You are on page 1of 8

PERSpECTIVES

(Table 1; Supplementary Table 2). Of these


OPINION
14 biomarker assays, only nine (GSTP1,
APC, RASSF1, NDRG4, BMP3, two SEPT9
Analysis of DNA methylation in biomarkers, SHOX2, and MGMT) have
been included in one or more clinical
cancer: location revisited guidelines2–7. Moreover, only two tests have
received FDA approval: Cologuard (NDRG4
and BMP3) for the analysis of stool DNA
Alexander Koch   , Sophie C. Joosten, Zheng Feng, Tim C. de Ruijter, samples collected as part of colorectal cancer
Muriel X. Draht, Veerle Melotte, Kim M. Smits, Jurgen Veeck, screening protocols and Epi proColon
James G. Herman, Leander Van Neste, Wim Van Criekinge, Tim De Meyer and (SEPT9) for analysis of blood samples taken
for the same purpose. This rate of translation
Manon van Engeland
of ~0.8% (14 commercially available DNA
Abstract | Changes in DNA methylation in cancer have been heralded as promising methylation-​based assays versus an estimated
targets for the development of powerful diagnostic, prognostic, and predictive 1,800 reported in various publications) is in
line with the numbers presented by George
biomarkers. Despite the existence of more than 14,000 scientific publications Poste8 in 2011 (100 biomarkers used in the
describing DNA methylation-​based biomarkers and their clinical associations in clinic versus 150,000 biomarker-​related
cancer, only 14 of these biomarkers have been translated into a commercially publications) and Scott E. Kern9 in 2012
available clinical test. Methodological and experimental obstacles are both major (<1% of cancer biomarkers reported in
causes of this disparity , but the genomic location of a DNA methylation-​based publications reach the clinic) (Fig. 1).
The above-​mentioned translational
biomarker is an intrinsic and essential property that also has an important and
success rate and the current public
often overlooked role. Here, we examine the importance of the location of DNA discussion on the reliability and efficiency
methylation for the development of cancer biomarkers, and take a detailed look of scientific research10 warrant a reflection
at the genomic location and other relevant characteristics of the various on why so few DNA methylation-​based
biomarkers with commercially available tests. We also emphasize the value of biomarkers are transitioned from initial
publicly available databases for the development of DNA methylation-​based discovery and publication to a clinical
test. The answer to this question is
biomarkers and the importance of accurate reporting of the full methodological complex, multifaceted, and mainly
details of research findings. relates to small effect sizes and a lack of
substantial clinical value, as well as various
In 2003, Peter Laird1 expressed his features of cancer (the query was executed methodological caveats (such as biased
belief that the then recent advances in in January 2018, Supplementary Table 1). patient selection, improper study design
our understanding of the role of DNA This search returned 14,743 research articles, and data analysis, lack of validation, and/
methylation in human cancer could one representing an estimated 1,800 markers, on or inappropriate reporting) that prevent
day result in a multitude of powerful the basis of manual curation of representative the thorough evaluation of the clinical
DNA methylation-​based biomarkers, subsets of these publications. However, only value of a biomarker. Furthermore, the
particularly for use as cancer diagnostics. 14 DNA methylation-​based biomarker assays financial investment needed to develop
This belief was based on a number are currently commercially available and are an assay quantifying a specific biomarker
of characteristics of changes in DNA designed to measure the methylation of only or biomarkers, including trials providing
methylation that make them promising 13 genes in total: glutathione S-​transferase P validation and exploring clinical utility, is a
biomarkers: early and frequent occurrence (GSTP1); adenomatous polyposis coli protein major obstacle. These issues are not unique
in cancer, easy detection by well-​ (APC); ras association domain-​containing to DNA methylation-​based biomarkers.
established techniques, stability of DNA protein 1 (RASSF1); N-​myc downregulated Researchers in the field of biomarker
methylation in fixed samples over time, gene 4 (NDRG4); bone morphogenetic research in general face similar problems,
presence in various bodily fluids, and protein 3 (BMP3); septin-9 (SEPT9); short although only now are such problems
cell-​type specificity. stature homeobox protein 2 (SHOX2); twist-​ beginning to be addressed8,9,11,12.
Interest in changes in DNA methylation related protein 1 (TWIST1); homeobox Apart from the established issues
as cancer biomarkers has resulted in a large protein OTX1 (OTX1); one cut domain surrounding clinical translation, we
number of scientific publications. To more family member 2 (ONECUT2); methylated-​ believe another often overlooked barrier
accurately define this number, we performed DNA-protein-​cysteine methyltransferase exists between the discovery and clinical
a search of the PubMed database for (MGMT); branched-​chain-amino-​acid implementation of DNA methylation-​based
publications that describe DNA methylation-​ aminotransferase, cytosolic (BCAT1); and biomarkers in patients with cancer. In 2011,
based biomarkers, in relation to clinical DNA-​binding protein Ikaros (IKZF1) we addressed the importance of the genomic

Nature Reviews | Clinical Oncology


© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
PerspeCTIves

Table 1 | Commercially available assays designed to measure Dna methylation-​based cancer biomarkers
genes indication Hypermethylated Correlation with Promoter Cpg Clinical FDa Refs
in tumour? gene expression? region? island? guideline? approved?
ConfirmMDx
GSTP1 Diagnosis and prognosis of Yes Yes Yes Yes Yes2 No 57

prostate cancer
APC Yes No No No
RASSF1 Yes No Yes Yes
Cologuard
NDRG4 Early detection of CRC Yes Yes Yes Yes Yes3–5 Yes 77,78

