Identification and validation of highly frequent CpG island hypermethylation in colorectal adenomas and carcinomas

Identification and validation of highly frequent CpG islandhypermethylation in colorectal adenomas and carcinomas

Bodil Øster1, Kasper Thorsen1, Philippe Lamy1,2, Tomasz K. Wojdacz3, Lise Lotte Hansen3, Karin Birkenkamp-Demtroder1,

Karina D. Sørensen1, Søren Laurberg4, Torben F. Ørntoft1 and Claus L. Andersen1

1 Department of Molecular Medicine, Aarhus University Hospital, Skejby, Denmark2 Bioinformatics Research Center (BiRC), Aarhus University, Aarhus, Denmark3 Institute of Human Genetics, Aarhus University, Aarhus, Denmark4 Department of Surgery P, Aarhus University Hospital, Aarhus, Denmark

In our study, whole-genome methylation arrays were applied to identify novel genes with tumor specific DNA methylation of

promoter CpG islands in pre-malignant and malignant colorectal lesions. Using a combination of Illumina

HumanMethylation27 beadchips, Methylation-Sensitive High Resolution Melting (MS-HRM) analysis, and Exon arrays

(Affymetrix) the DNA methylation pattern of ~14,000 genes and their transcript levels were investigated in six normal

mucosas, six adenomas and 30 MSI and MSS carcinomas. Sixty eight genes with tumor-specific hypermethylation were

identified (p < 0.005). Identified hypermethylated sites were validated in an independent sample set of eight normal

mucosas, 12 adenomas, 40 MSS and nine MSI cancer samples. The methylation patterns of 15 selected genes,

hypermethylated in adenomas and carcinomas (FLI1, ST6GALNAC5, TWIST1, ADHFE1, JAM2, IRF4, CNRIP1, NRG1 and EYA4), in

carcinomas only (ABHD9, AOX1 and RERG), or in MSI but not MSS carcinomas (RAMP2, DSC3 and MLH1) were validated using

MS-HRM. Four of these genes (MLH1, AOX1, EYA4 and TWIST1) had previously been reported to be hypermethylated in CRC.

Eleven genes, not previously known to be affected by CRC specific hypermethylation, were identified and validated. Inverse

correlation to gene expression was observed for six of the 15 genes with Spearman correlation coefficients ranging from

20.39 to 20.60. For six of these genes the altered methylation patterns had a profound transcriptional association,

indicating that methylation of these genes may play a direct regulatory role. The hypermethylation changes often occurred

already in adenomas, indicating that they may be used as biomarkers for early detection of CRC.

Epigenetic abnormalities, including global changes in DNAmethylation, have been observed in many types of cancer,including colorectal cancer (CRC). Aberrant methylation in

the colon can be observed already in early premalignantlesions as well as in tumor-adjacent normal-appearing mu-cosa, rendering DNA methylation attractive as a molecularmarker for early detection. A considerable fraction (70–80%)of the sporadic colorectal cancers with microsatellite instabil-ity (MSI) have increased promoter methylation of certaingenes, a phenomenon termed CpG island methylator pheno-type (CIMP), that defines a distinctive subset of colorectalcancers.1,2 Altogether, promoter hypermethylation of a widerange of genes seems to be a general phenomenon in alltypes of CRC, and it is thought that, while global hypome-thylation may induce genomic instability, localized hyper-methylation may promote tumorigenesis through silencing oftumor suppressor genes.

Although the triggering mechanisms behind aberrant DNAmethylation in cancer development are only poorly under-stood, hypermethylation of CpG sites has been correlated toendpoints like recurrence,3–5 progression,6 and chemores-ponse,7,8 which further substantiates the importance of an in-depth understanding of this epigenetic modification in cancer.

Much of what is known about the importance of DNAmethylation in cancer was gained through small- and moder-ate-scale analysis of gene promoters. Hence, with the recentadvances in genome characterization technologies that enableelucidation of DNA methylation levels of the whole genome,

Key words: DNA methylation, genome-wide, colorectal cancer, gene

expression, microarray

Abbreviations: CIMP: CpG island methylator phenotype; CRC:

colorectal cancer; HNPCC: hereditary nonpolyposis colon cancer;

IHC: immunohistochemistry; MS-HRM: methylation sensitive high

resolution melting; MSI: microsatellite instability; MSS: microsatellite

stability; qRT-PCR: quantitative real-time RT–PCR; TMA: tissue

microarray; UBC: ubiquitin C

Additional Supporting Information may be found in the online

version of this article.

Grant sponsors: John and Birthe Meyer Foundation, Danish

Council for Independent Research Medical Sciences, Lundbeck

Foundation, EC project GENICA, Danish Ministry of the Interior

and Health

DOI: 10.1002/ijc.25951

History: Received 1 Oct 2010; Accepted 27 Dec 2010; Online 11

Mar 2011

Correspondence to: Torben F. ørntoft, Department of Molecular

Medicine, Aarhus University Hospital, Skejby, Brendstrupgaardsvej

100, DK-8200 Aarhus N, Denmark, Fax: þ45-8678-2108,, E-mail:

[email protected]

Early

Detection

andDiagn

osis

Int. J. Cancer: 129, 2855–2866 (2011) VC 2011 UICC

International Journal of Cancer

IJC

an improved understanding of the role of DNA methylationin cancer will likely be gained.

For validation of the methylation candidates identified ingenome-wide studies and for their translation into the clinicalsetting, labor- and cost-efficient technologies are needed toallow high-throughput assessment of single-locus methylationchanges. Methylation-Sensitive High Resolution Melting(MS-HRM) has recently emerged as a promising method forthis purpose, and was employed in our study.

In our study CpG island methylation was obtained usingInfinium HumanMethylation27 BeadChips, and gene expres-sion was investigated to identify differential methylation pat-terns between noncancer colorectal tissue and colorectaladenomas or adenocarcinomas. This enabled us to identifygenes not previously associated with aberrant methylation inCRC and to relate the methylation levels of these genes tothe corresponding RNA expression levels.

