Genome-wide unmasking of epigenetically silenced genes in lung adenocarcinoma from smokers and never smokers

Summary

Lung cancer displays striking demographic, clinicopathological and molecular distinctions between smokers and never smokers. This study evaluated genome-wide abnormalities in methylation and identified multiple, novel, epigenetically silenced genes, with some differing in prevalence by smoking status and stage of lung cancer.

Abstract

Lung cancer in never smokers (NS) shows striking demographic, clinicopathological and molecular distinctions from the disease in smokers (S). Studies on selected genetic and epigenetic alterations in lung cancer identified that the frequency and profile of some abnormalities significantly differ by smoking status. This study compared the transcriptome of lung adenocarcinoma cell lines derived from S (n = 3) and NS (n = 3) each treated with vehicle (control), histone deacetylation inhibitor (trichostatin A) or DNA methylation inhibitor (5-aza-2′-deoxycytidine). Among 122 genes reexpressed following 5-aza-2′-deoxycytidine but not trichostatin A treatment in two or more cell lines (including 32 genes in S-only and 12 NS-only), methylation was validated for 80% (98/122 genes). After methylation analysis of 20 normal tissue samples and 14 additional non–small cell lung cancer cell lines (total 20), 39 genes frequently methylated in normal (>20%, 4/20) and 21 genes rarely methylated in non–small cell lung cancer (≤10%, 2/20) were excluded. The prevalence for methylation of the remaining 38 genes in lung adenocarcinomas from S (n = 97) and NS (n = 75) ranged from 8–89% and significantly differs between S and NS for CPEB1, CST6, EMILIN2, LAYN and MARVELD3 (P < 0.05). Furthermore, methylation of EMILIN2, ROBO3 and IGDCC4 was more prevalent in advanced (Stage II–IV, n = 61) than early (Stage I, n = 110) tumors. Knockdown of MARVELD3, one of the novel epigenetically silenced genes, by small interfering RNA significantly reduced anchorage-independent growth of lung cancer cells (P < 0.001). Collectively, this study has identified multiple, novel, epigenetically silenced genes in lung cancer and provides invaluable resources for the development of diagnostic and prognostic biomarkers.

Introduction

Lung cancer, the most frequent cause of cancer incidence and mortality worldwide, is responsible for ~1.6 million new cases and 1.4 million deaths each year (1). Although smoking remains the major cause of the disease, an increasing number of never smokers (NS) who smoked none or <100 cigarettes in their lifetime are being diagnosed with lung cancer. The World Health Organization estimates about 25% of all lung cancer cases worldwide are NS, which indicates that lung cancer in NS (LCNS) is the 7th largest cause of cancer-related mortality claiming 300 000 lives per year (2). Current estimates in the USA also show that LCNS is one of the top-10 causes of cancer mortality accounting for 10 000–15 000 deaths a year (3). The etiology of LCNS is less defined and could range from familial susceptibility to exposure to environmental tobacco smoke (second-hand smoke) and/or one or more of the environmental pulmonary carcinogens such as asbestos, crystalline silica, radon, polycyclic aromatic hydrocarbons and certain heavy metals. Second-hand smoke has now been classified as a human carcinogen with an average excess risk of lung cancer among regularly exposed NS by 20–30% (4,5). Epidemiological studies identified striking demographic, clinicopathological and molecular differences between lung cancer in smokers (S) and NS, indicating that the disease may develop through distinct carcinogenesis pathways (6).

LCNS is frequently associated with female gender, East Asian ethnicity and adenocarcinoma (AC) histologic subtype. AC has surpassed squamous cell carcinoma over the last few decades and became the most common lung cancer subtype. The increased incidence of LCNS together with the predominance of AC in this group appears to contribute to this remarkable shift. Among the molecular changes in lung cancer that show distinction based on smoking habits, cancer driver mutations of EGFR, HER2, PI3K and ALK rearrangements are more commonly seen in LCNS. In contrast, KRAS and LKB1 mutations are more prevalent in S (7). These distinct mutation patterns and frequencies are primarily responsible for the dramatic differences in the responsiveness of S and NS lung cancer patients to chemotherapies targeting pathways regulated by these genes. For instance, being a NS is the strongest clinical predictor of benefit from the EGFR tyrosine kinase inhibitors (6). Emerging evidence indicates that the molecular alterations distinguishing lung cancer between S and NS are not limited to these mutations. Thu et al. (8), recently reported that LCNS show greater proportion of genomic alteration and copy number gains on chromosomes 5q, 7p and 16p compared with S. Another molecular aberration that is commonly seen in lung cancer and potentially contributes to the clinical disparities between lung tumors from S and NS is epigenetic abnormality.