BMP3 Yes Yes Yes Yes


Epi proColon/ColoVantage (Licensed to Quest Diagnostics and ARUP Laboratories) a

SEPT9a Early detection of CRC Yes No No Yes Yes4,7 Yes 79

Epi proColon a

SEPT9a Early detection of CRC Yes No No Yes Yes4,7 Yes 80

Epi proLung
SHOX2 Early detection of lung cancer Yes Yesb No Yes No No 81

AssureMDx
TWIST1 Risk of bladder cancer in Yes Yes Yes Yes No No 82

patients with haematuria


OTX1 Yes Yes b
No Yes
ONECUT2 Yes Yes b
No Yes
PredictMDx
MGMT Prediction of response to NAc Yes Yes Yes Yes6 No 20,83

temozolomide in glioblastoma
Colvera
BCAT1 Early detection of CRC Yes Yes Yes Yes No No 84

recurrence
IKZF1 Yes Yes Yes Yes
The Cancer Genome Atlas (TCGA) data for all biomarkers with commercially available assays were analysed for differential methylation between tumour and
nonmalignant samples and to determine whether or not methylation status at the location of the area covered by the assay is correlated with expression. We also
investigated whether or not the biomarkers are located in the promoter region of their corresponding genes and whether or not they are located in a CpG island.
Data on the clinical performance of these biomarkers, as reported in the literature, are listed in Supplementary Table 2. CRC, colorectal cancer ; NA , not available.
a
Two different SEPT9 biomarkers are available, which both target the same genomic region but at slightly different locations. In some publications, the SEPT9
biomarker is variably referred to as the ColoVantage test by Quest Diagnostics, as a test provided by ARUP Laboratories, or as the Epi proColon test provided by
Epigenomics. Both biomarkers have been patented by Epigenomics83,85, which has licensed one of them to Quest Diagnostics and ARUP Laboratories. bA positive,
instead of a negative, correlation exists between DNA methylation and gene expression of SHOX2, OTX1, and ONECUT2. cMGMT is a predictive marker, rendering
the comparison between tumour and nonmalignant samples less relevant. Furthermore, no nonmalignant samples of brain tissue are available in TCGA for
comparisons with glioblastoma specimens.

location of DNA methylation in relation to and conclude with recommendations for the CpG islands is an important step in the
gene expression and/or clinicopathological development of DNA methylation-​based development of a DNA methylation-​based
characteristics of cancers and emphasized biomarkers that might improve the currently biomarker, as illustrated in the following
how crucial this aspect is for biomarker limited translation of basic research into examples.
development13. We argued that both a better clinical advances.
understanding and more detailed analysis Diagnostic methylation markers.
of the clinical relevance of the genomic The importance of location Methylation of GSTP1 has been reported
location of DNA methylation is required to Traditionally, DNA methylation research as a promising diagnostic marker for
increase the number of such biomarkers that in general and DNA methylation-​based hepatocellular carcinoma (HCC) in several
can be successfully implemented in patient biomarker studies have predominantly been studies, but with widely varying levels of
care. In this Perspective, we hypothesize focused on the effects of hypermethylation specificity. This variability is hypothesized
that epigenomic data that are currently of promoter CpG islands in tumour-​ to result from analysis of different specific
publicly available offer an opportunity to suppressor genes14. However, even within a parts of the same promoter20. One group of
better select the optimal location of DNA single promoter region, not all CpG islands investigators conducted a detailed analysis
methylation-​based biomarkers. We closely are functionally equivalent. Transcriptional of methylation of the GSTP1 promoter
examine the genomic location and other silencing, for example, is often controlled using bisulfite sequencing and found
relevant characteristics of DNA methylation-​ by the methylation of one or more small that methylation of the 5ʹ region of the
based biomarkers with potential for clinical parts of a promoter, rather than by the entire promoter was significantly more specific
use in patients with cancer and for which an promoter region15–19. Thus, identifying in distinguishing HCC from nonmalignant
assay is currently commercially available, the precise location of clinically relevant liver diseases or liver tissue with no

www.nature.com/nrclinonc
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
PerspeCTIves

Number of biomarkers survival in four independent patient series


used in the clinic (all P ≤ 0.03)25.

14 Predictive methylation markers. In the


(0.8%)
initial studies that described the predictive
100 power of MGMT methylation in patients
(0.07%) <1% with glioblastoma, the methylation status
of MGMT was assessed using methylation-​
specific PCR (MSP), using primers covering
only a few CpG dinucleotides located
in the first exon and promoter regions of
the gene26–30. Direct correlations between
14,743
Published papers methylation and MGMT expression and/or
describing DNA- activity were not investigated. Following this
150,000 methylation based 100% finding, numerous investigators25–27 analysed
biomarker papers biomarkers cancer biomarkers
(Poste, 2011) (Kern, 2012) the promoter region of the MGMT gene in
~1,800 more detail, using quantitative techniques
biomarkers such as pyrosequencing31–33. The goal of
(Koch et al., 2018)
these studies was to gain greater insight into
specific methylation patterns, to identify the
Fig. 1 | Success rates for the clinical implementation of cancer biomarkers. The numbers of puta- CpG dinucleotides that are most important
tive biomarkers and related publications have been compared with the number of biomarkers used in in transcriptional regulation, and to identify
the clinic, with similar results obtained from different sources. The success rate of DNA methylation-​
the region or regions in which analysis of the
based cancer biomarkers (0.8%) was calculated by comparing the number of commercially available
biomarkers with the estimated total number of published biomarkers. A comprehensive search of the changes in methylation provides the highest
PubMed database for publications related to DNA methylation-​based cancer biomarkers returned clinical value. Conflicting results regarding
14,743 research papers. By counting the number of unique gene symbols in the corresponding the specific methylated region or regions,
abstracts (5,090), manually curating three random subsets of these abstracts (100 abstracts each) and or even individual CpG dinucleotides,
calculating the average proportion of cancer DNA methylation-​based biomarkers in these three sub- that correlate best with MGMT expression
sets (36.5%), we arrived at our estimate of ~1,800 unique biomarkers. The estimates of the translational have been published31–33. The region
success rate by Poste2 and Kern3 were derived based on the number of biomarkers used in the clinic encompassing CpGs 75–78, which is covered
versus the number of biomarker papers and the number of cancer biomarkers used in the clinic versus by the commercially available PredictMDx
the number of published biomarkers, respectively ; the exact calculation was not described in detail test, is currently the best studied and
by either author.
methylation of CpG dinucleotides in
this region has been repeatedly shown
established pathological alterations than investigators described an association to be associated with favourable patient
methylation of the 3ʹ end of the promoter between methylation of a small region prognosis33. This MGMT test has received a
(97.1% versus 60%; P < 0.001)20. The same in the promoter of the NMDAR2B gene, Tier 1 reimbursement code by the American
group obtained similar data on RASSF1 located just outside a CpG island, and Medical Association and was included in
methylation as an early detection marker for inferior overall and disease-​specific survival the 2013 National Comprehensive Cancer
HCC (72.9% specificity for promoter region outcomes in patients with oesophageal Network guidelines6, indicating that the test
1 compared with 27.1% for promoter region squamous cell carcinoma (HR 3.13, 95% should be widely used; however, the debate
2 and 38.6% for exon 1 at a fixed sensitivity CI 1.05–9.72; P ≤ 0.006), while DNA on the exact region of the promoter at which
of 90%; P < 0.0001)21. methylation at two other locations, located methylation provides the highest level of
further downstream of the TSS and within clinical value continues.
Prognostic methylation markers. In CpG islands, was not correlated with
previous research, we analysed three patient outcomes24. Genomic context
different genomic regions within CpG Interestingly, the methylation status of Methylation of specific CpG dinucleotides
islands of the GREM1 promoter and found single CpG dinucleotides, and not only within a single genomic region, such as a
that only methylation of CpGs in one of of wider regions that encompass multiple gene promoter, can regulate transcription
these regions was associated with inferior CpG dinucleotides, has been shown to as well as determining the clinical value of a
prognosis in patients with clear cell renal influence the regulation of gene expression biomarker. Therefore, identifying the most
cell carcinoma (RCC; hazard ratio (HR) and, therefore, the prognostic value of relevant genomic location of a potential
2.32, 95% CI 1.52–3.53; P = 0.001)22. Similar a DNA methylation-​based biomarker25. DNA methylation-​based biomarker is
results were observed in patients with A high-​resolution quantitative analysis important. Apart from location, the genomic
gastric cancer, in which hypermethylation of methylation of the entire ZAP70 gene context of DNA methylation including, for
of CpGs in a specific promoter region in samples from patients with chronic example, the presence of different functional
located close to the transcription start lymphocytic leukaemia (CLL) demonstrated regions, such as exons or enhancers, must
site (TSS), but not of CpGs in a different that methylation of a single CpG also be taken into account. The presence
location, in the MAL promoter was dinucleotide (located +223 from the TSS) is of these different regions is associated
associated with superior disease-​free particularly important for the transcriptional with characteristic methylation profiles.
survival (P = 0.03) and downregulation repression of ZAP70 and that methylation of Promoter CpG islands, for example, are
of gene expression (P = 0.01)23. Other this CpG is correlated with superior patient typically not methylated, whereas most gene

