Endocrine Research
Identification of a CpG Island Methylator Phenotype in Adrenocortical Carcinomas
Olivia Barreau,* Guillaume Assié,* Hortense Wilmot-Roussel, Bruno Ragazzon, Camille Baudry, Karine Perlemoine, Fernande René-Corail, Xavier Bertagna, Bertrand Dousset, Nadim Hamzaoui, Frédérique Tissier, Aurélien de Reynies, and Jérôme Bertherat
Institut National de la Santé et de la Recherche Médicale Unité 1016 (O.B., G.A., H.W .- R., B.R., C.B., X.B., B.D., F.T., J.B.), Institut Cochin; Centre National de la Recherche Scientifique Unité Mixte de Recherche 8104 (O.B., G.A., H.W .- R., B.R., C.B., X.B., B.D., F.T., J.B.); and Departments of Endocrinology (G.A., C.B., X.B., J.B.) and Pathology (F.T.), Center for Rare Adrenal Diseases (X.B., J.B.), Unit of Digestive and Endocrine Surgery (B.D.), and Oncogenetic Unit (N.H.), Assistance Publique Hôpitaux de Paris, Hôpital Cochin, 75014 Paris, France; Programme Cartes d’identité des Tumeurs (O.B., A.d.R.), Ligue Nationale Contre Le Cancer, 75013 Paris, France; Université Paris Descartes (O.B., G.A., H.W .- R., C.B., K.P., F.R .- C., X.B., B.D., F.T., J.B.), Sorbonne Paris Cité, 75006 Paris, France; and Rare Adrenal Cancer Network, Institut National du Cancer Cortico-Medulo Tumeurs Endocrines (X.B.), 75014 Paris, France
Purpose: DNA methylation is a mechanism for gene expression silencing in cancer. Limited infor- mation is available for adrenocortical tumors. Abnormal methylation at the IGF2/H19 locus is common in adrenocortical carcinomas. Our aim was to characterize the methylation in adreno- cortical carcinomas at a whole-genome scale and to assess its clinical significance and its impact on gene expression.
Experimental Design: Methylation patterns of CpG islands in promoter regions of 51 adrenocor- tical carcinomas and 84 adenomas were studied by the Infinium HumanMethylation27 Beadchip (Illumina, San Diego, CA). Methylation of 33 genes was studied by methylation-specific multiplex ligation-dependent probe amplification (MRC-Holland, Amsterdam, The Netherlands) in 15 car- cinomas. Gene expression data were available for 87 tumors from a previous study (HG-U133Plus2.0 AffymetrixGeneChip; Affymetrix, Santa Clara, CA). Clinical information, including patient features and survival, were available for all tumors.
Results: Methylation was higher in carcinomas than in adenomas (t test P = 3.1 x 10-9). Unsu- pervised clustering of DNA methylation profiles identified two groups of carcinomas, one with an elevated methylation level, evoking a CpG island methylator phenotype (CIMP). The subgroup of hypermethylated carcinomas was further divided in two subgroups, with different levels of meth- ylation (CIMP-high and CIMP-low). This classification could be confirmed by methylation-specific multiplex ligation-dependent probe amplification. Hypermethylation was associated with a poor survival (Cox model P = 0.02). The transcriptome/methylation correlation showed 1741 genes (of 12,250) negatively correlated; among the top genes were H19 and other tumor suppressors (PLAGL-1, G0S2, and NDRG2).
Conclusions: This genome-wide methylation analysis reveals the existence of hypermethylated adrenocortical carcinomas, with a poorer prognosis. Hypermethylation in these tumors is impor- tant for silencing specific tumor suppressor genes. (J Clin Endocrinol Metab 98: 0000-0000, 2013)
* O.B. and G.A. contributed equally to this study. Abbreviations: CIMP, CpG island methylator phenotype; COMETE, Cortico-Medulo Tumeurs Endocrines; MS-MLPA, methylation-specific multiplex ligation-dependent probe amplification.
A drenocortical tumors are frequent, most often now- adays discovered as adrenal incidentalomas. A ma- jority of these tumors are adenomas and can be responsible for steroid excess. By contrast, adrenocortical carcinoma is a rare disease with an estimated annual incidence of two per million (1). The overall prognosis is poor but hetero- geneous, with a 5-yr survival not exceeding 40% in most series (1).
Using candidate genes approaches, several molecular pathways have been involved in adrenocortical malig- nancy so far, including IGF-2 overexpression, tumor pro- tein p53 (TP53) mutations, and Wnt/B-catenin signaling alterations (2).
More recently, genomic approaches were applied to these carcinomas. Gene expression profiles by transcrip- tome analysis showed major differences between adeno- mas and carcinomas (3-5). Within the carcinomas, unsu- pervised transcriptome classification identified two groups, associated with different outcomes (3, 4). Within the group of poor prognosis, the carcinomas could be fur- ther classified in three different subgroups, two of them harboring a cardinal molecular alteration (6), one associ- ated with p53 inactivation, the second with ß-catenin ac- tivation; no specific molecular defect has been identified in the third subgroup so far.
In addition to gene expression, genomics now covers a large spectrum of alterations, including chromosomal al- terations, DNA sequence modifications, and epigenetic alterations. The latter have been poorly investigated so far. DNA methylation is an epigenetic modification, e.g. a change that is not accompanied by any change in the DNA sequence, and usually occurs in the context of CpG di- nucleotides. Two types of changes in DNA methylation patterns can be observed in cancer: a global hypomethy- lation associated with increased chromosomal instability, loss of parental imprinting, and reactivation of transpos- able elements and, alternatively, a hypermethylation of CpG islands located in the promoter regions of specific genes, which has conventionally been associated with transcriptional silencing of tumor suppressor genes (7). In some common cancer types, such as breast or colon can- cers, pangenomic methylation studies have associated DNA methylation patterns with histological tumor grade and with specific subsets of tumors with distinct patho- genic mechanisms and prognosis (8, 9). In colon cancer, methylation analyses have identified subsets of hyperm- ethylated tumors, a phenomenon called CpG island methylator phenotype (CIMP) (10). The CIMP was fur- ther categorized as CIMP-high and CIMP-low, depending on the degree of hypermethylation (11). The CIMP is thought to drive the abnormal expression of a large num-
ber of genes (12). Subsequently, a similar phenotype has been described in several other neoplasms (8).
