DNA Methylation Profiling Identifies Global Methylation Differences and Markers of Adrenocortical Tumors

Abstract

Context:

It is not known whether there are any DNA methylation alterations in adrenocortical tumors.

Objective:

The objective of the study was to determine the methylation profile of normal adrenal cortex and benign and malignant adrenocortical tumors.

Methods:

Genome-wide methylation status of CpG regions were determined in normal (n = 19), benign (n = 48), primary malignant (n = 8), and metastatic malignant (n = 12) adrenocortical tissue samples. An integrated analysis of genome-wide methylation and mRNA expression in benign vs. malignant adrenocortical tissue samples was also performed.

Results:

Methylation profiling revealed the following: 1) that methylation patterns were distinctly different and could distinguish normal, benign, primary malignant, and metastatic tissue samples; 2) that malignant samples have global hypomethylation; and 3) that the methylation of CpG regions are different in benign adrenocortical tumors by functional status. Normal compared with benign samples had the least amount of methylation differences, whereas normal compared with primary and metastatic adrenocortical carcinoma samples had the greatest variability in methylation (adjusted P ≤ 0.01). Of 215 down-regulated genes (≥2-fold, adjusted P ≤ 0.05) in malignant primary adrenocortical tumor samples, 52 of these genes were also hypermethylated.

Conclusions:

Malignant adrenocortical tumors are globally hypomethylated as compared with normal and benign tumors. Methylation profile differences may accurately distinguish between primary benign and malignant adrenocortical tumors. Several differentially methylated sites are associated with genes known to be dysregulated in malignant adrenocortical tumors.

Adrenocortical carcinomas (ACC) are rare malignancies that have an annual incidence of 0.5–2 cases per million, but adrenal neoplasms are common and are identified in up to 14% of the population (1–4). ACC has a very poor prognosis with only a 15–45% survival at 5 yr (5). The incidence of ACC is highest in the fifth and sixth decade of life; however, a high incidence of ACC has been found in children in southern Brazil, which was associated with a germline mutation in p53 (6, 7). More than half of ACC are considered functional when excess hormones are produced, and Cushing's syndrome is the most common clinical manifestation (8, 9). The majority of ACC cases are sporadic; however, several genetic syndromes have been associated with ACC including Beckwith-Wiedemann and Li-Fraumeni syndromes, Carney complex, familial adenomatous polyposis, congenital adrenal hyperplasia, and multiple endocrine neoplasia type 1 (9, 10). Without the clear presentation of local invasion or distant metastasis, the diagnosis and pathological distinction between benign and malignant adrenocortical tumors can be difficult. Histologically, the nine Weiss parameters (0–9 score) are commonly used for diagnosis, with Weiss scores of 2 or less being considered benign and 3 or greater being considered malignant (10). However, given the subjective nature of the Weiss histological features, particularly with regard to scores of 2–3 that are of indeterminate classification and the most common group of adrenocortical tumors, genomic markers that can accurately diagnose ACC, determine prognosis, and shed light on the molecular basis of ACC could have important clinical ramifications (8, 11).

Several genetic changes at the chromosomal level (losses and gains) have been found by comparative genomic hybridization in addition to loss of heterozygosity and allelic imbalances in the majority of sporadic ACC cases (6, 7, 12). The most common changes are gains at chromosomes 4, 5, 12, and 19 and losses at chromosomes 1, 2, 11, 17, and 22 (6, 7, 12). In addition, high throughput gene expression analysis has confirmed that overexpression of IGF2, a fetal growth factor at the imprinted chromosome 11p15 locus, is an important marker of ACC that is up-regulated in 90% of ACC cases and is regulated by both genetic and epigenetic factors (6, 9, 11, 13). Another important imprinted gene in the 11p15 locus is H19, a non-protein-coding RNA that is thought to inhibit IGF-II expression (12). The autocrine/paracrine growth effects of IGF-II occurs via the IGF-I receptor (IGF1R) (12). When IGF-II binds the IGF1R, intracellular networks that affect cell survival and mRNA translation such as the phosphatidylinositol 3-kinase-AKT-mammalian target of rapamycin pathway and proliferation via the RAF-MAPK pathway are activated (14). In addition to p53, IGF2, and H19 have been implicated in the pathogenesis of ACC as well as SF1, GATA6, p57^kip2, and MEN1 genes, β-catenin (CTNNB1), and the Wnt and ACTH-cAMP-protein kinase A pathways and other growth factors (TGF, FGF, VEGF, and EGF) (6, 10, 12).