Nature Reviews | Clinical Oncology


© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
PerspeCTIves

bodies, particularly exons, are extensively CpG islands in tumour cells accompanied methylation-​based biomarkers that currently
methylated34. Research published in 2012 by downregulation of the associated gene) have commercially available detection assays
shows that alterations in DNA methylation reflect the current best practice for the (Table 1) are indeed located in the most
in both intergenic regions and intragenic development of DNA methylation-​based clinically relevant genomic region or regions.
regions are modulated in development and biomarkers. A link between methylation and Testing this hypothesis is straightforward
disease and that these alterations are actively gene expression provides a clear biological owing to the universal availability of
involved in the regulation of transcription35. rationale, although it is important to note large-​scale cancer genomics databases.
Methylation at non-​promoter sites has that a DNA methylation-​based biomarker
a role in the regulation of several vital does not necessarily have to be linked with Biomarker performance in TCGA
processes, including splicing, the expression a measurable change in gene expression. Publicly available databases such as The
of alternative transcripts from alternative Of the commercially available biomarkers Cancer Genome Atlas (TCGA) already have
promoters, and enhancer activation, for example, four (APC, RASSF1, and demonstrated value for the cancer research
although such DNA modifications remain both SEPT9 biomarkers) do not show a community and have proved helpful in
a largely unexplored area for biomarker correlation between DNA methylation and unravelling both the genetic and epigenetic
development. However, potential clinical expression of the associated gene (Table 1). background of cancer (see, for example, the
utility has already been demonstrated for Cancer-​associated hypermethylation comprehensive molecular classifications of
methylation-​based biomarkers in certain can occur on silent genes marked by the colorectal42 and prostate43 cancers). Data
non-​promoter locations. For example, presence of bivalent chromatin, thus provided by TCGA are publicly available
patients with gastric cancer who are positive preventing their activation rather than and can be used in biomarker research to
for long-​range epigenetic silencing (LRES), actively suppressing their expression38,39. generate hypotheses or as a discovery or
which encompasses multiple neighbouring This blocking of activation still implies a validation data set. Several online tools, such
genes, of the chromosomal region 15q25 link with expression even if no statistical as the UCSC Xena browser44, cBioPortal45,
have been shown to have a lower risk association is observed between the DNA and MEXPRESS46, have enabled researchers
of disease recurrence (HR 0.6; 95% CI methylation status of a gene and its basal to easily inspect (and analyse) TCGA data
0.38–0.94; P = 0.027)36. Aberrant enhancer level of expression. Unravelling such an for their biomarker of interest.
hypermethylation has also been shown to association would require additional We used TCGA data to examine the
be prognostic for survival in patients with analyses such as re-​expression experiments 14 methylation-​based biomarkers with
RCC (log-​rank P < 0.05)37. With advances involving a demethylating agent or the commercially available assays (Table 1). A
in our understanding of the biology and analysis of different tissues. similar analysis of potential biomarkers
consequences of these types of DNA Other important considerations in that failed to reach the clinic could provide
methylation, we will undoubtedly see an the development of a methylation-​based valuable insight into the pitfalls of the
increase in the amount of published data biomarker include cell-​type specificity and biomarker development process, but given
available on the role of methylation-​based the optimal number of CpG dinucleotides to the considerable technical challenges such
biomarkers located outside of promoter cover. For example, methylation of potential an analysis would pose, we decided to focus
regions. Similar to methylation-​based cancer biomarker genes has been observed on the small set of biomarkers for which
biomarkers located in promoter regions, the in peripheral blood samples containing assays are commercially available. When
development of non-​promoter methylation-​ no malignant cells40, thus emphasizing the evaluating the genomic location examined
based biomarkers will also face the challenge importance of verifying the specificity of any using the ConfirmMDx test to measure
of determining the precise genomic region potential biomarker. Regarding the number the methylation status of GSTP1, together
associated with changes in gene expression of CpG motifs to cover in a biomarker-​based with that of regions interrogated by other
and/or clinical outcomes. assay, some publications advocate exploiting primers with published data indicating
the regional nature of DNA methylation their use in prostate adenocarcinoma
Biomarker development by measuring the methylation of several samples (Fig. 2; Supplementary Table 3),
For methylation of a specific genomic neighbouring CpG repeats using a single all assays were found to target the same
location to be a good biomarker, a clinically assay41, although the findings of other genomic region surrounding the TSS,
significant difference in methylation studies19 demonstrate that methylation of a despite the existence of small differences
between the groups of interest has to single CpG dinucleotide can be responsible in the exact genomic location analysed.
be detected: for example, in tumour for transcriptional repression as well as Each Infinium 450k probe within the
versus nontumour tissue for a diagnostic determining the clinical relevance of a plotted genomic window revealed a clear
biomarker; between tumour samples from biomarker. This observation emphasizes the difference in methylation between tumour
patients with high-​risk disease versus low-​ importance of identifying not only the and nonmalignant tissue samples, as well
risk disease for a prognostic marker; or optimal genomic location but also as a strong negative correlation with GSTP1
between responders versus nonresponders the optimal number of CpG dinucleotides expression levels, indicating the presence of
for a predictive marker. The associated to be analysed by each individual DNA a 1-kb-​wide region with clinical relevance,
gene should — ideally — be differentially methylation-​based biomarker assay. which is targeted by all illustrated markers.
expressed between both groups. A classic In order to investigate whether Overall, eight assays examine methylation
example of a candidate diagnostic biomarker our strategy of measuring CpG of biomarkers located in the canonical
would be a hypermethylated CpG-​rich hypermethylation in the promoter regions promoter regions of their corresponding
region of the promoter of a tumour of downregulated genes is the best possible genes (GSTP1, RASSF1, NDRG4, BMP3,
suppressor gene that is downregulated in method of determining the clinical TWIST1, MGMT, BCAT1, and IKZF1),
tumour cells. We believe that these features relevance of DNA methylation, we explored 12 assays examine biomarkers located
(a promoter containing hypermethylated the hypothesis that the 14 cancer DNA within 2 kbp of the TSS (GSTP1, RASSF1,

