Abstract

   The importance of epigenetic regulation has been increasingly
   recognized in the development of cancer. In this study, we investigated
   the impact of smoking, a major risk factor of lung cancer, on DNA
   methylation by comparing the genome-wide DNA methylation patterns
   between lung adenocarcinoma samples from six smokers and six
   nonsmokers. We identified that smoking-induced DNA methylations were
   enriched in the calcium signaling and neuroactive ligand receptor
   signaling pathways, which are closely related to smoking-induced lung
   cancers. Interestingly, we discovered that two genes in the
   mitogen-activated protein kinase signaling pathway (RPS6KA3 and ARAF)
   were hypomethylated in smokers but not in nonsmokers. In addition, we
   found that the smoking-induced lung cancer-specific DNA methylations
   were mostly enriched in nuclear activities, including regulation of
   gene expression and chromatin remodeling. Moreover, the smoking-induced
   hypermethylation could only be seen in lung adenocarcinoma tissue but
   not in adjacent normal lung tissue. We also used differentially
   methylated DNA loci to construct a diagnostic model to distinguish
   smoking-associated lung cancer from nonsmoking lung cancer with a
   sensitivity of 88.9% and specificity of 83.2%. Our results provided
   novel evidence to support that smoking can cause dramatic changes in
   the DNA methylation landscape of lung cancer, suggesting that
   epigenetic regulation of specific oncogenic signaling pathways plays an
   important role in the development of lung cancer.

   Keywords: lung cancer, epigenome, methylation, tumor suppressor gene,
   smoking

Introduction

   Lung cancer is the leading cause of cancer-related death worldwide. The
   estimated 5-year survival rate ranges from 6% to 16%, depending on the
   subtype.[38]^1 The tumorigenesis of lung cancer not only involves
   various genetic alterations, such as point mutations, deletions, and
   gene amplifications, but also alterations in epigenetics, including DNA
   methylation and histone modifications. One of the most studied
   epigenetic regulations is cytosine methylation, which does not affect
   the primary structure of chromatin, but does affect the secondary
   structure, which is critical to regulation of gene expression.

   In lung cancer, upregulation of DNA methyltransferase 1 (DNMT1) protein
   has been shown to correlate with increased hypermethylation of tumor
   suppressor genes and downregulation of tumor suppressor
   proteins.[39]^2^–[40]^4 Recently, Belinsky evaluated the methylation of
   genes isolated from blood samples in a cohort of women with different
   levels of risk for lung cancer.[41]^5 They identified that p16Ink4,
   MGMT, and RASSF1A were hypermethylated in the high-risk cohort. The
   methylation pattern of specific genes may be useful in early detection
   of lung cancer. In another study, Feng et al studied the DNA
   methylation status of 27 genes from 49 patients with non-small cell
   lung cancer and found that some cancer-specific changes in DNA
   methylation pattern can only be seen in tumor tissues but not in
   preneoplastic tissues.[42]^6

   Smoking, as a well known risk factor for lung cancer, has been widely
   recognized as a carcinogen with a significant role in tumorigenesis of
   the lung. Many studies have demonstrated that smoking can induce
   genomic instability by producing genetic mutations and altering
   epigenetic modifications. Hypermethylation of promoter regions have
   frequently been observed in smokers with and without lung cancer, which
   is consistent with the higher level of DNMT1 in lung samples from
   smokers when compared with that from nonsmokers.[43]^3^,[44]^7^,[45]^8
   The cumulative smoking dose (pack-years) has been shown to correlate
   well with the frequency of methylation in cancer-free heavy
   smokers.[46]^9CCND2 and APC were frequently hypermethylated in both
   cancerous and noncancerous lung tissues of smokers with non-small cell
   lung cancer, suggesting that hypermethylation of these genes may be
   associated with environmental factors, such as chronic smoking. Other
   recent research showed that methylation of RASSF1A is associated with
   exposure to smoke in lung cancer.[47]^10^,[48]^11 Hypermethylation of
   FHIT has been shown to be an early event in smoking-associated squamous
   cell lung cancer.[49]^12

