HYPERMETHYLATION IN LUNG CANCER: APPLICATION TO MARKER DEVELOPMENT

Although it would appear that the effects of hypomethylation
in lung cancer are modest, hypermethylation of promoter
CpG islands is widely observed. 12–16,38 Hypermethylation is
associated with transcriptional shutdown. 3 This could happen
directly through steric interference of methylated cytosines with
transcription factor and cofactor binding sites, or indirectly,
through the attraction of methyl-binding proteins to the DNA,
which in turn recruit HDAC enzymes and other epigenetic
modifiers (see Fig. 7.2). 67 In lung cancer, hundreds of studies
have been devoted to the characterization of hypermethylation
events. One of the driving forces behind this research is
the desire to identify DNA methylation markers for early lung
cancer detection. 14,15,38 DNA hypermethylation analyses could
yield powerful candidate markers for lung cancer because only
a small region of each gene needs to be interrogated, and DNA
is a PCR-amplifiable substance that can be detected in bodily
fluids. 14,15,38 Successful development of markers for cancer is a
long process that should culminate in a randomized case-control
study that demonstrates a reduction in mortality. 68 The process
for the development of DNA methylation loci into markers for
early lung cancer detection is diagrammed in the right panel of
Figure 7.3.
Most DNA methylation studies in lung cancer have focused
on NSCLC, which makes up about 85% of all lung cancers.
Small cell lung cancer, a very aggressive cancer with poor survival,
is considered by many to be a poor candidate for the development
of early detection molecular makers due to the rapid
progression of the disease. In contrast, NSCLC patients, which
include the major groups adenocarcinoma ( 40%), squamous
cell carcinoma ( 30%), large cell carcinoma ( 10%), and miscellaneous
other histological subtypes such as carcinoids and
neuroendocrine cancers ( 5%), 69 could benefit importantly if
cancers that would normally lead to death could be detected at
an early stage. 70 A comparison of methylation profiles of SCLC
and NSCLC cell lines and tumors indicates that hypermethylation
profiles are distinct for these two groups. 71–73 Not surprisingly,
differences between hypermethylation profiles of NSCLC
histological subtypes have also been observed, 15,20,72,74–77
meshing with other molecular and clinicopathological differences
found in these tumor types. 78–81 This suggests that a panel
of DNA methylation markers would be optimal, and that this
panel should include pan-lung cancer markers as well as ones for
distinct histological subtypes. Indeed, a panel of markers would
be needed even for a single subtype because penetrance of molecular
changes in cancer is usually less than 100%; it would be
unexpected to find one marker with very high sensitivity and
specificity. 20,74
The first step in molecular marker development is the
identification of promising candidate markers. 68 In the case of
DNA methylation markers for lung cancer, it is of high priority
to identify frequently methylated genes or loci (we refer
to the CpG island section we are probing as a locus , because a
given gene can be probed in multiple areas within a single or
multiple CpG islands). These loci should also show substantially
increased methylation levels over those found in healthy
tissues. Thus, the initial focus should be on penetrance and
DNA methylation levels. Because even noncancerous lung tissue
from long-term smokers may have accumulated substantial
methylation caused by age and environmental exposure, 77,82–85
many labs, including ours, have chosen to compare cancer tissues
to this type of “high-background” control tissue (referred
to here as adjacent nontumor lung [AdjNTL]) (Table 7.1). This
ensures that identified hypermethylation markers are indeed
cancer-specific and not merely indicative of environmental exposure.
(Comparison to healthy lung from nonsmokers would
be of use for the development of risk markers or identification
of candidates for chemoprevention treatments [perhaps even
epigenetic ones], once these become available.)