www.nature.com/nrclinonc
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
PerspeCTIves

Methylation location of a specific probe (β-​value ≥ 0.2)


Correlation (β-value) and those who did not (β-​value < 0.2).
GSTP1 1 1
Only data from patients who received
0.5 0.75 temozolomide were included in this
analysis. We observed a significant
0 0.5 difference in overall survival at three probe
locations (log-​rank test, FDR < 0.05), one
–0.5 0.25 of which corresponded with the location
covered by the commercially available
–1 0
MGMT assay, thereby confirming the
GSTP1 known predictive power of MGMT as a
1 marker (Supplementary Fig. 3).
4
2 3 Of the 14 biomarkers, 10 (GSTP1,
6 5 NDRG4, BMP3, SHOX2, TWIST1,
7
9 8 OTX1, ONECUT2, MGMT, BCAT1,
and IKZF1) are located in a region
67351000 67351500 67352000 where we found a statistically significant
Genomic location correlation between methylation and
gene expression by analysing TCGA
Genomic Spearman Wilcoxon data (Spearman correlation, FDR < 0.05).
Samples annotation correlation test Interestingly, for methylation of SHOX2,
Normal Gene P < 0.05 P < 0.05 OTX1, and ONECUT2, this correlation
Tumour Transcript P ≥ 0.05 P ≥ 0.05 was positive instead of negative. Several
Primer investigators describe the complex, but
CpG(i) often positive, relationships between gene-​
body methylation and gene expression in
Fig. 2 | genomic location of GSTP1 Dna methylation biomarkers. All GSTP1 methylation-​based general47, and a positive association has
biomarkers in this figure cluster together at the canonical transcription start site (note that for the been reported for methylation of SHOX2
ConfirmMDx assay , only information on the amplicon was available; all other primer annotations can specifically48. Overall, our findings, while
be found in Supplementary Table 3). On inspection of The Cancer Genome Atlas (TCGA) DNA meth- not definitive evidence, imply that our
ylation and gene expression data for GSTP1 in prostate adenocarcinoma samples, we can see that all proposed strategy is indeed a valid approach
displayed Infinium 450k probes reveal differences in methylation between nonmalignant and tumour to identifying clinically relevant regions.
tissue samples (as assessed by Wilcoxon test, false discovery rate (FDR) < 0.05), and that the methyla-
TCGA data come with two important
tion data of each probe were negatively correlated with GSTP1 expression (as assessed by Spearman
limitations. First, DNA methylation is
correlation, rs < −0.5, FDR < 0.05). The y-​axis on the right shows the β-​values; a horizontal bar was drawn
at the median β-​value for each probe, and the vertical lines indicate the 25–75% interquartile range. measured using Infinium 450k microarrays,
The y-​axis on the left shows rs. Hypermethylation of the CpG island (CpG(i)) in the promoter of GSTP1 restricting the resulting data to the particular
is a prime example of a DNA methylation-​based biomarker that is associated with downregulation of genomic locations of the probes used in the
the corresponding gene in cancer. Details on the primer sequences were sourced from the following array, which might not necessarily capture
publications: 1 (ref.57); 2 (refs58,59); 3 (refs60–70); 4 (ref.71); 5 (ref.72); 6 (ref.73); 7 (ref.74); 8 (ref.75); the most relevant methylation sites. Second,
and 9 (ref.76). for predictive markers, which are associated
with a specific treatment outcome, relevant
data on both treatment received and patient
NDRG4, BMP3, TWIST1, MGMT, BCAT1, 2, 4–12), the only exception being the outcomes are not available in TCGA. Despite
IKZF1, APC, both SEPT9 markers, and predictive MGMT marker, for which these limitations, data from TCGA or other
SHOX2), and all but one of these biomarkers the comparison of nonmalignant and large-​scale cancer genomics projects might
(APC) are located in a CpG island (Table 1; tumour tissue samples is limited by a lack still provide valuable information on, for
Supplementary Figs 1–12). of suitable nonmalignant tissue samples. example, the presence of DNA methylation
In an exploratory analysis, we then Ideally, we would have compared the or a possible link between DNA methylation
considered whether the biomarkers MGMT methylation profiles of patients and gene expression. Ongoing efforts by
examined using commercially available who responded to treatment with international organizations, such as the
assays were hypermethylated in tumour temozolomide with those of patients International Cancer Genome Consortium
samples and whether their methylation who did not respond to this agent, in (ICGC), to create comprehensive cancer
status correlated with the expression of line with the hypothesized effects of genomics databases might help to resolve
the associated gene by analysing the MGMT methylation on responsiveness these issues.
specific probes used at and around to alkylating agents, although data on Importantly, none of the standard
the location of each biomarker. We found these responses are not readily available in bisulfite-​based technologies, including
a statistically significant difference in TCGA. Instead, we performed a survival the Infinium microarrays, can distinguish
methylation between nonmalignant analysis on TCGA data for all Infinium between DNA methylation and DNA
and tumour samples at the location of 450k probes surrounding the location hydroxymethylation, another epigenetic
each marker (Wilcoxon test, Benjamini of the MGMT marker, in which we mark linked with tumorigenesis49, unless
& Hochberg false discovery rate compared the overall survival outcomes a specific protocol is used to identify
(FDR) < 0.05) (Supplementary Figs 1, of patients who had methylation at the hydroxymethylation50. Moreover, other

