Biology of Human Tumors

DNAMethylationSignatureRevealsCellOntogenyof Renal Cell CarcinomasGabriel G. Malouf1,2, Xiaoping Su3, Jianping Zhang3, Chad J. Creighton4, Thai H. Ho5,6,Yue Lu7, No€el J.-M. Raynal8, Jose A. Karam9, Pheroze Tamboli10, Frederick Allanick2,Roger Mouawad2, Jean-Philippe Spano1,2, David Khayat1,2, Christopher G.Wood9,Jaroslav Jelinek7, and Nizar M. Tannir11

Abstract

Purpose:DNAmethylation is a heritable covalentmodificationthat is developmentally regulated and is critical in tissue-typedefinition. Although genotype–phenotype correlations have beendescribed for different subtypes of renal cell carcinoma (RCC), it isunknown if DNA methylation profiles correlate with morpho-logical or ontology based phenotypes. Here, we test the hypoth-esis that DNA methylation signatures can discriminate betweenputative precursor cells in the nephron.

Experimental Designs: We performed deep profiling of DNAmethylation and transcriptome in diverse histopathological RCCsubtypes and validated DNA methylation in an independentdataset as well as in The Cancer Genome Atlas Clear Cell andChromophobe Renal Cell Carcinoma Datasets.

Results: Our data provide the first mapping of methylomeepi-signature and indicate that RCC subtypes can be groupedinto two major epi-clusters: C1, which encompasses clear-cell

RCC, papillary RCC, mucinous and spindle cell carcinomasand translocation RCC; C2, which comprises oncocytoma andchromophobe RCC. Interestingly, C1 epi-cluster displayed 3-fold more hypermethylation as compared with C2 epi-cluster.Of note, differentially methylated regions between C1 and C2epi-clusters occur in gene bodies and intergenic regions,instead of gene promoters. Transcriptome analysis of C1epi-cluster suggests a functional convergence on Polycombtargets, whereas C2 epi-cluster displays DNA methylationdefects. Furthermore, we find that our epigenetic ontogenysignature is associated with worse outcomes of patients withclear-cell RCC.

Conclusion: Our data define the epi-clusters that can discrim-inate between distinct RCC subtypes and for the first time definethe epigenetic basis for proximal versus distal tubule derivedkidney tumors. Clin Cancer Res; 1–11. �2016 AACR.

IntroductionRenal cell carcinoma (RCC) is a heterogeneous disease

with at least 12 subtypes of RCC based on the World HealthOrganization (WHO) classification, with distinct clinical out-comes ranging from malignant (i.e., clear-cell RCC, papillaryRCC) to benign (i.e., oncocytoma; ref. 1). For both clear-cellRCC (ccRCC) and papillary RCC (pRCC), patients can pres-ent with either localized or metastatic disease and medianoverall survivals for the most common histology ccRCC isapproximately 2 years (2, 3). Chromophobe RCC is usuallyindolent with a greater than 90% 10-year cancer-specificoverall survival, although some patients can develop metas-tases (3, 4).

Ontogenetically, each RCC subtype is classified on morpho-logical and histological biomarkers; however, some biomarkerscan be expressed in more than one subtype, making classifica-tion difficult (5–7). The precursor cell in the renal tubularsystem can affect the RCC subtype with studies suggesting thatccRCC and pRCC arise from the proximal tubules and onco-cytoma and chromophobe arise from the distal tubules (8, 9).Given the common precursor cells that give rise to RCC,regulatory factors that maintain normal cell identity and cellproliferation may influence the RCC phenotype (5–7, 10).DNAmethylation is a key DNAmodification that allows proper

1Department of Medical Oncology, Groupe Hospitalier Piti�e-Sal-petri�ere, University Pierre and Marie Curie (Paris VI), Institut Universi-taire de canc�erologie, AP-HP, Paris, France. 2AVEC Foundation Lab-oratory, Groupe Hospitalier Piti�e-Salpetri�ere, Paris, France. 3Depart-ment of Bioinformatics and Computational Biology, The University ofTexas MD Anderson Cancer Center, Houston, Texas. 4Department ofMedicine, Baylor College of Medicine, Houston, Texas. 5Division ofHematology and Medical Oncology, Mayo Clinic, Scottsdale, Arizona.6Center for Individualized Medicine, Epigenomics Group, Mayo Clinic,Rochester, Minnesota. 7Department of Molecular Carcinogenesis, TheUniversity of Texas MD Anderson Cancer Center, Smithville, Texas.8Fels Institute, Temple University School of Medicine, Philadelphia,Pennsylvania. 9Department of Urology, The University of Texas MDAnderson Cancer Center, Houston,Texas. 10Department of Pathology,The University of Texas MDAnderson Cancer Center, Houston, Texas.11Department of Genitourinary Medical Oncology, The University ofTexas MD Anderson Cancer Center, Houston, Texas.

Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).

Gabriel G. Malouf andXiaoping Su share first authorship and senior authorship ofthis article. These authors contributed equally to this article.

CorrespondingAuthors:N. Tannir, TheUniversity of TexasMDAndersonCancerCenter, 1515 Holcombe Boulevard, Houston, TX 77030. Phone: 713-563-7265;Fax: 713-745-0422; E-mail: ntannir@mdanderson.org; and G. Malouf, E-mail:gabriel.malouf@psl.aphp.fr

doi: 10.1158/1078-0432.CCR-15-1217

�2016 American Association for Cancer Research.

ClinicalCancerResearch

www.aacrjournals.org OF1

regulation of gene expression and stable gene silencing allow-ing to define cell state (11). The importance of DNA methyl-ation alterations in cancer has been extensively studied, withthe discovery of both hypermethylation within promoterregions of certain tumor-suppressor genes and long-rangehypomethylation which broadly contributes to cell transfor-mation (12, 13). DNA methylation is considered to be stable,when compared with gene expression (14).

We hypothesized that we could identify a DNA methylationsignature that discriminates between benign andmalignant RCC.In this article, we applied for thefirst time, to our knowledge, next-generation sequencing to study targeted DNA methylation in

subtypes of RCC. Our analysis reveals two distinct RCC epi-clusters consistent each with their cell ontogeny.

Patients and MethodsPatient characteristics and tumor selection

Patient samples for the training set were related to 22 primaryRCCs with different histologies treated at the University of Texas,MD Anderson Cancer Center. Specimens were gathered in accor-dancewith the institutional policies. All patients providedwritteninformed consent. The study was approved by MD AndersonCancer Center institutional review board. DNA was obtainedexclusively from frozen nephrectomy, and none of the patienthad been treated with chemotherapy or radiation prior to surgery.Tumors were selected solely on the basis of availability. Further-more, DNA was extracted from 2 specimens of normal kidneytissue adjacent to renal cancers. Overall, our cohort included 3ccRCC, 5 pRCC (3 type I and 2 type II), 5 tRCC confirmedcytogenetically, 2 oncocytomas, 3 chromophobes, 2 MTS, and2 RCC with sarcomatoid differentiation. All the hematoxylin andeosin-stained (HES) slides from surgical specimens have beenreviewed by one uropathologist (P.T.). Out of the 22 cases, RNAwas available for 11 cases, including 2 oncocytoma, 3 chromo-phobe, 2 tRCC, 1 ccRCC, 2 papillary type 1 and 1MTS. The clinicalannotations of those samples are provided in Table 1.

