Carcinogenesis vol.33 no.7 pp.1286–1293, 2012
doi:10.1093/carcin/bgs157
Advance Access Publication May 2, 2012
Oncogenic human papillomavirus imposes an instructive pattern of DNA methylation
changes which parallel the natural history of cervical HPV infection in young women
Sarah M.Leonard, Wenbin Wei1, Stuart I.Collins2,
Merlin Pereira1, Afaf Diyaf1, Christothea ConstandinouWilliams1, Lawrence S.Young1, Sally Roberts1 and
Ciarán B.Woodman1,*
1
School of Cancer Sciences, College of Medical and Dental Sciences and
Cancer Research UK Clinical Trials Unit, School of Cancer Sciences,
University of Birmingham, Edgbaston, Birmingham, B15 2TT, UK
2
*
To whom correspondence should be addressed.Tel: +01 21 4147610;
Fax: +01 21 4158781;
Email: [email protected]
The contribution of early virus-induced epigenetic changes to
human papillomavirus (HPV)–associated carcinogenesis is poorly
understood. Using genome-wide methylation array profiling and
a cell-based model, which supports replication of HPV episomes,
we found that transfection of primary human foreskin keratinocytes with episomal forms of high-risk HPV types was followed
by upregulation of the DNA methyltransferases, DNMT1 and
DNMT3B, and changes in the methylation status of cellular genes
many of which are reported to be differentially methylated in cervical neoplasia. HPV16- and HPV18-associated changes were not
randomly distributed across the genome, but clustered at specific
chromosomal locations which mapped on to known HPV integration sites and to chromosomal regions lost and gained in highgrade cervical neoplasia. Methylation changes were directed in
part by the same cis-acting factors that appear to direct methylation changes in cancer, the presence of a bivalent chromatin mark
in human embryonic stem cells and promoter CpG content; these
associations explain much of the ontological profile of genes found
to have increased methylation following HPV16 transfection. We
were also able to show, using sequential samples from a cohort of
young women with incident HPV16 infections, that the detection
in cervical samples of methylated forms of the tumour suppressor
gene, RARB, often parallels the natural history of cervical HPV
infection. Our findings suggest that further investigation of the
distribution and determinants of early virus-induced epigenetic
reprogramming will provide important insights into the pathogenesis of virus-associated malignancy.
any, remains uncertain (12,13). Henken et al. (2007), for example,
found that keratinocytes containing integrated forms of high-risk
HPV types only begin to accumulate methylation changes following
immortalization (14). Apostilodou et al. (2010) commenting on the
failure to detect methylation differences in cervical smears taken from
HPV positive and HPV negative disease-free women speculated that
HPV infection per se is not the driving force for the majority of DNA
methylation alterations observed in early cervical neoplasia (15,16).
Therefore, in this study, we define using genome-wide promoter
arrays, the topographical distribution of the methylation changes
that follow the transfection of primary human foreskin keratinocytes
(PHFK) with HPV16 and HPV18 using a virus replication model
which unlike that shown previously supports replication of viral episomes (17). We investigate the determinants of this distribution, and
the potential implications of these methylation changes for HPV-associated carcinogenesis. Finally, we determine using sequential observations generated by a well-characterized cohort of young women
who had recently embarked on sexual activity how the detection of
methylated forms of a tumour suppressor gene implicated in cervical carcinogenesis varies with the natural history of cervical HPV
infection.
Materials and methods
Introduction
Transfection of PHFK with HPV16 and HPV18 genomes
Early passage neonatal PHFK isolated from a single donor were co-transfected
with recircularized genomes of either HPV16 (clone 114/B) (kindly provided
by E-M de Villiers, DKFZ, Heidelberg, Germany) or HPV18 and a plasmid
containing a neomycin resistance gene, as described previously (18,19). After
a brief selection of cells in G418- and serum-containing media, emergent colonies were pooled and cultured without drug selection for between 6 and 9
population doublings on gamma-irradiated J2 3T3 fibroblasts feeders prior
to extraction of genomic DNA. The presence of episomal forms of the HPV
genomes was confirmed by the assessment of E2 integrity and by Southern
blot, as described previously (20). Organotypic raft cultures were prepared
as shown previously (19) and harvested after 13 days of growth. After fixation in formaldehyde, the rafts were embedded in paraffin and 4 μm sections
stained with haematoxylin and eosin to assess morphology (Propath UK Ltd).
For immunohistochemistry, slides were de-paraffinized, and low-temperature
antigen retrieval was performed as described previously (21). DNMT1 was
detected using the rabbit polyclonal antibody (ab19905, Abcam), the specificity of which was confirmed using a blocking peptide (ab21999, Abcam).
DNMT3B was detected using the rabbit polyclonal antibody (HPA001595,
Sigma); no blocking peptide is available for this antibody but its specificity
has already been confirmed using protein arrays (22).
DNA methylation in mammalian cells occurs at the 5′-position of
cytosine within CpG dinucleotides. Although under-represented
across the genome, CpG dinucleotides are enriched in CpG islands
that are genomic regions of about 1 kilobase found within or near
promoters or first exons in approximately 70% of human genes (1).
Although the ordered imposition and removal of DNA methylation
is critical for embryonic development, X-chromosome inactivation,
parental imprinting and cellular differentiation, aberrant DNA methylation also contributes to cancer initiation and progression.
DNA methylation is catalysed by three DNA methyltransferases:
DNMT1, DNMT3A and DNMT3B (2). Oncogenic viruses including
hepatitis B, hepatitis C, Kaposi’s sarcoma-associated herpesvirus and
Epstein–Barr virus (EBV) are known to interact with, and to modulate
the expression of the DNMT, resulting in the transactivation and transrepression of both cellular and viral genes (3–8). The HPV16 oncoproteins, E6 and E7, have been shown to increase the expression and
activity of DNMT1; the former through suppression of p53, and the
latter through disruption of the pRB/E2F pathway that regulates the
DNMT1 promoter (9–11). However, the contribution of early virusinduced methylation changes to HPV-associated carcinogenesis, if
Methylation microarray experiments
Genomic DNA was isolated from untransfected PHFK and from PHFK transfected with HPV16 or HPV18 using cell lysis, standard phenol/chloroform
extraction and ethanol precipitation. DNA quantity and quality was analysed
using a Nanodrop ND-1000 spectrophotometer and an Agilent Bioanlayzer,
respectively; both untransfected and transfected passages were found to be of
high-quality. Methylated DNA was immunoprecipitated (MeDIP) using the
MethylCollector™ Ultra kit (Active Motif). Whole-genomic DNA amplification was performed on methyl enriched samples before fragmentation and
hybridization to promoter tiling arrays (Affymetrix).
