Class I HDACs Are Mediators of Smoke Carcinogen–Induced Stabilization of DNMT1 and Serve as Promising Targets for Chemoprevention of Lung Cancer

Epigenetic alterations have been identified as key events in the pathogenesis of non–small cell lung cancer (NSCLC) carcinogenesis (1). Aberrant methylation is common in lung cancer precursor lesions such as dysplasia and atypical adenomatous hyperplasia (2, 3). Detection of aberrant methylation in the sputum of either current or former smokers can serve as a marker for increased lung cancer risk (4). The DNA methyltransferase (DNMT)1 mediates the transfer of acetyl-groups from S-adenosyl-methionine to cytosine residues in the DNA and is required for maintenance of DNA methylation (5). Smoke carcinogen exposure leads to increased DNMT1 protein expression and subsequent de novo methylation (6–8). Inhibition or knockdown of DNMT1 leads to decreased colony formation of transformed bronchial epithelial cells (7) and decreased tumor counts in mouse models of carcinogen-induced lung cancer (9), implicating DNMT1 as a critical mediator of early smoking related bronchial carcinogenesis.

DNMT1 functions during S phase and its protein levels are tightly regulated throughout the cell cycle (10). DNMT1 turnover is regulated by posttranslational modifications such as acetylation (11–13), phosphorylation (14), and methylation (15, 16). The dominant mechanism for increased DNMT1 protein expression after carcinogen exposure has not yet been determined (6–8).

Histone deacetylases (HDAC) were originally identified as transcriptional repressors, counteracting the activity of histone acetyltransferases that activate transcription by acetylation of histone tails, thereby loosening the DNA/core histone interaction and providing a permissive chromatin state for transcriptional machinery. To date several classes of HDACs have been identified: class I HDACs (HDAC-1, -2, -3, and -8) are mostly localized in the nucleus and target proteins involved in the regulation of gene transcription. HDAC3 is unique in this list because it is also found in the cytoplasm. Here, it is involved in src signaling and has been found to act as a chaperone for the TR2 receptor in promyelocytic leukemia (17).

In lung cancer, increased mRNA levels of HDAC1 have been associated with higher stage and worse prognosis (18, 19). Similar findings have been reported for HDAC3 (20). Recently, genome-wide analyses of copy number changes, DNA methylation patterns, and gene expression changes in a large set of lung cancers of adeno- and squamous cell (SCC) origin revealed important differences in gene expression patterns between these 2 histologic subtypes. Furthermore, in silico compound screens suggested that the pattern of gene expression signature associated with SCC in particular may be altered by HDAC inhibition (21).

Clinically, HDAC inhibitors have shown promise for the treatment of advanced NSCLC either in combination with chemotherapy (22) or in combination with the demethylating agent 5-azadeoxycytidine (23). These studies focused on advanced cases of lung cancer and show a somewhat limited efficacy of HDAC inhibition in treating lung cancer in this setting. Here, we focused on the role of HDACs in early lung carcinogenesis. We used a model of long-term exposure of human bronchial epithelial cells to smoke carcinogens and examine the relationship between HDACs and DNMT1 during, and to investigate strategies to target, this critical early event in bronchial carcinogenesis.

We identified a biochemical interaction between DNMT1 and class I HDACs after carcinogen exposure as the primary mechanism responsible for DNMT1 protein stabilization and upregulation. Furthermore, we find a significant increase in DNMT1 and class I HDACs in primary lung tumor samples, and characterize the effects of HDAC inhibitors in overcoming tobacco-induced epigenetic changes. This study points to a potential role for HDAC inhibition in chemoprevention of aerodigestive carcinogenesis.

DNMT1 (NM_001130823.1), HDAC1, and HDAC3 cDNAs were obtained from Open Biosystems. An N-terminally truncated DNMT1 (encoding aa 121-1616) cDNA and shSET7 and control vectors were gifted by Dr. P. Vertino (24). HDAC2 cDNA was purchased from DNASU (ORFeome consortium). cDNAs were cloned into pDEST51 vectors expressing a C-terminal H6-V5 tag (Invitrogen) and into pDEST-27 vectors expressing an N-terminal glutathione S-transferase (GST) tag. Vectors were transfected using Lipofectamine-2000 (Invitrogen).