For the validation dataset, 41 new samples, including 4 normalkidneys and 37 tumor samples were collected from Piti�e-Sal-petri�ere Hospital. Specimens were gathered in accordancewith the institutional policies. All patients provided writteninformed consent. The study was approved by Piti�e-Salpetri�ereethical committee. Distribution of tumor samples in this inde-pendent dataset was as follows: ccRCC (n ¼ 16), chromophobeRCC (n ¼ 6), oncocytoma (n ¼ 8), papillary type II RCC (n ¼ 6),and translocation RCC (n ¼ 1).

Table 1. Clinicopathological annotations of RCC samples selected in our study

Histology Sample-ID Sex

TNMpathologicalstage

Totaltags(N) RNAseq

CpG sites (n) withat least 10 tagscoverage per site

CpG sites (n) withat least 100 tagscoverage per site

MITF/TFE translocation RCC T1 F T4N2M1 85,062,596 Yes 65,050 43,590T2 M T1N2M1 75,410,891 No 98,341 38,456T3 M T3bN0M0 71,422,259 No 139,443 44,191T24 F T3aN0M0 99,019,782 No 162,226 43,215T34 M T3bN1M0 57,537,059 Yes 158,229 44,375

ccRCC T11 F T2N0M0 118,594,689 Yes 156,156 42,552T12 M T2N0M0 42,845,148 No 53,989 33,982T15 F T1bN0M0 86,725,973 No 65,104 40,630

Papillary RCC (type I) T14 M T1aN0M0 80,878,268 No 57,256 46,228T20 M T2aNxM0 125,615,959 Yes 59,059 42,120T21 M T4N1M0 91,521,140 Yes 118,971 45,886

Papillary RCC (type II) T7 M T1bN0M0 78,384,802 No 151,757 37,930T8 F T3bN2M0 77,404,215 No 54,928 45,572

Mucinous and spindle cell RCC T13 M T1N0M0 51,426,774 No 148,280 44,584T19 F T1bNxM0 27,216,549 Yes 51,127 38,557

Sarcomatoid RCC T22 F T4N1M1 59,402,641 No 121,837 46,118T23 M T4N1M0 140,885,594 No 98,104 44,956

Oncocytoma T45 M T1aNxM0 79,100,507 Yes 82,991 36,167T46 F T1aNxM0 77,335,441 Yes 87,629 46,205

Chromophobe T10 F T2N0M0 82,187,775 Yes 61,940 61,940T16 M T2aNxM0 73,994,602 Yes 78,618 45,828T17 F T1bNxM0 38,917,059 Yes 51,653 34,117

Normal N4 F 87,459,889 Yes 76,348 43,590N8 F 84,567,697 Yes 61,256 41,514

NOTE: All cases have been assessed for DNA methylation using DREAM technique. The details of sequencing tags and the read coverage are reported for DREAM.Selected RCC cases assessed by RNAseq are also specified.

Translational Relevance

Genotype–phenotype correlations have been described fordifferent subtypes of renal cell carcinoma (RCC), but it isunknown if epigenetics can define cell ontology across diversehistological kidney subtypes. Herein, we demonstrate thatRCC can be divided epigenetically in two epi-clusters thatcorrelate with kidney cell ontogeny. While C1 epi-clusterencompasses clear-cell RCC and papillary RCC, C2 epi-clusterwas composed of chromophobe RCC and oncocytomas. Ofnote, C1 epi-cluster displayed 3-fold more hypermethylationas compared with C2 epi-cluster, consistent with functionalconvergence on Polycomb targets. Finally, our epigeneticontogeny signature was associated with worse outcomes ofpatients with clear-cell RCC. These data provide clues for theepigenetic basis of proximal versus distal tubule derived kid-ney tumors and suggest that interest of using epigenetictherapy in the most threatening subtypes of kidney cancers.

Malouf et al.

Clin Cancer Res; 2016 Clinical Cancer ResearchOF2

Digital restriction enzyme analysis of methylationAnalysis of DNA methylation was performed using Digital

Restriction Enzyme Analysis of Methylation (DREAM), which isbased on sequential DNA digestion with a pair of methylation-blocked and methylation-tolerant restriction enzymes SmaI/XmaI. These end sequences are repaired and then analyzed byultra-deep next-generation sequencing; thereafter, the methyla-tion status of each individual CCCGGG sites across the genomecan be determined quantitatively (15). Sequencing was per-formed using Illumina Hiseq2000 at MD Anderson core facility.Mapping tags and algorithm of analysis have been previouslyreported (15, 16).

Genome annotation of DREAM data and statistical analysisGenomic regions were defined according to NCBI coordinates

downloaded from the UCSC web site (hg18 version). Promoterswere defined as regions between�2,000bp from the transcriptionstart site (TSS) to þ2,000 bp from TSS for each Refseq transcript.To calculate a gene promoter methylation, we averaged themethylation level of all CpG sites located between �2,000 bpand 2,000 bp from TSS.

To calculate CpG sites differentially methylated between C1and C2 epi-clusters, a t test was performed and CG sites wereranked according to their smallest P value. A P value of less than0.05 was considered statistically significant.

RNA sequencing and analysisRNA extraction for 11 RCC samples, for which materials were

available, was done using the RNeasy Kit (Qiagen) according tothe manufacturer's instructions. RNA sequencing was per-formed using Illumina Hiseq2000 according to manufacturerinstructions and after library preparation according to theNuGen Ovation RNA-Seq System V2 protocol. For the analysis,we first counted the overlaps between the mapped reads andgenomic features, such as genes/exons using htseq-count scriptdistributed with the HTSeq package. We then performedbetween-sample normalization when testing for differentialexpression. To do so, we used the scaling factor normalizationmethod as it preserves the count nature of the data and hasbeen shown to be an effective means of improving the detec-tion of differential expression (17). To perform the normali-zation and the test for differential expression with a negativebinomial model between conditions, we chose to use theBioconductor package DESeq (version 1.11.0; ref. 18). Specif-ically, for each gene, a generalized linear model (GLM) was fitto compare the expressions of the two epi-clusters C1 and C2.The Benjamini–Hochberg method was used to control the falsediscovery rate (FDR; ref. 19).

Pathway analysis toolsGREAT prediction was performed using default settings (20).

We first define the differentially methylated regions (DMR) asgenomic regions located at þ and �50 base pair from differen-tially methylated CpG sites. The background used was genomicregions covered by DREAM. A significantly enriched pathway wasdefined as the one with at least 10 hyper foreground genes, a foldchange of at least 2 and a qFDR less than 0.05.