Many of the genes studied early on did not show high
methylation frequencies, 15 but more recent efforts by several
groups to examine much larger collections of genes have
yielded a number of panels that might deliver high sensitivity
and specificity, based on the examination of tissues
(Table 7.1). 20,49,51,58,74,75,86–92 Some these panels contain
genes that were identified early on (such as CDKN2A/p16,
MGMT, and RASSF1), 74,75,88 but many new loci have been
added to the repertoire, including homeotic genes involved
in development, such as members of the HOX and PAX
families. 19,20,51,58,90,92 The latter group of methylated loci
agrees with the observation that genes occupied in embryonic
stem cells by transcriptionally repressive polycomb group
complexes appear to be prone to methylation in cancer. 93 The
significance of methylation of these genes is unclear, since it is
thought that they may already have been silent in noncancerous
lung, but their involvement hints at the potential role of stem
cells in lung cancer development. 93 Whether the hypermethylation
of potential DNA methylation markers is functional or
not (i.e., leads to transcriptional silencing) is not relevant, as
long as penetrance is high and hypermethylation is associated
with the presence of cancer. Many of the marker panels in
Table 7.1 must still be validated on independent tumor sets,
and their ability to identify lung cancer independently of gender,
histological subtype, racial/ethnic group, and/or stages of
cancer must be further scrutinized (Fig. 7.3, right panel). Once
that is accomplished, they can be taken to the next phase of
marker development: clinical assay validation. 68 For these panels
to function in early lung cancer detection, they must be
detectable in patient remote media: bodily fluids that could
carry methylated DNA molecules from the cancer and that
could be sampled relatively noninvasively.
Detection of DNA Methylation Markers in Bodily
Fluids Potential remote media for lung cancer detection are
coughed-up sputum (spontaneously collected from smokers or
induced in never-smokers or ex-smokers), blood (plasma or
serum), bronchoalveolar lavage (BAL, a saline rinse that can be
collected during bronchoscopy), bronchial brushings, and exhaled
breath condensate (EBC, collected as condensation from
breath using a cooling device). 38 DNA methylation markers
have been detected in sputum, plasma, serum and BAL, and
data from a large number of studies is summarized in Table 7.2
(the table was compiled with special attention to BAL studies,
which have provided the most promising results to date; key
results are indicated in bold). One problem with many studies
is the lack of control subjects, which makes results difficult to
interpret. In addition, some studies cite frequencies based on
the number of methylation-positive remote samples found in
patients in which the tumor is positive. Although this is helpful
to determine the experimental sensitivity of the test, it does
not provide a good estimate of clinical sensitivity.
There are no published reports of detection of DNA
methylation markers in EBC, but microsatellite alterations
and p53 mutations have been detected in this material when
collected from lung cancer patients 94,95 While the microsatellite
alterations in EBC matched those of the lung tumors, the
p53 mutations detected in EBC and the corresponding lung
cancer did not match. 95 Thus, the DNA in EBC may not be
from the tumor, but might derive from elsewhere in the lung
or the throat and mouth. 96 Similar concerns apply to sputum,
with the added caveat that sputum is thought to provide
samples of DNA and cells from central lung areas, and thus
might favor detection of squamous lung cancer over the generally
more peripherally located adenocarcinoma. In one sputum
study, a high fraction of samples was positive for LAMC2 and
SFRP1methylation when using nested MSP (in which a preamplification
is incorporated to increase signal). 17 However,
a very large fraction of the controls was also positive, perhaps
related to the many amplification cycles.
Blood (plasma or serum) would be the easiest bodily
fluid to obtain for screening, but analyses to date indicate the
medium is not very sensitive (Table 7.2). In addition, DNA
methylation signatures observed could arise from anywhere in
the body. However, the new high-throughput DNA methylation
profiling technologies might make it feasible to identify
lung cancer –specific DNA methylation signatures. This would
require the profiling of DNA methylation in all other common
types of cancer, something that is ongoing in different
laboratories. Any identified potential lung cancer –specific
markers would need to be evaluated in other cancer types
using standardized methods. Alternatively, the combination
of blood-based methylation signatures with high-resolution
imaging (LDSCT) might be sufficient to address the issue
of organ of origin. It remains a question what size the tumor
must be in order to shed sufficient DNA into the blood for
remote detection.
Of the remote media tested, BAL appears the most promising,
showing sensitivities for individual loci approaching
50% or higher. Combination of markers into panels will help
boost sensitivity, as exemplified by studies of Grote and coworkers.
97,98 The finding that combined methylation analysis
of CDKN2A and RARB detects lung cancer cases with a sensitivity
of 69% and a specificity of 87% is highly encouraging. 98
Combination of APC, CDKN2A, and RASSF1 also showed
promise, detecting 63% of central and 44% of peripheral cancers
and exhibiting a very low background of 1/102 cases with
benign lung disease. 97 Based on the detection of methylation
in BAL from noncancer patients in a number of studies, it
would be important to use quantitative measurements and to
set a cutoff value for positive methylation. 97–100 The fact that
the collection of lavage fluid can be directed to a particular
area of the lung makes it especially suited to combine with
imaging approaches. This, in addition to the promising results
obtained to date, suggest that analysis of DNA methylation in
BAL might be the key to early lung cancer detection.