More recently, changes at the epigenetic level have been implicated in carcinogenesis and found to be diagnostic and prognostic markers. Epigenetics refers to heritable changes in gene expression that are not due to changes in DNA. The best defined epigenetic change is DNA methylation of cytosines, by DNA methyltransferase enzymes. Cytosines associated with guanines are called CpG dinucleotides, and these are generally found in CpG-rich regions called CpG islands. CpG islands are defined as regions of greater than 500 bp that have guanine cytosine content of greater than 55% (15). Up to 60% of CpG islands are in the 5′ regulatory (promoter) regions of genes (16–18). However, CpG islands that are not in promoter regions can also be found within coding regions and noncoding regions of genes, which may be targets for de novo methylation in cancer and aging (15). DNA methylation affects a number of different cellular processes including apoptosis, cell cycle, DNA damage repair, growth factor response, signal transduction, and tumor architecture, all of which can contribute to the initiation and progression of cancer (19).

To our knowledge, there have been no studies on methylation in ACC. Because of the potential for understanding DNA methylation changes that may be associated with ACC and that differential DNA methylation status may serve as diagnostic markers and or targets for therapy for ACC, we performed genome-wide DNA methylation analysis using a platform with 485,421 cytosine probe sites. We characterized the methylome of 87 adrenocortical tissue samples (19 normal, 48 benign, eight primary malignant, and 12 metastatic) and determined the correlation of gene-methylation status with gene expression levels in benign vs. malignant adrenocortical tissues samples.

Materials and Methods

Tissue samples

Adrenocortical tissue samples were collected as described previously (13, 20). Eighty-seven tissue samples were obtained at surgical resection and were immediately snap frozen and stored at −80 C. Normal adrenal glands were obtained at the time of nephrectomy for organ donation and immediately snap frozen and stored at −80 C. Demographic, clinical, and pathological information was collected after written informed consent under an institutional review board approved protocol. Tumors were classified as adrenocortical carcinoma when there was gross local invasion or distant metastasis was present at diagnosis or developed during follow-up. Benign adrenocortical tissue samples were classified if the tumor was localized at presentation, and there was no evidence of local or distant recurrent disease after follow-up of an average of 2.1 yr (range 1–10 yr) (Table 1).

Table 1.

Clinical features of tissue samples used for methylation profiling

	Benign tumor	Primary tumor	Metastatic tumor
Number of samples	48	8	12
Age (yr), mean ± sem	48.9 ± 1.0	42.3 ± 5.3	57.2 ± 4.8
Gender (women/men)	35/13	5/3	11/1
Type of tumor	48 primary	8 primary	7 locoregional recurrence, 2 liver metastases, 2 lung metastases, 1 abdominal wall metastasis
Syndrome
Hypercortisolism	16	6	7
Hyperaldosteronism	16	0	0
Nonfunctioning	16	2	5

Methylation profiling of tissue samples

Frozen adrenocortical tissue was sectioned for DNA isolation, and total DNA was extracted using DNA STAT-60 (Tel-Test Inc., Friendswood, TX) or DNeasy blood and tissue kit (QIAGEN, Valencia, CA). DNA quality was determined using a NanoDrop 2000c spectrophotometer (Thermo Fisher Scientific Inc., Wilmington, DE).