In adrenocortical tumors, DNA methylation has been studied at the 11p15 locus, one of the loci associated with the Beckwith-Wiedmann syndrome (OMIM 130650), which predisposes to adrenocortical carcinomas. The H19 promoter methylation is associated with H19 underex- pression and IGF-2 overexpression (13). This region is submitted to parental imprinting. The maternal allele is normally not methylated and expresses H19, whereas the paternal allele is methylated and expresses IGF-2. In Beck- with-Wiedemann patients, the maternal allele is inacti- vated, resulting in a high IGF-2 expression associated with a low H19 expression. In sporadic adrenocortical carci- nomas, 11p15 is also commonly altered with a loss of the maternal allele and a duplication of the paternal allele. This alteration is thought to be responsible for the high IGF-2 expression level in adrenocortical carcinomas, one of the most common features in these tumors (14).
Recently, Rechache et al. (15) reported a pangenomic methylation study in adrenocortical tumors. The authors reported a wide spectrum of methylation alterations in cancer compared with adenomas and normal adrenal tu- mors. The profiles greatly varied depending on the type of methylation (in CpG islands or around), and on the genomic regions considered (in gene promoters or in in- tergenic regions). Another recent pangenomic methyl- ation study (16) identified hypermethylated genes com- paring benign and malignant adrenocortical tumors.
In the current study, we move forward in the fine mo- lecular characterization of adrenocortical carcinomas and identify two subgroups of adrenocortical carcinomas with hypermethylation, evoking a CIMP. We show that hyper- methylation is associated with a poor prognosis, has an impact on gene expression, and is associated with specific molecular subtypes of carcinomas previously defined by transcriptome studies.
Patients and Methods
Patients
The tumors were prospectively collected in the tumor bank of the Cortico-Medulo Tumeurs Endocrines (COMETE) network. One hundred thirty-five tumors, collected between 1993 and 2005 by the Cochin team, were included in this study. Samples were dissected by the pathologist immediately after tumor re- moval, snap frozen, and kept in liquid nitrogen. Tumor staging was performed using the ENSAT (European Network for the Study of Adrenal Tumors) classification (17). Hormonal inves- tigations, imaging, and patient follow-up were done as previ- ously reported (4).
For each patient, diagnosis, tumor weight, size, and classifi- cation were determined by pathological examination. Malig-
nancy was assessed according to Weiss criteria (18): for each tumor, a Weiss score (0-9) was determined by a single experi- enced pathologist (F.T.). Tumors with Weiss scores of 0 or 1 were considered benign. Tumors with a Weiss score of 2 were con- sidered benign in the absence of androgen secretion and of un- determined malignancy in the presence of androgen secretion (19). Tumors with Weiss scores of 3 or more were considered malignant.
Informed signed consent for the analysis of the tumor and for access to the data collected was obtained from all the patients as part of the COMETE prospective protocols, and the study was approved by the institutional review board of the Cochin Hospital.
Tumor DNA preparation
Tumor samples (10-50 mg) were powdered under liquid ni- trogen. DNA was extracted and purified by cesium chloride gra- dient ultracentrifugation or by proteinase K digestion and eth- anol extraction, followed by a clean-up step on columns (QIAGEN, Courtaboeuf, France). The assessment of nucleic ac- ids purity and integrity was performed using the measure of the absorbance with the NanoDrop ND-1000 spectrophotometer, and evaluation of the electrophoretic profile using agarose gel electrophoresis.
Methylation analysis
Data generation and normalization
Bisulfite-converted genomic DNA was analyzed using the In- finium HumanMethylation27 Beadchip (Illumina Inc., San Di- ego, CA), which interrogates 27,578 highly informative CpG sites located within the proximal promoter regions of 14,475 genes.
The methylation data were processed using the Bioconductor lumi package (20). The methylation data were color balance checked, and background corrected, and inter-samples were nor- malized using quantile normalization based on the assumption that the intensity distribution of the pooled methylated and un- methylated probes are similar for different samples. Eighty-nine CpG sites with more than 25% of samples having detection P values <0.05 were filtered out before the analysis.
Quantification of the methylation
Methylation values were expressed as the ratio of the meth- ylated alleles over the unmethylated alleles. For each CpG locus, this ratio was quantified using the M-value, calculated as the log2 ratio of the intensities of the methylated and unmethylated probes (21). The full dataset can be downloaded from the Eu- ropean Bioinformatics Institute (http://www.ebi.ac.uk/ arrayexpress/, experiment E-MTAB-815).
Methylation level of each tumor was calculated as the mean of M-values for CpGs within CpG islands of all chromosomes excepted chromosomes X and Y.
Clustering analysis
Unsupervised hierarchical clustering analysis of carcinomas was performed using the R Stats Package, with the Manhattan metric and Ward method, and selecting the 2824 CpGs in CpG islands with the highest methylation variability in carcinomas (defined as a SD >1.5).
Validation of the methylation phenotype by methy- lation-specific multiplex ligation-dependent probe amplification (MS-MLPA)
Methylation was confirmed by MS-MLPA using the SALSA MLPA ME002 tumor suppressor-2 and SALSA MLPA ME003 tumor suppressor-3 probe mixes, with the EK1-Cy5 SALSA MLPA EK1 Cy5 reagent kit (MRC Holland, Amsterdam, The Netherlands). The two probe mixes contain a total of 54 MS- MLPA probes that detect the methylation status of promoter regions of 42 different tumor suppressor genes. These two kits were chosen for their enrichment in hypermethylated genes (identified in the hypermethylated tumors with the whole-ge- nome methylation data obtained from the BeadChips). The two MLPA kits were used according to the manufacturer’s protocol, starting from 500 ng genomic DNA.
The proportion of methylated alleles was determined by the ratio of the HhaI-digested product (a methylation-sensitive re- striction enzyme) over the undigested product. For that aim, a capillary electrophoresis of the amplification products, both the digested and the undigested products, was run in a Beckman Ceq 8800 genetic analysis system (Beckman-Coulter, Villepinte, France). After electrophoresis, quantification was performed by measuring the peak heights, using the manufacturer’s software.