Aberrant promoter CpG island hypermethylation is one of the most common and well-established epigenetic abnormalities in cancer. Studies from our group and others have compared epigenetic silencing of selected candidate genes in lung cancer from S and NS. Toyooka et al. (9) evaluated the relationship between aberrant methylation of five tumor suppressor genes with mutually exclusive EGFR and KRAS mutations in lung cancer. The results revealed that the probability of having EGFR mutation was significantly lower among those with p16 ^INK4A and CDH13 methylation than in those without. In contrast, KRAS mutation was significantly higher in p16 ^INK4A methylated than unmethylated cases. The methylation index of these genes was significantly lower in EGFR mutant than wild-type cases and was marginally higher in KRAS mutant than wild-type cases. These results highlight differences in the involvement of epigenetic alterations between the EGFR- and KRAS-mediated tumorigenesis and suggest different interactions between genetic and epigenetic changes in lung carcinogenesis by smoking status (9). Similarly, our data from evaluation of the promoter CpG islands of 26 candidate genes revealed that aberrant methylation of three genes (TNFRSF10C, BHLHB5 and BOLL) were significantly more prevalent in lung AC from NS than S (10). Taken together, these genetic and epigenetic alterations support the premise that lung tumors in S and NS may develop through distinct molecular mechanisms. However, since these studies are based on the analysis of selected candidate genes, the genome-wide similarities and differences between lung tumors from S and NS are largely unknown.

Therefore, the objective of this study was to conduct a systematic, genome-wide screening of lung AC from S and NS for genes that are epigenetically silenced through aberrant promoter CpG island methylation. First, lung AC cell lines derived from S (n = 3) and NS (n = 3) were each treated with vehicle (0.006% ethanol in culture medium), the class I and II histone deacetylase inhibitor, trichostatin A (TSA), or the DNA methyltransferase inhibitor, 5-aza-2′-deoxycytidine (DAC). The genome-wide changes in gene expression were evaluated using a transcriptome array and compared between control versus TSA-treated cells and control versus DAC-treated cells. Genes that were reexpressed following DAC but not TSA treatment were selected as candidates for DNA methylation-mediated repression. A series of criteria targeting gene promoter regions with novel, aberrant and more prevalent methylation in lung cancer was applied to prioritize the selection of top candidates for validation. Multiple methylation assays were used to validate aberrant methylation of selected genes using normal human bronchial epithelial cells (NHBEC), non–small cell lung cancer (NSCLC) cell lines and primary lung tumors from S and NS. The relationship between methylation and gene silencing, the potential use of epigenetic drugs to induce reexpression of the silenced gene and the contribution of epigenetic repression to lung carcinogenesis were determined for a subset of aberrantly methylated genes based on novelty.

Materials and methods

Tissue samples and cell lines

Lung tumors from 172 AC patients were obtained from frozen tumor banks at Johns Hopkins and the Mayo Clinic. Normal lung tissue obtained most distant from the tumor in the resected lobe was available for a subset of these cases. NHBEC were collected from 10 cancer-free smokers at the University of New Mexico through diagnostic bronchoscopy (11). Peripheral blood mononuclear cells (PBMC) were obtained from 10 healthy donors. All samples were obtained with written informed consent from patients, and the study was approved by the institute’s Ethics Committee. Five human bronchial epithelial cell lines (HBEC1, 2, 3, 13 and 14) immortalized as described (12) were obtained from Drs Shay and Minna, Southwestern Medical Center, Dallas, TX. Twenty NSCLC cell lines, H23, H1568, H1993, H2023, H2085, H2228, H358, H441, H460, SKLU1, A549, H1838, H1299, H3255, Calu3, Calu6, H1435, H1975, HCC827 and HCC4006 obtained from and authenticated by the American Type Culture Collection (Manassas, VA) were also used. Experiments were conducted in cell lines passed for a maximum of 6 months postresuscitation.

DNA methylation and gene expression analysis

DNA extraction and modification were done as described (13) and 40ng of modified DNA was used per PCR. Combined bisulfite modification and restriction analysis (CoBRA), methylation-specific PCR (MSP) and bisulfite sequencing assays were performed as described (13). CoBRA and MSP primer sequences and amplification conditions for the final 38 genes are described in Supplementary Tables S1 and S2, available at Carcinogenesis Online, respectively. The CoBRA primers and amplification conditions were used for bisulfite sequencing. For genome-wide gene expression analysis, lung AC cell lines derived from S (H23, H1568 and H1993) and NS (H2023, H2085 and H2228) were each treated with growth medium containing 0.006% ethanol for 18h (vehicle), 300nM for 18h (TSA) or 500nM every 12h for 96h (DAC); RNA was extracted and the changes in gene expression were analyzed using the Agilent whole transcriptome array as described previously (13). For gene-specific expression, RNA was isolated as described (14) and 3 µg total RNA was reverse transcribed using the High Capacity cDNA Reverse Transcription Kit from Applied Biosystems (Foster City, CA) according to the manufacturer’s protocol. TaqMan assays from Applied Biosystems were used to quantify expression of each gene relative to the house keeping gene ACTB (β-actin) that was multiplexed in each reaction as an endogenous control. Samples were run at least twice in duplicates and expression of each target gene relative to the endogenous control (ΔCT) and the reference control samples (ΔΔCT) was calculated in fold change as described (15).