For Ingenuity Pathway Analysis (IPA), default settings wereused to analyze genes differentially expressed between C1 and C2epi-clusters. Pathways differentially expressed were ranked

according to their smallest P values and were considered statis-tically significant if their P value was less than 0.05.

Unsupervised clustering for promoter DNA methylationHierarchical clustering analysis was performed for CpG sites

located in promoters CGI and outside CGI, with at least 10 tags ofcoverage across samples, using the Pearson correlation coefficientas the distance metric and Ward's linkage rule.

Analysis of gene expression between C1 and C2 epi-clustersWe performed unsupervised hierarchical clustering using all

genes of the 11 RCC samples for which RNA seq were available.For Gene Set Enrichment Analysis (GSEA), gene sets were down-loaded from the Broad Institute MSigDB web site (21). Gene setpermutations were used to determine statistical enrichment of thegene sets using the differentially expressed genes between C1 andC2 epi-clusters. The association of C1 and C2 epi-clusters expres-sion patterns with those of specific regions of the nephron wasdone as previously reported (22) using datasets of Cheval andcolleagues (23).

Validation using TCGA datasetsWe extracted The Cancer Genome Atlas (TCGA) data from 271

samples of ccRCC and 66 chromophobes for which both DNAmethylation data assessed by Infinium 450K and RNAseq wereavailable (22, 24). Promoter DNA methylation was defined asregions spanning � 1,000 base pair from TSS. Supervised clus-tering using the 56 genes epi-signature were performed on all thewhole TCGA dataset (n ¼ 337) for both DNA methylation andgene expression. Accuracy of the method was then calculated forits ability to distinguish ccRCC from chromophobe cases.

Analysis of associations between the epigenetic signature andoverall survival of patients with ccRCCs

Supervised principal components (SPC) analysis was used toexamine association between the gene expression of the 56genes belonging to the epi-signature and overall survival of 463ccRCC cases from TCGA (25, 26). This method has previouslybeen used to examine associations between gene expressionprofiling data and survival in ccRCC (25, 26). We randomlydivided TCGA ccRCC RNA-seq samples into two datasets, oneof which was used as the training set (232 patients), andanother as validation set (231 patients). First, a modifiedunivariate Cox score was calculated for the association betweengene expression for each gene and overall survival, and geneswhose Cox score exceeded a threshold that best predictedsurvival were used to carry out supervised principal compo-nents analysis (PCA). To determine the Cox threshold, thetraining set was split and principal components were derivedfrom one half of the samples, and then used in a Cox model topredict survival in the other half. By varying the threshold ofCox scores and using 2-fold cross-validation, this process wasrepeated 10 times, and a threshold of 0 (averaged over 10separate repeats of this procedure) was used to generate theprincipal components subsequently used to predict outcome,where all 56 genes were used to build the model.

For each case, we used the first principal component in aregression model to calculate a SPC risk score that represents thesumof theweighted gene expression levels for each of the 56 gene.To validate the SPC predictor, we computed risk scores for each of

Epigenetic Basis of Renal Cell Carcinomas

www.aacrjournals.org Clin Cancer Res; 2016 OF3

the 231 cases in validation set using the model developed in the232 TCGA training set, and tested whether these scores werecorrelated with survival. To examine the role of individual genesin determining outcome, we computed importance scores forgenes. The importance score is equivalent to the correlationbetween each gene and the first supervised principal component.Higher positive importance score means higher risk (worse sur-vival), and lower negative importance score means lower risk(better survival).

ResultsDNA methylation is conserved among normal kidney samples

To determine the variability of DNA methylation in normalkidney, we compared the differences in DNA methylationbetween normal kidneys from two different individuals usingDREAM (15). Using a minimum threshold of 10 tags/CG sites,we identified common 36,035 CG sites in the two normalkidneys. Of note, 91.6% (n ¼ 33,018/36,035) of those siteshad at least a minimum coverage of 100 tags, allowing us toanalyze quantitatively subtle DNA methylation changes.

DNAmethylation levels for each CG site were then plotted toidentify correlations between the two normal kidneys. Wefound a high correlation (Spearman R ¼ 0.95, P < 0.0001),suggesting there are few variations in DNA methylation in thenormal kidneys. The estimated FDR of DREAM is 0.7% and2.1%, using 15% and 20% as a difference methylation cutoff,respectively. This was consistent with our previous results usingblood and breast samples (15, 16). To determine whether DNAmethylation alters gene expression, we examined the associa-tion of DNA methylation with promoter activity. Consistentwith our previous observations, we showed that promoterswith no methylation (�1%) were highly expressed (Supple-mentary Fig. S1A); conversely, those that had methylationlevels >1% were repressed (Supplementary Fig. S1A). Interest-ingly, promoters with methylation �10% were completelyrepressed (Supplementary Fig. S1A). Although the frequencyof methylated CG sites outside CGI was higher than in CGI,similar findings were observed for promoters located outsideCGIs (Supplementary Fig. S1B), suggesting little differencebetween the correlation of DNA methylation and expressionwhether promoters were located in CGI or outside CGI.

Unsupervised clustering of DNA methylation profilesdistinguishes RCC subtypes

To determine if RCC subtypes are associated with DNA meth-ylation signatures, we analyzed DNA methylation of 22 RCCtumors with different subtypes using DREAM, with a median of77million tags obtained by sample (range, 27–141million; Table1). We assessed quantitative DNA methylation of a median65,000 CpG sites (range, 51,000–156,000) per sample, with atleast 10 tags coverage per CpG site; when we filtered for CpG siteswith at least 100 tags coverage, we identified a median 43,000CpG sites (range, 33,000–62,000) covered per sample (Table 1).After merging all the samples, we ended upwith 36,035 CpG sitesfor which quantitative DNA methylation calculated on the basisof at least 10 tags coverage perCpGsite (Supplementary Table S1);of note, the majority of those CpG sites had more than 100 tagscoverage with a median sequencing depth of�1,850 tags/CG site(range, 10–697,312). Out of those, 18,800 and 16,566 sites werelocated in CpG islands and outside CpG islands, respectively. We

then filtered for CpG sites located in promoter CpG islands (n ¼13,593) and promoter outside CpG islands (n ¼ 3,086).

We then performed an unsupervised hierarchical clustering forpromoter DNAmethylation using CG sites. Our analysis revealedtwo similar epi-clusters independently of whether CG sites werelocated in CGI (CG islands) or outside (Fig. 1A and B). Interest-ingly, the first epi-cluster "C1" was associated with ccRCC, pRCC,MITF/TFE translocation RCC (tRCC), mucinous and spindle cellcarcinomas (MTS) and one sarcomatoid RCC case. On the otherhand, the second epi-cluster "C2" comprised oncocytoma, chro-mophobe, and one sarcomatoid sample. Sarcomatoid dediffer-entiation may occur in various RCC subtypes and after patholog-ical re-review of the sarcomatoid cases, we identified the sarco-matoid case in the C1 epi-cluster arose from ccRCC. Similarly, thesarcomatoid case in the C2 epi-cluster arose from chromophobeRCC. PCA confirmed the existence of the two epi-clusters (notshown).

