power of CIMP was further confirmed in Italian and Swedish NBL cases.37, 38 Currently, in order to evaluate the clinical utility of CIMP as a prognostic marker, a prospective study is being conducted.

Comparison of CIMP with MYCN amplification

MYCN amplification is known as the strongest prognostic marker for NBLs and is one of the first molecular markers used in practice.39, 40, 41 The presence of MYCN amplification is therefore used as a biomarker for stratification of NBLs in practice and trials. Importantly, almost all NBLs with MYCN amplification displayed CIMP (37/38 in Japanese and 23/23 in German NBL cases) while some NBLs without MYCN amplification also displayed CIMP35, 36 (Figure 2a). The cases with MYCN amplification showed poor overall survival with HR of 9.5 (95% CI=4.4–21) and 12 (95% CI=4.9–29) in Japanese and German NBL cases, respectively.35, 36 Therefore, NBL cases were classified into three groups: (J1) NBL cases without CIMP nor MYCN amplification (N=72), (J2) NBL cases with CIMP without MYCN amplification (N=30) and (J3) NBL cases with both CIMP and MYCN amplification (N=37) in Japanese NBL cases. German NBL cases were also classified into three groups (G1–G3) in the same manner (Figure 2b). The three groups, respectively, showed step-wise increases of risk, and notably, among NBLs without MYCN amplification (J1, J2 in Japanese and G1, G2 in German NBL cases), NBLs with CIMP showed worse prognosis with a HR of 12 (95% CI=2.6–59) in Japanese and 4.5 (95% CI=1.3–16) in German NBL cases. The almost complete inclusion of NBLs with MYCN amplification in those with CIMP indicates that these two abnormalities are very closely associated with each other. However, it is still unknown whether CIMP causes MYCN amplification, MYCN amplification causes CIMP, these two must coexist for development of NBLs, or if there is a shared upstream event.

Figure 2
figure 2

Comparison of CpG island methylator phenotype (CIMP) and MYCN amplification. (a) The relationship between neuroblastomas (NBLs) with CIMP and those with MYCN amplification are shown by Venn diagrams. Almost all NBLs with MYCN amplification (37/38 in Japanese and 23/23 in German NBL cases) displayed CIMP, and some additional cases had CIMP only. (b) Kaplan–Meier analysis of (J1, G1) NBL cases without CIMP or MYCN amplification (N=72 in Japanese and N=95 in German NBL cases), (J2, G2) NBL cases with CIMP without MYCN amplification (N=30 in Japanese and N=27 in German NBL cases), (J3, G3) NBL cases with both CIMP and MYCN amplification (N=37 in Japanese and N=23 in German NBL cases). Among NBLs without MYCN amplification (J1, J2 in Japanese and G1, G2 in German NBL cases), CIMP also had a significant and strong prognostic marker with a hazard ratio of 12 (95% confidence interval (CI)=2.6–59; P=0.002) in Japanese and 4.5 (95% CI=1.3–16; P=0.02) in German NBL cases.

Full size image

Mechanism for the association between CIMP and poor prognosis

CGIs of the PCDHB family are located in their gene body, and their methylation was not associated with gene expression levels.35 This suggested that, although methylation of the PCDHB family was closely associated with poor survival, simultaneous methylation of promoter CGIs was mechanistically involved in the poor survival (Figure 3). Indeed, in our study of Japanese NBL cases, methylation of promoter CGIs of the RASSF1A, BLU, CYP26C1, FERD3L, CRYBA2 and PCDHGC4 were methylated at significantly higher incidences in NBLs with CIMP, indicating that methylation silencing of tumor-suppressor genes was indeed associated with CIMP.8, 35

Figure 3
figure 3

A likely mechanism for the association between CpG island methylator phenotype (CIMP) and poor prognosis. In neuroblastomas (NBLs) with and without CIMP, the exonic CGIs of the PCDHB family and promoter CGIs of tumor-suppressor genes typically showed methylation statuses as in this scheme. NBLs with CIMP had methylation of the exonic CGIs of the PCDHB family consistently, and that of promoter CGIs of multiple tumor-suppressor genes with lower frequencies. Although methylation silencing of tumor-suppressor genes was considered to be responsible for the poor prognosis of NBLs with CIMP, methylation of individual genes had less sensitivity of CIMP than methylation of the PCDHB family. On the other hand, NBLs without CIMP did not have methylation of the exonic CGIs of the PCDHB family or that of promoter CGIs of tumor-suppressor genes, and thus were considered to have a good prognosis.