Material and MethodsTumor samples and controls

Fresh frozen tissue samples (Table 1) were obtained frompatients diagnosed with colorectal adenomas or adenocarcino-mas. Inclusion criteria were no other cancer than CRC, no indi-cations of heredity, and no radio- or chemotherapy before sur-gical resection. Normal colon tissues were obtained frompatients who had colorectal surgery for other reasons than CRC(9 diverticulosis, 2 constipation, 1 appendectomy, 1 lipoma and1 rectal prolapse). Informed written consent was obtained fromall patients, and research protocols were approved by the Cen-tral Denmark Region Committees on Biomedical Research

Ethics. All tissue samples were collected immediately after sur-gery and snap frozen in Tissue-TekVR O.C.T. (Optimal CuttingTemperature) Compound (Sakura Finetek, Vaerloese, Den-mark), and stored at minus 80�C.

Analysis of DNA methylation

Genomic DNA from serial cryosections was extracted usingPuregene DNA purification kit (Gentra Systems, Plymouth,MN). Tumor cell percentage was determined individually bytwo trained persons. When necessary, tumor biopsies were mac-roscopically trimmed to enrich the fraction of neoplastic cells toa minimum of 60% before DNA isolation. Median cancer cellpercentage was 80%. One microgram of DNA was bisulfitemodified using EpiTect Bisulfite Kit (Qiagen, Copenhagen, Den-mark) for MS-HRM, and using EZ-96 DNA Methylation D5004(Zymo Research, Orange, CA) for microarrays and bisulfitesequencing. Microsatellite instability status was determinedusing a pentaplex polymerase chain reaction with five quasimo-nomorphic mononucleotide repeats, as previously described.9

Bisulfite sequencing. Bisulfite modified DNA was amplifiedusing primers designed with MethPrimer (http://www.uroge-ne.org/methprimer/index1.html) (Supporting Information Ta-ble 1) and TEMPase DNA Polymerase (Ampliqon, Skov-lunde, Denmark). PCR products were gel-purified and clonedusing TOPO TA Cloning VR Kit for Sequencing (Invitrogen,-Taastrup, Denmark). PCR amplification for sequencing wasperformed directly on the colonies with M13 primers (DNATechnology, Risskov, Denmark) and TEMPase DNA Poly-merase (Ampliqon). For each gene, 10 clones were randomlyselected and sequenced using BigDye terminator cycle

Table 1. Patient characteristics of the two populations

Screening set (array) Validation set (MS-HRM)

Normal Adenoma MSS MSI Normal Adenoma MSS MSI

Male/Female 4/2 4/2 15/9 1/5 3/5 5/7 13/27 1/8

Mean age, years 55 (46–71) 61 (52–73) 71 (55–85) 72 (64–80) 60 (39–76) 65 (49–79) 70 (30–89) 74 (59–86)

Location

Right 1 1 7 6 0 5 10 7

Left 4 4 11 0 6 4 13 1

Rectum 0 1 6 0 2 3 17 1

N/A 1 – – – – – – –

Differentiation

Well – – 3 0 – – 4 0

Moderate – 5 17 4 – 1 28 5

Poor – 1 3 2 – 11 8 4

N/A – – 1 – – – – –

Stage

I - – 0 0 – – 10 0

II – – 12 4 – - 10 8

III – – 12 2 – – 10 0

IV – – 0 0 – – 10 1

Early

Detection

andDiagn

osis

2856 CpG island hypermethylation in colorectal cancer


sequencing kit and a 3130xl Genetic Analyzer (Applied Bio-systems, Foster City, CA).

Whole genome methylation analysis. Bisulfite modified DNAwas whole genome amplified and hybridized to InfiniumHumanMethylation27 BeadChips (Illumina, San Diego, CA)overnight as described by the manufacturer. BeadChips werescanned with a BeadXpress Reader instrument (Illumina) anddata analyzed using Bead Studio Methylation ModuleSoftware (Illumina). Methylation levels were provided in betavalues, with a beta value of 0 corresponding to no methyla-tion, and 1 corresponding to full methylation. The screeningset for whole genome methylation analysis consisted ofcolorectal tissue from six noncancer patients, six adenomas(all MSS), six MSI cancers (stage II and III), 12 MSS stage IIcancers and 12 MSS stage III cancers.

Methylation-sensitive high resolution melting analysis (MS-

HRM). The validation set consisted of tissue from 8 non-cancer patients, 12 adenomas (all MSS), nine MSI cancers, 10stage I, 10 stage II, 10 stage III and 10 stage IV MSS cancers.Amplification of bisulfite modified DNA was performed intriplicates with primers designed according to guidelines pub-lished by Wojdacz et al.10 (Supporting Information Table 2).PCR conditions were 15 ng bisulfite modified DNA, 4.8 llLightCyclerVR 480 HRM Master Mix (Roche, Hvidovre, Den-mark), 3 mM MgCl2, 0.5 nM of primer F and primer Reach, in a total volume of 10 ll. Standard curves from bisul-

fite modified templates were prepared by mixing 100% meth-ylated (CpGenomeTM Universal Methylated DNA, Millipore,Copenhagen, Denmark) in a background of unmethylatedDNA (periferal blood DNA). The standard curve rangedfrom 0% methylated, through 5, 25, 50, 75 to 100% methyl-ated bisulfite converted DNA. Standard curves and no tem-plate controls were included in each experimental run. As anegative control, genomic unmodified DNA from a pool ofeight healthy individuals was tested once with each primerpair. The PCR reaction and high resolution melting was per-formed essentially as described in10 using a LightScannerinstrument (Idaho Technology, Salt Lake City, Utah). MS-HRM data were normalized with the LightScannerVR Instru-ment & Analysis Software to compensate for varying startingfluorescence levels. Based on the standard curves, patientdata was classified into different methylation categories bytwo independent observers. The K coefficient was calculatedto evaluate interobserver agreement of the scored methylationlevels for all 15 analyzed genes. The coefficients ranged from0.91 to 1.0, indicating very good to excellent agreement.

