Integration of High-Resolution Methylome and Transcriptome ...Childhood Acute Lymphoblastic Leukemia Stephan Busche 1 , Bing Ge 2 , Ramon Vidal 3 , Jean-Fran¸cois Spinella 3 , Virginie

Molecular and Cellular Pathobiology

Integration of High-Resolution Methylome andTranscriptome Analyses to Dissect Epigenomic Changes inChildhood Acute Lymphoblastic Leukemia

Stephan Busche1, Bing Ge2, Ramon Vidal3, Jean-Francois Spinella3, Virginie Saillour3, Chantal Richer3,Jasmine Healy3, Shu-Huang Chen1, Arnaud Droit5, Daniel Sinnett3,4, and Tomi Pastinen1,2

AbstractB-cell precursor acute lymphoblastic leukemia (pre-B ALL) is the most common pediatric cancer. Although

the genetic determinants underlying disease onset remain unclear, epigenetic modifications including DNAmethylation are suggested to contribute significantly to leukemogenesis. Using the Illumina 450K array, weassessed DNA methylation in matched tumor-normal samples of 46 childhood patients with pre-B ALL,extending single CpG-site resolution analysis of the pre-B ALL methylome beyond CpG-islands (CGI).Unsupervised hierarchical clustering of CpG-site neighborhood, gene, or microRNA (miRNA) gene-associatedmethylation levels separated the tumor cohort according to major pre-B ALL subtypes, and methylation inCGIs, CGI shores, and in regions around the transcription start site was found to significantly correlate withtranscript expression. Focusing on samples carrying the t(12;21) ETV6–RUNX1 fusion, we identified 119subtype-specific high-confidence marker CpG-loci. Pathway analyses linked the CpG-loci–associated geneswith hematopoiesis and cancer. Further integration with whole-transcriptome data showed the effects ofmethylation on expression of 17 potential drivers of leukemogenesis. Independent validation of arraymethylation and sequencing-derived transcript expression with Sequenom Epityper technology and real-time quantitative reverse transcriptase PCR, respectively, indicates more than 80% empirical accuracy of ourgenome-wide findings. In summary, genome-wide DNAmethylation profiling enabled us to separate pre-B ALLaccording to major subtypes, to map epigenetic biomarkers specific for the t(12;21) subtype, and through acombined methylome and transcriptome approach to identify downstream effects on candidate drivers ofleukemogenesis. Cancer Res; 73(14); 4323–36. �2013 AACR.

IntroductionAcute lymphoblastic leukemia (ALL) is the most frequent

cancer in children, accounting for approximately 25% of allpediatric cancers (1). B-progenitor cell (pre-B), the most prev-alent form of ALL in children (�80% of cases), arises from animmature lymphoid progenitor cell that acquires genomiclesions leading tomalignant transformation (1, 2). Cytogeneticanalyses detect primary gross chromosomal abnormalities in75% of B-cell precursor ALL (pre-B ALL) cases, allowing further

classification into subtypes based on recurrent translocationsor aneuploidies (3). The t(12;21) resulting into the ETV6–RUNX1 fusion and high hyperdiploidy (>50 chromosomes),each account for approximately 25% of cases. Less frequenttranslocations include t(9;22), generating the BCR–ABL fusion(2%–4% frequency), t(1;19) TCF3-PBX1 (2%–6%), and t(4;11)MLL-AF4 (1%–2%; ref. 3). Current treatment regimens forchildhoodALL rely on risk classification systems incorporatingmolecular genetic subtype information, and risk-adapted ther-apies have improved overall survival rates. However, due todismal cure rates in high-risk groups, relapse, and drug resis-tance resulting in treatment failure, pre-B ALL remains theleading cause of cancer-related death in children (1, 4). Inaddition, more than two thirds of survivors develop chronic orlate-occurring health problems linked to their disease and itstreatment (5, 6).

DNA methylation is a frequent epigenetic modificationthat occurs almost exclusively in the context of cytosinemethylation, and is crucially involved in the regulation ofgene expression (7). Promoter hypermethylation of tumorsuppressor genes, hypomethylation of oncogenes, andalterations of microRNA (miRNA) gene methylation havebeen recognized to favor cancer development (7, 8).Although primary genetic lesions are considered important

Authors' Affiliations: 1Department of Human Genetics, McGill Univer-sity; 2McGill University and G�enome Qu�ebec Innovation Centre;3Research Center, CHU Sainte-Justine; 4Department of Pediatrics,Universit�e de Montr�eal, Montr�eal; and 5Department of Molecular Med-icine, Centre de Recherche du CHUQ, Universit�e Laval, Qu�ebec City,Qu�ebec, Canada

Note: Supplementary data for this article are available at Cancer ResearchOnline (http://cancerres.aacrjournals.org/).

CorrespondingAuthor:Daniel Sinnett, Division ofHematology-Oncology,Research Center, CHU Sainte-Justine, 3175 Cote Sainte-Catherine, Mon-tr�eal, QC H3T 1C5, Canada. Phone: 514-345-4931, ext. 3282; Fax: 514-345-4731; E-mail: [email protected]

doi: 10.1158/0008-5472.CAN-12-4367

�2013 American Association for Cancer Research.

CancerResearch

www.aacrjournals.org 4323

on August 28, 2020. © 2013 American Association for Cancer Research. cancerres.aacrjournals.org Downloaded from

Published OnlineFirst May 30, 2013; DOI: 10.1158/0008-5472.CAN-12-4367

http://cancerres.aacrjournals.org/

initiating events in leukemogenesis, it is now evident thatcooperating epigenetic alterations including DNA methyla-tion modifications contribute significantly to the phenotypeof leukemia cells and disease progression (8).

DNA methylation in childhood ALL has been intensivelystudied andmethylation profiles have been used to distinguishALL from healthy samples, other leukemias, or ALL subtypes(9–23). Single CpG-site resolution studies focused on analysisof CpG-islands (CGI), promoter regions, and regions aroundthe transcription start site (TSS), and recent large-scaleapproaches screening methylation levels of thousands ofprotein-coding genes associated with these regions foundinverse correlation with gene expression in ALL cell lines andprimary cells (9, 14–18, 20, 22–24). However, recent studies incolon cancer and induced pluripotent stem cells identifiedsubstantial alterations of DNA methylation to occur outsideof CGIs (25, 26).

The ALL methylome assessed earlier might overlook impor-tant ALL subtype classifiers and prognostic markers. The aimof our study was to provide a detailed map of childhood pre-BALL methylation, and to identify novel pre-B ALL subtype-specific methylation markers within and outside the classicCpG regions. We further investigated the downstream effectsof DNA methylation alterations to find clues of pathwaysperturbed in leukemia, and intersected methylome and tran-scriptome data to identify methylation events affecting puta-tive genes driving leukemogenesis.

Materials and MethodsPatient samples

Bone marrow or peripheral blood samples isolated at diag-nosis (leukemia material before treatment) and remission(matched normal material following treatment, derived 1month to 2 years after diagnosis) from 46 children diagnosedwith pre-B ALL were analyzed in this study (Table 1). Allpatients were of French-Canadian origin to decrease geneticheterogeneity. The tumor material contained high levels ofleukemic blasts (Table 1), whereas remission samples wereblast-free. Patients were from the established Quebec child-hood ALL (QcALL) cohort, and were diagnosed between 1999and 2010 at the Sainte-Justine University Health Center (UHC;Montreal, QC, Canada). The Sainte-Justine UHC InstitutionalReview Board approved the research protocol and informedconsent was obtained from all participating individuals and/ortheir parents.

DNA methylation analysisGenomic DNA (gDNA) was isolated from frozen cell pellets

using a genomic DNA purification kit (Gentra Systems), and500 ng of it was used for bisulfite conversion using the EZ DNAMethylation Kit (Zymo Research) according to the manufac-turer's protocols. The modified gDNA was processed asdescribed in the Infinium Assay Methylation Protocol GuideRev. C (November 2010), and analyzed on Infinium Human-Methylation450 BeadChips (Illumina), measuring DNA meth-ylation at single CpG-site resolution based on genotyping ofC/U polymorphisms. As measure of methylation, we chose theb value, ranging from 0 (no methylation at any allele) to 1.0

(complete methylation of both alleles). We classified themethylation status of probes with b < 0.3 as unmethylated,with 0.3 � b � 0.7 as hemi-methylated, and with b > 0.7 asmethylated (27). For genomic annotation of BeadChip probes,RefSeq gene and CGI annotations were obtained from theUniversity of California, Santa Cruz (UCSC) hg18 referencegenome [National Center for Biotechnology Information(NCBI) Reference Sequence Database Release 36], whereasmiRNA annotations were obtained from miRBase (http://www.mirbase.org) version 17. We excluded probes located onthe Y chromosome, binding multiple genomic regions, orbinding targets with known sequence variants (SupplementaryTable S1). Themethylation data have been deposited in NCBI'sGene Expression Omnibus: GSE38235.