MS-MLPA methylation level was calculated for each tumor as the mean of the methylation values for 33 genes identified as hypermethylated in the hypermethylated tumors with the whole- genome methylation analysis.
Gene expression
HG-U133 Plus 2.0 AffymetrixGeneChip arrays (Af- fymetrix, Santa Clara, CA) gene expression data were avail- able for 87 of 135 tumors, including 53 adenomas and 34 carcinomas, as previously described (4).
Statistical analysis
Analyses were performed using R statistical software (R Stats Package) (22).
The comparison between subgroups of tumors was per- formed with Welch’s t test or Wilcoxon test depending on the number of samples for quantitative variables and with Fisher’s exact test for qualitative variables.
Differentially methylated CpGs or expressed genes between subgroups were identified using ANOVA followed by limma moderated t tests.
For each CpG site, the correlation between expression and methylation was performed with the Pearson correlation test.
Gene ontology enrichment analyses were performed using DAVID Bioinformatics Resources version 6.7 (http://david.abcc. ncifcrf.gov/tools.jsp) (23).
Survival analyses were performed in adrenocortical carcino- mas using Cox models (for quantitative variables) and log rank models (for qualitative variables) and considering the overall survival.
Multiple-testing P values were adjusted following the Benja- mini and Hochberg method.
Results
Patient characteristics
The main clinical, hormonal, and pathological charac- teristics of the 135 patients with adrenocortical tumors are summarized in Table 1.
| TABLE 1. Patient and tumor characteristics | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| Carcinomas (n = 51) | P valuec | ||||||||
| Non-CIMP (group 1, n = 26) | CIMP (group 2) | P valueb | Total | Adenomas (n = 84) | |||||
| CIMP-low (group 2A, n = 17) | CIMP-high (group 2B, n = 8) | P valueª | Total CIMP (n = 25) | ||||||
| Age (yr) | 0.93 | 0.79 | 0.41 | ||||||
| Median | 41.5 | 44 | 43 | 43 | 43 | 48 | |||
| Range | 15-81 | 25-77 | 15-73 | 15-77 | 15-81 | 22-78 | |||
| Sex | 0.39 | 0.05 | 0.02 | ||||||
| Male | 3 (11.5%) | 5 (29.4%) | 4 (50%) | 9 | 12 (23.5%) | 7 (8.3%) | |||
| Female | 23 (88.5%) | 12 (70.6%) | 4 (50%) | 16 | 39 (76.5%) | 77 (91.7%) | |||
| Hormone secretion | 0.14 | 1 | 0.01 | ||||||
| Yes | 19 (73.1%) | 15 (88.2%) | 8 (100%) | 23 | 42 (82%) | 51 (61%) | |||
| No | 7 (26.9%) | 2 (11.8%) | 0 | 2 | 9 (18%) | 33 (39%) | |||
| Tumor size (cm)d | 0.96 | 6.7e-3 | 1.3 × 10-12 | ||||||
| Median | 8 | 14 | 14.5 | 14.5 | 10 | 3.6 | |||
| Range | 4-25 | 5.5-24 | 4.5-20 | 4.5-24 | 4-25 | 2-10 | |||
| Weiss score | 0.1 | 9.5e-3 | <2.2 × 10-16 | ||||||
| Median | 4 | 6 | 7 | 7 | 6 | 0 | |||
| Range | 2-9 | 2-9 | 6-8 | 2-9 | 2-9 | 0-2 | |||
| ENSAT staging | 0.4 | 0.02 | 1.45 × 10-11 | ||||||
| 1 or 2 | 19 (73.1%) | 8 (47.1%) | 2 (25%) | 10 | 29 (56.9%) | 84 (100%) | |||
| 3 or 4 | 7 (26.9%) | 9 (52.9%) | 6 (75%) | 15 | 22 (43.1%) | 0 | |||
a CIMP-low vs. CIMP-high carcinomas.
b CIMP- vs. non-CIMP carcinomas.
” Carcinomas vs. adenomas.
d The size of two carcinomas was not available.
Carcinomas are hypermethylated on a whole- genome scale
The CpG island methylation level varied among adre- nocortical tumors. Globally, carcinomas were more meth- ylated than adenomas (P = 3.1 × 10-9). Unsupervised clustering of carcinomas separated them in two groups with different methylation levels (Fig. 1A). Group 1 car- cinomas were slightly hypermethylated compared with adenomas (P = 0.03). Group 2 carcinomas were hyper- methylated compared with both the adenomas and the carcinomas from group 1 (Fig. 1B, P = 5.7 × 10-14 and 5.9 × 10-14, respectively) and evoked a CIMP. Group 2 itself was further divided in two subgroups, one accumu- lating the carcinomas with the highest hypermethylation, identified as the CIMP-high subgroup and the other sub- sequently identified as the CIMP-low subgroup.
Validation of hypermethylation by MS-MLPA
The CIMP status identified by unsupervised cluster- ing could be validated in 15 carcinomas by MS-MLPA. The CpG methylation level of 33 genes measured by MS-MLPA showed a strong correlation with the whole- genome methylation level measured by the BeadChip (correlation coefficient = 0.85, P = 5.9 × 10-5, Sup- plemental Fig. 1, published on The Endocrine Society’s Journals Online web site at http://jcem.endojournals.
org). The non-CIMP, CIMP-low, and CIMP-high could also be individualized using MS-MLPA.
Hypermethylation in carcinomas is associated with a poor prognosis
The global level of methylation in CpG islands was associated with survival (P = 0.02). In agreement, CIMP carcinomas were associated with a poorer prognosis than non-CIMP (log rank P = 0.04, Fig. 2A). Considering the subgroup of CIMP-high carcinomas, the prognosis was even poorer compared with non-CIMP carcinomas (log rank P = 0.006, Fig. 2B).
The prognostic value associated with the methylation of each CpG was also tested. A total of 667 CpGs (553 genes) showed a significant association between their methylation level and overall survival (P < 0.05, Supple- mental Table 1).
Hypermethylated genes
Non-CIMP, CIMP-low, and CIMP-high carcinomas are different in terms of global methylation level. We sought for the specific genes behind these variable meth- ylation levels. For that aim, CIMP-high carcinomas were compared with non-CIMP carcinomas, showing 4498 CpG sites (3235 genes) significantly hypermethylated in CIMP-high carcinomas (P < 0.05, Supplemental Table 2).