Gene knock down, cell proliferation and growth in soft agar

Cells were transfected with control (negative control #1) or gene-specific small interfering RNAs (siRNAs) from Applied Biosystems using Lipofectamine 2000 (Invitrogen, Santa Clara, CA) as described (16). Cell survival and migration were compared between control and gene-specific siRNA-transfected cells using tetrazolium reduction assay and wound closure assays, respectively, as described (14). Briefly, for cell migration, cells were plated in 6-well plates, transfected the next day, equal size wounds created 24 h posttransfection and pictures taken at 0, 24, 48, 72 and 96 h postwounding and wound closure was measured and calculated from the pictures. For cell proliferation, cells were plated in 24-well plates and transfected the next day and the number of live cells were compared at 24, 48, 72 and 96 h posttransfection using tetrazolium reduction assay. For growth in soft agar, cells were plated and transfected in 6-well plates and transferred into soft agar 24h posttransfection at a density of 6 × 10³ cells per well in 6-well plates. After 7 days in soft agar, cell colonies were fixed with methanol, stained with trypan blue and counted. Each experiment was conducted in triplicate and the results are shown as mean ± standard error of the mean. Genome-wide changes in gene expression between siControl- and siMARVELD3-transfected cells were evaluated 48 h posttransfection using Agilent whole transcriptome array as described (14).

Statistical analysis

Gene methylation and patient characteristics including age, gender, smoking status, tumor histology and performance were summarized with mean and standard deviation for continuous variables and proportions for categorical variables. Survival time was calculated from time of diagnosis until death or last follow-up and used to employ Kaplan–Meier plots, the log rank test and proportional hazards models. The association between methylation and patient characteristics was assessed by Fisher’s exact test. The effect of siRNA knockdown (siControl versus siTarget-Gene) on level of gene expression was compared using one-way analysis of variance. Tukey’s and Dunnett’s methods were used for pairwise and treatment-control comparison adjustments, respectively. The impact of potential outliers on the one-way analysis of variance values was controlled using non-parametric Wilcoxon Rank-sum test. All analyses were conducted in SAS 9.2.

Results

Genome-wide unmasking of epigenetically silenced genes in lung cancer

Lung AC cell lines derived from S (H23, H1568 and H1993) and NS (H2023, H2085 and H2228) were used for pharmacological reexpression of epigenetically silenced genes. Each cell line was treated with vehicle (control), TSA or DAC as described and genome-wide changes in gene expression was measured using the Agilent transcriptome array. The levels of gene expression in TSA- or DAC-treated cells were compared with the corresponding control cell line and the relative changes in gene expression across the entire transcriptome were determined as fold changes. The X and Y coordinates of each dot in Figure 1A represent the relative change in expression of 45°000+ transcripts in TSA- and DAC-treated H23, respectively. Due to the focus of this study, to identify genes primarily silenced by DNA methylation, top-tier genes whose expression increased by >2-fold (y-axis > 1.0 log base 2) after DAC treatment, but showed little or no change after TSA treatment (x-axis −0.5 to 0.5 log base 2), were selected for the study. A total of 448 genes met these criteria at least in two of the six cell lines. A strategy was developed (see Figure 1B) to prioritize the validation of novel aberrantly methylated genes that are commonly silenced in lung cancer. Thus, genes that are poorly annotated, without a promoter CpG island (as defined by Takai and Jones (17) and detailed under Figure 1B), already well-characterized for methylation in lung cancer, or normally regulated through methylation as seen in X chromosome, imprinted, or developmentally regulated, were excluded. Based on these criteria, 135 top candidate genes including 36 S-only [reexpressed in 2–3 S but none (0/3) NS-derived cell lines], 15 NS-only (0/3 S and 2–3 NS) and 84 common (≥1 S and ≥1 NS cell line) were selected for methylation analysis (Supplementary Table S3, available at Carcinogenesis Online).

Fig. 1.

Pharmacological unmasking of epigenetically silenced genes in lung cancer. (A) Genome-wide changes in gene expression profile of the lung AC cell line H23 treated with TSA or DAC. The Agilent transcriptome array was used to interrogate level of expression across the whole genome and expression of each gene in TSA- or DAC-treated cells was compared with Control. The X and Y coordinates of each dot represent expression changes of a gene in the TSA- and DAC-treated cells, respectively. Genes located in section II of the plot were considered the top-tier candidates for methylation-mediated repression. (B) A strategy developed to prioritize the selection of the most common, novel, aberrantly methylated genes in lung cancer for validation studies.