Nature Reviews | Clinical Oncology


© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
PerspeCTIves

Box 1 | Recommendations demonstration of its clinical utility, are


equally important but are not the focus of
• The optimal genomic location of a potential DNA methylation biomarker and the number of these recommendations.
CpGs to cover should be critically evaluated. The location of a DNA methylation-​based
biomarker can affect clinical relevance. Conclusions
• Publicly available, large-​scale cancer (epi)genomics databases should be used to identify the DNA methylation-​based biomarkers were
genomic location that is best associated with the relevant clinical features. heralded as the next ‘big thing’ in cancer
• All assay protocols should be accurately and thoroughly reported, including the exact genomic biomarker research more than a decade
location of the biomarker. ago, but so far, they have struggled to live
up to the expectations. Several possible
explanations are available as to why so few
approaches to finding clinically relevant biomarkers comparable to, for example, DNA methylation-​based biomarkers have
changes in DNA methylation are available, the Minimum Information About a been commercialized, but we believe that
such as the location-​agnostic machine Microarray Experiment standard the complex relationship between DNA
learning methods used in the EPICUP for microarray experiments54 and based methylation and its precise genomic location
classifier, which was developed to identify on the nomenclature proposed by the is one of the main obstacles. In order to
and classify the origins of cancers of Human Genome Variation Society55,56 reduce the difference between the number of
unknown primary51. could help alleviate these issues. published biomarkers and the number
of biomarkers that are used in the clinic, we
Reporting Recommendations formulated three simple recommendations
The clinical translation of a methylation-​ In summary, the analysis of the for the development of DNA methylation-​
based cancer biomarker does not depend commercially available assays designed based biomarkers. Biomarkers developed
on identifying the most clinically relevant to quantify various methylation-​based according to these recommendations are
genomic location alone: for observations biomarkers, together with the suboptimal not guaranteed to be either reproducible
regarding genomic location to be useful assay design regarding genomic location or clinically relevant, although, at least
they must also be published at a sufficient regularly encountered in the literature, the first barrier in the DNA methylation-​
level of detail. Reproducibility, one of the prompted us to establish a few simple based biomarker pipeline, determining
pillars of scientific research, relies on the recommendations that will hopefully the clinically most relevant region, can
detailed and transparent reporting of all facilitate the development and clinical be overcome though analysis of publicly
findings, data, and protocols, including translation of DNA methylation-​based available data. In order to systematically
the genomic location of methylation markers (Box 1). Our first and most evaluate the performance of DNA
biomarkers10,52,53. Accurate reporting seems important recommendation is to critically methylation-​based biomarkers and to
obvious but, in practice, improvements are evaluate both the optimal genomic support the accurate future reporting
still possible52. In our opinion, details of the location of a potential methylation-​ of both the discovery and validation of
genomic locations of all methylation-​based based biomarker and the number of such biomarkers, we are developing a
biomarkers with commercially available CpG dinucleotides to cover in the assay centralized atlas containing data on all
assays were reported in sufficient detail, used to measure DNA methylation, published biomarkers, as well as a biomarker
apart from GSTP1, APC, and RASSF1, for regardless of whether this biomarker reporting standard. We hope that our
which only the genomic location of the is located in a promoter CpG island or recommendations and continuing efforts
MSP amplicon was published and not the in another functional region of a gene. will help improve the performance of DNA
primer sequences. We recommend that, in The examples described above provide methylation-​based markers, promote
future, publications contain details such overwhelming evidence that the location reproducibility, and decrease the amount of
as specific primer sequences and assay of a DNA methylation-​based biomarker research waste that is currently produced in
protocols, preferably in the main text of absolutely influences its clinical relevance. this field.
the article (and not as supplementary Second, we recommend incorporating
information), to facilitate the replication of publicly available large-​scale epigenomics Alexander Koch   1, Sophie C. Joosten2, Zheng Feng1,
the findings. Reporting only the genomic data into methylation-​based biomarker Tim C. de Ruijter2, Muriel X. Draht1, Veerle Melotte1,
Kim M. Smits1,2, Jurgen Veeck2, James G. Herman3,
location of the primers or amplicons and/ development and validation processes Leander Van Neste1, Wim Van Criekinge4,
or referencing previously published primers in order to improve predictions of the Tim De Meyer5 and Manon van Engeland1*
without providing the exact sequences genomic region that is best associated 1
Department of Pathology, GROW — School for
either burdens the interested reader with with the relevant clinical features. Oncology and Developmental Biology, Maastricht
the additional task of retrieving the primer Third, in order to facilitate independent University Medical Center, Maastricht, Netherlands.
sequences or necessitates the design of replication and validation, we urge all 2
Division of Medical Oncology, Department of Internal
new assays, which might differ from those investigators to accurately report the Medicine, GROW — School for Oncology and
used in the original publication. Finally, methylation-​based biomarker assay Developmental Biology, Maastricht University Medical
Center, Maastricht, Netherlands.
having to contact the authors to obtain protocols used, including details of the
3
Lung Cancer Program, University of Pittsburgh Cancer
primer sequences or assay protocols exact genomic location of the biomarker.
Institute, The Hillman Cancer Center, Pittsburgh,
only results in further complications Other steps in the development process, PA, USA.
in terms of both data exchange and in such as the choice and design of an 4
BioBix: Laboratory of Bioinformatics and
the reproducibility of research — two appropriate assay, comparisons of the Computational Genomics, Department of Mathematical
cornerstones of good science. A reporting performance of any new biomarker to Modelling, Statistics and Bioinformatics, Ghent
standard for cancer DNA methylation-​based the gold standard (if available), and/or University, Ghent, Belgium.