Unsupervised clustering of nonpromoter CG sites locatedwithin and outside CGI yielded similar results (not shown),suggesting not only that tissue specificity is related to CGI butthat regulatory elements differentiating between proximal anddistal tubules are dispersed across the genome. These findingsupport distinct DNA methylation signatures associated withhistological RCC subtypes based on their cell ontogeny in thenephron.

C1 epi-cluster is characterized by coordinated methylation ofgene promoters

To determine the gene pathways regulated by DNA meth-ylation, we evaluated the differentially methylated promotersbetween the C1 and C2 epi-clusters. Using 15% difference ofmethylation as cutoff, we identified 40 CpG sites that har-bored lower methylation in C2 epi-cluster as compared withC1 epi-cluster. Interestingly, there was no CG site unmethy-lated in C1 epi-cluster with concurrent gain of DNA methyl-ation in the C2 epi-cluster, suggesting aberrant and coordi-nated DNA methylation in C1 epi-cluster. Importantly, theobserved gain of DNA methylation in C1 epi-cluster was notassociated with gene repression, except for the PLEK2 gene.This might be related to the fact that the majority of thoseCpG sites had measurable levels of methylation in C2 (�1%)and were thus already repressed, consistent with previousreports (27).

Our data indicate that promoters with some methylation(>1%) level were repressed when compared with those withoutany methylation level (Supplementary Fig. S1), suggesting thatsubtle changes in DNAmethylation may impact gene expression.Using the following criteria, at least 2% methylation gain forunmethylated CG sites (�1%) or demethylation below or equalto 1% for methylated CG sites (�2%), we estimated the FDR ofour method to be 0.01. In addition, as few CpG sites wereidentified using 15% difference methylation cutoff, we thusdecided to analyze CpG sites in CGI which were differentiallymethylated between C1 and C2 epi-clusters (irrespective of theirmethylation difference levels) and with consistent changes in theexpression of their related genes (Fig. 2A). Overall, 971 out of11,943 CG sites were differentially methylated (P < 0.05), regard-less of theirmethylation levels in C1 andC2 epi-clusters (Fig. 2A).Out of those, 95 CpG sites related to 73 genes were also differ-entially expressed between C1 and C2 clusters (P < 0.05; Fig. 2A;Supplementary Table S2). To evaluate for an association between

Malouf et al.

Clin Cancer Res; 2016 Clinical Cancer ResearchOF4

CpG site DNA methylation and gene expression, we divided the95 CG sites in C2 epi-cluster into two subgroups: the first encom-passes "unmethylated" (�1% methylation) CpG sites (n ¼ 63)while the second one containedmethylated (>1%)CpG sites (n¼32). Interestingly, the majority of unmethylated CpG belongingto C2 epi-cluster showed gains of DNA methylation in C1 epi-cluster, consistent with gene expression repression (Spearman r¼�0.46 (P < 0.0001; Fig. 2B). Average methylation gain wasestimated to be as low as 2.1% (P < 0.0001), suggesting thatsmallmethylation gainsmight be correlatedwith gene repression.Likewise, we found an opposite correlation between genes thathad "detectable level" of methylation in C2 epi-cluster and geneexpression, although this was less strong than for our "unmethy-lated" subgroup (r ¼ �0.37, P ¼ 0.03; Fig. 2C). IPA showed thatthose genes were enriched for IL8 (P ¼ 4.6�10�3), CXCR4 (P ¼0.02), and mTOR signaling pathways (P ¼ 0.03; SupplementaryTable S3).

RCC subtypes derived from the distal tubules display defects inDNA methylation

To identify differentially methylated CpG sites between the C2epi-cluster relative to both normal kidney and C1 epi-cluster, weused a differential DNAmethylation cutoff of 15% (FDR¼ 0.03).We identified a trend toward lower DNA methylation in 2.4% (n¼ 834/35366) of CG sites in C2 epi-cluster as compared withnormal kidneys; in contrast, 0.6% (n ¼ 220/35,366) of themgained DNAmethylation as compared with normal kidneys (Fig.

3A). This trend was not observed when we compared C1 epi-cluster to normal kidneys, as 3.8% (n¼ 1,363/35,366) and 3.2%(n ¼ 1,122/35,366) of differentially methylated CG sites in C1epi-cluster gained and lost DNA methylation, respectively, ascompared with normal kidneys (Fig. 3B).

The same trendwasmaintainedwhen differentiallymethylatedCG sites were analyzed between C2 epi-cluster relative to C1 epi-cluster. Indeed, genome-wide, C1 epi-cluster displayed 3-foldmore hypermethylation (5.6%; n ¼ 1,872/33,536) than hypo-methylation (2.2%; n ¼ 773/35,366) as compared with C2 epi-cluster. Of note, the majority of those CG sites were located farfrom the transcription start sites, mainly in gene bodies andintergenic regions (Fig. 3C). GREAT Gene Ontology (GO) toolsshow that differentiallyMethylatedRegions (DMR)were enrichedfor genes involved in negative regulation of angiogenesis, regu-lation of epithelial cell differentiation, and regulation of ARFprotein signal transduction (Fig. 3D; Supplementary Table S4).This is interesting because genes involved in epithelial differen-tiation are key genes for the development of kidney, such as PAX2,PAX8, GATA3, and LHX1. Furthermore, genes involved in theregulation of angiogenesis are key therapeutic targets in ccRCC.Finally, GREAT identified enrichment for two motifs in thoseDMR; the first matches to the peroxisome proliferative activatedreceptor, alpha (PPARA) gene (P ¼ 1.46e�11) and the second tothe CCAAT/enhancer binding protein (C/EBP), beta (CEBPB)gene (P¼ 3.5e�4; Supplementary Table S5). Importantly,CEBPBis one of the major transcription factors controlling the

Figure 1.

A, Unsupervised clustering of DNA methylation using CpG sites located in promoter CpG islands. B, Unsupervised clustering of DNA methylation using CpG siteslocated outside promoter CpG islands. Note that the analysis revealed 2 epi-clusters with C1 containing almost tumors with benign potential (except onesarcomatoid RCC case) and C2 containing tumors with potential malignant behavior.

Epigenetic Basis of Renal Cell Carcinomas

www.aacrjournals.org Clin Cancer Res; 2016 OF5

differentiation of a range of cell types (28). We thus conclude thatDMRs between C1 andC2 epi-clusters contain key genes involvedin epithelial differentiation, with several putative binding sites ofenhancers.

