Powered by - Future Medicine · detection and quantification of cancer-specific meth-ylation changes might help to diagnose and subtype UC, especially conveniently from urinary samples,

Cancer biomarkers TOP ARTICLES SUPPLEMENT

CONTENTSREVIEW: DNA methylation in urothelial carcinoma Epigenomics (Epub ahead of print)

REVIEW: Long noncoding and circular RNAs in lung cancer: advances and perspectives Epigenomics Vol. 8 Issue 9

REVIEW: Epigenetic drift, epigenetic clocks and cancer risk Epigenomics Vol. 8 Issue 5

RESEARCH ARTICLE: Epigenome-wide analysis of DNA methylation reveals a rectal cancer-specific epigenomic signature Epigenomics Vol. 8 Issue 9

REVIEW: The role of epigenetics and long noncoding RNA MIAT in neuroendocrine prostate cancer Epigenomics Vol. 8 Issue 5

Powered by

http://www.futuremedicine.com/doi/full/10.2217/epi-2016-0064









http://www.futuremedicine.com/doi/full/10.2217/epi.16.6

http://www.futuremedicine.com/doi/full/10.2217/epi.16.6

https://www.epigenomicsnet.com/

Epigenomics (Epub ahead of print) ISSN 1750-1911

part of

DNA methylation in urothelial carcinoma

Wolfgang A Schulz*,1 & Wolfgang Goering2

1Department of Urology, Medical Faculty,

Heinrich Heine University Duesseldorf,

Moorenstr. 5, 40225 Düsseldorf,

Germany 2Department of Pathology, Medical

Faculty, Heinrich Heine University

Duesseldorf, Germany

*Author for correspondence:

Tel.: +49 (0)211 81 15845

Fax: +49 (0)211 81 15846

[email protected]

Review

10.2217/epi-2016-0064 © 2016 Future Medicine Ltd

Epigenomics

Review 2016/09/308

10

2016

DNA methylation alterations are common in urothelial carcinoma, a prevalent cancer worldwide caused predominantly by chemical carcinogens. Recent studies have proposed sets of hypermethylated genes as promising diagnostic and prognostic biomarkers from urine or tissue samples, which require validation. Other studies have revealed intriguing links between specific carcinogens and DNA methylation alterations in cancer tissues or blood that might clarify carcinogenesis mechanisms and aid prevention. Like DNA methylation alterations, mutations in chromatin regulators are frequent, underlining the importance of epigenetic changes. However, the relations between the two changes and their functions in urothelial carcinogenesis remain unclear. Transcription factor genes with altered methylation deserve particular interest. Elucidating the functional impact of methylation changes is a prerequisite for their therapeutic targeting.

First draft submitted: 25 May 2016; Accepted for publication: 4 August 2016; Published online: 14 September 2016

Keywords: bladder cancer • carcinogenesis • chromatin regulator • CpG-island hypermethylation • DNA methylation • retroelement hypomethylation • transcription factors • urine biomarkers • urothelial carcinoma

Urothelial carcinomaUrothelial carcinoma (UC) is the most com-mon cancer of the urinary bladder, afflicting also other segments of the urinary tract. UC is predominantly caused by chemical car-cinogens originating from cigarette smok-ing, various occupational and environmental exposures, certain drugs and herbal rem-edies, as well as endogenous sources. Rarer histological subtypes of bladder cancer, with different etiologies, include squamous cell carcinoma (SCC) and adenocarcinoma. SCC in particular arises in the context of chronic bladder inflammation, most often in patients affected by schistosomiasis.

The frequency and impact of UC are often underestimated, but in fact its incidence ranges among the top ten cancers world-wide; in many industrialized countries, it is the fourth or fifth most common cancer in

males. Its underestimation probably ensues, because this very heterogeneous disease com-prises two subclasses with quite different clinical behavior. Most UCs are hyperplastic tumors with a papillary structure that are restricted to the urothelial layer (stage pTa). These tumors can be removed by transure-thral surgery, are usually not life-threatening and are therefore not efficiently registered in many countries. However, papillary UCs are not generally harmless. They often occur multifocally, have a strong tendency to recur, and especially high-grade tumors may prog-ress to invasive stages. Therefore, the patients do require treatment and long-time moni-toring for recurrences and progression by regular cystoscopies. Invasive UCs represent 20–30% of the cases and especially carcino-mas having invaded the muscle layers of the bladder (muscle-invasive [MI-UC], stages

For reprint orders, please contact: [email protected]

10.2217/epi-2016-0064 Epigenomics (Epub ahead of print) future science group

Review Schulz & Goering

≥pT2) often metastasize and require radical surgery. Surgery can be supported by neoadjuvant or adjuvant cisplatin-based combination chemotherapy, which is, however, not curative in metastatic cases. Thus, about 120,000 persons are affected by bladder cancer and about 40,000 die of the disease in the EU annually. UC causes not only substantial morbidity and mortal-ity, but also incurs considerable costs for the health sys-tems due to the prolonged treatments and long-term monitoring.

Unfortunately, few advances have been made in improving diagnostic and therapeutic procedures for UC over the last decades. The established therapeu-tic modalities like surgery and cisplatin-based che-motherapy have evidently been pushed to their limits and patient survival has only marginally improved. No molecularly targeted therapy has so far achieved substantial benefit in the clinic and none is being used routinely for UC [1]. Therapeutic antibodies target-ing T-cell activation checkpoints have shown promise for the treatment of advanced stage UC [2], but they require further development and predictive biomark-ers for these therapies are unsatisfactory. Diagnostic procedures still rely essentially on cystoscopy, since no robust biomarker assays are available for tumor detec-tion, monitoring, prognosis and prediction of response to therapy.

The widespread underestimation of the disease may also contribute to the poor funding of research on bladder cancer compared with other major cancers such as prostate, breast or colorectal cancer [3]. Perhaps for that reason, comprehensive studies of UC genomics have only recently been conducted and published [4–6]. These studies have substantiated the consensus that molecular changes differ between the papillary and MI carcinomas. In keeping with their hyperprolifera-tive phenotype, papillary UC appears to be driven by mutations in growth factor receptors and signal trans-duction pathway components also involved in nor-mal urothelial development and regeneration, such as FGFR3, EGFR, ERBB2, PI3K and HRAS [7]. They are complemented by changes preventing senescence such as promoter mutations activating the hTERT gene and deletions of the CDKN2A locus [7]. MI-UC are char-acterized by severe defects in cell cycle control, includ-ing RB1 or CDKN2A deletions and mutations, ampli-fication and overexpression of E2F1, E2F3, CCND1/Cyclin D1 or CCNE1/Cyclin E, and by inactivation of p53 [7].

Combined analysis of genetic alterations and expres-sion profiles has led to the recognition of molecular subtypes within the group of invasive UC [7], of which two are relatively well defined. The ‘basal’ subtype is characterized by expression of cytokeratins, EGFR and

other markers of the basal urothelial layer; its expres-sion profile reflects among others the activity of the transcription factor p63 [8,9]. Basal UC often expresses aberrantly markers of squamous differentiation. The ‘luminal’ subtype expresses markers of advanced uro-thelial differentiation like uroplakins and cytokeratin 20, harbors mutations including those found in papil-lary tumors, and its expression profile reflects the activ-ity of transcription factors involved in urothelial dif-ferentiation, such as FOXA1, GATA3 and PPARγ [8,9]. It is still contentious whether these subtypes arise as a consequence of different mutations or from different cells of origin, as it has recently been proposed that the urothelium originates from two distinct precursor cell populations [10]. Notably, the basal and luminal sub-types comprise around two thirds of all invasive UC and no consensus has been reached on the classification of the remaining cancers [11].

An unexpected finding in the recent comprehensive genomic studies was a high prevalence of mutations in chromatin regulators, possibly the highest across all cancer types [4–6]. These inactivating mutations affect at varying frequencies and often alternatively the his-tone acetyltransferases CREBBP/CBP, EP300/p300, the histone methyltransferases MLL1-3, the his-tone demethylase KDM6A/UTX and the chromatin remodelers ARID1A and SMARCA4. These factors collaborate in one epigenetic pathway, which controls gene activation during development and differentia-tion [12,13]. These mutations, together with the DNA methylation alterations discussed below, suggest that epigenetic alterations are particularly important in UC pathogenesis.

DNA methylation in cancerDNA methylation as defined in this review is the C5 methylation of cytosine bases in a CpG dinucleotide; other modifications are poorly studied in UC. DNA methylation serves as an epigenetic mark to organize the complex human genome. It is introduced by DNA methyltransferases that use S-adenosylmethionine (SAM) as a methyl group donor. Among the three human enzymes, DNMT1 is considered to be mainly responsible for maintaining patterns of DNA methyla-tion following DNA replication, whereas DNMT3A and DNMT3B are primarily responsible for introduc-ing DNA methylation at novel sites, but also appear to support DNMT1 in its maintenance function. Notably, all three proteins have additional functions in organizing chromatin, beyond their enzymatic activity [14].

DNA methylation at cytosine C5 is biologically more stable than many other epigenetic marks, but can be removed by passive or active mechanisms. Passive

10.2217/epi-2016-0064www.futuremedicine.comfuture science group

DNA methylation in urothelial carcinoma Review

demethylation occurs if newly incorporated cytosines are not remethylated following DNA replication and DNA methylation becomes diluted over successive rounds of replication. Active mechanisms comprise removal of methylcytosine by nucleotide excision repair or specific oxidation to 5-hydroxymethylcyto-sine catalyzed by TET dioxygenases. Hydroxymeth-ylcytosine may persist or eventually be replaced by cytosine via further oxidation and base excision repair pathways [15].

DNA methylation is unequally distributed across the human genome. Overall, approximately 70–80% of the CpG sites are methylated, in particular those in heterochromatic repeats, retroelements, most inter-genic regions and in the bodies of transcribed genes. Notably, intergenic regions of the genome often have a relatively low CpG density [16]. Clustered CpG sites located around the transcriptional start site or the 3′-end of genes, designated CpG islands, are usually completely unmethylated. Adjacent less CpG-rich sequences, termed CpG shores, and enhancer sequences, are often partly methylated. Differential methylation at enhancers accompanies cell differentia-tion and methylation of selected promoters is impor-tant for the determination of cell lineages and for differentiation from stem cells [17].

In cancers, DNA methylation patterns are generally disturbed, albeit to various extents, compared with the corresponding normal tissues (Figure 1A). In many cancers, overall methylcytosine content is decreased, which is reflected in the hypomethylation of various sequences, especially of retroelements and CpG-rich satellites as well as of gene bodies. Conversely, some CpG islands, which are otherwise completely unmeth-ylated, become aberrantly hypermethylated [18]. As in normal tissue differentiation, CpG-island shores and enhancers are often differentially methylated in can-cers [19,20]. DNA hypermethylation of CpG-island promoters is commonly associated with transcrip-tional repression of the affected promoter, leading to decreased gene expression or alternative promoter usage. Conversely, global hypomethylation in cancer may promote genomic instability, disturb splicing and allow re-expression of silenced genes and retroele-ments [18]. In many instances, altered DNA methyla-tion interacts with changes in other epigenetic regu-lators, such as chromatin regulator proteins [21]. In particular, in many cancers, a subset of genes prone to DNA hypermethylation are Polycomb targets. These genes are often marked by H3K27 trimethylation and weakly expressed in normal tissues. Many encode developmental regulators, especially transcription fac-tors [18]. Their hypermethylation likely involves tar-geting of DNA methylation by Polycomb complexes,

but may be associated paradoxically with loss of Poly-comb binding after DNA hypermethylation has been established [18].

DNA methylation changes can be exploited for can-cer diagnostics, because they can distinguish cancer from normal tissues, different cancer types or subtypes from each other and in some instances are related to tumor progression and thereby to prognosis. Moreover, DNA methylation can be analyzed in body fluids and formalin-fixed paraffin-embedded (FFPE) tissues by several highly sensitive and specific analytical tech-niques. Hypermethylation of CpG islands, in partic-ular, lends itself to clinical applications because it is essentially absent in normal tissues and can be highly cancer-type specific.

Directions of research on DNA methylation in UCConsidering this briefly sketched background, there are several obvious questions concerning DNA meth-ylation in UC. With respect to clinical applications, detection and quantification of cancer-specific meth-ylation changes might help to diagnose and subtype UC, especially conveniently from urinary samples, as all UC abut on the lumen of the urinary tract. Sensitive and specific methylation biomarkers should be particu-larly helpful for monitoring patients with nonmuscle invasive (NMI) UC for recurrences to reduce the num-ber of cystoscopies. Given the biological and clinical heterogeneity of UC, methylation-based biomarkers distinguishing subtypes and providing prognostic information are desirable. Considering the limited effi-cacy of chemotherapy and of immunotherapy, drugs targeting DNA methylation might be explored. With respect to the disease pathophysiology, one would like to evaluate to which extent DNA methylation changes contribute functionally to the disease. Also, consid-ering the high prevalence of mutations in chromatin regulators in UC, one wonders how these alterations relate to and interact with those in DNA methylation. In the following sections, we will attempt to address such questions as far as they can be answered by recent publications (Figure 1B).

As in other cancer entities, research on DNA meth-ylation in UC has gained impetus from the introduc-tion of novel technologies permitting to study DNA methylation more comprehensively. Instead, research before 2010 mainly dealt with candidate genes, includ-ing established and presumed tumor suppressors that were supposed to be silenced by DNA hypermethyl-ation, such as RASSF1A, RARB, CDKN2A, PYCARD and the SFRP genes. Widespread hypomethylation was observed especially in LINE-1 retroelements. This research and the studies using next-generation tech-

10.2217/epi-2016-0064 Epigenomics (Epub ahead of print)

Figure 1. Concepts on DNA methylation in urothelial carcinomas. (A) Simplified depiction of conventional concepts on DNA methylation alterations in cancer: DNA hypermethylation accompanies silencing of, among others, tumor suppressor genes while DNA hypomethylation may promote hyperactivity of oncogenes. Hypomethylation of retroelements contributes primarily to genomic instability. These concepts are also thought to apply to urothelial carcinoma (UC). (B) Emerging concepts of DNA methylation alterations in UC: accumulating evidence suggests that chemical carcinogens may act not only as mutagens, but also disturb DNA methylation. DNA methylation changes in UC may occur as a consequence of mutated or deregulated chromatin regulators, or might be directly elicited by carcinogen action. Many consistently hypermethylated genes encode developmentally regulated transcription factors. In this fashion, DNA hypermethylation might contribute to the altered differentiation and the molecular subtypes in UC. Right hand: deeper understanding of the relation between specific carcinogens and DNA methylation alterations in UC might improve prevention. Hypermethylation provides excellent biomarkers for diagnostics. Epigenetic therapies might reverse not only silencing of tumor suppressors, but also aberrant differentiation in UC.

Hypermethylation of tumor suppressor

genes

Hypomethylation of oncogenes

Prevention

Diagnostics

Epigenetic therapies

Carcinogens

Genomic instabilityTumor cell proliferation,

survival and spread

Hypomethylation of retroelements

Mutation andderegulation of

chromatin regulators

Hypermethylation of transcription

factor genes

Altered differentiation Molecular subtypes

A

B

future science group


nologies until 2013 have been comprehensively sum-marized in several excellent journal reviews [22–24] and in a book chapter [25]. Many more recent studies have applied techniques interrogating thousands of CpG sites across the genome to evaluate DNA methylation changes in UC in a less biased fashion. Initially, espe-cially the Infinium 27K bead array was widely used, which assesses methylation at about 15,000 genes; more recently, the much larger 450K array has been

applied, which covers many more CpG sites (roughly 3% of all), especially outside CpG islands.

DNA methylation changes as biomarkers for UCSince 2012, several studies using novel techniques have aimed at identifying diagnostic or prognostic biomark-ers. These, together with recent validation studies are summarized in Table 1. Additionally, methylation



Table 1. Summary of recent methylation biomarker studies on urothelial carcinoma.

Study (year) Screening approach Validation design Relevant genes Purpose Ref.

Ibragimova et al. (2014)

HM27 array FFT: – 64 NMI-UC – 37 MI-UC – 6 NU

BS-Pyroseq: – Subset of UC

TET2, MLL3, ACTL6B, CIDEA, TRPA1, ITPKB

Profiles, T vs N, MI vs NMI, HG NMI vs LG NMI

[26]

Kandimalla et al. (2012)

Agilent 244K CpG-island array FFT: – 29 NMI-UC – 15 MI-UC

Custom Golden Gate Methylation assay FFPE tissues 24 + 41 NMI-UC

TBX2, TBX3, GATA2, ZIC4

Prognosis for pTa tumors

[27]

Yeh et al. (2015)

HM27 array FFT: – 3 LG-UC – 4 HG-UC – Normal urothelial cells

qMS-PCR FFT (64 NMI-UC, 37 MI-UC, 6 NU), Urine (54 NMI-UC, 15 MI-UC, 28 healthy; 21 NMI, 12 MI, 28 healthy)

ZNF671, IRF8, SFRP1

Detection in urine, Prognosis

[28]

Kitchen et al. (2016)

HM450 array FFT: – 21 NMI-UC (HG) – 4 NU

BS-Pyroseq FFT: – 18 LG-UC – 30 HG-UC

ATP5G2, IRX1, VAX2

Distinction, HG NMI vs N, HG NMI vs LG NMI

[29]

Chihara et al. (2013)

Re-evaluation of GoldenGate Methylation array data FFT (52 NMI-UC, 39 MI-UC, 34 TA-N, 12 NU)

BS-Pyroseq FFT (18 NMI-UC, 35 MI-UC, 25 TA-NU, 21 NU), urine (37 NMI-UC, 38 MI-UC, 28 healthy)

HOXA9, SOX1, SPP1, IFNG

Detection in urine

[30]

Sacristan et al. (2014)

Candidate TSGs MS-MPLA FFPE: – 251 NMI-UC

RARB, CD44, PAX5A, GSTP1, IGSF4, PYCARD, CDH13, TP53, GATA5, RB1

Prognosis [31]

Beukers et al. (2015)

Validation of former screen BS-SNap-Shot FFPE: – 192 NMI-UC

TBX2, TBX3, GATA2

Prognosis [32]

Kim et al. (2013)

HM27 array FFT: – 18 NMI-UC – 6 NU

BS-Pyroseq FFT: – 181 NMI-UC

HOXA9, ISL1, ALDH1A3

Prognosis [33]

Garcia-Baquero et al. (2014)

Candidate TSG MS-MPLA FFPE: – 38 NMI-UC – 23 MI-UC

SFRP5, PRDM2, BNIP3, CACN1G

Prognosis [34]

Su et al. (2014)

Validation of former screen BS-Pyroseq Urine: – 90 NMI-UC

SOX1, L1-MET, IRAK3

Detection Prediction of recurrence

[35]

Wang et al. (2016)

Candidates from screens in literature

qMS-PCR Urine: – 80 NMI-UC – 50 MI-UC – 129 other diseases – 53 healthy

POUF4F2, PCDH17

Detection [36]

BS-Pyroseq: Bisulfite pyrosequencing; FFPE: Formalin-fixed paraffin-embedded tissues; FFT: Fresh frozen tissues; HG: High grade; HM27: Illumina 27K bead array; HM450: Illumina 450K bead array; LG: Low grade; MI: Muscle-invasive; MPLA: Multiplex ligation-dependent probe amplification; MS: Methylation-specific; NMI: Nonmuscle invasive; NU: Normal bladder/urothelium; TA: Tumor-adjacent; UC: Urothelial carcinoma.



array techniques have been used to define methylation subtypes in UC.

Ibragimova et al. [26] obtained DNA methylation profiles by the 27K array from 101 pTa (papillary), pT1 (early invasive) and muscle-invasive UC at initial presentation. In keeping with previous studies, MI-UC had clearly different DNA methylation patterns from the NMI cases. Overall, 729 CpG islands were sig-nificantly hypermethylated and expression of many of the respective genes was downregulated. Interestingly, 11 tumors, mostly NMI-UC, displayed exceptionally widespread hypermethylation and in most of these the DNA demethylase gene TET2 was hypermethyl-ated. This latter finding hints at a novel mechanism underlying frequent hypermethylation (CpG-island methylator phenotype [CIMP] epigenotype) in some UC. Notably, CIMP tumors have also been detected by other studies in UC [6,27,40].

The publication by Aine et al. [41] belongs to a series of papers in which the Swedish investigators aim at the definition of molecular subtypes of UC by genomic approaches. Here, they used methylated DNA immunoprecipitation followed by hybridiza-tion to Nimblegen arrays (MeDIP-chip) on 98 UC across all stages and grades. On the basis of 5453 dif-ferentially methylated regions they ultimately derived three different patterns of altered DNA methylation. These were predicted to be related to chromatin states and transcription factor binding. Interestingly, three major methylation subgroups, too, were identified by the comprehensive genomic study of 131 UC (mostly muscle-invasive) by the TCGA consortium [6]. Like-wise consistent with other studies (see below) and the TCGA data, Aine et al. [41] observed characteristic

methylation changes in HOX gene clusters, especially in the HOXA cluster.

Following the investigation of 7 UC using the 27K methylation microarray, Yeh et al. [28] investi-gated one differentially methylated transcription fac-tor gene, ZNF671. Its methylation was validated by bisulfite pyrosequencing in a larger cohort of cancer tissues and cell lines and its relation to histopathologi-cal and clinical parameters was considered. The util-ity of ZNF671 methylation as a biomarker in urine was studied by quantitative methylation-specific PCR. Combined ZNF671, IRF8 and SFRP1 methyla-tion yielded a sensitivity of 96% and high specific-ity. These values are comparable to those achieved by other combinations of two or three methylation markers previously reported (summarized in [23]) and are notably better than those of presently approved urine assays for UC [22]. Finally, epigenetic repression of ZNF671 in UC cell lines was demonstrated and its reexpression inhibited cell growth and invasion. The authors ascribed this inhibition to the repression of stem cell genes like KIT, OCT4 and NANOG, which – in our hands – are, however, feebly expressed in UC cell lines.

Recently, Kitchen et al. [29] searched for DNA methylation events specific to high-grade NMI-UC, an important subgroup with risk of progression to MI disease. They used 450K arrays to define hypermeth-ylated CpG islands in 250 genes, from which they chose 25 candidates to compare high grade and low grade NMI-UC; differential methylation at several of these genes was validated in bisulfite pyrosequencing assays. For several genes, moreover, expression was inversely related to DNA methylation.

Study (year) Screening approach Validation design Relevant genes Purpose Ref.

Fantony et al. (2015)

Replication study qMS-PCR Urine 209 suspected UC, according to cystoscopy: – 52 positive – 12 unclear – 145 negative

TWIST1, NID2 Detection [37]

Monteiro-Reis et al. (2014)

Validation of former screen qMS-PCR FFT (57 UTUC, 36 NU) Urine (22 suspected UC, 20 healthy)

GDF15, TMEFF2, VIM

Detection Prognosis, UTUC

[38]

Xiong et al. (2015)

Candidates from literature MS-PCR FFPE: – 229 NMI-UC – 458 MI-UC

CDH1, HSPA2, RASSF1A, BRCA1, THBS1, GDF15

Staging, Prognosis, UTUC

[39]

BS-Pyroseq: Bisulfite pyrosequencing; FFPE: Formalin-fixed paraffin-embedded tissues; FFT: Fresh frozen tissues; HG: High grade; HM27: Illumina 27K bead array; HM450: Illumina 450K bead array; LG: Low grade; MI: Muscle-invasive; MPLA: Multiplex ligation-dependent probe amplification; MS: Methylation-specific; NMI: Nonmuscle invasive; NU: Normal bladder/urothelium; TA: Tumor-adjacent; UC: Urothelial carcinoma.

Table 1. Summary of recent methylation biomarker studies on urothelial carcinoma (cont.).



Chihara et al. [30] aimed at developing a methyla-tion biomarker panel for detection of UC using urine. They reanalyzed data from a previous study on 91 tumor, 34 tumor-adjacent normal and 12 normal blad-der tissues obtained by the Golden Gate Methylation Cancer Panel. Hyper- and hypomethylation changes at 11 genes were validated in 53 tumor, 25 tumor-adjacent and 21 normal bladder tissues using bisulfite pyrosequencing. Finally, these assays were applied to urine sediments from 73 UC patients and 18 healthy controls yielding up to 97% sensitivity and 100% specificity with the entire panel, but also good results with some subpanels. Hypermethylated loci included HOXA9 and SOX1, whereas IFNG and SPP1 loci were hypomethylated.

Stratification of NMI-UC was also the purpose of two studies investigating methylation of selected genes in FFPE tissues. Sacristan et al. [31] employed a mul-tiplex MS-ligation PCR for 25 typical cancer-hyper-methylated genes to a series of 251 samples, identify-ing suitable candidates, including PAX5A and RB1. Interestingly, hypermethylation was overall frequent in STK11, RARB and GATA5, three genes with plausible, but poorly investigated functions in urothelial carcino-genesis. Various panels of genes were related to clini-cal parameters, such as grading, staging and prognosis, but the results should be viewed with caution because of multiple testing issues. Beukers et al. [32] performed a validation study on 192 FFPE samples based on previ-ously defined candidate genes [27], which again, nota-bly, included transcription factor genes like GATA2, TBX2 and TBX4. Combined methylation values strati-fied the patient population in three groups with sig-nificantly different risks of progression to MI-UC. Yet another new methylation marker panel for NMI-UC including HOXA9, ISL1 and ALDH1A3 was proposed by Kim et al. [33] based on a 27K microarray analysis and subsequent verification by bisulfite pyrosequenc-ing in overall 181 tumor samples. In a prospective study, a larger set of methylated genes categorized as tumor suppressors was assessed for use as prognostic biomarkers yielding new candidates such as CACNA1, PRDM2, BNIP3 and a known candidate, SFRP1 [34].

Whereas FFPE or fresh-frozen samples from surgi-cally resected tumor tissues are suitable for prognostic methylation biomarker assays, samples for diagnostic biomarkers should be obtained by noninvasive or mini-mally invasive procedures from urine or blood. Most studies have defined methylation biomarker panels applicable to urine sediment [23], which contains tumor cells, but is free of factors in the urine supernatant that might damage DNA or interfere with PCR assays. However, a few studies on blood samples have been conducted and, taken together, these indicate that

analysis of circulating free DNA in plasma might pro-vide additional clinically useful information (reviewed in [42]).

Several recent studies report successful validation of previously developed urinary biomarker panels. For instance, Su et al. [35] used quantitative evaluation of DNA methylation at six genes to predict tumor recur-rence in patient with NMI-UC. A three-marker assay, including an interesting hypomethylation marker, L1-MET, was found vastly superior to cytology and even cystoscopy. A study on Chinese patients [36] developed a two-marker test for detection of UC yielding greater than 90% sensitivity and specificity. Importantly, this study included 130 UC patients (in development and validation cohorts combined) and a number of con-trols with other urological diseases that may confound UC diagnosis, such as prostate cancer and urolithiasis (see [23] for a discussion). Also notable is that one of the methylated genes assayed encodes yet another HOX-related transcription factor, POU4F2; the other gene is a member of the protocadherin cluster, which is aber-rantly methylated in many cancers [43]. One of the few multi-institutional studies to date attempted to vali-date a previously suggested two-gene set of methyla-tion biomarkers in patients with hematuria, a potential indicator of UC, or under surveillance for UC recur-rence [37]. This study is notable for failing to replicate the results of previous reports. Evidently, more such replication studies should be performed.

Detection of UC in the upper urinary tract (UTUC), in other words, ureters and renal pelves, which are less easily accessible to cystoscopic instruments, can be challenging. UTUC is a less frequent disease, with genetic alterations that largely overlap with those found in bladder UC, but some peculiarities, such as a higher frequency of microsatellite instability [44]. Regardless, a panel of urinary DNA methylation biomarkers suitable for detecting bladder UC has been reported to accu-rately identify UTUC, too, with high sensitivity and specificity [38] and gene methylation panels have been suggested as prognostic and staging biomarkers for this particular disease [39].

In summary, various panels of urinary methyla-tion markers have been suggested by multiple studies which could be employed to detect UC in populations at risk, but would be particularly valuable to monitor patients for recurrences and thereby reduce the number of cystoscopies. In addition, a few methylation marker panels have been reported that can be used on urine samples or FFPE-materials to assess the risk for disease recurrence and especially for progression of NMI-UC toward MI stages. Notably, DNA from urinary sedi-ment cells can concurrently be used for methylation and mutation analysis; thus, inclusion of characteristic



and frequent mutations like those affecting FGFR3 might yield additional specificity and sensitivity [45,46]. All reported methylation markers have been claimed to be superior to current methods for detection or prog-nostication of UC. However, while this claim is plausi-ble and DNA methylation assays in general have many advantages, these assays will not become embedded in clinical routine, unless validated in thorough multi-institutional studies with large numbers of patients and relevant controls, including patients with other urological afflictions. Perhaps the best way to achieve this aim would be a head-to-head comparison of the most reasonable proposed UC DNA methylation bio-marker panels by standardized protocols, which could be funded by an international agency.

Standardization is required for sample collection and DNA extraction, but especially for DNA methylation analysis. Techniques for DNA methylation assays may be confounded by DNA quality and impurities, but especially by biological heterogeneity, in other words, allelic methylation or partial and variable methylation patterns along a sequence [47]. DNA methylation assay techniques differ in their sensitivity to both types of confounders. For instance, the widely used elegant and convenient MS-PCR techniques, unlike pyrosequenc-ing, do not allow for a control of complete bisulfite conversion and are unreliable with partially methyl-ated sequences. MS-PCR techniques are suitable for estimating the overall frequency of hypermethylation at a locus across a sample set, but they do not yield a reliable result for each individual sample [48]. Stan-dardization across several laboratories appears to be better achievable with techniques like amplicon bisul-fite sequencing or bisulfite pyrosequencing providing quantitative information on several individual CpG sites and containing internal controls for DNA qual-ity and bisulfite conversion [49]. Ultimately, affordable deep and third-generation sequencing could replace these techniques in the next decade.

DNA methylation & urothelial carcinogenesisPresumably because of the evident advantages pro-vided by analyzing urine, relatively few attempts have been made to detect UC-specific methylation changes in free circulating DNA from plasma and their bio-marker potential remains uncertain (discussed in [42]). Instead, a number of large-scale investigations have addressed the question whether DNA hypomethyl-ation in blood cells might indicate an increased risk for the development or the presence of UC. Most of these studies have specifically investigated methylation of LINE-1 retroelements, which might be considered as a surrogate marker for overall genomic methylation [50]. Of note, LINE-1 hypomethylation is a well-established

characteristic of UC [51]; in a recent comparison of sev-eral common cancers, LINE-1 methylation overall and at specific individual elements was most frequently and most strongly diminished in UC tissues [52]. Obviously, methylation measured in blood cells reflects essentially the methylation of leukocyte genomes. Any significant molecular changes will therefore result from shifts in the leukocyte population or in gross disturbances in the methylation of hematopoietic cells (see [53] for a relevant discussion). Thus, changes in overall DNA methylation in blood cells suggest an underlying dis-ease process that may necessitate surveillance and preventive intervention, for example with respect to smoking behavior, nutrition or exposure to environ-mental factors. For instance, LINE-1 hypomethylation in granulocytes was found to be related to exposure toward the disinfectant-by-product trihalomethane and bladder cancer risk in the Spanish Bladder Cancer study/EPICURO [54]. In general, however, the value of LINE-1 hypomethylation in blood for assessment of UC and other cancers is not established [52,55].

Like LINE-1 hypomethylation in blood cells, spe-cific methylation changes in UC tumor tissues may be indicative of specific carcinogens, and in some cases can provide insights into mechanisms of carcinogen-esis. For instance, several studies have attempted to relate DNA methylation changes in UC tissues to smoking, a major risk factor [24,50,56–58]. More directly, the ability of cigarette smoke components to induce DNA methylation changes was demonstrated by genome-wide analyses in a cell culture model of UC progression [56]. Similarly, LINE-1 hypomethylation was linked to oxidative stress in bladder tissues [59] and to SAM depletion by oxidative stress in urothelial can-cer cells [60], although the extent of the methylation changes was relatively small.

Alterations of DNA methylation may be particularly relevant in the carcinogenic action of arsenic, which induces some DNA damage, but more pertinently disturbs epigenetic regulation, including DNA meth-ylation [61,62]. Exposure to arsenic from environmental sources, particularly contaminated drinking water, is a major risk factor for several cancers, especially UC, in several regions across the world [61]. Arsenic metabo-lism requires methyl groups from SAM. Depletion of SAM and accumulation of S-adenosyl-homocysteine therefore ought to lead primarily to DNA hypometh-ylation. Indeed, hypomethylation of LINE-1 retroele-ments was reported in leukocytes of exposed persons, but was modulated by nutrition, such as supply of folate, another methyl group carrier (reviewed in [61]). Surprisingly, two studies on DNA methylation changes in UC associated with arsenic-exposure revealed sub-stantial promoter hypermethylation. Yang et al. [63]



compared 14 arsenic-associated and nonassociated UC each using the HM27 array. They observed 208 CpG sites with considerably stronger methylation, but only 23 with lower methylation in arsenic-associated UC. After adjustment for stage and age, methylation differences in five genes (CTNNA2, KLK7, NPY2R, ZNF132 and KCNK17) remained significant. Simi-larly, an investigation of UC tissues and exfoliated urinary cells from Mexican patients with high arsenic exposure by a methylation-dependent immunoprecipi-tation technique revealed 49 significantly hypermeth-ylated genes [64], of which almost half were also fre-quently hypermethylated in the TCGA UC cohort [6]. This parallel could indicate that arsenic is a more common factor in urothelial carcinogenesis than hith-erto assumed or that different carcinogenic pathways converge on common target genes in UC.

Of note, a different histological subtype of blad-der cancer, SCC, is caused by chronic inflammation in the bladder, especially by chronic infections with Schistosoma mansoni in certain countries like Egypt. Comparing DNA methylation patterns and genetic alterations in this disease with those in UC might yield important insights into the mechanisms of carcino-genesis in both types of disease, as common changes might point at biological factors relevant for tumor growth, whereas differential changes might highlight the respective specific mechanisms of carcinogenesis. Unfortunately, only a few studies on DNA methylation in bladder SCC have been conducted to date, usually investigating candidate genes [65].

In summary, thus, DNA methylation changes in bladder cancer may often be related to carcinogen exposure, possibly in a carcinogen-specific manner. Elucidation of these relations could therefore help to clarify the etiology and pathogenesis of UC and to better understand the mechanisms causing genetic and epigenetic changes. Ideally, DNA methylation signatures might spot the involvement of specific car-cinogens, like the mutational signature of aristolochic acids has revealed the unexpectedly large impact of these herbal toxins on carcinogenesis in the urinary tract [66]. Ultimately, such insights might lead to better cancer prevention and identify populations at risk.

Biological relevance of DNA methylation changes in UCRelatively few papers in the last years have addressed the biological relevance of DNA methylation in UC, owing perhaps to the attraction of promising trans-lational applications of DNA methylation analyses in UC diagnostics, therapy and prevention, but likely also to the insufficient funding for basic research on bladder cancer. Notably, the TCGA consortium has

conducted DNA methylation analysis by the 450K array on a large number of tumors (as of May 2016, 412 tumors plus 21 controls), but the results received short mention in two major papers on the genom-ics of UC [5,6]. Thus, while some attempts have been made [40], more comprehensive and detailed analyses of the relation between genomic changes and DNA meth-ylation changes based on next-generation data need to be published. Likewise, further systematic investiga-tions of the relation of DNA methylation alterations to gene-expression changes based on genome-wide DNA methylation analyses and RNA sequencing would be valuable to better delineate the contribution of DNA methylation to UC pathogenesis.

The mechanisms underlying the prominent changes in DNA methylation patterns in UC are likewise poorly understood. As discussed above, some meth-ylation changes in UC might be induced by chemical carcinogens like arsenic, which is thought to induce hypomethylation by depletion of SAM. Indeed, global DNA hypomethylation is highly prevalent in UC, but to an extent that can hardly be ascribed to arsenic carcinogenesis only. Thus, the cause of global hypo-methylation in UC remains largely as unexplained as a decade ago [67]. Moreover, unexpectedly, upon closer analysis, arsenic appears to induce also gene-specific hypermethylation [63,64].

Several studies [6,26–27,40] have identified a subgroup of UC with conspicuously increased CpG-island hypermethylation, resembling the CIMP-like epigeno-type observed in other cancer types. In those, a CIMP epigenotype is caused by various mechanisms [68,69], none of which appears to straightforwardly apply to UC. In colorectal cancers, CIMP is quite strictly associated with BRAFV600E mutations, which do not occur in UC. In leukemias and glioblastoma, CIMP is a consequence of mutations in IDH1/2 or TET1/2 genes that inhibit DNA demethylation via the hydroxymethylcytosine pathway [69]. Indeed, the genome of UCs, like many other cancers, is depleted of hydroxymethylcytosine [70]. However, apart from the report on an association of a CIMP epigenotype with TET2 hypermethylation [26], there is little evidence for changes in these genes in UC. In fact, according to the TCGA data (accessed via cBioPortal), TET1–3 are mutated in a few cases each, but otherwise rather over-expressed. Mutations of IDH1/2 are likewise rare. Of note, overexpression of each of the DNMTs occurs in some cases, but not generally.