Validation of aberrant methylation in lung cancer

CoBRA assays were developed and used to accurately determine the methylation status of CpG islands within the promoter regions of 122 genes. Multiple assay designs and optimization attempts failed for 13 of the 135 genes. Methylation of 98 genes was confirmed in one or more NSCLC cell line, whereas 24 genes were unmethylated in all cell lines equating to over 80% (98/122) success rate of methylation discovery in lung cancer. Similarly, tumor specificity of methylation of these 98 genes was screened by CoBRA using NHBEC and PBMC obtained from five cancer-free donors. A total of 23 genes, 7 methylated in nearly all PBMC but not NHBEC and 16 methylated in both NHBEC and PBMC, were further excluded for being methylated in normal tissue. Due to the high sensitivity of MSP assays (0.1%) (18), tumor specificity of methylation of the remaining 75 genes was further tested by MSP using NHBEC and PBMC samples from 10 cancer-free donors. An additional 16 genes were further excluded due to methylation in 2 or more normal samples. Of the remaining 59 genes that show tumor-specific methylation (methylated in ≤20% normal samples), 21 additional genes were also excluded for being rarely (2/20, 10%) methylated in NSCLC cell lines. The final 38 genes that show tumor-specific methylation in at least two NSCLC cell lines by both assays were evaluated by MSP using 172 primary lung AC from 38 current smokers (CS), 59 former smokers (FS) and 75 NS. The overall prevalence for methylation of these genes in lung AC ranged from 8 to 89% and aberrant methylation is shown for the first time for 30 genes in lung cancer. This includes 12 genes not previously recognized as regulated by methylation in any cancer type (Table I). The density and distribution of methylation across the promoter CpG island of a subset of these genes was determined through bisulfite sequencing and validated the methylation data from CoBRA and MSP assays (Figure 2A–D, Supplementary Figure S1A–D, available at Carcinogenesis Online). CoBRA assays using a subset of genes with densely methylated promoter CpG islands showed partial demethylation following DAC but not Vehicle or TSA treatment (Supplementary Figure S1E and F, available at Carcinogenesis Online).

Table I.

Prevalence for aberrant methylation of genes in lung AC from current, former and NS

Gene name	Genomic loci	# CpG island	Methylation n (%)
			Total (n = 172)	CS (n = 38)	FS (n = 59)	NS (n = 75)
ACP5 a	19p13.2	35	53/169 (31)	13/37 (35)	18/57 (32)	22/75 (29)
ADRBK2 a	22q12.1	41	43/168 (26)	7/37 (19)	19/58 (33)	17/73 (23)
BNIP3	10q26.3	181	42/171 (25)	4/37 (11)	14/59 (24)	24/75 (32)
CNRIP1 b	2p14	75	110/171 (64)	20/38 (53)	39/58 (67)	51/75 (68)
CCDC50 a	3q28	81	69/171 (40)	12/38 (32)	21/58 (36)	36/75 (48)
CPEB1 b	15q25.2	218	88/172 (51)	27/38 (71)	28/59 (47)	33/75 (44)
CST6 b	11q21.2	59	80/171 (47)	9/38 (24)	22/58 (38)	49/75 (65)
CYP1B1	2p22.2	300	23/169 (14)	3/38 (8)	6/58 (10)	14/75 (19)
DFNA5 b	7p15.3	102	18/167 (11)	2/36 (6)	10/56 (18)	6/75 (8)
DKK1	10q21.1	64	103/171 (60)	23/38 (61)	31/58 (53)	49/75 (65)
DOCK10 a	2q36.2	95	13/170 (8)	3/37 (8)	7/58 (12)	3/75 (4)
EMILIN2	18p11.32	146	45/172 (26)	13/38 (34)	18/59 (31)	14/75 (19)
GATA2	3q21.3	514	159/165 (96)	31/34 (91)	53/56 (95)	75/75 (100)
GJB2 b	13q13.11	145	74/172 (43)	14/38 (37)	21/59 (36)	39/75 (52)
GPX7 b	1p32.3	91	26/170 (15)	7/37 (19)	11/58 (19)	8/75 (11)
GSTT2 a	22q11.23	55	109/172 (63)	25/38 (66)	38/59 (64)	48/75 (64)
HES5 b	1p36.32	105	35/169 (21)	6/37 (16)	12/57 (21)	17/75 (23)
HTRA3	4p16.1	100	103/172 (60)	23/38 (61)	38/59 (64)	42/75 (56)
HTRA4 a	8p11.22	78	83/170 (49)	13/38 (34)	24/57 (42)	46/75 (61)
IGDCC4 b	15q22.31	89	72/172 (42)	16/38 (42)	29/59 (49)	27/75 (36)
LAYN a	11q23.1	110	96/172 (56)	17/38 (45)	29/59 (49)	50/75 (67)
LOX	5q23.1	136	81/172 (47)	18/38 (47)	27/59 (46)	36/75 (48)
MARVELD3 a	16q22.2	104	75/171 (44)	12/38 (32)	19/58 (33)	44/75 (59)
MSRB3 b	12q14.3	88	42/170 (25)	8/37 (22)	10/58 (17)	24/75 (32)
MSX1	4p16.2	96	76/171 (44)	20/38 (53)	25/58 (43)	31/75 (41)
MT1M b	16q12.2	68	112/172 (65)	20/38 (53)	37/59 (63)	55/75 (73)
PAK6 b	15q15.1	116	54/172 (31)	10/38 (26)	17/59 (29)	27/75 (36)
PDE4D a	5q11.5	120	152/170 (89)	29/37 (78)	52/58 (90)	71/75 (95)
RBP1 b	3q23	63	87/171 (51)	21/38 (55)	28/58 (48)	38/75 (51)
ROBO3 b	11q24.2	111	116/172 (67)	21/38 (55)	42/59 (71)	53/75 (71)
SYT13 a	11p11.2	52	129/170 (76)	25/36 (69)	45/59 (76)	59/75 (79)
TMEFF1 a	9q31.1	132	24/169 (14)	1/37 (3)	7/57 (12)	16/75 (21)
TMEM22 b	3q22.3	135	81/172 (47)	19/38 (50)	29/59 (49)	33/75 (44)
TNFSF12 b	17p13.1	48	23/170 (14)	6/37 (16)	6/58 (10)	11/75 (15)
TPM1 b	15q22.2	154	91/169 (53)	17/37 (46)	32/57 (56)	42/75 (56)
VAV3 b	1p13.3	78	44/172 (26)	13/38 (34)	14/59 (24)	17/75 (23)
ZNF365 b	10q21.2	67	51/171 (30)	12/37 (32)	18/59 (31)	21/75 (28)
ZNF655 a	7q22.1	90	17/171 (10)	5/38 (13)	8/58 (14)	4/75 (5)