C1 epi-cluster transcriptome is enriched for activation of genesrelated to Polycomb

To define cell ontogeny of RCC subtypes, we examined ourgene expression data using an external gene expression datasetof normal kidney cells microdissected from distinct regions ofthe nephron (23). Through supervised analysis, we globallycompared the gene expression data of the C1 and C2 epi-clusters with those of each cell sample in the nephron structure,and found high mRNA expression correlations for C1 and C2with proximal and distal tubules of the nephrons, respectively(Fig. 4A).

To understand pathways that are specifically altered in ourRCC epi-clusters, we thus compared differentially expressedgenes between C1 and C2 epi-clusters using available RNAseqdata. Overall, we identified 4.7% (n ¼ 606/12,877) of genesthat were upregulated in C1 epi-cluster as compared with 4.9%(n ¼ 627/12,877) of them that were downregulated. GSEArevealed that upregulated pathways in C1 epi-cluster wereenriched for epithelial–mesenchymal transition (EMT; FDR <0.0001), stem cell signature (Fig. 4B) and Polycomb targetgenes (FDR < 0.0001; Fig. 4C). Furthermore, EZH2 targets werefound among the top altered pathways consistent with activa-tion of EZH2 in the C1 epi-cluster (Fig. 4D; FDR < 0.0001). Wethus looked at EZH2 expression and confirmed its overexpres-sion in C1 as compared with C2 epi-cluster (log2 foldchange¼2.7, P ¼ 0.001; Fig. 4E). These data highly suggestthat the difference between C1 and C2 epi-clusters might berelated to the activation of the Polycomb repressive complex 2(PRC2) within C1 epi-cluster.

Discovery of genes with promoter DNAhypermethylation in RCC

To discover genes repressed by DNA methylation and poten-tially novel candidate tumor suppressor genes in RCC, we ana-lyzed genes that gain promoter DNA methylation in our dataset.Promoter DNA methylation was defined as genes with less than2% methylation in normal kidney and that gain DNA methyla-tion�10%methylation in cancer. Overall, 713 out of 4,558 genesgain DNAmethylation as compared with normal kidneys. Out ofthose, 126 genes gained DNAmethylation in C2 epi-cluster while688 genes gained DNA methylation in C1 epi-cluster. DAVIDfunctional pathway analysis revealed that genes that gain DNAmethylation are involved in development (P ¼ 1.8 � 10�9) andcell differentiation (P ¼ 1.2 � 10�5). Furthermore, those wererelated to Homeobox genes (P ¼ 1.37 � 10�7) and displayedsequence-specificDNAbinding (P¼ 2.07� 10�6; SupplementaryTable S6).

As genes that gain DNA methylation in cancer have beenshown to be Polycomb targets in embryonic stem cells andadult stem/progenitor cells (29), we thus analyzed our resultsusing the chromatin marks of the embryonic stem (ES) cellsand found that 487 (68.3%) of gene promoters gaining DNAmethylation in our RCC samples were marked by H3K27me3in ES cells. These data are consistent with the existence of a"DNA hypermethylation module" in RCC and which has beenpreviously shown to encompass a portion of the Polycombtarget genes (29).

We thenmatched the list of our 126genesmethylated inC2 epi-cluster with the list of known tumor suppressor genes (30) andidentified 11 candidate genes. Out of them, only IRX1 wasmethylated in 2 out of 3 chromophobe RCCs.

On the other hand, 47 genes out of the 688 genes that gainedpromoter DNA methylation in C1 epi-cluster C1 had been pre-viously described as TSG. Out of the 205 genes methylated in at

Figure 2.

A, Flow chart showing the number ofCG sites in promoter CG islandscovered by DREAM and the numberof differentially methylated CGbetween C1 and C2 epi-clusters.B, Correlation between differentiallymethylated CpG sites between C2and C1 epi-clusters and geneexpression changes. Herein, onlyunmethylated CpG sites (�1%) of C2epi-cluster located in promoter CGislands and both differentiallymethylated and expressed ascompared with C1 epi-cluster aredepicted. C, Correlation betweendifferentially methylated CpG sitesbetween C2 and C1 epi-clusters andgeneexpression changes. Herein, onlymethylated CpG sites (>1%) of C2 epi-cluster located in promoter CG islandsand both differentially methylatedand expressed as compared with C1epi-cluster are depicted. Note thatfew of those methylated CG sites lostDNAmethylation in C1 epi-cluster andbecome expressed (green), while themajority gains DNA methylation andget repressed.

Malouf et al.

Clin Cancer Res; 2016 Clinical Cancer ResearchOF6

least 2 RCC samples, we identified 2 known TSGs, IRX1 andSOX11. IRX1 was methylated in 4 RCC cases and SOX11 wasmethylated in 3 RCC cases. The list of those 205 genes is reportedin Supplementary Table S7. We then asked if any of the 205 geneswhich we identified as methylated at least 2 RCC samples are alsomethylated and repressed in a large independent cohort of ccRCCfrom The Cancer GenomeAtlas (TCGA) project. Overall, 25 geneswere identified as methylated and repressed in at least 2% ofccRCC samples, themajority of themnever reported previously inccRCC. Those geneswere related to EMT (i.e., FOXC2 andOLMF1)and stimulation of endothelial cell proliferation and angiogenesis(i.e., SMOC2 and CLEC14A) pathways. The list of those genes,including the prevalence of their methylation in the ccRCCdataset, is reported in Supplementary Table S8. Finally, wefocused on the 5 tRCC and identified 85 genes that gained DNAmethylation (Supplementary Table S7). Out of those, 15 geneswere methylated in at least two samples (i.e., WNT3A, IRX4). Ofnote, the tumor suppressor gene IRX4 gained promoter DNA

methylation in the 2metastatic tRCC cases,whichwas not the casefor the three remaining cases.

Validation of the epi-signature in an independent datasetWe then asked if the 73 genes thatwe identified in our dataset as

differentially methylated and expressed between C1 and C2 epi-clusters could distinguish chromophobe from ccRCC, using anindependent data of RCC subtypes using Infinium 450K arrays.This dataset encompasses 41 new samples, including 4 normalkidneys and 37 tumor samples. Distribution of tumor samples inthis independent dataset was as follows: ccRCC (n ¼ 16), chro-mophobe RCC (n¼ 6), oncocytoma (n¼ 8), papillary type II RCC(n¼ 6), and translocation RCC (n¼ 1). To do so, we first filteredfor gene promoters with data available for bothDNAmethylationb-values and gene expression; out of those 73 genes, data wereavailable for 56 genes (Supplementary Table S9). Supervisedclustering analysis using our 56 genes signature confirmed ourprevious findings about classification of kidney tumors in two

100

80

60

40

20

0

100

80

60

40

20

00 20 40 60 80 100 0 20 40 60 80 100

Avg

met

hyla

tio

n C

2 (%

)

Avg

met

hyla

tio

n C

1 (%

)

Methylation normal (%) Methylation normal (%)

35%

30%

25%

20%

15%

10%

5%

0%

Reg

ion

-gen

e as

soci

atio

ns

Distance to TSS (kb)

<–50

0–5

00 to

–50

–50

to –

5

–5 to

0

0 to

5

5 to

50

50 to

500

>500

Foreground genomic regions

Background genomic regions

0 1 2 3 4 5 6 7 8 9

9.09

7.04

6.55

6.51

6.26

6.18

5.99

5.74

5.59

4.84

–log10 (Hypergeometric P)

Fat cell differentiation

Negative regulation of angiogenesis

Regulation of epithelial cell differentiation

Hormone biosynthetic process

Response to endoplasmic reticulum stress

Negative regulation of myeloid cell differentiation

Regulation of ARF protein signal transduction

Muscle organ morphogenesis

ER-nucleus signaling pathway

Cardiac muscle tissue morphogenesis

A

C D

B

Figure 3.