Full size image

To strengthen this hypothesis, we further analyzed methylation of promoter CGIs of genes whose silencing was reported to be involved in development or progression of NBLs, namely CASP8, EMP3, HOXA9, NR1I2 and CD44. CASP8 is an anti-apoptotic gene, and its methylation was reported to be associated with poor survival with HR of 5.3 (P=0.008).42 Also, methylation of EMP3, HOXA9, NR1I2 and CD44 were associated with poor survival with a P-value of 0.014, 0.03, 0.04 and 0.049, respectively.42, 43, 44, 45 We found that CIMP was associated with methylation of multiple promoter CGIs, mainly CASP8 and NR1I2, but had stronger prognostic power than methylation of individual genes.46 These results strengthened the hypothesis that CIMP leads to poor prognosis by induction of methylation of promoter CGIs of various tumor-suppressor genes at low incidences.

CIMP as a potential target of epigenetic therapy

Poor prognosis of NBLs with CIMP is likely to be caused by silencing of multiple genes due to methylation of their promoter CGIs. As silenced genes are multiple and variable among individual NBLs, it was hypothesized that simultaneous demethylation of multiple genes could be effective for treatment of NBLs (Figure 4). Indeed, treatment of NBL cell lines with a demethylating drug, 5-aza-CdR, enhanced the sensitivity to the differentiation effect by 13-cis-retinoic acid.8

Figure 4
figure 4

The concept of demethylating drug treatment for neuroblastomas (NBLs). In individual NBLs (tumor A–C), multiple but different tumor-suppressor genes are methylation-silenced, and reversal of these individual genes is difficult. By the use of a demethylating drug, multiple tumor-suppressor genes become simultaneously demethylated, and NBL cells are expected to show better responses to differentiation and cytotoxic agents.

Full size image

In addition to CIMP, aberrant histone modifications or modifying enzymes are also emerging as potential therapeutic targets of NBLs. For example, lysine-specific demethylase 1 (LSD1), a histone H3 lysine4 (H3K4) demethylase, is highly expressed in poorly differentiated NBLs, and inhibition of LSD1 using a monoamine oxidase inhibitor, tranylcypromine, resulted in growth suppression of NBLs in vitro and in vivo.47 To improve selectivity for LSD1 over monoamine oxidase inhibitor, LSD1-selective inhibitors were developed,48 and they are expected to show high anticancer efficacy and low toxicity in normal cells.

Clinically, a phase I study of 5-aza-CdR with doxorubicin and cyclophosphamide in children with NBLs and other solid tumors was conducted in the United States,49 and low-dose 5-aza-CdR (5 mg m−2) turned out to have tolerable toxicity in children. However, doses of 5-aza-CdR capable of producing clinically relevant biologic effects were not well tolerated. Different combinations between 5-aza-CdR and other drugs, such as differentiating drugs (13-cis-retinoic acid) or other epigenetic drugs (LSD1 inhibitors and histone deacetylase inhibitors), could improve the effectiveness of the demethylating drug for NBLs.

Conclusions

The usefulness of aberrant DNA methylation in cancer diagnostics and therapeutics is now coming into practice. In NBLs, CIMP has prognostic power in cases without MYCN amplification, and its strong prognostic power was validated in German, Italian and Swedish NBL cases. Combinations of a DNA-demethylating drug with a differentiating drug has been shown to be effective for NBLs with CIMP in vitro, and the appropriate dose and appropriate combination is expected to improve survival of NBL cases. Epigenetic diagnosis and therapeutics are becoming or have become an important choice for cancer patients.