Array methylation validation with the SequenomEpityper

Array methylation was validated by quantitative analyses ofcleavage products of bisulfite-converted DNA by matrix-assisted laser desorption/ionization–time-of-flight (MALDI-TOF) mass spectrometry (28) using gDNA from 12 patientswhose material was also analyzed in the array experiment(Table 1). For PCR, target-specific primer pairs were designedusing the EpiDesigner tool (http://www.epidesigner.com) and50 ng bisulfite-converted gDNA was used as input. Duplicatesand unreliably detected CpG units having high or low masswere removed from the analysis.

RNA expression analysisTotal RNA of 17 pre-B ALL tumor samples (Table 1) was

extracted from cell pellets using the mirVana Kit (Life Tech-nologies). Six microgram RNA (RNA integrity number > 6)spiked with External RNA Controls Consortium (ERCC) RNASpike-InControls (LifeTechnologies)were rRNA-depletedusingthe RiboMinus Eukaryote Kit (Life Technologies), and cDNAlibraries were prepared using the SOLiD Total RNA-seq Kit(Life Technologies). Whole-transcriptome paired-end sequenc-ing (50 bp � 35 bp) was conducted using the SOLiD 4 System(Life Technologies). RNA expression levels were assessed asnormalized FPKM (fragments per kilobase per million sequenc-ed reads) by running Cufflinks in quantification-only mode,using the UCSC RefSeq database (downloaded November 1,2010) with additional coordinates for the ERCC spike-in controlas reference (Supplementary Table S2).

RNA expression validation by quantitative real-timeRT-PCR

Tumor RNA from the same 12 patients validated for arraymethylation was used for expression validation by real-timequantitative reverse transcriptase PCR (qRT-PCR; primersequences given in the Supplementary Data). First-strandcDNA synthesis was conducted using the Verso cDNA Syn-thesis Kit (Thermo Scientific) with anchored oligo-dT primersaccording to the manufacturer's protocol. qRT-PCR assayswere run on the Rotor-Gene 6000 real-time rotary analyzer(Corbett Life Sciences) using the Platinum SYBR Green qRT-PCR SuperMix-UDG (Life Technologies) according to themanufacturer's protocol with the following cycling conditions:

Busche et al.

Cancer Res; 73(14) July 15, 2013 Cancer Research4324




Table 1. Histopathologic and clinical characteristics of patients with pre-B ALL

Subtypea ID SexBlastrate, %b

Age,mob

Height,cmb

Weight,kgb

WBC,109/Lb,c Riskd Eventse ADf

Hyperdiploidy (HHD) 356 M 63 124 137 35.8 9.8 High R, D372 M 87 99 133 30.3 2.1 Standard390 F 72 90 120 21.1 10.0 Standard618 F 97 36 94 12.8 2.9 Standard663 F 86 28 96 15.2 25.1 Standard670 M 93 33 98 14.7 5.5 Standard R761 M 88 30 96 15 3.2 Standard764 M 75 55 112 21.6 59.2 High R R, V768 M 95 53 100 16 4.9 Standard775 F 98 41 102 15.7 50.7 High777 F 73 33 93 13.6 61.7 High R, V801 M 93 48 108 19.3 3.9 Standard R815 M 80 63 106 19.2 6.1 Standard R, V819 M 97 48 98 16.8 98.2 High R, V

Trisomy 5 595 M 73 155 161.5 45 48.8 High R, D

Hypodiploidy 605 M 88 141 148 48 30.2 High

t(12;21) 373 F 75 55 96.5 13 208.0 High375 F 86 32 88 13.3 141.4 High379 F 93 28 90 12.2 4.5 Standard392 M 55 84 123 23.6 11.6 Standard R, V411 F 87 57 110.5 18.1 3.8 Standard R, V420 M 94 57 107.5 19.4 127.0 High R, V443 F 90 32 94.5 14.2 103.9 High580 F 94 52 105 16.6 35.3 Standard614 F 75 48 105 17.1 33.1 High R, V657 F 62 31 90 11.2 17.1 Standard685 F 86 63 107 19.7 83.5 High753 F 95 61 111 18.2 2.9 Standard R, V817 F 93 67 110 16.9 3.3 Standard R

t(12;21) and trisomy 21 718 F 66 142 130 44.2 30.1 High

t(9;22) 790 M 75 171 170 60.1 14.3 High R825 F 91 138 157 43.3 65.5 High R

t(1;19) 424 M 94 46 104 19 15.2 Standard

t(4;11) 757 F 97 127 149 35.3 7.7 High R, V

t(9;12) and monosomy 7 752 M 30 31 86 13.2 37.8 Standard

No abnormal finding (NAF) 381 F 67 32 99 16.9 27.7 High384 M 94 102 130 30 3.1 Standard431 M 98 144 156.5 44.4 7.1 High579 M 87 75 127 25 183.1 High R, D608 M 72 47 107 20 81.5 High617 F 77 46 111 20 23.6 Standard R, V762 F 97 133 156 34.8 274.6 High767 F 79 81 128 23.6 3.6 Standard794 M 89 168 156 37.2 280.6 Very High802 M 92 58 107 19.8 40.5 Standard R, V820 M 31 101 96 15.6 19.3 Standard R

aDetails on the cytogenetic analyses are given in the Supplementary Data.bData were obtained at diagnosis.cWBC, white blood cell count.dThe risk groupswere determinedand the patientswere treated according to theDana-FarberCancer Institute (Boston,MA)ChildhoodALL consortium protocols effective at the time of diagnosis.eR, relapse; D, death. For 2 relapse cases death did not occur within the follow-up period.fAdditional datasets available: R, RNA-sequencing; V, Epityper and qRT-PCR.

Epigenomic Alterations in Childhood ALL

www.aacrjournals.org Cancer Res; 73(14) July 15, 2013 4325




4 minutes at 95�C, 40 cycles� (20 seconds at 95�C, 30 secondsat 55�C, and 30 seconds at 72�C) followed by the dissociationprotocol at 72�C. Transcript expression was analyzed in tri-plicates. 18S and glyceraldehyde-3-phosphate dehydrogenase(GAPDH) were used for normalization.

Statistical data analysisData analyses were conducted in R version 2.12.2 (29)

applying standard functions and the "lattice" and "RColor-Brewer" packages. Heatmaps were generated with the "heat-map" function, wherein "hclust" was used to compute theunsupervised clustering analyseswith average linkage asmeth-od of clustering and one minus the correlation coefficient(Spearman r) as dissimilarity structure. Weighted gene corre-lation network analysis (WGCNA) was conducted in R version2.14.2 by using theWGCNAR package for weighted correlationnetwork analysis (30). Default settings were applied andunsigned networks were constructed; the soft-thresholdingpower was adjusted to 7, the minimum module size set to 30.

Ingenuity pathway analysisIngenuity pathway analysis (IPA; Ingenuity Systems) content

version 12402621 was used. We conducted core analyses usingapproach-specific reference datasets, considered direct andindirect relationships, included endogenous chemicals, and setthresholds to a maximum of 35 molecules per network and 25networks per analysis. We included all available data sources,selected for the human species, tissue, and primary cells,experimentally observed molecules and/or relationships only,and selected stringent filters when applicable.