^aAberrant gene methylation in any other cancer is reported for the first time.

^bAberrant gene methylation in lung cancer is reported for the first time.

Fig. 2.

Validation of aberrant methylation within the promoter CpG islands of selected genes. CoBRA assays revealed aberrant methylation of EMILIN2 (A) and CPEB1 (B) promoters in some lung AC cell lines (H1838 and SKLU1) and primary tumors (AC#s). Methylated samples show bands of smaller digested fragments in the BstU-1 + lanes. Methylation was not detected in NHBEC (e.g. N#30). Bisulfite sequencing was used to assess the density and distribution of methylation across the promoter CpG islands of EMILIN2 (C) and CPEB1 (D) and confirmed methylation results from CoBRA and MSP assays. Each oval represents a CpG. Ovals in a row indicate the number of CpGs sequenced across the promoter CpG island of a clone from a particular sample. At least five clones per sample were sequenced. Filled and open ovals represent CG (methylated) and TG (unmethylated) sequences, respectively.

Association of gene methylation with disease progression

The potential association between methylation of specific genes with disease progression was evaluated. Among the 172 tumors analyzed, 110 were obtained from early stage (Stage IA and IB) and 61 from advanced stage (Stages II–IV) lung AC patients, whereas the stage was unknown for one patient. As expected, Kaplan–Meier curves, log rank test and Cox regression model analyses after adjustment for age, gender, race and smoking habits showed that patients whose tumors were resected at early stage survived significantly longer (P < 0.001) compared with patients whose tumors were resected at an advanced stage (Figure 3A). The median overall survival and the 5-year survival rates from the date of lung cancer diagnosis, respectively, were 86 months and 54% for early stage compared with 29 months and 36% for advanced stage patients. Methylation of three genes, EMILIN2, ROBO3 and IGDCC4, was found to be more prevalent in advanced stage lung tumors (41, 52 and 77%) compared with early stage tumors (18, 35 and 62%), respectively (Table II). Logistic regression analysis after adjustment for age, gender, race and smoking status revealed that the prevalence for methylation of EMILIN2 (P = 0.004; odds ratio: 0.346, 95% confidence interval: 0.168–0.715) and ROBO3 (P = 0.05; odds ratio: 0.488, 0.235 – 1.010) was significantly higher in advanced stage tumors (Table II). The higher prevalence for methylation of IGDCC4 in the advanced than early stage tumors showed borderline significance (P = 0.07; odds ratio: 0.552, 0.287–1.061). However, despite the strong association seen between stage and overall survival in these patients, methylation of these genes was not significantly associated with patient survival (data not shown). The prevalence for methylation of the remaining genes did not significantly differ between early and advanced tumors (Supplementary Table S4, available at Carcinogenesis Online).

Fig. 3.

Promoter methylation repressed gene expression. (A) As expected, Kaplan–Meier Curve showed that lung AC patients with early stage (Stage I) tumor at the time of diagnosis and resection survived significantly better than patients with advanced (Stage II–IV) tumor. Expression of CPEB1 (B) and ROBO3 (C) in cell lines with methylated promoter was dramatically repressed or completely silenced compared with normal human bronchial epithelial or lung cancer cell lines with unmethylated promoter. Expression of CPEB1 (D) and ROBO3 (E) could be restored or significantly increased in cells with methylated promoter primarily by DAC and in some cases by TSA treatment.

Table II.

Genes with significantly different methylation prevalence by stage

Genes	Methylation n (%)		P values		Odds ratio
	Early (Stage I)	Advanced (Stage II–IV)	Fisher exact test	Logistica regression
EMILIN2	20/110 (18%)	25/61 (41%)	0.002	0.004	0.35 (0.17–0.72)
IGDCC4	39/110 (35%)	32/61 (52%)	0.04	0.07	0.55 (0.29–1.06)
ROBO3	68/110 (62%)	47/61 (77%)	0.06	0.05	0.49 (0.24–1.01)

^aMethylation status of each gene in early stage lung tumors was used as a reference to calculate the P values after adjustment for age, gender, race and smoking habits.