Gene expression analysis

RNA was extracted from serial cryosections using RNeasyMini Kit (Qiagen). Macroscopical trimming was performedwhen necessary to obtain a tumor percentage of at least 60%.Median cancer cell percentage was 80%. For gene expressionanalysis, 100 ng of total RNA, from the 42 patients used for

Table 2. Summary of hypermethylated genes selected for validation

Gene symbol Microarray CpG ID Gene name Methylation fold change

Group 1 Normal versus adenoma and cancer

FLI1 Cg17872757 Friend leukemia virus integration 1 26.1

ST6GALNAC5 Cg13823136 ST6 (alpha-N-acetyl-neuraminyl-2,3-beta-galactosyl-1,3)-N-acetylgalactosaminide alpha-2,6-sialyltransferase 5

6.7

TWIST1 Cg26312150 Twist homolog 1 3.8

ADHFE1 Cg08090772 Alcohol dehydrogenase, iron containing, 1 7.1

JAM2 Cg03382304 Junctional adhesion molecule 2 5.3

IRF4 Cg12741420 Interferon regulatory factor 4 6.9

CNRIP1 Cg07080358 Cannabinoid receptor interacting protein 1 4,1

NRG1 Cg17457560 Neuregulin 1 3.4

EYA4 Cg20286200 Eyes absent homolog 4 13.6

Group 2 MSS vs. MSI

DSC3 Cg11832722 Desmocollin 3 6.01

MLH1 Cg13846866 MutL homolog 1 68.11

RAMP2 Cg26990660 Receptor (G protein-coupled) activity modifying protein 2 9.51

Group 3 Normal vs. adenoma vs. cancer

AOX1 Cg02144933 Aldehyde oxidase 1 9.5

RERG Cg03028472 RAS-like, estrogen-regulated, growth inhibitor 5.7

ABHD9 Cg05488632 Abhydrolase domain containing 9 2.5

Fold change is calculated as mean microarray methylation value for MSS cancers relative to normal. The MS-HRM assay for MLH1 does not includethe Illumina CpG site.1Fold change is calculated as mean microarray methylation value for MSS cancers relative to MSI cancers.

Early

Detection

andDiagn

osis

Øster et al. 2857


whole genome methylation analysis, was labeled using Gene-Chip Whole Transcript (WT) Sense Target Labeling Assay(Affymetrix, Santa Clara, CA) and hybridized overnight toHuman Exon 1.0 ST Arrays (Affymetrix) according to themanufacturer’s instructions. Arrays were scanned in an Affy-metrix GCS 3000 7G scanner. To avoid batch effects, all sam-ples were labeled and scanned in random order. Data analysiswas performed using GeneSpring GX 10 software (AgilentTechnologies, Naerum, Denmark). Samples were quantile-nor-malized using iterPLIER16 with transcript level core (17881transcripts) and by using antigenomic background probes.

Quantitative RT-PCR

Quantitative real-time RT–PCR (qRT–PCR) was performed intriplicates on a ABI 7500 Fast Real Time System (Applied Bio-systems) using the relevant (TaqMan or SYBR Green) MasterMix (Applied Biosystems). Predesigned TaqMan assays tar-geting all known isoforms of JAM2 (Hs00221894_m1)and ADHFE1 (Hs00329084_m1) were obtained from AppliedBiosystems. For normalization, the gene Ubiquitin C (UBC)was employed. The suitability of UBC as a normalization genefor analysis of normal mucosa and CRC specimen sample setsand the UBC primer sequences have been published earlier.11

Sequencing

After purification using QIAquick PCR Purification Kit(Qiagen), sequencing of selected samples was performedusing BigDye terminator cycle sequencing kit and a 3130xlGenetic Analyzer (Applied Biosystems) and primers used forMS-HRM analysis.

Immunohistochemistry

Immunohistochemistry (IHC) was performed essentially asdescribed previously.12 In brief, sections were deparaffinizedand rehydrated. Endogenous peroxidase activity was blocked,and antigen retrieval was performed by heating. Nonspecificbinding was blocked, and tissue sections were stained with po-lyclonal rabbit anti-NRG1 (1:120, ab2994, Abcam, Cambridge,UK) and with EnVision anti-Rabbit (EnVisionþ System, HRPLabeled Polymer Anti-Rabbit, K4003, Dako Denmark A/S).Finally, sections were incubated with chromogen and counter-stained in haematoxylin before mounting for microscopy. Aprotein band of � 44 kDa was obtained by western blottingwith anti-NRG1 antibody and tumor tissue from patients withhigh NRG1 mRNA expression, confirming the specificity of theantibody (data not shown).

Scoring of NRG1 IHC staining

A tissue micro array (TMA) consisting of 48 tumor coresand 42 normal colon tissue cores was scored independentlyby two observers (B.O. and C.L.A.). The presence of cytoplas-mic and/or membranous NRG1 was evaluated in epithelialcells of the normal mucosa and in cancer cells. The intensityof the staining was scored in four categories; negative, weak,moderate or strong staining. Lost specimens or specimens

that did not contain epithelial cells or tumor cells wereexcluded from evaluation. A K coefficient of 0.64 indicatedsubstantial agreement between the observer’s scoring of TMAcore IHC staining. Scoring by two independent observers wasfollowed by consensus scoring.

Statistical analysis

A Mann-Whitney U test was used to assess for each CpG,whether two groups had the same distribution of methyla-tion. To eliminate probes that did not work and other diverg-ing values, beta values in each group were trimmed forextremes before the calculation of the average beta value.Next, only genes with the highest 5% tail of the absolute av-erage beta value differences between the two groups wereselected. Thus CpGs with a p-value below 0.005 and with thehighest beta value differences between the two groups wereused for identification of genes with statistically significantcancer-specific changes in methylation. Comparison of arraymethylation results with MS-HRM results was performedwith a Mann-Whitney U test and a Chi2 test for trend.Spearman coefficients were calculated to assess the correla-tion between methylation and gene expression.

ResultsAnalysis of whole genome methylation

Genome-wide CpG island methylation profiles of sixnonCRC patients (normal), six adenomas (all MSS), six MSI(stage II and III cancers) and 24 MSS (stage II and III) can-cers (Table 1) were generated using Illumina InfiniumHumanMethylation27 arrays. From these profiles a total of95 genes were identified that, compared to normals, had sig-nificantly altered CpG island methylation patterns (p <

0.005) already in adenomas (MSS), which persisted in MSScarcinomas. Of these, 64 were hypermethylated, and the resthypomethylated (Supporting Information Table S4). Of the14,475 genes queried on the methylation array only fourgenes were identified, which showed significantly altered pro-moter methylation patterns in MSS carcinomas compared toMSS adenomas. These four genes were all hypermethylated.