ResultsGlobal methylation landscapes

Methylation profiles were measured in 46 French-Canadianpre-B ALL patients (Table 1) on Illumina 450K BeadChips. Foreach patient, tumor material was isolated at diagnosis and thecorresponding normal material at remission. Taking intoaccount the 392,904 autosomal probes on the chip thatmapped uniquely to the genome and showed no geneticvariation, we observed a bimodal median methylation leveldistribution (Supplementary Fig. S1A–S1I) with probes beingeither unmethylated (40.1% in normal/35.7% in tumor) ormethylated (47.5% in normal/50.9% in tumor), with a medianmethylation level of 0.63 in normal and 0.70 in tumor samples(Supplementary Fig. S1J). We annotated autosomal probeswith respect to their location (Fig. 1A). Each probe wasassigned to its CpG-site neighborhood being located in a CGI,CGI shore, CGI shelf, or open sea, and toward the closest TSS ofRefSeq and miRBase-annotated genes located within 1.5 kb or200 bp upstream of the TSS (TSS1500 or TSS200, respectively),or intragenic. For RefSeq genes, intragenic regionswere furthersubcategorized in 50-untranslated region (UTR), exon 1, genebody, and 30-UTR. Irrespective of disease status, we detectedlocation-specific CpG-site methylation trends: CGIs were pre-dominantly unmethylated, shores showed dichotomous meth-ylation, and shelves and open sea were essentially methylated.In terms of RefSeq gene context, sites within the TSS1500,TSS200, 50-UTR, and exon 1were predominantly unmethylated,

whereas the remaining intragenic sites were mostly methyl-ated. Sites located in miRNA gene promoter and intragenicregions tended to be methylated (Supplementary Fig. S1K andS1L). The X-chromosomal median methylation level distribu-tion allowed a clear gender-based sample separation withhemi-methylated patterns characteristic for X-chromosomeinactivation detected in females (Supplementary Fig. S1M–S1O), and median methylation levels of 0.54 in remissionsamples derived from females (0.64 in tumor) and 0.25 inmales (0.41 in tumor). Overall, the increase in median tumormethylation levels is driven by (i) a decrease in unmethylatedand concordant increase in methylated sites, and (ii) a quan-titative increase toward higher methylation levels in predom-inantly methylated samples (Supplementary Fig. S1A–S1L).

Across samples, individual CpG-site methylation variationwas low for the majority of probes: in every CpG-site contextmore than 56.5% (up to 98.9%) of probes displayed a SD smallerthan 0.1 (Supplementary Table S3). However, we observedsignificant differences between tumor and normal material(Supplementary Table S3). A SD < 0.1 was observed for morethan 95.3% of CpG-sites in every CpG-site context in normalsamples, compared with only 73.6% in tumor samples. Incontrast, high CpG-site methylation variation (SD > 0.2) wasobserved for at least 7.4% of the probeswhen summed in tumormaterial, and was rarely observed in normal samples.

Despite these differences, we note that DNA methylationpatterns are tissue specific (7). The analyzed tumor materialconsisted mainly of leukemic blasts, which resemble lymphoidprogenitor cells within the early developmental stages of B-lymphocytes (2). In contrast, the corresponding normal mate-rial was a mixture of untransformed lymphocytes and mono-cytes. Therefore, some of the detected methylation alterationsmight arise from differences in tissue origin rather than be truedisease-associated methylation changes, and the lower cross-sample CpG-site methylation variation in normal samplesmight be a result of blunted variability in heterogeneoustissues. Also, patients undergo chemotherapy and remissionsamples might exhibit treatment-biased methylation patterns,which outlast the disease. To circumvent these biases, wedecided to conduct comparisons between tumor samples toidentify tumor and subtype-specific methylation alterations.

Methylation profile-based pre-B ALL subtypestratification

Sample-dependent CpG-site methylation level variationshould bemost revealing for methylation profile-based sampledifferentiation into specific subgroups. Autosomal probesshowing highest variance among pre-B ALL tumor samplesand passing filters were selected (Supplementary Table S4).Unsupervised hierarchical clustering of the detected methyl-ation levels clearly separated samples carrying the t(12;21)translocation from other pre-B ALL subtypes (Fig. 1B). Asecond broad cluster was formed by 10 high hyperdiploid(HHD) samples aligned next to 4 samples categorized asdisplaying no abnormalfinding (NAF). One sample categorizedas t(12;21) aligned next to the t(12;21) sample with additionaltrisomy 21, with both of these samples exhibiting amethylationprofile distinct from other t(12;21) tumors. We conducted the

Busche et al.





same analysis using methylation data from the normal sam-ples: these clustered independently of their tumoral counter-part's cytogenetic characteristics (Supplementary Fig. S3A),

indicating that ALL subtypes have distinct global CpG-meth-ylation signatures irrespective of the patients' germline geneticbackground.

Figure 1. Cluster analysis of pre-B ALL sample-derived methylation values for subtype stratification and correlation of methylation and gene expressionlevels. A, probe annotation regions as outlined in the text. Each probe was assigned to its CpG-site neighborhood and toward the closest TSS ofRefSeq and miRBase-annotated genes. B and C, heatmaps illustrate an unsupervised hierarchical clustering analysis of autosomal (B) and miRNA(C) gene-associated methylation levels exhibiting highest cross-sample variance including all study samples. The dissimilarity structure wasdetermined as one minus Spearman rank correlation coefficient and average linkage was used as method of clustering. In the dendrogram on the left,each horizontal line represents a sample color coded according to clinically determined molecular subtype. The color key is given in the legend below B.In the top dendrogram, each vertical line represents a CpG-site, with detected methylation levels colored according to the scale bar above B.Methylation levels detected at 3,929 probes (top 1%; B) and 63 probes (top 2%; C) were used for heatmap generation. D, Spearman rank correlationr was determined for methylation and transcript expression levels in each indicated CpG-site context. Black diamonds, individual patient sample-derived Spearman r values; red diamonds, the median r within the specified region.






We applied the same clustering approach to methylationlevels detected in individual CpG-site neighborhoods and generegions (Supplementary Table S4 and Supplementary Fig. S2A–S2P). In all CpG-site contexts, we obtained subtype separationssimilar to those described earlier: t(12;21) samples formed aseparate cluster, and an additional cluster was formed by theHHD samples. Clustering of methylation levels detected on theX-chromosome followed the same separation patterns (Supple-mentary Fig. S2Q and S2R); however, interpretation for maleswas difficult as only 2 individuals carried the t(12;21) translo-cation. Normal samples showed no subtype-dependent cluster-ing in different CpG-contexts (Supplementary Fig. S3B–S3S).

Subtype stratification based onmiRNA genemethylationprofiles

Deregulation of miRNA expression in acute leukemia hasbeen associated with altered miRNA gene promoter methyl-ation levels (8). The BeadChip covers 683 autosomal miRNAgene loci with a total of 3,126 probes (Supplementary Table S1).We tested whether the detected methylation levels displayinghighest cross-sample variance allow hierarchical clustering-based subtype separation (Supplementary Table S4). Twelve of13 samples carrying the t(12;21) translocation formed adistinctcluster, which aligned next to a second cluster consisting ofnine HHD samples interspersed with 2 NAF samples (Fig. 1C).Clustering of region-specific methylation levels did notimprove separation efficiency (Supplementary Fig. S2S andS2T). Normal samples clustered irrespective of their tumoralcounterpart's subtype (Supplementary Fig. S3T–S3V).

Correlation of DNA methylation and gene expressionlevels

DNA methylation is involved in the regulation of geneexpression (7). We observed a decrease in median methylationlevels for CpG-sites located within 1 kb of the TSS (Supple-mentary Fig. S4A). When we compared methylation levels andgene expression in 17 patient samples whose transcriptomeshad been determined by RNA-sequencing (Table 1), we foundinverse correlation between methylation at CpG-sites locatedwithin the TSS1500 and gene expression. The unmethylated/methylated ratio of the highest median expression quartile oftranscripts is close to 9 (Supplementary Fig. S4B), whereas forthe lowest expression quartile this ratio dips to 2 (Supplemen-tary Fig. S4C). We quantified the methylation/expressioncorrelation in different CpG-site regions determining Spear-man rank correlation r for each patient sample (Fig. 1D). Weobserved the highest inverse correlations for CpG-sites locatedwithin exon 1, and slight but significant positive correlationsfor the 30-UTR. Further inverse correlations were observed inthe TSS1500, TSS200, 50-UTR, CGIs, CGI shores, and genebodies. The detected methylation and expression levels in CGIshelves were not correlated (Supplementary Tables S5 and S6).

t(12;21)-Specific DNA methylation patterns and geneexpression

Unsupervised clustering clearly separated samples carryingthe t(12;21) translocation from other molecular subtypes(Fig. 1B, 1C, and Supplementary Fig. S2). Weighted gene

correlation network analysis (WGCNA) is based on unsuper-vised clustering and can be used to link clustering-derivedmodules of highly interconnected genes to external sampletraits (30). It has recently been applied to investigate correla-tion patterns among CpG-sites (31). We used WGCNA toidentify modules of interconnected CpG-sites related to thet(12;21) subtype. As input data, we used probes showinghighest cross-sample methylation level variance (as for Fig.1B) and excluded the t(12;21) sample with the additionaltrisomy 21. Figure 2A displays the cluster dendrogram gener-ated by WGCNA with assigned modules; the brown moduledisplayed highest correlation with the t(12;21)-subtype (r ¼0.98; P ¼ 7e�169; Fig. 2B). This module contains 240 probes,124 of which displaying lower and 116 displaying highermedian methylation levels in samples carrying the t(12;21)translocation when compared with the other samples (Sup-plementary Table S7A). These probes were associated with 133genes; their annotations are given in Supplementary Table S8A.