Cel files were generated using scanned images of microarray chips and the
Affymetrix GeneChip Operating Software, and analysed using the TileMap software (23). Probe-level data were quantile-normalized and probe-level test statistics for two-sample comparisons calculated using a hierarchical empirical Bayes
model. The hidden state of each probe was inferred using unbalanced mixture
subtraction and hidden Markov model algorithms with the posterior probability cut-off set to 0.5. The Tilemap peak detection algorithm was used to detect
hypomethylated and hypermethylated regions. These regions were required to
contain at least 11 probes. Regions shorter than 100 bp or those containing less
than five continuous probes passing the cut-off were discarded. Adjacent regions
were merged when the gap between them was less than 201 bp and the number
of intervening probes failing to pass the cut-off less than 6. No further cut-offs
© The Author 2012. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]
1286
HPV-induced methylation changes
These arrangements were approved by the research ethical committee for the
then Central Birmingham Health Authority (Reference number 1290).
Cervical samples. From our natural history cohort, we had identified a subset
(n 5 60) of women in whom Q-PCR had previously been used to measure virus
copy number in serial samples following an incident HPV16 infection. From
this subset, 23 women were selected for an investigation of how the detection of
methylated forms of TSG changed following the first detection of HPV16 DNA
or with ‘clearance’ of the cervical infection (24). Of the 23 women potentially
contributing to the analysis of incident infection, two were excluded because
their first HPV positive sample either tested negative for the house-keeping gene
GAPDH or insufficient DNA from this sample was available for testing. Of
the 19 women potentially contributing to the analysis of clearance of HPV16
infection, three were excluded because there was insufficient DNA for testing in
either their last HPV positive or their first HPV negative sample.
Fig. 1. Changes in the topography and intensity of DNMT1 and DNMT3B
expression in organotypic rafts before and after transfection with HPV18.
The basal layer is highlighted with an arrow. Rafts prepared from two
different donors showed a similar pattern of expression of which this is a
representative example.
were applied. When calculating the false discovery rate, unbalanced mixture
subtraction was used. All probe sequences were mapped to the genome assembly (Hg18). Genomic regions were visualized using the Affymetrix Integrated
Human Genome browser. The methylation array was validated by performing MeDIP-coupled quantitative-PCR (Q-PCR) on 17 genes. When validating
methylation changes predicted on the array, primers were designed within the
region of predicted change, using primer3 design software. Each Q-PCR reaction was performed using 12.5 µl SYBR green Q-PCR master mix (Qiagen),
20 ng of methyl enriched DNA and 7.5 pmol of primers (SupplementaryTable S1
is available at Carcinogenesis Online). Q-PCR assays were performed in triplicate using an ABI Prism 7700 sequence detection system (Applied Biosystems).
The computer programme R was used for simulating the distribution of concordantly methylated genes, and the probability of the observed number of adjacent
genes showing similar changes in methylation status was calculated using the
binomial distribution. When exploring the determinants of methylation change,
estimates of association for categorical data were obtained using odds ratios,
with chi-squared statistics and P-values calculated as appropriate.
Detection of methylated forms of RARB in cervical samples using nested
quantitative methylation-specific PCR (Q-MSP)
Study design. The study design and characteristics of the study population have
been described elsewhere (20,24). In brief, 2011 women aged 15–19 years
were recruited from a single Brook Advisory Centre (a family planning clinic)
in Birmingham, UK, between 1988 and 1992, and asked to re-attend at intervals of 6 months; follow-up ended on 31 August 1997. At recruitment, a standardized interview questionnaire was used to construct a detailed social, sexual
and behavioural risk-factor profile, including smoking. Sexual and smoking
histories were updated at each follow-up visit. In addition, at each visit, two
cervical cytological samples were taken using the same Ayres spatula; the
first was used to prepare a smear for immediate cytological evaluation and
the second was placed into 10 ml of phosphate buffered saline and stored at
−80°C for subsequent virological examination. All women in whom a cervical
cytological abnormality was identified were immediately referred to a dedicated research clinic for histological examination, irrespective of the severity
of that abnormality. Colposcopic and cytological surveillance were maintained
in these women, and treatment was postponed, until there was histological
evidence of high-grade cervical intraepithelial neoplasia (CIN2 or CIN3), at
which point women left the study.
Ethical approval. Informed oral consent was obtained from all women by one of
only two gynaecological clinical research fellows who were involved in subject
recruitment and in the management of patients with cytological abnormality.
Nested Q-MSP. A total of 149 cervical samples were tested for methylated
forms of RARB. Genomic DNA (200 ng) was bisulphite converted using the
EZ DNA methylation kit (Zymo Research). The nested Q-MSP method along
with the primers and probes used in this analysis were as described in Fackler
et al. (2004) (25). In the first PCR reaction, 10 µl of bisulphite DNA was added
to 25 µl of PCR master mix (Promega) and 25 pmol of forward and reverse
nested PCR primers. The PCR products were measured following purification
(Qiagen) and 20 ng of product was added to Q-MSP reaction buffer containing
12.5 µl Taqman master mix (Applied Biosystems), 10 pmol of methylated or
unmethylated primers and 2 pmol of methylated or unmethylated probe. Assays
were performed in triplicate using an ABI Prism 7700 sequence detection system (Applied Biosystems). For each reaction plate, the following were used to
generate standard curves and to provide controls: (i) serially diluted methylated
and unmethylated copy number standards (ii) no template control and (iii) SiHa
(a known methylated control). The percentage of methylation was calculated as:
percentage methylated 5 100 × [number of copies of methylated DNA/(number
of copies of methylated DNA + number of copies of unmethylated DNA)]. Wilcoxon tests were used when comparing these continuous data.
Results
Changes in the pattern of expression of the DNA methyltransferases
in organotypic rafts established following HPV transfection of PHFK
recapitulates that observed in cervical neoplasia
In PHFK organotypic rafts, DNMT1 expression is restricted to nuclei
of cells of the basal and early differentiating layers but downregulated
in more differentiated cells. However, in rafts established following
transfection of PHFK with HPV18, DNMT1 nuclear staining is seen
not only in the basal layer but also in suprabasal cells extending to
the stratum granulosum, with intense DNMT1 staining in some of
these cells. Changes in the topography of DNMT3B expression were
similar to those observed for DNMT1, although the overall intensity
of staining was weaker (Figure 1). No consistent changes were seen
in the expression of DNMT3A in organotypic rafts following HPV
transfection (data not shown).