Ninety-two genes with significant methylation differencesbetween MSS and MSI carcinomas were also identified (Sup-porting Information Table S4). For all 92 genes, MSI cancershad higher methylation levels than MSS cancers.

The general picture, when comparing normal colon tissuewith adenoma and cancer, and MSI with MSS cancer, wasmarked differences in methylation levels between groups, asexemplified for the genes selected for validation (Fig. 1a andTable 2). For several genes, the methylation levels observedin normal and cancer tissue did not show any overlap. Oneexample was CNRIP1 that was hypermethylated in six out ofsix adenomas and 24 out of 24 MSS cancers with methylationlevels ranging between beta values of 0.60 and 0.86, whereasbeta values in normal tissue ranged between of 0.09 and 0.34(Fig. 1b). Such genes could therefore be regarded as candidatebiomarkers for CRC detection.

Early

Detection

andDiagn

osis



Figure 1. DNA methylation array results for 15 hypermethylated genes in the screening set. N ¼ normal mucosa (n ¼ 6), A ¼ adenomas

(n ¼ 6), MSS ¼ MSS cancers (n ¼ 24), MSI ¼ MSI cancers (n ¼ 6), þ or Pos ¼ positive control, � or Neg ¼ Negative control. (a) Heatmaps

of 15 genes selected for validation. Blue ¼ unmethylated, yellow ¼ methylated, black ¼ intermediate methylation. (b) Nine genes with

statistical difference in methylation between normals and (adenomas þ cancer). The median methylation values are indicated with black

horizontal lines. (c) Three genes with statistical difference in methylation between MSS and MSI cancers. (d) Three genes with statistical

difference in methylation between normals and cancer, but not between normals and adenomas.

Of the 64 significantly (p < 0.005) hypermethylated genes inadenomas and MSS cancers, 15 genes (EYA4),13,14

CDKN2A,15,16 TFPI-2,17 ITGA4,18 ESR1,19 SLIT3,20 SOX17,21

LRFN5,22 SFRP1,23 BNC1,24 ZNF447,25 THBD,25 SIX6,26

UNC5C,27,28 and TWIST129 were previously reported to havealtered DNA methylation patterns in CRC. These were excludedfrom further analysis, except EYA4 and TWIST1, which wereincluded as positive controls for the validation procedure. EYA4is known for cancer-specific hypermethylation in esophagus30

and colon cancer, and TWIST1 is known for cancer-specific hy-permethylation in cervical neoplasia,31 breast,32 bladder,33

lung,34 gastric cancer35 and recently also in CRC.29 Seven genes,hypermethylated in adenomas and MSS cancers (p < 0.005)and not previously associated with methylation changes in colo-rectal cancer (FLI1, ST6GALNAC5, ADHFE1, JAM2, IRF4,CNRIP1 and NRG1), were selected for validation (Fig. 1b).

Among the 92 genes which were hypermethylated (p <

0.005) in MSI compared to MSS patients, several genes wereidentified that previously had been used as CpG island meth-ylator phenotype (CIMP) markers (SOCS1, RUNX3)36 as wellas LOX, a gene recently used to identify MSI-high CRCpatients.25 Three genes from this group were selected for vali-dation; RAMP2, DSC3 and MLH1 (Fig. 1c). The average dif-ference in beta values between MSI and MSS cancers were0.51 and 0.50 respectively for RAMP2 and DSC3 methylation,with only low levels of methylation in normal tissue andadenomas (Fig. 1c and Table 2). MLH1, a well-known markerfor DNA hypermethylation in MSI cancers, was included as acontrol to support the identification of novel methylationmarkers.

From the third group of genes, which had methylationpatterns in the adenomas that either resembled the methylationpattern of normal tissue or had a methylation pattern in-between normal and cancer, three genes were selected for vali-dation; AOX1, RERG and ABHD9 (Fig. 1d). ABHD9 and AOX1have previously shown cancer related methylation changes.25,37

Validation of infinium array technology

using bisulfite sequencing

To make a technical validation of the methylation resultsobtained, three genes were selected for bisulfite sequencing.Three patients with low, intermediate, or high methylationwere identified, and for each patient 10 clones were bisulfitesequenced. All sequenced fragments contained the CpG-siteinterrogated by the methylation array as well as several adja-cent CpG-sites. When looking specifically at the microarrayCpG-site, bisulfite sequencing supported the array data inall cases (Fig. 2 and Supporting Information Fig. S1). Asexpected, bisulfite sequencing of single clones revealed heter-ogeneous methylation in some clones and in some casesunmethylated molecules. We cannot exclude that the pres-ence of noncancer cells contributed to the small fraction ofunmethylated clones.

Validation of methylation array results using

methylation-sensitive high resolution melting (MS-HRM)

To confirm differential methylation in the generated groups,additional validation of the 15 selected genes was carried outusing MS-HRM. Initially, 9 patients from the array set wereanalyzed using MS-HRM. For all 15 genes and all 9 patients,MS-HRM data was in excellent concordance with the arrayresults (Supporting Information Table 3), supporting thatMS-HRM was an effective method for validation in an inde-pendent patient set.

Next, methylation status of the 15 genes was assessedusing MS-HRM on an independent sample set consisting ofeight non-CRC patients (normals), 12 adenomas (six tubularand six villous, all MSS), nine MSI (stage II and III cancers)and 40 MSS samples (stage I, II, III and IV cancers, 10 ofeach). Based on melting curve analysis, methylation levelswere scored in the intervals: 0–5, 6–25, 26–50, 51–75 and76–100%. The expected differences in methylation could bevalidated (p < 0.05) in all cases in the group with adenoma-and cancer-specific hypermethylation (Fig. 3a) and the MSSversus MSI group (Fig. 3b). P-values exceeding 0.05 wereonly observed in the third group of potential cancer-progres-sion genes (Fig. 3c). For all three genes in this category, dif-ferences in methylation between adenomas and MSS cancerswere not statistically significant (ABHD9 p ¼ 0.06, AOX1p ¼ 0.20 and RERG p ¼ 0.47). Nevertheless, the MS-HRMdata for ABHD9 and AOX1 (Fig. 3c) showed increasing levelsof methylation, first in the adenomas, and even more pro-nounced in the cancers. A chi-square test for trend confirmedthis observation (ABHD9 p ¼ 0.0006, AOX1 p ¼ 0.0005).Further, when looking into RERG methylation levels in thevalidation set, RERG resembled the majority of the hyperme-thylated genes in the screening set with aberrant methylationbeing prominent already in the adenomas.