Besides their use for early detection, DNA methylation
markers identified in bodily fluids could be utilized for risk
assessment and monitoring of recurrence. In the case of quantitative
markers, cutoff values could be stratified to distinguish
between methylation detected in normal tissue from nonsmokers,
histologically normal tissue of cases prior to diagnosis,
lung cancer, or recurring lung cancer. Several studies have
shown that methylation can be detected long before the cancer
becomes clinically apparent. 101–103 However, as noted previously,
the sensitivity of the least invasive approaches (sputum,
blood) has not been high, and use of BAL would require bronchoscopy,
a semi-invasive procedure.
Functional Implications of DNA Hypermethylation
For the purposes of early detection, the functional consequences
of hypermethylation are not important. However, for
the purposes of prognostication or providing tailored therapies,
whether observed DNA methylation events have functional
consequences could be of highly relevant. This idea is supported
by the prognostic utility of expression arrays. 104 Figure 7.3 (left
panel) outlines the approach to determine whether promoter
CpG island hypermethylation of a gene is functionally significant.
After lack of expression has been verified by mRNA and/
or protein analysis, 5 aza-deoxycytidine treatment of cell lines
can be used to determine if the gene can be reactivated by demethylation.
A potential caveat of such experiments is that reactivation
could be the indirect consequence of demethylation
of other genes. Next, reexpression of the gene in cancer cells
in which the gene was silenced, and silencing (e.g., through
targeted RNAi transfection) of the gene in cells in which the
gene is still expressed will help determine the role of the gene
in cancer development and progression. In the silencing experiment,
the choice of cells is important (primary, immortalized,
transformed) and should be influenced by the perceived stage
of cancer development at which the gene of interest is thought
to play a role. Experiments such as these have implicated a variety
of hypermethylated genes in lung cancer. 12,13,15,38,105 It
should be noted that methylation of genes that were already
silent in lung tissue might not be a functional event per se,
but might still be informative, as it could provide hints to the
origin of the cancer or the involvement of stem cells. 93
Genes that appear to be silenced by methylation and that
limit any aspect of the transformed phenotype when reactivated
can be excellent tools for prognostication or targets for
therapy; the testing of associations between clinical data and
DNA methylation status in patient populations could provide
markers for survival or response to therapy. In addition, the
silenced genes could point to the involvement of molecular
pathways that might be targeted by new drug therapies. An
important caveat of studies of associations between DNA
methylation and clinicopathological variables is that a correction
should be applied for multiple hypothesis testing when
multiple new loci and clinical parameters are examined. 106
Although many genes/loci silenced in lung cancer by
DNA methylation have been studied to date, only a handful
have been analyzed in depth. The most intensive focus
has been on CDKN2A/p16, a gene encoding an inhibitor
of cyclin- dependent kinases 4 and 6, which in turn bind to
cyclin D1 and promote the phosphorylation and inactivation
of the retinoblastoma gene product, RB. RB is a key cell
cycle regulator that is frequently inactivated in SCLC. 25 In
contrast in NSCLC, it is CDKN2A that is inactivated in the
majority of tumors, and in many cases, this occurs through
promoter hypermethylation. 14,15,107,108 Methylation of the
CDKN2A promoter CpG island appears to be a very early
change in the development of squamous cell lung cancer as
well as adenocarinoma. 16,102,109 In a cohort of high-risk longterm
smokers from whom sputum was examined for seven
DNA methylation markers, hypermethylation of CDKN2A
was most strongly associated with lung cancer risk. 17 These
observations are consistent with the idea that disruption of cell
cycle regulation is an important early event in the transition
from normalcy to cancer, an observation that is emphasized
by the fact that human bronchial epithelial (HBE) cells can
be immortalized through CDK4 activation in combination
with overexpression of telomerase. 110 It is intriguing that in
NSCLC, the CDKN2A promoter CpG island appears to be
the weak link in the regulatory pathway, and it is tempting
to speculate that this might be linked to occupancy of this
region by polycomb complexes in stem cells. 111 Methylation
of the gene appears to become more pronounced during
progression 109 and is associated with an unfavorable prognosis
in lung adenocarcinoma. 112 These observations fit well with a
recent report exploring the potential link between methylation
of several genes and risk of progression in stage I NSCLC. 18 In
this study, 51 patients with stage I NSCLC who had a recurrence
within 40 months after surgery were matched for age,
sex, surgery date, and stage with 116 patients who did not have
a recurrence. The odds ratio for recurrence was found to be
significantly elevated when CDKN2A was methylated in the
tumor, regional nodes (N1) or mediastinal (N2) nodes of the
patients, or when CDH13 was methylated in the mediastinal
nodes. Combination of CDKN2A and CDH13 methylation
in the tumor and mediastinal lymph nodes was associated with
an odds ratio of recurrent cancer of 15. In a separate validation
set of 20 cases, the authors observed an association between
methylation of CDKN2A and CDH13 (individually as well
as together) in regional nodes, with an odds ratio for recurrence
of 8 for methylation of each gene alone, and of 19 when
both genes were methylated. These compelling results indicate
that methylation profiling may have important applications in
prognostication (Fig. 7.4).
