G Model JCV-2906; No. of Pages 6
ARTICLE IN PRESS Journal of Clinical Virology xxx (2013) xxx–xxx
Contents lists available at ScienceDirect
Journal of Clinical Virology journal homepage: www.elsevier.com/locate/jcv
Quantitative measurement of Human Papillomavirus type 16 L1/L2 DNA methylation correlates with cervical disease grade Dean Bryant a , Amanda Tristram a , Triantafillos Liloglou b , Samantha Hibbitts a , Alison Fiander a , Ned Powell a,∗ a b
School of Medicine, Cardiff University, UK Department of Molecular and Clinical Cancer Medicine, University of Liverpool, UK
a r t i c l e
i n f o
Article history: Received 8 July 2013 Received in revised form 13 September 2013 Accepted 28 October 2013 Keywords: Human Papillomavirus HPV Cervical cancer Methylation
a b s t r a c t Background: Persistent infection with Human Papillomavirus (HPV) type 16 causes the majority of cervical cancers. Genital HPV infection is very common, but neoplastic progression is uncommon. There is an urgent need to identify biomarkers associated with cervical neoplasia that can be used to triage women who test positive for HPV. Objectives: To assess the ability of quantitative measurement of HPV16 DNA methylation to separate samples of different cytology grades and cervical cancers, and determine which of the assessed regions of the HPV genome and individual CpGs are most informative. Study design: DNA methylation was quantified by pyrosequencing of bisulphite converted DNA from liquid based cytology samples from 17 women with normal cytology and 20 women with severe dyskaryosis, and from fixed tissue from 24 women with cervical cancer. Methylation was assessed in the HPV Long Control Region (LCR), E2 and L1/L2 regions. Results: In cervical cancers, increased HPV DNA methylation was present in all regions. Increased methylation was also observed in severely dyskaryotic relative to normal samples, but only in the E2 and L1/L2 regions. The ability of methylation based classifiers to separate the three classes of material was assessed by ROC curve analyses. The best separation between normal and dyskaryotic samples was achieved by assessment of the L1/L2 CpGs at nucleotide positions 5600 and 5609 (AUC = 0.900, 95% CI: 0.793–1). Conclusions: This study demonstrates the potential of quantification of HPV DNA methylation as a biomarker of cervical neoplasia. An algorithm considering methylation at specific L1/L2 CpGs appeared the most promising model. © 2013 Elsevier B.V. All rights reserved.
1. Background DNA methylation is the covalent addition of a methyl group to the 5 position of cytosine or adenine, and occurs primarily at CpG dinucleotides. Dynamic methylation of DNA is a fundamental epigenetic mechanism that plays an important role in cancer and facilitates interaction between genotype and environment [1]. Cervical cancer is caused by persistent infection with high risk Human Papillomavirus (hrHPV). During carcinogenesis, substantial changes in methylation are observed in both the host cell
Abbreviations: HPV, Human Papillomavirus; LBC, liquid based cytology; CIN, cervical intraepithelial neoplasia; LCR, long control region; ROC, Receiver Operating Characteristic; AUC, Area Under Curve. ∗ Corresponding author at: HPV Research Group, School of Medicine, Cardiff University, Heath Park, Cardiff CF14 4XN, UK. Tel.: +44 02920 744742; fax: +44 2920 744399. E-mail address: [email protected] (N. Powell).