The concordant results obtained by two dissimilar methodsfor methylation detection strongly supported our findings.The arrays determined the methylation status of one specificCpG site although the technology depends on homogeneousmethylation in those cases where multiple CpG sites are con-tained within the microarray probe. The MS-HRM methodrequires the presence of at least one CpG site to be able todistinguish the methylation differences, but a number of CpGsites is preferred as this eases the interpretation. In some sam-ples heterogeneous methylation complicated the interpretationof MS-HRM melting curves. Thus, selected samples fromgenes with heterogeneous methylation were sequenced toinquire the methylation status of the CpG site interrogated bythe array (data not shown). In all cases the methylation pat-terns obtained by the arrays could be confirmed. Interestingly,heterogeneous methylation was more frequently observed inadenomas than in cancer samples. This observation was sup-ported by the bisulfite sequencing results, which revealedmore heterogeneous methylation at the individual CpG siteswithin each clone in the adenoma samples compared to can-cer tissue (Fig. 2 and Supporting Information Fig. S1).

Early

Detection

andDiagn

osis



Correlation between methylation status

and gene expression analysis

The role of DNA methylation in silencing of gene expressionis well known. Therefore, we studied the effect of DNA hyper-methylation on gene expression in the screening set usingexpression microarrays. An inverse correlation between meth-ylation and gene expression was observed for six (IRF4,ADHFE1, JAM2, NRG1, FLI1 and MLH1) of the 15 selectedgenes with Spearman correlation coefficients from –0.39 to –0.60. The lacking inverse correlation between methylation andgene expression for some of the genes (DSC3, AOX1, ABHD9and RERG) may be explained by the fact that these geneswere expressed at very low levels in all samples, if expressedat all. For the remaining genes, intermediate to high levels ofexpression was maintained in the presence of cancer-specifichypermethylation (ST6GALNAC5, TWIST1, CNRIP1, EYA4and RAMP2). Lack of inverse correlation between hypermeth-

ylation and gene expression has previously been reported forTWIST1 in breast cancer,38 and recently in CRC.29

The inverse correlation between JAM2 and ADHFE1 geneexpression and methylation was validated with RT-qPCR ona sample set consisting of 19 samples from the validation set.Expression levels of the 19 samples from the validation setwere inversely correlated with methylation levels of the samesamples (Fig. 4a), thus, supporting the microarray data.

Protein expression of NRG1 determined by IHC

Our mRNA expression data indicated that NRG1 was silencedby hypermethylation in CRC tumors. To investigate if hyper-methylation was also associated with silencing at the proteinlevel we performed immunohistochemical analysis of NRG1protein expression in 48 carcinoma and 42 normal mucosasamples. NRG1 protein expression was significantly decreasedin tumor tissue compared to normal mucosa (p < 0.001,

Figure 2. Technical validation of FLI1 using bisulfite sequencing. Top: 10 clones from each of three individual patients, which had been

analyzed using methylation arrays were bisulfite sequenced. Black boxes represent methylated CpG-sites, white boxes represent

unmethylated CpG-sites. X ¼ not determined. The CpG-site interrogated by the array is marked with a rectangle. Methylation array data

are shown in the lower part of the figure. Arrows indicate the samples used for technical validation by bisulfite sequencing. N ¼ normal

mucosa, A ¼ adenomas, Cancer ¼ MSS and MSI cancers, þ ¼ positive control, � ¼ negative control. Bottom: FLI1 gene structure with CpG

island location and indication of analyzed region.

Early

Detection

andDiagn

osis

Øster et al. 2861


Figure 3. Validation of methylation levels of 15 hypermethylated genes in an independent validation set using MS-HRM. N ¼ normal mucosa

(n ¼ 8), A ¼ adenomas (n ¼ 12), MSS ¼ MSS cancers (n ¼ 40), and MSI ¼ MSI cancers (n ¼ 9). Methylation index 1 to 5 represents the

methylation intervals 0–5, 6–25, 26–50, 51–75 and 76–100% methylation. (a) Nine genes with statistical difference in methylation between

normals and (adenomas þ cancer). Array results for all nine genes could be validated using MS-HRM. (b) Three genes with statistical

difference in methylation between MSS and MSI cancers. For all three genes array results could be validated using MS-HRM. (c) Three genes

where adenomas had methylation levels that resembled the methylation levels observed in normal mucosa, whereas MSS cancers were

significantly hypermethylated. Array results could be validated for ABHD9 and AOX1 using MS-HRM. RERG methylation in adenoma samples

resembled methylation levels in MSS cancer samples. Statistical significant differences are marked with an asterisk (p < 0.05).

Fisher’s exact), consistent with the hypothesis that NRG1 issilenced by hypermethylation in CRC (Fig. 4b).

DiscussionIn our study, methylation arrays identified novel as well aspreviously described hypermethylated genes, 15 of whichwere validated in an independent sample set using MS-HRM.Cancer-specific changes in methylation patterns could dividethe hypermethylated genes into three separate groups, where

group 1 genes were hypermethylated in cancers as well asadenomas, group 2 genes were hypermethylated in MSI can-cers when compared to MSS cancers, and group 3 geneswere hypermethylated in cancers but not adenomas. Of thenine genes from group 1, seven had not previously beenassociated with methylation changes in colorectal cancer andmay hold the potential of being used as novel cancer bio-markers. As 20–25% of patients with CRC already have disse-minated disease at the time of diagnosis, there is a need for

Figure 4. Gene and protein expression. (a) Inverse correlation between methylation and gene expression results for ADHFE1 and JAM2. N ¼normal mucosa, A ¼ adenomas, Cancer ¼ MSS and MSI cancers. Methylation and gene expression results obtained using arrays are shown

to the left. Methylation levels determined using MS-HRM and gene expression levels using qPCR are shown to the right. (b) The protein

expression of NRG1 in matched normal colon mucosa (left) and tumor tissue (right) from two representative patients. Positive expression of

NRG1 is observed as a brown color in the cytoplasm.