Human bronchial epithelial cells immortalized with hTERT and cdk4 (3KT; ref. 25) were exposed to 500 μmol/L methylnitrosourea (MNU) and 50 nmol/L benz(a)pyrene (BaP) for 24 hours per week with 6 days of outgrowth after exposure (7). 3KT cells treated in this manner for 31 weeks are designated T31. 3KT cells were exposed to vehicle control and cultured in parallel for the same duration to account for possible changes in epigenetic gene regulation induced by long-term cultures. Stable lines expressing DNMT1 were established with treatment of transfected T31 cells with 10 μmol/L blasticydin. Stable HDAC3 overexpressing cells were established by treating transfected 3KT cells with 10 μmol/L blasticydin. In some experiments, valproic acid (VPA) was added at 0.1 to 1 mmol/L for indicated time. MG-132 was added at 2.5 μmol/L for 16 hours or 25 μmol/L for 3 hours as indicated. All cell lines used were tested for mycobacterium contamination by Bionique testing labs. 3KT and T31 were authenticated through short tandem repeat analysis by Biosynthesis Inc.

3KT and carcinogen exposed cells were seeded at 1,000 cells per well in 0.35% agarose containing growth medium into 6-well cell-culture dishes coated with 0.5% bottom agar containing RPMI/10% FBS/1%pen-strep. Feeding medium was added to the top agar once solidified and replenished 3 times weekly. Sodium valproate (VPA) was added in the top agar and feeding medium as indicated. Cells were allowed to grow for 21 days, stained with crystal violet, and scored on a low-power objective light microscope.

Cells expressing H6-tagged truncated DNMT1 were lysed in NTA lysis buffer (50 mmol/L sodium phosphate, 300 mmol/L NaCl, 10 mmol/L Imidizole, 0.05% Tween-20 pH 8) containing phenylmethylsulfonylfluoride (PMSF), protease, and phosphatase inhibitors. After sonication and clarification by centrifugation, DNMT1 was affinity purified using nickel-nitrilotriacetic acid magnetic agarose (Qiagen). Cells expressing GST-tagged HDAC and V5-tagged DNMT1 proteins were lysed in 1% NP40 buffer (50 mmol/L Tris pH 8, 150 mmol/L NaCl, 0.5 mmol/L EDTA, 1% NP40) containing PMSF, protease, and phosphatase inhibitors. GST-tagged proteins were isolated using a glutathione sepharose resin (GE healthcare) whereas V5-tagged constructs were immunoprecipitated using V5 antibodies (Invitrogen) and eluted with V5 peptide (Alpha Diagnostic). In Coimmunoprecipitation experiments, DNAseI (900 U/mL; Invitrogen) was added before precipitation to rule out the possibility that a protein–protein interaction may be mediated by chromatin.

Cells were synchronized with 5 μmol/L aphidocholine for 24 hours, released for the indicated time (0, 6, 12, and 24 hours), and fixed in 70% ice cold ethanol. Cells were stained with 7-aminoactinomycin D (7AAD) 250 ng/mL in the presence of 100 μg/mL RNase A. Cells were counted on a BD-FACSCANTO II instrument and analyzed on DIVA and FLOW-Jo software.

Cells were lysed in 1× cell lysis buffer (Cell Signaling), containing complete protease inhibitor and Phostop (Roche) and 1 mmol/L PMSF. Cells were sonicated briefly and lysates clarified by centrifugation. Following SDS-PAGE and semi-dry transfer, the following antibodies were used: acetylated lysine, EZH2, G9a, H3K27Me3, (Cell Signaling), H3K9Me3 (Abcam) H3K4Me2, acetyl H3K9/K14 (Millipore), HDAC1 (Immunechem and Cell Signaling), HDAC2 (Santa Cruz, Cell Signaling), HDAC3 (Cell Signaling, Abgent), V5 (Invitrogen), and DNMT1 (Abcam and EU101 generously provided by Dr. P. Vertino). Total H3 (Upstate), β-actin and β-tubulin (Sigma) glyceraldehyde-3-phosphate dehydrogenase (GAPDH; Cell Signaling) were used as loading controls depending on the application and molecular weights of target proteins in the experiment.