One microgram of DNA was bisulfite converted using the EZ DNA methylation gold kit (Zymo Research Corp., Irvine, CA) according to the manufacturer's protocol with a modified thermocycling procedure as suggested by Illumina (San Diego, CA) (16 cycles of 95 C for 30 sec, 50 C for 60 min). Four microliters (∼600 ng) of the bisulfite converted DNA were assayed on Infinium HumanMethylation450 BeadChips using the Illumina Infinium HD methylation assay kit (Illumina) (21). These chips assess the methylation status at greater than 485,000 individual CpG sites encompassing 99% of reference sequence genes and 96% of CpG islands (19). Each DNA sample first underwent an overnight isothermal whole-genome amplification step. Amplified DNA was fragmented, precipitated, and resuspended. Samples were hybridized to BeadChips overnight at 48 C in an Illumina hybridization oven. Using an automated protocol on the Tecan Evo robot (Tecan Group Ltd., Mannedorf, Switzerland), hybridized arrays were processed through a single-base extension reaction on the probe sequence using DNP- or biotin-labeled nucleotides, with subsequent immunostaining. The BeadChips were then coated, dried, and imaged on an Illumina HiScanSQ. Image data were extracted using the Genome Studio version 2010.3 methylation module (Illumina). Beta values were calculated at each locus (β = intensity of methylated allele/intensity of unmethylated allele + intensity of methylated allele + 100) followed by analysis with R package to normalize the data. The X and Y chromosomes' methylation data were excluded from the results. Quality control inclusion valuation depended on hybridization detection P values of less than 0.05. One benign sample failed to meet quality control standards and was thus excluded from subsequent data analysis.

mRNA microarray of tissue samples

Frozen adrenocortical tissue was sectioned for RNA isolation, and a serial section was stained using hematoxylin-eosin to confirm diagnosis and tumor content of greater than 80%. Total RNA was extracted from homogenized frozen tissue using Trizol reagent (Invitrogen, Carlsbad, CA) and was purified using an RNeasy minikit (QIAGEN). One microgram of total RNA was used for amplification and labeling with the MessageAmp amplified RNA kit (Ambion Inc., Foster City, CA). Fragmented and labeled complementary RNA (12 μg) was hybridized to a gene chip (Affymetrix Human Genome U133 plus 2.0 GeneChip; Affymetrix Inc., Santa Clara, CA) for16 h at 45 C. The gene chip arrays were stained and washed (Affymetrix Fluidics Station 400; Affymetrix) according to the manufacturer's protocol. The probe intensities were measured using an argon laser confocal scanner (GeneArray scanner; Hewlett-Packard, Palo Alto, CA).

Data and statistical analyses

Data analysis for the mRNA microarrays was carried out using the Affymetrix GeneChip operating software (Affymetrix) to process the raw microarray data. To generate intensity values in the log2 scale, R/Bioconductor statistics were used for each probe set using the robust multiarray average method with default variables (22–24). For the class comparison (benign vs. malignant), the limma package in R/Bioconductor was used to calculate the moderated t statistics and the associated P values and the log posterior odds ratio (B statistic) that a gene is differentially expressed compared with not differentially expressed (25). The P values were adjusted for multiple testing by controlling for the false discovery rate using the Benjamini-Hochberg method (26).

Data analysis for the methylation BeadChip arrays was carried out by extracting image data using the Genome Studio version 2010.3 methylation module (Illumina). Beta values were calculated at each locus followed by analysis with R package and Partek software (Partek Inc., St. Louis, MO). For the comparison of the different tissue type groups, ANOVA was used based on the M values converted from corresponding beta values from each locus, and P values were adjusted for multiple testing by controlling for the false discovery rate using the Benjamini-Hochberg method (27). Pathway and biological function analysis was conducted using Ingenuity Pathway Analysis software (Ingenuity Systems Inc., Redwood City, CA).

Results

Methylation profile of human adrenocortical tissue samples

Unsupervised hierarchical cluster analysis was performed on 19 normal, 47 benign, eight primary malignant, and 12 metastatic tissue samples. Primary malignant ACC and metastatic tissue samples were globally hypomethylated compared with normal and benign tissue samples (adjusted P ≤ 0.01) (Fig. 1).