   On the other side, some studies have shown that smoking is not clearly
   associated with an altered DNA methylation pattern in lung cancer
   patients. Tommasi et al investigated the effects of chronic exposure to
   a prototype smoke-derived carcinogen, benzo[a]pyrene diol epoxide
   (B[a]PDE), on DNA methylation in genomic regions that have been
   previously demonstrated to be important in lung cancer.[50]^13 They
   demonstrated that chronic treatment of normal human cells in vitro with
   B[a]PDE did not alter the DNA methylation pattern in the genomic
   regions relevant to lung cancer within a time frame that preceded
   cellular transformation.[51]^13 Another study by Scesnaite et al did
   not identify significant changes in the pattern of DNA methylation
   associated with smoking.[52]^14 Hillemacher et al investigated the
   effect of smoking on the global DNA methylation pattern in 298 genomic
   DNA samples, but did not find a direct effect of smoking on global DNA
   methylation.[53]^15 However, these epigenetic studies were mostly
   focused on special gene groups or genome loci rather than systematic
   epigenome-wide analysis of DNA methylation pattern. Moreover, these
   studies relied on a qualitative methylation-specific polymerase chain
   reaction method that is subjective, relying on detection of a band from
   electrophoresis, which does not distinguish low-level from high-level
   methylation.[54]^16^–[55]^18

   In this study, we used an epigenome-wide screening method based on
   Illumina 27K DNA methylation microarray technology to study how smoking
   contributes to the development of lung cancer from a DNA methylation
   point of view. We identified differentially methylated genes by
   comparing the global DNA methylation patterns between lung
   adenocarcinoma samples from smokers and nonsmokers. Our study provides
   an insightful perspective on smoking-associated DNA methylation and its
   role in tumorigenesis of the lung.

Materials and methods

Sample preparations

   We obtained lung adenocarcinoma cancer tissues and corresponding
   adjacent normal lung tissues from 12 patients diagnosed with stage
   I–IIIa lung adenocarcinoma (six smokers and six nonsmokers) at the
   Shanghai Chest Hospital. All 12 patients enrolled in this study were
   untreated before they underwent complete tumor resection at the
   Shanghai Chest Hospital in 2010. Tumor staging was based on biopsy and
   conducted by pathologists in the Shanghai Chest Hospital. Baseline
   characteristics and tumor stages for all patients are listed in
   [56]Supplementary Table 1. This study was approved by the ethics
   committee of the Shanghai Chest Hospital and the School of Medicine,
   Shanghai Jiao Tong University. All patients provided their written
   informed consent.

Illumina 27K DNA methylation microarray

   The experiment was performed according to Illumina’s protocol. Briefly,
   the procedures included:
    1. Bisulfite treatment, whereby 1 μg of genomic DNA was used in
       bisulfite conversion to convert the unmethylated cytosine into
       uracil.
    2. Genomic DNA amplification, where the bisulfite-treated DNA was
       subjected to whole genome analysis by random hexamer primer and
       Phi29 DNA polymerase. The products were then enzymatically
       fragmented, purified from dNTPs, primers, and enzymes, before being
       applied to microarray chips.
    3. Hybridization and single-base extension. There were two bead types
       for each CpG site per locus in the chip, which were differentiated
       by different bead types. A total of 200,000 beads were available in
       the chip. Hybridization was followed by single-base extension with
       hapten-labeled dideoxynucleotides.
    4. Fluorescence staining and chip scanning. Multilayered
       immunohistochemical assays were performed by repeated rounds of
       staining with a combination of antibodies to differentiate the two
       types. After staining, the chip was scanned to show the intensities
       of the unmethylated and methylated bead types. The fluorescence
       intensity ratios between the two bead types were then calculated.
    5. Analysis of methylation data. The scanned microarray images were
       analyzed using BeadStudio software (Illumina, San Diego, USA),
       which performed statistical tests on the results and normalized the
       raw data to reduce the effects of experimental variation and
       background.