www.nature.com/nrclinonc
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
PerspeCTIves

5
Department of Mathematical Modelling, Statistics carcinoma. Clin. Cancer Res. 13, 6658–6665 50. Stewart, S. K. et al. oxBS-450K: a method for
and Bioinformatics, Ghent University, Ghent, Belgium. (2007). analysing hydroxymethylation using 450K BeadChips.
25. Claus, R. et al. Quantitative DNA methylation analysis Methods 72, 9–15 (2015).
*e-​mail: manon.van.engeland@mumc.nl identifies a single CpG dinucleotide important for 51. Moran, S. et al. Epigenetic profiling to classify cancer
ZAP-70 expression and predictive of prognosis in of unknown primary: a multicentre, retrospective
https://doi.org/10.1038/s41571-018-0004-4
chronic lymphocytic leukemia. J. Clin. Oncol. 30, analysis. Lancet Oncol. 17, 1386–1395 (2016).
Published online xx xx xxxx 2483–2491 (2012). 52. Iqbal, S. A., Wallach, J. D., Khoury, M. J., Schully, S. D.
26. Esteller, M. et al. Inactivation of the DNA-​repair gene & Ioannidis, J. P. Reproducible research practices and
1. Laird, P. W. The power and the promise of DNA MGMT and the clinical response of gliomas to alkylating transparency across the biomedical literature. PLoS
methylation markers. Nat. Rev. Cancer 3, agents. N. Engl. J. Med. 343, 1350–1354 (2000). Biol. 14, e1002333 (2016).
253–266 (2003). 27. Hegi, M. E. et al. Clinical trial substantiates the 53. Freedman, L. P., Venugopalan, G. & Wisman, R.
2. Carroll, P. R. et al. NCCN Guidelines Insights: Prostate predictive value of O-6-methylguanine-​DNA Reproducibility2020: Progress and priorities.
Cancer Early Detection, Version 2.2016. J. Natl methyltransferase promoter methylation in F1000Res 6, 604 (2017).
Compr. Canc. Netw. 14, 509–519 (2016). glioblastoma patients treated with temozolomide. 54. Brazma, A. et al. Minimum information about a
3. Exact Sciences. Clinical guidelines and quality Clin. Cancer Res. 10, 1871–1874 (2004). microarray experiment (MIAME)-toward standards
measures. Cologuard http://www.cologuardtest.com/ 28. Hegi, M. E. et al. MGMT gene silencing and benefit for microarray data. Nat. Genet. 29, 365–371
hcp/about-​cologuard/clinical-​guidelines (2017). from temozolomide in glioblastoma. N. Engl. J. Med. (2001).
4. US Preventive Services Task Force. et al. Screening 352, 997–1003 (2005). 55. den Dunnen, J. T. et al. HGVS Recommendations for
for colorectal cancer: US Preventive Services Task 29. Wick, W. et al. Temozolomide chemotherapy alone the Description of Sequence Variants: 2016 Update.
Force recommendation statement. JAMA 315, versus radiotherapy alone for malignant astrocytoma Hum. Mutat. 37, 564–569 (2016).
2564–2575 (2016). in the elderly: the NOA-08 randomised, phase 3 trial. 56. Monk, D. et al. Recommendations for a nomenclature
5. The American Cancer Society medical and editorial Lancet Oncol. 13, 707–715 (2012). system for reporting methylation aberrations in
content team. Colorectal cancer screening tests. The 30. Malmstrom, A. et al. Temozolomide versus standard imprinted domains. Epigenetics https://doi.org/10.108
American Cancer Society http://www.cancer.org/ 6-week radiotherapy versus hypofractionated 0/15592294.2016.1264561 (2016).
content/cancer/en/cancer/colon-​rectal-cancer/detection-​ radiotherapy in patients older than 60 years with 57. Van Neste, L. et al. A tissue biopsy-​based epigenetic
diagnosis-staging/screening-​tests-used.html (2017). glioblastoma: the Nordic randomised, phase 3 trial. multiplex PCR assay for prostate cancer detection.
6. National Comprehensive Cancer Network. NCCN Lancet Oncol. 13, 916–926 (2012). BMC Urol. 12, 16 (2012).
Clinical Practice Guidelines in Oncology: Older Adult 31. Malley, D. S. et al. A distinct region of the MGMT CpG 58. Jeronimo, C. et al. Quantitation of GSTP1 methylation
Oncology. NCCN http://www.nccn.org/professionals/ island critical for transcriptional regulation is in non-​neoplastic prostatic tissue and organ-​confined
physician_gls/pdf/senior.pdf (2017). preferentially methylated in glioblastoma cells and prostate adenocarcinoma. J. Natl Cancer Inst. 93,
7. Epigenomics AG. China FDA names Epigenomics’ xenografts. Acta Neuropathol. 121, 651–661 (2011). 1747–1752 (2001).
blood-​based Septin9 colorectal cancer test a most 32. Everhard, S. et al. Identification of regions correlating 59. Hoque, M. O. et al. Quantitative methylation-​specific
innovative medical product for 2015. Epigenomics MGMT promoter methylation and gene expression in polymerase chain reaction gene patterns in urine
http://www.epigenomics.com/china-​fda-names-​ glioblastomas. Neuro Oncol. 11, 348–356 (2009). sediment distinguish prostate cancer patients from
epigenomics-blood-​based-septin9-colorectal-​ 33. Bienkowski, M. et al. Clinical Neuropathology practice control subjects. J. Clin. Oncol. 23, 6569–6575
cancer-test-​innovative-medical-​product-2015/ (2016). guide 5-2015: MGMT methylation pyrosequencing in (2005).
8. Poste, G. Bring on the biomarkers. Nature 469, glioblastoma: unresolved issues and open questions. 60. Altimari, A. et al. Diagnostic role of circulating free
156–157 (2011). Clin. Neuropathol. 34, 250–257 (2015). plasma DNA detection in patients with localized prostate
9. Kern, S. E. Why your new cancer biomarker may never 34. Jones, P. A. Functions of DNA methylation: islands, cancer. Am. J. Clin. Pathol. 129, 756–762 (2008).
work: recurrent patterns and remarkable diversity in start sites, gene bodies and beyond. Nat. Rev. Genet. 61. Cairns, P. et al. Molecular detection of prostate cancer
biomarker failures. Cancer Res. 72, 6097–6101 13, 484–492 (2012). in urine by GSTP1 hypermethylation. Clin. Cancer Res.
(2012). 35. Kulis, M. et al. Epigenomic analysis detects widespread 7, 2727–2730 (2001).
10. Munafò, M. R. et al. A manifesto for reproducible gene-​body DNA hypomethylation in chronic lymphocytic 62. Crocitto, L. E. et al. Prostate cancer molecular
science. Nat. Hum. Behav. 1, 0021 (2017). leukemia. Nat. Genet. 44, 1236–1242 (2012). markers GSTP1 and hTERT in expressed prostatic
11. Ioannidis, J. P. A. & Bossuyt, P. M. M. Waste, leaks, 36. Kang, J. Y. et al. Identification of long-​range epigenetic secretions as predictors of biopsy results. Urology 64,
and failures in the biomarker pipeline. Clin. Chem. 63, silencing on chromosome 15q25 and its clinical 821–825 (2004).