and Human Papillomavirus (HPV) genomes. Quantification of these changes may facilitate both diagnosis and prognostication [2,3]. Due to its high sensitivity and potential applicability to selfcollected samples, testing for hrHPV is likely to replace cytology as the primary screening method to prevent cervical cancer [4,5]. However HPV testing has significantly lower specificity than cytology, and therefore triage testing is required to prevent excessive referrals for colposcopy. Initially, cytology is likely to be used [6] but a molecular test that would allow reflex testing of HPV positive samples would potentially streamline workflows and increase efficiency. Unlike cytology, a molecular test could potentially be applicable to self-collected samples. Quantitative measurement of HPV DNA methylation shows significant promise as a simple test for triage of HPV positive women. Limited prospective data shows that in women infected with HPV16, the level of methylation of specific CpGs increases gradually with viral persistence and is highest in women with high grade neoplasia [3]. Cross-sectional studies have also demonstrated a trend for increased methylation with disease progression, but the results
1386-6532/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jcv.2013.10.029
Please cite this article in press as: Bryant D, et al. Quantitative measurement of Human Papillomavirus type 16 L1/L2 DNA methylation correlates with cervical disease grade. J Clin Virol (2013), http://dx.doi.org/10.1016/j.jcv.2013.10.029
G Model JCV-2906; No. of Pages 6
ARTICLE IN PRESS D. Bryant et al. / Journal of Clinical Virology xxx (2013) xxx–xxx
2
of these studies vary depending on the specific CpGs investigated, the assay used, and the type of material examined [7–10]. The aims of the current study were to develop and validate pyrosequencing assays to quantitatively assess HPV16 DNA methylation in the regions identified as potentially significant in previous studies and apply these assays to clinically relevant Liquid Based Cytology (LBC) samples. DNA from cervical cancers was assessed to determine the degree of association of HPV methylation with disease grade and analyse patterns of methylation across early and late stages of the disease. The study also aimed to identify the regions and CpGs of the HPV genome, at which DNA methylation level was most associated with disease grade. 2. Objectives The primary aim was to assess the ability of quantitative measurement of HPV16 DNA methylation, by pyrosequencing of bisulphite converted DNA, to separate samples of different cytology grades and cervical cancers. The secondary aim was to determine which of the assessed CpGs and regions of the HPV genome correlated best with disease grade. 3. Study design 3.1. Samples and DNA extraction Methylation of HPV DNA was investigated in 40 LBC samples and 27 cervical cancers. Twenty LBC samples had normal cytology and 20 severe dyskaryosis. DNA was extracted by overnight digestion in 50 mM Tris with 1 mg/ml Proteinase K followed by heat inactivation (100 ◦ C, 5 min) and centrifugation. HPV typing was performed using GP5+/6+ PCR ELISA [11,12]. For the cervical cancers, DNA was extracted using the Qiagen DNeasy Blood and Tissue kit (Qiagen, Hildesheim, Germany) without xylene washes [13]. Research Ethics Committee approval was obtained for use of all clinical material. 3.2. Sodium bisulphite treatment and pyrosequencing For each sample, sodium bisulphite treated DNA (500 ng) was prepared using the EZ DNA Methylation Kit (Zymo Research Corporation, CA, USA) according to the manufacturer’s instructions. Assays were designed to target two regions of the HPV16 LCR, as published studies suggested that methylation of the LCR may influence HPV gene expression; and to the E2 and L1/L2 overlap regions which have previously been identified as potentially clinically significant [14]. Assays were designed using PyroMark Assay Design SW 2.0 (Qiagen, Hilden, Germany). PCRs were performed using the primers described in Table 1 using ZymoTaq Premix hot start Taq (Zymo Research Corporation, CA, USA). The pyrosequencing reactions were conducted using PyroMark Gold Q96 Reagents and PyroMark Q96 ID Instrument (Qiagen, Hilden, Germany). 3.3. Statistical analyses All statistical analyses were performed using PASW statistics 18 and Minitab 16 statistical software. ROC curves and AUC were calculated for mean values of combinations of CpGs, representing regions of the HPV genome or selected CpGs. Kruskal–Wallis tests were performed to compare median methylation among disease groups for each HPV region, and each CpG tested. To account for multiple comparisons, P-values were corrected using the Benjamini–Hochberg FDR method to ensure that the expected proportion of falsely positive associations remained below 0.05 [15].