Analysis of DNA methylation data

   We used the MethyLumiSet Bioconductor package to handle the methylation
   microarray data as a data container in Methyllumi R. We studied the
   pattern of missing values and performed data imputation. Probes were
   discarded if they lay on the sex chromosomes. We then performed the
   Wilcoxon test to find significant differential DNA methylation regions
   between: normal lung tissues from smokers and normal lung tissues from
   nonsmokers to identify smoking-specific DNA methylation; and between
   lung cancer tissues from smokers and lung cancer tissues from
   nonsmokers to identify smoking-associated methylation changes that are
   specific to lung cancer.

Gene ontology and pathway analysis

   We used hypergeometric analysis to identify the differentially
   methylated region (DMR) and obtained the DMR gene-enriched functional
   pathways. Gene function annotation was performed using resources from
   Gene Ontology. Pathway resources were obtained from KEGG and BioCarta.
   Gene functional cluster analysis was performed using the DAVID system.

Construction of a diagnostic model for classifying smoking-associated and
nonsmoking lung cancer

   We used the proper signature selection method, which combines the
   nearest centroid and genetic algorithm fitting tuning approach, to
   search for the best DNA methylation loci as diagnostic markers. Genetic
   algorithms are variable search procedures that are based on the
   principle of evolution by natural selection. The procedure works by
   evolving sets of variables (chromosomes) that fit certain criteria from
   an initial random population via cycles of differential replication,
   recombination, and mutation of the fittest chromosomes.[57]^19 The
   concept of using in silico evolution for resolution of optimization
   problems was introduced by John Holland in 1975.[58]^19

Results

Identification of smoking-specific DNA methylation in lung adenocarcinoma

   To investigate how smoking affects DNA methylation in lung
   adenocarcinoma, we used a DNA methylation microarray to compare the
   global DNA methylation pattern between lung adenocarcinoma samples from
   smokers and nonsmokers (six samples for each group) to identify
   smoking-induced DNA methylations that are specific to lung
   adenocarcinoma ([59]Figure 1A). We identified 137 DMRs, including a
   number of genes having been previously shown to function in smoking and
   cancer, such as ADRB3, GALR1, and TRPC1. Of the 137 DMRs, 48 were
   hypermethylated in smokers with lung adenocarcinoma compared with
   nonsmokers with lung adenocarcinoma, while 89 of the 137 DMRs were
   hypomethylated. To filter out the smoking-induced DNA methylations that
   are not specific to lung adenocarcinoma, we also compared the global
   DNA methylation pattern for adjacent normal lung tissues from the same
   six smokers and six nonsmokers. We discovered 90 DMRs that were
   specifically induced by smoking, including two pro-oncogenic genes from
   the Ras/Raf/MEK/ERK (mitogen-activated protein kinase, MAPK) signaling
   pathway (RPS6KA3 and ARAF). Interestingly, all 90 DMRs were
   hypomethylated loci ([60]Figure 1B and [61]C). By comparing the two
   sets of results, we found that 27 DMRs (all hypomethylated) can be
   found in both sets of results, indicating that these 27 DMRs are
   smoking-specific changes in DNA methylation that present in both lung
   cancer tissues and normal lung tissues of smokers. After subtracting
   the 27 common DMRs from the 137 DMRs that we identified in the
   comparison between lung cancer tissues from smokers and nonsmokers, we
   obtained 110 smoking-induced DNA methylations specific to lung
   adenocarcinoma, including 48 hypomethylated DMRs and 62 hypermethylated
   DMRs ([62]Figure 1B and [63]C). The full list of the differentially
   methylated genes can be found in [64]Supplementary Tables 2–[65]4.

Figure 1.

   [66]Figure 1
   [67]Open in a new tab

   Differences in global DNA methylation pattern in lung cancer between
   smokers and nonsmokers. The global DNA methylation pattern in normal
   lung tissues from smokers (SLT) was compared with that from nonsmokers
   (NSLT), producing 90 differentially methylated regions. The global DNA
   methylation pattern of lung tumor tissues from smokers (STT) was
   compared with that from nonsmokers (NSTT), resulting in 137
   differentially methylated regions. Twenty-seven loci can be found in
   both result sets. The numbers of hypermethylation and hypomethylation
   loci are also shown. Abbreviation: vs, versus.