A, Square graph of CpG sites (green: CpG islands; red: outside CpG islands) with statistically significant difference of DNA methylation between C2 epi-cluster andnormal kidneys. Note the tendency toward global hypomethylation of C2 epi-cluster with almost no gain of DNA methylation in CGI. B, Square graph ofCpG siteswith statistically significant difference of DNAmethylation between C1 epi-cluster and normal kidneys. Note the tendency toward both hypomethylation ofCpG sites located outside CGI and hypermethylation of CpG sites in CGI. C, Distribution of differentially methylated CpG sites between C1 and C2 epi-clustersaccording to the distance from transcription start site (TSS) as assessedbyGREAT tool.D,GeneOntology functional annotations for differentiallymethylated regionsbetween C1 and C2 epi-clusters as identified by GREAT tool.

Epigenetic Basis of Renal Cell Carcinomas

www.aacrjournals.org Clin Cancer Res; 2016 OF7

major epi-clusters C1 and C2 (Supplementary Fig. S2). Of note,C2 contains the majority of oncocytoma and chromophobe RCCcases (n ¼ 11/13) as compared with C1 that encompasses theremaining RCC subtypes, including ccRCC, papillary RCC, andtranslocation RCC (n ¼ 21/24; P ¼ 0.0001).

Independent validation of the epi-signature in TCGA datasetWe furthermore performed a supervised clustering based on

either DNA methylation (Fig. 5A) or gene expression (Fig. 5B)using this ontogeny epi-signature in an independent datasetfrom TCGA project encompassing 271 ccRCC and 66 chro-mophobe RCC. Importantly, hierarchical clustering revealedtwo distinct clusters with 94.4% and 97.6% accuracy fordistinguishing chromophobe from ccRCC, respectively.Indeed, using gene expression signature, 97% (n ¼ 64/66)of chromophobe and 97.8% (n ¼ 265/271) ccRCC wereclassified correctly; on the other hand, using DNA methylationsignature, 86.4% (n ¼ 57/66) chromophobe and 96.3% (n ¼261/271) ccRCC were classified correctly. Of note, severalccRCC that clustered with chRCC using our epi-genetic sig-nature were misclassified and were chRCC or papillary ccRCCfeatures, as we recently demonstrated in our long non-coding

RNA subtype classification of ccRCC (ref. 6; SupplementaryTable S10).

Epigenetic signature is associated with outcome ofpatients with ccRCCs

We then asked whether epigenetic signature is capable ofpredicting outcome for patients with ccRCC. To do so, we thusexamined the association between the gene expression of the 56genes defining the epi-signature and overall survival of 463patients with ccRCC from TCGA. First, we randomly dividedTCGA ccRCC RNA-seq samples into two datasets, one of whichwas used as the training set (232 patients) and another asvalidation set (231 patients). For each case, we used the firstprincipal component to calculate an SPC risk score (Supplemen-tary Table S11). To validate the SPC predictor, we computed riskscores for each of the 231 cases in the validation set using themodel developed in the 232 TCGA training set (SupplementaryFig. S3A). The risk score based on the supervised PCA wassignificantly associated with poor outcome in the validationcohort (Log-rank test P ¼ 0.0004; Supplementary Fig. S3B).Patients were then classified as being high or low risk accordingto the calculated SPC risk score. To examine the role of individual

Figure 4.

A, Heat maps showing intersample correlations (red, positive) between mRNA profiles of RCC (columns) belonging to C1 and C2 epi-clusters and mRNA profiles ofnephron anatomical sites (rows). Association of C1 and C2 epi-clusters expression patterns with those of specific regions of the nephron is depicted. The8 kidney nephron regions evaluated are Glom, glomerulus; S1 and S3, the proximal tubule; mTAL, medullary thick ascending limb of Henle's loop; cTAL, cortical thickascending limb of Henle's loop; DCT, distal convoluted tubule; CCD, cortical collecting duct; OMCD, outer medullary collecting duct. B–D, GSEA for differentiallyexpressed genes in C1 as compared with C2 epi-clusters. E, Box plot for EZH2 gene expression levels in normal kidneys and C1 and C2 epi-clusters.

Malouf et al.

Clin Cancer Res; 2016 Clinical Cancer ResearchOF8

genes in determining outcome, we computed importance scoresfor genes. Finally, we asked whether there are associationsbetween mutational load, ccRCC TCGA transcriptomic four-sub-group classification and somatic mutations; we discovered thatour 56-gene signature was highly correlated with TCGA transcrip-tomic classification (Supplementary Table S12) but not withmutational load. These data highly suggest that our epigeneticontogeny signature is correlated with ccRCC transcriptomic clas-sification and is useful in predicting outcome of patients withccRCC.

DiscussionDNA methylation is a heritable covalent modification that is

developmentally regulated, controls stemness, and is critical intissue-type definition. Our data demonstrate the utility of ana-lyzing the DNA methylation profiles of RCC subtypes that arisefrom either proximal or distal cells of the nephron to define anepigenetic basis of cell ontogeny in kidney cancers. Our keyfindings include (i) RCC subtypes can be grouped into twomajorepi-clusters: C1, which encompasses ccRCC, pRCC, MTS, tRCC;C2, which comprises oncocytoma, chromophobe RCC; (ii) dif-ferentially methylated regions between C1 and C2 occur in genebodies and intergenic regions, instead of gene promoters with afunctional convergence on Polycomb targets; (iii) methylationdefects in C2 epi-clusters; (iv) the identification of a 56-gene epi-

signature charting kidney cell ontogeny that was predictive ofoutcomes in ccRCC.

Historically, the cells of origin of different RCC subtypes havebeen controversial, and hypotheses are based on morphologicalsimilarities between RCC subtypes and proximal or distal cells ofthe nephron (7). Those observations have been corroborated byHeidelberg classification which added specific genetic aberrationsto morphology (8). Recent studies that incorporate DNA meth-ylation profiling have identified chromophobe RCC-specificmethylation or differentially methylated regions between RCCtumors and adjacent uninvolved kidney (31, 32).Our data furtherdefine the epi-clusters that can discriminate between additionalRCC subtypes (ccRCC, pRCC, MTS, tRCC vs. oncocytoma, chro-mophobe RCC) and, for the first time, to our knowledge, define aepigenetic basis for proximal versus distal derived tumors using atraining and an independent dataset.