In addition to the clustering-based WGCNA approach, weconducted a Mann–Whitney rank sum test to identify signif-icantly altered DNA methylation patterns specific for thet(12;21) subtype. In this analysis, we included all probes thatpassed our filters and compared probe methylation betweensamples with and without the t(12;21) translocation. Weselected significantly differing probes [false discovery rate(FDR) ¼ 5%] with median b values < 0.3 in one and > 0.7 inthe other group. This led to the identification of 169 hypo- and140 hypermethylated probes in the t(12;21) subtype (Supple-mentary Table S7B and S7C), associated with 85 and 75 genes,respectively. Annotations of probe-associated genes are givenin Supplementary Table S8B and S8C. A common set of 119probes and 62 genes were identified by both approaches andare highlighted in Supplementary Tables S7A and S8A, respec-tively. Less than 50% of the differentially methylated CpG-sitesare located in CGIs or around the TSS (Fig. 2C and D).

Noprobes in the brownmodulewere associatedwithmiRNAgenes. In the Mann–Whitney approach, the only identifiedprobe associated with a miRNA, cg19463199, indicated hypo-methylation in the TSS1500 of the miRNA-922 gene in additionto its association with KIAA0226. However, miRNA gene meth-ylation-based clustering separated t(12;21) from other sub-types (Fig. 1C). To look for evidence of differential methylationat miRNA loci, we lowered stringency of our probe selectioncriteria (FDR 10%; Db > 0.25) and identified two additionalCpG-sites that confirmed hypomethylation of miRNA-922and five other differentially methylated miRNA loci within thet(12;21) subtype (Table 2).

To identify methylation alterations that translate into geneexpression differences in the t(12;21) subtype, we selectedprobes located in regions with observed highest absolutemedian correlation coefficients (Supplementary Table S5;jmedian rj > 0.1) and compared the methylation trend ofassociated genes with RNA-sequencing–derived transcriptexpression estimates of the available 6 t(12;21) and 11 othersubtype patient samples. Of the 133 probe-associated genesidentified by the WGCNA approach, 49 and 43 were classifiedas uniformly hypo- and hypermethylated, respectively (Sup-plementary Table S9A). Of these, 7 hypomethylated genes

Busche et al.





(IGF2BP1, DSC3, FAM19A1, TMED6, BEST3, SIDT1, and OPN3)and 1 hypermethylated gene (SPSB1) displayed significantlyaltered expression levels in t(12;21) (Table 3). In every case,significant expression alteration inversely correlated withmethylation trend.Mann–Whitney analysis identified 85 genes as hypo- and

75 genes as hypermethylated specifically in the t(12;21)-subtype, of which 60 hypo- and 52 hypermethylated geneswere associated with probes in regions with high absolutecorrelation (Supplementary Table S9B and S9C). Sixteenhypomethylated genes (IGF2BP1, DSC3, FAM19A1, TMED6,TCFL5, CHL1, EPOR, SOX11, BEST3, FUCA1, SIDT1, HLA-DPB1, MYH10, CPNE2, NRN1, and GRK7) and 1 hypermethy-lated gene (FBXL15) showed significantly altered transcriptexpression levels (Table 3). In all instances, hypomethylationwas associated with increased transcript expression. How-ever, for hypermethylated gene FBXL15 transcript expres-

sion was slightly elevated in t(12;21) samples (Table 3). Ofthe 17 identified hypomethylated genes, six were jointlyidentified by both approaches (IGF2BP1, DSC3, FAM19A1,TMED6, BEST3, and SIDT1).

Overall, the median expression levels of genes hypermethy-lated in t(12;21) (b > 0.7) and unmethylated in the othersubtypes (b < 0.3) did not differ significantly (P ¼ 0.87 forWGCNA; P¼ 0.65 for Mann–Whitney; Fig. 2E). In contrast, weobserved significantly increased median transcript expressionlevels only for hypomethylated genes in t(12;21). This suggeststhat the unmethylated genes in the other subtypes areexpressed at low levels similar to methylated genes.

Pathway analysis of differentially methylated genes int(12;21)

IPA was conducted to determine biologic and disease-asso-ciated pathways affected by differentialmethylation in t(12;21).

Figure 2. DNA methylation and expression patterns specific for samples carrying the t(12;21)-translocation. A, WGCNA-derived methylation level clusteringdendrogramobtainedby average linkage hierarchical clustering, with dissimilarity basedon topologic overlap. Each vertical line in the dendrogram representsone CpG-site, with highly interconnected CpG-sites grouped together in branches. Individual branches are assigned to colored modules determined by the"Dynamic Tree Cut" function, and identified modules with very similar methylation profiles were merged (merged dynamic). As input data, methylationvalues detected at the 3,929 autosomal probes displaying highest cross-sample variancewere used (same as used to generate Fig. 1B). B, the brownmodulewas identified to be most significantly associated with the t(12;21) subtype. Shown is a scatter plot of methylation probe significance versus brown modulemembership. Absolute Pearson r (cor) with P value is indicated above the plot. C and D, location of probes identified by the WGCNA or Mann–Whitneyapproach as t(12;21)-specifically hypo- and hypermethylated annotated to CpG-site neighborhood (C) or regional gene context (D). E, significance of mediantranscript expression level alterations between hypo- and hypermethylated genes in t(12;21) and other samples. Mann–Whitney rank sum tests wereconducted to determine statistical significance betweenmedian expression levels of hypo- and hypermethylated genes identified in Supplementary Table S9.P values derived for genes identified in the WGCNA and Mann–Whitney approaches are highlighted in yellow and blue boxes, respectively; significantdifferences (P < 0.05) are highlighted in bold.






Genes differentially methylated in regions displaying highabsolute correlation between methylation and expressionlevels were considered (see above). The WGCNA modulescontain both hypo- and hypermethylated genes.We conducteda core analysis including all differentially methylated genes(Supplementary Table S9A) and weighted gene hypo- or hyper-methylation as transcriptional up- or downregulation, respec-tively, with all probe-associated genes considered in theWGCNA approach used as reference dataset. "Cell Cycle,Cancer, Gastrointestinal Disease" was identified as the top-associated network, followed by "Cellular Movement, Hema-tological System Development and Function, Immune CellTrafficking" (Supplementary Fig. S5A and S5B and Supplemen-tary Table S10A).

In the Mann–Whitney approach, we generated 2 gene listsseparating hypo- and hypermethylated genes. We conductedIPA for genes of each list separately (Supplementary TableS9B and S9C), weighted differential methylation as previ-ously, and used probe-associated genes interrogated in theMann–Whitney approach as reference dataset. "Cell Cycle,Cell Death, Cancer" was identified as top network associatedwith hypomethylation (Supplementary Fig. S5C and Supple-mentary Table S10B), and "Cellular Development, CellularGrowth and Proliferation, Hematological System Develop-ment and Function" with hypermethylation (SupplementaryFig. S5D and Supplementary Table S10C).

Finally, we investigated networks generated by genes jointlyidentified in WCGNA and Mann–Whitney approaches (Sup-plementary Table S9B and S9C); the top-associated networkwas "Cell Death, Cell-To-Cell Signaling and Interaction, Hema-tological System Development and Function" (SupplementaryFig. S5E and Supplementary Table S10D).