A mechanism prevalent in many cancers that results in preferential DNA hypermethylation of Polycomb target genes evidently contributes to hypermethylation in UC as well [71]. EZH2, the histone methyltransfer-ase subunit of the Polycomb complex PRC2 is over-



expressed in many UC (reviewed in [72]). However, it is unknown to which extent EZH2 overexpression promotes DNA hypermethylation in UC as well as whether overexpressed EZH2 might exert additional PRC-independent functions, as in prostate cancer [73].

Notably, many of the chromatin regulators inac-tivated by mutations in UC are ‘trithorax-like’ fac-tors antagonizing Polycomb activity during develop-ment and differentiation, especially the MLL H3K4 methyltransferases and the H3K27me3 demethylase UTX/KDM6A. Conceivably, alternatively to Poly-comb overactivity, loss of these antagonists might lead to gene silencing during urothelial carcinogenesis and precipitate DNA hypermethylation. Unfortunately, no direct experiments addressing this hypothesis have so far been published and the target genes of the mutated chromatin regulators in urothelial cells are presently unknown.

Undisputedly, however, as emphasized above, many genes frequently hypermethylated in UC encode developmental transcription factors regulated by Poly-comb complexes, especially homeobox genes [27,32,40]. Several studies have reported coordinated expression and epigenetic changes in HOX gene clusters [41,74], as well as in other developmentally regulated gene clus-ters like the imprinted gene cluster on chromosome 14q32 [75]. Detailed analyses of HOX gene expression in UC accordingly reveal severely disturbed expression patterns [76,77].

As in normal development [17], many changes in DNA methylation observed in UC may be more consequence than cause of gene expression changes. However, DNA hypermethylation at key develop-mental regulators, together with expression changes of transcription factors like FOXA1, p63 and PPARγ, may be crucial for urothelial carcinogenesis by dis-turbing cellular identity and urothelial differen-tiation [9]. There is substantial evidence from other cancer types that a subset of DNA methylation alterations defines tumor subtype-specific lineages by redirecting or blocking differentiation [78–80]. This hypothesis needs experimental investigation in the case of UC. Notably, the relevant DNA methyla-tion changes will likely differ between UC molecu-lar subtypes, especially if these should indeed arise from different precursor cells. Indeed, UC stem cells appear to be remarkably diverse [81]. Methylation subtypes have been defined in several studies on UC, but it is not yet clear how they relate to the molecu-lar subtypes defined by gene expression and muta-tional patterns. Aine et al. [41] found a certain degree of correspondence, but further studies will certainly follow, as the definition of UC molecular subtypes becomes refined.

DNA methylation-directed therapy in UCDNA methylation inhibitors, especially the cytidine analogues decitabine (5-deoxy-azacytidine) and aza-cytidine, are used in the treatment of myelodysplas-tic syndrome and certain leukemias and have recently been demonstrated to provide clinical benefit in indi-vidual patients with solid cancers [82,83]. These cyti-dine analogues are routinely used in experimental research on UC, but no clinical trials are underway. DNA methylation inhibitors are postulated to impede tumor growth by inducing the reexpression of tumor suppressors or pro-apoptotic genes silenced by hyper-methylation. However, although some genes with antineoplastic functions have been demonstrated to become reexpressed in UC cell lines following DNA methylation inhibitor treatment (see [28,84] for recent examples), no major tumor suppressor is known to be predominantly inactivated by hypermethylation in UC. CDKN2A, for instance, is typically inactivated by deletion or point mutations rather than hypermethyl-ation. Moreover, according to recent findings in other cancers, DNA methylation inhibitors may act more by inducing re-expression of retroelements that subse-quently elicit an interferon response leading to apopto-sis [85,86]. Whether this mechanism applies in UC with their already extensive hypomethylation of retroele-ments [51,52] remains to be determined. Clearly, a bet-ter understanding of the functional consequences of altered DNA methylation is a prerequisite for therapy by DNA methylation inhibitors in UC. Notably, DNA methylation inhibitors or other ‘epigenetic inhibi-tors’ might be employed in combination therapies to prevent or reverse resistance to cytotoxic or targeted chemotherapeutic drugs.

Conclusion & future perspectiveThrough many recent studies using next-generation techniques, a large amount of data has been assembled providing a much better description of DNA methyla-tion alterations in UC. In the next years, now, several major important questions can be addressed.

How can the promising candidate DNA methylation biomarker panels for UC detection described by several groups be optimally employed in clinical practice? The first step should be validation of selected promising panels in multicenter studies using standardized pro-tocols on large patient populations, including patients with urological diseases that in routine practice may confound diagnostics.

How reliable are the proposed DNA methylation biomarkers for NMI-UC stratification and prognosis across diverse populations? Validation of the proposed marker panels by independent groups should be the first priority.



How do DNA methylation alterations in UC relate to carcinogen exposures and could they be used to delineate etiological factors? Hints that certain altera-tions of DNA methylation in UC tissues or in blood may indicate exposure to specific carcinogens should be followed up with the aim of developing yield spe-cific carcinogen methylation signatures that might ultimately help to spot exposures and prevent UC.

Which molecular mechanisms cause DNA methyla-tion alterations in UC? While some DNA methylation alterations in UC have parallels in other cancers, the underlying mechanisms appear to differ and require fur-ther experimental clarification. Likewise, the causes for the high prevalence of retroelement hypomethylation in UC require further experimental clarification.

How do DNA methylation alterations relate to the frequent mutations in chromatin regulators in UC? It is tempting to speculate that alterations in these epigen-etic regulators may contribute to and may interact with DNA methylation alterations in UC. This hypothesis needs to be addressed by dedicated experiments.

How do DNA methylation alterations relate to molecular subtypes of UC? The essential data to answer this question are already available and it should

become answered following further refinement of the definitions of UC molecular subtypes.

To what extent do DNA methylation alterations contribute to neoplastic properties of urothelial tumors? Much more work is required to answer this question, especially on the relation between disturbed DNA methylation and urothelial differentiation in UC. This issue is crucial to understanding the patho-genesis of UC and for gauging therapies targeting DNA methylation.

AcknowledgementsThe authors are grateful for helpful comments on the

manuscript by MJ Hoffmann and G Niegisch.

Financial & competing interests disclosureThe authors have no relevant affiliations or financial involve-

ment with any organization or entity with a financial inter-

est in or financial conflict with the subject matter or mate-

rials discussed in the manuscript. This includes employment,

consultancies, honoraria, stock ownership or options, expert

testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this

manuscript.

Executive summary

• Urothelial carcinoma (UC) is the most frequent histological subtype of bladder cancer, a highly prevalent malignancy worldwide. Improvements in diagnostics and treatment of this disease are urgently needed.

• A focus of research on DNA methylation alterations in UC has been on the development of diagnostic and prognostic biomarkers. Diagnostic biomarkers are especially required to monitor for recurrence and in risk populations. Prognostic biomarkers are especially required for nonmuscle-invasive stages of the disease to determine the risk of recurrences and progression.

• Recent investigations have employed array-based analysis of DNA methylation across the genome. Several promising biomarker candidates have been proposed, consisting typically of small sets of hypermethylated genes. These proposed biomarker sets require independent validation by standardized procedures and robust DNA methylation assays in multi-institutional studies as a prerequisite to their clinical implementation.

• Although the mechanisms underlying DNA methylation alterations in UC remain unclear, several recent studies hint at a relation between hypermethylation or hypomethylation events in tumor tissues or blood cells and exposure to specific carcinogens.

• Recent studies have identified UC subtypes according to their patterns of DNA methylation alterations. One current research topic is how these methylation subtypes relate to molecular subtypes recently defined by mutational and expression analyses.

• Mutations in chromatin regulators are prevalent in UC. Their functional impact and their relationship to DNA methylation alterations is unknown and should constitute a central topic in UC research.

• The functional effects of DNA methylation alterations in UC are likewise largely unexplored. However, recent data suggest that DNA methylation alterations affect especially developmental transcription factors and may thereby impinge on altered differentiation in these cancers.

ReferencesPapers of special note have been highlighted as: • of interest; •• of considerable interest

1 Ikeda S, Hansel DE, Kurzrock R. Beyond conventional chemotherapy: emerging molecular targeted and immunotherapy strategies in urothelial carcinoma. Cancer Treat. Rev. 41(8), 699–706 (2015).

2 Netto GJ. Role for anti-PD-L1 immune checkpoint inhibitor in advanced urothelial carcinoma. Lancet 387(10031), 1881–1882 (2016).

3 Carter AJ, Nguyen CN. A comparison of cancer burden and research spending reveals discrepancies in the distribution of research funding. BMC Public Health 12, 526 (2012).



4 Guo G, Sun X, Chen C et al. Whole-genome and whole-exome sequencing of bladder cancer identifies frequent alterations in genes involved in sister chromatid cohesion and segregation. Nat. Genet. 45(12), 1459–1463 (2013).

5 Kim J, Akbani R, Creighton CJ et al. Invasive bladder cancer: genomic insights and therapeutic promise. Clin. Cancer Res. 21(20), 4514–4524 (2015).

6 Cancer Genome Atlas Research Network. Comprehensive molecular characterization of urothelial bladder carcinoma. Nature 507(7492), 315–322 (2014).

7 Knowles MA, Hurst CD. Molecular biology of bladder cancer: new insights into pathogenesis and clinical diversity. Nat. Rev. Cancer 15(1), 25–41 (2015).

8 Choi W, Porten S, Kim S et al. Identification of distinct basal and luminal subtypes of muscle-invasive bladder cancer with different sensitivities to frontline chemotherapy. Cancer Cell 25(2), 152–165 (2014).

9 Adam RM, Degraff DJ. Molecular mechanisms of squamous differentiation in urothelial cell carcinoma: a paradigm for molecular subtyping of urothelial cell carcinoma of the bladder. Urol. Oncol. 33(10), 444–450 (2015).

10 Van Batavia J, Yamany T, Molotkov A et al. Bladder cancers arise from distinct urothelial sub-populations. Nat. Cell Biol. 16(10), 982–991 (2014).

11 Lerner SP, Mcconkey DJ, Hoadley KA et al. Bladder cancer molecular taxonomy: summary from a consensus meeting. Bladder Cancer 2(1), 37–47 (2016).

12 Van Der Meulen J, Speleman F, Van Vlierberghe P. The H3K27me3 demethylase UTX in normal development and disease. Epigenetics 9(5), 658–668 (2014).

13 Laugesen A, Helin K. Chromatin repressive complexes in stem cells, development, and cancer. Cell Stem Cell 14(6), 735–751 (2014).

14 Hamidi T, Singh AK, Chen T. Genetic alterations of DNA methylation machinery in human diseases. Epigenomics 7(2), 247–265 (2015).

15 Kohli RM, Zhang Y. TET enzymes, TDG and the dynamics of DNA demethylation. Nature 502(7472), 472–479 (2013).

16 Jones PA. Functions of DNA methylation: islands, start sites, gene bodies and beyond. Nat. Rev. Genet. 13(7), 484–492 (2012).

17 Smith ZD, Meissner A. DNA methylation: roles in mammalian development. Nat. Rev. Genet. 14(3), 204–220 (2013).

18 Baylin SB, Jones PA. Epigenetic determinants of cancer. Cold Spring Harb. Perspect. Biol. doi:10.1101/cshperspect.a019505 (2016) (Epub ahead of print).

19 Hansen KD, Timp W, Bravo HC et al. Increased methylation variation in epigenetic domains across cancer types. Nat. Genet. 43(8), 768–775 (2011).

20 Bell RE, Golan T, Sheinboim D et al. Enhancer methylation dynamics contribute to cancer plasticity and patient mortality. Genome Res. 26(5), 601–611 (2016).

21 Plass C, Pfister SM, Lindroth AM, Bogatyrova O, Claus R, Lichter P. Mutations in regulators of the epigenome and their connections to global chromatin patterns in cancer. Nat. Rev. Genet. 14(11), 765–780 (2013).

22 Reinert T. Methylation markers for urine-based detection of bladder cancer: the next generation of urinary markers for diagnosis and surveillance of bladder cancer. Adv. Urol. 2012, 503271 (2012).

23 Kandimalla R, Van Tilborg AA, Zwarthoff EC. DNA methylation-based biomarkers in bladder cancer. Nat. Rev. Urol. 10(6), 327–335 (2013).

•• AcomprehensiveanddetailedsummaryofDNAmethylation-basedbiomarkerstudiesinurothelialcarcinoma(UC)until2012.

24 Besaratinia A, Cockburn M, Tommasi S. Alterations of DNA methylome in human bladder cancer. Epigenetics 8(10), 1013–1022 (2013).

•• AnexcellentoverviewofbiologicalaspectsofDNAmethylationalterationsinUCuntil2012.

25 Schulz WA, Koutsogiannouli EA, Niegisch G, Hoffmann MJ. Epigenetics of urothelial carcinoma. Methods Mol. Biol. 1238, 183–215 (2015).

26 Ibragimova I, Dulaimi E, Slifker MJ, Chen DY, Uzzo RG, Cairns P. A global profile of gene promoter methylation in treatment-naive urothelial cancer. Epigenetics 9(5), 760–773 (2014).

27 Kandimalla R, Van Tilborg AA, Kompier LC et al. Genome-wide analysis of CpG island methylation in bladder cancer identified TBX2, TBX3, GATA2, and ZIC4 as pTa-specific prognostic markers. Eur. Urol. 61(6), 1245–1256 (2012).

28 Yeh CM, Chen PC, Hsieh HY et al. Methylomics analysis identifies ZNF671 as an epigenetically repressed novel tumor suppressor and a potential non-invasive biomarker for the detection of urothelial carcinoma. Oncotarget 6(30), 29555–29572 (2015).

29 Kitchen MO, Bryan RT, Emes RD et al. Quantitative genome-wide methylation analysis of high-grade non-muscle invasive bladder cancer. Epigenetics 11(3), 237–246 (2016).

30 Chihara Y, Kanai Y, Fujimoto H et al. Diagnostic markers of urothelial cancer based on DNA methylation analysis. BMC Cancer 13, 275 (2013).

31 Sacristan R, Gonzalez C, Fernandez-Gomez JM, Fresno F, Escaf S, Sanchez-Carbayo M. Molecular classification of non-muscle-invasive bladder cancer (pTa low-grade, pT1 low-grade, and pT1 high-grade subgroups) using methylation of tumor-suppressor genes. J. Mol. Diagn. 16(5), 564–572 (2014).

32 Beukers W, Kandimalla R, Masius RG et al. Stratification based on methylation of TBX2 and TBX3 into three molecular grades predicts progression in patients with pTa-bladder cancer. Mod. Pathol. 28(4), 515–522 (2015).

33 Kim YJ, Yoon HY, Kim JS et al. HOXA9, ISL1 and ALDH1A3 methylation patterns as prognostic markers for nonmuscle invasive bladder cancer: array-based DNA methylation and expression profiling. Int. J. Cancer 133(5), 1135–1142 (2013).

34 Garcia-Baquero R, Puerta P, Beltran M et al. Methylation of tumor suppressor genes in a novel panel predicts clinical outcome in paraffin-embedded bladder tumors. Tumour Biol. 35(6), 5777–5786 (2014).



35 Su SF, De Castro Abreu AL, Chihara Y et al. A panel of three markers hyper- and hypomethylated in urine sediments accurately predicts bladder cancer recurrence. Clin. Cancer Res. 20(7), 1978–1989 (2014).

36 Wang Y, Yu Y, Ye R et al. An epigenetic biomarker combination of PCDH17 and POU4F2 detects bladder cancer accurately by methylation analyses of urine sediment DNA in Han Chinese. Oncotarget 7(3), 2754–2764 (2016).

37 Fantony JJ, Abern MR, Gopalakrishna A et al. Multi-institutional external validation of urinary TWIST1 and NID2 methylation as a diagnostic test for bladder cancer. Urol. Oncol. 33(9), 387.e1–e6 (2015).

• Araremulti-institutionalvalidationstudyofmethylationbiomarkersinUC.

38 Monteiro-Reis S, Leca L, Almeida M et al. Accurate detection of upper tract urothelial carcinoma in tissue and urine by means of quantitative GDF15, TMEFF2 and VIM promoter methylation. Eur. J. Cancer 50(1), 226–233 (2014).

39 Xiong G, Liu J, Tang Q et al. Prognostic and predictive value of epigenetic biomarkers and clinical factors in upper tract urothelial carcinoma. Epigenomics 7(5), 733–744 (2015).

40 Lauss M, Aine M, Sjodahl G et al. DNA methylation analyses of urothelial carcinoma reveal distinct epigenetic subtypes and an association between gene copy number and methylation status. Epigenetics 7(8), 858–867 (2012).

41 Aine M, Sjodahl G, Eriksson P et al. Integrative epigenomic analysis of differential DNA methylation in urothelial carcinoma. Genome Med. 7(1), 23 (2015).

• AninterestingattemptatrelatingDNAmethylationchangestoothergenomicalterationsandgeneexpressioninUC,notablyidentifyingcoordinatedchangesinHOXclusters.

42 Ellinger J, Muller SC, Dietrich D. Epigenetic biomarkers in the blood of patients with urological malignancies. Expert. Rev. Mol. Diagn. 15(4), 505–516 (2015).

43 Banelli B, Romani M. Quantitative methylation analysis of the PCDHB gene cluster. Methods Mol. Biol. 1315, 189–200 (2015).

44 Skafianos JP, Cha EK, Iyer G et al. Genomic characterization of upper tract urothelial carcinoma. Eur. Urol. 68(6), 970–977 (2015).

45 Serizawa RR, Ralfkiaer U, Steven K et al. Integrated genetic and epigenetic analysis of bladder cancer reveals an additive diagnostic value of FGFR3 mutations and hypermethylation events. Int. J. Cancer 129(1), 78–87 (2011).

46 Kandimalla R, Masius R, Beukers W et al. A 3-plex methylation assay combined with the FGFR3 mutation assay sensitively detects recurrent bladder cancer in voided urine. Clin. Cancer Res. 19(17), 4760–4769 (2013).

47 Mikeska T, Candiloro IL, Dobrovic A. The implications of heterogeneous DNA methylation for the accurate quantification of methylation. Epigenomics 2(4), 561–573 (2010).

48 Alnaes GI, Ronneberg JA, Kristensen VN, Tost J. Heterogeneous DNA methylation patterns in the GSTP1 promoter lead to discordant results between assay technologies and impede its implementation as epigenetic

biomarkers in breast cancer. Genes (Basel) 6(3), 878–900 (2015).

• Whiledealingwithbreastcancer,thisthoroughcomparisonofDNAmethylationassaytechniquesishighlyrelevanttothedevelopmentofUCdiagnostics.

49 Blueprint Consortium. Quantitative comparison of DNA methylation assays for biomarker development and clinical applications. Nat. Biotechnol. 34(7), 726–737 (2016).

50 Tajuddin SM, Amaral AF, Fernandez AF et al. Genetic and non-genetic predictors of LINE-1 methylation in leukocyte DNA. Environ. Health Perspect. 121(6), 650–656 (2013).

51 Kreimer U, Schulz WA, Koch A, Niegisch G, Goering W. HERV-K and LINE-1 DNA methylation and reexpression in urothelial carcinoma. Front. Oncol. 3, 255 (2013).

52 Nuesgen N, Goering W, Dauksa A et al. Inter-locus as well as intra-locus heterogeneity in LINE-1 promoter methylation in common human cancers suggests selective demethylation pressure at specific CpGs. Clin. Epigenetics 7(1), 17 (2015).

• Adetailedcomparativere-evaluationofretroelementhypomethylationinUCandothercancers.

53 Leidinger P, Backes C, Dahmke IN et al. What makes a blood cell based miRNA expression pattern disease specific? – a miRNome analysis of blood cell subsets in lung cancer patients and healthy controls. Oncotarget 5(19), 9484–9497 (2014).

54 Salas LA, Villanueva CM, Tajuddin SM et al. LINE-1 methylation in granulocyte DNA and trihalomethane exposure is associated with bladder cancer risk. Epigenetics 9(11), 1532–1539 (2014).

55 Barchitta M, Quattrocchi A, Maugeri A, Vinciguerra M, Agodi A. LINE-1 hypomethylation in blood and tissue samples as an epigenetic marker for cancer risk: a systematic review and meta-analysis. PLoS ONE 9(10), e109478 (2014).

56 Brait M, Munari E, Lebron C et al. Genome-wide methylation profiling and the PI3K-AKT pathway analysis associated with smoking in urothelial cell carcinoma. Cell Cycle 12(7), 1058–1070 (2013).

57 Marsit CJ, Koestler DC, Christensen BC, Karagas MR, Houseman EA, Kelsey KT. DNA methylation array analysis identifies profiles of blood-derived DNA methylation associated with bladder cancer. J. Clin. Oncol. 29(9), 1133–1139 (2011).

58 Hoyos-Giraldo LS, Escobar-Hoyos LF, Saavedra-Trujillo D et al. Gene-specific promoter methylation is associated with micronuclei frequency in urothelial cells from individuals exposed to organic solvents and paints. J. Expo. Sci. Environ. Epidemiol. 26(3), 257–262 (2016).

59 Patchsung M, Boonla C, Amnattrakul P, Dissayabutra T, Mutirangura A, Tosukhowong P. Long interspersed nuclear element-1 hypomethylation and oxidative stress: correlation and bladder cancer diagnostic potential. PLoS ONE 7(5), e37009 (2012).

60 Kloypan C, Srisa-Art M, Mutirangura A, Boonla C. LINE-1 hypomethylation induced by reactive oxygen species is mediated via depletion of S-adenosylmethionine. Cell Biochem. Funct. 33(6), 375–385 (2015).



61 Bustaffa E, Stoccoro A, Bianchi F, Migliore L. Genotoxic and epigenetic mechanisms in arsenic carcinogenicity. Arch. Toxicol. 88(5), 1043–1067 (2014).

62 Riedmann C, Ma Y, Melikishvili M et al. Inorganic arsenic-induced cellular transformation is coupled with genome wide changes in chromatin structure, transcriptome and splicing patterns. BMC Genomics 16, 212 (2015).

63 Yang TY, Hsu LI, Chiu AW et al. Comparison of genome-wide DNA methylation in urothelial carcinomas of patients with and without arsenic exposure. Environ. Res. 128, 57–63 (2014).

64 Rager JE, Tilley SK, Tulenko SE et al. Identification of novel gene targets and putative regulators of arsenic-associated DNA methylation in human urothelial cells and bladder cancer. Chem. Res. Toxicol. 28(6), 1144–1155 (2015).

• ProvidingevidenceforarsenicexposureasacauseofDNAmethylationchangesinUCinawell-characterizedpopulationandsuggestingitswidespreadimpact.

65 Eissa S, Swellam M, El-Khouly IM et al. Aberrant methylation of RARbeta2 and APC genes in voided urine as molecular markers for early detection of bilharzial and nonbilharzial bladder cancer. Cancer Epidemiol. Biomarkers Prev. 20(8), 1657–1664 (2011).

66 Poon SL, Huang MN, Choo Y et al. Mutation signatures implicate aristolochic acid in bladder cancer development. Genome Med. 7(1), 38 (2015).

67 Hoffmann MJ, Schulz WA. Causes and consequences of DNA hypomethylation in human cancer. Biochem. Cell Biol. 83(3), 296–321 (2005).

68 Weisenberger DJ. Characterizing DNA methylation alterations from the cancer genome atlas. J. Clin. Invest. 124(1), 17–23 (2014).

69 Hughes LA, Melotte V, De Schrijver J et al. The CpG island methylator phenotype: what’s in a name? Cancer Res. 73(19), 5858–5868 (2013).

70 Munari E, Chaux A, Vaghasia AM et al. Global 5-hydroxymethylcytosine levels are profoundly reduced in multiple genitourinary malignancies. PLoS ONE 11(1), e0146302 (2016).

71 Beukers W, Hercegovac A, Vermeij M et al. Hypermethylation of the Polycomb group target gene PCDH7 in bladder tumors from patients of all ages. J. Urol. 190(1), 311–316 (2013).

72 Martinez-Fernandez M, Rubio C, Segovia C, Lopez-Calderon FF, Duenas M, Paramio JM. EZH2 in bladder bancer, a promising therapeutic target. Int. J. Mol. Sci. 16(11), 27107–27132 (2015).

73 Xu K, Wu ZJ, Groner AC et al. EZH2 oncogenic activity in castration-resistant prostate cancer cells is Polycomb-independent. Science 338(6113), 1465–1469 (2012).

74 Vallot C, Stransky N, Bernard-Pierrot I et al. A novel epigenetic phenotype associated with the most aggressive pathway of bladder tumor progression. J. Natl Cancer Inst. 103(1), 47–60 (2011).

• EvidenceforcoordinatedregionalepigeneticchangesinUC,especiallyatHOXgeneclusters.

75 Greife A, Knievel J, Ribarska T, Niegisch G, Schulz WA. Concomitant downregulation of the imprinted genes DLK1 and MEG3 at 14q32.2 by epigenetic mechanisms in urothelial carcinoma. Clin. Epigenetics 6(1), 29 (2014).

76 Eriksson P, Aine M, Veerla S, Liedberg F, Sjodahl G, Hoglund M. Molecular subtypes of urothelial carcinoma are defined by specific gene regulatory systems. BMC Med. Genomics 8, 25 (2015).

77 Heubach J, Monsior J, Deenen R et al. The long noncoding RNA HOTAIR has tissue and cell type-dependent effects on HOX gene expression and phenotype of urothelial cancer cells. Mol. Cancer 14, 108 (2015).

78 Sproul D, Kitchen RR, Nestor CE et al. Tissue of origin determines cancer-associated CpG island promoter hypermethylation patterns. Genome Biol. 13(10), R84 (2012).

79 Oakes CC, Seifert M, Assenov Y et al. DNA methylation dynamics during B cell maturation underlie a continuum of disease phenotypes in chronic lymphocytic leukemia. Nat. Genet. 48(3), 253–264 (2016).

80 Guillamot M, Cimmino L, Aifantis I. The impact of DNA methylation in hematopoietic malignancies. Trends Cancer 2(2), 70–83 (2016).

81 Ho PL, Kurtova A, Chan KS. Normal and neoplastic urothelial stem cells: getting to the root of the problem. Nat. Rev. Urol. 9(10), 583–594 (2012).

82 Karahoca M, Momparler RL. Pharmacokinetic and pharmacodynamic analysis of 5-aza-2’-deoxycytidine (decitabine) in the design of its dose-schedule for cancer therapy. Clin. Epigenetics 5(1), 3 (2013).

83 Nie J, Liu L, Li X, Han W. Decitabine, a new star in epigenetic therapy: the clinical application and biological mechanism in solid tumors. Cancer Lett. 354(1), 12–20 (2014).

84 Rose M, Schubert C, Dierichs L et al. OASIS/CREB3L1 is epigenetically silenced in human bladder cancer facilitating tumor cell spreading and migration in vitro. Epigenetics 9(12), 1626–1640 (2014).

85 Roulois D, Loo Yau H, Singhania R et al. DNA-Demethylating agents target colorectal cancer cells by inducing viral mimicry by endogenous transcripts. Cell 162(5), 961–973 (2015).

86 Chiappinelli KB, Strissel PL, Desrichard A et al. Inhibiting DNA methylation causes an Interferon response in cancer via dsRNA including endogenous retroviruses. Cell 164(5), 1073 (2016).

1275Epigenomics (2016) 8(9), 1275–1287 ISSN 1750-1911

part of

Long noncoding and circular RNAs in lung cancer: advances and perspectives

Weijia Xie1, Shuai Yuan1, Zhifu Sun‡,2 & Yafei Li*,‡,1

1Department of Epidemiology, College

of Preventive Medicine, Third Military

Medical University, Chongqing, People’s

Republic of China 2Department of Health Sciences

Research, Mayo Clinic, Rochester,

MN, USA


Tel.: +86 236 875 2293

[email protected]‡Authors contributed equally

Review


Epigenomics

Review 2016/08/308

9

2016

Better understanding and management of lung cancer are needed. Although much has been learned from known protein coding genes, long noncoding RNAs (lncRNAs), a relatively new and fast evolving large family of transcripts, have recently generated much attention for new discoveries. LncRNAs play critical regulatory functions and are emerging as new players in tumorigenesis and phenotypic determinators of lung cancer. In this review, we highlight the latest development of lncRNAs, including circular RNAs in lung cancer. We start with well-characterized lncRNAs and circular RNAs as an oncogene or tumor suppressor and then extend our discussion on the impact of SNPs in lncRNA on its functions and lung cancer risk and the clinical applications of lncRNAs as biomarkers and therapeutic targets.

First draft submitted: 6 April 2016; Accepted for publication: 21 June 2016; Submitted online: 2 September 2016

Keywords: biomarkers • circular RNA • long noncoding RNA • lung cancer • oncogenic lncRNA • tumor suppressive lncRNA

Lung cancer is one of the most common human cancers and is the leading cause of cancer-related deaths around the world [1]. Mainly as the result of late-stage diagnosis and insensitiv-ity to chemotherapy, the 5-year survival rate of lung cancer has been staggering at about 15% [2]. Better understanding of its develop-ment, progression and metastasis mechanisms can help identify new biomarkers for early diagnosis and more effective treatments. There are two major types of lung cancer with distinct clinical management and outcomes: small-cell lung cancer (SCLC), which is more aggressive with chemotherapy as the mainstay of treat-ment, and NSCLC, a more heterogeneous group with good outcome in early stage treated by surgery. NSCLC accounts for >80–85% of all lung cancer cases [3].

The relationship between protein-coding genes and lung cancer has been widely stud-ied over the past decades. However, the protein-coding genes are coded from only <2% of the human genome while 85% of

the human genome sequences are tran-scribed into noncoding RNAs, a new and poorly understood RNA family [4,5]. Based on transcript lengths, long noncoding RNAs are loosely classified into two categories: short noncoding RNAs (<200 nucleotides: e.g., miRNA, snRNA, snoRNS, siRNA and piRNA) and long noncoding RNAs (lncRNA, >200 nucleotides). LncRNAs have attracted much attention due to their large number and unknown functions. Although 14,880 lncRNAs coded from 9277 loci have been defined in the latest GENCODE project [6], novel lncRNAs are constantly being discovered.

The lncRNAs can be divided into seven subcategories by their genomic location (Figure 1A): exonic-sense lncRNAs overlap with the exonic region of a protein-coding gene and are transcribed from the same strand; exonic-antisense lncRNAs overlap with the exonic region of a protein-coding gene but transcribed from the opposite strand;


1276 Epigenomics (2016) 8(9)

Figure 1. Long noncoding RNA categories based on genomic locations and biological functions. (A) Long noncoding RNAs (lncRNAs) categorized based on their genomic locations: lncRNAs (red arrows) and their genomic positions are shown relative to the protein-coding genes (blue arrows) on a chromosome (black line). The exons of lncRNAs are represented by solid red boxes. The exons of protein-coding genes are represented by solid blue boxes (for each gene, only two exons are shown for simplicity of the illustration). The arrowheads point to the direction of transcription. The percentage of each category of lncRNAs are shown in brackets (based on Derrien et al. [6]). (B) LncRNAs categorized based on their biological functions. LncRNA (red hairpin) can function as a (I) signal; (II) decoy; (III) guide or (IV) scaffold to participate in the regulatory mechanisms. The gene names of lung cancer-related lncRNAs are given as examples for each category.

Exonic sense (<5%)

Exonic antisense (16%)

Intronic sense (4%)

Intronic antisense (15%)

Overlapping (1%)

Bidirectional (<5%)

>1000 bp >1000 bp

<1000 bp

Intergenic (64%)

SOX2OT

ANRILZXF1GHSROSGAS6-AS1

SPRY44-IT1

HOTAIRCCAT2MALAT1H19

I. Signal

II. Decoy

III. Guide

IV. Scaffold

CCAT2

MALAT1

H19

ANRIL

ANRIL

HOTAIR

HOTAIRSPRY44-IT1

MEG3TUG1

MEG3TUG1

Protein E

Protein D

Protein C

Protein C

Protein A

Protein B

Protei

n D

Protein F

Protein G

Activate gene expression

Methylation

Methylation

Suppressgene

expression


Review Xie, Yuan, Sun & Li

intronic lncRNAs are encoded wholly inside the introns of protein-coding genes and they can also be subdivided into sense and antisense classes; overlapping lncRNAs contain a protein-coding gene within its intron; bidi-rectional lncRNAs are located on the opposite strand from a neighboring protein-coding gene <1000 base pairs away; and intergenic lncRNAs are not in the proximity (i.e., >1000 base pairs away) of any protein-coding genes [6]. Functionally, lncRNA can be catego-rized into four general classes (Figure 1B): signals, where they serve as molecular signals by cell-specific expres-sion and response to diverse stimuli; decoys, where they bind to a protein to sequester it from its targets; guides, where they bind to a protein and then direct the ribonu-cleoprotein complex to specific targets in order to regu-late gene expression either in cis or in trans; and scaf-folds, where they act as a platform to assemble different molecules to interact or function together [4,7].

The dysregulation of lncRNAs in cancer has been well recognized, including lung cancer [7,8]. Moreover, recent research suggested that circular RNAs (cir-cRNAs) may also play an important role in the initiation and development of lung cancer. Although there has been no established classification for circRNAs, most consider them as a subcategory of lncRNA because the vast majority of known circRNAs do not translate into proteins, their lengths being greater than 200 bp, and perform regulatory functions, for example, working as miRNA sponges and transcriptional regulators [9,10].

In this review, we highlight the recent advances about lncRNAs in lung cancer. We discuss the lung cancer-related lncRNAs and emerging circRNAs; the effects of SNPs in lncRNAs on lung cancer risk; and the latest development of using lncRNAs as potential diagnostic and prognostic biomarkers and therapeutic targets.

www.futuremedicine.com 1277future science group

lncRNA & lung cancer Review

Lung cancer-related lncRNAsNumerous lncRNAs have been implicated in lung can-cer, but only a handful of them are well characterized for their biological functions and underlying mechanisms of action as summarized in Table 1. Like protein-coding genes, lncRNAs can also be categorized into oncogenic lncRNAs and tumor suppressive lncRNAs according to their dysregulated expression in cancer cells.

Oncogenic lncRNAs in lung cancerHOTAIRHOTAIR is a 2.4-kb antisense lncRNA transcribed within the HOXC gene cluster on chromosome 12 [44,45] where it is coexpressed along with the HOXC genes. It acts in trans to repress the transcription of HOXD genes on chromosome 2 [11,44,46]. HOTAIR is shuttled from chromosome 12 to chromosome 2 by PRC2. Its 5′ domain binds to PRC2, whereas its 3′ domain binds to LSD1/CoREST/REST complex, acting like a scaffold to assemble PRC2 and LSD1 into a complex and guide it to the HOXD gene cluster on chromosome 2 [46]. The binding of PRC2/LSD1 complex then leads to H3K27 methylation and H3K4 demethylation for epigenetic silencing of HOXD genes in multiple tis-sues [44]. HOTAIR is a well-known cancer-related lncRNA, highly expressed in NSCLC, SCLC as well as various other human cancers [8]. The knockdown of HOTAIR in cancer cells led to decreased proliferation activity and decreased metastasis in vitro [12,13]. Recent studies demonstrated its role in the chemoresistance to cisplatin in NSCLC, suggesting that HOTAIR can be used as a therapeutic target [14].

MALAT1MALAT1, also known as nuclear-enriched transcript 2, is an 8.7 kb intergenic lncRNA located on chromosome 11q13 [47]. The transcription of MALAT1 is regulated by the tumor suppressive protein p53, which binds to the promoter of MALAT1 and represses its transcrip-tion [15]. MALAT1 regulates mRNA transcription by sequestering pre-mRNA splicing factors from their targets, like a ‘molecular sponge’ decoy [16]. MALAT1 has been significantly associated with metastasis poten-tial and poor prognosis in NSCLC patients, especially in early-stage metastasizing patients [47,48]. In addi-tion to NSCLC, upregulation of MALAT1 has also been observed in multiple cancerous tissues, including human cancers of the breast, uterus, prostate, colon and liver [17]. Its expression showed positive associa-tions with proliferation and metastasis of tumor cells. RNA interference-mediated silencing of MALAT1 reduced the in vitro migration of lung adenocarcinoma cells by regulating the expression of motility-related genes [49].

CCAT2CCAT2 is a 1.7-kb intergenic lncRNA located on chro-mosome 8q24. The 8q24.21 genomic region is a highly conserved segment found in 28 species-conservation tracks (University of California, Santa Cruz Genome Browser [50]). CCAT2 was initially discovered in colon cancer as a novel lncRNA which promoted invasion and metastasis [18]. More recently, Qiu et al. reported a 7.5-fold upregulation of CCAT2 in NSCLC tissues compared with paired adjacent normal lung tissues, but the overexpression was only significant in lung ade-nocarcinoma but not in squamous cell carcinoma [19]. The exact role of CCAT2 in cancer is still not quite clear. Ling et al. demonstrated the involvement of CCAT2 in the WNT-signaling pathway through inter-acting with TCF7L2 to upregulate MYC, miR-17-5p and miR-20a [18].