   To gain a functional insight from our data, we performed a Gene
   Ontology analysis of all candidate genes. As a result, we found that
   the 90 smoking-induced DMRs in the normal lung tissues of smokers were
   enriched in pathways including energy metabolism, phosphorylase kinase
   activity, and ubiquitin thiolesterase activity ([68]Table 1). The 110
   smoking-induced DMRs specific to lung cancer were mostly enriched in
   biological processes associated with transcription factor activity,
   chromatin remodeling, and neural signal processing ([69]Table 2). Using
   the data we obtained from this study, we proposed a model to depict the
   developmental progress of smoking-induced lung cancer from an
   epigenetic point of view ([70]Figure 2).

Table 1.

   Pathways enriched with differentially methylated genes from the
   comparison between normal lung tissues from smokers and nonsmokers
   Category Term Count % P
   GOTERM_BP_FAT GO:0006006∼glucose metabolic process 4 7.843137255
   0.011044244
   GOTERM_MF_FAT GO:0004689∼phosphorylase kinase activity 2 3.921568627
   0.014170587
   GOTERM_MF_FAT GO:0004221∼ubiquitin thiolesterase activity 3 5.882352941
   0.019246991
   GOTERM_BP_FAT GO:0019318∼hexose metabolic process 4 7.843137255
   0.0202083
   GOTERM_BP_FAT GO:0005996∼monosaccharide metabolic process 4 7.843137255
   0.029419949
   GOTERM_MF_FAT GO:0016790∼thiolester hydrolase activity 3 5.882352941
   0.034057248
   GOTERM_MF_FAT GO:0000166∼nucleotide binding 12 23.52941176 0.04365391
   GOTERM_MF_FAT GO:0048487∼beta-tubulin binding 2 3.921568627 0.044661571
   [71]Open in a new tab

Table 2.

   Pathway enrichment analysis for smoking-induced lung cancer-specific
   differentially methylated genes
   Term Count % P
   GO:0003700∼transcription factor activity 10 27.77777778 0.000326
   GO:0019932∼second messenger-mediated signaling 5 13.88888889
   0.001931974
   GO:0043565∼sequence-specific DNA binding 7 19.44444444 0.00275705
   GO:0048168∼regulation of neuronal synaptic plasticity 3 8.333333333
   0.003050078
   GO:0050804∼regulation of synaptic transmission 4 11.11111111
   0.003636035
   GO:0051969∼regulation of transmission of nerve impulse 4 11.11111111
   0.004521842
   GO:0031644∼regulation of neurologic system process 4 11.11111111
   0.005055558
   GO:0003677∼DNA binding 13 36.11111111 0.005249171
   GO:0030528∼transcription regulator activity 10 27.77777778 0.007137687
   GO:0048167∼regulation of synaptic plasticity 3 8.333333333 0.009379318
   GO:0003002∼regionalization 4 11.11111111 0.010129184
   GO:0006325∼chromatin organization 5 13.88888889 0.010398849
   GO:0006334∼nucleosome assembly 3 8.333333333 0.015765103
   GO:0031497∼chromatin assembly 3 8.333333333 0.016846638
   GO:0030182∼neuron differentiation 5 13.88888889 0.017077758
   GO:0048015∼phosphoinositide-mediated signaling 3 8.333333333
   0.017214047
   GO:0065004∼protein-DNA complex assembly 3 8.333333333 0.018336737
   GO:0034728∼nucleosome organization 3 8.333333333 0.019102083
   GO:0007389∼pattern specification process 4 11.11111111 0.022739147
   GO:0051276∼chromosome organization 5 13.88888889 0.023859659
   GO:0006323∼DNA packaging 3 8.333333333 0.029291461
   GO:0044057∼regulation of system process 4 11.11111111 0.033103168
   GO:0006333∼chromatin assembly or disassembly 3 8.333333333 0.03405299
   GO:0034622∼cellular macromolecular complex assembly 4 11.11111111
   0.035597851
   GO:0048666∼neuron development 4 11.11111111 0.041790392
   GO:0006355∼regulation of transcription, DNA-dependent 9 25.00000000
   0.042198386
   GO:0044093∼positive regulation of molecular function 5 13.88888889
   0.043420144
   GO:0051252∼regulation of RNA metabolic process 9 25.00000000
   0.047283874
   GO:0034621∼cellular macromolecular complex subunit organization 4
   11.11111111 0.04750734
   [72]Open in a new tab

Figure 2.