Transfection of PHFK with episomal forms of high-risk HPV is
followed by widespread methylation changes in cellular genes
To investigate the consequences of these HPV-induced changes in the
expression of the DNA methyltransferases, genome-wide methylation
profiling was performed on PHFK following their transfection with
episomal forms of HPV16 and HPV18. Table I reveals a significant
increase in the methylation of 5607 and 2837 genes, following transfection of PHFK with HPV16 and HPV18, respectively; a decrease
in the methylation of 3568 and 4160 genes and both an increase
and a decrease in methylation in non-overlapping regions in a further 2295 and 1023 genes (Table I). Of those genes found to have
either increased or decreased methylation following transfection with
HPV18, 87% and 63%, respectively, were concordantly changed following HPV16 transfection. Methyl enriched Q-PCR confirmed the
changes in methylation status predicted on the promoter array in all
17 genes selected for further validation (Figure 2). These 17 genes
were selected for further validation because they had either been
reported to be methylated in cervical neoplasia, or they were components of the WNT signalling pathway which is known to be disrupted in cervical neoplasia, or were themselves epigenetic r­ egulators
(Supplementary Table S2 is available at Carcinogenesis Online) (26).
1287
S.M.Leonard et al.
Table I. Distribution of methylation changes in cellular genes following transfection of PHFK with episomal HPV16 and episomal HPV18
Methylation status following transfection of PHFK with episomal HPV18
Unchanged
(n 5 11865)
Methylation status following
transfection of PHFK with
episomal HPV16
Increased
(n 5 2837)
Decreased
(n 5 4160)
Increased and decreased
in non-over-lapping
regions(n 5 1023)
Unchanged (n 5 8415)
7815
97
497
6
2476
150
131
Increased (n 5 5607)
2850
Decreased (n 5 3568)
905
11
2613
39
Increased and decreased in
non-over-lapping regions (n 5 2295)
295
253
900
847
Fig. 2. Methylation changes on the array validated using MeDIP-coupled Q-PCR.
Genes with increased methylation following transfection with either
episomal HPV16 or HPV18, or both, were enriched for those identified in candidate gene studies as aberrantly methylated in cervical
neoplasia (OR 5 2.0, χ2 (1 df) 5 9.7, P 5 0.0019, Supplementary
Table S2, which lists all publications describing methylated forms of
genes methylated in cervical neoplasia, is available at Carcinogenesis
Online). We also determined the frequency with which genes found
to be differentially methylated by HPV18 in our initial methylation
analysis were also differentially methylated in three other donors
(A, B and C), which had previously been transfected with HPV18
(27). For a number of these genes, concordant changes in methylation
status were seen in keratinocytes isolated in these additional donors
(SupplementaryFigure S1 is available at Carcinogenesis Online). For
1288
example, hypomethylation of SFRP1 and DLC1 was confirmed in all
four donors and hypermethylation of CADM1 and MBD1 in three.
Methylation ‘hotspots’ map on to HPV integration sites and
chromosomal regions lost in high-grade cervical neoplasia
Although earlier candidate gene studies suggested that hypermethylation
events are focal in nature, recent genome-wide surveys have revealed
that co-ordinated methylation changes over large regions are a common
phenomenon at different sites of cancer (28,29). When we looked for evidence of clustering of methylation changes in HPV-transfected PHFK,
we found that 64% and 50% of those genes with increased methylation
following HPV16 and HPV18 transfection, respectively, were in clusters consisting of two or more adjacent genes that were concordantly
HPV-induced methylation changes
Fig. 3. HPV16 and HPV18 hotspots map onto HPV integration sites and regions of reported chromosomal loss in high-grade CIN. Panels A and B map the
distribution of hypermethylation and hypomethylation hotspots following transfection of PHFK. Panel C maps those HPV16 hypermethylation hotspots that
are also reported to be HPV integration sites, and panel D, those HPV18 hypermethylation hotspots reported to be regions of chromosomal loss in women with
HG-CIN.
changed. Of genes found to have decreased methylation following transfection with HPV16 or HPV18, 49% and 53%, respectively, were similarly clustered. For both hypermethylated and hypomethylated genes,
the probability of observing this proportion of adjacent genes with a
concordant change in methylation status, or one greater is less than 1 in
100 000, suggesting that HPV16- and HPV18-associated methylation
changes are non-randomly distributed across the genome (Supplementary Figure S2 is available at Carcinogenesis Online).
We next explored whether methylation changes clustered at certain
cytogenetic locations. A cytoband was considered a methylation hotspot when it included significantly (P  0.001) more genes with either
increased or decreased methylation following HPV transfection than
would have been expected from the overall frequency of these changes
observed on the array. Of 806 cytobands listed on the array, 36 and 35
were found to be HPV16 and HPV18 hypermethylation hotspots, respectively, with 23 of these hotspots common to both HPV types; 27 HPV16
and 16 HPV18 hypomethylation hotspots were also identified of which
9 were common to both types (Figure 3A and B, Supplementary Table
S3A is available at Carcinogenesis Online). Furthermore, HPV-associated hypermethylation hotspots appear to cluster in telomeric regions.
Consistent with recent evidence that epigenetic and genetic changes
are globally co-ordinated, and with the possibility that early HPVinduced epigenetic changes may predispose to later genetic events
relevant to virus-associated carcinogenesis, we found that HPV16-associated hypermethylation hotspots were enriched for sites where HPV is
known to be integrated into the genome (OR 5 2.4, χ2(1 df) 5 5.012,
P 5 0.0252, Figure 3C, Supplementary Table S3B is available at Carcinogenesis Online) (30–35). We also found that HPV16 and HPV18
hypermethylation hotspots were enriched for regions of chromosomal
loss previously identified in women with high-grade CIN using comparative genomic hybridization (HPV16 hotspots: OR 5 4.9, χ2(1
df) 5 7.1, P 5 0.0078; HPV18 hotspots: OR 5 10.5, χ2(1 df) 5 26.3,
P  0.0001, Figure 3D, Supplementary Table S3C is available at
Carcinogenesis Online) (36–38). Of the 27 HPV16 and 16 HPV18
hypomethylation hotspots identified, 6 were also listed as regions of
amplification in the same analysis of women with high-grade CIN;
HPV16 but not HPV18 ­hypomethylation hotspots were enriched for
these regions of amplification (χ2 (1 df) 5 4.98, P  0.05) (Supplementary Table S3D is available at ­Carcinogenesis Online).