Table 1 Primer sequences and CpGs sequenced. LCR (Enhancer) Sense primer Antisense primer Sequencing primer Sequenced CpGs Amplified DNA
BTN-ATTGTATTATGTGTAATTATTGAA CCAAAAATATATACCTAACAAC CCAAAAATATATACCTAACAAC nt 7691, 7679, 7673 104 bp (nt 7611–7714)
LCR (Promoter) Sense primer Antisense primer Sequencing primer Sequenced CpGs Amplified DNA
GTAAAATTGTATATGGGTGT BTN-TAAAATATCTACTTTTATACTAACC TAATTTATGTATAAAATTAAGG nt 31, 37, 43, 52, 58 156 bp (nt 7832–83)
E2 Sense primer Antisense primer Sequencing primer Sequenced CpGs Amplified DNA
GTGAAATTATTAGGTAGTATTTGG BTN-CAACAACTTAATAATATAACAAAAA GTGAAATTATTAGGTAGTA nt 3411, 3414, 3416, 3432, 3435, 3447, 3461, 3472 148 bp (nt 3378–3525)
L1/L2 Sense primer Antisense primer Sequencing primer Sequenced CpGs Amplified DNA
BTN-TTATTGTTGATGTAGGTGATTT CCCAATAACCTCACTAAACAACC TAACCTCACTAAACAACCAA nt 5600, 5606, 5609, 5615 118 bp (nt 5551–5668)
BTN refers to the position of a biotin label. All nucleotide positions are relative to NC001526.1
Receiver Operating Characteristic (ROC) analysis was used to examine the relationship between the sensitivity and specificity using different cut-off values, and to investigate which features of the methylation data were potentially the most informative. ROC analyses were performed in several conformations, first, as non-cancer versus cancer, then as normal cytology versus severe dyskaryosis and cancer, and finally as normal cytology versus severe dyskaryosis. DNA methylation was considered as: mean methylation of all CpGs tested; mean methylation of each region; and three combinations of the four individual CpGs with the greatest differences in mean DNA methylation between the normal cytology and severe dyskaryosis groups. The performance of each combination of CpGs was assessed by comparing the Area Under Curve (AUC). 4. Results The pyrosequencing assays performed well with both LBC and FFPE material. The assay includes multiple controls (bisulphite conversion, complete extension of product, positive and negative template controls, background noise, reference sequence errors, etc.) and 6 samples that produced low quality data were excluded. The final analysis included 17 samples with normal cytology, 20 samples with severe dyskaryosis and 24 cancers (19 squamous cell carcinomas, 3 adenocarcinomas, 1 adenosquamous carcinoma and 1 carcinosarcoma). The mean ages for the three groups were 26.1 years for normal cytology, 31.4 years for dyskaryotic cytology, and 51.1 years for cancer. Within each disease group, there was no discernible relationship between patient age and HPV DNA methylation. DNA methylation varied among CpGs and regions, and also with severity of disease (Fig. 1). The HPV genome was most methylated in DNA derived from cervical cancers and least methylated in DNA derived from samples with normal cytology. Median DNA methylation was significantly different among the disease grades for all four regions tested (Kruskal–Wallis; P < 0.001 in all instances). Similarly, median DNA methylation of each CpG was significantly different among disease grades (P < 0.002). Samples with severe dyskaryosis had levels of methylation that were typically intermediate between
Please cite this article in press as: Bryant D, et al. Quantitative measurement of Human Papillomavirus type 16 L1/L2 DNA methylation correlates with cervical disease grade. J Clin Virol (2013), http://dx.doi.org/10.1016/j.jcv.2013.10.029
ARTICLE IN PRESS
G Model JCV-2906; No. of Pages 6
D. Bryant et al. / Journal of Clinical Virology xxx (2013) xxx–xxx
3
Table 2 AUC values for ROC analyses. Classifier
Area Under Curve (95% CI)
All CpGs E2 L1/L2 LCR 5600, 5609 3411, 3432 5600, 5609, 3411, 3432
Outcome 1: Cancer Outcome 2: Normal cytology and severe dyskaryosis
Outcome 1: Cancer or severe dyskaryosis Outcome 2: normal cytology
Outcome 1: Severe dyskaryosis Outcome 2: normal cytology
0.979 (0.944–1.000) 0.983 (0.956–1.000) 0.945 (0.878–1.000) 0.935 (0.867–1.000) 0.940 (0.861–1.000) 0.988a (0.966–1.000) 0.983 (0.955–1.000)
0.893 (0.800–0.985) 0.803 (0.680–0.926) 0.922 (0.839–1.000) 0.530 (0.367–0.693) 0.933 (0.860–1.000) 0.876 (0.779–0.973) 0.934a (0.866–1.000)
0.828 (0.689–0.968) 0.685 (0.508–0.862) 0.882 (0.763–1.000) 0.295 (0.121–0.469) 0.900a (0.793–1.000) 0.802 (0.657–0.947) 0.895 (0.790–1.000)
AUC and asymptotic 95% confidence intervals are shown for the potential classifiers. a Highest point estimate.