Distinct aberrant methylation of genes by smoking history

The relationship between aberrant gene methylation and patients’ smoking history was compared between lung AC from CS, FS and NS. The prevalence for methylation of five genes (CPEB1, CST6, EMILIN2, LAYN and MARVELD3) showed significant difference based on smoking status (Table I). Logistic regression analysis after adjustment for age, gender, race and stage of lung cancer revealed that methylation of CPEB1 and EMILIN2 was significantly more prevalent in CS compared with NS (Table III). Whereas methylation of CPEB1 in FS was comparable with NS, methylation of EMILIN2 in FS was similar to CS. In contrast, methylation of CST6, LAYN, and MARVELD3 was more prevalent in NS than either CS or FS. The impact of lifetime cigarette smoke exposure on methylation regardless of current smoking status was compared between ever smokers (CS + FS) and NS. Similarly, the relationship between methylation and current smoking status was compared between CS and current nonsmokers (FS + NS). As shown in Supplementary Table S5, available at Carcinogenesis Online, methylation of all five genes was significantly different between ever smokers and NS, whereas only CPEB1 and CST6 showed statistically significant difference between CS and current nonsmokers.

Table III.

Genes with significantly different methylation prevalence by smoking status

Genes	Smoking status	P valuesa	Odds ratioa
CPEB1	Current smoker	0.001	0.20 (0.07–0.53)
	Former smoker	0.26	0.63 (0.29–1.41)
CST6	Current smoker	0.0001	7.44 (2.72–20.3)
	Former smoker	0.001	3.86 (1.70–8.80)
EMILIN2	Current smoker	0.04	0.34 (0.13–0.94)
	Former smoker	0.15	0.52 (0.21–1.28)
LAYN	Current smoker	0.06	2.32 (0.95–5.66)
	Former smoker	0.05	2.22 (1.01–4.85)
MARVELD3	Current smoker	0.011	3.28 (1.31–8.25)
	Former smoker	0.008	2.97 (1.34–6.60)

^aMethylation status of each gene in never smokers was used as a reference to calculate the P values using generalized logistic regression after adjustment for age, gender, race and stage.

Reexpression of methylation-silenced genes with epigenetic drugs

The impact of promoter CpG island methylation on the level of gene expression was evaluated for a subset of genes and the responsiveness of these genes to TSA or DAC was tested in vitro. Lung cancer cell lines with defined methylation status were selected for each gene and the level of gene expression was compared between methylated and unmethylated cell lines. As shown for CPEB1 and ROBO3 genes in Figure 3B and C, NSCLC cell lines with a densely methylated promoter show strongly repressed or completely silenced expression compared with NSCLC cell lines in which these genes are unmethylated. For example, CPEB1 expression in three NSCLC cell lines (H23, H2228 and Calu3) where the promoter of the gene is unmethylated (Figure 2B and D) is ~200- to 1000-fold higher than its expression in HBEC2 (Figure 3B). A plausible explanation for this dramatic induction of CPEB1 is given in the discussion. In contrast, in H358, Calu6 and SKLU1 where the promoter of CPEB1 is densely methylated (Figure 2B and D), the mean expression of CPEB1 is about 7-fold lower than in HBEC2 (Figure 3B). Genes that are silenced by methylation also show the most response to epigenetic drug treatments. For example, the expression of CPEB1 in Calu6 and SKLU1 was, respectively, increased by an average of 10-fold or 600-fold following TSA or DAC treatment compared with vehicle-treated cells (Figure 3D). In contrast, neither TSA nor DAC induced further expression of CPEB1 in the cell lines H23 and Calu3 where the gene is unmethylated. Similar repression by methylation (Supplementary Figure S1A and C, available at Carcinogenesis Online) and reexpression mainly by DAC and to a lesser extent by TSA was demonstrated for ROBO3 (Figure 3C and E).

Impact of epigenetic silencing on cancer cell properties

MARVELD3 is one of the genes whose methylation is reported for the first time in cancer and significantly associated with patients’ smoking habits. Reduced expression of MARVELD3 that is associated with Snail-induced epithelial–mesenchymal transition (EMT) and poorly differentiated pancreatic cancer has been recently reported (19). Thus, it was selected for functional studies. CoBRA and bisulfite sequencing demonstrated dense methylation of MARVED3 promoter CpG island in some NSCLC cell lines and TaqMan assays revealed significantly reduced or completely silenced expression in most NSCLC cell lines including some with unmethylated (e.g. H23) or partially methylated (e.g. SKLU1) promoters compared with HBECs (Figure 4A–C). Expression of MARVELD3 could be partially or completely restored by DAC or TSA treatment supporting epigenetic repression (Figure 4C and D). To mimic the epigenetic silencing of MARVELD3 in lung cancer cell lines, siRNA was used to transiently knock down its expression in Calu3 and A549 cell lines. MARVELD3 expression was reduced by 65 to 89% in Calu3 and A549 cells transfected with siMARVELD3 compared with siControl for at least 8 days post transfection (Figure 4E). The impact of this repression on cell proliferation/survival and anchorage-independent growth was evaluated using tetrazolium reduction assay and soft agar assays, respectively. Knock down of MARVELD3 did not affect cell proliferation/survival, but surprisingly, dramatically reduced growth in soft agar in both cell lines (Figure 4F and G and Supplementary Figure S2A and B, available at Carcinogenesis Online). Reduction of MARVELD3 expression was also strongly associated with loss of CDH1 (E-cadherin) and moderate increase of SNAI2 (Snail2) expression (Supplementary Figure S2C, available at Carcinogenesis Online). Comparison of genome-wide gene expression changes in Calu3 cells transfected with these siRNAs showed that siMARVELD3 altered the expression of 656 genes by >2-fold compared with siControl (Supplementary Table S6, available at Carcinogenesis Online). Pathway analysis of these differentially expressed genes revealed significant (P < 0.001) effects on 23 different pathways, primarily involving the immune response (12 pathways), cell adhesion and EMT (3), development (3), cytokine production (2) and metabolism (2) (Figure 4H).