Fig. 1.

Unsupervised heirarchical cluster analysis of normal, benign, primary malignant, and metastatic tissue samples using ANOVA and an adjusted value of P ≤ 0.01. Primary ACC and metastatic tissue samples are globally hypomethylated. Red regions represent hypermethylation and blue regions are hypomethylation.

Differentially methylated sites were found in both coding and noncoding regions of DNA. Using an adjusted P ≤ 0.01 and beta value differences of Δβ of −0.20 or less or Δβ of 0.20 or greater, the smallest methylation differences were found between normal and benign tissue samples (104 total differentially methylated sites), 67.3% of which were hypermethylated in benign tissue samples. The largest differences were between primary and metastatic ACC samples compared with normal tissue samples (24,229 and 21,736 differentially methylated sites, respectively), and these were 81.3 and 80.8% hypomethylated in the malignant tissue samples, respectively (Fig. 2A). The next largest differences in methylation patterns were between primary and metastatic ACC samples compared with benign tissue samples (13,727 sites and 11,849 differentially methylated sites, respectively), and these sites were 64.3 and 64.6% hypomethylated in the malignant tissue samples, respectively. In contrast, primary ACC samples compared with metastatic samples had only 3799 differentially methylated sites, and these were 59.0% hypermethylated in primary ACC samples (Fig. 2B).

Fig. 2.

Differential methylation across all tissue comparisons. A, Normal compared with benign tissue samples has the least number of differences in methylation, and they are predominantly hypermethylated (104 total, 34 hypomethylated, and 70 hypermethylated). Primary and metastatic ACC samples compared with normal tissue samples have the greatest number of differences in methylation, and they are predominantly hypomethylated (primary ACC vs. normal, 24,229 total, 19,689 hypomethylated 4,540 hypermethylated, and metastatic vs. normal samples 21,736 total, 17,569 hypomethylated, and 4,167 hypermethylated). B, Primary and metastatic ACC samples compared with benign tissue samples have the next largest differences in methylation, and they are also predominantly hypomethylated (primary ACC vs. benign, 13,727 total, 8,824 hypomethylated, and 4,903 hypermethylated and metastatic vs. benign, 11,849 total, 7,650 hypomethylated, and 4,199 hypermethylated). Primary ACC samples compared with metastatic samples, however, have only 3799 differentially methylated sites, and they are predominantly hypermethylated (1556 hypomethylated and 2243 hypermethylated).

Using principal component analysis to determine the global methylation patterns of 19 normal, 47 benign, eight primary malignant, and 12 metastatic tissue samples and ANOVA with an adjusted P ≤ 0.01, normal and benign adrenocortical tissue samples clustered more closely with less variation. Primary malignant and metastatic tissue samples cluster separately from the normal and benign samples and each other and had greater variability between each sample (Fig. 3).

Fig. 3.

Principal component analysis of normal, benign, primary malignant, and metastatic tissue samples. Using ANOVA and an adjusted value of P ≤ 0.01, 44.5% of differentially methylated genes can separate the four tissue categories. Normal and benign tissue samples cluster more closely and have less variation across the individual samples. Primary malignant and metastatic tissue samples cluster separately from the normal and benign samples and each other and have more variability across each sample.

Analysis of the methylation pattern of benign adrenocortical tumor samples by functional status (cortisol secreting, aldosterone secreting, and nonfunctioning) showed different methylation patterns. Aldosterone-secreting tumor samples compared with nonfunctioning samples had 397 differentially methylated CpG sites (98 hypermethylated, 299 hypomethylated sites). Only eighteen differentially methylated CpG sites were found between cortisol-secreting tumor samples compared with aldosterone-secreting tumor samples, and all of these sites were hypermethylated. We found no significant differences in the methylation pattern between cortisol secreting tumors and nonfunctioning tumor samples.