963–972 (2017). implication in gastric cancer. Am. J. Pathol. 185, 63. Goessl, C., Muller, M., Heicappell, R., Krause, H. &
12. van Gool, A. J. et al. Bridging the translational 666–678 (2015). Miller, K. DNA-​based detection of prostate cancer in
innovation gap through good biomarker practice. Nat. 37. Hu, C. Y. et al. Kidney cancer is characterized by blood, urine, and ejaculates. Ann. NY Acad. Sci. 945,
Rev. Drug Discov. 16, 587–588 (2017). aberrant methylation of tissue-​specific enhancers that 51–58 (2001).
13. van Vlodrop, I. J. et al. Analysis of promoter CpG are prognostic for overall survival. Clin. Cancer Res. 64. Goessl, C. et al. DNA-​based detection of prostate
island hypermethylation in cancer: location, location, 20, 4349–4360 (2014). cancer in urine after prostatic massage. Urology 58,
location! Clin. Cancer Res. 17, 4225–4231 (2011). 38. Klutstein, M., Nejman, D., Greenfield, R. & Cedar, H. 335–338 (2001).
14. Baylin, S. B., Herman, J. G., Graff, J. R., Vertino, P. M. DNA methylation in cancer and aging. Cancer Res. 76, 65. Gonzalgo, M. L., Nakayama, M., Lee, S. M.,
& Issa, J. P. Alterations in DNA methylation: a 3446–3450 (2016). De Marzo, A. M. & Nelson, W. G. Detection of GSTP1
fundamental aspect of neoplasia. Adv. Cancer Res. 72, 39. Easwaran, H. et al. A DNA hypermethylation module methylation in prostatic secretions using combinatorial
141–196 (1998). for the stem/progenitor cell signature of cancer. MSP analysis. Urology 63, 414–418 (2004).
15. Gonzalgo, M. L. et al. The role of DNA methylation in Genome Res. 22, 837–849 (2012). 66. Gonzalgo, M. L., Pavlovich, C. P., Lee, S. M. &
expression of the p19/p16 locus in human bladder 40. Kristensen, L. S., Raynor, M. P., Candiloro, I. & Nelson, W. G. Prostate cancer detection by GSTP1
cancer cell lines. Cancer Res. 58, 1245–1252 (1998). Dobrovic, A. Methylation profiling of normal individuals methylation analysis of postbiopsy urine specimens.
16. Homma, N. et al. Spreading of methylation within reveals mosaic promoter methylation of cancer-​ Clin. Cancer Res. 9, 2673–2677 (2003).
RUNX3 CpG island in gastric cancer. Cancer Sci. 97, associated genes. Oncotarget 3, 450–461 (2012). 67. Jeronimo, C. et al. Quantitative GSTP1
51–56 (2006). 41. Lehmann-​Werman, R. et al. Identification of tissue-​ hypermethylation in bodily fluids of patients with
17. Deng, G., Chen, A., Hong, J., Chae, H. S. & Kim, Y. S. specific cell death using methylation patterns of prostate cancer. Urology 60, 1131–1135 (2002).
Methylation of CpG in a small region of the hMLH1 circulating DNA. Proc. Natl Acad. Sci. USA 113, 68. Rogers, C. G. et al. High concordance of gene
promoter invariably correlates with the absence of E1826–E1834 (2016). methylation in post-​digital rectal examination and
gene expression. Cancer Res. 59, 2029–2033 (1999). 42. Cancer Genome Atlas Network. Comprehensive post-​biopsy urine samples for prostate cancer
18. Licchesi, J. D. et al. Transcriptional regulation of Wnt molecular characterization of human colon and rectal detection. J. Urol. 176, 2280–2284 (2006).
inhibitory factor-1 by Miz-1/c-​Myc. Oncogene 29, cancer. Nature 487, 330–337 (2012). 69. Roupret, M. et al. Promoter hypermethylation in
5923–5934 (2010). 43. Cancer Genome Atlas Research Network. The circulating blood cells identifies prostate cancer
19. Yoshikawa, H. et al. SOCS-1, a negative regulator of molecular taxonomy of primary prostate cancer. Cell progression. Int. J. Cancer 122, 952–960 (2008).
the JAK/STAT pathway, is silenced by methylation in 163, 1011–1025 (2015). 70. Sunami, E. et al. Multimarker circulating DNA assay
human hepatocellular carcinoma and shows growth-​ 44. Goldman, M., Craft, B., Zhu, J. & Haussler, D. The for assessing blood of prostate cancer patients. Clin.
suppression activity. Nat. Genet. 28, 29–35 (2001). UCSC Xena system for cancer genomics data Chem. 55, 559–567 (2009).
20. Jain, S. et al. Impact of the location of CpG visualization and interpretation [abstract]. Cancer Res. 71. Bastian, P. J. et al. CpG island hypermethylation
methylation within the GSTP1 gene on its specificity as 77 (Suppl.), 2584 (2017). profile in the serum of men with clinically localized
a DNA marker for hepatocellular carcinoma. PLoS 45. Cerami, E. et al. The cBio cancer genomics portal: an and hormone refractory metastatic prostate cancer.
ONE 7, e35789 (2012). open platform for exploring multidimensional cancer J. Urol. 179, 529–534 (2008).
21. Jain, S. et al. Differential methylation of the genomics data. Cancer Discov. 2, 401–404 (2012). 72. Bryzgunova, O. E., Morozkin, E. S., Yarmoschuk, S. V.,
promoter and first exon of the RASSF1A gene in 46. Koch, A., De Meyer, T., Jeschke, J. & Van Criekinge, W. Vlassov, V. V. & Laktionov, P. P. Methylation-​specific
hepatocarcinogenesis. Hepatol. Res. 45, MEXPRESS: visualizing expression, DNA methylation sequencing of GSTP1 gene promoter in circulating/
1110–1123 (2015). and clinical TCGA data. BMC Genomics 16, 636 extracellular DNA from blood and urine of healthy
22. van Vlodrop, I. J. et al. Prognostic significance of (2015). donors and prostate cancer patients. Ann. NY Acad.
Gremlin1 (GREM1) promoter CpG island 47. Ehrlich, M. & Lacey, M. DNA methylation and Sci. 1137, 222–225 (2008).
hypermethylation in clear cell renal cell carcinoma. differentiation: silencing, upregulation and modulation 73. Chuang, C. K. et al. Hypermethylation of the CpG
Am. J. Pathol. 176, 575–584 (2010). of gene expression. Epigenomics 5, 553–568 (2013). islands in the promoter region flanking GSTP1 gene
23. Buffart, T. E. et al. MAL promoter hypermethylation as a 48. Zhang, Y. A. et al. SHOX2 is a potent independent is a potential plasma DNA biomarker for detecting
novel prognostic marker in gastric cancer. Br. J. Cancer biomarker to predict survival of WHO Grade II-​III prostate carcinoma. Cancer Detect Prev. 31, 59–63
99, 1802–1807 (2008). diffuse gliomas. EBioMedicine 13, 80–89 (2016). (2007).
24. Kim, M. S. et al. A promoter methylation pattern in 49. Jeschke, J., Collignon, E. & Fuks, F. Portraits of TET-​ 74. Ellinger, J. et al. CpG island hypermethylation at
the N-​methyl-D-​aspartate receptor 2B gene predicts mediated DNA hydroxymethylation in cancer. Curr. multiple gene sites in diagnosis and prognosis of
poor prognosis in esophageal squamous cell Opin. Genet. Dev. 36, 16–26 (2016). prostate cancer. Urology 71, 161–167 (2008).