The same cis-acting factors which direct DNA methylation in cancer
determine the distribution of methylation changes in HPV-transfected
PHFK
As these results are consistent with the possibility that HPV-associated
methylation changes may like those described in cancer come about
through an instructive mechanism, we looked for cis-acting factors that
might determine their distribution. In melanoma, prostate cancer and in
non-Hodgkin’s lymphomas, genome-wide analyses show that those genes
that become hypermethylated in cancer are significantly more likely to
have a high CpG content whereas those that become hypomethylated have
a low CpG content (39–41). Using the same classification of overall promoter CpG content first reported by Weber et al. (2007), we found that
compared with genes with a low CpG content, those with an intermediate
or high CpG content were more likely to have increased methylation and
less likely to have decreased methylation following transfection of PHFK
with episomal HPV16 [χ2(2 df) = 592.8, P < 0.0001; χ2(2 df) = 122.6,
1289
S.M.Leonard et al.
Table II. Change in methylation status following transfection of PHFK with
episomal HPV16 is related to promoter CpG content and to the pattern of
histone marking in HES cells
Change in methylation status
Increased
CpG content*
Low (n 5 4126)
501 (12.1%)
990 (24%)
490 (24.5%)
354 (17.7%)
High (n 5 8673)
3233 (36.9%) 1297 (14.8%)
715 (14.6%) 1147 (23.5%)
3278 (32.6%) 1688 (16.8%)
785 (45%)
187 (10.7%)
Risk of increased methylation increases with increasing CpG content and is
greatest for genes bivalently marked by H3K4me3 and H3K27me3 in HES
cells; risk of decreased methylation falls with decreasing CpG content and is
lowest for bivalently marked genes.
*Analysis restricted to genes listed on Weber et al., array (2007) (42).
**Analysis restricted to genes listed on Zhao et al., array (2008) (47).
P < 0.0001, Table II (42)]. Similar trends were seen following HPV18
transfection (SupplementaryTable S4 is available at Carcinogenesis
Online).
In both cancer and ageing, aberrantly hypermethylated promoters have been associated with a key set of developmentally regulated
genes that are held in a ‘transcription-ready’ state in embryonic stem
cells by a bivalent promoter chromatin pattern consisting of the active
histone mark, H3K4me3, and the repressive mark, H3K27me3 (43–
47). Using a published genome-wide map of these histone modifications in human embryonic stem (HES) cells, we found that bivalently
marked genes are more likely to have increased methylation and less
likely to have decreased methylation following transfection of PHFK
with episomal HPV16 (χ2(2 df) 5 254, P  0.0001; χ2(2 df) 5 71.2,
P  0.0001, Table II (47). Consistent with these findings, genes with
increased methylation following HPV16 transfection were also found
to be enriched for a subset of polycomb target genes found to undergo
methylation with age and to be significantly more methylated in cervical smears taken from women with HPV-associated CIN than in
normal cervical cells (OR 5 2.9, χ2(1 df) 5 19.9, P  0.0001, Supplementary Table S5 is available at Carcinogenesis Online) (48).
Although genes with a high CpG content are more likely to be bivalently marked in HES cells than those with a low CpG content (χ2(1
df) 5 289.1, P  0.0001), both CpG content and the presence of a
bivalent chromatin mark in HES cells appear to be independent predictors of methylation changes in HPV16-transfected PHFK (Table III).
The detection of methylated forms of RARB parallels the natural
history of cervical HPV infection in young women
Providing evidence of a causal association between HPV infection and
the acquisition of methylation changes in vivo is more problematic.
Evidence adduced from cross-sectional surveys in patients with cervical intraepithelial neoplasia is unsatisfactory in so far as this study
design cannot distinguish those methylation changes that are a consequence of the disease process from those that are directly attributable
to the exposure. To establish a chain of causality linking HPV exposure to a given epigenetic change and to a particular disease outcome
requires the generation of longitudinal observations in women who
are both disease-free and HPV negative at the beginning of the period
of observation. We were able to explore the temporal relationship
between the first detection of cervical HPV infection and the subsequent appearance of methylation changes in a cohort of young women
(methods section QV). We chose as our candidate genes for our in
vivo methylation studies, the TSG, RARB and CADM1, methylated
forms of which have been consistently detected in cervical samples
1290
CpG content
Decreased
Intermediate (n 5 1999)
Marked by
Neither (n 5 4883)
H3K4 or by
H3K4 (n 5 10 061)
H3K4me3K27me3
H3K4me3K27me3
in HES**
(n 5 1745)
Table III. CpG content and a bivalent pattern of histone marking in HES
cells independently predict methylation change in PHFK transfected with
HPV16
Histone marking
in HES cells
Change in methylation status
Increased
Decreased
Low
269 (9.0%) 747 (24.9%)
Neither (n 5 2999)
128 (22.6%) 128 (22.6%)
H3K4me3 (n 5 566)
44 (34.6%)
22 (17.3%)
H3K4me3K27me3 (n 5 127)
Intermediate Neither (n 5 618)
146 (23.6%) 119 (19.3%)
209 (23.1%) 175 (19.3%)
H3K4me3 (n 5 905)
63 (32.8%)
19 (19.9%)
H3K4me3 K27me3 (n 5 192)
High
145 (38.4%)
52 (13.8%)
Neither (n 5 378)
2318 (34.6%) 1063 (15.8%)
H3K4me3 (n 5 6707)
108 (9.5%)
H3Kme34K27me3 (n 5 1135) 566 (49.9%)
For each level of CpG content, genes that are bivalently marked by H3K4me3
and H3K27me3 in HES cells have the highest risk of increased methylation
following transfection of PHFK with episomal HPV16. For genes with
low and high CpG content, bivalently marked genes have the lowest risk
of decreased methylation following HPV16 transfection. This analysis is
restricted to genes listed on both the Weber et al. (2007) and Zhao et al.