Early

Detection

andDiagn

osis

Øster et al. 2863


moving the time of diagnosis towards less progressed stages.An important feature of biomarkers for cancer screening isthe ability to detect changes in the bowel at the earliest possi-ble stages, which is a characteristic of the group 1 typehypermethylated genes.

To rule out that the observed methylation differences wereage-associated, we compared the 15 genes selected for valida-tion with a recently published list of aging-associated hyperme-thylated genes.39 Indeed, 12 of the 15 genes was not associatedwith aging-associated hypermethylation, confirming that themethylation changes were cancer rather than age-related. Onlythree of the 15 genes; TWIST1, EYA4 and DSC3 were on thelist of hypermethylated genes associated with aging. TWIST1and EYA4 have previously been associated with cancer-specifichypermethylation in other studies.29–35 Further, DSC3 was onlymethylated in the MSI-cancers, a group, whose age-distributionmatched the unmethylated MSS cancers. Overlap between our15 selected hypermethylated genes and the aging-associationstudy was consistent with the fact that promoters associatedwith aging frequently are hypermethylated in cancers.

When investigating the potential functional impact of theidentified cancer specific methylation changes on gene expres-sion, 40% (6/15) of the selected genes showed methylation pat-terns that correlated inversely with gene expression. Impor-tantly, several of these validated hypermethylated genes encodeproteins that bear the potential of being implicated in cancerdevelopment. Of the genes with the best inverse correlationbetween methylation and gene expression, Junctional adhesionmolecule 2 (JAM2) is involved in cell-cell adhesion and cell-extracellular matrix interactions, functions that are expected tobe greatly affected during tumorigenesis. The well-characterizedgene MLH1, which is involved in mis-match repair, has alreadybeen associated with the development of hereditary nonpolypo-sis colon cancer (HNPCC).40 FLI1 and EYA4 are transcriptionalactivators and coactivators, respectively, which play a role inthe regulation of gene expression. These functions are alsoexpected to be affected in cancer development.

NRG1 has been proposed to have both oncogene and tu-mor suppressor functions. In breast cancer cell lines, reductionof NRG1 levels by stable transfection of siRNA constructsincreases net cell proliferation and supports a role of NRG1 asa tumor suppressor.41 In melanoma, however, NRG1 has beenreported to promote proliferation and invasion, and inhibitdifferentiation.42

The inverse correlation between NRG1 mRNA and proteinexpression and DNA methylation in colorectal tissue is sup-

ported by similar findings in breast cancer, in which NRG1 isfrequently silenced by hypermethylation.41

Lack of correlation between DNA methylation and geneexpression was also observed and may have different explan-ations. A mechanism termed long-range epigenetic silencingdescribed for CRC and prostate cancer may explain why lessthan half of the validated genes had an inverse correlationbetween DNA methylation and gene expression. Hypermeth-ylation of large regions typically spanning 1-2 Mb seem to bea common phenomenon in carcinogenesis.43,44 Hence, if largecontiguous regions are hypermethylated, it is possible thatgenes that already were silenced by other mechanismsbecome hypermethylated as a bystander effect. Hypermethyl-ation of highly expressed genes may be explained by the factthat only few CpG sites were interrogated for each gene, andthe interrogated sites may not be the sites involved in regula-tion of the gene. In addition, the presence of alternative tran-scription start sites may also be involved.

Although several genes with known cancer-specific altera-tions in DNA methylation were identified, we did not con-firm all previously published CRC-specifically hypermethy-lated genes using these arrays. This may be explained by thefact that for each gene represented on the array, in averageonly two CpG sites were interrogated, most of which local-ized in CpG-islands in the promoter area. If different CpGsites have been interrogated in the individual studies, discrep-ancies in methylation levels may have been observed. More-over, not all functionally important DNA methylation eventstake place in promoter regions. Irizarry et al. have reportedthat most methylation alterations occur in sequences up to 2kb away from the target gene, which they term ‘‘CpG islandshores.’’22 Genes that are regulated by distant methylationalterations will not be discovered using the arrays that wehave used in our study.

In conclusion, using whole genome methylation arrays, wehave identified genes not previously associated with cancer-spe-cific hypermethylation in CRC. The majority of the identifiedgenes were methylated already in the adenomas rendering thesegenes potential biomarkers for screening purposes. Future stud-ies will determine whether the methylation status of these genescan be detected in for example feces or blood samples.

AcknowledgementsThe authors thankMs. Anita Roest, Ms. Pamela Celis, Ms. Inge-Lis Thorsen,Ms. Anne Slotsdal, Ms. Hanne Steen, Ms. Lisbet Kjeldsen, Ms. Karen Bihland Ms. Bente Devantie for their excellent technical assistance, and the staffat the Department of Pathology at Aarhus University Hospital.

References

1. Ahuja N, Mohan AL, Li Q, Stolker JM,Herman JG, Hamilton SR, Baylin SB, Issa JP.Association between CpG island methylationand microsatellite instability in colorectalcancer. Cancer Res 1997;57:3370–4.

2. Toyota M, Ahuja N, Ohe-Toyota M,Herman JG, Baylin SB, Issa JP. CpG

island methylator phenotype in colorectalcancer. Proc Natl Acad Sci USA 1999;96:8681–6.

3. Issa JP, Zehnbauer BA, Kaufmann SH, BielMA, Baylin SB. HIC1 hypermethylation isa late event in hematopoietic neoplasms.Cancer Res 1997;57:1678–81.

4. Dominguez G, Carballido J, Silva J, SilvaJM, Garcia JM, Menendez J, Provencio M,Espana P, Bonilla F. p14ARF promoterhypermethylation in plasma DNA as anindicator of disease recurrence in bladdercancer patients. Clin Cancer Res 2002;8:980–5.

Early

Detection

andDiagn

osis



5. Wei SH, Chen CM, Strathdee G,Harnsomburana J, Shyu CR, RahmatpanahF, Shi H, Ng SW, Yan PS, Nephew KP,Brown R, Huang TH. Methylationmicroarray analysis of late-stage ovariancarcinomas distinguishes progression-freesurvival in patients and identifies candidateepigenetic markers. Clin Cancer Res 2002;8:2246–52.