The DREAM technique that incorporates next-generationsequencing to analyze DNA methylation allows accurate quanti-tation, thus allowing us to detect differences through the kidneymethylome. First, we identified two main epi-clusters; C1 epi-cluster encompasses ccRCC, pRCC,MTS, and tRCC. C2 epi-clustercomprises oncocytomaandchromophobeRCC. Interestingly, ourdata support an epigenetic basis for the Heidelberg classification,suggesting that MTS and tRCC may also arise from proximaltubules. Second, we discovered that much of the differentiallymethylated regions distinguishing C1 and C2 epi-clusters occuroutside of gene promoters; motif analysis uncovers binding sites

DNA Methylation

RCC Subtypes

Gene expression

ChromophobeccRCC

Subtype Subtype2

1

0

–1

–2

0.8

0.6

0.4

0.2

A B

Figure 5.

A, Supervised clustering of promoter DNA methylation using the 56 genes epi-signature in TCGA dataset of ccRCC and chromophobe samples. B, Supervisedclustering of gene expression using the 56 genes epi-signature in TCGA dataset of clear-cell cell carcinomas and chromophobe samples.

Epigenetic Basis of Renal Cell Carcinomas

www.aacrjournals.org Clin Cancer Res; 2016 OF9

related to transcription factors PPARA andCEBPB; of note, PPARAhas been reported to be expressed predominantly in proximaltubules andmedullary thick ascending limbs, in contrast to PPARgamma, which is exclusively expressed in medullary collectingduct and papillary urothelium (33). Furthermore, CEBPB isconsidered to be one of themajor transcription factors controllingthe differentiation of a range of cell types (28). Altogether, thesedata corroborate the dynamic regulation of CpG DNA methyla-tion during normal development that mainly occurs at distalregions from the transcription start sites (34). Third,weobserved amethylation defect in C2 epi-cluster consistent with a recentreport using Infinium arrays in a selected cohort of chromophobe(31), and most significantly we discovered that differentiallyexpressed genes between C1 and C2 epi-clusters were targets ofPolycomb; as expected, the methyltransferase EZH2 was over-expressed in C1 epi-cluster, which might be related to largerH3K27me3 domains in C1 epi-cluster as compared with C2. Wespeculate that activation of the PRC2 in the proximal tubules cellof origin might explain the aggressiveness of derived tumorsubtypes in C1 epi-cluster, in contrast to distal tubules cell oforigin giving rise to oncocytoma and chromophobe, which haveusually good outcome. We thus suggest that epigenetic therapiesmay have utility against a broad range of themost life-threateningkidney cancers, independent of genotype and morphologicalphenotype.

Finally, we have established an epi-signature ontogeny-basedclassifier composed of 56 genes. Using an independent cohortfrom TCGA comprising ccRCC and chromophobe samples, wevalidated the high accuracy of this ontogeny epi-signature fordistinguishing tumors arising either from the proximal tubules orfrom the distal tubules of the nephron. Of note, this is consistentwith the distinct pattern of genetic mutations according to kidneycancer ontogeny. Indeed, themutation rates per exome for ccRCC(n¼ 43) and pRCC (n¼ 59) are by far higher than chromophobe(n ¼ 17) and oncocytoma (n ¼ 15) RCC (35). One explanationmight be chromatin organization of kidney cell of origin asmutations rates in cancer genomes are associated with epigeneticarchitecture of human cells (36); indeed, heterochromatin-asso-ciated histone modification can account for more than 40% ofmutation-rate variation (36). Our 56 genes epi-signature was alsoable to predict survival of patients with ccRCC. One explanationmight be that ccRCC arising from distal part of the proximalnephron harbor a good outcome as recently showed by Buttnerand colleagues using TCGA datasets (37).

In summary, we define herein an epigenetic basis of kidneytumors ontogeny and provide the first mapping of methylomeepi-signature across different histological RCC subtypes. In addi-tion, we showed that our epigenetic ontogeny signature is alsocapable of predicting patients outcome.

Disclosure of Potential Conflicts of InterestJ.A. Karam is a consultant/advisory boardmember for Pfizer. N.M. Tannir is a

consultant/advisory board member for Exelixis, GlaxoSmithKline, Novartis,and Pfizer and reports receiving commercial research grants from Bristol-MyersSquibb, Exelixis, GlaxoSmithKline, Novartis, and Pfizer. No potential conflictsof interest were disclosed by the other authors.

Authors' ContributionsConception and design: G.G. Malouf, T.H. Ho, J.-P. Spano, C.G. Wood,N.M. TannirDevelopment of methodology: G.G. Malouf, T.H. Ho, F. Allanick, J. Jelinek,N.M. TannirAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): G.G. Malouf, J.A. Karam, P. Tamboli, R. Mouawad,C.G. Wood, N.M. TannirAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis):G.G.Malouf, X. Su, J. Zhang, C.J. Creighton, T.H. Ho,Y. Lu, N.J.-M. Raynal, J.A. Karam, R. Mouawad, J.-P. Spano, C.G. Wood,J. Jelinek, N.M. TannirWriting, review, and/or revision of the manuscript: G.G. Malouf, X. Su,T.H. Ho, N.J.-M. Raynal, J.A. Karam, P. Tamboli, R. Mouawad, J.-P. Spano,C.G. Wood, J. Jelinek, N.M. TannirAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): G.G. Malouf, F. Allanick, R. Mouawad, N.M.TannirStudy supervision: G.G. Malouf, D. Khayat, C.G. Wood, N.M. Tannir

Grant SupportThe authors acknowledge the AVEC Foundation for their support of research

in kidney cancer. T.H. Ho is supported by funding from the ASCO YoungInvestigator Award from the Kidney Cancer Association and NIH grant K12CA90628. C.J. Creighton is supported by NIH CA125123 grant. This work wasalso supported by NIH/NCI award number P30 CA016672 and the Genitouri-nary Cancers Program of the CCSG shared resources at MD Anderson CancerCenter.

The costs of publication of this articlewere defrayed inpart by the payment ofpage charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received June 5, 2015; revised April 28, 2016; accepted May 12, 2016;published OnlineFirst June 2, 2016.

References1. Lopez-Beltran A, Scarpelli M, Montironi R, Kirkali Z. 2004WHO classifi-

cation of the renal tumors of the adults. Eur Urol 2006;49:798–805.2. Rini BI, Campbell SC, Escudier B. Renal cell carcinoma. Lancet

2009;373:1119–32.3. Amin MB, Paner GP, Alvarado-Cabrero I, Young AN, Stricker HJ, Lyles RH,

et al. Chromophobe renal cell carcinoma: histomorphologic characteristicsand evaluation of conventional pathologic prognostic parameters in 145cases. Am J Surg Pathol 2008;32:1822–34.