A
B
:
:
Whole genome CpG methylation level (mean M-values)
:
i
:
Group 1 non-CIMP
Group 2A CIMP-low
Group 2B CIMP-High
:
a
·
2
2
:
Carcinomas group 1 non-CIMP (n=26)
Carcinomas group 2A CIMP-Jow (n=17)
Carcinomas group 2B CIMP-High (n=8)
Adenomas (n=84)
Similarly, CIMP-low carcinomas were compared with non-CIMP carcinomas, showing 3490 CpG sites (2589 genes) significantly hypermethylated in CIMP-low carci- nomas (P < 0.05, Supplemental Table 3). A substantial overlap between the CIMP-high and CIMP-low hyperm- ethylated genes was observed: 74% of CIMP-low-associ- ated CpG sites were also hypermethylated in CIMP-high carcinomas.
To identify the genes hypermethylated in the carcino- mas, methylation in carcinomas was also compared with the adenomas. Non-CIMP carcinomas and CIMP carci- nomas showed 2737 hypermethylated CpGs (2206 genes) and 7485 hypermethylated CpGs (5372 genes), respec- tively. Among these, 1659 CpGs (1443 genes) were shared between CIMP and non-CIMP carcinomas (Supplemental Table 4).
A
B
100
Non-CIMP
100
Non-CIMP
CIMP
1
CIMP-Low
CIMP-High
80
80
1
Overall survival (%)
Overall survival (%)
++
60
60
40
40
20
20
0
0
0
20
40
60
80
100
120
140
0
20
40
60
80
100
120
140
Time (months)
Time (months)
Genes down-regulated by hypermethylation
Promoter hypermethylation is expected to lower the gene expression in cis. Among 12,250 genes with meth- ylation and expression data both available, a significant proportion was negatively correlated (1741 genes; 14% with P < 0.05). Gene ontology analyses of these genes showed an enrichment for immune response (P = 1.5 X 10-9) and regulation of cell proliferation (P = 7.1 X 10-6), two sets of genes usually considered as important for tumorigenesis. The top 25 genes with inverse correla- tion are listed in Table 2 and include H19 and several other known or putative tumor suppressor genes such as PLAGL-1, G0S2, and NDRG2.
About a hundred genes in the genome are submitted to parental imprinting. These imprints are specific DNA methylation patterns. Among the genes studied, 64 are known to be parentally imprinted. Of these 64 genes, 20 (31%) were present among the 1741 negatively correlated genes (P = 1 × 10-4, Supplemental Table 5).
We next focused on genes both down-regulated and hypermethylated. We found 818 and 1096 such genes in CIMP-high and CIMP-low carcinomas, which is higher than the 437 genes in non-CIMP carcinomas (P < 2 X 10-16, Fig. 3A). These numbers are impacted by the higher
methylation level in CIMP carcinomas. To correct for this effect, the proportion of hypermethylated genes among the down-regulated genes was compared with the propor- tion of hypermethylated genes among the genes that are not down-regulated. This showed a significant enrichment of hypermethylated genes among the down-regulated genes in all the carcinomas (Fig. 3B, P < 2 × 10-16 for CIMP-high and CIMP-low carcinomas, and P = 8 X 10-13 for non-CIMP carcinomas).
Hypermethylated carcinomas accumulate in specific transcriptome subgroups
As a reminder, the transcriptome classification discrimi- nated the adenomas (cluster C2) from carcinomas (cluster C1); the cluster of carcinomas was subdivided into two groups with different outcomes, the C1A, corresponding to carcinomas with poor prognosis, and C1B corresponding to carcinomas with good prognosis; finally, the cluster C1A of poor-prognosis carcinomas was subdivided into three sub- groups, one associated with p53 inactivation (C1A-p53), one with ß-catenin activation (C1A-ß-catenin), and one with a still unidentified molecular alteration (C1A-x) (4, 6) (Fig. 4).
Global methylation showed various levels among these transcriptome-based subgroups (Fig. 4). Within the sub-
| Gene | Correlation coefficient (Pearson) | Adjusted P values (FDR) | Chromosomal location | Gene/protein information |
|---|---|---|---|---|
| H19 | -0.83 | 1.17 × 10-28 | 11p15.5 | Noncoding RNA, tumor suppressor, imprinted gene |
| GSTM1 | -0.83 | 6.34 × 10-28 | 1p13.3 | Detoxification |
| GSTP1 | -0.81 | 6.10 × 10-26 | 11q13 | Detoxification |
| G0S2 | -0.8 | 4.01 × 10-25 | 1q32.2 | Proapoptotic, tumor suppressor |
| GSTT1 | -0.8 | 9.24 × 10-25 | 22q11.23 | Detoxification |
| RAB34 | -0.8 | 1.24 × 10-24 | 17q11.2 | RAB family of proteins, small GTPases |
| GYPC | -0.72 | 9.78 × 10-18 | 2q14-q21 | Integral membrane glycoprotein, role in regulating the mechanical stability of red cells |
| GIPC2 | -0.71 | 6.50 × 10-17 | 1p31.1 | Protein with central PDZ domain |
| PLAGL1 | -0.71 | 6.86 × 10-17 | 6q24-q25 | Tumor suppressor, imprinted gene |
| LY6D | -0.71 | 1.11 × 10-16 | 8q24-qter | Lymphocyte antigen 6 complex |
| PCOLCE | -0.7 | 2.64 × 10-16 | 7q22 | Procollagen C-endopeptidase enhancer |
| NDN | -0.7 | 3.65 × 10-16 | 15q11.2-q12 | Negative growth regulator, imprinted gene |
| AMT | -0.7 | 5.20 × 10-16 | 3p21.2-p21.1 | Component of the glycine cleavage system |
| LGALS3BP | -0.7 | 5.20 × 10-16 | 17q25 | ß-Galactoside-binding protein |
| APOC1 | -0.69 | 7.16 × 10-16 | 19q13.2 | Apolipoprotein C-I |
| TM7SF2 | -0.69 | 1.99 × 10-15 | 11q13 | Reduction of C14-unsaturated sterols in cholesterol biosynthesis |
| PPAPDC3 | -0.69 | 2.38 × 10-15 | 9q34.13 | Phosphatidic acid phosphatase type 2 domain containing 3 |
| PTPN7 | -0.69 | 2.63 × 10-15 | 1q32.1 | Protein tyrosine phosphatase |
| SCNN1A | -0.68 | 2.85 × 10-15 | 12p13 | Sodium channel, non-voltage-gated 1«-subunit |
| HSD3B2 | -0.68 | 6.97 × 10-15 | 1p13.1 | 3ß-Hydroxysteroid dehydrogenase |
| ACAA2 | -0.68 | 6.98 × 10-15 | 18q21.1 | Catalyzes the last step of the mitochondrial fatty acid ß-oxidation spiral |
| CTSZ | -0.68 | 7.24 × 10-15 | 20q13 | Lysosomal cysteine proteinase |
| PYGM | -0.68 | 7.44 × 10-15 | 11q12-q13.2 | Muscle enzyme involved in glycogenolysis |
| KRT8 | -0.68 | 8.57 × 10-15 | 12q13 | Keratin 8 |
| NDRG2 | -0.67 | 1.16 × 10-14 | 14q11.2 | Tumor suppressor |
FDR, False discovery rate.