H19H19 is a 2.3-kb intergenic lncRNA located in an imprinted genomic region 11p15.5. H19 is only expressed from the maternally inherited chromo-some [20]. Increased expression of H19 has been asso-ciated with hypomethylation of the promoter of H19. Loss of imprinting and H19 overexpression, together with hypomethylation of its promoter region, are fre-quently observed in lung cancer development [51]. H19 induces the expression of genes that are important for cell differentiation and proliferation, for example, c-jun and JNK1/2 in the JNK pathway [20]. Moreover, H19 act as a molecular sponge by sequestering miRNA let-7, the low expression of which has been associated with poor prognosis of lung cancer and multiple other cancers [52]. The role of H19 in cancer is still under debate. However, in human, most studies have shown that H19 plays oncogenic roles. H19 overexpression has been correlated with poor prognosis not only in lung [51], but also in bladder [21] and gastric cancers [53].

Tumor-suppressive lncRNAs in lung cancerMEG3MEG3 is 6.9 kb in length and located on 14q32.2. It is expressed from the maternal chromosome in many normal tissues [30]. MEG3 is believed to be a tumor-suppressive lncRNA because its expression is decreased in various human tumors, including lung cancer tis-sues [30]. It correlated negatively with poor prognosis in lung cancer [31]. Overexpression of MEG3 decreases proliferation and induces apoptosis of lung cancer cells in vitro and impeded tumorigenesis in vivo [32]. Decreased expression of MEG3 has been observed in cisplatin-resistant A549 lung cancer cells. In vitro experiment demonstrated that upregulation of MEG3 increased the sensitivity of A549 cells to cisplatin

1278 Epigenomics (2016) 8(9) future science group


Table 1. Lung cancer-related long noncoding RNAs.

LncRNA Type Function Pathway Mechanisms Ref.

Oncogenic (upregulated)

HOTAIR Intergenic Scaffold guide PRC2 Recruits PRC2 and LSD1/CoREST/REST complex to repress HOXD gene transcription

[11–14]

MALAT1(NEAT2) Intergenic Decoy scaffold guide

PRC2 p53

Enhances expression of PRC2-independent target genes such as SUZ12 and EZH2

[15–17]

CCAT2 Intergenic Signal Wnt Interacts with TCF7L2 to upregulate MYC, miR-17-5p, and miR-20a

[18,19]

H19 Intergenic Decoy p53, JNK

Induces the expression of c-jun and JNK1/2 in the JNK pathway and suppresses the expression of the β-5, β-3 and α-4 integrins

[20,21]

SCAL1 Intergenic Scaffold Nrf-2 Induced by cigarette smoke in airway epithelial cells in an NRF2-dependent manner

[22]

ANRIL Antisense Scaffold guide

PRC2 Binds to and recruits PRC2 to repress the expression of p15INK4B locus to and silencing of p15(INK4B)

[23]

LncRNA-DQ786227

Intronic Unknown Unknown High expression in lung cancer cell lines A549 and QG56

[24]

UCA1/CUDR Intergenic Decoy AKT mTOR

Upregulates the expression of miR-193a-3p target gene ERBB4 through competitively ‘sponging’ miR-193a-3p

[25]

ZXF1 Antisense Unknown Unknown Enhanced lncRNA ZXF1 expression in human lung adenocarcinoma was associated with the lymph node metastasis, tumor pathological stage and survival time

[26]

SOX2OT Sense Scaffold PRC2 Modulates expression levels of the G2/M transition-regulating proteins in lung cancer cells. Regulates but not directly binds to EZH2 protein

[27]

GHSROS Antisense Scaffold guide

Unknown Induces migration of A549 and NCI-H1299 NSCLC cell lines

[28]

CARLo-5 Intergenic Scaffold Unknown Induces cell proliferation in H1975 cells through suppressing G0/G1 arrest in cell cycle regulation Upregulates mesenchymal marker including vimentin and fibronectin and downregulates E-cadherin

[29]

Tumor suppressive (downregulated)

MEG3 Intergenic Scaffold guide

P53 PRC2 Wnt TGF-β

Upregulates the tumor suppressive protein p53 by downregulating MDM2 which is a p53 inhibitor. Guides the PRC2 complex to targets by forming a RNA–DNA triplex structure at the binding site

[30–33]

TUG1 Intergenic Scaffold guide

PRC2 P53 AkT MAPK

Binds to the PRC2 complex and guides it to the HOXB7 promoter to suppress its expression

[34–36]

SPRY4-IT1 Intronic Scaffold PRC2 Mediates cell growth, proliferation and apoptosis [37]

GAS6-AS1 Antisense Unknown Unknown GAS6-AS1 level is inversely correlated with GAS6 mRNA level

[38]

BANCR Intergenic Scaffold Unknown Induces E-cadherin expression, while suppresses N-cadherin, vimentin and MMP-2 expression in epithelial–mesenchymal transition

[39]

LncRNA: Long noncoding RNA.



through activation of the WNT/β-catenin signaling pathway, whereas downregulation of MEG3 caused the opposite effect [31,33]. MEG3 can upregulate the tumor suppressive protein p53 by downregulating the expression of mouse double minute 2 homolog, which is an E3 ubiquitin-protein ligase that inhibits of p53 transcription [32]. MEG3 also interacts with the PRC2 complex by binding to the EZH2 subunit and then guides the PRC2 complex to the regulatory elements of the target genes by forming a RNA–DNA triplex structure at the binding site [54].

TUG1TUG1 is a 5.6-kb intergenic lncRNA located on 22q12.2. The expression of TUG1 is induced by the tumor suppressive protein p53. Like HOTAIR, MALAT1 and MEG3, TUG1 also interacts with the PRC2 complex. TUG1 binds to the EZH2 subunit in the PRC2 complex and guides it to the promoter region of homeobox B7 (HOXB7) [55,56]. HOXB7 is a known oncogene that promotes cell proliferation through acti-vating AkT and MAPK pathways. The binding of the TUG1/PRC2 complex to HOXB7 promoter leads to the downregulation of HOXB7 expression in NSCLC tissues. Inhibition of TUG1 could upregulate HOXB7 expression [55].

TUG1 was significantly downregulated in lung can-cer tissues compared with the corresponding normal lung tissues. Downregulation of TUG1 also correlates with advanced pathological stage, greater tumor size and shorter survival time in both lung squamous cell carcinoma and lung adenocarcinoma [56]. Aberrant expressions of TUG1 have also been observed in vari-ous types of cancer, for example, bladder cancer [34], hepatocellular carcinoma [35], esophageal squamous cell carcinoma [36] and osteosarcoma [57]. Intrigu-ingly, TUG1 is upregulated in those cancers instead of downregulated like in lung cancer. This suggests that

TUG1 may have tissue-specific expression patterns and perform different functions in various human tumors.

SPRY4-IT1SPRY4-IT1 is a 0.7-kb lncRNA located on 5q31.3. SPRY4-IT1 is transcribed from an intron of SPRY4. The downregulation of SPRY4-IT1 occurs through tran-scriptional repression mediated by EZH2 that directly binds to SPRY4-IT1 promoter and induces H3K27me3 modification [37]. In a recent study, Sun et al. showed that downregulation of SPRY4-IT1 promoted A549 cell migration and invasion in vitro, whereas overex-pression promoted apoptosis [37]. Additionally, in vivo experiments in nude mice indicated that ectopic over-expression of SPRY4-IT1 was able to reduce the num-ber of metastatic nodules compared with the control group. The expression level of SPRY4-IT1 was down-regulated in lung cancer tissues compared with nor-mal lung tissues. Sun et al. also demonstrated that the downregulation of SPRY4-IT1 correlated with larger tumor size, advanced pathological stage and lymph node metastasis in NSCLC patients.

Lung cancer-related circRNAsCircular RNA (circRNA) is a newly identified cat-egory of lncRNAs without the polyadenylated tail. According to how they are generated, circRNAs can be divided into two subtypes: the one that gener-ated by a noncanonical ‘backsplice circularization’ of exons and another from intron lariats that fail to be debranched after splicing but covalently circularized with 2′,5′-phosphodiester bond between a splice donor site and a branch point site (circular intronic long noncoding RNAs [ciRNAs]) [45,58,59].

Circular RNAs were originally thought to be rare byproducts of splicing errors without biological func-tion. However, the recent studies through RNA sequencing by Memczak et al. and Jeck et al. identified

LncRNA Type Function Pathway Mechanisms Ref.

AK126698 Intergenic Unknown Wnt Knockdown of AK126698 increased protein levels of NKD2 and phospho-β-catenin, activated canonical Wnt signaling pathway and induced cisplatin resistance in A549 cell line

[40]

LncRNA-LET Intergenic Scaffold NF90 CDC42

Promotes NF90 protein ubiquitination and subsequent degradation. Decreased the CDC42 and increased the DUSP1 protein levels under normoxic conditions

[41]

GAS5 Intergenic Scaffold decoy

P53 E2F1 mTOR

Upregulated p53 expression and downregulated transcription factor E2F1 expression in A549 cells

[42] [43]

LncRNA: Long noncoding RNA.

Table 1. Lung cancer-related long noncoding RNAs (cont.).

1280 Epigenomics (2016) 8(9)

Figure 2. CDR1as-mediated regulation of miR-7. The circular RNA CDR1as functions as a miR-7 reservoir that sequesters miR-7 from their target genes. The perfectly complementary binding by miR-671 would induce CDR1as cleavage and release miR-7, which, in turn, leads to the repression of miR-7 target genes.

Tumor suppressor gene

miR-7

miR-671

CDR1as



>2000 of circRNAs in human genome [60,61]. It is get-ting clear that back-splicing derived circRNAs can function as miRNA sponges which fine-tunes the level of miRNA-mediated regulation of gene expression by blocking the miRNA [62]. The functional and clini-cal implications of circular RNAs in lung cancer have started emerging. Circular RNAs demonstrate tissue-specific induction in lung during human fetal develop-ment [63]. Hansen et al. [64] reported that a circRNA CDR1as was frequently expressed in lung carcinomas, which was coexpressed with miR-7 [65]. The miRNA miR-7 has been found to be associated with poor prog-nosis of lung cancer [66], and the inhibition of miR-7 caused increased apoptosis in lung cancer cell lines [67]. CDR1as contains 74 binding sites for miR-7. These binding sites are not perfect matches; hence, CDR1as is able to sponge away a substantial amount of miR-7 without being digested by RNA-induced silencing complex [60]. Another miRNA, miR-671, which shows near-perfect complementarity with CDR1as would trig-ger endonucleolytic cleavage and release many miR-7, which are then active for regulation (Figure 2) [68]. Apart from CDR1as, Liu et al. discovered that a cir-cRNA produced from a back-splicing event at the ZEB1 gene locus can sponge miR-200 [69]. miR-200 is a well-known lung cancer-related miRNA [70] and has been shown to regulate the expression of ZEB1 gene in lung adenocarcinoma in a double negative feedback loop [71]. These lines of evidence suggest that the ZEB1 circRNA may influence lung cancer by competing with miR-200.

Although our understanding of circular RNAs is still in its infancy, the bioinformatics tools and resources that can help us to conduct more research and inter-pret circular RNAs are quickly emerging. CircBase is public database where investigators can search, browse and download genomic annotations of circRNAs [72].

CircNet database allows the users to explore circRNA expression profiles and miRNA target networks [69]. CircInteractome database can be used to map miRNA-binding sites and RNA-binding proteins on circRNAs, design junction-spanning primers for circRNA detec-tion and design siRNAs for circRNA silencing experi-ments [73]. Circ2Trait database hosts information of 1951 circRNAs with a complete putative interaction network of miRNA–circRNA–mRNA and their asso-ciations with 105 different diseases. Circ2Trait also provides a database of disease-associated SNPs mapped on circRNA loci [74].

Associations of lncRNA SNPs with lung cancerThe dysregulation of lncRNAs in cancer raises the question of whether SNPs located in the lncRNA genes would affect their expression, and if so, whether these SNPs are associated with cancer risks in general popu-lation. Evidence is mounting that even a small point mutation in an lncRNA is able to change its structure or expression level, thus affect its regulatory func-tions [18,75]. For example, Zhang et al. demonstrated that the expression of lncRNA TUG1 in lung cancer was regulated by tumor suppressive protein p53, but only by the wild type p53 [56]. The DNA-binding domain in p53 contained a frequent cancer-associated point mutation R175H. The R175H-mutant-type p53 was not able to influence TUG1 expression [75]. Another example is the well-known SNP rs6983267 in NSCLC-related lncRNA CCAT2. The risk allele G of rs6983267 is associated with increased CCAT2 expression [18]. In a very recent study, Gong et al. per-formed an association study of 13 SNPs in the well-characterized lncRNAs with lung cancer susceptibil-ity. Even with a limited sample size of 498 cases and



213 controls, they detected a significant association between CCAT2 SNP rs6983267 and lung cancer susceptibility [76]. Like protein-coding genes, the asso-ciation of SNPs with lncRNA gene expression can be established through long intergenic noncoding RNAs expression quantitative trait loci (lincRNA eQTL). Kumar et al. showed that many genome-wide associa-tion study (GWAS) SNPs are associated with disease predisposition via their effects on lincRNA expression. Moreover, lincRNA eQTL demonstrated a strong tis-sue dependency [77]. Popadin et al. demonstrated that the cis-eQTLs generally have a larger effect size on expression levels of lincRNAs and closer to the tran-scription start sites in comparison with the cis-eQTLs of protein-coding genes [78]. Considerable overlaps between the promoters of lincRNAs and the enhanc-ers of protein-coding genes were observed [78]. In both studies, the main difficulty of identifying lincRNA eQTLs was caused by the high tissue specificity and low abundance of lincRNAs.

LncRNAs as lung cancer diagnostic & prognostic biomarkersThe tissue and cancer specific expression of lncRNAs makes them ideal biomarkers. As shown in Table 2, HOTAIR, MALAT1, CCAT2 and H19 are upregulated and MEG3, TUG1 and SPRY4-IT1 are downregulated in lung cancer. Their clinical utilities as biomarkers have been explored in previous studies. An early study by Ji et al. demonstrated that MALAT1 was signifi-cantly associated with metastasis in NSCLC patients and could be used as a prognostic biomarker for over-all survival (OS) for patients diagnosed in early-stage NSCLC [47]. A subsequent study by Schmidt et al. further indicated that MALAT1 can be utilized as an independent prognostic parameter for survival for both lung squamous cell carcinoma and adenocarcinoma patients [79]. Weber et al. evaluated the effectiveness of MALAT1 as a blood-based biomarker for NSCLC and showed that MALAT1 had the characteristics of a good diagnostic biomarker in terms of easy accessibil-ity in body fluids and high specificity (96%). How-ever, the issue came with its low sensitivity of 56% and combined use with other biomarkers may improve the diagnostic performance [80]. When MALAT1 was combined with SPRY4-IT1 and ANRIL, a much higher diagnostic performance than each lncRNA alone was achieved (AUC: 0.876; sensitivity: 82.8%; specificity: 92.3%) [81]. Unlike MALAT1 which was overexpressed in both lung squamous cell carcinoma and adenocarcinoma, CCAT2 was highly expressed in adenocarcinoma only, which made it a potential lung adenocarcinoma-specific diagnostic biomarker [19]. Low expression of MEG3 and BANCR can both be

used as a biomarker for poor prognosis (shorter OS) in NSCLC patients [32,39]. Also, the decreased expres-sion of GAS6-AS1 with lymph node metastasis status may be used as a biomarker of late stage and unfavor-able survival [38]. A score-based stratification of lung cancer patients by multiple lncRNAs may also be use-ful. Zhou et al. identified eight lncRNAs and then formulated a risk score based on the expression data. By dividing NSCLC patients into high- and low-risk groups using the median risk score as a cut-off value, they found that patients in the high-risk group had a significantly shorter OS than those in the low-risk group (median OS: 1.67 vs 6.06 years, log-rank test p = 4.33 × 10-9) [82].

In addition to the well-characterized (or known) lncRNAs, White et al. analyzed the publicly available RNA-seq data of lung cancer patients from The Can-cer Genome Atlas for novel lncRNAs [58]. The study detected 3452 novel lncRNAs, 111 of which showed strong differential expression between lung and adja-cent normal tissue. Moreover, 27 of the 111 lncRNAs were differentially expressed between lung squamous cell carcinoma and adenocarcinoma tumors, which could potentially serve as cell histologic type-specific biomarkers. They also tried to validate some well-characterized lung cancer lncRNAs, such as HOTAIR, MALAT1, ANRIL, SCAL1, H19 and MEG3, but failed, as most of these lncRNAs were either very weakly expressed (e.g., HOTAIR, ANRIL, SCAL1 and MEG3) or showed no altered expression (e.g., MALAT1 and H19) in The Cancer Genome Atlas lung cancer cohorts. SCAL1 was the only known lung cancer lncRNA that has been picked up among the list of 111 lncRNAs. It is worth noting that the data used in this study were from poly(A) selected RNA-seq and their focus was inter-genic lncRNAs. Nonintergenic and poly(A) negative lncRNAs would be missed, such as circRNAs.

As a newly identified member of the lncRNA fam-ily, circRNA may have its potential as a promising biomarker for cancer detection. Li et al. found abun-dant and stable presence of >1000 circRNAs in human serum exosomes. They examined various cancer cell lines, including lung cancer, and detected an enrich-ment of circRNA in exosomes in lung, colon, stomach, breast and cervical cancers cells by qRT-PCR analysis. CircRNAs are known to bind miRNAs which are also abundant in exosomes, for example, CDR1as circRNA is known to act as an miR-7 sponge [60,62], which is a known lung cancer-associated miRNA [86]. To prove the role of CDR1as as miR-7 sponge in exosomes, Li et al. introduced miR-7 mimics into cells and observed a significant downregulation of CDR1as circRNA in exosomes. They also showed that the circRNAs in exo-somes were readily measurable in human blood even



Table 2. Lung cancer-related long noncoding RNAs as diagnostic, prognostic and predictive biomarkers.

LncRNA Diagnostic Prognostic Predictive Ref.

HOTAIR Upregulated in NSCLC, SCLC

Upregulation associated with poor survival in NSCLC

Upregulation decreases chemosensitivity to cisplatin

[8,14]

MALAT1 (NEAT2) Upregulated in NSCLC Upregulation associated with metastasis potential and poor survival in NSCLC

Upregulation decreases chemosensitivity to gemcitabine

[47,48,83]

CCAT2 Upregulated in LUAD but not LUSC

Low-predictive efficiency for lymph node metastasis unless combined with serum tumor biomarkers carcinoembryonic antigen

Unknown [18,19]

H19 Upregulated in NSCLC Upregulation increased tumor cell invasion

Unknown [20,21]

SCAL1 Upregulated in smokers and lung cancer cell lines

Unknown Regulated by NRF2 which is a transcription factor that can predict chemoresistance in NSCLC

[22,84]

ANRIL Upregulated in NSCLC Upregulation associated with tumor-node metastasis stages, tumor size and poor prognosis

Unknown [23]

UCA1/CUDR Upregulated in NSCLC patients resistant to EGFR-TKIs and lung cancer cell lines

Unknown Overexpression may induce EGFR-TKIs resistance by activating the AkT/mTOR pathway which is a therapeutic target in cancer treatment

[85]

ZXF1 Upregulated in LUAD Upregulation increased invasion, metastasis and poor survival in LUAD

Unknown [26]

SOX2OT Upregulated in LUSC but not LUAD

Upregulation associated with poor survival

Unknown [27]

CARLo-5 Upregulated in NSCLC Upregulation increased proliferation, migration and invasion in NSCLC cell lines

Unknown [29]

MEG3 Downregulated in NSCLC

Low expression associated with poor survival in NSCLC

Downregulation decreases chemosensitivity to cisplatin

[30–33]

TUG1 Downregulated in NSCLC

Low expression associated with advanced pathological stage, greater tumor size and shorter survival time in both LUAD and LUSC

Unknown [34–36]

SPRY4-IT1 Downregulated in NSCLC

Low expression associated with advanced pathological stage, greater tumor size, lymph node metastasis and poor survival in NSCLC

Unknown [37]

GAS6-AS1 Downregulated in NSCLC

Low expression associated with poor survival in late-stage NSCLC

Unknown [38]

BANCR Downregulated in NSCLC

Low expression associated with poor survival in NSCLC

Unknown [39]

LncRNA: Long noncoding RNA; LUAD: Lung adenocarcinoma; LUSC: Lung squamous cell carcinoma; SCLC: Small-cell lung cancer; TKI:Tyrosine kinase inhibitor.



after a 24-h incubation at room temperature [87], which is a favorable characteristic of a biomarker.

These findings suggest that lncRNAs may become particularly useful in noninvasive screening as poten-tial circulating biomarker for cancer diagnosis and prognosis. However, there are also some challenges of using lncRNAs as biomarkers. For instance, the amount of lncRNAs in plasma is generally low. There-fore, unlike short noncoding RNAs, such as miR-NAs, most lncRNAs are not detectable in plasma by standard methods, such as microarrays or quantita-tive PCR [5]. More studies are required to examine whether lncRNAs are better predictive biomarkers than protein-coding genes or other noncoding RNAs such as miRNAs.

LncRNAs as lung cancer therapeutic targetsThe functions of lncRNAs can be modulated by dis-rupting the interaction between lncRNAs with either protein factors or small ligands, or inhibit lncRNAs using siRNAs, antisense oligonucleotide, ribozyme and aptamer [88]. Numerous studies have explored the possibilities of using lncRNA as drug targets. For example, Chen et al. demonstrated that both HOTAIR and AK126698 participated in chemotherapy resis-tance to cisplatin in NSCLC patients: upregulation of HOTAIR and downregulation of AK126698 correlated with increased cisplatin resistance. Therefore, silencing HOTAIR and sensitizing AK126698 may be an efficient therapeutic intervention to alleviate cisplatin resistance in NSCLC patients [7]. It has been shown that HOTAIR knockdown reduced proliferation and enhanced radiosensitivity in pancreatic cancer cells [89]. Further researches are required to investigate whether HOTAIR knockdown would also affect the radiosensitivity in lung cancer cells. On the other hand, Liu et al. demonstrated that the overexpression of MEG3 in A549 cells increased their chemosensitivity to cisplatin, whereas knockdown MEG3 in A549 cells decreased the chemosensitivity, sug-gesting that MEG3 can be used as a potential therapeutic target for lung cancer chemotherapy [31].

So far, several lines of evidence have indicated that lncRNAs are very promising as novel therapeutic tar-

gets in lung cancer and other cancers. However, unlike miRNAs, there is still a paucity of clinical applica-tion of lncRNAs in translational oncology. Some lncRNAs have been used as predictive biomarkers in clinical trials. For example, in breast cancer patients, a study has been under Phase II and III trials about using an mRNA–lncRNA signature to predict the efficacy of various combinations of chemotherapy drugs (NCT02641847). Similar study designs could be applied to lung cancer patients to create personal-ized chemotherapy based on the patients’ lncRNA pro-files related to chemoresistance (Table 2). In NSCLC, mTOR-targeted therapy has been tested on Phase I trial (NCT01390818 [90]), therefore, lung cancer-related lncRNAs such as UCA1 and GAS5 may help select patients who are eligible for mTOR-targeted therapy because of the association between UCA1 and GAS5 expression with mTOR signaling pathway [25,42,43].

Apart from being used as predictive biomarkers in therapeutic applications, the lncRNA-based drugs for lung cancer patients remain poor. There is one prom-ising example about the lung cancer-related lncRNA H19. The H19 promoter has been built into a DNA plasmid BC-819 upstream of a Diphtheria Toxin ‘A’ sequence. H19 is substantially expressed in cancer-ous cells. When BC-819 enters a cancerous cell, the Diphtheria Toxin ‘A’ chain is synthesized under the regulation of H19 promoter to kill cancerous cells [91]. Phase I and II trials have been ongoing in patients with bladder and pancreatic cancer (NCT01878188; NCT01413087). So far, there has been no clinical trial for lung cancer, but a mouse model study suggested that inhaled BC-819 reduced primary lung cancer growth in mice induced by intravenous injection of A549-C8-luc cells [92].

Some factors should be taken into consideration for designing lncRNAs as therapeutic targets: the turn-over rate and expression levels of lncRNAs should be relatively high, because it is not feasible to target transiently expressed or low abundant lncRNAs; drugs targeting lncRNAs should be specific and avoid inter-fering with other molecules with similar secondary and tertiary structure; by modulating the lncRNA levels,

LncRNA Diagnostic Prognostic Predictive Ref.

AK126698 Unknown Unknown Downregulation decreases chemosensitivity to cisplatin in A549 cell line

[40]

GAS5 Downregulated in NSCLC

Low expression associated with larger tumor size and poor survival in NSCLC

Downregulation decreases chemosensitivity to trastuzumab

[43]

LncRNA: Long noncoding RNA; LUAD: Lung adenocarcinoma; LUSC: Lung squamous cell carcinoma; SCLC: Small-cell lung cancer; TKI:Tyrosine kinase inhibitor.

Table 2. Lung cancer-related long noncoding RNAs as diagnostic, prognostic and predictive biomarkers (cont.).



miRNAs that interact with the target lncRNAs in the ceRNA network may also be influenced.

Conclusion & future perspectiveLung cancer-related lncRNAs are still an emerging field, with only a handful of them being character-ized for their biological functions and clinical implica-tions. They are known to participate in a wide range of biological mechanisms and dynamically expressed in tissue-specific manner. Their expression levels are gen-erally lower, and under weaker evolutionary constraint than protein-coding genes.

As shown in Table 1, it is worth noting that about half of the lung cancer-related lncRNAs directly inter-act with PRC2 complex or involved in the p53 pathway. The frequent dysregulation of PRC2-related lncRNAs has been pointed out before by Khalil et al. [93]. By genome-wide RNA immunoprecipitation analy-sis, they identified that approximately 20% of the lncRNAs involved in various pathological processes are bound to PRC2. Further investigations about the roles of lncRNAs in PRC2 and p53 pathways may discover potential therapeutic targets in tumor progression.

More research is required to investigate SNPs that affect the expression levels and biological func-tions of lncRNAs. Taking together the evidence from eQTLs and ENCODES regulatory tracks, we may have a deeper insight into the regulation net-works between lncRNAs and genetic variations. This would also help us to pin down the functional varia-tions in the noncoding regions identified in previous GWAS studies.

The poor prognosis of lung cancer can be improved by biomarkers with high sensitivity, specificity and easy accessibility in noninvasive screening. In these aspects, lncRNAs are very promising as new biomarkers in lung cancer. Their diagnostic power could be improved by combining several lung cancer-related lncRNAs into a biomarker panel. In addition, lncRNAs may prove use-ful as predictive markers for chemotherapy sensitivity in tailored anticancer treatment.

The full potential of using lncRNAs in cancer therapy has not yet been explored because only a few lncRNAs have their functions thoroughly investi-gated. The future of lncRNA therapy relies on a better understanding of their precise biological functions.

Executive summary

Lung cancer-related long noncoding RNAs• Long noncoding RNAs (lncRNAs) account for a vast majority of noncoding transcripts in human genome and

the dysregulation of lncRNAs takes part in the tumorigenesis of lung cancer through a variety of mechanisms.• More than 20 lncRNAs have been well characterized in lung cancer, including oncogenic lncRNAs (such as

HOTAIR, MALAT1, CCAT2 and H19) and tumor suppressive lncRNAs (such as MEG3, TUG1, SPRY4-IT1 and H19). Most of them are also known to be associated with various other cancers.

• Lung cancer-related lncRNAs are commonly involved in PRC2, p53 and WNT-signaling pathways.Lung cancer-related circular RNAs• Circular RNA (circRNA) is a newly identified category of lncRNAs. More than 2000 of cirRNAs have been

identified in human genome.• CircRNA CDR1as has been associated with lung cancer. It functions as a miRNA sponge to bind miR-7, which is a

known lung cancer-associated miRNA.The association of lncRNA SNPs with lung cancer• Several SNPs in lung cancer-related lncRNAs have been associated with lung cancer susceptibility.• Previous studies have demonstrated that SNPs could affect lncRNA expression and/or their binding capacity

with regulatory proteins and miRNAs.LncRNAs as lung cancer diagnostic & prognostic biomarkers• LncRNAs are qualified as useful biomarkers for their easy accessibility in body fluids and highly specific

expression in tumor tissues. Several well-characterized lncRNAs have been identified as effective and stable blood-based biomarkers for the diagnosis or prognosis of lung cancer.

• The issue of low sensitivity of using a single lncRNA as a biomarker could be improved by combining the expression profile of several lncRNAs and formulate a risk score-based evaluation.

LncRNAs as lung cancer therapeutic targets• LncRNAs associated with chemotherapy resistance in lung cancer patients are particularly interesting

therapeutic targets (e.g., HOTAIR, AK126698 and MEG3). Although lncRNAs have not been applied in therapeutic interventions in lung cancer patients, some of them have been used as predictive biomarkers on clinical trials to predict the clinical response to certain drugs in chemotherapy.

Summary & future directions• Numerous lncRNAs have been identified to play oncogenic or tumor suppressive roles in lung cancer and

various other cancers. A better understanding of the functions of lncRNAs would help realize their therapeutic potential.



Financial & competing interests disclosureThis work is partly supported by the National Natural Sci-

ence Foundation of China (No. 81171903 and No. 81472190

to Y Li), the Chongqing Natural Science Foundation of China

(No. cstc2015jcyjBX0110 to Y Li) and Mayo Clinic Center for

Individualized Medicine. The authors have no other relevant

affiliations or financial involvement with any organization or

entity with a financial interest in or financial conflict with the

subject matter or materials discussed in the manuscript apart

from those disclosed.


manuscript.

References1 Jemal A, Bray F, Center MM et al. Global cancer statistics.

CA Cancer. J. Clin. 61(2), 69–90 (2011).

2 Polanski J, Jankowska-Polanska B, Rosinczuk J et al. Quality of life of patients with lung cancer. Onco Targets Ther. 9, 1023–1028 (2016).

3 Reck M, Popat S, Reinmuth N et al. Metastatic non-small-cell lung cancer (NSCLC): ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann. Oncol. 25(Suppl. 3), iii27–iii39 (2014).

4 Fatica A, Bozzoni I. Long non-coding RNAs: new players in cell differentiation and development. Nat. Rev. Genet. 15(1), 7–21 (2014).

5 Skroblin P, Mayr M. ‘Going long’: long non-coding RNAs as biomarkers. Circ. Res. 115(7), 607–609 (2014).

6 Derrien T, Johnson R, Bussotti G et al. The GENCODE v7 catalog of human long noncoding RNAs: analysis of their gene structure, evolution, and expression. Genome Res. 22(9), 1775–1789 (2012).

7 Chen J, Wang R, Zhang K et al. Long non-coding RNAs in non-small-cell lung cancer as biomarkers and therapeutic targets. J. Cell. Mol. Med. 18(12), 2425–2436 (2014).

8 Sang H, Liu H, Xiong P et al. Long non-coding RNA functions in lung cancer. Tumour Biol. 36(6), 4027–4037 (2015).

9 Qu S, Yang X, Li X et al. Circular RNA: a new star of noncoding RNAs. Cancer Lett. 365(2), 141–148 (2015).

10 Zhang Y, Yang L, Chen LL. Life without a tail: new formats of long noncoding RNAs. Int. J. Biochem. Cell. Biol. 54, 338–349 (2014).

11 Gupta RA, Shah N, Wang KC et al. Long non-coding RNA HOTAIR reprograms chromatin state to promote cancer metastasis. Nature 464(7291), 1071–1076 (2010).

12 Nakagawa T, Endo H, Yokoyama M et al. Large noncoding RNA HOTAIR enhances aggressive biological behavior and is associated with short disease-free survival in human non-small-cell lung cancer. Biochem. Biophys. Res. Commun. 436(2), 319–324 (2013).

13 Liu XH, Liu ZL, Sun M et al. The long non-coding RNA HOTAIR indicates a poor prognosis and promotes metastasis in non-small-cell lung cancer. BMC Cancer 13, 464 (2013).

14 Liu Z, Sun M, Lu K et al. The long noncoding RNA HOTAIR contributes to cisplatin resistance of human lung adenocarcinoma cells via downregualtion of p21(WAF1/CIP1) expression. PLoS ONE 8(10), e77293 (2013).

15 Ma XY, Wang JH, Wang JL et al. Malat1 as an evolutionarily conserved lncRNA, plays a positive role in regulating proliferation and maintaining undifferentiated status of

early-stage hematopoietic cells. BMC Genomics 16, 676 (2015).

16 Tripathi V, Shen Z, Chakraborty A et al. Long noncoding RNA MALAT1 controls cell cycle progression by regulating the expression of oncogenic transcription factor B-MYB. PLoS Genet. 9(3), e1003368 (2013).

17 Wei Y, Niu B. Role of MALAT1 as a prognostic factor for survival in various cancers: a systematic review of the literature with meta-analysis. Dis. Markers 2015, 164635 (2015).

18 Ling H, Spizzo R, Atlasi Y et al. CCAT2, a novel noncoding RNA mapping to 8q24, underlies metastatic progression and chromosomal instability in colon cancer. Genome Res. 23(9), 1446–1461 (2013).

19 Qiu M, Xu Y, Yang X et al. CCAT2 is a lung adenocarcinoma-specific long non-coding RNA and promotes invasion of non-small-cell lung cancer. Tumour Biol. 35(6), 5375–5380 (2014).

20 Ayesh S, Matouk I, Schneider T et al. Possible physiological role of H19 RNA. Mol. Carcinog. 35(2), 63–74 (2002).

21 Ariel I, de Groot N, Hochberg A. Imprinted H19 gene expression in embryogenesis and human cancer: the oncofetal connection. Am. J. Med. Genet. 91(1), 46–50 (2000).

22 Thai P, Statt S, Chen CH et al. Characterization of a novel long noncoding RNA, SCAL1, induced by cigarette smoke and elevated in lung cancer cell lines. Am. J. Respir. Cell Mol. Biol. 49(2), 204–211 (2013).

23 Nie FQ, Sun M, Yang JS et al. Long noncoding RNA ANRIL promotes non-small-cell lung cancer cell proliferation and inhibits apoptosis by silencing KLF2 and P21 expression. Mol. Cancer Ther. 14(1), 268–277 (2015).

24 Gao L, Mai A, Li X et al. LncRNA-DQ786227-mediated cell malignant transformation induced by benzo(a)pyrene. Toxicol. Lett. 223(2), 205–210 (2013).

25 Wang HM, Lu JH, Chen WY et al. Upregulated lncRNA-UCA1 contributes to progression of lung cancer and is closely related to clinical diagnosis as a predictive biomarker in plasma. Int. J. Clin. Exp. Med. 8(7), 11824–11830 (2015).

26 Zhang L, Zhou XF, Pan GF et al. Enhanced expression of long non-coding RNA ZXF1 promoted the invasion and metastasis in lung adenocarcinoma. Biomed. Pharmacother. 68(4), 401–407 (2014).

27 Hou Z, Zhao W, Zhou J et al. A long noncoding RNA Sox2ot regulates lung cancer cell proliferation and is a prognostic indicator of poor survival. Int. J. Biochem. Cell. Biol. 53, 380–388 (2014).

28 Whiteside EJ, Seim I, Pauli JP et al. Identification of a long non-coding RNA gene, growth hormone secretagogue receptor opposite strand, which stimulates cell migration in non-small-cell lung cancer cell lines. Int. J. Oncol. 43(2), 566–574 (2013).



29 Luo J, Tang L, Zhang J et al. Long non-coding RNA CARLo-5 is a negative prognostic factor and exhibits tumor pro-oncogenic activity in non-small-cell lung cancer. Tumor Biol. 35(11), 11541–11549 (2014).

30 Zhou Y, Zhang X, Klibanski A. MEG3 noncoding RNA: a tumor suppressor. J. Mol. Endocrinol. 48(3), R45–R53 (2012).

31 Liu J, Wan L, Lu K et al. The long noncoding RNA MEG3 contributes to cisplatin resistance of human lung adenocarcinoma. PLoS ONE 10(5), e0114586 (2015).

32 Lu KH, Li W, Liu XH et al. Long non-coding RNA MEG3 inhibits NSCLC cells proliferation and induces apoptosis by affecting p53 expression. BMC Cancer 13, 461 (2013).

33 Xia Y, He Z, Liu B et al. Downregulation of Meg3 enhances cisplatin resistance of lung cancer cells through activation of the WNT/beta-catenin signaling pathway. Mol. Med. Rep. 12(3), 4530–4537 (2015).

34 Tan J, Qiu K, Li M et al. Double-negative feedback loop between long non-coding RNA TUG1 and miR-145 promotes epithelial to mesenchymal transition and radioresistance in human bladder cancer cells. FEBS Lett. 589(20 Pt B), 3175–3181 (2015).

35 Huang MD, Chen WM, Qi FZ et al. Long non-coding RNA TUG1 is up-regulated in hepatocellular carcinoma and promotes cell growth and apoptosis by epigenetically silencing of KLF2. Mol. Cancer 14, 165 (2015).