Primers for bisulfite sequencing, and methylation-specific PCR (MSP) were designed using MSPPRIMER.org (26). Sequences are available upon request. Genomic DNA was bisulfite converted using EZ DNA Methylation Kit (Zymo Research) and subjected to nested MSP. Briefly, unbiased primers outside of the CpG island were used to pre-amplify bisulfite converted DNA. The resultant amplicon was further amplified with primers specific for the methylated or unmethylated form of the expected amplicon. Bisulfite sequencing was performed using a nested primer set designed for MSP using bisulfite converted genomic DNA as above. PCR reactions were cloned into pCR2.1topo (Life Technologies) and sequenced by Genewiz Inc.

RNA was extracted from cell lines by PureLink RNA Mini Kit (Life Technologies) with DNaseI treatment on column. cDNA was reverse transcribed using Superscript III First Strand Synthesis Kit (Life Technologies). Quantitative real-time PCR (qRT-PCR) was performed on a StepOne Plus (ABI) thermocycler unit using either TaqMan reagents (for SFRP2) or SYBR green mastermix (Bio-Rad or Life Technology, respectively; for HDAC1 to HDAC3 mRNA, HDAC1 to HDAC3 preMRNA, 18S RNA, DNMT1, G9A, β-actin, and GAPDH). We observed no significant difference in housekeeping gene levels across treatments and cell types (not shown).

High-throughput methylation profiling was done using the Illumina GoldenGate Methylation Cancer Panel I microarray platform. DNA quality control using picogreen, bisulfite conversion using the EZ DNA Methylation Kit (Zymo) and array hybridization according to manufacturer's specifications were performed by Emory Integrated Genomics Core facility.

Statistical analysis of Illumina Golden Gate methylation data was carried out using R. Sample quality checks on the data were carried out using the bioconductor package methylumi (27). Following quality control checks, the replicates were summarized into a composite value using singular value decomposition 2 algorithm. Significant differentially methylated CpG sites were identified using the nonparametric hypothesis testing method called hypothesis-based analysis of microarrays (HAM; refs. 28 and 29). HAM method consists of 2 steps. In the first step, the significance of a given hypothesis is assessed over the entire dataset. If significant, the next step identified a list of differentially methylated CpG sites. Pathway analysis was carried out on significant differentially methylated genes in comparisons of interest using Metacore from Thomson Reuters.

siRNA knockdowns of HDAC1, HDAC2, and HDAC3 were performed with the following reagents: (HDAC1, s73—Life Technologies; sc-44208—Santa Cruz Biotechnology; HDAC2, sc-29345—Santa Cruz Biotechnology; HDAC3, s16878—Life Technologies; Control A—Santa Cruz Biotechnology) 40 to 80 pmol of siRNA duplex were added per experiment. Lipofectamine 2000 was used as transfection reagent.

Archived paraffin-embedded samples of 20 patients with early stage lung cancer who underwent surgical resection of NSCLC at Emory University affiliated hospitals between May and August 2000 were obtained from the Lung and Thoracic Malignancies Satellite Cancer Tissue Bank at the Winship Cancer Institute of Emory University. Matched tumor and surrounding histologically normal tissue blocks were available on 18 patients.

Sectioning of the tissue samples and immunohistochemistry (IHC) was performed by the Cancer Tissue and Pathology Shared Resource of the Winship Cancer Institute using the following antibodies and dilutions: DNMT1 (Abcam; 1:1,000), HDAC1 (1:1000), HDAC2 (1:1000; Cell Signaling), and HDAC3 (1:1000; Abgent) proliferating cell nuclear antigen (PCNA; 1:1,000; Cell Signaling). Staining occurred on a fully automated stainer after standard antigen retrieval steps as previously described. A horseradish-peroxidase labeled secondary anti-rabbit antibody was used in 1:1,000 dilution. Staining was evaluated by 2 independent observers and quantified using a weighted index [intensity (scale 0–2) x% stained; ref. 30].

Differences between continuous variables were analyzed by Student t test.