A
Carcinomas CIMP-High
Carcinomas CIMP-Low
Carcinomas non-CIMP
.
.
.
Expression fold change (log2, CIMP-High vs adenomas)
Expression fold change (log2, CIMP-Low vs adenomas)
Expression fold change (log2, no-CIMP vs adenomas)
-
-
2
·
-
A
”
”
”
1
+
+
-4
a
2
4
6
.
2
Đ
2
4
.
B
-2
₫
2
4
6
8
DNA methylation fold change (log2, CIMP-High vs adenomas)
DNA methylation fold change (log2, CIMP-Low vs adenomas)
DNA methylation fold change (log2, non-CIMP vs adenomas)
B
100
Proportion of hypermethylated genes (%)
80
p =< 2e-16
p =< 2e-16
60
40
p=8e-13
20
underexpressed genes
not underexpressed genes
0
CIMP-High
CIMP-Low
non-CIMP
groups of carcinomas of poor prognosis, almost all the carcinomas from the C1A-p53 (nine of 10) and the C1A-x (four of four) subgroups showed a CIMP (Fig. 4). All the CIMP-high carcinomas belong to these two subgroups. In contrast, a non-CIMP pattern was observed in the poor- prognosis C1A-B-catenin subgroup and the good-prog- nosis C1B subgroup (P = 4.8 × 10-4).
Discussion
In this study, we described the genome-wide patterns of methylation in a large series of adrenocortical tumors, us- ing an array-based technology to investigate the methyl- ation status of 19,252 CpG within CpG islands, repre- senting 14,475 genes. Overall, the pangenomic analysis showed a much higher methylation level in carcinomas
than in adenomas. Another study has identified 212 CpG islands hypermethylated in adrenocortical carcinomas compared with adenomas (16). A recent study reported a global hypomethylation in carcinomas compared with ad- enomas (15). This apparent discrepancy can be explained by the difference in the CpGs that were interrogated. In- deed, hypomethylation reported by Rechache et al. (15) is related to CpGs both outside of the CpG islands and in intergenic regions that are poorly represented in our study. In addition, restricting the analysis to CpG islands in gene regions, Rechache et al. (15) also identified a hypermeth- ylation in carcinomas compared with adenomas, in agree- ment with our results. Methylation in promoter regions has an important impact on gene expression (24). The physiopathological role of the isolated CpGs in intergenic regions is currently poorly understood, as far as we know.
Transcriptome-based tumor classification
C1 (carcinomas)
C2 (adenomas, n=54)
C1A (carcinomas of poor prognosis)
C1B (n=20) (carcinomas of good prognosis )
C1A-p53 (n=10) (p53 inactivation)
C1A-x (n=4) (unidentified alteration)
C1A-Bcatenin (n=6) (Bcatenin activation)
Methylation-based classification
0.8
Whole genome CpG methylation level (mean of M-values)
0.6
0.4
0.2
.
0.0
-0.2
The methylation level in adrenocortical carcinomas is heterogeneous. This study identified a subtype of carci- nomas with a high CpG islands methylation level. Sub- types of carcinomas associated with specific methylation patterns have already been reported in cancers from other organs, such as breast (9), colon (10), ovary (25), stomach (26), liver (27), pancreas (28), esophagus (29), kidney (30), and melanoma (31). The group of highly methylated carcinomas described here parallels the CIMP first de- scribed as a distinct subset of colorectal tumors (10). Sub- sequently, a similar phenotype has been described in a wide range of other neoplasms (8). In some tumor types, CIMP has been shown to be associated with poor survival (27, 31). Similarly, the hypermethylated adrenocortical carcinomas group is associated with a poor prognosis, especially in the CIMP-high group. The poor prognosis in the CIMP-low group seems to be attenuated compared with CIMP-high (Fig. 2B), although no significant differ- ence could be reached when comparing the CIMP-high and CIMP-low group (log rank P = 0.12). Beyond sur- vival, no difference was found between CIMP-high and CIMP-low tumors, regarding the tumor size or the tumor stage. Of note, we also performed an unsupervised clus- tering of the adenomas and could identify two subgroups
with different methylation levels. These two groups did not differ in the tumor size, in the secretion status, or in the patients’ sex (data not shown).
Tumors with intermediate Weiss score cannot be un- ambiguously classified. In this series, four tumors with a Weiss score of 2 with androgen secretion (undetermined malignancy) were included in the carcinomas group. In- deed, three of these four tumors had previously been an- alyzed by transcriptome (4), and unsupervised clustering unambiguously classified these three tumors within the carcinomas. In addition, removing these tumors did not change the clustering based on the methylation or the poorer survival associated with hypermethylation pre- sented here (data not shown).
Methylation has an important impact on gene expres- sion. An inverse correlation between methylation and ex- pression levels of many genes has been found in our study, including known tumor suppressor genes. The strongest inverse correlation was observed for H19. H19 is a gene submitted to parental imprinting. H19 is underexpressed in adrenocortical carcinomas due to the hypermethylation of the imprinting control region resulting from the loss of the maternal allele and the duplication of the imprinted paternal allele (loss of heterozygosity in 11p15) (13, 14).