Fig. 4.

Epigenetic silencing of MARVELD3 modulates cancer cell phenotypes. CoBRA (A) and bisulfite sequencing (B) were used to demonstrate methylation of MARVELD3 promoter CpG island in lung cancer. (C) Compared with NHBEC, MARVELD3 is markedly repressed or completely silenced in most NSCLC cell lines including some with unmethylated promoter. (D) In cell lines where MARVELD3 was silenced with densely methylated promoter (e.g. H1299), expression could only be restored by DAC. However, TSA treatment was also as effective as DAC in reexpressing MARVELD3 in cell lines where the gene is repressed without methylation of its promoter (e.g. SKLU1), suggesting repression through histone deacetylation. (E) Compared with Calu3 cells transfected with negative control siRNA (siControl), reduction of MARVELD3 expression by siMARVELD3 for up to 7 days posttransfcetion did not affect cell proliferation/survival (F) but significantly reduced growth in soft agar (G) and modulated multiple pathways (H).

Discussion

This study employed pharmacological reexpression of epigenetically silenced genes, compared expression changes across the whole genome of lung cancer cells derived from S and NS and identified multiple, novel, abnormally methylated genes in lung cancer. Aberrant methylation of 38 genes was validated in large sets of lung ACs. Among these, methylation of 30 genes in lung cancer is described for the first time including 12 genes whose methylation in cancer as a whole is novel. Aberrant methylation of three genes (EMILIN2, ROBO3 and IGDCC4) was significantly associated with disease progression, whereas methylation of five genes (CPEB1, CST6, EMILIN2, LAYN and MARVELD3) differs significantly based on patients’ smoking habits. The impact of methylation on transcriptional repression and the utility of epigenetic drugs to reexpressing silenced genes are demonstrated for a subset of genes. Finally, among the genes with novel methylation in cancer, the impact of epigenetic repression of MARVELD3 on cancer cell property (anchorage-independent growth) was demonstrated. Our group and others have demonstrated that aberrant methylated genes detected in clinical samples could serve as important biomarkers for predicting lung cancer risk, early detection and prognosis and are currently working on improving the sensitivity and specificity of such assays (20–23). Thus, the discovery and validation of novel methylation changes in a large sample of primary lung tumors provide a new resource of candidate biomarkers.

Our approach was to improve the accuracy of discovering genes silenced through cytosine methylation by excluding TSA responsive genes that are primarily repressed via histone modification. The rationale for this strategy came from previous studies that show, unlike DAC, TSA alone could not reexpress genes silenced through promoter hypermethylation (24,25). The study achieved over 80% (98/122) success rate in discovering true methylation, indicating a potential use of this approach for the discovery of methylation-silenced genes in cancer with a relatively high specificity. However, this strategy might have overlooked some methylated genes that respond to both or neither treatments. CST6, which was discovered by the array as a top-tier (methylation candidate) gene in each (3/3) NS-derived lung AC cell lines but none (0/3) S-derived lung AC cell lines is a good example. The top-tier status (>2-fold increased expression after DAC, but little or no change after TSA treatments) of CST6 only in the NS cell lines was validated by TaqMan assays (Supplementary Figure S3A, available at Carcinogenesis Online). Among the S cell lines, CST6 expression was increased >2-fold by both TSA and DAC in H23 and H1993, whereas H1568 responded to neither treatment. However, CoBRA assays revealed that the promoter CpG island of CST6 is methylated in all S and NS cell lines (Supplementary Figure S3B, available at Carcinogenesis Online), indicating that our array-based method overlooked its methylation in S cell lines. This misclassification could be overcome by combining pharmacological reexpression assays with direct assessment of methylation using genome-wide methylation arrays.

The discovery of genes whose methylation differs based on disease progression and smoking status will help to improve the classification of patients on the basis of these categories. The significantly higher prevalence for methylation of EMILIN2, ROBO3 and IGDCC4 in advanced lung AC than in early stage disease indicates the potential use of these epigenetic changes as prognostic biomarkers. EMILIN2 is an extracellular matrix glycoprotein that can directly bind to cell surface death receptors (mainly DR4) and induce cell death via the extrinsic apoptotic pathway (26,27). Aberrant methylation of EMILIN2 has been previously reported in breast, lung, colorectal and prostate cancers indicating that epigenetic inactivation of this extracellular matrix glycoprotein is a common feature among solid tumors (28). Although this is the first report of ROBO3 methylation in lung cancer, aberrant methylation of this gene has been previously reported in cervical cancer (29). Similar to our findings in lung cancer, the prevalence for methylation of ROBO3 was also increased during disease progression from 0% in normal tissue (n = 41) and low-grade squamous intraepithelial lesions (n = 47) to 10% in high-grade squamous intraepithelial lesions (n = 39) and 36% in cervical cancer samples (n = 118).