To model early events in smoke carcinogen–driven transformation of bronchial epithelia, we exposed immortalized human bronchial epithelial cells (here after 3KT) with low levels of the smoke-related carcinogens MNU and B(a)P for 1 day per week followed by 6 days of outgrowth. After 16 weeks of exposure, immortalized bronchial epithelial cells acquired the ability to grow as anchorage-independent colonies, a hallmark of ongogenic transformation (Fig. 1A). This transformation was accompanied by an increase in protein levels for several epigenetic repressors: the DNA methyltransferase DNMT1, the histone methyltransferases G9A [responsible for histone H3 dimethylation on lysine 9 (H3K9me2)], as well as the class I HDAC1 to HDAC3. Associated with these changes were profound reductions in global levels of histone marks associated with active gene transcription such as H3-acetylation (H3-Ac; Fig. 1B). Together, these changes indicate that tobacco-carcinogen exposure is associated with a more repressive chromatin pattern.

To determine the mechanism by which expression levels of the epigenetic repressors are controlled, we first used qRT-PCR to investigate steady-state mRNA levels. We observed that the increase in DNMT1 protein expression levels is not accompanied by an increase in steady-state mRNA levels. In contrast, we observed a profound increase in steady-state mRNA levels for HDAC1 to HDAC3 and G9A, suggesting that either transcription or changes in mRNA stability are responsible for the observed increase in the expression of these genes (Fig. 1C). To differentiate between transcriptional regulation versus posttranscriptional mRNA stabilization, we also analyzed the unspliced primary transcripts of HDAC1 to HDAC3 (Supplementary Fig. S1). After smoke carcinogen exposure, an increase in the primary transcripts for HDAC1 and HDAC2 were observed, indicating increased levels of these 2 HDACs are transcriptionally mediated. For HDAC3 however, no increase in primary transcript RNA was observed in T31 cells compared with 3KT cells, indicating that its increase in steady-state mRNA levels is likely regulated posttranscriptionally.

The observed increase in protein levels of these epigenetic repressors raised the question if these changes are sufficient to lead to permanent epigenetic reprogramming in carcinogen exposed cells. We therefore analyzed a set of candidate genes for DNA hypermethylation by MSP. We chose a set of candidate genes known to be epigenetically silenced in cancer, and examined the methylation status of their promoters in both long-term carcinogen-exposed cells and their vehicle exposed counterparts. Aberrant DNA hypermethylation was found in 6 of 15 (40%) previously unmethylated promoters. In noncarcinogen-exposed 3KT cells, only the GATA4 promoter was methylated (Table 1). As expected, aberrant DNA methylation of the SFRP2 promoter was associated with transcriptional repression (Fig. 1D).

We next performed an unbiased methylation analysis of 1,505 CpG dinucleotides, representing 807 target genes using the Illumina GoldenGate microarray platform. After adjusting for false-positive discoveries (FDR < 0.05), we found statistically significant increased methylation in 42 genes after carcinogen exposure for 31 weeks versus long-term cultured but not carcinogen-exposed cells (T31 cells vs. 3KT). In contrast, no significant hypomethylation was observed at any gene (Supplementary Fig. S2 and Table S1). Because the GoldenGate array is enriched for probes for genes whose promoters have been shown to be hypermethylated in cancer, the lack of detectable hypomethylation on this array may be because of this bias. Pathway analysis revealed that hypermethylated candidate genes are mostly involved in cell proliferation and cell-cycle control (Supplementary Table S2). Together these data suggest that tobacco-carcinogen exposure mediates aberrant de novo DNA methylation through upregulation of DNMT1.

DNMT1 expression is usually increased after DNA replication and reaches peak levels in S and G₂ phase, where DNMT1 physiologically mediates maintenance methylation of the newly synthesized daughter strand (10). Because deregulated cell-cycle expression of DNMT1 has been linked to cellular transformation, we examined DNMT1 expression in 3KT and T31 cells at various stages of the cell cycle after release from aphidocholin block. Interestingly, a cell-cycle independent increase in DNMT1 expression was observed in carcinogen transformed T31 cells but not in regular 3KT cells. In T31 cells, high levels of DNMT1 expression were observed in all stages of the cell cycle starting with G₀. This increase in cell-cycle independent DNMT1 expression was mirrored by a similar increase in cell-cycle independent HDAC3 protein expression (Fig. 2A). To determine if a direct correlation exists between HDAC3 and DNMT1 expression, we engineered a 3KT cell line, which either stably expressed empty vector or HDAC3. HDAC3 expression was sufficient to induce the previously observed cell-cycle independent increase of DNMT1 expression (Fig. 2B), suggesting that HDAC3 may stabilize the DNMT1 protein. To further test this hypothesis, we performed coimmunoprecipitation experiments after overexpressing HDAC1 to HDAC3 and DNMT1 constructs. We found that DNMT1 associates with HDAC1 and HDAC3 (Fig. 2C) and that overexpression of these HDACs stabilized both the V5-tagged DNMT1 protein as well as native DNMT1 (Figs. 2B and Supplementary Figs. S3 and S4). Moreover, siRNA knockdown of HDAC2 and HDAC3 led to a profound reduction in DNMT1 protein levels. Although knockdown of HDAC1 alone had little effect on DNMT1 levels, it potentiated the effects of HDAC2 knockdown on DNMT1 levels (Fig. 2D). These findings indicate not only that HDACs mediate DNMT1 protein stability, but also that there is a certain redundancy in the activity of these 3 class I HDACs to stabilize DNMT1.