The presence of H19 at the top of the inverse correlation between methylation and gene expression in our study confirmed the validity of our strategy. Nineteen other im- printed genes are present among genes negatively regu- lated by methylation. But imprinting control regions of these genes are not always well defined, and we cannot confirm that the observed inverse correlation is due to the hypermethylation of the imprinting control regions. Among other genes with a strong inverse correlation, we identified detoxification genes (GSTP1, GSTM1, and GSTT1) and known tumor suppressor genes (G0S2, PLAGL1, and NDRG2). The glutathione-S-transferases superfamily plays an important role in protecting cells from carcinogens by detoxifying cytotoxic and carcino- genic agents. GSTP1 hypermethylation is associated with reduced expression and aggressive disease in different neo- plasms (32-34). Detoxification mechanisms play also a pivotal role in determining tumor cell responses to plati- num-based chemotherapy. Patients with tumoral GSTP hypermethylation and reduced expression could respond better to chemotherapy (35). Platinum-based chemother- apy has been established as a reference treatment for met- astatic adrenocortical carcinomas (36). Whether glutathi- one-S-transferase methylation impacts the response to platinum chemotherapy in adrenocortical carcinomas re- mains to be determined. G0S2 encodes a mitochondrial protein that specifically interacts with Bcl-2 and promotes apoptosis by preventing the formation of protective Bcl- 2/Bax heterodimers. Ectopic expression of G0S2 induces apoptosis in diverse human cancer cell lines in which en- dogenous G0S2 is normally epigenetically silenced (37). PLAGL1 is an imprinted tumor suppressor gene encoding an important inducer of cell cycle arrest and apoptosis and is widely expressed in adult tissues, with the most abun- dant expression detected in the pituitary gland, kidney, placenta, and adrenal gland (38). Loss of PLAGL1 ex- pression has been reported in a number of human tumors, including breast, ovary and prostate cancer, pituitary ad- enoma, pheochromocytoma, squamous cell carcinoma of the head and neck, basal cell carcinoma, and hemangio- blastoma. NDRG2, a member of the N-myc downstream- regulated gene family, is down-regulated or undetectable in many human cancers. Studies have found that NDRG2 is able to inhibit proliferation and enhance apoptosis in many malignant tumors (39). Thus, methylation could be an important mechanism in gene deregulation during ad- renocortical tumorigenesis. This is further supported by experimental evidence. A few studies have demonstrated that the treatment of the human adrenocortical cell line NCI-H295R with the DNA methylation inhibitor decit- abine decreases the proliferation rate and inhibits cell in- vasion, even at low doses (40, 41).
Rather unexpectedly, we also observed 559 genes (4.6%) with a positive correlation between methylation and expression levels (data not shown). This positive cor- relation has also been reported in other studies dealing with transcriptome and methylation (9). Several hypoth- eses can be raised. First, the effect of CpG methylation on gene expression depends on the localization of the CpG relative to the promoter region (42). Some methylated CpG might block the fixation of transcriptional inhibitors or might down-regulate inhibitory antisense transcripts. An- other hypothesis is the occurrence of a modified DNA base, called 5-hydroxymethylcytosine, discovered recently (43). This base could regulate gene expression differently from methylcytosine. Enzymatic or bisulfite-based approaches cannot discriminate between hydroxy- or 5-methylcytosine, due to structural similarity (44).
Interestingly, this study demonstrated that the differ- ence in the methylation profiles parallels the transcrip- tome-based classification. Previous studies showed that the subgroups of adrenocortical carcinomas identified by the transcriptome were associated with different gene ex- pression patterns, unraveling different molecular mecha- nisms (especially p53 inactivation or Wnt/ß-catenin acti- vation). These subgroups were also associated with different outcomes (3, 4, 6). This study shows that these subgroups are also associated with different methylation patterns. Especially, the tumors of two subgroups of ad- renocortical carcinomas associated with a poor prognosis (C1A-p53 and C1A-x) are hypermethylated, whereas the third subgroup of poor prognosis (C1A-ß-catenin) and the subgroup of carcinomas of good prognosis (C1B) contain only a few hypermethylated carcinomas. This distribution of hypermethylation among the different subgroups of carcinomas is probably responsible for the prognostic value of hypermethylation. However, considering that some tumors of poor prognosis (from the C1A-ß-catenin subgroup) are not hypermethylated, this prognostic value can be anticipated to be less efficient than the predictions based on gene expression. Comparative studies are needed, especially in a prospective setting, to precisely as- sess the respective performances of these markers.
Conclusion
This genome-wide methylation study in adrenocortical tumors reveals the existence of hypermethylation or CIMP in some carcinomas. The CIMP status parallels the tran- scriptome-based classification of carcinomas. These re- sults reinforce the molecular classification based on the genomic analysis of adrenal tumors and give new insights in the pathogenesis of the various subtypes of carcinomas. Finally, the CIMP status also provides prognostic infor-
mation. The relevance of this prediction in prospective clinical settings remains to be specified.
Acknowledgments
We thank the Genomic Platform of Institut Cochin (Franck Le- tourneur) for their technical support, Jacqueline Metral and Jac- queline Godet from the Ligue Nationale Contre le Cancer for the organization of the Cartes d’Identité des Tumeurs program, Vé- ronique Duchossoy from the oncogenetic Unit of Cochin Hos- pital for MS-MLPA validations, Eric Clauser for helpful discus- sions, the members of our laboratories and the COMETE and ENSAT networks for support and discussions, and all the staff of the clinical departments of Cochin Hospital who were involved in patient care.
Address all correspondence and requests for reprints to: Jérôme Bertherat, Assistance Publique Hôpitaux de Paris, Hôpital Cochin, 75014 Paris, France. E-mail: jerome. bertherat@inserm.fr.
This work is part of the program Cartes d’Identité des Tumeurs (CIT, http://cit.ligue-cancer.net) funded and developed by the Ligue Nationale Contre le Cancer; it was supported in part by the Conny-Maeva foundation, the Plan Hospitalier de Re- cherche Clinique (AOM06179) to the COMETE Network, the Recherche Translationnelle DHOS/INCA 2009 (RTD09024), and the European Union Seventh Framework Program (FP7/ 2007-2013) under Grant Agreement 259735 (project ENS@ T-Cancer). O.B. is recipient of an Institut National de la Santé et de la Recherche Médicale (INSERM) fellowship. G.A. is recip- ient of INSERM support (contrat d’interface).