Methylation distribution and classification analysis

Methylated cytosines can be in CpG islands, shores, shelves, open sea, and sites surrounding transcription sites [−200 to −1500 bp, 5′ untranslated region (UTR), and exons 1] for coding genes as well as gene bodies and 3′UTR and other/open sea regions derived from genome-wide association studies (28, 29). Shores are considered regions 0–2 kb from CpG islands, shelves are regions 2–4 kb from CpG islands, and other/open sea regions are isolated CpG sites in the genome that do not have a specific designation. When comparing the different tissue groups, the benign samples compared with normal tissue samples had the lowest number of methylation CpG probe site differences, but these differences showed the highest percentage of probe sites in other/open sea regions (88%) and the lowest in the island (2%), shore (4%), or shelf (6%) regions. Primary ACC compared with normal, benign, and metastatic tissue samples had a range of 49–61% CpG methylation differences in the other/open sea regions, 13–19% of probes in the island regions, 16–23% in the shores, and 8–10% in the shelves. Similarly, metastatic tissue samples compared with normal and benign samples had a 53–63% number of differentially methylated CpG sites in the intergenic regions, 15–22% within CpG islands, 14–18% in the shores, and 7–8% in the shelves (Fig. 4A).

Fig. 4.

A, Percentage differences of CpG methylation probe location and functional genomic differences between normal, benign, primary, and metastatic malignant tissue samples. CpG locations in island, shore, shelf, and other. B, Methylation differences by functional genomics (promoter, body, 3′UTR, and intergenic regions). C, Methylation differences by RNA coding and noncoding regions. Normal and benign tissue samples have the least number of differentially methylated sites, but these sites are predominantly in the other/open sea and promoter regions of the genome as well as RNA coding regions.

The CpG sites were then separated based on hypermethylated or hypomethylated status. Although benign tissue samples compared with normal tissue have the lowest total number of methylation differences, it occurs in100% of hypermethylated sites and 65% of hypomethylated sites in other/open sea regions. Hypermethylated CpG sites in primary and metastatic ACC samples compared with both normal and benign tissue were predominantly in islands (42–39% for primary and 52% for metastatic ACC tissues), but the hypomethylated sites in these two comparisons were predominantly in other/open sea regions (70% for primary, 75% for metastatic ACC tissues). In addition, primary ACC compared with the metastatic samples had hypermethylated sites predominantly in the other/open sea region (65%), but the hypomethylated sites were predominantly in shores (34%), islands (31%), and shelves (28%) (Supplemental Fig. 1A, published on The Endocrine Society's Journals Online web site at http://jcem.endojournals.org).

Functional genomic distribution also varied across the tissue comparisons. Although benign and normal tissue samples had the lowest number of methylation probe site differences, the difference in distribution of the CpG sites was the greatest. The differential methylation of probes that distinguished benign from normal tissue samples was highest in the promoter regions of genes (70%) and lowest in the body (23%), 3′UTR (1%) and intergenic regions (7%) compared with other tissue group comparisons. Primary ACC compared with normal, benign, and metastatic tissue samples had a range of 26–29% of differentially methylated probes in promoter regions, 31–36% in the body, 3–4% in the 3′UTR regions, and 31–47% in intergenic regions. Metastatic compared with normal and benign tissue samples had 25–26% of differentially methylated probes in promoter regions, 30–32% in the body, 3% in the 3′UTR region, and 39–42% in intergenic regions (Fig. 4B).

When the CpG sites were then separated based on hypermethylated or hypomethylated status, benign compared with normal samples had 88% of hypermethylated sites localized in promoter regions and 50% of hypomethylated sites in body regions. Primary and metastatic ACC samples compared with normal and benign tissues had most of the hypermethylated sites in promoter (36–35% primary and 39–37% metastatic respectively), body (33% primary, 31–34% metastatic, respectively) and intergenic regions (28% primary, 27–26% metastatic, respectively); however, the hypomethylated sites were predominantly in intergenic regions (43–42% primary, 46% metastatic, respectively). Primary compared with metastatic ACC samples, however, had hypermethylated sites that were 38% in intergenic regions and 41% of hypomethylated sites in gene bodies (Supplemental Fig. 1B).