36 Xu Y, Wang J, Qiu M et al. Upregulation of the long noncoding RNA TUG1 promotes proliferation and migration of esophageal squamous cell carcinoma. Tumour Biol. 36(3), 1643–1651 (2015).

37 Sun M, Liu XH, Lu KH et al. EZH2-mediated epigenetic suppression of long noncoding RNA SPRY4-IT1 promotes NSCLC cell proliferation and metastasis by affecting the epithelial–mesenchymal transition. Cell Death Dis. 5, e1298 (2014).

38 Han L, Kong R, Yin DD et al. Low expression of long noncoding RNA GAS6-AS1 predicts a poor prognosis in patients with NSCLC. Med. Oncol. 30(4), 694 (2013).

39 Sun M, Liu XH, Wang KM et al. Downregulation of BRAF activated non-coding RNA is associated with poor prognosis for non-small-cell lung cancer and promotes metastasis by affecting epithelial–mesenchymal transition. Mol. Cancer. 13, 68 (2014).

40 Yang Y, Li H, Hou S et al. The noncoding RNA expression profile and the effect of lncRNA AK126698 on cisplatin resistance in non-small-cell lung cancer cell. PLoS ONE 8(5), e65309 (2013).

41 Yang F, Huo XS, Yuan SX et al. Repression of the long noncoding RNA-LET by histone deacetylase 3 contributes to hypoxia-mediated metastasis. Mol. Cell 49(6), 1083–1096 (2013).

42 Shi X, Sun M, Liu H et al. A critical role for the long non-coding RNA GAS5 in proliferation and apoptosis in non-small-cell lung cancer. Mol. Carcinog. 54(Suppl. 1), E1–E12 (2015).

43 Li W, Zhai L, Wang H et al. Downregulation of LncRNA GAS5 causes trastuzumab resistance in breast cancer.

Oncotarget doi: 10.18632/oncotarget.8413 (2016) (Epub ahead of print).

44 Rinn JL, Kertesz M, Wang JK et al. Functional demarcation of active and silent chromatin domains in human HOX loci by noncoding RNAs. Cell 129(7), 1311–1323 (2007).

45 Hauptman N, Glavac D. Long non-coding RNA in cancer. Int. J. Mol. Sci. 14(3), 4655–4669 (2013).

46 Tsai MC, Manor O, Wan Y et al. Long noncoding RNA as modular scaffold of histone modification complexes. Science 329(5992), 689–693 (2010).

47 Ji P, Diederichs S, Wang W et al. MALAT-1, a novel noncoding RNA, and thymosin beta4 predict metastasis and survival in early-stage non-small-cell lung cancer. Oncogene 22(39), 8031–8041 (2003).

48 Luo JH, Ren B, Keryanov S et al. Transcriptomic and genomic analysis of human hepatocellular carcinomas and hepatoblastomas. Hepatology 44(4), 1012–1024 (2006).

49 Tano K, Mizuno R, Okada T et al. MALAT-1 enhances cell motility of lung adenocarcinoma cells by influencing the expression of motility-related genes. FEBS Lett. 584(22), 4575–4580 (2010).

50 UCSC Genome Browser. http://genome.ucsc.edu/

51 Kondo M, Suzuki H, Ueda R et al. Frequent loss of imprinting of the H19 gene is often associated with its overexpression in human lung cancers. Oncogene 10(6), 1193–1198 (1995).

52 Kallen AN, Zhou XB, Xu J et al. The imprinted H19 lncRNA antagonizes let-7 microRNAs. Mol. Cell 52(1), 101–112 (2013).

53 Li H, Yu B, Li J et al. Overexpression of lncRNA H19 enhances carcinogenesis and metastasis of gastric cancer. Oncotarget 5(8), 2318–2329 (2014).

54 Mondal T, Subhash S, Vaid R et al. MEG3 long noncoding RNA regulates the TGF-beta pathway genes through formation of RNA–DNA triplex structures. Nat. Commun. 6, 7743 (2015).

55 Yang L, Lin C, Liu W et al. ncRNA- and Pc2 methylation-dependent gene relocation between nuclear structures mediates gene activation programs. Cell 147(4), 773–788 (2011).

56 Zhang EB, Yin DD, Sun M et al. P53-regulated long non-coding RNA TUG1 affects cell proliferation in human non-small-cell lung cancer, partly through epigenetically regulating HOXB7 expression. Cell Death Dis. 5, e1243 (2014).

57 Zhang Q, Geng PL, Yin P et al. Down-regulation of long non-coding RNA TUG1 inhibits osteosarcoma cell proliferation and promotes apoptosis. Asian Pac. J. Cancer Prev. 14(4), 2311–2315 (2013).

58 White NM, Cabanski CR, Silva-Fisher JM et al. Transcriptome sequencing reveals altered long intergenic non-coding RNAs in lung cancer. Genome Biol. 15(8), 429 (2014).

59 Zhang Y, Zhang XO, Chen T et al. Circular intronic long noncoding RNAs. Mol. Cell 51(6), 792–806 (2013).

60 Memczak S, Jens M, Elefsinioti A et al. Circular RNAs are a large class of animal RNAs with regulatory potency. Nature 495(7441), 333–338 (2013).

http://genome.ucsc.edu/



61 Jeck WR, Sorrentino JA, Wang K et al. Circular RNAs are abundant, conserved, and associated with ALU repeats. RNA 19(2), 141–157 (2013).

62 Hansen TB, Jensen TI, Clausen BH et al. Natural RNA circles function as efficient microRNA sponges. Nature 495(7441), 384–388 (2013).

63 Szabo L, Morey R, Palpant NJ et al. Statistically based splicing detection reveals neural enrichment and tissue-specific induction of circular RNA during human fetal development. Genome Biol. 16, 126 (2015).

64 Hansen TB, Kjems J, Damgaard CK. Circular RNA and miR-7 in cancer. Cancer Res. 73(18), 5609–5612 (2013).

65 Hansen TB, Wiklund ED, Bramsen JB et al. miRNA-dependent gene silencing involving Ago2-mediated cleavage of a circular antisense RNA. EMBO J. 30(21), 4414–4422 (2011).

66 Chou YT, Lin HH, Lien YC et al. EGFR promotes lung tumorigenesis by activating miR-7 through a Ras/ERK/Myc pathway that targets the Ets2 transcriptional repressor ERF. Cancer Res. 70(21), 8822–8831 (2010).

67 Cheng AM, Byrom MW, Shelton J et al. Antisense inhibition of human miRNAs and indications for an involvement of miRNA in cell growth and apoptosis. Nucleic Acids Res. 33(4), 1290–1297 (2005).

68 Guo JU, Agarwal V, Guo H et al. Expanded identification and characterization of mammalian circular RNAs. Genome Biol. 15(7), 409 (2014).

69 Liu YC, Li JR, Sun CH et al. CircNet: a database of circular RNAs derived from transcriptome sequencing data. Nucleic Acids Res. 44(D1), D209–D215 (2016).

70 Del Vescovo V, Grasso M, Barbareschi M et al. MicroRNAs as lung cancer biomarkers. World J. Clin. Oncol. 5(4), 604–620 (2014).

71 Yang Y, Ahn YH, Chen Y et al. ZEB1 sensitizes lung adenocarcinoma to metastasis suppression by PI3K antagonism. J. Clin. Invest. 124(6), 2696–2708 (2014).

72 Glazar P, Papavasileiou P, Rajewsky N. circBase: a database for circular RNAs. RNA 20(11), 1666–1670 (2014).

73 Dudekula DB, Panda AC, Grammatikakis I et al. CircInteractome: a web tool for exploring circular RNAs and their interacting proteins and microRNAs. RNA Biol. 13(1), 34–42 (2015).

74 Ghosal S, Das S, Sen R et al. Circ2Traits: a comprehensive database for circular RNA potentially associated with disease and traits. Front Genet. 4, 283 (2013).

75 Soussi T, Ishioka C, Claustres M et al. Locus-specific mutation databases: pitfalls and good practice based on the p53 experience. Nat. Rev. Cancer 6(1), 83–90 (2006).

76 Gong WJ, Yin JY, Li XP et al. Association of well-characterized lung cancer lncRNA polymorphisms with lung cancer susceptibility and platinum-based chemotherapy response. Tumour Biol. 37(6), 8349–8358 (2016).

77 Kumar V, Westra HJ, Karjalainen J et al. Human disease-associated genetic variation impacts large intergenic non-coding RNA expression. PLoS Genet. 9(1), e1003201 (2013).

78 Popadin K, Gutierrez-Arcelus M, Dermitzakis ET et al. Genetic and epigenetic regulation of human lincRNA gene expression. Am. J. Hum. Genet. 93(6), 1015–1026 (2013).

79 Schmidt LH, Spieker T, Koschmieder S et al. The long noncoding MALAT-1 RNA indicates a poor prognosis in non-small-cell lung cancer and induces migration and tumor growth. J. Thorac. Oncol. 6(12), 1984–1992 (2011).

80 Weber DG, Johnen G, Casjens S et al. Evaluation of long noncoding RNA MALAT1 as a candidate blood-based biomarker for the diagnosis of non-small-cell lung cancer. BMC Res. Notes 6, 518 (2013).

81 Hu X, Bao J, Wang Z et al. The plasma lncRNA acting as fingerprint in non-small-cell lung cancer. Tumour Biol. 37(3), 3497–504 (2015).

82 Zhou M, Guo M, He D et al. A potential signature of eight long non-coding RNAs predicts survival in patients with non-small-cell lung cancer. J. Transl. Med. 13, 231 (2015).

83 Jiao F, Hu H, Han T et al. Long noncoding RNA MALAT-1 enhances stem cell-like phenotypes in pancreatic cancer cells. Int. J. Mol. Sci. 16(4), 6677–6693 (2015).

84 Tung MC, Lin PL, Wang YC et al. Mutant p53 confers chemoresistance in non-small-cell lung cancer by upregulating Nrf2. Oncotarget 6(39), 41692–41705 (2015).

85 Cheng N, Cai W, Ren S et al. Long non-coding RNA UCA1 induces non-T790M acquired resistance to EGFR-TKIs by activating the AKT/mTOR pathway in EGFR-mutant non-small-cell lung cancer. Oncotarget 6(27), 23582–23593 (2015).

86 Barshack I, Lithwick-Yanai G, Afek A et al. MicroRNA expression differentiates between primary lung tumors and metastases to the lung. Pathol. Res. Pract. 206(8), 578–584 (2010).

87 Li P, Chen S, Chen H et al. Using circular RNA as a novel type of biomarker in the screening of gastric cancer. Clin. Chim. Acta 444, 132–136 (2015).

88 Li CH, Chen Y. Targeting long non-coding RNAs in cancers: progress and prospects. Int. J. Biochem. Cell. Biol. 45(8), 1895–1910 (2013).

89 Taucher V, Mangge H, Haybaeck J. Non-coding RNAs in pancreatic cancer: challenges and opportunities for clinical application. Cell. Oncol. (Dordr.). doi:10.1007/s13402-016-0275-7 (2016) (Epub ahead of print).

90 Mamdani H, Induru R, Jalal SI. Novel therapies in small-cell lung cancer. Transl. Lung Cancer Res. 4(5), 533–544 (2015).

91 Smaldone MC, Davies BJ. BC-819, a plasmid comprising the H19 gene regulatory sequences and diphtheria toxin A, for the potential targeted therapy of cancers. Curr. Opin. Mol. Ther. 12(5), 607–616 (2010).

92 Hasenpusch G, Pfeifer C, Aneja MK et al. Aerosolized BC-819 inhibits primary but not secondary lung cancer growth. PLoS ONE 6(6), e20760 (2011).

93 Khalil AM, Guttman M, Huarte M et al. Many human large intergenic noncoding RNAs associate with chromatin-modifying complexes and affect gene expression. Proc. Natl Acad. Sci. USA 106(28), 11667–11672 (2009).


part of

Review


Epigenomics

Review 2016/04/308

5

2016

It is well-established that the DNA methylation landscape of normal cells undergoes a gradual modification with age, termed as ‘epigenetic drift’. Here, we review the current state of knowledge of epigenetic drift and its potential role in cancer etiology. We propose a new terminology to help distinguish the different components of epigenetic drift, with the aim of clarifying the role of the epigenetic clock, mitotic clocks and active changes, which accumulate in response to environmental disease risk factors. We further highlight the growing evidence that epigenetic changes associated with cancer risk factors may play an important causal role in cancer development, and that monitoring these molecular changes in normal cells may offer novel risk prediction and disease prevention strategies.

First draft submitted: 4 January 2016; Accepted for publication: 17 February 2016; Published online: 22 April 2016

Keywords: aging • cancer • cancer risk • DNA methylation • epigenetic • epigenetic clock

DNA methylation, a promising epigenetic cancer biomarkerEpigenetics can be defined as the study of mitotically heritable changes in gene regula-tion and cellular phenotype that cannot be explained by changes in DNA sequence [1]. Among the most important epigenetic modi-fications are those which affect DNA directly, mainly through covalent addition of a methyl (-CH

3) group at cytosines of CG dinucleo-

tides (referred to commonly as ‘CpGs’), although such DNA methylation (DNAm) can also occur in a non-CpG context [2]. Most CpGs in the human genome are methyl-ated, occurring mainly in intergenic regions, repetitive elements and gene bodies. The unmethylated form occurs preferentially in the context of CpG islands (CGIs, regions of particular high CpG density), which colocal-ize with gene promoters [3]. CpG sites located in regions just outside CGIs (termed shores and shelves), or in distal regulatory elements, notably enhancers, exhibit the highest vari-ability in DNAm [4,5]. DNAm can control

gene expression, with promoter DNAm typi-cally associated with gene silencing [3,6]. On the other hand, unmethylated gene promot-ers can associate with either active or inactive (poised) expression states, depending on the levels of nearby histone marks [7].

One of the most remarkable features of the DNAm landscape is that it gets reset during human embryogenesis, subsequently playing an essential role in development and tissue differentiation [6]. Specifically, DNAm in a differentiated cell of a given lineage is thought to play a critical role in irreversibly silencing genes that are not required for spec-ification of that lineage [5]. It further plays a key role in determining enhancer func-tion and transcription factor binding dur-ing development [5]. Once acquired, DNAm constitutes a metastable modification, which is maintained during cell division due to the action of DNA methyltransferase enzymes. However, the fidelity of the DNAm copying machinery is significantly lower than that of its DNA counterpart, which may result in

Epigenetic drift, epigenetic clocks and cancer risk

Shijie C Zheng‡,1,2, Martin Widschwendter3 & Andrew E Teschendorff*,‡,1,3,4

1CAS Key Lab of Computational

Biology, CAS-MPG Partner Institute

for Computational Biology, Shanghai

Institute for Biological Sciences, Chinese

Academy of Sciences, Shanghai 200031,

China 2University of Chinese Academy of

Sciences, 19A Yuquan Road, Beijing

100049, China 3Department of Women’s Cancer,

University College London, 74 Huntley

Street, London, WC1E 6AU, UK 4Statistical Cancer Genomics, UCL Cancer

Institute, University College London,

72 Huntley Street, London, WC1E 6BT, UK


[email protected]‡Authors contributed equally



Review Zheng, Widschwendter & Teschendorff

‘epimutations’ every time a cell divides [1]. The rate of such epimutations has been estimated to be as high as 10-5 per cytosine per cell division (cf. with a mutation rate of bases within CpG nucleotides of ∼10-7 per cell division [8,9]), and may result in either loss of DNAm (‘hypomethylation’) at sites that are normally methyl-ated, or in DNAm gains (‘hypermethylation’) at sites that are usually unmethylated [10].

Importantly, DNAm changes have been seen in a wide range of complex diseases, including cancer [1,11]. Specifically, two cancer hallmarks are hypermethyl-ation of gene promoters, often affecting tumor sup-pressor genes and hypomethylation of intergenic regions [11]. Because of this, and because it constitutes a metastable, directly amplifiable, DNA-based mark (as opposed to, e.g., RNA ‘snapshot’ measurements, which are strongly time-dependent), DNAm offers great potential as a cancer biomarker [12–15]. Importantly, DNAm is also highly malleable, and has been shown to be influenced by many environmental exposures, including diet and levels of in utero nutrients [16–18]. Thus, DNAm represents not only an attractive bio-marker for risk prediction and early detection of com-plex disease, but also offers to improve our understand-ing of the interface between environmental risk factors and disease phenotypes [1,18,19].

Epigenetic driftAlthough age-associated DNAm changes affect-ing individual genes in normal tissue have long been observed [20,21], one of the first studies to explore this phenomenon beyond single genes was a 2005 study by Fraga et al. [22]. This study compared DNAm profiles of a number of monozygotic twins of different ages, and observed that while newborn twins exhibited effectively identical methylomes, adult twins showed divergent patterns with the level of divergence increas-ing with age. Due to the nonlongitudinal nature of this study, as well as limitations in sample size and genomic coverage, no apparent pattern of DNAm divergence was observed, with the loci exhibiting divergence within a twin pair not overlapping with the corre-sponding loci defined by another twin pair. Thus, the authors referred to the observed divergence in DNAm within twin pairs with the term ‘epigenetic drift’, to highlight the apparent stochastic nature of age-asso-ciated DNAm changes. However, in what follows we shall use a more general definition of epigenetic drift to encompass any type of age-associated DNAm change, be it of a stochastic nature or not.

Larger and higher genome-coverage studies subse-quently confirmed that the DNAm landscape of nor-mal cells changes substantially with age [23–27]. Several important novel insights were obtained from these

studies. First, age-associated DNAm alterations do not happen randomly across the genome. For instance, it was observed that age-associated hypermethylation is more likely to happen at sites that carry bivalent [26] or PRC2 repressive marks [25], as defined in human embryonic or adult stem cells. By contrast, age-associated hypomethylation appears to target strong enhancers and active promoters [28]. Second, specific age-associated DNAm changes occur independently of tissue and cell-type, and this seems to be particularly true for age-associated hypermethylation, and specifi-cally for the PRC2-enriched component [25,28–29]. For instance, an age-associated PRC2-marked 69 CpG DNAm signature derived in blood was shown to cor-relate with chronological age in other normal tissue types (e.g., lung and ovarian tissue) [25], and another age-related module enriched for PRC2 members was found to co-vary with age in brain and blood [30]. This cross-tissue independence not only demonstrates that a component of drift reflects an underlying universal mechanism, but also that drift in a complex tissue is not entirely the result of underlying alterations in cell type proportions. Indeed, this has been shown explicitly, as similar age-associated DNAm alterations are observed in different subsets of purified blood cells [26]. Third, age-associated DNAm alterations are also seen in adult stem cell populations, notably mesenchymal stem cells and hematopoietic stem/progenitor cells [25,31–33]. This supports the view that a component of epigenetic drift accrues in the underlying stem cell population of a given tissue, giving rise to the corresponding observed changes in the differentiated cells that make up the bulk of the tissue.

Horvath’s epigenetic clockThe consistency and robustness of age-associated DNAm alterations across different tissue and cell types led to a number of studies to attempt predict the chronological age of an individual [34–37]. While some of these DNAm-based age predictors have been derived in specific tissues [34,35], Horvath derived a multitissue age predictor, which he then validated in a large number of independent datasets, encompass-ing in total over 8000 samples from over 50 different tissue and cell types [36]. This multitissue age predic-tor consists of 353 CpGs, and achieved a remarkable, clock-like accuracy on independent data with a median absolute deviation error of less than ±5 years. Although no direct comparison with other biological assays has yet been performed, it would appear that Horvath’s ‘epigenetic clock’ may achieve substantially higher accuracies than those based on measuring telomere length or other molecular features such as T-cell DNA rearrangements [38–42]. Moreover, although studies


Epigenetic drift, epigenetic clocks and cancer risk Review

have found age-associated copy number [43,44], muta-tional [45] and gene expression [46] signatures, none of these have yet been developed into accurate age pre-dictors, suggesting that these other types of molecular profiles may not be as relevant as DNAm for predicting chronological age.

DNAm-based age predictors containing far fewer than 353 marker CpGs have also been reported [34,35,37,42]. For instance, one study showed that DNAm values at only three CpG sites can accurately predict chronologi-cal age [42]. Another study indicated that 74% of the variation in chronological age can be predicted with as few as two CpG loci [35]. However, these age pre-dictors have only been tested in specific tissues, or not extensively tested in other tissue types [36,47]. Another DNAm-based age predictor is the one developed by Hannum et al. [34], which trained its predictor on one of the largest whole blood datasets encompassing over 650 samples. While this 71 CpG age predictor also achieved high accuracy in independent blood datasets, recalibration of the predictor was necessary to achieve comparable accuracies in other tissue types [34].

There are a number of reasons why age predictors like that of Hannum et al. may not achieve accura-cies comparable to those of Horvath’s clock. First of all, none of the age predictors developed so far made explicit adjustment for age-associated changes in cell-type composition [48,49]. Since age-associated changes in tissue composition will vary from one tissue type to another, deriving an age predictor from one tissue type only, without correcting for changes in cell-type composition, will certainly ‘bias’ the predictor toward the tissue of origin. Because Horvath’s epigenetic clock was trained on data from over 30 different tissue and cell types, this epigenetic clock is unlikely to have been confounded by tissue-specific age-associated changes in tissue composition. Second, an age predictor like that of Hannum et al. is likely to capture age-cumula-tive effects of specific endogenous and environmental factors, which are specific to blood-tissue and there-fore not generalizable to other tissue types. In contrast, Horvath’s clock, having been derived across so many different tissue types, is unlikely to have been con-founded by these other tissue-specific effects. Third, tissue-specific DNAm levels could confound age pre-dictors derived from one tissue type, when assessed in other tissue types, therefore, requiring recalibra-tion [34]. A final reason could be that highly accurate quantification of chronological age from DNAm pro-files may require a substantial number of CpGs, as is the case with Horvath’s clock [36]. This would seem necessary if there is a substantial element of stochastic-ity underlying epigenetic drift [22]. Indeed, although it is clear that some genomic loci are more likely to

undergo epigenetic drift than others, a more realistic picture is that of each CpG in the genome carrying an intrinsic probability of acquiring age-associated DNAm changes. Thus, the robustness of Horvath’s clock stems in part from it measuring an aggregate level of absolute deviation in DNAm over a relatively large and specific set of 353 CpGs. In fact, the clock’s accuracy only requires that a significant number of the 353 CpGs exhibit the expected DNAm deviations in a given sample, in order for the average deviation to then represent a meaningful number. Comparing two separate samples (e.g., different tissues from the same individual, or samples from identically aged individu-als, e.g., twins), the specific subset of 353 CpGs that are altered in each sample may differ substantially. Thus, a highly accurate molecular clock is possible despite a level of underlying stochasticity, provided the clock is defined over a sufficient number of loci. Nev-ertheless, future high-coverage whole-genome bisulfite sequencing studies may pinpoint a few ‘nonstochastic’ loci, which, not unlike ‘lighthouse beacons’, keep track of chronological age with an accuracy comparable, or even exceeding that of Horvath’s clock.

Significance & interpretation of the epigenetic clockAlthough Horvath’s epigenetic clock appears to pro-vide, on average, a highly accurate measure of chrono-logical age in seemingly healthy tissues, it is clear that for specific samples, large deviations/errors are also observed. This has led to the proposal that the devia-tion between DNAm age (‘DNAm-age’), that is, the value predicted by the epigenetic clock, and chrono-logical age may be informative of the true ‘biological’ age of a tissue [36]. Thus, the biological age of a tis-sue in an individual is not only a function of the per-son’s chronological age, but also a function of other additional endogenous and exogenous factors, some of which may cause age acceleration, whereas others may cause age deceleration (see e.g., [19]).

A number of studies have explored whether biologi-cal age, specifically the difference between DNAm-age and chronological age, appears aggravated in tis-sues associated with disease phenotypes [50–54]. For instance, increased ‘age acceleration’, that is, tissues from individuals where DNAm-age is higher than the chronological age, has been observed in the human liver of obese individuals [50], in HIV-1-infected individu-als [52] and in Down syndrome patients [53]. DNAm age in blood has also been shown to correlate with physical and cognitive fitness [54]. The significance of DNAm-age in the context of epithelial cancer is, however, less evident, since age acceleration is not observed across all cancer types [47,55]. The interpretability of DNAm-age



in affected tissues is also complicated due to potential confounding effects of the disease itself.

Other studies have, therefore, explored the possibil-ity that Horvath’s DNAm-age may be predictive of future disease risk. For instance, DNAm-age in blood has been found to be higher in men compared with women, a result which is consistent, in principle, with men’s average lower longevity [34,36]. DNAm-age in the blood of postmenopausal women has been shown to correlate with the prospective risk of lung cancer [56]. A recent study also found DNAm-age in blood to be predictive of all-cause mortality [57]. Although not explicitly using DNAm-age, another study showed how DNAm of CpGs defining an age-associated DNAm signature in blood [25] always exhibited hyper-variability in cervical normal cells, which 3 years later progressed to a high-grade cervical intraepithelial neo-plasia [58]. Although all these results support the notion that DNAm-age could indicate disease risk for a num-ber of different diseases, the reproducibility of these findings in independent cohorts still needs to be dem-onstrated, specially in those studies where progression of the DNAm changes could not be assessed.

A pressing unanswered question is the biological mechanism(s) underpinning the epigenetic clock. An initial attractive hypothesis would be that it constitutes a ‘mitotic clock’, measuring the number of cell divi-sions incurred by long-lived stem cells [59]. Under this model, incomplete maintenance of DNAm patterns by DNA methyltransferase enzymes (e.g. DNMT1) during DNA replication would lead to epimutations. This interpretation, however, cannot explain the abil-ity of the clock to accurately predict chronological age across tissue types that differ so widely in their over-all proliferation and turnover rates, including highly proliferative tissues such as colon and nonproliferative ones, such as brain [36]. Thus, while it is entirely plau-sible that components of other age-associated DNAm signatures may be mitotic in nature, as suggested by Beerman and Rossi [32] and Issa [59], the same does not appear to hold for Horvath’s epigenetic clock. Instead, the epigenetic clock may reflect the indirect effects of the work performed by an epigenetic maintenance system, although at present, it is unclear what this epigenetic maintenance system may actually be [36].

Epigenetic drift & cancer risk factor DNAm signaturesAmong risk factors for cancer, age is special, not only because it is the main risk factor for most cancer types, but because it indirectly captures the effects of age-associative cumulative exposure to exogenous and endogeneous risk factors. Thus, epigenetic drift may reflect molecular alterations caused by genetic and

environmental risk factors. This in turn implies that if one wishes to study the effect a cancer risk factor may have on the DNA methylome, that adjustment for age is paramount. Over the last few years, many stud-ies have explored the impact of major cancer risk fac-tors on the DNA methylome of normal cells (Table 1). These studies include the effect of: HPV infection in normal cervical smears [58], smoking in blood and buc-cal tissue [60–65], BRCA1 mutation in blood [66], sun-light (UV) exposure in skin [67,68], obesity in blood and adipose tissue [69–71], inflammation (inflammatory bowel disease) in colon tissue [72–74], alcohol intake in blood [75] and asbestos exposure in blood (Table 1) [23]. For other more specific chemical exposures, see [16].

Interestingly, if one focuses on the hypermethyl-ated components of these cancer risk-factor DNAm signatures, one observes that PRC2/bivalently marked sites are often significantly enriched (Table 1). This is the case for age [25–27], HPV infection [58], smok-ing [62], obesity [69], BRCA1 mutation [66] and inflam-mation [73]. The biological significance of this com-mon PRC2 enrichment is, however, unclear. First of all, most of these enrichments have been established in relation to PRC2/bivalent marks, as determined in the human embryonic stem cell ground state, which is clearly not the most relevant one. Nevertheless, results have been shown to carry over to the corresponding PRC2/bivalent marks obtained in relevant adult stem cell populations, as for instance in the case of CD133+ hematopoietic progenitor cells [25]. Second, most of the PRC2 targets represent transcription factors that are normally not expressed in the tissue of interest [81]. However, this does not exclude the possibility that a few key tissue-specific transcription factors are silenced through promoter hypermethylation. Therefore, it is entirely plausible that age-associated cumulative DNAm changes at PRC2 targets, which often encode key developmental and tissue-specific transcription factors, may result in deregulation of normal homeo-stasis, which is a key cancer hallmark [82–84].

A common cancer risk factor epigenetic signature?The common enrichment for PRC2 sites among dif-ferent cancer risk DNAm signatures also suggests that similar sites may be affected, irrespective of the risk fac-tor. Although a comprehensive analysis of the overlap of such risk-factor DNAm signatures is still lacking, there are already some hints that signatures predictive for one risk factor may also be predictive for another. For instance, one study showed how an age-associated DNAm signature, involving PRC2 marked CpG sites that become hypermethylated with age, could discrim-inate cervical neoplasias from age-matched normal



samples, suggesting that specific loci which undergo age-related DNAm changes do so also in response to HPV infection (the major risk factor for cervical cancer development) [25]. A more recent pan-cancer study compared a smoking-associated DNAm signa-ture derived in buccal (epithelial) tissue with DNAm changes in cancer, and found that a DNAm-based smoking index (SMKI) constructed from this buccal DNAm signature, was highest in smoking-related lung cancers, but, surprisingly, also higher in every single cancer type compared with its respective normal tis-sue, including cancers which are not smoking-associ-ated (e.g., endometrial cancer) [62]. That the SMKI is highest for smoking-associated lung cancer strongly supports the view that the smoking DNAm signature captures effects specific to the smoke carcinogens. On the other hand, that the SMKI is higher in every single cancer type compared with its corresponding normal tissue also suggests that a significant component of the smoking DNAm signature captures effects which are not specific to smoking. These additional nonspecific effects, therefore, suggest that other cancer risk factors

(e.g., a high estrogen to progesterone ratio associated with obesity in the case of endometrial cancer [85]) may actively cause DNAm changes in normal tissue, which are similar to those induced by smoking. However, without an improved understanding of the biological mechanisms by which cancer risk factors may actively cause DNAm changes in normal cells, the existence of a common ‘causal’ cancer risk factor DNAm signature remains speculative.

A much more likely explanation as to why DNAm signatures associated with different cancer risk fac-tors may overlap, or indeed why the smoking buccal DNAm signature reported in [62] is aggravated in all cancer types, is that these signatures contain a com-mon ‘mitotic clock’ component (Figure 1), which would appear accelerated both in normal cells exposed to inflammation (e.g., buccal cells exposed to smoke carcinogens), as well as in highly proliferative cancer cells [59]. Interestingly, this ‘mitotic clock’ component also seems to be particularly well-defined at PRC2 sites, which are usually unmethylated in normal cells, but which would acquire stochastic hypermethylation

Table 1. Table lists major cancer risk factors for which epigenome-wide association studies have been conducted in a number of tissue types.

Cancer risk factor Normal tissue Platform PRC2/bivalent enrichment?

Ref.

Age Blood (WB + purified) 27k and 450k Y [25,26]

Colon 27k Y [76]

Adipose 450k Y [69]

Brain 27k Y [28,30,77]

Kidney 27k Y [28]

Muscle 27k Y [28]

Buccal (saliva) 27k Y [29]

HPV Cervix 27k Y [58]

BRCA1 mutation Blood (WB) 27k Y [66]

Smoking Blood (WB) 450k Y [60]

Buccal 450k Y [62]

Obesity/BMI Blood CHARM and 450k N [70,71]

Adipose 450k Y [69]

Alcohol Blood (PBMCs) 27k N [75,78]

UV light Skin 27k and 450k N [67,68]

EBV Blood (B cell) WGBS Y [79]

IBD Intestine/colon CGI agilent Y [80]†

We also indicate if a DNA methylation signature for the cancer risk factor was enriched for PRC2 or bivalently marked sites as determined in human embryonic stem cells. This enrichment is in the hypermethylated part of the signature since these PRC2/bivalent sites are normally unmethylated in the control samples. We also list some of the references where enrichment for PRC2/bivalent sites was Y, or those where findings were N.†In mice.450k/27k: Illumina Human Methylation 450k/27k beadchip; BMI: Body mass index; CGI: CpG island, EBV: Epstein–Barr virus; HPV: Human papilloma virus; IBD: Inflammatory bowl disease; N: Negative; PBMC: Peripheral blood mononuclear cells; WB: Whole blood; WGBS: Whole-genome bisulfite sequencing; UV: Ultraviolet, Y: Observed.

710 Epigenomics (2016) 8(5)

Figure 1. The three major components of epigenetic drift. We define epigenetic (DNAm) drift as any type of age-associated DNAm change. This epigenetic drift has a cell-intrinsic and tissue-type independent component, termed ‘Horvath’s epigenetic clock’, which predicts chronological age with a remarkably high degree of accuracy. Another cell-intrinsic but tissue-type dependent component of DNAm drift is representing an ‘epigenetic mitotic clock’, which measures the cumulative number of cell divisions that the stem cell population of the tissue has undergone. The tick rate of this epigenetic mitotic clock may be influenced by endogenous (e.g., genetic risk factors) and exogenous (e.g., environmental risk factors) factors. These same cell-extrinsic factors may cause other types of active DNAm alterations, for instance, as seen for the AHRR gene in response to smoking or those that may be mediated by hormonal factors or viral infections. DNAm: DNA methylation.

CGCG

CG

CG

CG

CG

CG

CG CG CG

CG CG

CGCG

CG

CG

CG

CG

CG

CG CG CG

CG CG

The three components of epigenetic drift

Chronological age

Horvath’s epigeneticclock(cell intrinsic andtissue independent)

1

23Epigenetic mitotic clocks(cell intrinsic andtissue specific)

Age-cumulativeexposures(cell extrinsic andtissue specific)

Cancerrisk

DNAmethylation

Epigenetic ‘drift’



in a more proliferative state (Figure 2) [25,32,59,82]. This PRC2 enriched hypermethylation signature is also the one which is seen in aged stem cells and during hema-topoietic ontogeny, further supporting a mitotic clock interpretation (Figure 2) [31–32,86–88].

DNAm-based predictors of cancer riskSeveral studies have explored the possibility of using DNAm signatures in easily accessible nonepithelial tis-sues such as blood to predict the prospective risk of epithelial cancer. For instance, a number of studies have shown that the prospective risk of breast cancer can be predicted from whole blood DNAm profiles, yet the resulting AUCs or odd ratios are low (AUCs typically between 0.5 and 0.65), and therefore, only of marginal significance [66,89–91]. As mentioned ear-lier, DNAm-age in blood has also been shown to be predictive of lung cancer with a cancer incidence haz-

ard ratio of 1.5 (p = 0.003) [56]. Focusing on a small set of inflammatory genes, which included IL6, IFN, TLR2 and ICAM1, another recent study has shown how DNAm of these genes in blood could predict the prospective risk of prostate cancer with an incidence hazard ratio of approximately 1.5 [92]. Using four lon-gitudinal cohorts, it has also been shown that DNAm-age in blood predicts all-cause mortality in later life, with age accelerations of 5 years or higher carrying a 16% increased mortality risk [57]. However, the influ-ence of cancer-related mortality relative to mortality caused by other diseases in the context of DNAm-age needs to be more carefully assessed.

Other studies have shown how DNAm signatures derived in epithelial cells can predict the risk of neopla-sia or invasive cancer. For instance, one study showed how DNAm variability in cytologically normal cervical smears, collected 3 years in advance of diagnosis, could

www.futuremedicine.com 711

Figure 2. Epigenetic stem cell model of oncogenesis. Figure depicts how age-associated DNA methylation changes, which preferentially target genes marked by the PRC2 complex, accrue in underlying stem cells as a function of both age and exposure to cancer risk factors, leading at first to intrasample epigenetic nonclonal heterogeneity. Subsequent additional epigenetic and genetic changes can then give rise to epigenetic clonal mosaicism, from which subsequently a cancer clone can arise. The rate at which DNAm changes accrue at PRC2 targets will be determined by an epigenetic mitotic clock, with the rate influenced by endogenous and exogenous risk factors (see Figure 1).

A common epigenetic cancer risk signature?

Normal stem cellReversible PRC2 targetgene repression

Age+

Cancer risk factors

Stem cell with stochasticPRC2 target gene methylation

Age+

cancer riskfactors

Intrasampleepigeneticmosaicism

+mutations

Differentiation

Differentiated cellDNAmethylation‘Fixed’ stem cell with

PRC2 target gene methylationand irreversible suppression

Age+

cancer riskfactors

Mutations+

dominantclonal

expansion

Cancer cellSilenced PCGT genes

Intra-sampleepigeneticdiversity

CG

CG

CG

CG

CG

CG

CG

CG

CG

CG

CG

CG

CG

CG

CG

CG

CG

CG

CH3 CH3 CH3

EED

EZH2SUZ12

EED

EZH2SUZ12

H3-K27

PCG

CG

CG

CG

CG

CG

CG

CG

CG

CG

CH3 CH3 CH3

EED

EZH2SUZ12

EED

EZH2SUZ12

H3-K27

PCGDNMT

CG

CG

CG

CG

CG

CG

CG

CG

CG

CH3 CH3 CH3

EED

EZH2SUZ12

EED

EZH2SUZ12

H3-K27

PCGDNMT

DNAmethylation

CG

CG

CG

CG

CG

CG

CG

CG

CG

CH3 CH3 CH3

CO-REP

HDACMBD

H3-K27

DNMT



predict the prospective risk of a cervical intraepithelial neoplasia of grade 2 or higher, with an AUC of approxi-mately 0.66 (p < 0.05) [58]. Another study showed how a smoking DNAm signature, as derived in buccal cells,

and assessed in lung carcinomas in situ could predict the risk of progression to invasive lung cancer with an AUC of 0.88 [62]. While all of these results offer the exciting prospect of using DNAm in easily accessible



tissues to predict cancer, larger studies will be needed to assess the potential for clinical application.