Among the genes whose methylation in lung cancer varies based on patients’ smoking habits, methylation of CPEB1 and EMILIN2 were more prevalent in current smokers than NS. CPEB1 is an important mediator of cell senescence and regulator of messenger RNA translation. It binds to responsive messenger RNAs including the TP53 messenger RNA and regulates polyadenylation that is essential for translation (reviewed in ref. 30). Cpeb1 knockout mouse embryo fibroblasts become immortal by avoiding senescence, whereas ectopic Cpeb1 expression restores senescence in the knockout mouse embryo fibroblasts and induces premature senescence in the intact mouse embryo fibroblasts (31,32). Similarly, CPEB1 knockdown in human skin and lung cells reduced TP53 protein levels by 50% and caused bypass of senescence (33). Epigenetic silencing of CPEB1 in cancer was first observed in multiple myeloma where 50% of primary tumors and 67% of cell lines showed aberrant methylation (34). Subsequent studies show frequent methylation of CPEB1 in gastric (35) and lung (this study) cancer, suggesting it might be a common abnormality in solid tumors. We found that CPEB1 expression in some lung AC cell lines where its promoter is unmethylated such as H23, H2228 and Calu3 is much higher (on average ~500-fold) compared with normal bronchial epithelial cells. This raises the question as to how these cells could continue to proliferate in the presence of such a high expression of senescence-causing gene. Two independent studies demonstrated that CPEB1-induced senescence is dependent on p53 and increased expression of CPEB1 in the absence of p53 was unable to induce senescence (32,33). Interestingly, TP53 is mutated in each of these three lung AC cell lines indicating a plausible mechanism by which these cells could avoid CPEB1-mediated senescence. We speculate that the dramatic increase in CPEB1 expression in these cells could be a futile attempt of suppressing proliferation through CPEB1-mediated senescence.

This study also identified that methylation of CST6, LAYN and MARVELD3 was significantly more prevalent in NS than either CS or FS. Epigenetic silencing of CST6 via promoter methylation and its impact on the tumor suppressor role of the gene have been described for a number of solid tumors (36–40). In contrast, aberrant methylation of LAYN and MARVELD3 in lung or any other cancer is reported for the first time. LAYN encodes for hyaluronan receptor that is expressed on the apical side of human airway epithelium. A recent study indicated that cigarette smoke induces fragmentation of hyaluronan, which binds to its receptor LAYN to promote a signaling pathway leading to downregulation of E-cadherin and loss of epithelial barrier function (41). However, the role of epigenetic silencing of LAYN during lung or other carcinogenesis remains to be discovered.

MARVELD3 is a member of the tight junction-associated occludin family of proteins that form transmembrane component of tight junctions in epithelial and endothelial cells (42). Depletion of MARVELD3 in well-differentiated pancreatic cancer cells using siRNA increased permeability and weakened cell–cell contact and barrier functions. MARVELD3 is also repressed during Snail-induced EMT and during pancreatic cancer progression (19). Our data show that siRNA-mediated repression of MARVELD3 in Calu3 and A549 cell lines dramatically repressed anchorage-independent growth. Although the mechanism by which MARVELD3 contributes to these apparently contradicting tumor suppressor and promoter properties in pancreatic and lung cancer, respectively, is not clear, these phenotypes have been shown to coexist during EMT. Oyanagi et al. (43) recently used a three-dimensional (3D) collagen gel culture, which mimics a physiological environment, and demonstrated that induction of EMT in carcinoma cell lines including A549 not only enhanced cell motility and adhesiveness to fibronectin and collagen but also extremely suppressed their growth rates in soft agar and collagen gel cultures. The marked repression of CDH1 (E-cadherin) and moderate increase of SNAI2 (Snail2) expression we found in lung cancer following MARVELD3 knockdown suggests that the dramatic loss of growth of these cells in soft agar could be part of the EMT process. Together, our studies have identified many new epigenetically regulated genes that probably contribute to the development and progression of lung cancer.

Supplementary material

Supplementary Tables S1–S6 and Figures S1–S3 can be found at http://carcin.oxfordjournals.org/

Funding

Supported by National Institute of Health (R01 ES008801, R01CA089551, R01CA095568).

Conflict of Interest Statement: None declared.

Glossary

Abbreviations:

AC

adenocarcinoma

CoBRA

combined bisulfite modification and restriction analysis

CS

current smokers

DAC

5-aza-2′-deoxycytidine

EMT

epithelial–mesenchymal transition

FS

former smokers

LCNS

lung cancer in NS

MSP

methylation-specific PCR

NS

never smokers

NSCLC

non–small cell lung cancer

NHBEC

normal human bronchial epithelial cells

PBMC

peripheral blood mononuclear cells

siRNAs

small interfering RNAs

TSA

trichostatin A.