RNA coding content compared with other genome content of the differentially methylated probe sites in the comparisons again showed that although the total number of differentially methylated sites is lowest in benign compared with normal samples, these show the highest percentage distribution in the RNA category (93%), compared with other genomic content (7%). Primary ACC tissue samples compared with normal, benign, and metastatic tissue samples had 59–67% of the differentially methylated sites in RNA coding regions of the genome and 33–41% in other regions. Metastatic compared with normal and benign tissue samples had 56–59% (respectively) differentially methylated probes in RNA coding regions and 41–44% (respectively) in other regions of the genome (Fig. 4C).

Separating the methylation differences in the tissue comparisons based on hypermethylation or hypomethylation revealed that in all tissue comparisons, regardless of methylation status, the predominance of CpG methylation differences were in RNA coding regions. The greatest number of both hyper- and hypomethylated sites in the RNA coding regions were in the benign compared with normal tissue comparison (hypermethylated 97%, hypomethylated 85%). The next largest number of hypermethylated sites in RNA coding regions was in the primary and metastatic ACC compared with normal and benign tissue comparisons (70% primary, 71–72% metastatic, respectively). The hypomethylated sites in primary and metastatic ACC compared with normal and benign tissue samples were 56–57% primary and 52% metastatic, respectively. Primary ACC compared with metastatic tissues had hypermethylated sites (60%) and hypomethylated sites (76%) in RNA coding regions (Supplemental Fig. 1C).

Unsupervised cluster analysis of primary malignant and benign tissue samples

Because it is sometimes difficult to distinguish between primary malignant and benign adrenocortical neoplasms, we performed an unsupervised hierarchical cluster analysis between these two groups, using data from ANOVA analysis and cutoff values of P ≤ 0.01 and Δβ of −0.20 or less or Δβ of 0.20 or greater. The primary ACC and benign tissue samples almost completely clustered separately based on their methylation differences with the exception of one ACC sample, which consistently clustered with benign tissue samples, and one benign tissue sample, which had extensive necrosis (Supplemental Fig. 2).

Differential methylated regions of chromosomes in primary ACC and benign adrenocortical tumors

Comparative genomic hybrization studies of adrenocortical tumors have previously shown frequent chromosomal gains and losses comparing ACC and benign tumors (7, 30). Thus, we analyzed the chromosomal regions that showed differentially methylation between benign and primary ACC tumor samples using cutoff values of P ≤ 0.01 and Δβ of −0.20 or less or Δβ of 0.20 or greater. We found clusters of differentially methylated sites in select chromosomal regions: hypermethylated regions in primary ACC samples in chromosomes 1q, 2p, 2q, 5p, 6p, 7p, 7q, 10q, 11p, 12p, 12q, 18q, and 21q and hypomethylated regions in chromosome 5q and 12q (Supplemental Fig. 3).

Genes differentially methylated between primary malignant and benign adrenocortical samples

Given the difficulty of using histopathology in distinguishing benign and malignant samples, we found that at the epigenetic level, there were several differentially methylated probe sites that separated these two categories. With the exception of one ACC sample that clustered with benign tissue samples, all ACC samples had robust differential hypermethylation of probe sites for KCTD12 sites (mean difference across three sites of 2.65, Δβ = 0.35), KIRREL (mean difference 2.14, Δβ = 0.34), SYNGR1 (mean difference, 1.65, Δβ = 0.27), and NTNG2 (mean difference 2.50, Δβ = 0.38). Furthermore, other genes implicated in the pathogenesis of ACC such as GATA6 had probe sites that were significantly hypermethylated (mean difference across three sites of 1.99, Δβ = 0.32). And TP53, and β-catenin (CTNNB1), each had one site that was hypomethylated (mean difference −1.44, Δβ = 0.22 and −1.49, Δβ = 0.21, respectively).