Epigenetic drift: is it functional & causal?DNAm changes that correlate with age and other cancer risk factors and which can predict prospective cancer risk offer great biomarker potential. However, whether these associations are purely correlative or causative is still unclear. In the context of aging, age-associated DNAm changes may contribute to the well-known age-associated decline of stem cell function, and may underlie phenomena such as the well-known myeloid skewing of an aged hematopoietic system [93,94] and immunosenescence [95]. However, with the excep-tion of a few genes, it has been extremely difficult to pinpoint age-associated alterations in gene expression or gene function, which are due to corresponding alterations at the DNAm level. Indeed, although there have been some reports of correlations between DNAm and mRNA expression in blood [34,96], the studies were either small [96] or did not use matched samples or did not correct for changes in blood cell-type com-position [34]. Using two large but unmatched blood DNAm and mRNA expression datasets, and correcting for cellular heterogeneity, a recent study showed that one reason why there might not be a strong correla-tion is because age-associated DNAm changes appear to act by stabilizing pre-existing ‘baseline’ expression levels [97]. Thus, age-associated hypermethylation in blood is preferentially observed in promoters of genes that are normally not expressed in blood, and vice-versa, hypomethylation is observed in promoters of genes that are normally expressed [97]. Thus, if this result was to generalize to other tissue types, this suggests that most of the epigenetic drift is probably not functional. Given that epigenetic drift might already begin during embryogenesis [17,98] and that the rate of drift appears to be maximal before sexual maturity [99,100], one could speculate that most of the drift ought to be passive, as otherwise, it would bring forward the onset of complex diseases to coincide with the reproductive period, which would be highly undesirable and probably not selected for on evolutionary grounds [19]. Consistent with this, epigenetic drift also appears to target mostly peripheral nodes in protein interaction networks, avoiding essen-tial and housekeeping genes, and specifically targeting transcription factors, most of which are not expressed in any given tissue type [101].

Thus, it is tantalizing to speculate that epigenetic drift generally does not affect gene function, but that it may occasionally ‘hit’ a key transcription factor, for instance, one that is critical for maintaining healthy homeostasis of a given tissue type, thus increasing cancer risk [94,102]. Recent work in the hematopoietic

system supports this model, where DNAm alterations that are seen to accrue with age in blood and which may affect key lineage-specifying transcription factors, appear aggravated in myelodysplastic syndrome, with further increases in DNAm observed in acute myeloid leukemias [94]. Genetic mutations in key epigenetic regulators, which are seen as a function of age in pre-leukemic clonal expansions, and which can modify the normal DNAm landscape, provide further indirect support for such a model [103,104]. The WNT-signaling pathway, a key stem cell pathway, which is observed to undergo epigenetic deregulation with age [29,105–106], with DNAm-induced silencing of WNT-signaling antagonists potentially tipping the homeostatic bal-ance toward increased self-renewal [106], provides further evidence for how drift could affect normal homeostasis.

Another more concrete example of how drift could increase cancer risk, and which may serve as a gen-eral paradigm for several other cancer types, is the one provided by HAND2 in endometrial cancer [107]. Although the promoter of HAND2 has not yet been conclusively shown to undergo hypermethylation with age in endometrial tissue, it does undergo hyper-methylation with age in both blood [94,97] and in the colon of mice [24], suggesting that hypermethylation of its promoter with age is a tissue-wide phenomenon. This is significant, not only because age is a main risk factor for endometrial cancer, but because HAND2 plays a critical role in mediating the tumor suppres-sive effects of progesterone, the main tumor suppres-sor pathway in this cancer type. In fact, a high body mass index, which is usually associated with a high estrogen to progesterone ratio, is the other main risk factor for endometrial cancer, and HAND2 antago-nizes the oncogenic activity of estrogen [107]. Given that promoter hypermethylation-induced silencing of HAND2 is observed in normal tissue adjacent to com-plex atypical hyperplasias (CAH), in CAH itself, and in endometrial cancer [107], this supports the view that HAND2 methylation is an early progressive event in endometrial carcinogenesis. Moreover, HAND2 dou-ble knock-out mice develop CAH within weeks [107], suggesting that HAND2 inactivation is a causal driver of endometrial cancer development. Thus, although HAND2 promoter hypermethylation with age may be observed in many different tissue types, it is only the silencing in endometrial tissue that would be of par-ticular functional consequence, increasing the risk of neoplastic transformation in that tissue.

Conclusion & future perspectiveA deeper understanding of epigenetic drift, defined here as any age-associated DNAm alteration, will



require an improved characterization of its underlying components. As proposed here, epigenetic drift has at least three major components (Figure 1): a cell-intrinsic and tissue-type independent component, which allows highly accurate quantification of chronological age, as exemplified by Horvath’s epigenetic clock; another cell-intrinsic, mitotic component whose tick rate will be tissue-dependent and influenced by endogenous (e.g., genetic risk factors) and exogenous (e.g., endo-crine factors, inflammation) factors; and a cell-extrin-sic, nonmitotic, active component, which will largely depend on tissue-type, and which also captures the age-cumulative effects of endogenous and exogenous risk factors.

Future studies will need to address a number of key questions to help shed further light on the nature of epigenetic drift and its components. Focusing on the epigenetic clock, it will be interesting to determine if other equally accurate clocks can be constructed and whether there is a minimum number of CpGs that are required to achieve such accuracy. Given that Hor-vath’s clock is largely based on Illumina 27k probes, it would be interesting to see if more accurate clocks can be constructed based on the newer 450k/EPIC and eventually also whole genome bisulfite sequencing technologies. The degree of stochasticity in the epigen-etic clock also needs to be assessed, for instance, by comparison of DNAm patterns at the 353 clock sites between different tissues (e.g., buccal and blood) from the same individual. Using longitudinal studies with multiple time points may also shed light on the tempo-ral nature of DNAm changes at these specific sites [108]. The underlying biological mechanism underpinning Horvath’s clock is another outstanding question, spe-cially given that it does not seem to represent a mitotic clock. In particular, it will be of interest to explore if the 353 CpG sites making up Horvath’s clock are located in special chromatin states, which are largely independent of tissue type.

Another key task for the future is the dissection of age-associated DNAm changes into those that are intrinsic to the aging process itself and those which reflect a cumulative exposure to environmental and lifestyle factors. To do this on human cohorts is an entirely nontrivial proposition, since properly adjusting for environmental and lifestyle factors is hard, specially given that we are still unaware of all factors that may impact on the epigenome and how this may vary across tissue types. Birth cohorts could help adjust for chron-ological age and thus help identify DNAm alterations, which are specific to environmental exposures [109]. Longitudinal studies profiling multiple tissues at mul-tiple time points will be illuminating, in particular those based on twins [110]. An alternative approach to

help identify DNAm alterations, which are intrinsic to aging, would be to perform studies on isogenic mice, where all mice are treated uniformly and kept under identical environmental conditions. It would be inter-esting to see if and how such changes vary according to tissue type, even within the same mice, and whether an analogous epigenetic clock for mice can be found. Sim-ilarly, the effect of a controlled environmental expo-sure on an isogenic mouse population could be stud-ied to determine which factors accelerate or decelerate DNAm-age [111].

Of particular importance for cancer, is the com-ponent of epigenetic drift, which represents tissue-specific mitotic clocks, measuring the number of stem cell divisions in the tissue. As shown by Tomasetti and Vogelstein in the context of genetic mutations, such a mitotic clock may serve to predict the life-time risk of a given tissue-type turning cancerous [112,113]. Such a mitotic clock would accelerate in response to specific cancer risk factors, inflammation and endocrine fac-tors being a few clear candidates [73–74,80,114–115], and thus may help explain the interindividual variation in cancer risk of a given tissue-type [15,59]. As argued ear-lier, PRC2 promoter sites may be particularly prone to acquisition of methylation marks during DNA repli-cation, and consistent with this, hypermethylation of PRC2 sites appears to represent a universal DNAm signature of aging, preneoplastic lesions and cancer (Figure 2) [25]. Although an explicit link between an epigenetic mitotic clock and cancer risk still needs to be demonstrated, we have already seen that spe-cific age and nonage related PRC2-enriched DNAm signatures in relevant epithelial cell types can predict the risk of certain cancers, including that of the cer-vix and lung. Therefore, we here propose that mitotic PRC2-enriched DNAm clocks, which correlate with the level of exposure to a generic cancer risk factor such as inflammation [73,74], may be particularly use-ful in the context of risk prediction or early detection (Figure 1 & 2). It follows from this that Horvath’s epi-genetic clock, which is not a mitotic clock, may not be that relevant for predicting the risk of a disease like cancer, which is universally characterized by an increased cell proliferation rate. Indeed, a highly opti-mized multitissue age predictor like Horvath’s clock, which was trained over many tissue types with highly different mitotic rates, cannot be that informative of cancer risk, because it would not be able to capture the mitotic effects (e.g., inflammation, hormonal factors) that promote neoplastic transformation (Figure 1).

Apart from the silencing of key tissue-specific tran-scription factors, another related mechanism, which could link a mitotic PRC2-enriched DNAm clock to cancer risk, is through an increase in intrasample epi-



genetic mosaicism (Figure 2). Increased intrasample stochastic epigenetic variation could provide the very early seeds for carcinogenesis, facilitating the emer-gence of clonal expansions, which eventually lead to neoplastic transformation and cancer [73,116–121] (Figure 2). Indeed, recent studies have already shown that epigenetic clonal heterogeneity may play a key role in increasing the risk of neoplastic transformation [122] and in determining clinical outcome [123,124]. Thus, measures of intrasample clonal epigenetic heterogene-ity may represent excellent general cancer biomarkers for early detection or risk prediction. It will, therefore, also be important to assess the relative level of stochas-tic versus nonstochastic epigenetic variation, which results from mitotic clocks operating at the stem cell level [10].

In relation to studies reporting DNAm signatures that predict the prospective risk of an epithelial cancer (including those which used the epigenetic clock), it is important to note that in most cases the predictions were obtained in a cell type (e.g., blood), which does not serve as the cell of origin for the cancer. Therefore, there is an urgent need to provide a mechanistic basis for these associations. One possibility is that DNAm changes in blood may represent an immune system defect, which could predispose individuals to the development of epithelial tumors like lung cancer [125]. Another possibility is that subtle alterations in blood tissue composition could be signaling an early response to the presence of preneoplastic cells in epithelial tis-sues. Such shifts in blood composition are certainly seen in patients with epithelial cancers [86,126], but how early in carcinogenesis such shifts might be detect-able is currently unclear. Curiously, most cancer-risk predictive DNAm signatures in blood have not been consistently linked to immune-system related path-ways [56,66,89–90], suggesting that there may be another basis for the observed associations. One appealing and exciting possibility is that DNAm changes asso-ciated with an exposure could be similar in different normal tissue types. That this may be so is supported by a study demonstrating that smoking induces simi-lar DNAm alterations in buccal and blood tissue, although, as expected, the effect of smoking is far stronger in the cells directly exposed to the carcinogen (i.e., the buccals) [62]. For instance, the AHRR gene is similarly affected in both blood and buccal tissue [62], supporting the view that the observed hypomethyl-ation at this gene locus represents an active response to the smoke carcinogen, a response which would be common across affected tissues. Likewise, age-associ-ated DNAm signatures are generally valid across many different tissue types. HPV infection also seems to be associated with very similar DNAm changes in cervi-

cal and head and neck cancer [127]. It will, therefore, be extremely important to determine how DNAm changes associated with exposures vary according to tissue type. It will be equally important to assess which endogenous (e.g., genetic risk factors) and exogenous (exposure-related) factors cause similar and dissimilar DNAm changes in any given normal tissue type. For a given cancer type, these questions are key in order to then decide which easily accessible normal tissue might be suitable (if any) as a surrogate for developing risk prediction or early detection biomarkers.

The functional significance of epigenetic drift also needs further in-depth study. The lack of wide-spread in-cis associations between age-associated DNAm and mRNA expression changes does not mean that there might not be a more intricate in-trans associa-tion. Indeed, besides PRC2 members, binding sites of other key transcription factors like CTCF or those of the repressor NRSF/REST, become preferentially hypermethylated with age [97], strongly suggesting that specific regulatory networks, which support a certain 3D chromatin architecture, may become disrupted with age [128]. Thus, it will be specially interesting to investigate the patterns of epigenetic drift in relation to distal regulatory elements, including enhancers, as this may shed further light on how drift may impact on homeostasis in aged tissue, or to investigate the pat-terns of epigenetic drift in a multilayer setting, which also includes all major histone marks [129].

Finally, it will be important to see if there are other examples like HAND2 in endometrial cancer. As we have seen, this gene undergoes age-associated hyper-methylation in normal tissue, and inactivation appears to be also a causal driver of early endometrial cancer development. The example of HAND2 is particularly enlightening, because HAND2 methylation in endo-metrial tissue links together the two main epidemi-ological risk factors for endometrial cancer: age and obesity. We propose that this example may also serve as a more general paradigm linking age-associated DNAm-induced alterations in transcription factor activity to a modulation of the response to an exog-enous cancer risk factor (in this case high estrogen lev-els), and therefore, to an increased cancer risk.

Financial & competing interests disclosureThe authors have no relevant affiliations or financial involve-

ment with any organization or entity with a financial inter-

est in or financial conflict with the subject matter or mate-

rials discussed in the manuscript. This includes employment,

consultancies, honoraria, stock ownership or options, expert

testimony, grants or patents received or pending, or royalties.


manuscript.



Executive summary

• DNA methylation (DNAm) drift in normal tissue reflects cell intrinsic and extrinsic mechanisms, some of which are tissue-independent, whereas others are tissue specific.

• The epigenetic clock describes cell-intrinsic age-associated DNAm alterations, which are tissue and cell-type independent, and which allows highly accurate prediction of the chronological age.

• Most of the epigenetic drift is nonfunctional, yet some of the drift may eventually affect the expression or binding affinity of transcription factors that are required for normal homeostasis.

• Epigenetic PRC2-enriched mitotic clock(s), which measure the cumulative rate of stem cell divisions in a tissue, and whose clock-rate may be affected by endogenous and exogenous factors, are of likely relevance for cancer-risk prediction.

• DNAm signatures associated with cancer risk factors and derived in easily accessible tissues such as blood and buccal tissue have been correlated with the prospective risk of epithelial neoplasia and invasive cancers.

ReferencesPapers of special note have been highlighted as: • of interest; •• of considerable interest

1 Petronis A. Epigenetics as a unifying principle in the aetiology of complex traits and diseases. Nature 465(7299), 721–727 (2010).

2 Lister R, Pelizzola M, Dowen RH et al. Human DNA methylomes at base resolution show widespread epigenomic differences. Nature 462(7271), 315–322 (2009).

3 Deaton AM, Bird A. CpG islands and the regulation of transcription. Genes Dev. 25(10), 1010–1022 (2011).

4 Irizarry RA, Ladd-Acosta C, Wen B et al. The human colon cancer methylome shows similar hypo- and hypermethylation at conserved tissue-specific CpG island shores. Nat. Genet. 41(2), 178–186 (2009).

5 Ziller MJ, Gu HC, Muller F et al. Charting a dynamic DNA methylation landscape of the human genome. Nature 500(7463), 477–481 (2013).

6 Reik W. Stability and flexibility of epigenetic gene regulation in mammalian development. Nature 447(7143), 425–432 (2007).

7 Ernst J, Kellis M. ChromHMM: automating chromatin-state discovery and characterization. Nat. Methods 9(3), 215–216 (2012).

8 Campbell CD, Chong JX, Malig M et al. Estimating the human mutation rate using autozygosity in a founder population. Nat. Genet. 44(11), 1277–1281 (2012).

9 Horsthemke B. Epimutations in human disease. Curr. Top. Microbiol. Immunol. 310 45–59 (2006).

10 Landan G, Cohen NM, Mukamel Z et al. Epigenetic polymorphism and the stochastic formation of differentially methylated regions in normal and cancerous tissues. Nat. Genet. 44(11), 1207–1214 (2012).

11 Feinberg AP, Ohlsson R, Henikoff S. The epigenetic progenitor origin of human cancer. Nat. Rev. Genet. 7(1), 21–33 (2006).

12 Beck S, Rakyan VK. The methylome: approaches for global DNA methylation profiling. Trends Genet. 24(5), 231–237 (2008).

13 Beck S. Taking the measure of the methylome. Nat. Biotechnol. 28(10), 1026–1028 (2010).

14 Rakyan VK, Down TA, Balding DJ, Beck S. Epigenome-wide association studies for common human diseases. Nat. Rev. Genet. 12(8), 529–541 (2011).

15 Widschwendter M, Jones A, Teschendorff AE. Epigenetics makes its mark on women-specific cancers – an opportunity to redefine oncological approaches? Gynecol. Oncol. 128(1), 134–143 (2013).

16 Feil R, Fraga MF. Epigenetics and the environment: emerging patterns and implications. Nat. Rev. Genet. 13(2), 97–109 (2011).

17 Heijmans BT, Tobi EW, Stein AD et al. Persistent epigenetic differences associated with prenatal exposure to famine in humans. Proc. Natl Acad. Sci. USA 105(44), 17046–17049 (2008).

18 Jirtle RL, Skinner MK. Environmental epigenomics and disease susceptibility. Nat. Rev. Genet. 8(4), 253–262 (2007).

19 Teschendorff AE, West J, Beck S. Age-associated epigenetic drift: implications, and a case of epigenetic thrift? Hum. Mol. Genet. 22(R1), R7–R15 (2013).

20 Ahuja N, Issa JP. Aging, methylation and cancer. Histol. Histopathol. 15(3), 835–842 (2000).

21 Ahuja N, Li Q, Mohan AL, Baylin SB, Issa JP. Aging and DNA methylation in colorectal mucosa and cancer. Cancer Res. 58(23), 5489–5494 (1998).

22 Fraga MF, Ballestar E, Paz MF et al. Epigenetic differences arise during the lifetime of monozygotic twins. Proc. Natl Acad. Sci. USA 102(30), 10604–10609 (2005).

• Oneofthefirstpaperstodemonstratethephenomenonofepigeneticdriftonagenome-widescale.

23 Christensen BC, Houseman EA, Marsit CJ et al. Aging and environmental exposures alter tissue-specific DNA methylation dependent upon CpG island context. PLoS Genet. 5(8), e1000602 (2009).

24 Maegawa S, Hinkal G, Kim HS et al. Widespread and tissue specific age-related DNA methylation changes in mice. Genome Res. 20(3), 332–340 (2010).

25 Teschendorff AE, Menon U, Gentry-Maharaj A et al. Age-dependent DNA methylation of genes that are suppressed in stem cells is a hallmark of cancer. Genome Res. 20(4), 440–446 (2010).

• Oneofthefirstpaperstoshowthatage-associatedDNAmethylationsignaturesareindependentoftissuetype,presentinstemcellpopulations,andwhichbecomeaggravatedincancer.

26 Rakyan VK, Down TA, Maslau S et al. Human aging-associated DNA hypermethylation occurs preferentially at



bivalent chromatin domains. Genome Res. 20(4), 434–439 (2010).

27 Heyn H, Li N, Ferreira HJ et al. Distinct DNA methylomes of newborns and centenarians. Proc. Natl Acad. Sci. USA 109(26), 10522–10527 (2012).

28 Day K, Waite LL, Thalacker-Mercer A et al. Differential DNA methylation with age displays both common and dynamic features across human tissues that are influenced by CpG landscape. Genome Biol. 14(9), R102 (2013).

29 West J, Beck S, Wang X, Teschendorff AE. An integrative network algorithm identifies age-associated differential methylation interactome hotspots targeting stem-cell differentiation pathways. Sci. Rep. 3, 1630 (2013).

30 Horvath S, Zhang Y, Langfelder P et al. Aging effects on DNA methylation modules in human brain and blood tissue. Genome Biol. 13(10), R97 (2012).

31 Bork S, Pfister S, Witt H et al. DNA methylation pattern changes upon long-term culture and aging of human mesenchymal stromal cells. Aging Cell 9(1), 54–63 (2010).

32 Beerman I, Bock C, Garrison BS et al. Proliferation-dependent alterations of the DNA methylation landscape underlie hematopoietic stem cell aging. Cell Stem Cell 12(4), 413–425 (2013).

•• VeryimportantpaperdemonstratingDNAmethylationalterationsduringhematopoieticontogeny.

33 Bocker MT, Hellwig I, Breiling A, Eckstein V, Ho AD, Lyko F. Genome-wide promoter DNA methylation dynamics of human hematopoietic progenitor cells during differentiation and aging. Blood 117(19), e182–e189 (2011).

34 Hannum G, Guinney J, Zhao L et al. Genome-wide methylation profiles reveal quantitative views of human aging rates. Mol. Cell 49(2), 359–367 (2013).

35 Bocklandt S, Lin W, Sehl ME et al. Epigenetic predictor of age. PLoS ONE 6(6), e14821 (2011).

36 Horvath S. DNA methylation age of human tissues and cell types. Genome Biol. 14(10), R115 (2013).

•• Landmarkpaperdemonstratingthatchronologicalagecanbepredictedwithhighaccuracyacrossmanyindependenttissuetypes.

37 Koch CM, Wagner W. Epigenetic-aging-signature to determine age in different tissues. Aging 3(10), 1018–1027 (2011).

38 Aubert G, Lansdorp PM. Telomeres and aging. Physiol. Rev. 88(2), 557–579 (2008).

39 Meissner C, Ritz-Timme S. Molecular pathology and age estimation. Forensic Sci. Int. 203(1–3), 34–43 (2010).

40 Zubakov D, Liu F, Van Zelm MC et al. Estimating human age from T-cell DNA rearrangements. Curr. Biol. 20(22), R970–R971 (2010).

41 Lin J, Epel E, Blackburn E. Telomeres and lifestyle factors: roles in cellular aging. Mutat. Res. 730(1–2), 85–89 (2012).

42 Weidner CI, Lin Q, Koch CM et al. Aging of blood can be tracked by DNA methylation changes at just three CpG sites. Genome Biol. 15(2), R24 (2014).

43 Laurie CC, Laurie CA, Rice K et al. Detectable clonal mosaicism from birth to old age and its relationship to cancer. Nat. Genet. 44(6), 642–650 (2012).

44 Jacobs KB, Yeager M, Zhou W et al. Detectable clonal mosaicism and its relationship to aging and cancer. Nat. Genet. 44(6), 651–658 (2012).

45 Alexandrov LB, Jones PH, Wedge DC et al. Clock-like mutational processes in human somatic cells. Nat. Genet. 47(12), 1402–1407 (2015).

46 De Magalhaes JP, Curado J, Church GM. Meta-analysis of age-related gene expression profiles identifies common signatures of aging. Bioinformatics 25(7), 875–881 (2009).

47 Lin Q, Wagner W. Epigenetic aging signatures are coherently modified in cancer. PLoS Genet. 11(6), e1005334 (2015).

48 Teschendorff AE, Menon U, Gentry-Maharaj A et al. An epigenetic signature in peripheral blood predicts active ovarian cancer. PLoS ONE 4(12), e8274 (2009).

49 Houseman EA, Accomando WP, Koestler DC et al. DNA methylation arrays as surrogate measures of cell mixture distribution. BMC Bioinformatics 13, 86 (2012).

50 Horvath S, Erhart W, Brosch M et al. Obesity accelerates epigenetic aging of human liver. Proc. Natl Acad. Sci. USA 111(43), 15538–15543 (2014).

51 Rickabaugh TM, Baxter RM, Sehl M et al. Acceleration of age-associated methylation patterns in HIV-1-infected adults. PLoS ONE 10(3), e0119201 (2015).

52 Horvath S, Levine AJ. HIV-1 infection accelerates age according to the epigenetic clock. J. Infect. Dis. 212(10), 1563–1573 (2015).

53 Horvath S, Garagnani P, Bacalini MG et al. Accelerated epigenetic aging in Down syndrome. Aging Cell 14(3), 491–495 (2015).

54 Marioni RE, Shah S, Mcrae AF et al. The epigenetic clock is correlated with physical and cognitive fitness in the Lothian Birth Cohort 1936. Int. J. Epidemiol. 44(4), 1388–1396 (2015).

55 Horvath S. Erratum to: DNA methylation age of human tissues and cell types. Genome Biol. 16, 96 (2015).

56 Levine ME, Hosgood HD, Chen B, Absher D, Assimes T, Horvath S. DNA methylation age of blood predicts future onset of lung cancer in the women’s health initiative. Aging 7(9), 690–700 (2015).

57 Marioni RE, Shah S, Mcrae AF et al. DNA methylation age of blood predicts all-cause mortality in later life. Genome Biol. 16(1), 25 (2015).

• ImportantpaperdemonstratingthatHorvath’sDNAmethylationagemeasuredinbloodisassociatedwithclinicalphenotypesinlargenumbersofsamples.

58 Teschendorff AE, Jones A, Fiegl H et al. Epigenetic variability in cells of normal cytology is associated with the risk of future morphological transformation. Genome Med. 4(3), 24 (2012).

• FirstpapertoshowthattheprospectiveriskofanepithelialcancercanbepredictedbasedonaDNAmethylationsignatureinnormalepithelialcells.



59 Issa JP. Aging and epigenetic drift: a vicious cycle. J. Clin. Invest. 124(1), 24–29 (2014).

60 Zeilinger S, Kuhnel B, Klopp N et al. Tobacco smoking leads to extensive genome-wide changes in DNA methylation. PLoS ONE 8(5), e63812 (2013).

61 Shenker NS, Ueland PM, Polidoro S et al. DNA methylation as a long-term biomarker of exposure to tobacco smoke. Epidemiology 24(5), 712–716 (2013).

62 Teschendorff AE, Yang Z, Wong A et al. Correlation of smoking-associated DNA methylation changes in buccal cells with DNA methylation changes in epithelial cancer. JAMA Oncol. 1(4), 476–485 (2015).

63 Gao X, Jia M, Zhang Y, Breitling LP, Brenner H. DNA methylation changes of whole blood cells in response to active smoking exposure in adults: a systematic review of DNA methylation studies. Clin. Epigenetics 7, 113 (2015).

64 Zhang Y, Schottker B, Florath I et al. Smoking-associated DNA methylation biomarkers and their predictive value for all-cause and cardiovascular mortality. Environ. Health Perspect. 124(1), 67–74 (2015).

65 Zhang Y, Schottker B, Ordonez-Mena J et al. F2RL3 methylation, lung cancer incidence and mortality. Int. J. Cancer 137(7), 1739–1748 (2015).

66 Anjum S, Fourkala EO, Zikan M et al. A BRCA1-mutation associated DNA methylation signature in blood cells predicts sporadic breast cancer incidence and survival. Genome Med. 6(6), 47 (2014).

67 Gronniger E, Weber B, Heil O et al. Aging and chronic sun exposure cause distinct epigenetic changes in human skin. PLoS Genet. 6(5), e1000971 (2010).

68 Vandiver AR, Irizarry RA, Hansen KD et al. Age and sun exposure-related widespread genomic blocks of hypomethylation in nonmalignant skin. Genome Biol. 16(1), 80 (2015).

69 Ronn T, Volkov P, Gillberg L et al. Impact of age, BMI and HbA1c levels on the genome-wide DNA methylation and mRNA expression patterns in human adipose tissue and identification of epigenetic biomarkers in blood. Hum. Mol. Genet. 24(13), 3792–3813 (2015).

70 Dick KJ, Nelson CP, Tsaprouni L et al. DNA methylation and body-mass index: a genome-wide analysis. Lancet 383(9933), 1990–1998 (2014).

71 Feinberg AP, Irizarry RA, Fradin D et al. Personalized epigenomic signatures that are stable over time and covary with body mass index. Sci. Transl. Med. 2(49), 49ra67 (2010).

72 Hasler R, Feng Z, Backdahl L et al. A functional methylome map of ulcerative colitis. Genome Res. 22(11), 2130–2137 (2012).

73 Issa JP. Epigenetic variation and cellular Darwinism. Nat. Genet. 43(8), 724–726 (2011).

74 Issa JP, Ahuja N, Toyota M, Bronner MP, Brentnall TA. Accelerated age-related CpG island methylation in ulcerative colitis. Cancer Res. 61(9), 3573–3577 (2001).

75 Philibert RA, Plume JM, Gibbons FX, Brody GH, Beach SR. The impact of recent alcohol use on genome wide DNA methylation signatures. Front. Genet. 3 54 (2012).

76 Noreen F, Roosli M, Gaj P et al. Modulation of age- and cancer-associated DNA methylation change in the healthy colon by aspirin and lifestyle. J. Natl Cancer Inst. 106(7), pii: dju161 (2014) (Epub ahead of print).

77 Watson CT, Disanto G, Sandve GK, Breden F, Giovannoni G, Ramagopalan SV. Age-associated hyper-methylated regions in the human brain overlap with bivalent chromatin domains. PLoS ONE 7(9), e43840 (2012).

78 Lam LL, Emberly E, Fraser HB et al. Factors underlying variable DNA methylation in a human community cohort. Proc. Natl Acad. Sci. USA 109(Suppl. 2), 17253–17260 (2012).

79 Hansen KD, Sabunciyan S, Langmead B et al. Large-scale hypomethylated blocks associated with Epstein-Barr virus-induced B-cell immortalization. Genome Res. 24(2), 177–184 (2014).

80 Hahn MA, Hahn T, Lee DH et al. Methylation of polycomb target genes in intestinal cancer is mediated by inflammation. Cancer Res. 68(24), 10280–10289 (2008).

81 Lee TI, Jenner RG, Boyer LA et al. Control of developmental regulators by Polycomb in human embryonic stem cells. Cell 125(2), 301–313 (2006).

82 Widschwendter M, Fiegl H, Egle D et al. Epigenetic stem cell signature in cancer. Nat. Genet. 39(2), 157–158 (2007).

83 Ohm JE, Mcgarvey KM, Yu X et al. A stem cell-like chromatin pattern may predispose tumor suppressor genes to DNA hypermethylation and heritable silencing. Nat. Genet. 39(2), 237–242 (2007).

84 Schlesinger Y, Straussman R, Keshet I et al. Polycomb-mediated methylation on Lys27 of histone H3 pre-marks genes for de novo methylation in cancer. Nat. Genet. 39(2), 232–236 (2007).

85 Amant F, Moerman P, Neven P, Timmerman D, Van Limbergen E, Vergote I. Endometrial cancer. Lancet 366(9484), 491–505 (2005).

86 Udler MS, Ahmed S, Healey CS et al. Fine scale mapping of the breast cancer 16q12 locus. Hum. Mol. Genet. 19(12), 2507–2515 (2010).

87 Rando TA, Chang HY. Aging, rejuvenation, and epigenetic reprogramming: resetting the aging clock. Cell 148(1–2), 46–57 (2012).

88 Brack AS, Rando TA. Intrinsic changes and extrinsic influences of myogenic stem cell function during aging. Stem Cell Rev. 3(3), 226–237 (2007).

89 Xu Z, Bolick SC, Deroo LA, Weinberg CR, Sandler DP, Taylor JA. Epigenome-wide association study of breast cancer using prospectively collected sister study samples. J. Natl Cancer Inst. 105(10), 694–700 (2013).

90 Van Veldhoven K, Polidoro S, Baglietto L et al. Epigenome-wide association study reveals decreased average methylation levels years before breast cancer diagnosis. Clin. Epigenetics 7(1), 67 (2015).

91 Severi G, Southey MC, English DR et al. Epigenome-wide methylation in DNA from peripheral blood as a marker of risk for breast cancer. Breast Cancer Res. Treat. 148(3), 665–673 (2014).



92 Joyce BT, Gao T, Liu L et al. Longitudinal study of DNA methylation of inflammatory genes and cancer risk. Cancer Epidemiol. Biomarkers Prev. 24(10), 1531–1538 (2015).

93 Beerman I, Bhattacharya D, Zandi S et al. Functionally distinct hematopoietic stem cells modulate hematopoietic lineage potential during aging by a mechanism of clonal expansion. Proc. Natl Acad. Sci. USA 107(12), 5465–5470 (2010).

94 Maegawa S, Gough SM, Watanabe-Okochi N et al. Age-related epigenetic drift in the pathogenesis of MDS and AML. Genome Res. 24(4), 580–591 (2014).

95 Grolleau-Julius A, Ray D, Yung RL. The role of epigenetics in aging and autoimmunity. Clin. Rev. Allergy Immunol. 39(1), 42–50 (2010).

96 Steegenga WT, Boekschoten MV, Lute C et al. Genome-wide age-related changes in DNA methylation and gene expression in human PBMCs. Age (Dordr.) 36(3), 9648 (2014).

97 Yuan T, Jiao Y, De Jong S, Ophoff RA, Beck S, Teschendorff AE. An integrative multi-scale analysis of the dynamic DNA methylation landscape in aging. PLoS Genet. 11(2), e1004996 (2015).

98 Lee HS. Impact of maternal diet on the epigenome during in utero life and the developmental programming of diseases in childhood and adulthood. Nutrients 7(11), 9492–9507 (2015).

99 Martino D, Loke YJ, Gordon L et al. Longitudinal, genome-scale analysis of DNA methylation in twins from birth to 18 months of age reveals rapid epigenetic change in early life and pair-specific effects of discordance. Genome Biol. 14(5), R42 (2013).

• Importantpaperdemonstratingthatepigeneticdriftcanbemeasuredinthefirstfewmonthsoflife.

100 Alisch RS, Barwick BG, Chopra P et al. Age-associated DNA methylation in pediatric populations. Genome Res. 22(4), 623–632 (2012).

101 West J, Widschwendter M, Teschendorff AE. Distinctive topology of age-associated epigenetic drift in the human interactome. Proc. Natl Acad. Sci. USA 110(35), 14138–14143 (2013).

102 Teschendorff AE. Epigenetic aging: insights from network biology. Aging 5(10), 719–720 (2013).

103 Genovese G, Kahler AK, Handsaker RE et al. Clonal hematopoiesis and blood-cancer risk inferred from blood DNA sequence. N. Engl. J. Med. 371(26), 2477–2487 (2014).

104 Adams PD, Jasper H, Rudolph KL. Aging-induced stem cell mutations as drivers for disease and cancer. Cell Stem Cell 16(6), 601–612 (2015).

105 Suzuki H, Toyota M, Carraway H et al. Frequent epigenetic inactivation of Wnt antagonist genes in breast cancer. Br. J. Cancer 98(6), 1147–1156 (2008).

106 Baylin SB, Ohm JE. Epigenetic gene silencing in cancer – a mechanism for early oncogenic pathway addiction? Nat. Rev. Cancer 6(2), 107–116 (2006).

•• Seminalreviewpaper,exposingtheideaofDNAmethylationalterationsasearlyeventsinpredisposingnormalcellstocarcinogenesis.

107 Jones A, Teschendorff AE, Li Q et al. Role of DNA methylation and epigenetic silencing of HAND2 in endometrial cancer development. PLoS Med. 10(11), e1001551 (2013).

108 Tabassum R, Sivadas A, Agrawal V, Tian H, Arafat D, Gibson G. Omic personality: implications of stable transcript and methylation profiles for personalized medicine. Genome Med. 7(1), 88 (2015).

109 Wadsworth M, Kuh D, Richards M, Hardy R. Cohort profile: The 1946 National Birth Cohort (MRC National Survey of Health and Development). Int. J. Epidemiol. 35(1), 49–54 (2006).

110 Bell JT, Spector TD. DNA methylation studies using twins: what are they telling us? Genome Biol. 13(10), 172 (2012).

111 Faulk C, Liu K, Barks A, Goodrich JM, Dolinoy DC. Longitudinal epigenetic drift in mice perinatally exposed to lead. Epigenetics 9(7), 934–941 (2014).

112 Tomasetti C, Vogelstein B, Parmigiani G. Half or more of the somatic mutations in cancers of self-renewing tissues originate prior to tumor initiation. Proc. Natl Acad. Sci. USA 110(6), 1999–2004 (2013).

113 Tomasetti C, Vogelstein B. Cancer etiology. Variation in cancer risk among tissues can be explained by the number of stem cell divisions. Science 347(6217), 78–81 (2015).

114 Hartnett L, Egan LJ. Inflammation, DNA methylation and colitis-associated cancer. Carcinogenesis 33(4), 723–731 (2012).

115 Suzuki H, Toyota M, Kondo Y, Shinomura Y. Inflammation-related aberrant patterns of DNA methylation: detection and role in epigenetic deregulation of cancer cell transcriptome. Methods Mol. Biol. 512 55–69 (2009).