   Figure 2
   [73]Open in a new tab

   Hypothesized model of smoking-induced lung cancer based on our data.

Diagnosis model for distinguishing smoking-associated from nonsmoking lung
cancer

   To translate our results into clinical practice, we aimed to use the
   smoking-specific DNA methylation pattern as a diagnosis tool to
   classify lung cancer samples associated with smoking and those that are
   not. We thus used a machine learning algorithm known as the nearest
   centroid approach, which is a combined model using a tuning strategy of
   genetic algorithms to select robust DNA methylation markers to classify
   smoking and nonsmoking lung cancers. We found that a set of
   differentially methylated DNA methylation loci can predict
   smoking-associated lung cancer and nonsmoking lung cancer with a
   sensitivity of 83.2% and a specificity of 88.9%. The overall accuracy
   and principal component analysis (PCA) of the classification model is
   shown in [74]Figure 3.

Figure 3.

   Figure 3
   [75]Open in a new tab

   Diagnostic model for classifying smoking-associated lung cancer and
   nonsmoking lung cancer.

   Notes: Classification accuracy is shown with the horizontal axis
   represents the individual samples grouped according to the disease
   class whereas the vertical axis represents the predicted classes. Black
   denotes nonsmokers and red denotes smokers.

   Abbreviations: PC, principal component; Sensit, sensitivity; specif,
   specificity.

Discussion

   Epigenetic regulation has emerged to become an important field in
   studying the development of cancer. In this study, we conducted an
   epigenome-wide screening of DNA methylation patterns in lung
   adenocarcinoma samples from smokers and nonsmokers. By comparing normal
   lung tissue samples between smokers and nonsmokers, we identified 90
   differentially methylated regions comprising genes functioning in the
   energy metabolism process and mitogenic kinase activity, suggesting
   that smoking alters cellular energy supply and cell cycle regulation in
   normal lung cells. Since both aberrant energy metabolism and a
   dysregulated cell cycle are the hallmarks of cancer, smoking makes lung
   cells prone to the development of neoplastic disease. Interestingly,
   all 90 DMRs were hypomethylated in normal lung tissues from smokers
   compared with those from nonsmokers. Global DNA hypomethylation has
   been well demonstrated in many types of cancers.[76]^20 However, unlike
   the close link between DNA hypermethylation and suppression of tumor
   suppressors, the role of DNA hypomethylation in the development of
   cancer is still elusive.[77]^21 Several lines of evidence support that
   DNA hypomethylation contributes to tumorigenesis by inducing genomic
   instability.[78]^21 Our results indicate that smoking preferentially
   induces hypomethylation in normal lung tissues, which may predispose
   smokers to the development of lung cancer by inducing genomic
   instability.

   By comparing smoking-specific DNA methylation with lung cancer-specific
   DNA methylation, we discovered 110 differentially methylated genes that
   were specifically associated with smoking-induced lung cancer. These
   genes are enriched in genetic and epigenetic regulatory pathways such
   as transcription factor activity and chromatin remodeling activity.
   This result indicates that smoking-induced tumorigenesis requires
   extensive nuclear reorganization and reflects the fundamental
   characteristic differences between smoking-induced lung cancer and lung
   cancer not related to smoking, and also provides a potential diagnostic
   method to distinguish smoking-induced lung cancer from lung cancer not
   related to smoking.

   Many interesting genes that may function in smoking and cancer were
   identified in this study. We found that both RPS6KA3 and ARAF genes are
   hypomethylated in smokers but not in nonsmokers. Both genes take part
   in the MAPK kinase signaling pathway. The Ras/Raf/MEK/ERK (MAPK) signal
   transduction cascade is an important regulator of cell proliferation.
   RAS gene products activate proteins in the RAF family, which consists
   of ARAF, BRAF and RAF-1. In a lung cancer study, Lee et al identified
   the mutation spectrum revealed by paired genome sequences from a lung
   cancer patient, and suggested a model for how the multiplicity of
   mutations within the MAPK signaling cascade acted together to drive
   constitutive progrowth signals.[79]^22 They found that MAPK signaling
   genes (SHC1, GRB2, SOS, ARAF, AP3K3, and ELK1) were upregulated in lung
   cancer.[80]^22^,[81]^23 In addition, hypomethylation of MAPK signaling
   members in smokers is an indicator of oncogenic changes that may be
   valuable targets for early cancer detection.[82]^22^,[83]^23