(2008) arrays (42,47).
taken from women with intraepithelial and invasive cervical disease
(49,50), In 17 of the 21 women who were found to have had an incident HPV16 infection using nested Q-PCR, the number of methylated forms of RARB/10 000 cells detected using nested Q-MSP was
found to be higher in the first HPV positive sample than in the HPV
negative sample immediately preceding it (Wilcoxon-ranked sign test,
P 5 0.0072, Table IV). In 16 women, HPV viral load either returned
to zero or decreased in the second of their last two HPV positive samples taken before the end of follow-up. In 13 of these women, the
decrease in viral load in the second of these two samples was associated with a contemporaneous decrease in the number of methylated
forms of RARB/10 000 cells (Wilcoxon-ranked sign test, P 5 0.0213,
Table IV). In two of the three women in whom there was no decrease
in the number of methylated forms of RARB following clearance of
HPV16 infection, both had a concurrent HPV18 infection which persisted beyond that visit at which HPV16 could no longer be detected
(data not shown). Thus, changes in the number of methylated forms
of RARB in clinical samples mirrored changes in HPV copy number. They are also consistent with those observed in our in vitro models. Following transfection of keratinocytes with episomal HPV16
and HPV18, Q-MSP revealed that the number of methylated forms
of RARB increased from 183/10 000 cells to 2839 and 3851/10 000
cells, respectively; these values lie within the range found in our clinical samples. A smaller number of samples were also tested for methylated forms of the TSG, CADM1, methylated forms of which have
been detected in women with CIN (16). However, although Q-MSP
confirmed an increase in the number of methylated forms of CADM1
from 419/10 000 cells to 6738 and 4552/10 000 cells following transfection of PHFK with episomal HPV16 and HPV18, respectively, we
found no compelling evidence to suggest that the number of these
forms of this gene in clinical samples vary with either the first detection or the subsequent clearance of HPV DNA (data not shown).
Discussion
We have shown that the transfection of PHFK with episomal forms
of two high-risk HPV types is followed by widespread methylation
changes in cellular genes, many of which are reported to be differentially methylated in cervical neoplasia. Consistent with ­observations
that the HPV16 E7 oncoprotein can augment DNMT1 activity we
have shown using organotypic raft cultures that HPV18 modulates
the expression of both DNMT1 and DNMT3B (12). Deregulation
of DNMT1 and DNMT3B expression is also seen in HPV16 and
HPV18-associated CIN3. We found DNMT1 to be consistently upreg-
HPV-induced methylation changes
Table IV. Number of methylated forms of RARB increases when HPV16 is first detected and decreases with clearance of viral infection
Number
Number of copies/1000
cells of HPV16 in:
Positive sample
2
69
73
128
240
244
330
350
490
519
556
561
662
744
910
928
932
938
1115
1430
1551
1803
Number of methylated
forms/10000 cells of RARB in:
Negative
7538
7
690
125
8274
16
397966
72
237
0.5
27
96
3238
30
29
0.4
11842
0
0
0
2
17
0
299
1
0
2
131
41
69
53
1
0
52
389932
84
43
469
0
77
0
277
Number of copies/1000 cells
of HPV 16 in:
Positive
Penultimate
291
0
1175
676
290
36
4237
3
0.3
382
346
85
112
1
1526
0.1
793
110
332
3
1
106
7538
100818158
ulated throughout CIN3 as described by Sawada et al. (2007) (51),
and DNMT3B to be expressed in all layers unlike normal epithelium
where it is restricted to nuclei of basal and parabasal cells (Supplementary Figure S3 is available at Carcinogenesis Online). As the HPV
genome does not code for a DNA methyltransferase, changes in the
viral methylome are likely to be dependent on the cell’s DNA methylation machinery. However, it has yet to be defined as to how HPVinduced changes in the activity of the DNA methyltransferases might
explain the progressive increase in DNA methylation content in tissue
taken from women who are HPV positive in the absence of any evidence of clinical disease, through those with CIN, to those with cervical carcinoma (52,53). Recently, it has also been suggested that in
order to ensure continuing cell viability, HPV ‘self-methylates’, thus
downregulating excessive oncogene activity (54).
Our observations also, for the first time, point to a pattern of
HPV-induced instructive methylation of cellular genes. HPV16- and
HPV18-associated changes were not randomly distributed across the
genome but clustered at certain chromosomal locations directed in
part by the same cis-acting factors that appear to direct methylation
changes in cancer, the presence of a bivalent chromatin mark in HES
cells and promoter CpG content.
These cis-acting factors also appear to explain much of the ontological
profile of the methylation changes observed following the transfection
of high-risk HPV types. For example, the top 10 signalling pathways
over-represented among genes with increased methylation following
transfection of PHFK with HPV16 are also enriched for genes with
a high CpG content and for those bivalently marked in HES cells
(Supplementary Table S6A is available at Carcinogenesis Online). They
include the WNT signalling pathway, components of which have been
shown to be silenced by DNA methylation in cervical neoplasia (26).
Similarly, of the top 10 biological processes over-represented among
genes hypermethylated following HPV16 transfection, all were enriched
for genes bivalently marked in HES, and eight were enriched for genes
with a high CpG content (Supplementary Table S6B is available at
Carcinogenesis Online). On the other hand, of the top 10 biological
processes that were significantly under-represented among genes
which were hypermethylated following HPV16 transfection, all were
found to be significantly depleted for genes with a high CpG content;
these processes included antigen presentation and processing, cellular
defence, immune response and complement activation (Supplementary
Next sample
Number of methylated
forms/10 000 cells of RARB in:
Penultimate
Next sample
3
22329
291
5
151
0
48806
9
948
0
679329
237
22671
0
6978
0.3
1431
404
1068
96
5537236
30
367
0
0
0
77
85
229
1
0
32
30
1382
1737
107
74
389932
84
43
469
0
0
603
8076
50
2
0
603
15
0
332
3
1
106
0
0
2
0
0
324
Table S6C is available at Carcinogenesis Online). Given the evolutionary
pressure to lose CpG content, it is tempting to speculate that the low
CpG content of many immune response genes may have evolved to
offer some protection against virus-induced deregulation of the DNA
methylation machinery. A more restricted set of pathways and biological
processes were found to be enriched significantly among genes which
were hypomethylated following HPV16 transfection (Supplementary
Table S7 is available at Carcinogenesis Online).
We have also shown using genome-wide methylation analysis
that infection of germinal centre B cells with another oncogenic
virus, EBV, imposes a similar pattern of instructive methylation (55).