One caveat with the design of studies of this kind is potential
confounding factors, such as tumor size. If a range is
used to categorize tumor size (e.g., 3 cm), the distribution
of tumor sizes within this range may not be equal for cases
(showing recurrence) and controls (no recurrence). Since the
actual size of the tumor may be a key factor in progression, it
would be important to examine this variable closely in relation
to methylation.
A second gene that has been extensively studied in DNA
methylation analysis is MGMT. 14,15 Mentioned previously as
a target of epigenetic regulation that could promote further
genetic changes, this DNA repair gene appears to be another
hot spot for early methylation, showing increasing methylation
as cells progress from a field defect to hyperplasia, and
further to adenocarcinoma. 16,109 Interestingly, methylation
was also found to be associated with tumor progression and
poor survival. 113,114 Conflicting reports about the preferential
methylation of MGMT in smokers 115 versus nonsmokers have
been made. 113
Of the genes listed in Table 7.2 as potential DNA methylation
markers in lung cancer, RARB is one of the two most promising
because it is frequently detected in BAL of cancer cases.
This gene encodes the retinoic acid receptor beta. Retinoids,
vitamin A, and its analogs, play important roles in development,
differentiation, proliferation, and apoptosis. 116 Retinoids have
been considered strong candidates for chemoprevention of lung
cancer, 116 an idea that makes sense considering the hypermethylation
of RARB (Table 7.1). Unfortunately, the outcome of
retinoid chemoprevention clinical trials was an increase rather
than decrease in the risk of lung cancer. 117 Nevertheless, in vitro
experiments suggest that retinoic acid can prevent the oncogenic
transformation of immortalized HBE cells. 118 Hypermethylation
of RARB was linked with retinoic acid resistance in an HBE cell
line and treatment with DNA methylation inhibitor azacitidine
restored the cells’ ability to respond to retinoic acid. 119 Like
methylation of MGMT, RARB hypermethylation appears to be
an early event in lung adenocarcinoma development, showing
low but detectable levels in adjacent lung and increasing methylation
in hyperplasia and adenocarcinoma. 109,120
Similarly to MGMT and RARB, RASSF1 methylation occurs
early, showing a progression in frequency of methylation
from the surrounding histologically normal tissue of a tumor, to
hyperplasia and to lung adenocarcinoma. 109 RASSF1 methylation
is frequently detectable in BAL and appears to be the most
specific marker examined in BAL to date (Table 7.2). The gene
encodes a putative RAS effector protein. 121 It lies on chromosome
3p21, in an area of common loss of heterozygosity in lung
cancer, and is frequently methylated in human malignancies including
lung cancer (Table 7.2). 14,15,121 The gene has alternative
first exons, alpha and gamma, each with a CpG island. The upstream
island (RASSF1A) is hypermethylated in lung cancer, and
its methylation is strongly correlated with expression of the delta
subfamily of the DNA methyltransferase 3B. 35 This DNMT3B
subfamily consists of at least seven splice variants. Knockdown
of DNMT3B4 in lung cancer cell lines resulted in reactivation
of the RASSF1A but not the CDKN2A promoter, implying that
DNMT3B isoforms could be involved in initiating promoterspecific
DNA methylation. RASSF1A methylation is suggested
to correlate with a poor prognosis, although this observation
should be confirmed by an independent study. 122
Besides CDKN2A, MGMT, RARB, and RASSF1, there
are many other genes that could be of interest, functionally
or therapeutically. Three genes worth mentioning briefly are
CDH13, OPCML, and miRNA-29 (miR-29). As mentioned
previously, hypermethylation of CDH13, encoding the cell adhesion
molecule heart cadherin, was associated with increased
risk of recurrence. 18 OPCML, encoding an opioid-binding cell
adhesion molecule-like, had been pegged as a suspected tumor
suppressor gene many years ago by Maneckjee and Minna 123
based on the apoptotic response of lung cancer cell lines to
opioids, which antagonized the growth stimulatory effect of
nicotine. The frequent and high methylation of OPCML in
both adenocarcinoma and squamous cell lung cancer suggests