Validation of t(12;21)-specific CpG-site methylation andgene expression

We used the Epityper technology and qRT-PCR to validatearray-derived CpG-site methylation and RNA-seq–derivedgene expression levels, respectively, in 12 previously analyzedpatient samples (Table 1). Methylation validation was targetedto CpG-sites associated with genes specifically methylated andexpressed in t(12;21) (IGF2BP1,DSC3, FAM19A1, BEST3, TCFL5,SIDT1, TMED6, and SPSB1). For the selected array-interrogatedTMED6-associated CpG-site, the Epityper experiment failed. Inall other instances, array methylation significantly correlated(median r ¼ 0.882) with Epityper methylation (Fig. 3A).Moreover, previously observed significant methylation differ-ences between samples carrying the t(12;21) translocation andother subtypes were confirmed for all interrogated CpG-sites, except for one (of two) FAM19A1-associated site atchr3:68,136,174 and the SPSB1-associated site at chr1:9,300,572(Fig. 3B). Failed validation of t(12;21)-specific methylation atthese loci might be due to false positive discovery, or, in caseof SPSB1, blurred by additional CpG-sites interrogated in theEpityper CpG unit. Because the Epityper technology quan-tifies CpG-site methylation on up to 600-bp long stretches oftemplate, we investigated methylation of CpG-sites in closeproximity to the array-interrogated sites. This identifiedt(12;21)-specific methylation of SPSB1, and multiple novelloci displaying t(12;21)-specific methylation were discovered(Fig. 3B), further supporting characterized subtype-specificgene methylation patterns.

Because of limited RNA availability, transcript expressionvalidation was targeted to a subset of the methylation-vali-dated genes (IGF2BP1, DSC3, FAM19A1, TCFL5, SIDT1, andTMED6). For all genes, we observe high inverse correlations

Table 2. miRNA genes differentially methylated in samples carrying the t(12;21) translocation

miRNAa Probe IDb Chrc LocationcmiRNAcontextd

CpG-siteneighborhoode

t(12;21)med. bf

Othersmed. bg P valueh

Hypomethylatedhsa-mir-922 cg19463199 3 198886256 TSS1500 Open sea 0.289 0.868 8.62e�07

cg10579986 3 198885848 TSS200 Open sea 0.503 0.922 8.77e�06cg15435900 3 198885797 Gene body Open sea 0.558 0.899 2.07e�06

hsa-mir-320b-1 cg04390865 1 117015533 TSS1500 South shelf 0.679 0.927 4.71e�05hsa-mir-4747 cg11970254 19 4883663 TSS1500 CGI 0.455 0.922 5.23e�05

Hypermethylatedhsa-mir-320a cg08264885 8 22159482 TSS1500 South shore 0.402 0.133 5.83e�07hsa-mir-1976 cg23955417 1 26753516 TSS1500 Open sea 0.841 0.304 4.71e�05hsa-mir-200c cg16642299 12 6941644 TSS1500 Open sea 0.703 0.282 5.53e�05

amiRNA name (miRBase version 17).bIllumina ID of CpG-site–interrogating probe associated with the miRNA gene.cChromosome number (Chr) and genomic coordinate (Location) of the cytosine interrogated by the probe.dLocation of the probe with respect to the miRNA gene.eLocation of the probe in the CpG-site neighborhood.fMedian methylation level across samples carrying the t(12;21) translocation.gMedian methylation level across all other samples.hMann–Whitney P value for difference in methylation between t(12;21) and other samples.

Busche et al.





between RNA-seq–derived and qRT-PCR–detected transcriptexpression (Fig. 4A), and previously observed subtype-specificexpression was validated for all except TMED6 (Fig. 4B).

DiscussionWe studied 46 childhood pre-B ALL sample pairs derived at

diagnosis (tumor) and after disease remission (normal) fromthe QcALL cohort. Importantly, we studied primary materialas culturing of cells might introduce methylation artifacts(32). For each sample, we assessed the methylation status of402,842 CpG-sites across the whole genome, including CpG-site contexts not previously assessed in ALL. We note that theIllumina 450K BeadChip interrogates 485,577 CpG-sites, butwe chose a stringent approach and discarded probes targetingmultiple genomic regions or containing known single-nucle-otide polymorphisms (SNP) to minimize nonepigenetic vari-ation. Whole-transcriptome sequencing was conducted for17 of the tumor samples. Array methylation and sequenc-

ing-derived transcript expression were validated with Seque-nom Epityper technology and qRT-PCR, respectively. Theseindependent tests indicate more than 80% empirical accuracyof our genome-wide findings (10 of 12 methylation and 5 of6 expression observations validated).

Unsupervised hierarchical clustering of methylation levelsdisplaying high cross-sample variance separated the tumorcohort according to major pre-B ALL subtypes. Overall, ourclustering results were similar to those obtained in a previousstudy conducted by Milani and colleagues investigating 1,536CpG-sites (15). The CpG-sites interrogated in the Milani andcolleagues' study were located in CGIs and around the TSS;their results validate our findings in the corresponding regions,and bringmoreweight to the clustering results we obtained forpreviously uninvestigated regions. We report for the first timepre-B ALL subtype separation based on the clustering ofmethylation levels detected in CGI shores or CGI shelves,associated with individual gene regions, or miRNA genes.Unexpectedly, subtype separation efficiencies achieved in

Table 3. Genes displaying significant hypo- or hypermethylation and differential expression in samplescarrying the t(12;21) translocation

# of CpG-sites Med. FPKM

Genea Accessionb Approachc WGCNAd MWe Trendf t(12;21)g Othersh P valuei

BEST3 NM_032735 WþM 1 1 Hypo 43442 4088 1.03e�02CHL1 NM_006614 M n/a 1 Hypo 6477 3398 3.07e�03CPNE2 NM_152727 M n/a 1 Hypo 64912 25368 2.02e�02DSC3 NM_024423 WþM 6 3 Hypo 15275 0 1.56e�02EPOR NM_000121 M n/a 1 Hypo 10508 2825 7.11e�03FAM19A1 NM_213609 WþM 2 2 Hypo 43213 9747 3.07e�03FBXL15j NM_024326 M n/a 3 Hyper 6991 2586 1.03e�02FUCA1 NM_000147 M n/a 1 Hypo 38738 21545 1.45e�02GRK7 NM_139209 M n/a 1 Hypo 4198 1955 4.99e�02HLA-DPB1 NM_002121 M n/a 1 Hypo 90807 0 1.58e�02IGF2BP1 NM_006546 WþM 6 8 Hypo 35148 0 7.13e�04MYH10 NM_005964 M n/a 1 Hypo 19942 13633 2.02e�02NRN1 NM_016588 M n/a 1 Hypo 38910 18048 2.73e�02OPN3 NM_014322 W 1 n/a Hypo 30895 14952 3.65e�02SIDT1 NM_017699 WþM 1 1 Hypo 17969 6750 1.45e�02SOX11 NM_003108 M n/a 1 Hypo 28661 11345 7.11e�03SPSB1 NM_025106 W 1 n/a Hyper 3708 7409 3.65e�02TCFL5 NM_006602 M n/a 1 Hypo 126736 23603 1.13e�03TMED6k NM_144676 WþM 1 2 Hypo 55113 13879 7.70e�03

aGene name according to HUGO Gene Nomenclature Committee.bRefSeq transcript accession number.cProbe-identifying approach: W, WGCNA; M, Mann–Whitney; WþM, both approaches.dNumber of gene-associated CpG-sites identified in the WGCNA approach (Supplementary Table S7A).eNumber of gene-associated CpG-sites identified in the Mann–Whitney approach (Supplementary Table S7B–S7C).fGene methylation trend in t(12;21).gMedian transcript expression across samples carrying the t(12;21) translocation.hMedian transcript expression across all other samples.iMann–Whitney P value for difference in transcript expression between t(12;21) and other samples.jHypermethylation in t(12;21) is associated with slightly increased expression.kt(12;21)-specific expression was not validated by qRT-PCR.






these novel regions were similar to the ones obtained previ-ously, indicating that cross-sample methylation level variancespecific to pre-B ALL spans a variety of genomic elements.

DNA methylation in CGIs, CGI shores, and at gene promo-ters is associated with gene silencing (7); we indeed detectstrong inverse correlations between methylation and tran-script expression levels in these regions. In contrast, intragenicmethylation is assumed to positively correlate with geneexpression (7). However, we observe a positive correlationonly for methylation levels detected in 30-UTR, whereas levelsin 50-UTR and exon 1 exhibit high, and in the gene body slight,negative correlations.