Although the distribution of EBV-associated hotspots differed from
that observed following transfection of PHFK with high-risk HPV
types, the risk of these virus-associated methylation changes was also
found to vary significantly with CpG content. It has also been suggested that aberrant de novo DNA methylation is, in part, dictated by
underlying sequence context of CpG islands. Of the 1639 CpG islands
classified as methylation prone by McCabe et al. (2009), using an
algorithm that integrates seven DNA sequence patterns with a marker
of polycomb repressive complex 2 occupancy (SUZ12), 918 were
listed on the Affymetrix promoter array (56). Of these CpG islands,
529 and 228 were found to have increased methylation in HPV16- and
HPV18-transfected cells, respectively. However, as this represents
only a small proportion of those genes found to have increased methylation following transfection with these high-risk HPV types, other
sequence-specific markers of methylation susceptibility may yet be
identified.
Although most attention has focused on the epigenetic consequences of the insertion of foreign DNA into mammalian genomes
and their contribution to oncogenic transformation, our results at least
raise the as yet unproven possibility that methylation hotspots may
be more susceptible to later genetic events. It will be of interest to
relate the site of any viral integration event, or of chromosomal loss,
which occur on long-term culture of these cell lines to the distribution of these early methylation changes. Furthermore, HPV-associated
hypermethylation hotspots appear to cluster in telomeric regions (Figure 3). This is also of interest not only because large-scale epigenomic
alterations localized to these regions have been reported in cancer but
also because methylation of transcriptionally repressive sequences
in the hTERT promoter and proximal exonic sequences is associated
1291
S.M.Leonard et al.
with deregulated hTERT transcription in HPV-immortalized cells and
cervical cancer cells (57,58).
Of course, it could be argued that we have described only those
methylation changes in cellular genes that follow the transfection
of keratinocytes from a single donor, albeit with two high-risk HPV
types, and that some of the observed changes are potentially unique
to that donor. Genome-wide profiling of other keratinocytes from
other donors would almost certainly reveal another set of methylation
changes, some of which would be common to those we have already
observed, whereas others would be different. Supplementary Figure 1,
available at Carcinogenesis Online, confirms that some methylation
changes are consistently observed following transfection of keratinocytes from other donors infected with episomal forms of the same
HPV type, whereas others are not. This heterogeneity reflects that
observed in vivo where no single TSG has ever been shown to be
methylated in more than 50% of the cervical cancers sampled. Similarly, it would be unreasonable to expect that all genes reported to be
methylated in cervical neoplasia will necessarily be methylated following the transfection of keratinocytes with high-risk HPV types;
some of the changes observed in vivo will be secondary to the onset
of a disease process that it must be remembered is not an inevitable
consequence of infection with high-risk HPV types. It should also be
remembered that although our in vitro models of viral replication may
be informative with respect to the acquisition of virus-induced epigenetic changes, only limited inferences can be drawn from these simple
unidirectional models about the natural history of epigenetic changes
associated with cervical HPV infection which more often than not is
a self-limiting infection.
However, we have shown for the first time in vivo albeit in a small
series, an association between the onset of infection with an oncogenic virus in young women who have recently embarked on sexual
activity, and an increase in the detection of methylated forms of a
TSG. Such inferences cannot be drawn from cross-sectional studies
where there is always some temporal ambiguity about the relationship
between exposure and disease, and which if performed in diseased
subjects, may be confounded by changes secondary to the disease
process. It remains to be determined whether the detection of methylated forms of RARB or of other TSG can refine the increased risk
of high-grade CIN in women in this cohort who tested positive for
high-risk HPV types. We also plan to explore using samples from this
cohort the impact of infection with high-risk HPV types on the methylation of cellular genes and the relationship of these changes to the
acquisition of disease. We found no evidence to suggest that HPV16
infection in these young women was followed by increased methylation of CADM1 consistent with the results of a recent cross-sectional
study (16). However, this finding does not exclude the possibility that
the detection of methylated forms of CADM1 in cervical smears may
still be a useful marker of disease status.
We also found that in many cases, ‘clearance’ of HPV16 infection
was associated with a sharp decline in the detection of methylated
forms of RARB and that in others, ‘persistence’ of infection was associated with the continuing detection of these methylated forms. This is
of interest because the ectopic expression of RARβ2 has been shown
to downregulate HPV18 transcription by selectively abrogating the
binding of AP-1 to the viral regulatory region in a ligand-independent
manner (59).
These observations may also have implications for attempts to
improve the efficiency of cervical screening programmes using such
DNA methylation markers. If the length of time during which methylated forms of these genes can be detected merely reflects the natural history of viral infection, then their detection might at first glance appear
to offer no advantage over HPV testing. An analogous scenario may be
seen in patients infected with Helicobacter pylori in whom reduced levels of TSG methylation follow successful eradication treatment (60,61).
In both the gastric and cervical mucosa, cell turnover will result in the
disappearance of methylated forms following clearance of infection
when these forms are restricted to progenitor and differentiating cells.
However, the continuing detection of these methylated forms following
clearance of HPV infection may indicate their presence in stem cells or
1292
alternatively, a continuing exposure to some other methylating agent,
for example, smoking which we have also shown to increase the risk of
detection of another TSG, CDKN2A in this population (62).
Establishing a chain of causality linking an environmental or a
behavioural exposure such as HPV infection to a given methylation
change and to a particular disease outcome in vivo is problematic.
To do so, it is first necessary to show that the exposure can induce
methylation changes in disease-free individuals (we have been able
to reveal evidence for such an association in our preliminary analysis
of our study cohort). Thereafter, it will be necessary to show that
exposure-induced changes ‘explain’ at least some of the subsequent
risk of disease. Only by generating longitudinal observations in
disease-free subjects can the risk of incident and progressive disease
associated with exposure-induced methylation changes be estimated.
It is also possible that the onset of incident disease may itself herald a
cascade of secondary methylation events, some of which may actually
be better markers of disease progression. For no site of cancer has
the risk of incident disease in disease-free individuals, or the risk
of disease progression in those with pre-malignant changes, been
related to the presence of epigenetic changes in baseline material.
However, the accessibility of the cervix uteri and the acceptability of
the sampling process does provide the opportunity for the repeated
observations that are required. As cervical cancer has a clearly
defined pre-malignant stage in which outcomes can be accrued within
an acceptable time-frame, it is a useful model for exploring how
epigenetic changes induced by behavioural and environmental risk
factors, e.g. HPV infection and cigarette smoking, contribute to the
initiation and progression of disease.