116 Hansen KD, Timp W, Bravo HC et al. Increased methylation variation in epigenetic domains across cancer types. Nat. Genet. 43(8), 768–775 (2011).

117 Feinberg AP, Irizarry RA. Evolution in health and medicine Sackler colloquium: stochastic epigenetic variation as a driving force of development, evolutionary adaptation, and disease. Proc. Natl Acad. Sci. USA 107(Suppl. 1), 1757–1764 (2010).

118 Feinberg AP. Epigenetic stochasticity, nuclear structure and cancer: the implications for medicine. J. Int. Med. 276(1), 5–11 (2014).

119 Wijetunga NA, Delahaye F, Zhao YM et al. The meta-epigenomic structure of purified human stem cell populations is defined at cis-regulatory sequences. Nat. Commun. 5, 5195 (2014).

120 Siegmund KD, Marjoram P, Woo YJ, Tavare S, Shibata D. Inferring clonal expansion and cancer stem cell dynamics from DNA methylation patterns in colorectal cancers. Proc. Natl Acad. Sci. USA 106(12), 4828–4833 (2009).

121 Shibata D. Clonal diversity in tumor progression. Nat. Genet. 38(4), 402–403 (2006).

122 Teschendorff AE, Liu X, Caren H et al. The dynamics of DNA methylation covariation patterns in carcinogenesis. PLoS Comput. Biol. 10(7), e1003709 (2014).

123 Landau DA, Clement K, Ziller MJ et al. Locally disordered methylation forms the basis of intratumor methylome



variation in chronic lymphocytic leukemia. Cancer Cell 26(6), 813–825 (2014).

124 Mazor T, Pankov A, Johnson BE et al. DNA methylation and somatic mutations converge on the cell cycle and define similar evolutionary histories in brain tumors. Cancer Cell 28(3), 307–317 (2015).

125 Issa JP. Age-related epigenetic changes and the immune system. Clin. Immunol. 109(1), 103–108 (2003).

126 Langevin SM, Houseman EA, Accomando WP et al. Leukocyte-adjusted epigenome-wide association studies of blood from solid tumor patients. Epigenetics 9(6), 884–895 (2014).

127 Lechner M, Fenton T, West J et al. Identification and functional validation of HPV-mediated hypermethylation in head and neck squamous cell carcinoma. Genome Med. 5(2), 15 (2013).

128 Xu Z, Taylor JA. Genome-wide age-related DNA methylation changes in blood and other tissues relate to histone modification, expression and cancer. Carcinogenesis 35(2), 356–364 (2014).

129 Roadmap Epigenomics Consortium, Kundaje A, Meuleman W et al. Integrative analysis of 111 reference human epigenomes. Nature 518(7539), 317–330 (2015).


part of

Epigenome-wide analysis of DNA methylation reveals a rectal cancer-specific epigenomic signature

Veronika Vymetalkova*,1,2, Pavel Vodicka1,2,3, Barbara Pardini4, Fabio Rosa4, Miroslav Levy5, Michaela Schneiderova6, Vaclav Liska3,7, Ludmila Vodickova1,2,3, Torbjörn K Nilsson8 & Sanja A Farkas9

1Institute of Experimental Medicine,

Academy of Sciences of the Czech

Republic, Prague, Czech Republic 2Institute of Biology & Medical Genetics,

1st Medical Faculty, Charles University,

Prague, Czech Republic 3Biomedical Centre, Faculty of Medicine

in Pilsen, Charles University, Czech

Republic 4Human Genetics Foundation, (HuGeF),

Torino, Italy 5Department of Surgery, 1st Faculty of

Medicine, Charles University & Thomayer

Hospital, Prague, Czech Republic 6Department of Surgery, General

University Hospital, Prague, Czech

Republic 7Department of Surgery, Teaching

Hospital & Medical School in Pilsen,

Charles University, Pilsen, Czech Republic 8Department of Medical

Biosciences/Clinical Chemistry, Umeå

University, Umeå, Sweden 9Department of Laboratory Medicine,

Örebro University; Örebro, Sweden


Tel.: +420 296 2251

[email protected]

Research Article


Epigenomics

Research Article 2016/08/298

9

2016

Aim: The aim of the present study is to address a genome-wide search for novel methylation biomarkers in the rectal cancer (RC), as only scarce information on methylation profile is available. Materials & methods: We analyzed methylation status in 25 pairs of RC and adjacent healthy mucosa using the Illumina Human Methylation 450 BeadChip. Results: We found significantly aberrant methylation in 33 genes. After validation of our results by pyrosequencing, we found a good agreement with our findings. The BPIL3 and HBBP1 genes resulted hypomethylated in RC, whereas TIFPI2, ADHFE1, FLI1 and TLX1 were hypermethylated. An external validation by TCGA datasets confirmed the results. Conclusion: Our study, with external validation, has demonstrated the feasibility of using specific methylated DNA signatures for developing biomarkers in RC.

First draft submitted: 20 April 2016; Accepted for publication: 28 June 2016; Published online: 16 August 2016

Keywords: DNA methylation • Illumina Human Methylation 450 BeadChip • rectal cancer

BackgroundColorectal cancer (CRC) is the third most common cause of cancer and the second lead-ing cause of cancer death in Europe and the USA [1]. From a clinical point of view, malig-nancies in the colon (CC) and the rectum (RC, comprising approximately 33%) rep-resent two distinct entities that require dif-ferent treatment strategies. The distinction between the CC and RC is largely anatomi-cal but it impacts both surgical and radio-therapeutic management with often different prognoses [2]. In contrast with CC, which has a low incidence of local recurrence and longer survival time, patients with RC have a higher incidence of recurrence requiring the addition of pelvic radiation therapy (chemo-radiation) [3,4]. As a consequence, the clini-cal management of patients with RC differs significantly from that of the CC in terms of surgical technique, the more frequent use of radiotherapy and method of chemotherapy administration [5].

There are some examples of studies that tried to clarify whether established CRC risk factors may or may not be risk factors for CC or RC separately [6]. For example, physical inactivity and body mass index have been associated with CC cancer but not with rec-tal cancer [7]. However, for RC only, very lim-ited data are available, since existing studies usually failed to separate these entities.

From a molecular point of view, the preva-lence of K-ras mutations and mutation pat-terns in the TP53 gene in RC differs from those seen in CC [8]. For all these reasons, RC and CC should preferably be analyzed separately to reduce the attenuation of risk estimates for RC in the studies.

At present, it is generally assumed that CRC arises as a consequence of an accumulation of genetic and epigenetic alterations, which transforms colonic epithelial cells into adeno-carcinoma cells. DNA methylation is an epi-genetic event that alters gene expression and that may lead to cancer and other human dis-



Research Article Vymetalkova, Vodicka, Pardini et al.

eases. Global DNA hypomethylation is a type of altered DNA methylation which often occurs in repetitive ele-ments of the genome such as long interspersed repeat sequences (LINE-1). Global DNA hypomethylation is associated with genomic instability and chromosome abnormalities [9,10]. Gene-specific methylation occurs at specific regions of the gene such as gene promoters and can either silence or activate the gene. It is gener-ally accepted dogma that CpG island hypermethylation leads to transcriptional gene silencing [11].

The epigenetic changes associated with CRC, espe-cially aberrant CpG island methylation in the pro-moter regions of tumor suppressor genes, have been tested in several studies [2]. In general, up to 10% of CpG islands in cancer epigenomes may be aberrantly methylated, which can lead to the silencing of thou-sands of gene promoters in the average cancer [12]. Moreover, CRC-associated aberrant methylation is not exclusively limited to CpG islands but may com-prise ‘CpG island shores’ or areas that are less dense in CpG dinucleotides within 2 kb upstream of a CpG island [13,14]. Methylation of the CpG island shores may also be associated with the transcriptional inactivation and expression of splice variants [15].

In the last decade, technologies for analyses on genome scale have progressed, and new tools have been implemented to characterize the full spectrum of molecular heterogeneity in many types of cancer cells [16]. With respect to CRC, genome-wide meth-ylation changes have been identified in the past few years [13,15,17].

To our knowledge, no published studies focused on epigenetic diversity in RC by using a state-of-the-art high-density methylation array. In the present study, CpG-level methylation of tumor and matched adjacent tissues from RC patients were analyzed using Infinium HumanMethylation450K BeadChips. This enabled us to characterize differentially methylated regions involved in RC pathogenesis and identify novel can-didate genes not previously associated with aberrant methylation in RC.

Materials & methodsClinical samplesThe study comprised 64 paired samples (tumor tissue and adjacent nonmalignant rectal mucosa [ANMRM]) from 32 patients with RC. We included only those patients that at the time of the collection, did not receive any adjuvant therapy yet. Clinical character-istics of patients are presented in Table 1. The Ethics committees of the Institute for Clinical and Experi-mental Medicine, the Thomayer Hospital, Prague (C.j. 786/09(G0-04-09), the General University Hospital, Prague (C.j. 12/11 Grant GAČR 1.LF UK) and the Teaching Hospital and Medical School, Pilsen (for project IGA MZCR NT14329) approved the study. All patients signed informed consent. The Ethical Review Board, Uppsala, Sweden approved the Swedish participation in the study.

DNA methylation arrayThe 64 RC and ANMRM specimens were homoge-nized by MagNA Lyser (Hoffmann-La Roche, Prague, Czech Republic) and the genomic DNA was extracted using the AllPrep DNA/RNA Isolation Kit (Qiagen, Prague, Czech Republic) protocol according to the manufacturer’s instructions. DNA samples were ana-lyzed with the Illumina Infinium HumanMethyl-ation450K BeadChip according to standard laboratory procedures obtained from Illumina, described in details by Sandoval et al. [18]. The BeadChips were read by an iScan scanner and the data collection was performed in the GenomeStudio software (version 1.0).

Data processingCpG sites with bead count less than three in more than 5% of the samples were removed, together with sites where more than 1% of the samples had a detection

Table 1. Patient’s clinical characteristics.

Characteristic n (total = 32)

Gender:

– Male 20

– Female 12

Age mean: 65.8

– Male 68.4

– Female 63.1

CIN status:

– MSS 32

– MSI 0

Tumor differentiation:

– Low 10

– Moderate 19

– High 3

TNM stage:

– Stage I 8

– Stage II 10

– Stage III 8

– Stage IV 6

Therapy:

– Neoadjuvant therapy 5

– Adjuvant therapy 9

CIN: Chromosomal instability; MSI: Microsatellite instable; MSS: Microsatellite stable.


DNA methylation in rectal cancer Research Article

p-value > 0.05 were filtered away from the raw data. The data were quantile normalized before the calcula-tion of β-values. Beta mixture quantile dilation was used to eliminate the probe type bias in the Illumina Infinium technology. The combination of Quantile normalization together with beta mixture quantile dilation was recently suggested as the most effective normalization strategy when dealing with Illumina HumanMethylation450 BeadChip data [17]. Genetic variations may affect probe hybridization; therefore, probe filtering for single nucleotide polymorphisms was applied according to the Supplementary Table 1 from Nordlund et al. [19]. After filtering, 434,749 CpG sites remained for further analysis. Probes on X and Y chromosomes were removed prior to calculating the Δβ values. The gene body regional analysis included all the CpG sites annotated to the same RefSeq gene (within the boundaries of 5′ and 3′ UTRs) and a new averaged β-value was calculated. When a CpG site was annotated to more than one gene, it was used in the average calculation for all present genes. A CpG locus was considered differentially methylated if the Δβ-value (between tumor and ANMRM tissue samples) ≥ |0.3| and the adjusted p-value < 0.05. A cut-off value of |0.2| represents the detection limit of differential methylation with 99% CI [20].

Array validation with pyrosequencingSix CpG sites targeted by the array were validated using the Pyrosequencing assay technology. CpG sites of interest were located on TIFPI2, HBBP1, ADHFE1, BPIL3, FLI1 and TLX1 genes, and 12 sample pairs pre-viously analyzed with the Illumina Infinium Human-Methylation450K BeadChip were analyzed for valida-tion. Five hundred nanograms of DNA were used for the bisulfite treatment using the EZ DNA Methylation Gold kit according to the instructions and eluted in 14 μl elution buffer (Zymo Research, Freiburg, Ger-many). PCR and sequencing primers were designed using Pyromark assay design software 2.0 (Qiagen). Fifty microliter PCR reaction was performed with the HotStarTaq DNA Polymerase Kit (Qiagen), con-taining 0.15 μmol/l of each primer, 1.25 units of Taq polymerase, 1.5–2.5 mM MgCl

2 and 0.1 mM each of

dGTP, dATP, dTTP, dCTP and approximately 40 ng of bisulfite-treated DNA was added as a template. The PCR primers, annealing temperatures and amplicon sizes are shown in Supplementary Table 1. All prim-ers were purchased from [21]. The PCR program was as follows: initial denaturation step of 5 min at 95°C, followed by 50 cycles of 45-s denaturation at 94°C, 45 s of annealing at 54 or 56°C with an extension of 45 s at 72°C and one cycle for 7 min at 72°C. After PCR, the samples were prepared for pyrosequencing

using the Vacuum Prep Workstation (Qiagen): 37 μl of the amplicon, 3 μl Streptavidin Sepharose HP Beads (Amersham Biosciences, UK) and 40 μl binding buffer (10 mmol/l Tris-HCl, 2 mol/l NaCl, 1 mmol/l EDTA, 0.1% Tween-20, Milli-Q (18.2 MΩ × cm) water, pH 7.6) were mixed and used in the Vacuum Prep work-station. The biotinylated amplicons were immobilized onto the Streptavidin sepharose beads and then passed through one denaturation and two washing steps using the Vacuum Prep Workstation according to a standard protocol. The amplicons were subsequently transferred to a plate containing sequencing primers (0.4 μmol/l) in 40 μl annealing buffer (20 mmol/l Tris-Acetate, 2 mmol/l Magnesium acetate, pH 7.6). Sequencing was performed using a Pyromark Gold Q96 Reagent Kit and a PSQ 96ID system (Qiagen). The nucleotide addition order was optimized by the Pyro Q-CpG software version 1.0.9 (Qiagen). Results were automatically analyzed using the same software.

External validation in TCGA rectal tumor & normal samplesResults from methylation profiling on RC patients were compared with an open access dataset of RC indi-viduals from the cancer genome atlas (TCGA) proj-ect. Level 3 genomic data from RC publicly available in TCGA were employed as validation test set. TCGA DNA methylation data were generated using the Illu-mina Infinium Meth450K platform and presented as β values, with 0 indicating 0% DNA methylation and β values of 1 indicating 100% DNA methylation [2]. Methylation data on 485,577 CpG sites from 98 rec-tal tumors and seven ANMRM rectal tissue samples were obtained from the TCGA Data Portal. To vali-date our results, we also investigated RNA expression data for RC patients (RNAseq. Level 2 data available at the time of manuscript preparation) by the Wan-derer tool [22], an interactive viewer to explore DNA methylation and gene expression data in TCGA. This set comprises 91 RC patients and nine ANMRM. The overlap of tissues for methylation and expres-sion data was the following: 89 RC patients and two ANMRM. The unadjusted p-values < 0.05 were con-sidered as statistically significant, since these analyses were hypothesis driven.

Gene functional classificationWe used the functional annotation tool available in the Database for Annotation, Visualization and Inte-grated Discovery to identify the gene ontology (GO) terms that were over-represented in the list of hyper- and hypomethylated genes. To narrow the number of genes in the list, we selected genes with differentially methylated CpG sites in the 5′UTR and upstream



gene region: in total, 806 hypermethylated genes and 302 hypomethylated genes were included.

Statistical analysisAll statistical analyses on data measured with the DNA methylation array were performed using the R-package environment [23]. Color balance adjustment and back-ground correction were performed using the methylumi package available from the Bioconductor project [24]. Filtering was performed using the ‘pfilter’ function with default settings from the R-package wateRmelon, also available from the Bioconductor project. The data were quantile normalized, using the ‘nanet’ method from the same R-package, prior to calculating the p-values for dif-ference in DNA methylation using the empirical Bayes’ moderated t-statistic [25].

The Benjamini–Hochberg method was used to adjust the p-values for multiple testing [26]. A simple t-test was used for TCGA 3 level methylation data.

ResultsData preprocessingWe have analyzed 64 tissue samples using the Illumina Methylation array (485,577 CpG sites investigated). Two samples failed to meet the criteria of the probe call rate and were excluded from further analysis along with their paired specimen. Due to the preoperative radiotherapy, rectal tumors may contain fairly variable number of cancer cells. For this reason, we initially performed a paired chromosomal analysis displaying DNA methyla-tion changes Δβ ≥ |0.3| and excluded the sample pairs 5, 11, 12, 18 and 29 with no changes in DNA methylation of this magnitude, Supplementary Figure 1. The inclu-sion of patients with Δβ ≤ |0.3| is in fact only likely to dilute the solid findings, not adding new, extra findings to the results. The exclusion of five sample pairs (<10% of patients) with Δβ-value ≤ |0.3| was, therefore, manda-tory according to the design. In total, seven sample pairs were removed.

Differentially methylated genesIn the principal component analysis based on the dif-ferential DNA methylation of the CpG loci there was a marked separation among RC and ANMRM (Figure 1). Applying the criterion Δβ ≥ 0.3, we found 5929 CpG sites differentially methylated in RC with the majority of them (4350) located within a gene (Figure 2A & B). Of these last, 3527 were hypermethylated and 823 hypo-methylated. These CpG sites mapped to 1192 differ-ent genes, mostly on chromosomes 1, 4, 6–8 and 13 (Figure 2A & B). Hypermethylation was predominantly observed in CpG islands. On the other hand, hypometh-ylation was predominantly found in intergenic regions, so called open sea (Figure 2C & D). We used Database

for Annotation, Visualization and Integrated Discovery to identify GO terms enriched in our list of differen-tially methylated genes. The hypomethylated genes were involved in processes such as receptor and membrane activity and the hypermethylated genes in multicellular organism development, neurogenesis or regulation of cellular processes (Table 2).

When performing an unguided gene regional analysis (including CpG sites from 5′ and 3′UTRs), we found 33 genes differentially methylated in RC. The specimens were divided into two clusters representing on the left, the ANMRM tissues and on the right, the RC (Figure 3): only two patients, RC9 and RC10 were misclassified.

We identified those genes with the highest quan-titative differences in methylation between RC and ANMRM. The unguided analysis was based primar-ily on the analyzed Δβ-values and CpG sites with an adjusted p-value < 0.05. The differential methylation for the top ten hypo- and hypermethylated CpG sites and genes ranged between -0.47 and -0.53, and from 0.55 to 0.62, respectively (Table 3). The majority of these CpG sites were located within gene body regions. Particu-larly, the most hypermethylated CpG sites in ADHFE1, TFPI2 and FLI1 genes exhibited a large number of sig-nificantly aberrantly methylated CpG sites in cancer tis-sue: 19 CpG sites out of 23 analyzed covering ADHFE1 gene; 23 out of 25 covering TFPI2 and 39 out of 47 in FLI1. RC tissues displayed significant hypermethylation in the upstream of the 5′ region of ADHFE1 (Figure 4) while hypomethylation was found in the gene body. Sev-eral CpG sites in this gene (cg01588438, cg20912169, cg09383816 and cg20040765) pointed to regions with considerable difference between tumor and healthy tissues (Δβ fractions of -0.26 and 0.57–0.62).

RC tissues also displayed significant CpGs hyper-methylation in the 5′upstream region of the FLI1 gene; in contrast CpG sites belonging to the so called S_Shore displayed hypomethylation (from cg01681098 till the end of the gene; Figure 4).

Other genes with a similar switch pattern were ZNF804B and ZNF793, whereas ZNF385B or RASSF2 showed the opposite direction of the switch (Figure 4).

The 3′UTR region of GPR85 in RC tissues showed a significant hypomethylation while the gene body and 5′UTR region had significant hypermethylation. Inter-estingly, the region 1500kb upstream of a transcription start site (TSS) again showed hypomethylation of RC (Figure 4).

Validation of the methylation array data by pyrosequencingThe differential methylation of six selected genes was validated by pyrosequencing in a subset of tis-sues already analyzed with the Illumina BeadChip


Figure 1. Principal component analysis composed of CpG sites after filtering for cancer. The tumor tissue (red) and adjacent nonmalignant (blue) rectal mucosa specimens differ in DNA methylation.

-0.4

-0.2

-0.1

0.0

0.1

0.2

0.3

-0.2 0.0 0.2

-0.145

-0.140

-0.135

-0.130

-0.125

y

x

z



arrays. Results were in agreement with the array data (Supplementary Figure 2). RC tissues displayed signifi-cant hypomethylation in the BPIL3 and HBBP1 genes, whereas in TIFPI2, ADHFE1, FLI1 and TLX1 tumors resulted as hypermethylated.

External validation of the DNA methylation & mRNA expressionThe most hypo- and hyper-methylated CpGs identified in our study, as well as 33 genes differentially methyl-ated from the unguided hierarchical clustering analy-sis (see heatmap, Figure 3) were investigated also in the TCGA datasets (98 rectal tumors and seven ANMRM rectal tissue) as an external validation. There was a good agreement in differences in DNA methylation between RC and ANMRM tissues among our and the TCGA datasets (Table 3 & Supplementary Table 2).

For RNA-seq, the TCGA dataset available on Wan-derer tool were for 91 rectal tumors and nine ANMRM

rectal tissues. When we overlapped methylation data and RNA-seq in Wanderer, the data in common were only for 89 rectal cancers and two ANMRM tissues.

Generally, the genes with hypermethylated CpG sites evinced lower expression levels in tumors when compared with their ANMRM (particularly ADHFE1, FLI1, NPY and ITGA4). Higher mRNA levels in tumors were observed for PRKAR1B, SND1 and TFPI2. For the genes with hypomethylated CpG sites, only CYP27A1 showed increased expression lev-els. On the other side, decreased expression levels were observed for ITGBL1 and MYBPC3 genes. The major-ity of these CpG sites were localized in the gene body or 5′UTR region, none of them were located in the promoter region of genes.

For those genes observed differentially methylated in the heat map, a general lower expression of all tested genes was observed in RC tissues when compared with their ANMRM counterparts (from p = 0.02

1198 Epigenomics (2016) 8(9)

Within genes

Within gene73%

Intergenic27%

All CpG Iocus

81%

19%

21 15

16

13

10

24

1 45

104

53

30

3

Hypermethylated Hypomethylated

Hypermethylated Hypomethylated

10

6 13

80 76

2 74

5

6

IslandN_ShoreS_ShoreN_ShelfS_ShelfOpen sea

TSS1500TSS2005ÚTRExon1Gene body3ÚTRIntergenic

TSS1500TSS2005ÚTRExon1Gene body3ÚTRIntergenic

IslandN_ShoreS_ShoreN_ShelfS_ShelfOpen sea

8%

8% 9%

9%

6%

3%

6%8%6%

1%1%4%3%1%2%

3%2%

1%5%7%

3%

3%

6%7%8%

6%

7%

5%8%6%

1%2%4%6%2%

2%

3% 2%

2% 6%

1%6%5%

5%

Distribution of hypermethylated CpGlocus

Distribution of hypomethylated CpGlocus

Chr1Chr9

Chr17

Chr2Chr10

Chr18

Chr3Chr11

Chr19

Chr4Chr12

Chr20

Chr5

Chr13Chr21

Chr6

Chr14Chr7

Chr15Chr8

Chr16Chr22

Chr1Chr9

Chr17

Chr2Chr10

Chr18

Chr3Chr11

Chr19

Chr4Chr12

Chr20

Chr5

Chr13Chr21

Chr6

Chr14Chr7

Chr15Chr8

Chr16Chr22

HypermethylatedHypomethylated




Figure 2. Genomic distribution of differentially methylated regions (see facing page). (A) Regional distribution of differentially methylated CpG sites in rectal cancer patients. (B) Chromosomal distribution of differentially methylated CpG sites in rectal cancer patients. (C) The distribution of the hypermethylated and hypomethylated CpG sites over seven gene categories: TSS1500, TSS200, 5′UTR, first exon, gene body, 3′UTR and intergenic regions. (D) The distribution of the hypermethylated and hypomethylated CpG sites over CpG islands, shores, shelves and open sea regions. For categorization, the CpG counts were normalized by the number of CpGs in the same category represented on the 450K array. The percentage of normalized CpG counts is indicated in the bars.



to p < 10-7). The only exception was for the TLX1 gene, where significantly higher expression levels were observed in RC in comparison to ANMRM (p < 10-6; Supplementary Figure 3). Particularly in the gene body region of TLX1, tumor hypermethylation was noticed in our study.

Using the TCGA database, we searched for corre-lations between methylation and RNA-seq expression data for the most hypo- and hyper-methylated CpGs identified in our study. RNA-seq expression data were available for a subset of RC patients in the TCGA proj-ect (89 tumor tissues and two ANMRM). Because of this limited number of ANMRM in TCGA dataset, we were not able to investigate whether changes in DNA methylation observed by us were correlated with respective transcript levels.

DiscussionThe design of the present study was guided by the insight of the different histological and molecular features among CC and RC tissues [28]. We focused on RC, which has in general been under-represented in epigenetic CRC studies. In the present study, RC tissues were characterized by differential methylation of a limited subset of genes showing altered methyla-tion throughout all regions of the gene. In addition, the absolute values of the differences (Δβ) were quite large, usually >0.50. These two features highlight that this limited set of genes constitutes a strong epigenetic signature for RC.

From the epidemiological and therapeutic point of view, CC and rectal tumors are considered as differ-ent entities [4]. They are functionally different and exposed to fecal matter for different time laps. RC, in particular, is exposed to stool in a more concentrated and direct way compared with CC. Also, as undigested matters pass through the CC, they are coated with alkaline mucus. The different levels of pH in proxi-mal and distal locations may influence susceptibility to environmental factors [29]. In addition, the peculiar microenvironment of the RC could have modulatory effects on tumor behavior in addition to promoter methylation and could also obscure any methylation differences [30].

The recently published data from the TCGA project suggested that the overall patterns of changes in meth-ylation, mRNA and miRNA are indistinguishable

between CC and RC [2]. Methylation data obtained from the clinical rectal paired samples studied in the present work may be adequate for use in comparative analyses in other RC methylation studies. The differ-ent epigenetic landscapes between adjacent nonmalig-nant mucosal tissues from the right CC, the left CC and the RC may contribute to determining which genes will show up with the largest Δβ when compar-ing the tumor and the ANMRM tissues from these different locations. It is the delta value, not the absolute one in the tumor tissues that determine which genes are players in the cancer process.

In our study, we found common RC-specific meth-ylation patterns consisting of 5929 CpGs that were sig-nificantly different from those of their healthy counter-parts. These CpG sites mapped 1192 different genes, mostly at chromosomes 1, 4, 6–8 and 13. GO and pathway analysis showed a significant enrichment of the genes containing hypermethylated CpG sites that were related to developmental and regulation activities such as regulation of metabolic or biosynthetic pro-cesses, of transcription or gene expression. By contrast, the hypomethylated CpG sites were related to signal transduction and receptor activity, suggesting that quite diverse cellular processes may be influenced by methylation events and participate in the development of pathological processes in RC.

There is scarce information on the role of the pro-posed gene signature in RC pathogenesis. The TCGA portal presents only few healthy tissues available for RC. The present study reports for the first time an equal representation of the tumor tissue and its healthy counterpart making the observed results solid. Moreover, TCGA cases are from an American mixed population. It is known that diet and other lifestyle risk factors may modify global and gene-specific DNA methylation [31–33]. There is evidence also of a differ-ential DNA methylation in various ethnicities and by gender which may be connected with differences in the dietary habits and in socioeconomic conditions [31–33]. Interestingly, the Czech group of cases derives from a relatively small country characterized by a homoge-neous population (all Caucasians) with individuals sharing a very similar diet. This is an important issue since the observed differential methylation in tumor tissues is, therefore, independent from important bias such as diet and ethnicity.



For two CpGs from our panel (cg1588438 and cg9383816), located in the TSS200 region of ADHFE1, we observed the same methylation pattern previously described in CC cancer patients only [15]. ADHFE1

encodes iron-containing alcohol dehydrogenase, an enzyme responsible for the oxidation of 4-hydroxybu-tyrate in mammals [34]. The hypermethylation of its promoter in CRC has been previously reported [15,35,36].

Table 2. Gene ontology enrichment of hyper- and hypomethylated CpG sites in the 5′UTR and upstream gene regions.

GO terms Number of genes (%) adj. p-value†

Hypomethylated genes

Cluster 1, enrichment score 11.7:

– GO:0004930∼G-protein-coupled receptor activity 44 (15) <0.001

– GO:0004888∼transmembrane receptor activity 51 (18) <0.001

– GO:0007186∼G-protein-coupled receptor protein signaling pathway 48 (17) <0.001


– GO:0004984∼olfactory receptor activity 29 (10) <0.001

– GO:0007606∼sensory perception of chemical stimulus 31 (11) <0.001

– GO:0007608∼sensory perception of smell 29 (10) <0.001


– GO:0004888∼transmembrane receptor activity 51 (18) <0.001

– GO:0004872∼receptor activity 57 (20) <0.001

– GO:0004871∼signal transducer activity 63 (22) <0.001

– GO:0060089∼molecular transducer activity 63 (22) <0.001

Hypermethylated genes


– GO:0007275∼multicellular organismal development 279 (33) <0.001

– GO:0048731∼system development 242 (29) <0.001

– GO:0048856∼anatomical structure development 254 (30) <0.001

– GO:0032502∼developmental process 291 (35) <0.001


– GO:0022008∼neurogenesis 103 (12) <0.001

– GO:0048699∼generation of neurons 95 (11) <0.001

– GO:0030182∼neuron differentiation 83 (10) <0.001


– GO:0031326∼regulation of cellular biosynthetic process 232 (28) <0.001

– GO:0009889∼regulation of biosynthetic process 233 (28) <0.001

– GO:0019219∼regulation of nucleobase, nucleoside, nucleotide and nucleic acid metabolic process

223 (27) <0.001

– GO:0051171∼regulation of nitrogen compound metabolic process 224 (27) <0.001

– GO:0080090∼regulation of primary metabolic process 243 (29) <0.001

– GO:0031323∼regulation of cellular metabolic process 250 (30) <0.001

– GO:0019222∼regulation of metabolic process 257 (31) <0.001

– GO:0045449∼regulation of transcription 199 (24) <0.001

– GO:0010556∼regulation of macromolecule biosynthetic process 209 (25) <0.001

– GO:0010468∼regulation of gene expression 206 (25) <0.001

– GO:0060255∼regulation of macromolecule metabolic process 224 (27) <0.001†p-values were adjusted using the Benjamini–Hochberg method.adj: Adjusted; GO: Gene ontology.


Fig

ure

3.

Hea

tmap

sh

ow

ing

dif

fere

nti

ally

met

hyl

ated

gen

es in

rec

tal m

uco

sa (

red

lett

ers)

an

d a

dja

cen

t n

on

mal

ign

ant

rect

al m

uco

sa (

blu

e le

tter

s).

-0.5

1:1

0.5

C1orf77

HOXA2

FOXD2

NKX2-2

MIR34B

GPR88

TFP12

PCDH8

TLX1

CHST2

MSC

LOC283392

C17orf46

EID3

MIR129-2

NEUROG3

RASSF2

GPR85

LOC283914

DEFB119

C20orf197

EIF3IP1

DEFB118

DEFB122

BEYLA

LOC339568

BPIL3

HBBP1

SAMSN1

ASCC2

IL22RA2

CAPSL

BMPR1B

ANCRM4

ANCRM1

ANCRM8

ANCRM9

ANCRM2

ANCRM14

ANCRM30

ANCRM24

ANCRM28

ANCRM25

ANCRM27

ANCRM17

ANCRM20

ANCRM23

ANCRM16ANCRM22

ANCRM19

ANCRM21

ANCRM15

ANCRM26

ANCRM13

ANCRM3

ANCRM6

ANCRM10

ANCRM7

RC10

RC9

RC26

RC13

RC24

RC16

RC25

RC7

RC6

RC27

RC15

RC3

RC2

RC1

RC30

RC22

RC14

RC23

RC21

RC8

RC17

RC28

RC19

RC20

RC4



1202 Epigenomics (2016) 8(9)

CpG site ID

0.70.60.50.40.30.20.10.0

1.00.90.8

cg11

8863

58

cg22

2358

73cg

1929

6671

cg17

4762

71cg

0903

7089

cg08

5906

01cg

1375

5070

cg06

1724

75

cg01

6810

98cg

1101

7065

cg17

8727

57

cg27

1983

38cg

2642

2472

cg17

5908

05

cg02

5360

65cg

0266

6257

cg16

0245

30

cg12

9407

47

cg08

8252

25cg

0379

8942

TSS 5ÚTR Body

ANMRMRC

FLI1

β-va

lue

0.70.60.50.40.30.20.10.0

1.00.90.8

cg11

6238

61

cg17

3323

26

cg13

1817

45

cg25

3619

07

cg24

0789

85

cg07

4827

95

cg24

5883

75

cg02

7118

01

cg15

1395

88

cg14

7329

98

cg23

2960

10

cg11

3785

02

cg16

2540

93

cg12

2839

16

cg04

8189

19

cg18

1540

14

TSS 5ÚTR 3ÚTRCpG site ID

ANMRMRC

ZNF793

Body

β-va

lue

0.70.60.50.40.30.20.10.0

1.00.90.8

cg01

5884

38

cg09

3838

16

cg18

0653

61

cg08

0907

72

cg19

2838

40

cg20

2954

42

cg20

9121

69

cg01

9881

29

cg26

6246

73

cg08

4906

24

cg25

0466

51

cg20

1800

50

cg20

0407

65

cg22

6286

08

cg11

5302

89

cg05

8502

05

TSS 5ÚTR

CpG site IDBody

ANMRMRC

ADHFE1

β-va

lue

0.70.60.50.40.30.20.10.0

1.00.90.8

cg09

5912

86cg

1274

5764

cg23

2197

20cg

2698

4626

cg16

1724

08cg

2347

4211

cg27

3198

98cg

2115

9370

cg11

1347

20cg

0318

3800

cg19

6827

86

cg13

3544

14

cg15

3251

54cg

0072

9133

cg06

8208

22cg

0310

8373

cg24

9758

98cg

1678

1484

cg22

9461

47

TSS 5ÚTRCpG site ID

Body

ANMRMRC

ZNF804B

β-va

lue

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.0

1.0

0.9

0.8

cg24

5756

76cg

1111

3760

cg26

2718

91cg

0927

9240

cg06

6956

11cg

1640

0999

cg10

0752

87

cg22

6577

80cg

1086

6755

cg01

4528

73

cg16

7052

45cg

1328

5968

cg10

6717

57cg

2236

3400

cg25

4988

15cg

0769

7891

cg19

1452

72cg

2498

3605

cg07

1290

67cg

1982

1128

cg13

2545

18cg

1775

3475

cg04

8689

62cg

1532

8131

cg14

9589

78cg

1349

0979

cg13

5736

53

TSS 5ÚTR

CpG site ID

3ÚTRBody

ANMRM

RC

ZNF385B

β-va

lue

RASSF2

0.70.60.50.40.30.20.10.0

1.00.90.8

cg16

8187

40

cg11

9888

31

cg16

8845

69

cg20

6562

61

cg07

9946

22

cg19

6143

21

cg12

9359

37

cg03

0873

72

cg04

1323

79

cg22

4852

89

cg26

6504

80

cg16

3857

58

cg03

5195

77

cg12

3894

61

cg02

0529

33

cg03

6051

16

cg19

2390

41

cg01

6130

77

cg14

7505

43

cg12

3228

52

cg23

7868

83

cg05

9145

37

TSS 5ÚTR

CpG site ID

3ÚTRBody

β-va

lue

ANMRM

RC




Figure 4. The different DNA methylation profile across the whole gene, covering TSS1500, TSS200, 5′UTR, 1st exon, gene body and 3′UTR regions (see facing page). The DNA methylation profile of the genes: (A) ADHFE1; (B) FLI1; (C) ZNF804B; (D) ZNF793; (E) ZNF385B; (F) RASSF2 and (G) GPR85. The y-axes show the absolute methylated fraction (β-value) of each CpG site. The x-axes show the CpG ID coordinates.



Additionally, a previous study reported an inverse cor-relation similar to our study between ADHFE1 meth-ylation and its expression [36]. Overall, the results on ADHFE1 methylation suggest a role of this gene in both CC and RC pathogenesis but the mechanism involved remains unclear. Kim et al. [37] showed that ADHFE1 transcripts exhibit differentiation-dependent expression during in vivo brown and white adipogen-esis. In another study, ADHFE1 was related to bacterial γ-hydroxybutyrate dehydrogenase and resulted having a conserved NAD-binding site [38].