To determine if the carcinogen-induced upregulation of DNMT1 is because of increased de novo synthesis or decreased protein turnover, we exposed 3KT and T31 cells to the protein synthesis inhibitor cycloheximide (2 μg/mL). Cycloheximide exposure was associated with an increase in steady-state DNMT1 protein levels, indicating that de novo peptide synthesis of destruction signals like ubiquitin may be required for DNMT1 protein degradation (Fig. S5). Although inhibition of class I HDACs with VPA led to almost complete suppression of DNMT1 levels, DNMT1 degradation after VPA exposure was fully inhibited after coexposure with the proteasome inhibitor MG-132, suggesting that the primary mechanism for DNMT1 stabilization through interaction with HDACs is by protection from proteasomal degradation (Fig. 3A).

We next sought to determine whether DNMT1 is a direct target for deacetylation by class I HDACs. A degradation-resistant DNMT1 construct lacking the N-terminal 110 amino acids (24) was C-terminally His6 and V5 tagged, and stably expressed in T31 cells. HDAC1 and HDAC2 were subsequently knocked down by siRNA either individually or in combination. After purification of DNMT1 via a NTA-column, DNMT1 acetylation status was determined by immunoblot using a pan acetyl-lysine antibody. Individual knockdowns of either HDAC1 or HDAC2 alone led to only a marginal increase in DNMT1 acetylation, which was much less pronounced than that induced by the class I HDAC inhibitor VPA. Combined HDAC1 and HDAC2 knockdown, however, resulted in greater levels of DNMT1 hyperacetylation comparable to that of VPA treatment; again showing that different class I HDACs have redundant roles in regulating DNMT1 acetylation and expression. Furthermore, knockdown of HDAC1 and/or HDAC2 in conjunction with VPA exposure increased the ubiquitination of DNMT1 (Fig. 3B), leading to the conclusion that the predominant mechanism of DNMT1 stabilization in carcinogen exposed cells is deacetylation and prevention of ubiquitination and proteasomal degradation.

To rule out additional indirect effects of VPA treatment on other mechanisms, we investigated if DNMT1 degradation could be mediated indirectly (i) by either inhibition of HSP90, whose deacetylation by HDAC1 has been previously shown to alter DNMT1 stability (12) or (ii) be dependent also on DNMT1 methylation, which has been previously reported to be a degradation signal similar to DNMT1 acetylation (16). First, we exposed 3KT and T31 cells to either VPA or the HSP90 inhibitor 17-N-allylamino-17-demethoxygeldanamycin (17-AAG) or a combination of both. In 3KT cells, both VPA and 17-AAG induced almost complete reduction of DNMT1 expression. In T31 cells, however, only VPA induced loss of DNMT1 expression. These findings demonstrate that VPA induced induction of DNMT1 degradation is HSP90 independent carcinogen-transformed cells (Fig. 3C). Similarly, VPA-induced DNMT1 degradation was also observed in T31 cells transfected with shRNA against the methyltransferase SET-7, which usually methylates DNMT1 at lysine-147 (16), proving that DNMT1 methylation is not required as intermediary step for VPA-induced degradation (Fig. 3D).