that it might function as a pan-lung cancer marker. 19,20 The
miR-29 family of three RNAs is an example of miRNAs that
are silenced by hypermethylation in lung cancer (in contrast
to the activated let-7a-3 mentioned earlier). 124 Expression
of these RNAs is inversely correlated with DNMT3a and 3b
in lung cancer, which appears to be mediated by targeting of
miR-29 to the 3 untranslated regions of the methyltransferase
mRNAs. Reactivation of miR-29 could be one way in which
methyltransferase expression and tumorigenic potential of
lung cancer cells could be mitigated, as illustrated by the reduced
tumor growth in nude mice of A549 lung cancer cells
transfected with miR-29. 124
From the data described previously, it is clear that progress
is being made in understanding the functional consequences and
clinical implications of DNA hypermethylation. However, much
work remains to be done to independently verify observations before
they can lead to clinical implementation (such as treatment
decisions based on methylation profiles). The general reversal of
methylation is already a clinical target though, with numerous
drugs that counteract DNA methylation under development and
a number of them in clinical trials (see later in this chapter). 21
Histone Modifications and Their Role in Lung
Cancer The link between DNA methylation and chromatin
structure is formed by proteins that bind directly or indirectly
to methylated DNA and modify the flexible histone N-termini
(Fig. 7.2). The nucleosomal core around which DNA is coiled
is composed of two molecules each of histones 2A, 2B, 3, and
4. The lysine and arginine-rich N-terminal regions extend
from the core and can be heavily decorated with monomethylation,
dimethylation, and trimethylation, acetylation, ubiquitination,
phosphorylation, and other modifications. 5,125
These modifications do not exist in isolation; functional and
physical cross talk ensures a complex web of epigenetic signals,
in which DNA methyltransferases, methyl-binding proteins,
histone variants, histone-modifying enzymes, and other chromatin
and transcriptional components play a role (Fig. 7.2). 126
Many of the enzymes that modify histones recognize other
modifications on the same or different histone tails, or on
DNA. For example, proteins that bind to methylated DNA
frequently carry additional domains that interact directly or
indirectly with histone-modifying proteins, such as deacetylases.
67 Acetylation of histones on lysine promotes active transcription.
On the one hand, this modification reduces positive
charge and minimizes the electrostatic attraction of the histone
tails for the DNA phosphate backbone, thereby relaxing chromatin
structure. In addition, acetylated histone N-terminal
tails are landing pads for bromodomain-containing proteins,
such as transcriptional coactivator p300/CBP associated factor
and TAF1, a component of the transcription initiation complex.
5 Multiple enzymes that add or remove acetyl groups exist
in the cell. In contrast to acetylation, methylation does not affect
histone tail charge, functioning by altering protein/protein
interactions. One or two methyl groups can be added to arginine
and up to three to lysine; the effects depend on the
modified position and the number of added methyl groups.
For example, histone 3 lysine 9 and lysine 27 trimethylation
(H3K9me3, H3K27me3) are repressive marks, while histone 3
lysine 4 trimethylation is found in transcribed regions.
In contrast to the abundance of information about DNA
methylation in lung cancer, relatively little is known about
how histone modification is affected; molecular changes on
the histone N-terminal regions are much more difficult to interrogate
in comparison to DNA methylation. The most commonly
used technique is formaldehyde cross-linking followed
by specific immunoprecipitation of particular histone modifications
and PCR-based or global (e.g., microarray, highthroughput
sequencing) characterization of the coprecipitated
DNA sequences. One recent study classified NSCLC patients
into seven distinct groups based on differential histone modifications
and observed differences in survival depending on histology
and histone 3 modifications. 127 This early study hints
at the potential use of this kind of epigenetic characterization
to guide treatment.