Disclosure Summary: The authors indicated no potential con- flicts of interest.
References
1. Fassnacht M, Allolio B 2009 Clinical management of adrenocortical carcinoma. Best Pract Res Clin Endocrinol Metab 23:273-289
2. Bertherat J, Bertagna X 2009 Pathogenesis of adrenocortical cancer. Best Pract Res Clin Endocrinol Metab 23:261-271
3. Giordano TJ, Kuick R, Else T, Gauger PG, Vinco M, Bauersfeld J, Sanders D, Thomas DG, Doherty G, Hammer G 2009 Molecular classification and prognostication of adrenocortical tumors by tran- scriptome profiling. Clin Cancer Res 15:668-676
4. de Reyniès A, Assié G, Rickman DS, Tissier F, Groussin L, René- Corail F, Dousset B, Bertagna X, Clauser E, Bertherat J 2009 Gene expression profiling reveals a new classification of adrenocortical tumors and identifies molecular predictors of malignancy and sur- vival. J Clin Oncol 27:1108-1115
5. Szabó PM, Tamási V, Molnár V, Andrásfalvy M, Tömböl Z, Farkas R, Kövesdi K, Patócs A, Tóth M, Szalai C, Falus A, Rácz K, Igaz P 2010 Meta-analysis of adrenocortical tumour genomics data: novel pathogenic pathways revealed. Oncogene 29:3163-3172
6. Ragazzon B, Libé R, Gaujoux S, Assié G, Fratticci A, Launay P, Clauser E, Bertagna X, Tissier F, de Reyniès A, Bertherat J 2010 Transcriptome analysis reveals that p53 and ß-catenin alterations occur in a group of aggressive adrenocortical cancers. Cancer Res 70:8276-8281
7. Kulis M, Esteller M 2010 DNA methylation and cancer. Adv Genet 70:27-56
8. Issa JP 2004 CpG island methylator phenotype in cancer. Nat Rev Cancer 4:988-993
9. Van der Auwera I, Yu W, Suo L, Van Neste L, van Dam P, Van Marck EA, Pauwels P, Vermeulen PB, Dirix LY, Van Laere SJ 2010 Array-based DNA methylation profiling for breast cancer subtype discrimination. PLoS One 5:e12616
10. Toyota M, Ahuja N, Ohe-Toyota M, Herman JG, Baylin SB, Issa JP 1999 CpG island methylator phenotype in colorectal cancer. Proc Natl Acad Sci USA 96:8681-8686
11. Hinoue T, Weisenberger DJ, Lange CP, Shen H, Byun HM, Van Den Berg D, Malik S, Pan F, Noushmehr H, van Dijk CM, Tollenaar RA, Laird PW 2012 Genome-scale analysis of aberrant DNA methyl- ation in colorectal cancer. Genome Res 22:271-282
12. Weisenberger DJ, Siegmund KD, Campan M, Young J, Long TI, Faasse MA, Kang GH, Widschwendter M, Weener D, Buchanan D, Koh H, Simms L, Barker M, Leggett B, Levine J, Kim M, French AJ, Thibodeau SN, Jass J, Haile R, Laird PW 2006 CpG island methy- lator phenotype underlies sporadic microsatellite instability and is tightly associated with BRAF mutation in colorectal cancer. Nat Genet 38:787-793
13. Gao ZH, Suppola S, Liu J, Heikkila P, Jänne J, Voutilainen R 2002 Association of H19 promoter methylation with the expression of H19 and IGF-II genes in adrenocortical tumors. J Clin Endocrinol Metab 87:1170-1176
14. Gicquel C, Bertagna X, Schneid H, Francillard-Leblond M, Luton JP, Girard F, Le Bouc Y 1994 Rearrangements at the 11p15 locus and overexpression of insulin-like growth factor-II gene in sporadic adrenocortical tumors. J Clin Endocrinol Metab 78:1444-1453
15. Rechache NS, Wang Y, Stevenson HS, Killian JK, Edelman DC, Merino M, Zhang L, Nilubol N, Stratakis CA, Meltzer PS, Kebebew E 2012 DNA methylation profiling identifies global methylation differences and markers of adrenocortical tumors. J Clin Endocrinol Metab 97:1004-1013
16. Fonseca AL, Kugelberg J, Starker LF, Scholl U, Choi M, Hellman P, Akerström G, Westin G, Lifton RP, Björklund P, Carling T 2012 Comprehensive DNA methylation analysis of benign and malignant adrenocortical tumors. Genes Chromosomes Cancer 51:949-960
17. Fassnacht M, Johanssen S, Quinkler M, Bucsky P, Willenberg HS, Beuschlein F, Terzolo M, Mueller HH, Hahner S, Allolio B 2009 Limited prognostic value of the 2004 International Union Against Cancer staging classification for adrenocortical carcinoma: pro- posal for a Revised TNM Classification. Cancer 115:243-250
18. Weiss LM 1984 Comparative histologic study of 43 metastasizing and nonmetastasizing adrenocortical tumors. Am J Surg Pathol 8:163-169
19. Terzolo M, Alì A, Osella G, Reimondo G, Pia A, Peretti P, Paccotti P, Angeli A 2000 The value of dehydroepiandrosterone sulfate mea- surement in the differentiation between benign and malignant ad- renal masses. Eur J Endocrinol 142:611-617
20. Du P, Kibbe WA, Lin SM 2008 lumi: a pipeline for processing Illu- mina microarray. Bioinformatics 24:1547-1548
21. Du P, Zhang X, Huang CC, Jafari N, Kibbe WA, Hou L, Lin SM 2010 Comparison of ß-value and M-value methods for quantifying methylation levels by microarray analysis. BMC Bioinformatics 11: 587
22. Team RDC 2011 R: a language and environment for statistical com- puting. R Foundation for Statistical Computing, ed. Vienna: R Foundation for Statistical Computing
23. Huang da W, Sherman BT, Lempicki RA 2009 Systematic and in- tegrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 4:44-57