In addition, several hypermethylated CpG sites include those associated with imprinted genes of the chromosome 11p15 locus, including IGF2 and H19 (mean ACC mean benign difference of 1.54, Δβ = 0.24 and 1.24, Δβ = 0.20, respectively). In addition, other genes associated with the IGF-II signaling pathway also had some probe sites that were hypermethylated such as CpG sites associated with IGF1R (mean difference across five probes 1.88, Δβ = 0.28) and AKT1 (mean difference across four probes, 1.80, Δβ = 0.25). RARRES2 and SLC16A9, two genes previously found to be robust diagnostic markers for ACC and underexpressed in ACC, were also found to have probe sites that were differentially methylated (13). RARRES2 had hypermethylated CpG sites (mean difference across five probes, 1.75, Δβ = 0.27), and SLC16A9 had hypermethylated sites (mean difference across two sites, 1.70, Δβ = 0.26).

Integrated analysis of genome-wide mRNA expression and methylation profile in primary malignant and benign adrenocortical tissue samples

In addition to methylation profiling of primary ACC and benign samples, genome-wide mRNA gene expression profiling was performed in a subset of samples (five ACC and 74 benign samples). We found 773 probes were differentially expressed (2-fold expression change and adjusted P ≤ 0.05). Of these, 215 were down-regulated genes and 558 were up-regulated. Of the down-regulated genes, 52 were also hypermethylated (ANOVA adjusted P ≤ 0.01 and Δβ ≤ −0.20 or Δβ ≥ 0.20). Two of these genes, RARRES2 and SLC16A9, were not only both significantly hypermethylated but also down-regulated in primary ACC samples (−11.98, adjusted P < 0.001, and −12.32, P < 0.001, respectively). GATA6 also had several hypermethylated sites and was down-regulated in gene expression (−3.64, adjusted P < 0.001).

When the 52 hypermethylated and down-regulated genes were analyzed using the Ingenuity Systems pathway analysis software, these genes were present in five biological function networks (Table 2). The network that had the greatest number of genes was the drug metabolism, endocrine system development and function, lipid metabolism biological function pathway, and it contained 16 genes, which were both hypermethylated and down-regulated including RARRES2 and GATA6. The next largest biological function pathway was the lipid metabolism, small molecule biochemistry, cell cycle network, which had 12 hypermethylated and down-regulated genes.

Table 2.

Biological function pathway analysis networks of hypermethylated and down-regulated genes

Biological function pathway	Score (number of other genes/hypermethylated and down-regulated genes from data sets)	Hypermethylated and downregulated genes (adjusted P < 0.01, β = 0.2; ≥ 2-fold, adjusted P < 0.05)
Drug metabolism, endocrine system development and function, lipid metabolism	38/16	ABCA1, CD55, CD74, COL4A3, GOS2, GATA6, HSD3B2, KCNQ1, MAP3K5, NCOA7, RAPGEF4, RARRES2, S100A6, SPTBN1, TNFSF13, TNS1
Lipid metabolism, small molecule biochemistry, cell cycle	27/12	ADCK3, ALDH3B1, CSDC2, CYP7B1, GIPC2, HOOK1, MEIS1, MLH3, MRPL33, NME5, RGNEF, TCIRG1
Lipid metabolism, small molecule biochemistry, energy production	21/10	AMPD3, B4GALT6, CAB39L, CD55, GYPC, NDRG4, RAB34, RBPMS, SEMA6A, TNFS1F2-TNFSF13
Cell-to-cell signaling and interaction, cellular assembly and organization, nervous system development and function	3/1	SLC16A9
Hematological disease, immunological disease, infectious disease	2/1	PHF11