4. Przybycin CG, Cronin AM, Darvishian F, Gopalan A, Al-Ahmadie HA, FineSW, et al. Chromophobe renal cell carcinoma: a clinicopathologic study of203 tumors in 200 patients with primary resection at a single institution.Am J Surg Pathol 2011;35:962–70.

5. Malouf GG, Monzon FA, Couturier J, Molinie V, Escudier B, Camparo P,et al. Genomic heterogeneity of translocation renal cell carcinoma. ClinCancer Res 2013;19:4673–84.

6. Malouf GG, Zhang J, Yuan Y, Comperat E, Roupret M, Cussenot O, et al.Characterization of long non-coding RNA transcriptome in clear-cellrenal cell carcinoma by next-generation deep sequencing. Mol Oncol2014;9:32–43.

7. Kuehn A, Paner GP, Skinnider BF, Cohen C, Datta MW, Young AN,et al. Expression analysis of kidney-specific cadherin in a widespectrum of traditional and newly recognized renal epithelial neo-plasms: diagnostic and histogenetic implications. Am J Surg Pathol2007;31:1528–33.

8. Kovacs G, Akhtar M, Beckwith BJ, Bugert P, Cooper CS, Delahunt B, et al.The Heidelberg classification of renal cell tumours. J Pathol 1997;183:131–3.

9. Oosterwijk E, Rathmell WK, Junker K, Brannon AR, Pouliot F,Finley DS, et al. Basic research in kidney cancer. Eur Urol 2011;60:622–33.

Malouf et al.

Clin Cancer Res; 2016 Clinical Cancer ResearchOF10

10. Lim E, Vaillant F, WuD, Forrest NC, Pal B, Hart AH, et al. Aberrant luminalprogenitors as the candidate target population for basal tumor develop-ment in BRCA1 mutation carriers. Nat Med 2009;15:907–13.

11. Jones PA.Functions of DNA methylation: islands, start sites, gene bodiesand beyond. Nat Rev Genet 2012;13:484–92.

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

13. Heyn H, Esteller M. DNA methylation profiling in the clinic: applicationsand challenges. Nat Rev Genet 2012;13:679–92.

14. Issa JP.DNA methylation as a clinical marker in oncology. J Clin Oncol2012;30:2566–8.

15. Jelinek J, Liang S, Lu Y, He R, Ramagli LS, Shpall EJ, et al. Conserved DNAmethylation patterns in healthy blood cells and extensive changes inleukemia measured by a new quantitative technique. Epigenetics2012;7:1368–78.

16. Malouf GG, Taube JH, Lu Y, Roysarkar T, Panjarian S, Estecio MR, et al.Architecture of epigenetic reprogramming following Twist1 mediatedepithelial–mesenchymal transition. Genome Biol 2013;14:R144.

17. RobinsonMD, Oshlack A. A scaling normalizationmethod for differentialexpression analysis of RNA-seq data. Genome Biol 2010;11:R25.

18. Anders S, Huber W. Differential expression analysis for sequence countdata. Genome Biol 2010;11:R106.

19. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practicaland powerful approach to multiple testing. J Royal Stat Soc Series B1995;57:289–300.

20. McLeanCY, BristorD,HillerM,Clarke SL, Schaar BT, LoweCB, et al. GREATimproves functional interpretation of cis-regulatory regions. Nat Biotech-nol 2010;28:495–501.

21. SubramanianA, TamayoP,Mootha VK,Mukherjee S, Ebert BL,GilletteMA,et al. Gene set enrichment analysis: a knowledge-based approach forinterpreting genome-wide expression profiles. Proc Natl Acad Sci U S A2005;102:15545–50.

22. Davis CF, Ricketts CJ, Wang M, Yang L, Cherniack AD, Shen H, et al.The somatic genomic landscape of chromophobe renal cell carcinoma.Cancer Cell 2014;26:319–30.

23. Cheval L, Pierrat F, Rajerison R, Piquemal D, Doucet A. Of mice and men:divergence of gene expression patterns in kidney. PLoS ONE 2012;7:e46876.

24. The Cancer Genome Atlas Research N. Comprehensive molecularcharacterization of clear cell renal cell carcinoma. Nature 2013;499:43–9.

25. Bair E, Tibshirani R. Semi-supervised methods to predict patient survivalfrom gene expression data. PLoS Biol 2004;2:E108.

26. Bair E HH, Debashis P, Tibshirani R. Prediction by Supervised PrincipalComponents. J Am Stat Assoc 2006;101:119–37.

27. Cedar H, Bergman Y. Linking DNAmethylation and histonemodification:patterns and paradigms. Nat Rev Genet 2009;10:295–304.

28. Nerlov C.The C/EBP family of transcription factors: a paradigm for inter-action between gene expression and proliferation control. Trends Cell Biol2007;17:318–24.

29. EaswaranH, Johnstone SE, VanNeste L,Ohm J,Mosbruger T,WangQ, et al.A DNA hypermethylationmodule for the stem/progenitor cell signature ofcancer. Genome Res 2012;22:837–49.

30. ZhaoM, Sun J, Zhao Z. TSGene: aweb resource for tumor suppressor genes.Nucleic Acids Res 2013;41:D970–6.

31. Slater AA, Alokail M, Gentle D, Yao M, Kovacs G, Maher ER, et al. DNAmethylation profiling distinguishes histological subtypes of renal cellcarcinoma. Epigenetics 2013;8:252–67.

32. Lasseigne BN, Burwell TC, Patil MA, Absher DM, Brooks JD, Myers RM.DNA methylation profiling reveals novel diagnostic biomarkers in renalcell carcinoma. BMC Med 2014;12:235.

33. Guan Y, Zhang Y, Davis L, Breyer MD. Expression of peroxisome prolif-erator-activated receptors in urinary tract of rabbits and humans. Am JPhysiol 1997;273:F1013–22.

34. ZillerMJ,GuH,Muller F,Donaghey J, Tsai LT, KohlbacherO, et al. Chartinga dynamic DNA methylation landscape of the human genome. Nature2013;500:477–81.

35. Durinck S, Stawiski EW, Pavia-Jimenez A,Modrusan Z, Kapur P, Jaiswal BS,et al. Spectrum of diverse genomic alterations define non-clear cell renalcarcinoma subtypes. Nat Genet 2014;47:13–21.

36. Schuster-Bockler B, Lehner B. Chromatin organization is a major influ-ence on regional mutation rates in human cancer cells. Nature2012;488:504–7.

37. Buttner F, Winter S, Rausch S, Reustle A, Kruck S, Junker K, et al. Survivalprediction of clear cell renal cell carcinoma based on gene expressionsimilarity to the proximal tubule of the nephron. Eur Urol 2015;68:1016–20.

www.aacrjournals.org Clin Cancer Res; 2016 OF11

Epigenetic Basis of Renal Cell Carcinomas