Supplementary material
Supplementary Tables S1–S7 and Figures S1–S3 can be found at
http://carcin.oxfordjournals.org/
Funding
This work was supported by INCA Project no. LSHC-CT-2005-018704
and Cancer Research UK, programme grant C427/A3913.
Acknowledgements
Conceived and designed the experiments: S.L., C.B.W., S.R. Performed
the experiments: S.L., M.P., A.D., C.C.W. Analyzed the data: W.W., S.I.C.,
C.B.W., S.L. Contributed reagents/materials/analysis tools: S.R., W.W. Wrote
the article and critically advised on the article: C.B.W., S.L., S.R., S.I.C., L.Y.
Conflict of Interest Statement: None declared.
References
1.Robertson,K.D. et al. (2000) DNA methylation in health and disease. Nat.
Rev. Genet., 1, 11–19.
2.Jurkowska,R.Z. et al. (2011) Structure and function of mammalian DNA
methyltransferases. Chembiochem., 24, 206–222.
3.Tsai,C.N. et al. (2002) The Epstein Barr virus oncogene product, latent
membrane protein 1, induces the downregulation of E-cadherin gene
expression via activation of DNA methyltransferases. Proc. Natl. Acad. Sci.
USA, 99, 10084–10089.
4.Seo,S.Y. et al. (2008) Epstein Barr virus latent membrane protein 1 suppresses
the growth-inhibitory effect of retinoic acid by inhibiting retinoic acid receptor-beta2 expression via DNA methylation. Cancer Lett., 270, 66–76.
5.Shamay,M. et al. (2006) Recruitment of the de novo DNA methyltransferase Dnmt3a by Kaposi’s sarcoma-associated herpesvirus LANA. Proc.
Natl. Acad. Sci. USA, 103, 14554–14559.
6.Park,I.Y. et al. (2007) Aberrant epigenetic modifications in hepatocarcinogenesis induced by hepatitis B virus X protein. Gastroenterology, 132,
1476–1494.
7.Arora,P. et al. (2008) Hepatitis C virus core protein downregulates E-cadherin expression via activation of DNA methyltransferase 1 and 3b. Cancer
Lett., 261, 244–252.
HPV-induced methylation changes
8.Zheng,D.L. et al. (2009) Epigenetic modification induced by hepatitis
B virus X protein via interaction with de novo DNA methyltransferase
DNMT3A. J. Hepatol., 50, 377–387.
9.McCabe,M.T. et al. (2005) Regulation of DNA methyltransferase 1 by the
pRb/E2F1 pathway. Cancer Res., 65, 3624–2632.
10. You,J.S. et al. (2008) Histone deacetylase inhibitor apicidin downregulates DNA methyltransferase 1 expression and induces repressive histone
modifications via recruitment of corepressor complex to promoter region in
human cervix cancer cells. Oncogene, 27, 1376–1386.
11. Au Yeung,C.L. et al. (2010) HPV-16 E6 upregulation of DNMT1 through
repression of tumor suppressor p53. Oncol. Rep., 24, 1599–1604.
12. Burgers,W.A. et al. (2007) Viral oncoproteins target the DNA methyltransferases. Oncogene, 26, 1650–1655.
13. Laurson,J. et al. (2010) Epigenetic repression of E-cadherin by human papillomavirus 16 E7 protein. Carcinogenesis, 31, 918–926.
14. Henken,F.E. et al. (2007) Sequential gene promoter methylation during
HPV-induced cervical carcinogenesis. Br. J. Cancer, 97, 1457–1464.
15.Kahn,S.L. et al. (2008) Quantitative methylation-specific PCR for the detection
of aberrant DNA methylation in liquid-based Pap tests. Cancer, 114, 57–64.
16. Apostolidou,S. et al. (2009) DNA methylation analysis in liquid-based
cytology for cervical cancer screening. Int. J. Cancer, 125, 2995–3002.
17. Frattini,M.G. et al. (1996) In vitro synthesis of oncogenic human papillomaviruses requires episomal genomes for differentiation-dependent late
expression. Proc. Natl. Acad. Sci. USA, 93, 3062–3067.
18. Kirnbauer,R. et al. (1993) Efficient self-assembly of human papillomavirus
type 16 L1 and L1-L2 into virus-like particles. J. Virol., 67, 6929–6936.
19. Wilson,R. et al. (2007) The full-length E1E4 protein of human papillomavirus type 18 modulates differentiation-dependent viral DNA amplification
and late gene expression. Virology, 362, 453–460.
20. Collins,S.I. et al. (2009) Disruption of the E2 gene is a common and early
event in the natural history of cervical human papillomavirus infection: a
longitudinal cohort study. Cancer Res., 69, 3828–3832.
21. Watson,R.A. et al. (2002) Changes in expression of the human homologue
of the Drosophila discs large tumour suppressor protein in high-grade premalignant cervical neoplasias. Carcinogenesis, 23, 1791–1796.
22. Uhlen,M. et al. (2010) Towards a knowledge-based Human Protein Atlas.
Nat. Biotechnol., 28, 1248–1250.
23. Ji,H. et al. (2005) TileMap: create chromosomal map of tiling array hybridizations. Bioinformatics, 21, 3629–3636.
24. Constandinou-Williams,C. et al. (2010) Is human papillomavirus viral load
a clinically useful predictive marker? A longitudinal study. Cancer Epidemiol. Biomarkers Prev., 19, 832–837.
25. Fackler,M.J. et al. (2004) Quantitative multiplex methylation-specific PCR
assay for the detection of promoter hypermethylation in multiple genes in
breast cancer. Cancer Res., 64, 4442–4452.
26. Chung,M.T. et al. (2009) Promoter methylation of SFRPs gene family in
cervical cancer. Gynecol. Oncol., 112, 301–306.
27. Knight,G.L. et al. (2004) Cooperation between different forms of the
human papillomavirus type 1 E4 protein to block cell cycle progression
and cellular DNA synthesis. J. Virol., 78, 13920–13933.
28. Keshet,I. et al. (2006) Evidence for an instructive mechanism of de novo
methylation in cancer cells. Nat. Genet., 38, 149–153.
29. Coolen,M.W. et al. (2010) Consolidation of the cancer genome into
domains of repressive chromatin by long-range epigenetic silencing
(LRES) reduces transcriptional plasticity. Nat. Cell Biol., 12, 235–246.
30. Matovina,M. et al. (2009) Identification of human papillomavirus type
16 integration sites in high-grade precancerous cervical lesions. Gynecol.