The above data suggest that smoke carcinogen–induced transformation in vitro is accompanied by alterations in the regulation of epigenetic regulators and that this in turn can impinge upon altered DNA methylation and gene expression profiles. To determine the relevance of these alterations to lung cancer in vivo, we analyzed DNMT1 and HDAC1 to HDAC3 protein expression by IHC in surgical specimens of 20 patients with resected NSCLC. All but 2 patients had tumor and surrounding normal lung tissue specimens available for analysis. Although all 4 genes have a ubiquitous baseline expression in normal lung, we observed a highly significant increase in the staining intensity for all 4 genes [DNMT1: median WI: 200 (tumor) vs. 110 (normal), P = 0.0007; HDAC1: median WI 30 (tumor) vs. 7.5 (normal), P = 0.0002; HDAC2: median WI: 60 vs. 10 (P = 0.033); HDAC3: median WI 180 vs. 160 (P = 0.029); Fig. 4]. These differences are not fully attributable to increases in proliferation as measured by PCNA staining median WI: 158 (tumor) vs. 148 (normal; P = 0.677; Supplementary Fig. S7). These findings strengthen and validate our in vitro observations by corroborating that the progression from carcinogen-exposed normal lung to invasive lung cancer is indeed accompanied by an increase in DNMT1 and HDAC1 to HDAC3 expression.

The above results point to a role for class I HDACs in stabilizing DNMT1. We therefore investigated the ability of VPA (a class I HDAC inhibitor) treatment to reverse carcinogen-induced alterations in our cell line model. VPA treatment led to a significant reduction in DNMT1, G9A, and HDAC1 to HDAC3 protein levels in T31 carcinogen-transformed 3KT cells. Global levels of H3-Ac and H3K4me2, which were markedly reduced by prior carcinogen exposure, were restored, whereas global levels of the repressive histone modifications H3K9me3 and H3K27me3 were unaffected (Fig. 5A). These histone changes are indicative of a potential global reprogramming effect of VPA in the carcinogen transformed bronchial epithelial cell background.

To determine the functional relevance of VPA exposure, we analyzed its effects on anchorage independent growth in soft-agar colony formation assays. Indeed, VPA treatment of carcinogen exposed cells at 0.1 and 0.5 mmol/L concentrations for 3 weeks led to a statistically significant 30% reduction in anchorage independent colony formation, indicating a powerful although not complete effect on the reversal of oncogenic transformation (Fig. 5B). These concentrations of VPA did not significantly affect cellular proliferation (Supplementary Fig. S9), suggesting that its effect is more likely through differentiation rather than cytostasis alone.

Because VPA stimulates the reversal of carcinogen-induced transcription and protein stabilization of epigenetic repressors, we next investigated if long-term VPA exposure used in the previous experiment can induce reprogramming of target genes subject to carcinogen-induced epigenetic silencing. In analogy to the soft-agar experiment, we exposed T31 cells to low doses of VPA (0.5 mmol/L) for 28 days. Bisulfite sequencing revealed the induction of significant hypomethylation of the SFRP2 promoter (P < 0.001; Fig. 5C). This promoter hypomethylation was accompanied by an increase in SFRP2 mRNA levels, indicating that VPA is capable of reversing carcinogen-driven hypermethylation, allowing for the reactivation of previously silenced genes (Fig. 5D) Similar results were observed for the RASSF1 promoter (Supplementary Fig. S10).