As samples carrying the t(12;21) translocation segregatedmost clearly from the other subtypes, we focused furtheranalyses on this subtype with the aim to identify subtype-specific epigenomic biomarkers and potential cooperatingepigenetic lesions driving leukemogenesis. The t(12;21) trans-

location creates the in-frame ETV6–RUNX1 fusion oncogene(syn. TEL-AML1), and the chimeric protein is assumed to act asregulation-resistant transcriptional repressor of RUNX1 targetgenes (33). However, recent studies investigating fusion pro-tein-dependent genome-wide expression changes report generepression and activation in similar amounts for protein-coding (34) and miRNA (35) genes.

We used two methodically distinct approaches, WGCNAand Mann–Whitney, to identify differential probe methylationin the t(12;21)-subtype. We identified a total of 430 uniqueCpG-sites specifically related to the t(12;21)-subtype, of which119 were jointly identified by both approaches. None of theseCpG-sites have previously been identified to be differentiallymethylated in pre-B ALL and all these sites might function asnovel epigenetic biomarkers specific for the t(12;21) subtype.Their ultimate suitability, however, will have to be confirmed inlarger study cohorts. Furthermore, more than 50% of these

Figure 3. Comparison of array- andSequenom Epityper–derivedmethylation. A, Epityper CpG unitquantitative methylation values areplotted against array-derivedb values. Genes, interrogated CpG-site location(s), and array-probe ID(s)are displayed. When the Epityper-interrogated CpG unit comprisedmore than one CpG-site, thearray-derived b values of bothCpG-sites were averaged andcompared with the Epityper-quantified CpG unit comprising thesesame two CpG-sites (�); ��, theEpityper-quantified CpG unitcomprised two neighboringCpG-sites, of which only the onehighlighted in black was interrogatedby the array. Spearman rankcorrelation r with significance level(P) is displayed. (Continued on thefollowing page.)

Busche et al.





sites are not located in CGIs or around the TSS, indicating thatCpG-site regions previously not investigated at single-siteresolution might harbor valuable biomarkers for pre-B ALL.In the Mann–Whitney approach, we also identified t(12;21)-

specific CpG-sites associated with miRNA genes. The moststatistically significant hit in the hypermethylated cohort wasmiRNA-320a, which was identified in a recent study (35) to bedownregulated together with miRNA-494 in t(12;21)-leukemia.Mechanistically, bothmiRNAs are coupled to the repression ofsurvivin, their inhibition promoting antiapoptotic signaling. Asecond significantly hypermethylated miRNA gene in t(12;21),miRNA-200c, was shown to exhibit strong tumor suppressorproperties partially by interactingwith the epigenetic regulatorand polycomb group protein BMI1 (36), which fulfills essentialroles in both normal and leukemic stem cells (37).CpG-sites identified as t(12;21)-specificwere associatedwith

231 genes (62 were jointly identified by both approaches). Ofthese, the methylation signatures of BMP4, CELSR1, DSC3,and PON2 were previously identified as subtype classifiers fort(12,21) (15). To our knowledge, the remaining 227 genes arenovel differentially methylated gene classifiers of the t(12;21)-subtype.Correlation analysis irrespective of molecular subtype

identified differential methylation of BMP4, CELSR1, DSC3,and PON2 to significantly inversely correlate with transcript

expression in ALL (15). We investigated methylation altera-tions translating into expression differences specific for thet(12;21) subtype, and identified one hyper- and 17 hypomethy-lated genes with inversely correlating transcript expression.Previous studies carrying out expression profiling of pre-B ALLidentified eight of the hypomethylated genes (DSC3, EPOR,FUCA1, HLA-DPB1, TCFL5, NRN1, IGF2BP1, BEST3) as cohort-specific subclassifiers upregulated in the t(12;21)-subtype (38–42). Transcriptional upregulation of SOX11 was previouslyidentified in t(12;21)- or TCF3 rearrangement-carrying ALL,and for t(12;21) upregulation was accompanied by a demethy-lated promoter (43). To our knowledge, FAM19A1, CHL1,SIDT1, CPNE2, GRK7, MYH10, and OPN3 are novel biomarkersidentified as specifically upregulated in the t(12;21)-subtype,and for all genes except SOX11, demethylation as mechanismunderlying augmented expression was previously unknown.The RNA-seq–determined t(12;21)-specific upregulation ofTMED6 expression could not be validated by qRT-PCR, pre-sumably due to a limited validation cohort size.

Several genes hypomethylated and upregulated in t(12;21)have been identified to have a functional impact on tumori-genesis. In pre-B leukemic cells, ETV6-RUNX1 binding tothe EPOR promoter was shown to promote cell survival (44).IGF2BP1 is suggested to enhance tumor cell proliferationby sustaining elevated MYC expression (45), TCFL5

Figure 3. (Continued.) B,Sequenom Epityper–derivedmethylation of genes identified tobe specifically methylated andexpressed in t(12;21). For eachEpityper-interrogated CpG unit themethylation status of individualscarrying (orange circles) andindividuals not carrying (bluecircles) the t(12;21)-translocation isdisplayed (group-specific medianmethylation indicated by an X).CpG unit comprising CpG-sitelocations are indicated on the left,CpG-sites also interrogated by thearray are highlighted in green.Mann–Whitney P valuescomparing methylation of t(12;21)-carriers with noncarriers areindicated on the right; insignificantvalues (P>0.05) are shaded in gray.






downregulation is associated with reduced cell viability ofhuman colorectal cancer cell lines (46), and CHL1 was shownto be differentially expressed during the development of seve-ral human cancers (47).

We identified only one gene, SPSB1, both hypermethylatedand downregulated in samples carrying the t(12;21) translo-

cation (Epityper validation of the array-interrogated probefailed; however, neighboring CpG-sites display t(12;21)-specificity; Fig. 4B). No link between SPSB1 and leukemiahas been previously reported, making it a novel biomarkergene in t(12;21).

Overall, we observed significantly increased median tran-script expression levels only for hypomethylated genes int(12;21), whereas hypermethylation did not significantlyimpact transcription. An explanation for this might be pro-vided by studies investigating nonhematologic cancers (48),which indicate that CGIs become de novo methylated at anearly stage of tumorigenesis and that most newly targetedgenes are already repressed before transformation. Mechanis-tically, abnormally expanding progenitor cells with flexiblePolycomb-group protein-mediated transcriptional repressionof CGI-associated genes as observed in embryonic stem cells,might shift toward stabler silencing via DNAmethylation. Thisshift from dynamic to static transcriptional repression mightprevent subsequent transcriptional activation needed for sig-nal transduction and differentiation.

As the ETV6–RUNX1 fusion gene was identified to directlybind the EPOR promoter (44), we hypothesized that thisinteraction might be generally coupled to promoter demeth-ylation. We tested for it but did not detect significant enrich-ment of hypomethylated genes identified in our approachwhen comparing with published RUNX1 chromatin immuno-precipitation (ChIP)-seq data generated in K562 lymphoblasts(49). However, K562 cells are derived from chronic myeloge-nous leukemia, and to ultimately clarify this hypothesis fusionprotein ChIP-seq in pre-B ALL needs to be carried out.

Both ETV6 and RUNX1 are known regulators of hemato-poiesis (33, 50). IPA analysis identified the functional network"Hematological System Development and Function" to besignificantly associated with differential gene methylation insamples carrying the t(12;21) translocation. Other significantassociations include "Cancer" and cellular functions deregu-lated during carcinogenesis, for example, "Cell Cycle," "CellDeath," and "Cellular Growth and Proliferation." TNF, a centralhub in several of these generated gene networks, is known to becritically involved in these functions (51). The most significantnetworks associated with Mann–Whitney–derived hypo- andhypermethylated genes identified TP53 as central hub. TP53itself is seldom found to be targeted by genetic alterations inALL, however, components of the TP53 pathway have beenshown to be frequently mutated in ALL (2), and a recent studyreported frequent epigenetic inactivation of TP53 pathwaygenes in ALL (21).