The methylation of TFPI2 and its loss of expression is a frequent event in human cancers [39,40], including CRC [35,41]. Recently, DNA methylation differences assessed by a targeted DNA microarray in RC tumors also identified TFPI2 as a potential methylation bio-marker in RC [42]. This potential tumor suppressor gene, a Kunitz-type serine proteinase inhibitor, protects the extracellular matrix of cancer cells from degradation and inhibits in vitro colony formation and proliferation [43]. A loss of the TFPI2 function could predispose cells toward a proinvasive program, consistent with an important role of this protein in later stages of carcinogenesis. TFPI2 belongs to the group of embryonic cell Polycomb group (PcG)-marked genes which participate in the 3D struc-ture of nuclear DNA [44] and may target genes with a characteristic ‘bivalent’ promoter chromatin structure containing both active and repressive histone modifica-tions. Such PcG-marked genes may be predisposed to methylation [45] and may thus be good targets for investi-gation as early diagnostic biomarkers. This phenomenon was confirmed by Glockner et al. [45] by the detection of TFPI2 methylation in stool DNA.

Concerning FLI1, an inverse correlation between the hypermethylation and gene expression was pre-viously observed [36]. This gene has also been found hypermethylated in CRC [46,47]. The FLI1 gene is a transcriptional activator playing a role in gene expres-sion regulation, which is also expected to be important in cancer development. All these data were from CRC studies and the similarity with our present results may indicate that RC and CC share common features in their pathogenesis. Interestingly, this particular gene resulted as strongly hypermethylated as well as in our previous study on HER2+ breast cancer [14].

A majority of the differentially methylated genes identified by us were not known to be CRC or RC related. These included several genes with large dif-ferences in β-values: PRKAR1B, TRBJ2-6, HOXA2, NKX2-2, PCDH8, TLX1 and MIR129-2. Interest-ingly, almost all of them presented altered methyla-tion profiles in several cancers, therefore their func-tional role in RC should be the subject of further research.

Aberrant methylation of the HOXA2 gene was recently observed in patients with non-small-cell lung carcinoma [48], nasopharyngeal carcinoma [49] and malignant and benign biliary tissues [50]. For NKX2-2, a hypomethylation was observed in glioblastoma multi-forme [51] and in breast cancer [52] but not in our previous study [14] where we instead found NKX2-4 and NKX2-6 to be hypermethylated. PCDH8 methylation is a frequent event in clear cell renal cell carcinoma [53], bladder [54], breast [14] and gastric cancers [55]. Tommasi et al. [56] and Lindqvist et al. [14] found a hypermethylated TLX1 in breast cancer. Finally, altered methylation of MIR129-2, a

0.70.60.50.40.30.20.10.0

1.00.90.8

cg21

8529

92

cg24

7866

58

cg25

2638

01

cg09

5066

75

cg10

1298

16

cg08

2206

49

cg01

4917

95

cg03

1196

39

cg06

9318

15

cg20

0190

19

cg20

5135

48

cg24

3104

31

cg13

2512

69

cg03

1777

35

cg14

5117

82

cg00

8513

77

TSS 5ÚTR

CpG site ID

3ÚTRBody

GPR85

β-va

lue

ANMRM

RC



tumor suppressive miRNA frequently methylated in lym-phoid but not myeloid malignancies, leads to its reversible silencing [57]. Recently, a strong hypermethylation was observed in HER2+ breast cancer [14].

For the other genes identified in the present study (FOXD2, GPR85, GPR88, CHST2, NEUROG3, DEF118, DEF118, DEF122, EIF3IP1, BEYLA, BPIL3, HBB1, ASCC2, IL22RA2, CAPSLB and HPR1B) no

data regarding methylation and cancer were available, and further cell biological studies of their functional roles could be potentially rewarding.

Crossvalidation against TCGA methylation dataset showed the reliability and reproducibility of the meth-ylation differences identified in our array study and confirmed the high discriminative and noncoincidental potential of the selected biomarkers.

Table 3. Ten most statistically significantly hypo- or hypermethylated CpG sites and genes according to the β-value identified in this study.

Gene name Present study TCGA data Chr CpG ID (Location)

Mean β-value Mean β-value

adj. p-value

Δβ RC Normal mucosa

adj. p-value

Δβ RC Normal mucosa

Hypomethylated CpG sites

MAP3K5 2.1 × 10-15 -0.53 0.19 0.72 1.4 × 10-16 -0.45 0.25 0.70 6 7,474,842Gene body

ITGBL1 3.2 × 10-14 -0.51 0.24 0.73 2.3 × 10-14 -0.50 0.16 0.66 13 11,838,152Gene body

MYBPC3 7.5 × 10-12 -0.50 0.46 0.96 2.5 × 10-67 -0.64 0.32 0.98 11 14,642,259Gene body

adj. to TRBVB 2.2 × 10-13 -0.48 0.30 0.78 1.8 × 10-35 -0.42 0.23 0.65 7 1,703,205

MAP2K2 1.7 × 10-13 -0.48 0.31 0.79 6.5 × 10-22 -0.53 0.21 0.74 19 14,573,876Gene body

adj. to Mafb 2.9 × 10-15 -0.48 0.15 0.64 6.2 × 10-11 -0.47 0.12 0.59 16 3,498,081

adj. to ILα 5.1 × 10-14 -0.48 0.36 0.84 5.1 × 10-10 -0.56 0.24 0.80 2 9,841,889

HPVC1 2.4 × 10-15 -0.48 0.33 0.81 1.3 × 10-13 -0.49 0.24 0.73 7 23,860,325TSS1500

MAP3K5 2.9 × 10-13 -0.48 0.39 0.87 NA NA NA NA 6 26,680,608Gene body

SPAG4L 1.4 × 10-11 -0.47 0.33 0.80 3.1 × 10-35 -0.59 0.29 0.88 20 2,510,8021st Exon, 5′UTR

CYP27A1 3.4 × 10-13 -0.47 0.24 0.71 2,930,667Gene Body

Hypermethylated CpG sites

ADHFE1 9.6 × 10-16 0.62 0.81 0.19 1.2 × 10-16 0.66 0.75 0.09 8 1,588,438TSS200

TFPI2 8.8 × 10-16 0.60 0.75 0.15 NA NA NA NA 7 16,934,178TSS200

ADHFE1 1.9 × 10-13 0.60 0.79 0.19 8.8 × 10-38 0.68 0.72 0.04 8 20,912,1695′UTR, 1st,

Exon

PRKAR1B 9.6 × 10-14 0.59 0.83 0.23 NA NA NA NA 7 18,601,1675′UTR, TSS200

TRBJ2–6 3.7 × 10-12 0.59 0.82 0.23 2.4 × 10-09 0.61 0.64 0.13 7 9,493,063CpG Island

PRKAR1B 7.3 × 10-14 0.58 0.80 0.27 NA NA NA NA 7 13,895,2355′UTR, TSS200

adj. to SOX-1 8.4 × 10-12 0.57 0.76 0.18 3.7 × 10-16 0.61 0.69 0.08 13 25,570,913CpG Island

ADHFE1 2.1 × 10-14 0.57 0.72 0.16 1.0 × 10-31 0.62 0.69 0.07 8 9,383,816TSS200

TFPI2 1.2 × 10-14 0.56 0.69 0.13 5.3 × 10-07 0.55 0.70 0.15 7 20,230,721Gene body

FLI1 4.0 × 10-10 0.56 0.74 0.18 2.6 × 10-23 0.63 0.70 0.07 11 11,017,065Gene

body,5′UTR

NPY 9.9 × 10-14 0.56 0.74 0.21 NA NA NA NA 7 16,964,348 TSS200

OPLAH 1.1 × 10-12 0.56 0.69 0.16 NA NA NA NA 8 17,698,295Gene body

ITGA4 9.9 × 10-14 0.56 0.62 0.09 NA NA NA NA 2 6,952,6715′UTR;1stExon

SND1 7.3 × 10-15 0.55 0.67 0.14 NA NA NA NA 7 9,296,001Gene body

Δβ, delta beta is the value of the differential methylation. Negative Δβ-values reflect hypomethylated status while positive Δβ an hypermethylated one.The adjusted p-value was considered significant when < 0.05 [27].adj.: Adjusted; Adj. to: Adjacent to gene from 5′ side; Body: Intragenic CpG sites; CGI: CpG island; CGI shore: Regions 2000 bp away from the CpG island; Chr: Chromosome number; CpG ID (Location): The coordinate of the CpG location according to the human genome build 37; NA: Data missing; RC: Rectal cancer.



The majority of differentially methylated CpG sites in our study were located in the gene body region. Intragenic DNA methylation may also affect the tran-scription from alternate promoters or the transcrip-tion of noncoding RNAs [58,59]. CpGs methylation in gene exons is a major cause of C to T transition muta-tions, leading to cancer causing mutations in somatic cells [60]. The functional role of intragenic DNA methylation needs further validation by expression analyses since the available data are conflicting [19]. For example, promoter methylation is inversely cor-related with expression, whereas methylation in the gene body is positively correlated with expression [58]. Thus, in mammals, it is the initiation of transcrip-tion but not transcription elongation that seems to be sensitive to DNA methylation silencing. This could be one of the reasons for lack of any inverse correla-tion of hypermethylated genes with their expression levels. However, the TCGA dataset contained very few paired ANMRM samples to go with the tumor tissues, which detracts from its usefulness to validate methylation/expression correlations.

During early stages of CRC or RC, epigenetic alterations appear to exceed the frequency of genetic mutations, suggesting their greater potential for the next generation of diagnostic biomarkers for the detection of increased risk of cancer transformation.

Our data may further contribute in understanding the role of aberrant methylation and other molecular mechanisms in RC pathogenesis. Collaborative efforts will ultimately result in the employment of epigeneti-cally based approaches to be commonly used to guide RC prevention and treatment. Limiting a study to a well-defined anatomical location such as the RC may reduce the noise levels in the array data studies and thereby increase the rate of successful identification of novel epigenetic biomarkers.

In conclusion, our large and sufficiently powered clinical study with independent external validation has demonstrated the feasibility of using specific methylated DNA signatures for developing putative diagnostic biomarkers in RC.

Future perspectiveOur data contribute to improved understanding of the role of gene-specific aberrant methylation in rec-tal cancer pathogenesis. Treating rectal cancer as an independent entity may improve discovery of bio-markers used for early detection and prognosis. In the future, new biomarker genes will be established and their association with patients’ survival will be addressed. The current study contributes to the estab-lishment of such new biomarkers, with the identifica-tion of BPIL3, HBBP1, TIFPI2, ADHFE1, FLI1 and TLX1 genes.

Supplementary dataTo view the supplementary data that accompany this paper

please visit the journal website at: www.futuremedicine.

com/doi/full/10.2217/epi-2016-0044

AcknowledgementsThe authors are very thankful to E Van Emburgh for his

technical support.

Financial & competing interests disclosureMethylation profiling was performed by the SNP&SEQ Tech-

nology Platform in Uppsala. The platform is part of Science

for Life Laboratory at Uppsala University and supported as

a national infrastructure by the Swedish Research Coun-

cil. This work was supported by the Internal Grant Agency

of the Czech Ministry of Health (NT 13424, NT/14329-3),

Czech Science Foundation (15-27580A; GA15-08239S),

COST LD14050 and by the European Regional Develop-

ment Fund (project number CZ.1.05/2.1.00/03.0076), Lions

Cancer Foundation and Nyckelfonden, Örebro läns landstin.

This study was also supported by the National Sustainability

Program I (NPU I) Nr. LO1503 provided by the Ministry of

Education Youth and Sports of the Czech Republic. The au-

thors have no other relevant affiliations or financial involve-

ment with any organization or entity with a financial interest

in or financial conflict with the subject matter or materials

discussed in the manuscript apart from those disclosed.


manuscript.

Executive summary

• Rectal cancer (RC) comprises about a third of all colorectal cancer cases and the location of RC makes it difficult to perform operative resection; therefore irradiation is used as an alternative treatment.

• The DNA methylation profile of 32 pairs of RC and adjacent nonmalignant rectal mucosa’s showed that majority of the CpG sites are hypermethylated in RC.

• The BPIL3 and HBBP1 genes were hypomethylated in rectal cancer, whereas TIFPI2, ADHFE1, FLI1 and TLX1 genes were hypermethylated.

• Cross-validation against TCGA methylation dataset showed the reliability and reproducibility of the methylation differences identified in our array study and confirmed the high discriminative and noncoincidental potential of the selected biomarkers.



References1 Haggar FA, Boushey RP. Colorectal cancer epidemiology:

incidence, mortality, survival, and risk factors. Clin. Colon Rectal Surg. 22(4), 191–197 (2009).

2 Comprehensive molecular characterization of human colon and rectal cancer. Nature 487(7407), 330–337 (2012).

3 Wolpin BM, Meyerhardt JA, Mamon HJ, Mayer RJ. Adjuvant treatment of colorectal cancer. CA Cancer J. Clin. 57(3), 168–185 (2007).

4 Minsky BD. Counterpoint: long-course chemoradiation is preferable in the neoadjuvant treatment of rectal cancer. Semin. Radiat. Oncol. 21(3), 228–233 (2011).

5 Zlobec I, Vuong T, Hayashi S et al. A simple and reproducible scoring system for EGFR in colorectal cancer: application to prognosis and prediction of response to preoperative brachytherapy. Br. J. Cancer 96(5), 793–800 (2007).

6 Wei EK, Giovannucci E, Wu K et al. Comparison of risk factors for colon and rectal cancer. Int. J. Cancer 108(3), 433–442 (2004).

7 Colditz GA, Cannuscio CC, Frazier AL. Physical activity and reduced risk of colon cancer: implications for prevention. Cancer Causes Control 8(4), 649–667 (1997).

8 Servomaa K, Kiuru A, Kosma VM, Hirvikoski P, Rytomaa T. p53 and K-ras gene mutations in carcinoma of the rectum among Finnish women. Mol. Pathol. 53(1), 24–30 (2000).

9 Sharma S, Kelly TK, Jones PA. Epigenetics in cancer. Carcinogenesis 31(1), 27–36 (2010).

10 Wilson AS, Power BE, Molloy PL. DNA hypomethylation and human diseases. Biochim. Biophys. Acta 1775(1), 138–162 (2007).

11 Esteller M. CpG island hypermethylation and tumor suppressor genes: a booming present, a brighter future. Oncogene 21(35), 5427–5440 (2002).

12 Pfeifer GP, Rauch TA. DNA methylation patterns in lung carcinomas. Semin. Cancer Biol. 19(3), 181–187 (2009).

13 Farkas SA, Vymetalkova V, Vodickova L, Vodicka P, Nilsson TK. DNA methylation changes in genes frequently mutated in sporadic colorectal cancer and in the DNA repair and Wnt/beta-catenin signaling pathway genes. Epigenomics 6(2), 179–191 (2014).

14 Lindqvist BM, Wingren S, Motlagh PB, Nilsson TK. Whole genome DNA methylation signature of HER2-positive breast cancer. Epigenetics 9(8), 1149–1162 (2014).

15 Naumov VA, Generozov EV, Zaharjevskaya NB et al. Genome-scale analysis of DNA methylation in colorectal cancer using Infinium HumanMethylation450 BeadChips. Epigenetics 8(9), 921–934 (2013).

16 The Cancer Genome Atlas Research Network (2012). http://cancergenome.nih.gov

17 Yousefi P, Huen K, Aguilar Schall R et al. Considerations for normalization of DNA methylation data by Illumina 450K BeadChip assay in population studies. Epigenetics 8(11), 1141–1152 (2013).

18 Sandoval J, Heyn H, Moran S et al. Validation of a DNA methylation microarray for 450,000 CpG sites in the human genome. Epigenetics 6(6), 692–702 (2011).

19 Nordlund J, Backlin Cl, Wahlberg P et al. Genome-wide signatures of differential DNA methylation in pediatric acute lymphoblastic leukemia. Genome Biol. 14(9), r105 (2013).

20 Bibikova M, Barnes B, Tsan C et al. High density DNA methylation array with single CpG site resolution. Genomics 98(4), 288–295 (2011).

21 Biomers.net. www.biomers.net

22 Wanderer. http://maplab.imppc.org/wanderer/#

23 Wang X, Kang DD, Shen K et al. An R package suite for microarray meta-analysis in quality control, differentially expressed gene analysis and pathway enrichment detection. Bioinformatics 28(19), 2534–2536 (2012).

24 Bioconductor. www.bioconductor.org

25 Smyth GK. Linear models and empirical bayes methods for assessing differential expression in microarray experiments. Stat. Appl. Genet. Mol. Biol. 3, Article3 (2004).

26 Benjamini Y, Hochberg Y. Controlling the false discovery rate – a practical and powerful approach to multiple testing. J. R. Stat. Soc. Series B (Methodological) 57(1), 289–300 (1995).

27 The Cancer Genome Atlas: TCGA Data Portal. https://tcga-data.nci.nih.gov/tcga/

28 Jass JR. Gastrointestinal polyposes: clinical, pathological and molecular features. Gastroenterol. Clin. North Am. 36(4), 927–946, viii (2007).

29 Iacopetta B. Are there two sides to colorectal cancer? Int. J. Cancer 101(5), 403–408 (2002).

30 Beggs AD, Jones A, El-Bahrawy M, Abulafi M, Hodgson SV, Tomlinson IP. Whole-genome methylation analysis of benign and malignant colorectal tumours. J. Pathol. 229(5), 697–704 (2013).

31 Zhang FF, Cardarelli R, Carroll J et al. Significant differences in global genomic DNA methylation by gender and race/ethnicity in peripheral blood. Epigenetics 6(5), 623–629 (2011).

32 Zhang FF, Santella RM, Wolff M, Kappil MA, Markowitz SB, Morabia A. White blood cell global methylation and IL-6 promoter methylation in association with diet and lifestyle risk factors in a cancer-free population. Epigenetics 7(6), 606–614 (2012).

33 Delgado-Cruzata L, Zhang W, Mcdonald JA et al. Dietary modifications, weight loss, and changes in metabolic markers affect global DNA methylation in Hispanic, African–American, and Afro-Caribbean breast cancer survivors. J. Nutr. 145(4), 783–790 (2015).

34 Kardon T, Noel G, Vertommen D, Schaftingen EV. Identification of the gene encoding hydroxyacid-oxoacid transhydrogenase, an enzyme that metabolizes 4-hydroxybutyrate. FEBS Lett. 580(9), 2347–2350 (2006).

35 Kim YH, Lee HC, Kim SY et al. Epigenomic analysis of aberrantly methylated genes in colorectal cancer identifies genes commonly affected by epigenetic alterations. Ann. Surg. Oncol. 18(8), 2338–2347 (2007).

36 Oster B, Thorsen K, Lamy P et al. Identification and validation of highly frequent CpG island hypermethylation

http://cancergenome.nih.gov/

www.biomers.net

http://maplab.imppc.org/wanderer/#

www.bioconductor.org

https://tcga-data.nci.nih.gov/tcga/



in colorectal adenomas and carcinomas. Int. J. Cancer 129(12), 2855–2866 (2011).

37 Kim JY, Tillison KS, Zhou S, Lee JH, Smas CM. Differentiation-dependent expression of Adhfe1 in adipogenesis. Arch. Biochem. Biophys. 464(1), 100–111 (2007).

38 Lyon RC, Johnston SM, Panopoulos A, Alzeer S, McGarvie G, Ellis EM. Enzymes involved in the metabolism of gamma-hydroxybutyrate in SH-SY5Y cells: identification of an iron-dependent alcohol dehydrogenase ADHFe1. Chem. Biol. Interact. 178(1–3), 283–287 (2009).

39 Hibi K, Goto T, Kitamura YH et al. Methylation of the TFPI2 gene is frequently detected in advanced gastric carcinoma. Anticancer Res. 30(10), 4131–4133 (2010).

40 Wang S, Xiao X, Zhou X et al. TFPI-2 is a putative tumor suppressor gene frequently inactivated by promoter hypermethylation in nasopharyngeal carcinoma. BMC Cancer 10, 617 (2010).

41 Ashktorab H, Rahi H, Wansley D et al. Toward a comprehensive and systematic methylome signature in colorectal cancers. Epigenetics 8(8), 807–815 (2013).

42 Exner R, Pulverer W, Diem M et al. Potential of DNA methylation in rectal cancer as diagnostic and prognostic biomarkers. Br. J. Cancer 113(7), 1035–1045 (2015).

43 Wong CM, Ng YL, Lee JM et al. Tissue factor pathway inhibitor-2 as a frequently silenced tumor suppressor gene in hepatocellular carcinoma. Hepatology 45(5), 1129–1138 (2007).

44 Sparmann A, Van Lohuizen M. Polycomb silencers control cell fate, development and cancer. Nat. Rev. Cancer 6(11), 846–856 (2006).

45 Glockner SC, Dhir M, Yi JM et al. Methylation of TFPI2 in stool DNA: a potential novel biomarker for the detection of colorectal cancer. Cancer Res. 69(11), 4691–4699 (2009).

46 Moon JW, Lee SK, Lee JO et al. Identification of novel hypermethylated genes and demethylating effect of vincristine in colorectal cancer. J. Exp. Clin. Cancer Res. 33, 4 (2014).

47 Lin PC, Lin JK, Lin CH et al. Clinical relevance of plasma DNA methylation in colorectal cancer patients identified by using a genome-wide high-resolution array. Ann. Surg. Oncol. 22(Suppl. 3), 1419–1427 (2015).

48 Heller G, Babinsky VN, Ziegler B et al. Genome-wide CpG island methylation analyses in non-small cell lung cancer patients. Carcinogenesis 34(3), 513–521 (2013).

49 Li HP, Peng CC, Chung IC et al. Aberrantly hypermethylated Homeobox A2 derepresses

metalloproteinase-9 through TBP and promotes invasion in Nasopharyngeal carcinoma. Oncotarget 4(11), 2154–2165 (2013).

50 Shu Y, Wang B, Wang J, Wang JM, Zou SQ. Identification of methylation profile of HOX genes in extrahepatic cholangiocarcinoma. World J. Gastroenterol. 17(29), 3407–3419 (2011).

51 Chiang JH, Cheng WS, Hood L, Tian Q. An epigenetic biomarker panel for glioblastoma multiforme personalized medicine through DNA methylation analysis of human embryonic stem cell-like signature. OMICS 18(5), 310–323 (2014).

52 Kamalakaran S, Varadan V, Giercksky Russnes HE et al. DNA methylation patterns in luminal breast cancers differ from non-luminal subtypes and can identify relapse risk independent of other clinical variables. Mol. Oncol. 5(1), 77–92 (2011).

53 Lin YL, Wang YL, Fu XL, Ma JG. Aberrant methylation of PCDH8 is a potential prognostic biomarker for patients with clear cell renal cell carcinoma. Med. Sci. Monit. 20, 2380–2385 (2014).

54 Lin YlL, Ma JH, Luo XL, Guan TY, Li ZG. Clinical significance of protocadherin-8 (PCDH8) promoter methylation in bladder cancer. J. Int. Med. Res. 41(1), 48–54 (2014).

55 Zhang D, Zhao W, Liao X, Bi T, Li H, Che X. Frequent silencing of protocadherin 8 by promoter methylation, a candidate tumor suppressor for human gastric cancer. Oncol. Rep. 28(5), 1785–1791 (2012).

56 Tommasi S, Karm DL, Wu X, Yen Y, Pfeifer GP. Methylation of homeobox genes is a frequent and early epigenetic event in breast cancer. Breast Cancer Res. 11(1), R14 (2009).

57 Wong KY, Yim RL, Kwong YL et al. Epigenetic inactivation of the MIR129-2 in hematological malignancies. J. Hematol. Oncol. 6, 16 (2013).

58 Jones PA. Functions of DNA methylation: islands, start sites, gene bodies and beyond. Nat. Rev. Genet. 13(7), 484–492 (2012).

59 Kulis M, Queiros AC, Beekman R, Martin-Subero JI. Intragenic DNA methylation in transcriptional regulation, normal differentiation and cancer. Biochim. Biophys. Acta 1829(11), 1161–1174 (2013).

60 Rideout WM 3rd, Coetzee GA, Olumi AF, Jones PA. 5-methylcytosine as an endogenous mutagen in the human LDL receptor and p53 genes. Science 249(4974), 1288–1290 (1990).


part of

Review

10.2217/epi.16.6 © 2016 Future Medicine Ltd

Epigenomics

Review 2016/04/308

5

2016

Neuroendocrine prostate cancer (NEPC) is the most lethal prostatic neoplasm. NEPC is thought to originate from the transdifferentiation of AR-positive adenocarcinoma cells. We have previously shown that an epigenetic/noncoding interactome (ENI) orchestrates cancer cells’ plasticity, thereby allowing the emergence of metastatic, drug-resistant neoplasms. The primary objective of this manuscript is to discuss evidence indicating that some components of the ENI (Polycomb genes, miRNAs) play a key role in NEPC initiation and progression. Long noncoding RNAs represent vast and largely unexplored component of the ENI. Their role in NEPC has not been investigated. We show preliminary evidence indicating that a lncRNA (MIAT) is selectively upregulated in NEPCs and might interact with Polycomb genes. Our results indicate that long noncoding RNAs can be exploited as new biomarkers and therapeutic targets for NEPC.

First draft submitted: 9 October 2015; Accepted for publication: 17 February 2016; Published online: 20 April 2016

Keywords: epigenetic/noncoding interactome • long noncoding RNAs • MIAT • neuroendocrine prostate cancer • Polycomb • transdifferentiation

Neuroendocrine prostate cancer: clinical & molecular featuresIn adult males, the prostate is a small acorn-shaped tissue with ductal-acinar histology surrounding the urethra at the base of the bladder. Its main function is to contribute secretory proteins to the seminal fluid [1]. The adult prostate is a pseudo-stratified epithelium composed of three main cell lineages (Figure 1, left panel):

• Secretory luminal cells are the predomi-nant cell type; these cells express keratins (K8, K18), the androgen receptor (AR) and secretory proteins such as prostate-specific antigen (PSA) and prostatic specific acid phosphatase (PSAP);

• Basal cells expressing K5 and K14 kera-tins and p63 are the second major cell type;

• Neuroendocrine cells (NEC) expressing chromogranin A (CHGA), synaptophy-sin (SYP) and neuropeptides are scat-tered throughout the basal layer and com-prise less than 1% of normal prostatic glandular epithelium [1–3].

Prostate cancer (PCa) represents the second most frequently diagnosed neoplasm and is the sixth leading cause of cancer-related deaths in males worldwide [4,5]. In keeping with the composition of prostate epithelium, more than 95% of PCas are classified as adenocar-cinomas, which show luminal phenotype and AR expression (Figure 1, middle panel) [6]. Endogenous androgens, mainly produced by the testis, bind to the AR and fuel prostate adenocarcinoma proliferation [7]. For this reason, androgen-deprivation therapy (also known as castration) is an effective therapeu-tic strategy for this disease. However, patients

The role of epigenetics and long noncoding RNA MIAT in neuroendocrine prostate cancer

Francesco Crea1,2, Erik Venalainen1, Xinpei Ci1,3, Hongwei Cheng1,3, Larissa Pikor4, Abhijit Parolia1, Hui Xue1, Nur Ridzwan Nur Saidy1, Dong Lin1,3, Wan Lam4, Colin Collins3 & Yuzhuo Wang*,1,3

1Experimental Therapeutics, BC Cancer

Agency Cancer Research Centre,

Vancouver, BC, Canada 2Department of Life Health & Chemical

Sciences, The Open University, Milton

Keynes, UK 3Vancouver Prostate Centre, Department

of Urologic Sciences, University of British

Columbia, Vancouver, BC, Canada 4Genetics Unit, Integrative Oncology,

BC Cancer Agency Cancer Research

Centre, Vancouver, BC, Canada


[email protected]



Figure 1. Evolution of prostatic neoplasms. Left panel: the normal prostate epithelium and its components. Middle panel: emergence of AR+ prostate cancer (purple nuclei, neoplastic cells; blue nuclei, normal cells). Right panel: NEPC transdifferentiation. The putative role of PcGs, miRNAs and lncRNAs in NEPC transdifferentiation is summarized. AR: Androgen receptor; lncRNA: Long noncoding RNA; NE: Neuroendocrine; NEPC: Neuroendocrine prostate cancer.

Normal prostate AR+ prostate cancer AR- NEPC

Luminal cell

Basal cell

Neuroendocrine cell

Neoplastic cells

1) Polycomb genes →Metastasis/drug resistanceand neural cell differentiation

2) lncRNAs →Chromatin modification and tumor suppressor silencing

3) miRNAs →Activation of NE gene programs and promoting metastasis + drug resistance

1

2

3


Review Crea, Venalainen, Ci et al.

invariably relapse despite castrate androgen levels (cas-tration-resistant PCa, CRPC) mainly via genetic and epigenetic alterations that facilitate ligand-independent AR activation, amplify the AR-dependent signaling or trigger different proliferative pathways [7]. CRPCs are characterized by substantially worse prognoses, but chemotherapeutics and newly approved hormonal treatments (e.g., enzalutamide [8] and abiraterone [9]) are still effective in prolonging patients’ survival at this stage.

Between 0.5 and 2% of newly diagnosed pros-tatic neoplasms are classified as neuroendocrine PCa (NEPC), which is insensitive to all forms of hormonal treatment [10]. The neuroendocrine phenotype is signif-icantly associated with lower AR expression [11]. How-ever, some reports indicate that AR expression might be retained in a relevant fraction of NEPCs [12,13]. NEPC is characterized by positive immunohisto-chemical (IHC) staining for CHGA, SYP and neu-ron-specific enolase. However, sparse NEPC cells are not immediately identifiable on IHC sections [3,14–15]. Furthermore, NEPC patients do not present elevated circulating PSA and PSAP levels. These two markers are important indexes to assess the potential presence, or monitor the progression of PCa [6,16]. As a result of these peculiarities, NEPC is often diagnosed at a metastatic stage [17]. In addition, no treatment has

demonstrated efficacy in extending the survival of NEPC patients. While median prostate adenocarci-noma survival is 125 months, median NEPC survival is only 7 months [18,19]. Hence, the phenotypic distinc-tion between NEPC and adenocarcinoma is extremely important from a clinical perspective.

There are two prevalent theories regarding the cel-lular origin of NEPC. One model hypothesizes that NEPCs originate from the transformation of prostate NECs that share a common origin with the luminal and basal prostatic cells. This model is based on the observation that PCa is often multifocal and that NECs are a normal component of the prostatic epithe-lium [10,20–21]. According to this model, environmental stress (e.g., androgen deprivation) favors the survival of these more proliferative and AR-negative NEPC cells. In keeping with this hypothesis, SV40 T-Ag express-ing NECs are able to generate NEPCs in a murine transgenic model [22]. Despite this convincing ratio-nale, the experimental and clinical evidence in support of this model is still limited [23]. Currently available evidence seems to favor another paradigm: under spe-cific conditions, adenocarcinoma cells acquire NEC markers and lose AR expression thereby transdiffer-entiating into NEPC cells (Figure 1, right panel). The mechanism by which adenocarcinoma cells acquire the NEPC phenotype is still not fully elucidated [24].


The role of epigenetics & long noncoding RNA MIAT in neuroendocrine prostate cancer Review

One model suggests that NEPC could originate from luminal cells expressing neuroendocrine (NE) genes, potentially due to input from surrounding NECs. NE transdifferentiation is primarily a mechanism of adaptive response/tumor resistance [10,25]. In vitro data demonstrate that LNCaP cells can be induced to transdifferentiate into NEPC cells by various stimuli such as androgen depletion, or supplementation with cAMP, cytokines, or growth factors [21]. More recently, we reported that a patient-derived prostate adenocarci-noma xenograft model (LTL331) developed complete NEPC relapse (LTL331R) upon castration. Nota-bly, the original hormone-sensitive adenocarcinoma and the derived NEPC exhibited matching genetic profiles [26]. Furthermore, an analysis of ERG rear-rangement and TP53 status in clinical samples with mixed NEPC/adenocarcinoma phenotype suggested a single clonal origin for the two PCa subtypes, thereby supporting the transdifferentiation model [13,27–31].

Regardless of its cellular origin, NEPC will likely become a major clinical issue in the near future. Although the diagnosis of de novo NEPC is rare, sparse NEPC clones often coexist with more abundant ade-nocarcinoma cells. Upon repeated cycles of hormonal therapy, NEPC clones can become the dominant pop-ulation [10,26,32]. AR-negative NEPC cells easily adapt to androgen deprivation and are highly proliferative (more than 50% of tumor cells in NEPCs are posi-tive for Ki67 IHC staining) [1,21,26]. It is therefore not surprising that NEPC cells replace adenocarcinoma cells after prolonged AR signaling suppression. It has been suggested that the emergence of more potent hormonal therapies (enzalutamide, abiraterone) might increase NEPC incidence [33]. Hence, a priority of future research will be to identify the molecular mech-anisms underlying NEPC emergence, and to identify viable therapeutic targets to prevent or at least delay the development of this incurable disease.

The epigenetic/noncoding interactome & its implication in the initiation & progression of neuroendocrine transdifferentiationAs highlighted in the previous paragraph, obtain-ing a more unified understanding of the molecular mechanisms that drive NEPC progression will enable us to identify novel therapeutic tools for this lethal disease. NEPC progression is often associated with genetic alterations including AR inactivation, the loss of specific tumor suppressors (RB1, PTEN, TP53), TMPRSS2–ERG rearrangement, and amplification of MYCN and AURKA oncogenes [21,34–35]. Although the roles of these irreversible genetic events have been well characterized, no effective targeted treatment has been developed so far. Emerging evidence indi-

cates that an epigenetic/noncoding interactome (ENI) could play a more fundamental function in NEPC ini-tiation and progression. We have previously proposed that the ENI confers unique plasticity to cancer cells, thereby allowing them to become metastatic and drug-resistant [28,36]. Notably, these two deadly features are hallmarks of NEPC. Here, we will discuss initial evi-dence suggesting that the ENI is implicated in NEPC initiation.

The ENI is comprised of two major components: epigenetic effectors (proteins) and noncoding RNAs (ncRNAs) [36]. Polycomb group (PcG) proteins are epigenetic effectors organized in multimeric com-plexes known as the Polycomb repressive complexes (PRCs) [37]. The two main PRCs (PRC1 and PRC2) act in concert to silence gene transcription. PRC2 functions to trimethylate H3K27me3 in the pro-moter region of a target gene, thus creating a repres-sive chromatin mark [38]. This histone modification is subsequently recognized by the chromodomain of the CBX Polycomb proteins (CBX2,4,6,7,8) [39] which facilitate the recruitment of the PRC1 to the chroma-tin [40]. PRC1 then monoubiquitylates H2AK119ub1 via its catalytic ligases RING1a and RING1b [41], thereby silencing transcription at target sites [42]. To date, many studies have shown that over-expression of the PcG proteins EZH2 and BMI1 facilitates metas-tasis in several cancers [43–45]. In addition, upstream and downstream miRNAs interact with EZH2 func-tion to promote drug resistance [46–48]. Taken together, these findings indicate the involvement of PcG pro-teins in two hallmarks of NEPC: metastasis and drug resistance. PcG proteins are also known to regulate stem cell differentiation and neurogenesis [49,50]. Loss of EZH2 in neuronal progenitor cells led to reduced proliferation and survival [51]. This evidence indicates that dysregulated PcG-mediated repression could play a role in neuroendocrine transdifferentiation (Figure 1, right panel).

Recently, we successfully developed the first-in-field patient tissue-derived xenograft model of com-plete NEPC transdifferentiation from prostate ade-nocarcinoma [36,52]. To identify the mechanisms of NEPC initiation, we conducted transcriptomic and genomic analyses on our ADT-induced NEPC model (LTL-331R) and on its hormone-sensitive predeces-sor (LTL-331). We found that the two models share identical genetic profiles, suggesting that genetic alterations may not exclusively drive NEPC transdif-ferentiation [36]. Interestingly, our analysis revealed that CBX2 and EZH2 (PcG members) were signifi-cantly upregulated in NEPC preclinical models and clinical samples [26]. This study also identified 185 PcG target genes that were significantly downregulated


Fold change

110100

1000

1000

0

100,

000

GEO value

0

500

1000

1500

Fea

ture

FA

M72

AM

IAT

AX

7471

04S

MP

D3

Adenocarcinoma

NEPC

Adenocarcinoma

NEPC

Adenocarcinoma

NEPC

Adenocarcinoma

NEPC

Adenocarcinoma

NEPC

Bone marrow

Liver

Heart

Spleen

Lung

Kidney

Skeletal muscle

Thymus

Brain

Spinal cord

Prostate

Pancreas

x2621x3023x2525x3042x3043x3051x3027x1783x2741x97 Tx3134x2740x2682x3071x2849x2620x2832x2858x3035x2743x2872x3050x3132x3026x3034x2761x3048x2844x7520x4240x7800x8740x8220x7820x7821

DB62AH51B56B51B57BB

B58AAB51DDBCB53DDB51B51B56C51DB50DC3

sc PT76

-3.0

-2.5

-1.00.0

1.5

2.0

Log 2 median-centered intensity

3.5

2.5

3.0

0.5

1.0

-0.5

-2.0

-1.5

12***

Pri

mar

yM

etas

tati

c

-3.0

-2.5

-0.02.0

4.5

5.0

Log 2 median-centered ratio

8.0

5.5

6.0

6.5

7.0

7.5

2.5

3.0

3.5

4.0

0.5

1.0

1.5

-2.0

-1.5

-1.0

-0.5

12****

Rb

w.t

.R

b m

uta

ted

Co

nce

pt

p-v

alu

eO

dd

s ra

tio

Pat

ien

ts (

n)

Rec

urre

nce

at5

year

s1.