24. Deaton AM, Bird A 2011 CpG islands and the regulation of tran- scription. Genes Dev 25:1010-1022
25. Strathdee G, Appleton K, Illand M, Millan DW, Sargent J, Paul J, Brown R 2001 Primary ovarian carcinomas display multiple methy- lator phenotypes involving known tumor suppressor genes. Am J Pathol 158:1121-1127
26. Toyota M, Ahuja N, Suzuki H, Itoh F, Ohe-Toyota M, Imai K,
Baylin SB, Issa JP 1999 Aberrant methylation in gastric cancer as- sociated with the CpG island methylator phenotype. Cancer Res 59:5438-5442
27. Li B, Liu W, Wang L, Li M, Wang J, Huang L, Huang P, Yuan Y 2010 CpG island methylator phenotype associated with tumor re- currence in tumor-node-metastasis stage I hepatocellular carci- noma. Ann Surg Oncol 17:1917-1926
28. Ueki T, Toyota M, Sohn T, Yeo CJ, Issa JP, Hruban RH, Goggins M 2000 Hypermethylation of multiple genes in pancreatic adeno- carcinoma. Cancer Res 60:1835-1839
29. Eads CA, Lord RV, Kurumboor SK, Wickramasinghe K, Skinner ML, Long TI, Peters JH, DeMeester TR, Danenberg KD, Danenberg PV, Laird PW, Skinner KA 2000 Fields of aberrant CpG island hypermethylation in Barrett’s esophagus and associated adenocar- cinoma. Cancer Res 60:5021-5026
30. McRonald FE, Morris MR, Gentle D, Winchester L, Baban D, Ra- goussis J, Clarke NW, Brown MD, Kishida T, Yao M, Latif F, Maher ER 2009 CpG methylation profiling in VHL related and VHL unrelated renal cell carcinoma. Mol Cancer 8:31
31. Tanemura A, Terando AM, Sim MS, van Hoesel AQ, de Maat MF, Morton DL, Hoon DS 2009 CpG island methylator phenotype pre- dicts progression of malignant melanoma. Clin Cancer Res 15: 1801-1807
32. Gumy-Pause F, Pardo B, Khoshbeen-Boudal M, Ansari M, Gayet- Ageron A, Sappino AP, Attiyeh EF, Ozsahin H 2012 GSTP1 hyper- methylation is associated with reduced protein expression, aggres- sive disease and prognosis in neuroblastoma. Genes Chromosomes Cancer 51:174-185
33. Chan QK, Khoo US, Chan KY, Ngan HY, Li SS, Chiu PM, Man LS, Ip PP, Xue WC, Cheung AN 2005 Promoter methylation and dif- ferential expression of T-class glutathione S-transferase in endome- trial carcinoma. J Mol Diagn 7:8-16
34. Nakayama M, Gonzalgo ML, Yegnasubramanian S, Lin X, De Marzo AM, Nelson WG 2004 GSTP1 CpG island hypermethylation as a molecular biomarker for prostate cancer. J Cell Biochem 91: 540-552
35. Noda E, Maeda K, Inoue T, Fukunaga S, Nagahara H, Shibutani M, Amano R, Nakata B, Tanaka H, Muguruma K, Yamada N, Yashiro M, Ohira M, Ishikawa T, Hirakawa K 2012 Predictive value of expression of ERCC 1 and GST-p for 5-fluorouracil/oxaliplatin che-
motherapy in advanced colorectal cancer. Hepatogastroenterology 59:130-133
36. Fassnacht M, Terzolo M, Allolio B, Baudin E, Haak H, Berruti A, Welin S, Schade-Brittinger C, Lacroix A, Jarzab B, Sorbye H, Torpy DJ, Stepan V, Schteingart DE, Arlt W, Kroiss M, Leboulleux S, Sperone P, Sundin A, Hermsen I, Hahner S, Willenberg HS, Tabarin A, Quinkler M, de la Fouchardière C, et al. 2012 Combination che- motherapy in advanced adrenocortical carcinoma. N Engl J Med 366:2189-2197
37. Welch C, Santra MK, El-Assaad W, Zhu X, Huber WE, Keys RA, Teodoro JG, Green MR 2009 Identification of a protein, G0S2, that lacks Bcl-2 homology domains and interacts with and antagonizes Bcl-2. Cancer Res 69:6782-6789
38. Varrault A, Ciani E, Apiou F, Bilanges B, Hoffmann A, Pantaloni C, Bockaert J, Spengler D, Journot L 1998 hZAC encodes a zinc finger protein with antiproliferative properties and maps to a chromo- somal region frequently lost in cancer. Proc Natl Acad Sci USA 95:8835-8840
39. Liu N, Wang L, Liu X, Yang Q, Zhang J, Zhang W, Wu Y, Shen L, Zhang Y, Yang A, Han H, Zhang J, Yao L 2007 Promoter meth- ylation, mutation, and genomic deletion are involved in the de- creased NDRG2 expression levels in several cancer cell lines. Biochem Biophys Res Commun 358:164-169
40. Liu J, Li XD, Vaheri A, Voutilainen R 2004 DNA methylation af- fects cell proliferation, cortisol secretion and steroidogenic gene ex- pression in human adrenocortical NCI-H295R cells. J Mol Endo- crinol 33:651-662
41. Suh I, Weng J, Fernandez-Ranvier G, Shen WT, Duh QY, Clark OH, Kebebew E 2010 Antineoplastic effects of decitabine, an inhibitor of DNA promoter methylation, in adrenocortical carcinoma cells. Arch Surg 145:226-232
42. van Vlodrop IJ, Niessen HE, Derks S, Baldewijns MM, van Criekinge W, Herman JG, van Engeland M 2011 Analysis of pro- moter CpG island hypermethylation in cancer: location, location, location! Clin Cancer Res 17:4225-4231
43. Tahiliani M, Koh KP, Shen Y, Pastor WA, Bandukwala H, Brudno Y, Agarwal S, Iyer LM, Liu DR, Aravind L, Rao A 2009 Conversion of 5-methylcytosine to 5-hydroxymethylcytosine in mammalian DNA by MLL partner TET1. Science 324:930-935
44. Huang Y, Pastor WA, Shen Y, Tahiliani M, Liu DR, Rao A 2010 The behaviour of 5-hydroxymethylcytosine in bisulfite sequencing. PLoS One 5:e8888