Discussion

Our understanding of the molecular events involved in adrenocortical carcinogenesis is limited, and there have not been studies evaluating the methylome of human adrenocortical tissue samples. Thus, we performed methylation profiling of human adrenocortical tissue samples to determine whether epigenetic differences exist in adrenocortical neoplasms and whether this approach could be used for molecular classification as well as to explain the aberrant expression of genes found to be dysregulated in ACC. We found that primary and metastatic ACC samples have a global hypomethylation pattern compared with normal and benign adrenocortical tissue samples. Moreover, select CpG site methylation status could reliably distinguish between the primary ACC and benign adrenocortical tissue samples. The methylation pattern of normal and benign tissue samples showed strong similarities and homogeneity across the samples, whereas the primary and metastatic tissue samples showed greater variation. There was also a significant difference in the methylation status of benign tumors by the functional status. We also found that clusters of differential methylation patterns in primary ACC and benign tumors samples were located in select chromosomal regions. In addition, we found that a subset of genes differentially expressed in primary ACC vs. benign adrenocortical tumors show an association with the methylation status, suggesting that epigenetic regulation of these gene may be an important mechanism.

Although the sequential progression from normal to benign to malignant adrenocortical lesions, as has been demonstrated in colorectal carcinoma, is not well established, we did observe a greater number of differential methylation from normal to benign to primary ACC to metastatic ACC. On the other hand, although benign samples had the lowest number of methylation CpG probe site differences compared with normal tissue samples, these differences showed the highest percentage of both total and hypermethylated and hypomethylated probe sites in other/open sea regions and the lowest in the island, shore, or shelf regions. Functional genomic distribution also varied across the tissue comparisons, and benign compared with normal tissue samples had the highest percentage of total and hypermethylated probes sites in the promoter regions of genes and lowest total methylated sites in the body, 3′UTR, and intergenic regions compared with other tissue group comparisons. Hypomethylated sites in benign compared with normal tissue samples were highest in body regions, and this was also the highest of all tissue comparisons. Furthermore, the benign compared with normal tissue comparison also showed the highest percentage distribution of total, hypermethylated, and hypomethylated sites in the RNA coding category compared with other genomic content. This suggests that although the methylation differences may be small, they may be in more biologically relevant regions of the genome.

When comparing the methylome of primary malignant and benign adrenocortical tissue samples, a clear signature of differentially methylated probe sites emerged that could classify these two categories. Therefore, determination of the methylation difference in certain probe sites in adrenocortical tumors may be a useful diagnostic adjunct to histopathology for localized primary ACC. Of the differentially methylated genes in primary ACC compared with benign tissue samples, several CpG sites were differentially methylated including those associated with KCTD12, KIRREL, SYNGR1, and NTNG2 and those in the chromosome 11p15 imprinted region including IGF2 and H19. In addition, other sites were in the IGF-II pathway, including, IGF1R that IGF-II binds to and AKT1, a downstream signaling molecule in the cell survival pathway (phosphatidylinositol 3-kinase-AKT-mammalian target of rapamycin) of IGF1R. Furthermore, we found other genes associated with ACC such as TP53 and CTNBB1. Both TP53 and CTNBB1 had hypomethylated sites. Finally, RARRES2 and SLC16A9, two previously described genes that are underexpressed in ACC, also had several hypermethylated sites in ACC tissue samples.

In the integrated genome-wide gene expression and methylation analysis of genes that were both down-regulated and hypermethylated, the biological function pathway network analysis revealed that the majority of the genes fell into the drug metabolism, endocrine system development and function, lipid metabolism network.

The mechanism of methylation alterations is an active area of investigation and may be related to genomic changes among many other factors (31). Although the goal of our study was not to determine the underlying mechanism of methylation alterations, we did find clusters of differentially methylated regions on chromosomes, some of which were in regions previously reported to have frequent genomic gains or losses (1q, 5p, 5q, 6p, 7p, 7q, 12q) in ACC tumor samples (30, 32, 33).

In conclusion, we found distinctly different methylation profiles in human adrenocortical tissue samples between normal, benign, primary malignant, and metastatic categories. Moreover, methylation profile differences may be able to classify primary ACC and benign adrenocortical tissue samples. Several of the differentially methylated sites were associated with genes known to be dysregulated or altered in ACC. Future gene-specific studies are needed to determine whether the methylation changes we identified regulate ACC initiation and progression and whether hypermethylated and down-regulated tumor suppressor genes could be effectively targeted by using demethylating agents.