Nature Reviews | Clinical Oncology


© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
PerspeCTIves

75. Payne, S. R. et al. DNA methylation biomarkers of bronchial aspirates. BMC Cancer 10, 600 Author contributions
prostate cancer: confirmation of candidates and (2010). A.K., S.C.J., and M.v.E. researched data for the article, all
evidence urine is the most sensitive body fluid for non-​ 82. van Kessel, K. E., Van Neste, L., Lurkin, I., Zwarthoff, E. C. authors made a substantial contribution to discussion of con-
invasive detection. Prostate 69, 1257–1269 & Van Criekinge, W. Evaluation of an epigenetic tent, A.K., S.C.J., and M.v.E. wrote the manuscript, and all
(2009). profile for the detection of bladder cancer in authors reviewed and/or edited the manuscript before
76. Trock, B. J. et al. Evaluation of GSTP1 and APC patients with hematuria. J. Urol. 195, 601–607 submission.
methylation as indicators for repeat biopsy in a high-​ (2016).
risk cohort of men with negative initial prostate 83. Lofton-​Day, C. & Ebert, M. Methods and nucleic Competing interests
biopsies. BJU Int. 110, 56–62 (2012). acids for the detection of colorectal cell M.v.E. receives research funding from MDxHealth. W.V.C. is
77. Melotte, V. et al. N-​Myc downstream-​regulated gene 4 proliferative disorders. ES2587068T3 (2016). a consultant of MDxHealth. The other authors declare no
(NDRG4): a candidate tumor suppressor gene and 84. Mitchell, S. M. et al. A panel of genes methylated with competing interests.
potential biomarker for colorectal cancer. J. Natl high frequency in colorectal cancer. BMC Cancer 14,
Cancer Inst. 101, 916–927 (2009). 54 (2014). Publisher’s note
78. Loh, K. et al. Bone morphogenic protein 3 inactivation 85. Devos, T. et al. Method for methylation analysis. Springer Nature remains neutral with regard to jurisdictional
is an early and frequent event in colorectal cancer EP2479289B1 (2007). claims in published maps and institutional affiliations.
development. Genes Chromosomes Cancer 47,
449–460 (2008). Acknowledgements Supplementary information
79. Warren, J. D. et al. Septin 9 methylated DNA is a The work of the authors is supported financially by the Maag Supplementary information is available for this paper at
sensitive and specific blood test for colorectal cancer. Lever Darm Stichting (MLDS, grant FP13-15) and an SU2C-​ https://doi.org/10.1038/s41571-018-0004-4.
BMC Med. 9, 133 (2011). DCS International Translational Cancer Research Dream
80. Lofton-​Day, C. et al. DNA methylation biomarkers for Team Grant (Stand Up To Cancer (SU2C)-AACR-​DT1415, Related links
blood-​based colorectal cancer screening. Clin. Chem. MEDOCC). SU2C is a programme of the Entertainment The Cancer Genome Atlas: https://cancergenome.nih.gov/
54, 414–423 (2008). Industry Foundation administered by the American The International Cancer Genome Consortium:
81. Schmidt, B. et al. SHOX2 DNA methylation is a Association for Cancer Research and the Universiteitsfonds http://icgc.org/
biomarker for the diagnosis of lung cancer based on Limburg/SWOL.

www.nature.com/nrclinonc
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

You might also like