6. el-Deiry WS, Nelkin BD, Celano P, YenRW, Falco JP, Hamilton SR, Baylin SB.High expression of the DNAmethyltransferase gene characterizeshuman neoplastic cells and progressionstages of colon cancer. Proc Natl Acad SciUSA 1991;88:3470–4.

7. de Caceres II, Cortes-Sempere M, MoratillaC, Machado-Pinilla R, Rodriguez-Fanjul V,Manguan-Garcia C, Cejas P, Lopez-Rios F,Paz-Ares L, de CJ, Nistal M, Belda-IniestaC, et al. IGFBP-3 hypermethylation-deriveddeficiency mediates cisplatin resistance innon-small-cell lung cancer. Oncogene 2010;29:1681–90.

8. Arnold CN, Goel A, Boland CR. Role ofhMLH1 promoter hypermethylation indrug resistance to 5-fluorouracil incolorectal cancer cell lines. Int J Cancer2003;106:66–73.

9. Suraweera N, Duval A, Reperant M, VauryC, Furlan D, Leroy K, Seruca R, IacopettaB, Hamelin R. Evaluation of tumormicrosatellite instability using fivequasimonomorphic mononucleotiderepeats and pentaplex PCR.Gastroenterology 2002;123:1804–11.

10. Wojdacz TK, Dobrovic A. Methylation-sensitive high resolution melting(MS-HRM): a new approach for sensitiveand high-throughput assessment ofmethylation. Nucleic Acids Res 2007;35:e41.

11. Andersen CL, Jensen JL, Orntoft TF.Normalization of real-time quantitativereverse transcription-PCR data:a model-based variance estimationapproach to identify genes suited fornormalization, applied to bladder andcolon cancer data sets. Cancer Res 2004;64:5245–50.

12. Birkenkamp-Demtroder K, Wagner L,Brandt SF, Bording AL, Gartner W,Scherubl H, Heine B, Christiansen P,Orntoft TF. Secretagogin is a novel markerfor neuroendocrine differentiation.Neuroendocrinology 2005;82:121–38.

13. Osborn NK, Zou H, Molina JR, Lesche R,Lewin J, Lofton-Day C, Klatt KK,Harrington JJ, Burgart LJ, Ahlquist DA.Aberrant methylation of the eyes absent 4gene in ulcerative colitis-associateddysplasia. Clin Gastroenterol Hepatol 2006;4:212–8.

14. Schatz P, Distler J, Berlin K, Schuster M.Novel method for high throughput DNAmethylation marker evaluation using PNA-

probe library hybridization and MALDI-TOF detection. Nucleic Acids Res2006;34:e59.

15. Herman JG, Merlo A, Mao L, Lapidus RG,Issa JP, Davidson NE, Sidransky D, BaylinSB. Inactivation of the CDKN2/p16/MTS1gene is frequently associated with aberrantDNA methylation in all common humancancers. Cancer Res 1995;55:4525–30.

16. Gonzalez-Zulueta M, Bender CM, YangAS, Nguyen T, Beart RW, Van TornoutJM, Jones PA. Methylation of the 5’ CpGisland of the p16/CDKN2 tumorsuppressor gene in normal andtransformed human tissues correlates withgene silencing. Cancer Res 1995;55:4531–5.

17. Glockner SC, Dhir M, Yi JM, McGarveyKE, Van NL, Louwagie J, Chan TA,Kleeberger W, de Bruine AP, Smits KM,Khalid-de Bakker CA, Jonkers DM, et al.Methylation of TFPI2 in stool DNA: apotential novel biomarker for the detectionof colorectal cancer. Cancer Res 2009;69:4691–9.

18. Ausch C, Kim YH, Tsuchiya KD,Dzieciatkowski S, Washington MK,Paraskeva C, Radich J, Grady WM.Comparative analysis of PCR-basedbiomarker assay methods for colorectalpolyp detection from fecal DNA. ClinChem 2009;55:1559–63.

19. Belshaw NJ, Elliott GO, Williams EA,Bradburn DM, Mills SJ, Mathers JC,Johnson IT. Use of DNA from humanstools to detect aberrant CpG islandmethylation of genes implicated incolorectal cancer. Cancer EpidemiolBiomarkers Prev 2004;13:1495–501.

20. Dickinson RE, Dallol A, Bieche I, Krex D,Morton D, Maher ER, Latif F. Epigeneticinactivation of SLIT3 and SLIT1 genes inhuman cancers. Br J Cancer 2004;91:2071–8.

21. Zhang W, Glockner SC, Guo M, MachidaEO, Wang DH, Easwaran H, Van NL,Herman JG, Schuebel KE, Watkins DN,Ahuja N, Baylin SB. Epigenetic inactivationof the canonical Wnt antagonist SRY-boxcontaining gene 17 in colorectal cancer.Cancer Res 2008;68:2764–72.

22. Irizarry RA, Ladd-Acosta C, Wen B, WuZ, Montano C, Onyango P, Cui H, GaboK, Rongione M, Webster M, Ji H, PotashJB, et al. The human colon cancermethylome shows similar hypo- andhypermethylation at conserved tissue-specific CpG island shores. Nat Genet2009;41:178–86.

23. Caldwell GM, Jones C, Gensberg K, Jan S,Hardy RG, Byrd P, Chughtai S, Wallis Y,Matthews GM, Morton DG. The Wntantagonist sFRP1 in colorectaltumorigenesis. Cancer Res 2004;64:883–8.

24. Shames DS, Girard L, Gao B, Sato M,Lewis CM, Shivapurkar N, Jiang A, Perou

CM, Kim YH, Pollack JR, Fong KM, LamCL, et al. A genome-wide screen forpromoter methylation in lung canceridentifies novel methylation markers formultiple malignancies. PLoS Med 2006;3:e486.

25. Yagi K, Akagi K, Hayashi H, Nagae G,Tsuji S, Isagawa T, Midorikawa Y,Nishimura Y, Sakamoto H, Seto Y,Aburatani H, Kaneda A. Three DNAmethylation epigenotypes in humancolorectal cancer. Clin Cancer Res 2010;16:21–33.

26. Model F, Osborn N, Ahlquist D,Gruetzmann R, Molnar B, Sipos F, GalambO, Pilarsky C, Saeger HD, Tulassay Z, HaleK, Mooney S, et al. Identification andvalidation of colorectal neoplasia-specificmethylation markers for accurateclassification of disease. Mol Cancer Res2007;5:153–63.