   ADRB3, a beta-adrenergic receptor, was hypermethylated in smokers with
   lung cancer but not in nonsmokers with lung cancer in this study.
   Recent studies have revealed that nitrosamine
   4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK) can directly bind
   and stimulate the beta-adrenergic receptor signaling pathway. NNK is
   derived from nicotine, and has been demonstrated to cause mutations in
   genes that affect cell regulation and proliferation. ADRB3 also
   functions in activation of the neuroactive ligand receptor interaction
   pathway. In an earlier study, the researchers identified other common
   genetic variants associated with the risk of lung cancer in the NNK
   disposition pathways, such as CYP2A13, the most active P450 for phase
   metabolic activation of NNK and the receptor (ADRB2) in its
   nongenotoxic pathway.[84]^24 Masi et al verified the mutagenic effects
   of NNK in an animal model and provided a way of identifying mutations
   not only in ADRB2 but also in other genes that may play an essential
   role in the development of lung adenocarcinoma.[85]^25 Therefore, it
   will be interesting to know if ADRB3 also has a role in smoking-induced
   lung cancer.

   GALR1 is another interesting gene we identified in this study that is
   specific to smoking-associated lung cancer. GALR1 is one of the three G
   protein-coupled receptors that function as the receptor for galanin,
   which is a multifunctional neuropeptide widely expressed in the
   mammalian central and peripheral nervous systems. Expression of galanin
   and its three receptors (GALR1–3) are silenced in several types of
   tumor.[86]^26 Hypermethylation of GALR1 has been correlated with
   reduced gene expression. Reported by Misawa et al, GALR1 is silenced in
   head and neck squamous cell carcinoma because of its hypermethylated
   promoter. The methylation status of the GALR1 promoter and the level of
   GALR1 gene expression have been well correlated in a large panel of
   head and neck squamous cancer cell lines and primary tumor
   specimens.[87]^26 Ectopic expression of GALR1 suppresses tumor cell
   proliferation through Erk1/2-mediated regulations on cyclin-dependent
   kinase inhibitors and cyclin D1.[88]^27 Interestingly, Jackson et al
   utilized a mouse model to show that galanin has a protective role
   against neuropeptide dependence, and these effects may be mediated by
   GALR1. In addition, several lines of evidence support that galanin acts
   through GALR1 to modulate the physiologic effects of nicotine.[89]^28
   However, the role of GALR1 in smoking-induced lung cancer is still
   unclear and deserves further investigation.

   Smoking-associated lung cancer often demonstrates clinical and
   pathologic characteristics different from lung cancers that are not
   associated with smoking. Hence, it has clinical value for classifying
   smoking-associated lung cancer and nonsmoking lung cancer. In this
   study, we used smoking-specific DNA methylation loci to construct a
   diagnostic model to predict whether development of lung cancer is
   associated with smoking. In this study, our results were obtained from
   lung adenocarcinoma, which is the most common subtype of lung cancer.
   Because different types of lung cancer have distinct clinicopathologic
   characteristics, it should be noted that the value of our diagnostic
   model for other types of lung cancer is still unknown. In the next
   stage of study, we will need to validate this model further in a larger
   patient cohort and in other types of lung cancer.

   In summary, we examined differences in global DNA methylation patterns
   between smoker and nonsmoker groups of lung cancer patients to identify
   differentially methylated genes associated with smoking-induced lung
   cancer. Our results provide novel evidence that smoking plays an
   important role in the development of lung cancer through DNA
   methylation.

Supplementary tables

   Supplementary tables 1–4
   [90]http://www.dovepress.com/cr_data/supplementary_file_51041.pdf.
   [91]OTT-6-1471-S.pdf^ (437.4KB, pdf)

Acknowledgments