Our data provide a compelling link between class I HDAC overexpression in lung cancer and DNMT1 protein stabilization, one of the key mediators of aberrant DNA methylation and widely believed to be an important oncogene in the initiation of cancer. Although DNMT1 protein expression is regulated by multiple different mechanisms, we show here that the dominant mechanism of VPA-induced DNMT1 degradation is through acetylation, ubiquitination, and proteasomal destruction. We have explored the contribution of alternative mechanisms such as DNMT1 methylation, AKT-phosphorylation, which has been reported to induce DNMT1 phosphorylation or through inhibition of HSP90. The differential impact of HSP90 inhibition on DNMT1 stability in 3KT cells and T31 cells is interesting and could be further explored. One possibility is that after carcinogen exposure other HSPs may be upregulated as well, limiting dependence on HSP90 alone. Finally, it has been reported that DNMT1 translation may be regulated by microRNAs. Although we have not studied the contribution of microRNAs specifically, the fact that loss of DNMT1 protein expression after VPA treatment can be inhibited both with cycloheximide and MG-132 point toward the proteasome as primary mechanism of DNMT1 degradation. DNMT1 targeting for cancer prevention has been challenging so far, because the only clinically available DNMT1 inhibitors 5-azacytidine and 5′-aza-deoxycitidine are nucleoside analogs, raising the theoretical risk that incorporation into the DNA could lead to secondary malignancies. Moreover, the need for daily subcutaneous injection of these agents would likely lead to low patient acceptance in the prevention setting. Our observation that VPA not only leads to an increase in histone acetylation, but also reverses other carcinogen-induced epigenetic changes such as G9A and DNMT1 upregulation and DNA hypermethylation ultimately leading to the re-expression of epigenetically silenced genes opens a new potential avenue for lung cancer chemoprevention. We have recently completed a large cohort study of US veterans with either current or past tobacco exposure, where long term use of VPA was associated with a significant reduction in smoking-related squamous cell carcinoma of the head and neck (31), supporting the potential clinical application of VPA for chemoprevention of smoking related malignancies of the upper aerodigestive tract. It should be noted that a benefit of VPA was only observed with long-term use of VPA (>3 years). This long duration is similar to that required in chemoprevention studies of other cancers such as breast cancer prevention with tamoxifen. Moreover, the long duration of exposure required may be an explanation why relatively short courses of HDAC treatment alone are insufficient to prevent lung cancer in carcinogen-induced mouse models (9).

The histone methyltransferases G9A and EZH2 are important transcriptional repressors. In particular, the interaction between G9A, H3K9me2, heterochromatin protein 1 (HP1), and DNMT1 has been hypothesized to direct de novo DNA methylation to loci previously marked by H3K9me3 (32). Demethylation after treatment of cancer cells with nucleoside DNMT inhibitors generally only yields transient demethylation, followed by gradual remethylation after drug withdrawal (33, 34). Because G9A has been implicated as potential mediator of de novo DNA methylation (32), the reduction in G9A protein levels we observed after HDAC inhibition are particularly important, because there might be a lesser tendency for target genes to become remethylated.

In summary, our data support a model (Fig. 5F) in which tobacco-related carcinogen-induced upregulation of HDAC1-3 mRNA and protein expression leads to increased stability of the oncogenic DNMT1 protein, thus enabling carcinogenic transformation. Moreover, our study provides strong rationale for the potential use of HDAC inhibitors as chemopreventive agents against lung cancer.

No potential conflicts of interest were disclosed.

The contents of this publication do not reflect views of the Department of Veterans Affairs or the United States Government.

Conception and design: S.A. Brodie, F.R. Khuri, J.C. Brandes

Development of methodology: S.A. Brodie, F.R. Khuri, J.C. Brandes

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. El-Kommos, M. Behera, G.L. Sica, J.C. Brandes

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.A. Brodie, A. El-Kommos, H. Kang, S.S. Ramalingam, K. Gandhi, J. Kowalski, F.R. Khuri, J.C. Brandes

Writing, review, and/or revision of the manuscript: S.A. Brodie, H. Kang, S.S. Ramalingam, K. Gandhi, F.R. Khuri, P.M. Vertino, J.C. Brandes

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Li, S.S. Ramalingam

Study supervision: P.M. Vertino, J.C. Brandes

Other: Provided materials, design, and interpretation of data: P.M. Vertino

The authors thank two talented (former) high-school students, S. Moon (now at CalTech) and A. Abid (now at Georgia Tech) for attempting some challenging aspects of this project. The authors also thank D. Martinson for microscopy help, D. Powell for technical advice, B. Gaudette for technical assistance on the flow cytometer, and members of the PV lab for critical comments.

This work was supported in part by the Department of Veterans Affairs, Veterans Health Administration, Office of Research and Development (Biomedical Laboratory Research and Development)-7IK2BX001283-02 to J.C. Brandes NCI-5 P50CA128613-02 Career Development Project to J.C. Brandes; SunTrust Scholar Award to J.C. Brandes; Cohen Family Scholar Award to J.C. Brandes; NCI-P30CA138292 pilot grant to J.C. Brandes. This research project was supported in part by the Emory University Integrated Cellular Imaging Microscopy Core of the Winship Cancer Institute Comprehensive Cancer Center grant, P30CA138292.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.