0E-1

32.

261

Rec

urre

nce

at

1yea

r6.

5E-4

3.2

173

ET

S fa

mily

fusi

on1.

6 E

-82.

258

Met

asta

tic P

Ca,

AR

am

pli

catio

n1.

6 E

-85.

531

Met

asta

tic P

Ca,

ET

S2

dele

tion

1.6

E-8

5.5

35

Met

asta

tic P

Ca,

ER

Gre

arra

ngm

ent

6.2E

-43.

335

MIA

T PC

AT

18

PC

GE

M1

MA

LAT

1P

CA

3

****

**




Figure 2. Identification of MIAT as a NEPC-specific lncRNA (see facing page). (A) Expression of MIAT and four other prostate cancer-associated lncRNA in 18 prostatic adenocarcinoma and three NEPC models established and maintained at the Living Tumor Laboratory [95]. Each transplantable model has been characterized by clinico-pathological examination. The expression of adenocarcinoma- (PSA, AR) and NEPC- (chromogranin A and synaptophysin) associated genes was assessed by immunohistochemistry (as described in Supplementary Figure 1 and in [29]). Gene expression is calculated based on microarray data described in [6]. p = 0.001 (unpaired, 2-tailed t-test). (B) Unsupervised hierarchical clustering discriminates six NEPC from all other prostate cancer samples, based on gene expression profiles (data taken from [1]). MIAT was included in the list of genes that discriminate the two prostate cancer subtypes (1886 genes upregulated in NEPC and 1010 genes upregulated in Adenocarcinoma). The six NEPC samples are on the right side of the dendrogram. Red means upregulation, blue means downregulation. (C) Expression of MIAT in a collection of non-neoplastic human tissues (GEO profile ID: 2896847, two samples per tissue). p < 0.0001 versus all other tissues (ANOVA and Dunnet’s post-test). (D) MIAT expression in primary (7) versus metastatic (6) prostate cancer samples. Fold change (FC): 7.9, p < 0.001. (E) MIAT expression in Rb wild-type (27) versus Rb mutated (2) prostate cancer samples. FC: 109.8, p < 0.0001. (F) Genes significantly associated with MIAT were uploaded to the Oncomine database to identify clinically relevant correlations in prostate cancer samples. SAM was performed as described in [5]. Genes were considered positively associated with MIAT if they displayed FC >2.0 and q < 0.001. We ranked the positively associated genes based on FC and uploaded the first 1000 to the Oncomine database. Oncomine software (Life Technology) was used for analysis and visualization in D–F. AR: Androgen receptor; ETS: GEO: Gene-expression omnibus; NEPC: Neuroendocrine prostate cancer; PCa: Prostate cancer.



indicating a relevant role of PcG complexes in NEPC. This ‘neuroendocrine-associated repression signature’ (NEARS) is associated with higher-grade neoplasms, metastatic progression and poor outcome in multiple clinical datasets [26]. In line with this model, we also found that the chromatin modifier DEK is upregulated in NEPC cells, and that targeting this gene reduces NEPC proliferation and migration [53]. Notably, PcG-targeting drugs are being developed and have been suc-cessfully tested in PCa preclinical models [45]. Hence, the deregulated expression of epigenetic effectors may offer viable drug targets for NEPC.

While some epigenetic effectors (e.g., PcGs) are hyper-activated during NEPC progression, others might be suppressed. The loss of the REST gene in CRPC promotes NEPC development [12]. REST is part of the KDM1A-coREST-REST histone modi-fying complex which is bound by HOTAIR, a long intergenic ncRNA that coordinates histone H3 lysine 27 methylation and lysine 4 demethylation [54]. Given that REST is commonly inactivated in NEPC and is responsible for repressing neuronal genes [55], aber-rant silencing of this gene could trigger neuronal dif-ferentiation programs in transdifferentiating cells [56]. Notably, epigenetic modifications are known to pre-cede genetic alterations in human neoplasms [57]. Therefore, we believe that the ENI plays a significant role in the initiating stages of NEPC transdifferentia-tion via epigenetic modification of downstream gene targets. These reversible epigenetic changes can in turn promote cellular plasticity and allow for more flexible adaptation to extreme conditions, including those associated with drug resistance and metastatic potential of NEPC.

The second crucial component of the ENI is repre-sented by ncRNAs [36]. Recent advancements in tran-

scriptome analysis support the notion that although approximately 90% of the genome is actively tran-scribed, only 2% of it encodes for proteins [58,59]. The remaining RNA molecules produced by the cells have been long considered transcriptional noise, which lacks relevant cellular functions [60,61]. More recently, a mul-titude of experimental studies have identified regula-tory ncRNAs that play functional roles in mammalian cells [62–64] and in cancer progression [65]. The category of regulatory ncRNAs includes long noncoding RNAs (lncRNAs) and miRNAs, both of which have been implicated in facilitating cancer metastasis [66,67] and drug resistance [68–70]. Emerging evidence suggests that ncRNAs might interact with epigenetic effectors to drive NEPC initiation and progression (Figure 1, right panel).

MiRNAs are small ncRNAs (<200 bp) approxi-mately 22 nt in length, and are produced by two RNase III proteins known as Drosha and Dicer [71]. The main function of miRNAs is to repress the trans-lation of proteins by binding to the 3′ untranslated region of their complementary (target) messenger RNA molecules. This action is completed via integration of mature miRNAs into the microRNA-induced silencing complex (miRISC or RISC) [72]. The role of miRNAs in cancer progression has been described before [73]. Ini-tial evidence suggests that miRNAs interact with epi-genetic effectors to drive NEPC initiation. For exam-ple, miR-124 is upregulated in response to loss of REST gene [55], a phenomenon observed in 50% of NEPC tumors [74]. Notably, miR-124 represses BAF53a, a chromatin remodeling protein that is essential for sup-pressing neuronal differentiation [75]. Therefore, loss of REST and subsequent upregulation of miR-124 could facilitate the activation of proneural genes in NEPC transdifferentiating cells. Other miRNAs could play



a multifaceted role in NEPC initiation and progres-sion. Let-7b targets the nuclear receptor TLX, thereby promoting neuronal differentiation programs [76]. On the other hand, TLX is over-expressed in high-grade PCa tissues and PCa cell lines, thereby promoting cell growth and inhibiting PTEN-induced senescence [77]. In addition, let-7b downregulation has been shown to facilitate the metastatic activity of established neuroen-docrine tumors [78]. Taken together, these data indicate that let-7b-mediated TLX silencing may constitute an initiating event of NE transdifferentiation, and that subsequent downregulation of this miRNA could trig-ger further NEPC development. Therefore, activation of neuronal genes potentially involved in NEPC trans-differentiation may be orchestrated by reversible and plastic mechanisms, which ensure timely activation and deactivation of specific genetic programs.

Long ncRNAs (lncRNAs) are often described as noncoding transcripts longer than 200 bp. Alike protein-coding genes, lncRNA expression is regu-lated by histone-post translational modifications and DNA methylation [79]. LncRNAs have been shown to modulate transcriptional programs by functioning as molecular scaffolds that target histone-modifying complexes to specific loci [54]. In keeping with this model, lncRNAs regulate histone methylation, inter-act with chromatin modifying proteins and influ-ence local gene expression via DNA-binding [80–82]. Examples of lncRNAs associated with chromatin modifying complexes include HOTAIR, AIR and Kcnq1ot1 [83–85].

As a result of their properties, lncRNAs can act in concert with the histone modifying complexes to repress transcription of potentially onco-suppressive genes (Figure 1, right panel). The lncRNAs PTENP1 and GAS5 have been identified as regulators of the tumor suppressor PTEN [86,87], whose loss is com-monly implicated in NEPC. In addition, the lncRNA H19 was previously shown to downregulate expres-sion of pRb in colorectal cancer cells [88]. Although the aforementioned lncRNAs have not been studied in NEPC, it is conceivable that some of them are impli-cated in the silencing of specific onco-suppressors dur-ing NEPC transdifferentiation.

Another key characteristic of NEPC is its elevated metastatic potential. There is mounting evidence to support the involvement of lncRNAs (e.g., SChLAP1, PCAT1 and MALAT1) in PCa invasion and metas-tasis [89–91]. In light of this evidence, it is likely that lncRNAs contribute to NEPC progression by promot-ing metastatic dissemination. Henceforth, lncRNAs may play a valuable and largely unexplored role in NE transdifferentiation and evolution. No direct evidence so far has supported this hypothesis. In the next para-

graph, we will discuss preliminary data on a lncRNA that seems to be specifically expressed in NEPC cells.

Discovery of MIAT as a NEPC-associated lncRNAIn the previous section, we have shown evidence sug-gesting that the ENI plays a key role in NEPC initia-tion and progression. Since the ENI mediates revers-ible changes in gene expression, its components are ideal drug targets [28,92]. While the importance of epigenetic effectors in NEPC is emerging, the role of ncRNAs in this disease is still largely unknown. LncRNAs represent a vast portion of our transcrip-tome (more than 50,000 unique sequences [93]) and have been described as a ‘gold mine’ for the discovery of new biomarkers and therapeutic targets [65]. To gain insights into the role of lncRNAs in NEPC, we ana-lyzed our collection of patient-derived PCa xenografts, searching for lncRNAs associated with NEPC. The parental adenocarcinoma (LTL331) and the relapsed NEPC line (LTL331R) described in the previous sec-tion, have been profiled though Agilent one-color microarrays. This platform includes 2497 lncRNA probes. The same platform has been used to profile our unique collection of patient-derived PCa xenografts, which includes androgen-dependent and independent adenocarcinomas as well as additional NEPC models (Supplementary Figure 1) [36].

In order to discover potentially relevant lncRNAs, we first identified transcripts showing a greater than twofold upregulation in LTL331R (NEPC) versus LTL331 (adenocarcinoma), and then cross-validated these against the lncRNAs specifically upregulated in a previously described cohort of NEPC cases [28]. We ranked those NEPC-specific transcripts based on their differential expression. The most highly upregulated transcript in this list was MIAT-Gomafu, a lncRNA previously described for its role in neural cell activa-tion [94]. Microarray analysis revealed that MIAT is exclusively expressed in NEPC patient-derived mod-els (Figure 2A, NEPC vs adenocarcinoma, difference between means = 6871 ± 1783; p = 0.001). Notably, the expression pattern of MIAT is unique when com-pared with other PCa-associated lncRNAs represented in the array (Figure 2A). Dramatic MIAT upregulation in LTL331R versus LTL331 was confirmed by qPCR (Supplementary Figure 2A). RNA fractionation experi-ments revealed that MIAT expression is restricted to the nucleus of NEPC cells (Supplementary Figure 2B). Unsupervised hierarchical clustering demonstrated that MIAT expression can efficiently discriminate NEPC and adenocarcinoma samples in a clinical cohort (Figure 2B).

According to the Ensembl database ([96], annotation release 62), the human MIAT locus maps on chromo-



some 22-q12.1 and it can be spliced into 21 different isoforms. In order to investigate the clinical relevance of MIAT, we interrogated different publically available gene expression databases. Microarray profiling of 12 normal human tissues indicated that, in physiologi-cal conditions, MIAT is significantly upregulated in neural cells (Figure 2C), thus confirming previous find-ings [94]. Oncomine database analyses revealed that MIAT is significantly upregulated in prostate cancer metastatic lesions (Figure 2D) and positively associated with Rb mutation (Figure 2E). Of note, a recent study reported higher Rb mutation rates in NEPC versus prostatic adenocarcinoma [97].

We then performed significance analysis of microar-ray (SAM) to identify transcripts positively and nega-tively associated with MIAT in prostate cancer samples. This analysis was performed on a publically available database including 131 primary and 19 metastatic PCa tissues [98]. In this dataset, MIAT was over-expressed in 6/131 (4.8%) primary and 2/19 (10.5%) metastatic samples (Z score >2.0 vs non-neoplastic prostate tissue). This consistent upregulation of MIAT in metastatic lesions is intriguing, but not fully in accordance with the NEPC-specificity of this gene (metastatic lesions often do not express NEPC markers). However, sam-ples showing MIAT upregulation also showed higher expression of the NEPC marker SYP (synaptophysin; odds ratio: 61.5, 95% CI: 7.3–514.6; p = 0.000527). Moreover, genes positively associated with MIAT were highly enriched for transcripts associated with poorer prognosis and with genomic alterations found in meta-static disease (Figure 2F). Interestingly, transcripts nega-tively associated with MIAT included androgen depen-dent genes, genes silenced in embryonic stem cells and HIF1 targets (Supplementary Table 1). As noted before, PcGs are epigenetic silencers that often interact with nuclear lncRNAs. PcGs are crucial for PCa stem cell

proliferation and metastatic dissemination [45]. In addi-tion, PcGs are known to interact with HIF1 [99]. For these reasons, we directly investigated the correlation between MIAT and PcG expression, finding that this lncRNA is significantly associated with CBX2 (linear regres-sion R2 = 0.45; p < 0.0001; Supplementary Figure 3A), a PcG member that our group identified as implicated in NEPC. According to our predictions, MIAT was also negatively associated with Rb expression (linear regres-sion R2 = 0.41; p < 0.0001, Supplementary Figure 3B). While our results indicate that MIAT variation is asso-ciated with CBX2 and Rb, they also suggest that MIAT is not primarily regulated by these proteins (R2 < 0.5). Further experimental studies are needed to dissect the molecular mechanisms of MIAT/CBX/Rb interaction.

Taken together, these data are the first demonstra-tion of a lncRNA specifically expressed in NEPC, and possibly implicated in this disease.

Conclusion & future perspectiveNEPC is an incurable disease. For this reason, the iden-tification of viable therapeutic targets is of paramount importance. We have shown evidence suggesting that the ENI plays a crucial role in NEPC development and pro-gression. While evidence on the role of epigenetic effec-tors was available in the literature, the role of ncRNAs (and particularly lncRNAs) has been overlooked so far. We postulated that some lncRNAs are implicated in NEPC initiation (onco-suppressor gene silencing) and progression (acquisition of metastatic and drug resistance potential). In line with our predictions, we found that MIAT expression is restricted to a small percentage of PCas, with high metastatic potential, poor prognosis and frequent Rb mutations. Notably, all those are hallmarks of NEPC [97,100]. In addition, we find strong indica-tions that MIAT transcripts in NEPC are restricted to the nucleus. Our data suggest that MIAT can interact

Executive summary

• Neuroendocrine prostate cancer (NEPC) is an androgen receptor-negative neoplasm, which is resistant to any available treatment

• NEPC originates from the transdifferentiation of androgen receptor-positive adenocarcinoma cells. The molecular mechanisms underpinning this phenomenon are still largely unknown.

• NEPC is an incurable disease. As a result of increasingly more aggressive hormonal treatments, NEPC incidence is sharply rising.

• The epigenetic/noncoding interactome (ENI) is composed of epigenetic effectors, miRNAs and long noncoding RNAs (lncRNAs).

• Current evidence indicates that some ENI components (miRNAs, Polycomb genes) are implicated in NEPC initiation and progression.

• By analyzing our patient-derived prostate cancer xenografts, we find evidence suggesting that at least one long noncoding RNA (MIAT) is specifically expressed in NEPCs and might interact with key oncogenic pathways (Polycomb).

• Our results indicate that the ENI (and particularly lncRNAs) are a potential ‘gold mine’ that will enable us to discover effective therapies for NEPC.



with Polycomb and Rb pathways, which may explain the association of MIAT expression with an aggressive PCa phenotype. Interestingly, previous data support an inter-action between MIAT and epigenetic modifiers in neural cells [101]. Future studies will investigate the molecular mechanisms by which MIAT mediates an aggressive phenotype, and the utility of MIAT and other lncRNAs as NEPC-specific therapeutic targets.

Supplementary dataTo view the supplementary data that accompany this paper

please visit the journal website at: www.futuremedicine.com/

doi/full/10.2217/epi.16.6

Financial & competing interests disclosureThis work was financially supported by Canadian Institutes

of Health Research, grant numbers: 102604-1, 119991-

1, 123449-1 (YZW); Michael Smith Foundation for Health

Research Fellowship number 5629 (FC); Prostate Cancer

Canada (YZW); Terry Fox Research Institute, grant number:

116129-1 (YZW). The authors have no other relevant affilia-

tions or financial involvement with any organization or entity

with a financial interest in or financial conflict with the subject

matter or materials discussed in the manuscript apart from

those disclosed.


manuscript.

ReferencesPapers of special note have been highlighted as: •ofinterest;••ofconsiderableinterest

1 Abate-Shen C, Shen MM. Molecular genetics of prostate cancer. Genes Dev. 14(19), 2410–2434 (2000).

2 Ousset M, Van Keymeulen A, Bouvencourt G et al. Multipotent and unipotent progenitors contribute to prostate postnatal development. Nat. Cell Biol. 14(11), 1131–1138 (2012).

3 Parimi V, Goyal R, Poropatich K, Yang XJ. Neuroendocrine differentiation of prostate cancer: a review. Am. J. Clin. Exp. Urol. 2(4), 273–285 (2014).

4 Jemal A, Bray F, Center MM, Ferlay J, Ward E, Forman D. Global cancer statistics. CA Cancer J. Clin. 61(2), 69–90 (2011).

5 Caicoya M. Prostate cancer screening in Europe. Lancet 385(9977), 1507 (2015).

6 Shen MM, Abate-Shen C. Molecular genetics of prostate cancer: new prospects for old challenges. Genes Dev. 24(18), 1967–2000 (2010).

7 Yuan X, Cai C, Chen S, Yu Z, Balk SP. Androgen receptor functions in castration-resistant prostate cancer and mechanisms of resistance to new agents targeting the androgen axis. Oncogene 33(22), 2815–2825 (2014).

8 Scher HI, Fizazi K, Saad F et al. Increased survival with enzalutamide in prostate cancer after chemotherapy. N. Engl. J. Med. 367(13), 1187–1197 (2012).

9 Fizazi K, Scher HI, Molina A et al. Abiraterone acetate for treatment of metastatic castration-resistant prostate cancer: final overall survival analysis of the COU-AA-301 randomised, double-blind, placebo-controlled Phase 3 study. Lancet Oncol. 13(10), 983–992 (2012).

10 Berman-Booty LD, Knudsen KE. Models of neuroendocrine prostate cancer. Endocr. Relat. Cancer 22(1), R33–R49 (2015).

11 Komiya A, Yasuda K, Watanabe A, Fujiuchi Y, Tsuzuki T, Fuse H. The prognostic significance of loss of the androgen receptor and neuroendocrine differentiation in prostate biopsy specimens among castration-resistant prostate cancer patients. Mol. Clin. Oncol. 1(2), 257–262 (2013).

12 Zhang X, Coleman IM, Brown LG et al. SRRM4 expression and the loss of REST activity may promote the emergence of

the neuroendocrine phenotype in castration-resistant prostate cancer. Clin. Cancer Res. 21(20), 4698–4670 (2015).

13 Lotan TL, Gupta NS, Wang W et al. ERG gene rearrangements are common in prostatic small cell carcinomas. Mod. Pathol. 24(6), 820–828 (2011).

14 Epstein JI, Amin MB, Beltran H et al. Proposed morphologic classification of prostate cancer with neuroendocrine differentiation. Am. J. Surg. Pathol. 38(6), 756–767 (2014).

15 Sun Y, Niu J, Huang J. Neuroendocrine differentiation in prostate cancer. Am. J. Transl. Res. 1(2), 148–162 (2009).

16 Okamoto R, Matsumoto K, Fukuzaki H, Teranobu O, Sumiya Y. [Backache, headache, pyrexia, systemic edema and numbness of extremities (serum reactions and serum protein freactionation): (periarteritis nodosa)]. Nihon Rinsho 35(Suppl. 2), 3022–3023, 3384–3025 (1977).

17 Aggarwal R, Zhang T, Small EJ, Armstrong AJ. Neuroendocrine prostate cancer: subtypes, biology, and clinical outcomes. J. Natl Compr. Canc. Netw. 12(5), 719–726 (2014).

18 Marcus DM, Goodman M, Jani AB, Osunkoya AO, Rossi PJ. A comprehensive review of incidence and survival in patients with rare histological variants of prostate cancer in the United States from 1973 to 2008. Prostate Cancer Prostatic Dis. 15(3), 283–288 (2012).

19 Wang HT, Yao YH, Li BG, Tang Y, Chang JW, Zhang J. Neuroendocrine Prostate Cancer (NEPC) progressing from conventional prostatic adenocarcinoma: factors associated with time to development of NEPC and survival from NEPC diagnosis-a systematic review and pooled analysis. J. Clin. Oncol. 32(30), 3383–3390 (2014).

20 Andreoiu M, Cheng L. Multifocal prostate cancer: biologic, prognostic, and therapeutic implications. Hum. Pathol. 41(6), 781–793 (2010).

21 Beltran H, Tomlins S, Aparicio A et al. Aggressive variants of castration-resistant prostate cancer. Clin. Cancer Res. 20(11), 2846–2850 (2014).

22 Garabedian EM, Humphrey PA, Gordon JI. A transgenic mouse model of metastatic prostate cancer originating from neuroendocrine cells. Proc. Natl Acad. Sci. USA 95(26), 15382–15387 (1998).



23 Terry S, Beltran H. The many faces of neuroendocrine differentiation in prostate cancer progression. Front. Oncol. 4, 60 (2014).

24 Wang ZA, Toivanen R, Bergren SK, Chambon P, Shen MM. Luminal cells are favored as the cell of origin for prostate cancer. Cell Rep. 8(5), 1339–1346 (2014).

25 Santoni M, Conti A, Burattini L et al. Neuroendocrine differentiation in prostate cancer: novel morphological insights and future therapeutic perspectives. Biochim. Biophys. Acta 1846(2), 630–637 (2014).

26 Clermont PL, Lin D, Crea F et al. Polycomb-mediated silencing in neuroendocrine prostate cancer. Clin. Epigenetics 7(1), 40 (2015).

27 Williamson SR, Zhang S, Yao JL et al. ERG–TMPRSS2 rearrangement is shared by concurrent prostatic adenocarcinoma and prostatic small cell carcinoma and absent in small cell carcinoma of the urinary bladder: evidence supporting monoclonal origin. Mod. Pathol. 24(8), 1120–1127 (2011).

28 Beltran H, Rickman DS, Park K et al. Molecular characterization of neuroendocrine prostate cancer and identification of new drug targets. Cancer Discov. 1(6), 487–495 (2011).

29 Hansel DE, Nakayama M, Luo J et al. Shared TP53 gene mutation in morphologically and phenotypically distinct concurrent primary small cell neuroendocrine carcinoma and adenocarcinoma of the prostate. Prostate 69(6), 603–609 (2009).

30 Scheble VJ, Braun M, Wilbertz T et al. ERG rearrangement in small cell prostatic and lung cancer. Histopathology 56(7), 937–943 (2010).

31 Kadakia KC, Tomlins SA, Sanghvi SK et al. Comprehensive serial molecular profiling of an “N of 1” exceptional non-responder with metastatic prostate cancer progressing to small cell carcinoma on treatment. J. Hematol. Oncol. 8(1), 109 (2015).

32 Yuan TC, Veeramani S, Lin MF. Neuroendocrine-like prostate cancer cells: neuroendocrine transdifferentiation of prostate adenocarcinoma cells. Endocr. Relat. Cancer 14(3), 531–547 (2007).

33 Buttigliero C, Tucci M, Bertaglia V et al. Understanding and overcoming the mechanisms of primary and acquired resistance to abiraterone and enzalutamide in castration resistant prostate cancer. Cancer Treat. Rev. 41(10), 884–892 (2015).

34 Logothetis CJ, Gallick GE, Maity SN et al. Molecular classification of prostate cancer progression: foundation for marker-driven treatment of prostate cancer. Cancer Discov. 3(8), 849–861 (2013).

35 Mosquera JM, Beltran H, Park K et al. Concurrent AURKA and MYCN gene amplifications are harbingers of lethal treatment-related neuroendocrine prostate cancer. Neoplasia 15(1), 1–10 (2013).

36 Lin D, Wyatt AW, Xue H et al. High fidelity patient-derived xenografts for accelerating prostate cancer discovery and drug development. Cancer Res. 74(4), 1272–1283 (2014).

• Describesthegenerationofthefirst-in-fieldin vivomodeloftransdifferentiationtoneuroendocrineprostatecancer(NEPC).

37 Bracken AP, Helin K. Polycomb group proteins: navigators of lineage pathways led astray in cancer. Nat. Rev. Cancer 9(11), 773–784 (2009).

38 Koppens M, Van Lohuizen M. Context-dependent actions of Polycomb repressors in cancer. Oncogene doi:10.1038/onc.2015.195 (2015) (Epub ahead of print).

39 Bernstein E, Duncan EM, Masui O, Gil J, Heard E, Allis CD. Mouse Polycomb proteins bind differentially to methylated histone H3 and RNA and are enriched in facultative heterochromatin. Mol. Cell Biol. 26(7), 2560–2569 (2006).

40 Simon JA, Kingston RE. Mechanisms of Polycomb gene silencing: knowns and unknowns. Nat. Rev. Mol. Cell Biol. 10(10), 697–708 (2009).

41 Di Croce L, Helin K. Transcriptional regulation by Polycomb group proteins. Nat. Struct. Mol. Biol. 20(10), 1147–1155 (2013).

42 Zhou W, Zhu P, Wang J et al. Histone H2A monoubiquitination represses transcription by inhibiting RNA polymerase II transcriptional elongation. Mol. Cell 29(1), 69–80 (2008).

43 Chang X, Sun Y, Han S, Zhu W, Zhang H, Lian S. MiR-203 inhibits melanoma invasive and proliferative abilities by targeting the Polycomb group gene BMI1. Biochem. Biophys. Res. Commun. 456(1), 361–366 (2015).

44 Chen DL, Zhang DS, Lu YX et al. microRNA-217 inhibits tumor progression and metastasis by downregulating EZH2 and predicts favorable prognosis in gastric cancer. Oncotarget 6(13), 10868–10879 (2015).

45 Crea F, Hurt EM, Mathews LA et al. Pharmacologic disruption of Polycomb Repressive Complex 2 inhibits tumorigenicity and tumor progression in prostate cancer. Mol. Cancer 10, 40 (2011).

46 Zhang Q, Padi SK, Tindall DJ, Guo B. Polycomb protein EZH2 suppresses apoptosis by silencing the proapoptotic miR-31. Cell Death Dis. 5, e1486 (2014).

47 Fan TY, Wang H, Xiang P et al. Inhibition of EZH2 reverses chemotherapeutic drug TMZ chemosensitivity in glioblastoma. Int. J. Clin. Exp. Pathol. 7(10), 6662–6670 (2014).

48 Liu L, Guo J, Yu L et al. miR-101 regulates expression of EZH2 and contributes to progression of and cisplatin resistance in epithelial ovarian cancer. Tumour Biol. 35(12), 12619–12626 (2014).

49 Corley M, Kroll KL. The roles and regulation of Polycomb complexes in neural development. Cell Tissue Res. 359(1), 65–85 (2015).

50 Egan CM, Nyman U, Skotte J et al. CHD5 is required for neurogenesis and has a dual role in facilitating gene expression and Polycomb gene repression. Dev. Cell 26(3), 223–236 (2013).

51 Zhang J, Ji F, Liu Y et al. Ezh2 regulates adult hippocampal neurogenesis and memory. J. Neurosci. 34(15), 5184–5199 (2014).



52 Akamatsu S, Wyatt AW, Lin D et al. The placental gene PEG10 promotes progression of neuroendocrine prostate cancer. Cell Rep. 12(6), 922–936 (2015).

53 Lin D, Dong X, Wang K et al. Identification of DEK as a potential therapeutic target for neuroendocrine prostate cancer. Oncotarget 6(3), 1806–1820 (2015).

54 Tsai MC, Manor O, Wan Y et al. Long noncoding RNA as modular scaffold of histone modification complexes. Science 329(5992), 689–693 (2010).

55 Conaco C, Otto S, Han JJ, Mandel G. Reciprocal actions of REST and a microRNA promote neuronal identity. Proc. Natl Acad. Sci. USA 103(7), 2422–2427 (2006).

56 Svensson C, Ceder J, Iglesias-Gato D et al. REST mediates androgen receptor actions on gene repression and predicts early recurrence of prostate cancer. Nucleic Acids Res. 42(2), 999–1015 (2014).

57 Kanwal R, Gupta K, Gupta S. Cancer epigenetics: an introduction. Methods Mol. Biol. 1238, 3–25 (2015).

58 Ponting CP, Belgard TG. Transcribed dark matter: meaning or myth? Hum. Mol. Genet. 19(R2), R162–R168 (2010).

59 Birney E, Stamatoyannopoulos JA, Dutta A et al. Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature 447(7146), 799–816 (2007).

60 Struhl K. Transcriptional noise and the fidelity of initiation by RNA polymerase II. Nat. Struct. Mol. Biol. 14(2), 103–105 (2007).

61 Ponjavic J, Ponting CP, Lunter G. Functionality or transcriptional noise? Evidence for selection within long noncoding RNAs. Genome Res. 17(5), 556–565 (2007).

62 Qu Z, Adelson DL. Evolutionary conservation and functional roles of ncRNA. Front. Genet. 3, 205 (2012).

63 Guttman M, Rinn JL. Modular regulatory principles of large non-coding RNAs. Nature 482(7385), 339–346 (2012).

64 Huarte M, Guttman M, Feldser D et al. A large intergenic noncoding RNA induced by p53 mediates global gene repression in the p53 response. Cell 142(3), 409–419 (2010).

65 Qi P, Du X. The long non-coding RNAs, a new cancer diagnostic and therapeutic gold mine. Mod. Pathol. 26(2), 155–165 (2013).

66 Crea F, Clermont PL, Parolia A, Wang Y, Helgason CD. The non-coding transcriptome as a dynamic regulator of cancer metastasis. Cancer Metastasis Rev. 33(1), 1–16 (2014).

67 Utikal J, Abba M, Novak D, Moniuszko M, Allgayer H. Function and significance of MicroRNAs in benign and malignant human stem cells. Semin. Cancer Biol. 35, 200–211 (2015).

68 Zhang CL, Zhu KP, Shen GQ, Zhu ZS. A long non-coding RNA contributes to doxorubicin resistance of osteosarcoma. Tumour Biol. doi:10.1007/s13277-015-4130-7 (2015) (Epub ahead of print).

69 Ma J, Dong C, Ji C. MicroRNA and drug resistance. Cancer Gene Ther. 17(8), 523–531 (2010).

70 Donzelli S, Mori F, Biagioni F et al. MicroRNAs: short non-coding players in cancer chemoresistance. Mol. Cell Ther. 2, 16 (2014).

71 Ha M, Kim VN. Regulation of microRNA biogenesis. Nat. Rev. Mol. Cell Biol. 15(8), 509–524 (2014).

72 Krol J, Loedige I, Filipowicz W. The widespread regulation of microRNA biogenesis, function and decay. Nat. Rev. Genet. 11(9), 597–610 (2010).

73 Acunzo M, Romano G, Wernicke D, Croce CM. MicroRNA and cancer – a brief overview. Adv. Biol. Regul. 57, 1–9 (2015).

74 Lapuk AV, Wu C, Wyatt AW et al. From sequence to molecular pathology, and a mechanism driving the neuroendocrine phenotype in prostate cancer. J. Pathol. 227(3), 286–297 (2012).

75 Yoo AS, Staahl BT, Chen L, Crabtree GR. MicroRNA-mediated switching of chromatin-remodelling complexes in neural development. Nature 460(7255), 642–646 (2009).

76 Zhao C, Sun G, Li S et al. MicroRNA let-7b regulates neural stem cell proliferation and differentiation by targeting nuclear receptor TLX signaling. Proc. Natl Acad. Sci. USA 107(5), 1876–1881 (2010).

77 Wu D, Yu S, Jia L et al. Orphan nuclear receptor TLX functions as a potent suppressor of oncogene-induced senescence in prostate cancer via its transcriptional co-regulation of the CDKN1A (p21WAF1/CIP1) and SIRT1 genes. J. Pathol. 236(1), 103–115 (2015).

78 Dossing KB, Binderup T, Kaczkowski B et al. Down-regulation of miR-129-5p and the let-7 family in neuroendocrine tumors and metastases leads to up-regulation of their targets Egr1, G3bp1, Hmga2 and Bach1. Genes 6(1), 1–21 (2014).

79 Chinnaiyan AM, Prensner JR. The emergence of lncRNAs in cancer biology. Cancer Discov. 1(5), 391–407 (2011).

80 Joh RI, Palmieri CM, Hill IT, Motamedi M. Regulation of histone methylation by noncoding RNAs. Biochim. Biophys. Acta 1839(12), 1385–1394 (2014).

81 Quinodoz S, Guttman M. Long noncoding RNAs: an emerging link between gene regulation and nuclear organization. Trends Cell Biol. 24(11), 651–663 (2014).

82 Vance KW, Ponting CP. Transcriptional regulatory functions of nuclear long noncoding RNAs. Trends Genet. 30(8), 348–355 (2014).

83 Khalil AM, Guttman M, Huarte M et al. Many human large intergenic noncoding RNAs associate with chromatin-modifying complexes and affect gene expression. Proc. Natl Acad. Sci. USA 106(28), 11667–11672 (2009).

• AnelegantdemonstrationoftheinteractionbetweenncRNAsandepigenetics.

84 Nagano T, Mitchell JA, Sanz LA et al. The air noncoding RNA epigenetically silences transcription by targeting G9a to chromatin. Science 322(5908), 1717–1720 (2008).

85 Pandey RR, Mondal T, Mohammad F et al. Kcnq1ot1 antisense noncoding RNA mediates lineage-specific transcriptional silencing through chromatin-level regulation. Mol. Cell 32(2), 232–246 (2008).

86 Guo X, Deng L, Deng K et al. Pseudogene PTENP1 suppresses gastric cancer progression by modulating PTEN. Anticancer Agents Med. Chem. 16(4), 456–464 (2016).



87 Hu L, Ye H, Huang G et al. Long noncoding RNA GAS5 suppresses the migration and invasion of hepatocellular carcinoma cells via miR-21. Tumour Biol. doi:10.1007/s13277-015-4130-7 (2015) (Epub ahead of print).

88 Tsang WP, Ng EK, Ng SS et al. Oncofetal H19-derived miR-675 regulates tumor suppressor RB in human colorectal cancer. Carcinogenesis 31(3), 350–358 (2010).

89 Prensner JR, Iyer MK, Sahu A et al. The long noncoding RNA SChLAP1 promotes aggressive prostate cancer and antagonizes the SWI/SNF complex. Nat. Genet. 45(11), 1392–1398 (2013).

•• Seminalpaperonepigenetic/noncodingmechanismsofprostatecancerprogression.

90 Prensner JR, Iyer MK, Balbin OA et al. Transcriptome sequencing across a prostate cancer cohort identifies PCAT-1, an unannotated lincRNA implicated in disease progression. Nat. Biotechnol. 29(8), 742–749 (2011).

91 Shen XH, Qi P, Du X. Long non-coding RNAs in cancer invasion and metastasis. Mod. Pathol. 28(1), 4–13 (2015).

92 Crea F, Danesi R, Farrar WL. Cancer stem cell epigenetics and chemoresistance. Epigenomics 1(1), 63–79 (2009).

93 Iyer MK, Niknafs YS, Malik R et al. The landscape of long noncoding RNAs in the human transcriptome. Nat. Genet. 47(3), 199–208 (2015).

94 Barry G, Briggs JA, Vanichkina DP et al. The long non-coding RNA Gomafu is acutely regulated in response to

neuronal activation and involved in schizophrenia-associated alternative splicing. Mol. Psychiatry 19(4), 486–494 (2014).

95 Living Tumor Laboratory. www.livingtumorlab.com

96 Ensembl Genome Browser. http://uswest.ensembl.org

97 Tan HL, Sood A, Rahimi HA et al. Rb loss is characteristic of prostatic small cell neuroendocrine carcinoma. Clin. Cancer Res. 20(4), 890–903 (2014).

98 Taylor BS, Schultz N, Hieronymus H et al. Integrative genomic profiling of human prostate cancer. Cancer Cell 18(1), 11–22 (2010).

99 Chang CJ, Yang JY, Xia W et al. EZH2 promotes expansion of breast tumor initiating cells through activation of RAF1-beta-catenin signaling. Cancer Cell 19(1), 86–100 (2011).

100 Lugnani F, Simone G, Biava PM, Ablin RJ. The role of neuroendocrine cells in prostate cancer: a comprehensive review of current literature and subsequent rationale to broaden and integrate current treatment modalities. Curr. Med. Chem. 21(9), 1082–1092 (2014).

101 Mercer TR, Qureshi IA, Gokhan S et al. Long noncoding RNAs in neuronal-glial fate specification and oligodendrocyte lineage maturation. BMC Neurosci. 11, 14 (2010).

www.livingtumorlab.com/

http://uswest.ensembl.org/

Documents

Powered by - Future Medicine · detection and quantification of cancer-specific meth-ylation changes might help to diagnose and subtype UC, especially conveniently from urinary samples,