Overall, this study illustrates the power of methylationprofiling to classify leukemic subtypes, and to identify sub-type-specific methylation biomarkers. However, most methyl-ation changes do not alter gene expression detectably, sodifferential methylation analysis seems suitable for epigeneticbiomarker discovery, but as tool to understand leukemogen-esis, it needs to be coupled with other genomic data (e.g., RNA-sequencing). We show that integrative analysis of global meth-ylation and transcriptome alterations in pre-B ALL subtypescan be applied to study downstream effects of individualgenomic rearrangements in tumors. Additional work on the

Figure 4. Comparison of RNA-seq– and qRT-PCR–derived transcriptexpression. A, for the indicated genes, RNA-seq–derived transcriptexpression levels (FPKM) are plotted against qRT-PCR DCt values[Ct target gene � Ct housekeeping gene (18S or GAPDH)]. Samplescarrying the t(12;21)-translocation are highlighted in orange, samplesof other subtypes in blue; n.d., target gene expression was notdetected by qRT-PCR. Spearman rank correlation r with significancelevel (P) derived from correlating FPKM to DCt (for each housekeepinggene separately) is displayed. B, qRT-PCR–derived median DCtsdetected for the indicated genes across samples carrying (orange)or not carrying (blue) the t(12;21)-translocation when normalized to18S (left) or GAPDH (right). Stars indicate statistical significance(Mann–Whitney; P < 0.05).

Busche et al.





group of hypomethylated and upregulated genes in samplescarrying the t(12;21) translocation we identified in thisstudy could reveal novel causally important processes inleukemogenesis.

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed.

Authors' ContributionsConception and design: S. Busche, D. Sinnett, T. PastinenDevelopment of methodology: S. Busche, R. Vidal, S.-H. Chen, T. PastinenAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): S. Busche, J.-F. Spinella, C. Richer, J. Healy, S.-H. Chen,A. Droit, D. Sinnett, T. PastinenAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): S. Busche, B. Ge, R. Vidal, V. Saillour, A. Droit,D. Sinnett, T. PastinenWriting, review, and/or revision of themanuscript: S. Busche, R. Vidal, J.-F.Spinella, J. Healy, A. Droit, D. Sinnett, T. PastinenAdministrative, technical, or material support (i.e., reporting or orga-nizing data, constructing databases): S. Busche, D. Sinnett, T. PastinenStudy supervision: T. Pastinen

AcknowledgmentsThe authors thank all the patients and their parents who consented to

participate in this study. The methylation assays were conducted at the McGillUniversity and G�enome Qu�ebec Innovation Centre, Montr�eal, Canada. Whole-transcriptome sequencing was conducted at the Child Health Genomics Plat-form of the Sainte-Justine UHC Research Center.

Grant SupportThis work was supported by research funds provided by Genome Quebec and

Genome Canada, the Terry Fox Research Institute, the Fonds de la Recherche enSant�e du Qu�ebec (FRSQ), the Canadian Institutes for Health Research (CIHR),and the Cole foundation. S. Busche is the recipient of a postdoctoral fellowshipfrom the R�eseau de M�edecine G�en�etique Appliqu�ee. R. Vidal is the recipient of apostdoctoral research fellowship from the Foreign Affairs and InternationalTrade Canada. J.-F. Spinella is the recipient of a Cole Foundation scholarship.D. Sinnett is a scholar of the Fonds de la Recherche en Sant�e du Qu�ebec.

The costs of publication of this article were defrayed in part by the paymentof page charges. This article must therefore be hereby marked advertisementin accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received December 11, 2012; revised April 11, 2013; accepted April 24, 2013;published OnlineFirst May 30, 2013.

References1. Cheok MH, Evans WE. Acute lymphoblastic leukaemia: a model for

the pharmacogenomics of cancer therapy. Nat Rev Cancer 2006;6:117–29.

2. Pui CH, Relling MV, Downing JR. Acute lymphoblastic leukemia.N Engl J Med 2004;350:1535–48.

3. Pui CH, Carroll WL, Meshinchi S, Arceci RJ. Biology, risk stratification,and therapy of pediatric acute leukemias: an update. J Clin Oncol2011;29:551–65.

4. Mullighan CG. Molecular genetics of B-precursor acute lymphoblasticleukemia. J Clin Invest 2012;122:3407–15.

5. Mody R, Li S, Dover DC, Sallan S, Leisenring W, Oeffinger KC, et al.Twenty-five-year follow-up among survivors of childhood acute lym-phoblastic leukemia: a report from the Childhood Cancer SurvivorStudy. Blood 2008;111:5515–23.

6. Oeffinger KC, Mertens AC, Sklar CA, Kawashima T, Hudson MM,Meadows AT, et al. Chronic health conditions in adult survivors ofchildhood cancer. N Engl J Med 2006;355:1572–82.

7. Portela A, Esteller M. Epigenetic modifications and human disease.Nat Biotechnol 2010;28:1057–68.

8. Chen J, Odenike O, Rowley JD. Leukaemogenesis: more than mutantgenes. Nat Rev Cancer 2010;10:23–36.

9. Davidsson J, Lilljebjorn H, Andersson A, Veerla S, Heldrup J, Beh-rendtz M, et al. The DNA methylome of pediatric acute lymphoblasticleukemia. Hum Mol Genet 2009;18:4054–65.

10. Roman-Gomez J, Jimenez-Velasco A, Agirre X, Prosper F, Heiniger A,Torres A. Lack of CpG island methylator phenotype defines a clinicalsubtype of T-cell acute lymphoblastic leukemia associated with goodprognosis. J Clin Oncol 2005;23:7043–9.

11. Gutierrez MI, Siraj AK, Bhargava M, Ozbek U, Banavali S, ChaudharyMA, et al. Concurrent methylation of multiple genes in childhood ALL:correlation with phenotype and molecular subgroup. Leukemia2003;17:1845–50.

12. Roman-Gomez J, Jimenez-Velasco A, Agirre X, Castillejo JA, NavarroG, Calasanz MJ, et al. CpG island methylator phenotype redefines theprognostic effect of t(12;21) in childhood acute lymphoblastic leuke-mia. Clin Cancer Res 2006;12:4845–50.

13. Takeuchi S, Matsushita M, Zimmermann M, Ikezoe T, Komatsu N,Seriu T, et al. Clinical significance of aberrant DNA methylation inchildhood acute lymphoblastic leukemia. Leuk Res 2011;35:1345–9.

14. Taylor KH, Kramer RS, Davis JW, Guo J, Duff DJ, Xu D, et al. Ultradeepbisulfite sequencing analysis of DNA methylation patterns in multiplegene promoters by 454 sequencing. Cancer Res 2007;67:8511–8.

15. Milani L, Lundmark A, Kiialainen A, Nordlund J, Flaegstad T, ForestierE, et al. DNA methylation for subtype classification and prediction of

treatment outcome in patients with childhood acute lymphoblasticleukemia. Blood 2010;115:1214–25.

16. Stumpel DJ, Schneider P, van Roon EH, Boer JM, de Lorenzo P,Valsecchi MG, et al. Specific promoter methylation identifies differentsubgroups of MLL-rearranged infant acute lymphoblastic leukemia,influences clinical outcome, and provides therapeutic options. Blood2009;114:5490–8.

17. Taylor KH, Pena-Hernandez KE, Davis JW, Arthur GL, Duff DJ, Shi H,et al. Large-scale CpGmethylation analysis identifies novel candidategenes and reveals methylation hotspots in acute lymphoblastic leu-kemia. Cancer Res 2007;67:2617–25.

18. Schafer E, Irizarry R, Negi S, McIntyre E, Small D, Figueroa ME, et al.Promoter hypermethylation in MLL-r infant acute lymphoblastic leu-kemia: biology and therapeutic targeting. Blood 2010;115:4798–809.

19. Scholz C, Nimmrich I, BurgerM, Becker E, Dorken B, LudwigWD, et al.Distinction of acute lymphoblastic leukemia from acute myeloid leu-kemia through microarray-based DNA methylation analysis. AnnHematol 2005;84:236–44.

20. Nordlund J, Milani L, Lundmark A, Lonnerholm G, Syvanen AC. DNAmethylation analysis of bone marrow cells at diagnosis of acutelymphoblastic leukemia and at remission. PLoS ONE 2012;7:e34513.

21. Vilas-Zornoza A, Agirre X, Martin-Palanco V, Martin-Subero JI, SanJose-Eneriz E, Garate L, et al. Frequent and simultaneous epigeneticinactivation of TP53 pathway genes in acute lymphoblastic leukemia.PLoS ONE 2011;6:e17012.

22. Dunwell T, Hesson L, Rauch TA, Wang L, Clark RE, Dallol A, et al. Agenome-wide screen identifies frequently methylated genes in hae-matological and epithelial cancers. Mol Cancer 2010;9:44.

23. Figueroa ME, Wouters BJ, Skrabanek L, Glass J, Li Y, Erpelinck-Verschueren CA, et al. Genome-wide epigenetic analysis delineatesa biologically distinct immature acute leukemia with myeloid/T-lym-phoid features. Blood 2009;113:2795–804.