Oncol., 113, 120–127.
31. Kraus,I. et al. (2008) The majority of viral-cellular fusion transcripts in
cervical carcinomas cotranscribe cellular sequences of known or predicted
genes. Cancer Res., 68, 2514–2522.
32. Peter,M. et al. (2006) MYC activation associated with the integration of
HPV DNA at the MYC locus in genital tumors. Oncogene, 25, 5985–5993.
33. Yu,T. et al. (2005) The role of viral integration in the development of cervical cancer. Cancer Genet. Cytogenet., 158, 27–34.
34. Wentzensen,N. et al. (2004) Systematic review of genomic integration sites
of human papillomavirus genomes in epithelial dysplasia and invasive cancer of the female lower genital tract. Cancer Res., 64, 3878–3884.
35. Poage,G.M. et al. (2010) Genetic and epigenetic somatic alterations in
head and neck squamous cell carcinomas are globally coordinated but not
locally targeted. PLoS One, 5, e9651.
36. Christensen,B.C. et al. (2010) Integrated profiling reveals a global correlation between epigenetic and genetic alterations in mesothelioma. Cancer
Res., 70, 5686–5694.
37. Wilting,S.M. et al. (2009) Chromosomal signatures of a subset of highgrade premalignant cervical lesions closely resemble invasive carcinomas.
Cancer Res., 69, 647–655.
38. Koga,Y. et al. (2009) Genome-wide screen of promoter methylation identifies novel markers in melanoma. Genome Res., 19, 1462–1470.
39.Martin-Subero,J.I. et al. (2009) A comprehensive microarray-based
DNA methylation study of 367 hematological neoplasms. PLoS One, 4,
e6986.
40. Gal-Yam,E.N. et al. (2008) Frequent switching of Polycomb repressive
marks and DNA hypermethylation in the PC3 prostate cancer cell line.
Proc. Natl. Acad. Sci. USA, 105, 12979–12984.
41. Richter,J. et al. (2009) Array-based DNA methylation profiling of primary
lymphomas of the central nervous system. BMC Cancer, 9, 455.
42. Weber,M. et al. (2007) Silencing potential and evolutionary impact
of promoter DNA methylation in the human genome. Nat. Genet., 39,
457–466.
43. Ohm,J.E. et al. (2007) A stem cell-like chromatin pattern may predispose
tumour suppressor genes to DNA hypermethylation and heritable silencing. Nat. Genet., 39, 237–242.
44. Schlesinger,Y. et al. (2007) Polycomb-mediated methylation on Lys27 of
histone H3 pre-marks genes for de novo methylation in cancer. Nat. Genet.,
39, 232–236.
45. Widschwendter,M. et al. (2007) Epigenetic stem cell signature in cancer.
Nat. Genet., 39, 157–158.
46. Rakyan,V.K. et al. (2010) Human aging-associated DNA hypermethylation occurs preferentially at bivalent chromatin domains. Genome Res., 20,
434–439.
47. Zhao,X.D. et al. (2007) Whole-genome mapping of histone H3 Lys4
and 27 trimethylations reveals distinct genomic compartments in human
embryonic stem cells. Stem Cell, 1, 286–298.
48. Teschendorff,A.E. et al. (2010) Age-dependent DNA methylation of genes
that are suppressed in stem cells is a hallmark of cancer. Genome Res., 20,
440–446.
49. Ivanova,T. et al. (2002) Methylation and silencing of the retinoic acid
receptor-beta 2 gene in cervical cancer. BMC Cancer, 21, 2–4.
50. De-Castro Arce,J. et al. (2007) Retinoic acid receptor beta silences
human papillomavirus-18 oncogene expression by induction of de novo
methylation and heterochromatinization of the viral control region. J. Biol.
Chem., 282, 28520–28529.
51. Sawada,M. et al. (2007) Increased expression of DNA methyltransferase 1
(DNMT1) protein in uterine cervix squamous cell carcinoma and its precursor lesion. Cancer Lett., 251, 211–219.
52.Kalantari,M. et al. (2008) Effects of cellular differentiation, chromosomal integration and 5-aza-2’-deoxycytidine treatment on human papillomavirus-16 DNA methylation in cultured cell lines. Virology, 374,
292–303.
53. Fernandez,A.F. et al. (2009) The dynamic DNA methylomes of doublestranded DNA viruses associated with human cancer. Genome Res., 19,
438–451.
54. De-Castro Arce,J. et al. (2012) Silencing of multi-copy HPV16 by viral
self-methylation and chromatin occlusion: a model for epigenetic virushost interaction. Hum. Mol. Genet., Jan 3. [Epub ahead of print]
55.Leonard,S. et al. (2011) Epigenetic and transcriptional changes which follow
Epstein-Barr virus infection of germinal centre B cells and their relevance to
the pathogenesis of Hodgkin’s lymphoma. J. Virol., 85, 9568–9577.
56. McCabe,M.T. et al. (2009) A multifactorial signature of DNA sequence
and polycomb binding predicts aberrant CpG island methylation. Cancer
Res.,69, 282–291.
57. Buckley,P.G. et al. (2011) Genome-wide DNA methylation analysis of
neuroblastic tumours reveals clinically relevant epigenetic events and
large-scale epigenomic alterations localized to telomeric regions. Int. J.
Cancer, 128, 2296–2305.
58. de Wilde,J. et al. (2010) hTERT promoter activity and CpG methylation in
HPV-induced carcinogenesis. BMC Cancer, 10, 271.
59. De-Castro Arce,J. et al. (2007) Retinoic acid receptor beta silences human
papillomavirus-18 oncogene expression by induction of de novo methylation and heterochromatinization of the viral control region. Biol. Chem.,
282, 28520–28529.
60. Nakajima,T. et al. (2010) Persistence of a component of DNA methylation
in gastric mucosae after Helicobacter pylori eradication. J. Gastroenterol.,
45, 37–44.
61. Sepulveda,A.R. et al. (2010) CpG methylation and reduced expression of
O6-methylguanine DNA methyltransferase is associated with Helicobacter
pylori infection. Gastroenterology, 138, 1836–1844.
62. Ma,Y.T. et al. (2011) Smoking initiation is followed by the early acquisition of epigenetic change in cervical epithelium: a longitudinal study. Br. J.
Cancer, 104, 1500–1504.
Received December 7, 2012; revised March 23, 2012; accepted April 21,
2012
1293