27. Bernet A, Mazelin L, Coissieux MM, GadotN, Ackerman SL, Scoazec JY, Mehlen P.Inactivation of the UNC5C Netrin-1receptor is associated with tumorprogression in colorectal malignancies.Gastroenterology 2007;133:1840–8.

28. Shin SK, Nagasaka T, Jung BH, MatsubaraN, Kim WH, Carethers JM, Boland CR,Goel A. Epigenetic and genetic alterationsin Netrin-1 receptors UNC5C and DCC inhuman colon cancer. Gastroenterology2007;133:1849–57.

29. Okada T, Suehiro Y, Ueno K, Mitomori S,Kaneko S, Nishioka M, Okayama N, SakaiK, Higaki S, Hazama S, Hirata H, SakaidaI, et al. TWIST1 hypermethylation isobserved frequently in colorectal tumorsand its overexpression is associated withunfavorable outcomes in patients withcolorectal cancer. Genes ChromosomesCancer 2010;49:452–62.

30. Zou H, Osborn NK, Harrington JJ, KlattKK, Molina JR, Burgart LJ, Ahlquist DA.Frequent methylation of eyes absent4 gene in Barrett’s esophagus andesophageal adenocarcinoma. CancerEpidemiol Biomarkers Prev 2005;14:830–4.

31. Feng Q, Balasubramanian A, Hawes SE,Toure P, Sow PS, Dem A, Dembele B,Critchlow CW, Xi L, Lu H, McIntosh MW,Young AM, et al. Detection ofhypermethylated genes in women with andwithout cervical neoplasia. J Natl CancerInst 2005;97:273–82.

32. Fackler MJ, McVeigh M, Evron E, GarrettE, Mehrotra J, Polyak K, Sukumar S,Argani P. DNA methylation of RASSF1A,HIN-1, RAR-beta, Cyclin D2 and Twist inin situ and invasive lobular breastcarcinoma. Int J Cancer 2003;107:970–5.

33. Renard I, Joniau S, van CB, Collette C,Naome C, Vlassenbroeck I, Nicolas H, deLJ, Straub J, Van CW, Hamida W, Hellel

Early

Detection

andDiagn

osis

Øster et al. 2865


M, et al. Identification and Validation ofthe Methylated TWIST1 and NID2 Genesthrough Real-Time Methylation-SpecificPolymerase Chain Reaction Assays for theNoninvasive Detection of Primary BladderCancer in Urine Samples. Eur Urol 2009.

34. Tsou JA, Galler JS, Siegmund KD, LairdPW, Turla S, Cozen W, Hagen JA, KossMN, Laird-Offringa IA. Identification of apanel of sensitive and specific DNAmethylation markers for lungadenocarcinoma. Mol Cancer 2007;6:70.

35. Kang GH, Lee S, Cho NY, GandamihardjaT, Long TI, Weisenberger DJ, Campan M,Laird PW. DNA methylation profiles ofgastric carcinoma characterized byquantitative DNA methylation analysis.Lab Invest 2008;88:161–70.

36. Weisenberger DJ, Siegmund KD, CampanM, Young J, Long TI, Faasse MA, Kang GH,Widschwendter M, Weener D, Buchanan D,Koh H, Simms L, et al. CpG islandmethylator phenotype underlies sporadicmicrosatellite instability and is tightlyassociated with BRAF mutation in colorectalcancer. Nat Genet 2006;38:787–93.

37. Cottrell S, Jung K, Kristiansen G, Eltze E,Semjonow A, Ittmann M, Hartmann A,

Stamey T, Haefliger C, Weiss G. Discoveryand validation of 3 novel DNAmethylation markers of prostate cancerprognosis. J Urol 2007;177:1753–8.

38. Gort EH, Suijkerbuijk KP, Roothaan SM,Raman V, Vooijs M, van der Wall E, vanDiest PJ. Methylation of the TWIST1promoter, TWIST1 mRNA levels, andimmunohistochemical expression ofTWIST1 in breast cancer. CancerEpidemiol Biomarkers Prev 2008;17:3325–30.

39. Rakyan VK, Down TA, Maslau S, AndrewT, Yang TP, Beyan H, Whittaker P,McCann OT, Finer S, Valdes AM, LeslieRD, Deloukas P, et al. Human aging-associated DNA hypermethylation occurspreferentially at bivalent chromatindomains. Genome Res 2010;20:434–9.

40. Bronner CE, Baker SM, Morrison PT,Warren G, Smith LG, Lescoe MK, Kane M,Earabino C, Lipford J, Lindblom A.Mutation in the DNA mismatch repairgene homologue hMLH1 is associated withhereditary non-polyposis colon cancer.Nature 1994;368:258–61.

41. Chua YL, Ito Y, Pole JC, Newman S, ChinSF, Stein RC, Ellis IO, Caldas C, O’Hare

MJ, Murrell A, Edwards PA. The NRG1gene is frequently silenced by methylationin breast cancers and is a strong candidatefor the 8p tumour suppressor gene.Oncogene 2009;28:4041–52.

42. Buac K, Xu M, Cronin J, WeeraratnaAT, Hewitt SM, Pavan WJ. NRG1/ERBB3 signaling in melanocytedevelopment and melanoma: inhibitionof differentiation and promotion ofproliferation. Pigment Cell Melanoma Res2009;22:773–84.

43. Frigola J, Song J, Stirzaker C, HinshelwoodRA, Peinado MA, Clark SJ. Epigeneticremodeling in colorectal cancer results incoordinate gene suppression across anentire chromosome band. Nat Genet 2006;38:540–9.

44. Coolen MW, Stirzaker C, Song JZ,Statham AL, Kassir Z, Moreno CS, YoungAN, Varma V, Speed TP, Cowley M,Lacaze P, Kaplan W, et al. Consolidationof the cancer genome into domains ofrepressive chromatin by long-rangeepigenetic silencing (LRES) reducestranscriptional plasticity. Nat Cell Biol2010;12:235–46.

Early

Detection

andDiagn

osis



Identification and validation of highly frequent CpG island hypermethylation in colorectal adenomas and carcinomas

Documents