24. Kuang SQ, Tong WG, Yang H, Lin W, Lee MK, Fang ZH, et al.Genome-wide identification of aberrantly methylated promoterassociated CpG islands in acute lymphocytic leukemia. Leukemia2008;22:1529–38.

25. IrizarryRA, Ladd-AcostaC,WenB,WuZ,MontanoC,OnyangoP, et al.The human colon cancer methylome shows similar hypo- and hyper-methylation at conserved tissue-specific CpG island shores. NatGenet 2009;41:178–86.

26. Doi A, Park IH, Wen B, Murakami P, Aryee MJ, Irizarry R, et al.Differential methylation of tissue- and cancer-specific CpG islandshoresdistinguishes human inducedpluripotent stemcells, embryonicstem cells and fibroblasts. Nat Genet 2009;41:1350–3.






27. Bell JT, Pai AA, Pickrell JK, Gaffney DJ, Pique-Regi R, Degner JF, et al.DNAmethylation patterns associate with genetic and gene expressionvariation in HapMap cell lines. Genome Biol 2011;12:R10.

28. Ehrich M, Nelson MR, Stanssens P, Zabeau M, Liloglou T, XinarianosG, et al. Quantitative high-throughput analysis of DNA methylationpatterns by base-specific cleavage and mass spectrometry. Proc NatlAcad Sci U S A 2005;102:15785–90.

29. Team RDC. R: a language and environment for statistical computing.Vienna, Austria: R Foundation for Statistical Computing; 2011.

30. Langfelder P, Horvath S. WGCNA: an R package for weighted corre-lation network analysis. BMC Bioinformatics 2008;9:559.

31. Bocklandt S, Lin W, Sehl ME, Sanchez FJ, Sinsheimer JS, Horvath S,et al. Epigenetic predictor of age. PLoS ONE 2011;6:e14821.

32. Sugawara H, Iwamoto K, Bundo M, Ueda J, Ishigooka J, Kato T.Comprehensive DNA methylation analysis of human peripheralblood leukocytes and lymphoblastoid cell lines. Epigenetics 2011;6:508–15.

33. Speck NA, Gilliland DG. Core-binding factors in haematopoiesis andleukaemia. Nat Rev Cancer 2002;2:502–13.

34. Fuka G, Kauer M, Kofler R, Haas OA, Panzer-Grumayer R. Theleukemia-specific fusion gene ETV6/RUNX1 perturbs distinct keybiological functions primarily by gene repression. PLoS ONE 2011;6:e26348.

35. Diakos C, Zhong S, Xiao Y, Zhou M, Vasconcelos GM, Krapf G, et al.TEL-AML1 regulation of survivin and apoptosis via miRNA-494 andmiRNA-320a. Blood 2010;116:4885–93.

36. Shimono Y, Zabala M, Cho RW, Lobo N, Dalerba P, Qian D, et al.Downregulation of miRNA-200c links breast cancer stem cells withnormal stem cells. Cell 2009;138:592–603.

37. Schuringa JJ, Vellenga E. Role of the polycomb group gene BMI1 innormal and leukemic hematopoietic stem and progenitor cells. CurrOpin Hematol 2010;17:294–9.

38. Ross ME, Zhou X, Song G, Shurtleff SA, Girtman K, WilliamsWK, et al.Classification of pediatric acute lymphoblastic leukemia by geneexpression profiling. Blood 2003;102:2951–9.

39. YeohEJ,RossME,Shurtleff SA,WilliamsWK,Patel D,MahfouzR, et al.Classification, subtype discovery, and prediction of outcome in pedi-atric acute lymphoblastic leukemia by gene expression profiling.Cancer Cell 2002;1:133–43.

40. Andersson A, Olofsson T, Lindgren D, Nilsson B, Ritz C, Eden P, et al.Molecular signatures in childhood acute leukemia and their correla-tions to expression patterns in normal hematopoietic subpopulations.Proc Natl Acad Sci U S A 2005;102:19069–74.

41. Nordlund J, KiialainenA, KarlbergO,BerglundEC,Goransson-KultimaH, Sonderkaer M, et al. Digital gene expression profiling of primaryacute lymphoblastic leukemia cells. Leukemia 2012;26:1218–27.

42. Stoskus M, Gineikiene E, Valceckiene V, Valatkaite B, Pileckyte R,Griskevicius L. Identification of characteristic IGF2BP expressionpatterns in distinct B-ALL entities. Blood Cells Mol Dis 2011;46:321–6.

43. Vegliante MC, Royo C, Palomero J, Salaverria I, Balint B, Martin-Guerrero I, et al. Epigenetic activation of SOX11 in lymphoid neo-plasms by histone modifications. PLoS ONE 2011;6:e21382.

44. Torrano V, Procter J, Cardus P, Greaves M, Ford AM. ETV6-RUNX1promotes survival of early B lineage progenitor cells via a dysregulatederythropoietin receptor. Blood 2011;118:4910–8.

45. Kobel M,Weidensdorfer D, Reinke C, Lederer M, Schmitt WD, Zeng K,et al. Expression of the RNA-binding protein IMP1 correlates with poorprognosis in ovarian carcinoma. Oncogene 2007;26:7584–9.

46. Sillars-Hardebol AH, Carvalho B, TijssenM, Belien JA, deWit M, Delis-vanDiemenPM, et al. TPX2 andAURKApromote 20q amplicon-drivencolorectal adenoma to carcinoma progression. Gut 2012;61:1568–75.

47. Senchenko VN, Krasnov GS, Dmitriev AA, Kudryavtseva AV, Aned-chenko EA, Braga EA, et al. Differential expression of CHL1 geneduring development of major human cancers. PLoS ONE 2011;6:e15612.

48. Baylin SB, Jones PA. A decade of exploring the cancer epigenome—biological and translational implications. Nat Rev Cancer 2011;11:726–34.

49. Pencovich N, Jaschek R, Tanay A, Groner Y. Dynamic combinatorialinteractions of RUNX1 and cooperating partners regulates megakar-yocytic differentiation in cell line models. Blood 2011;117:e1–14.

50. Wang LC, SwatW, Fujiwara Y, Davidson L, Visvader J, Kuo F, et al. TheTEL/ETV6 gene is required specifically for hematopoiesis in the bonemarrow. Genes Dev 1998;12:2392–402.

51. Aggarwal BB, Gupta SC, Kim JH. Historical perspectives on tumornecrosis factor and its superfamily: 25 years later, a golden journey.Blood 2012;119:651–65.

Busche et al.





2013;73:4323-4336. Published OnlineFirst May 30, 2013.Cancer Res Stephan Busche, Bing Ge, Ramon Vidal, et al. Lymphoblastic LeukemiaAnalyses to Dissect Epigenomic Changes in Childhood Acute Integration of High-Resolution Methylome and Transcriptome

Updated version

10.1158/0008-5472.CAN-12-4367doi:

Access the most recent version of this article at:

Material

Supplementary

http://cancerres.aacrjournals.org/content/suppl/2013/05/30/0008-5472.CAN-12-4367.DC1

Access the most recent supplemental material at:

Cited articles

http://cancerres.aacrjournals.org/content/73/14/4323.full#ref-list-1

This article cites 50 articles, 19 of which you can access for free at:

Citing articles

http://cancerres.aacrjournals.org/content/73/14/4323.full#related-urls

This article has been cited by 3 HighWire-hosted articles. Access the articles at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected]

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at

Permissions

Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://cancerres.aacrjournals.org/content/73/14/4323To request permission to re-use all or part of this article, use this link



http://cancerres.aacrjournals.org/lookup/doi/10.1158/0008-5472.CAN-12-4367

http://cancerres.aacrjournals.org/content/suppl/2013/05/30/0008-5472.CAN-12-4367.DC1

http://cancerres.aacrjournals.org/content/73/14/4323.full#ref-list-1

http://cancerres.aacrjournals.org/content/73/14/4323.full#related-urls

http://cancerres.aacrjournals.org/cgi/alerts

mailto:[email protected]

http://cancerres.aacrjournals.org/content/73/14/4323


Documents

Integration of High-Resolution Methylome and Transcriptome ...Childhood Acute Lymphoblastic Leukemia Stephan Busche 1 , Bing Ge 2 , Ramon Vidal 3 , Jean-Fran¸cois Spinella 3 , Virginie