cancer and normal cytology. Differences in methylation between adjacent CpGs seemed to be conserved in several areas (i.e. nt 3411–3432 and 5600–5615). There also appeared to be regional differences in DNA methylation, with the E2 region becoming heavily methylated in cancers and the L1/L2 region becoming progressively more methylated with increasing disease severity. It was also notable that whilst the LCR was not heavily methylated in most cancers, the promoter was typically more methylated than the enhancer. ROC analysis demonstrated that methylation of all the regions assayed, either individually or in combination, differentiated well between samples from patients with cancer and those from women showing normal or severely dyskaryotic cytology (Fig. 2A). The AUC was greatest for mean methylation of CpGs 3411 and 34,332 (AUC = 0.988, 95% CI: 0.966–1), but none of the regions performed substantially better than any other (Table 2). ROC analysis to distinguish cancer or severe dyskaryosis from normal cytology (Fig. 2B) showed that good separation was achieved by analysis of CpGs in the E2 and L1/L2 regions, but not of CpGs in the LCR region. Comparing AUC showed that greatest discrimination was achieved using mean methylation at nucleotides (nt) 3411, 3432, 5600 and 5609 (AUC = 0.934, 95% CI: 0.866–1), although other combinations of areas/CpGs were similarly discriminatory. The same trends were observed in ROC analysis with only severe dyskaryosis and normal cytology as the outcomes (Fig. 2C). DNA methylation of the L1/L2 region appeared to discriminate best, whilst assessment of the LCR region was uninformative. Comparison of AUC (Table 2)
80 Cancer 70 Severe dyskaryosis DNA methylation %
60 Normal 50 40 30
showed mean methylation at L1/L2 nucleotides 5600 and 5609 as providing best separation overall (AUC = 0.900, 95% CI: 0.793–1) but again, other combinations produced similarly high AUC. AUC for all individual CpGs assayed are shown in Supplementary Table 1. Supplementary material related to this article can be found, in the online version, at http://dx.doi.org/10.1016/j.jcv.2013.10.029. ROC curves are used to assess the performance of a biomarker at various cut off points, with the intended use of the biomarker informing the appropriate compromise between sensitivity and specificity. HPV DNA methylation is proposed as a triage for HPV positive samples in primary screening, hence the ROC curve comparing normal cytology and severe dyskaryosis LBC samples is most relevant (Fig. 2C). The two classifiers with the highest AUC (0.900 and 0.895) both achieved 100% sensitivity with 66.7% specificity (Table 3). 5. Discussion This study demonstrates that HPV DNA methylation can be successfully assessed by pyrosequencing of bisulphite treated DNA extracted from routine LBC samples. Most importantly, it showed that HPV DNA methylation increased with severity of disease, especially in the L1/L2 and E2 regions. The potential application of this technique is for triage of HPV positive LBC samples; and therefore evidence that it can be applied to routine LBC samples is a significant strength. The major advantages of pyrosequencing methylation analysis are: (i) delivery of fully quantitative results from a heterogeneous mix of DNAs; (ii) assessment of multiple CpGs within a single assay; and (iii) inclusion of multiple rigorous quality controls, which are extremely important in in vitro diagnostic (IVD) tests. Furthermore, the assay was highly reproducible (assessed in multiple runs with HPV16 positive CaSki cell line DNA; Supplementary Fig. 1). Supplementary material related to this article can be found, in the online version, at http://dx.doi.org/10.1016/j.jcv.2013.10.029. The absence of a confirmed histological endpoint was a limitation; LBC samples were collected during an anonymous HPV prevalence study [11] and it was not possible to link these with
20
LCR (Enhancer)
LCR (Promoter)
E2
5615
5609
5606
3472 5600
3447 3461
3435
3416 3432
3414
58 3411
52
43
37
7691 31
0
7673 7979
10
L1/L2
Table 3 Sensitivity and specificity to separate samples with severe dyskaryosis from those with normal cytology. Classifier (CpGs)
Sensitivity (%)
Specificity (%)
Cut off (%)
5600 and 5609
100 80 60
66.7 86.7 93.3
9.8 15.9 19.5
3411, 3432, 5600 and 5609
100 80 60
66.7 73.3 93.3
5.2 8.8 12.4
CpG and region Fig. 1. Mean level of HPV DNA methylation at each CpG. Cancer = triangles, severe dyskaryosis = squares and normal cytology = diamonds. Error bars indicate standard error. Data points have been linked to aid interpretation; no relationship is implied.
Classifiers were based on mean methylation for the CpGs shown.
Please cite this article in press as: Bryant D, et al. Quantitative measurement of Human Papillomavirus type 16 L1/L2 DNA methylation correlates with cervical disease grade. J Clin Virol (2013), http://dx.doi.org/10.1016/j.jcv.2013.10.029
ARTICLE IN PRESS
G Model JCV-2906; No. of Pages 6
D. Bryant et al. / Journal of Clinical Virology xxx (2013) xxx–xxx
4
1.0
B
0.9
0.7
0.7
0.6
0.6
0.5
0.5
0.4 0.3
0.4 0.3
0.2
0.2
0.1
0.1 0.0
0.2
0.4
0.6
0.8
1.0
1-specificity
0.0
0.0
0.2
0.4
0.6
0.8
1.0
1-specificity
1.0 0.9
All CpGs E2 L1/L2 LCR 5600, 5609 3411, 3432 5600, 5609, 3411, 3432
0.8 0.7 0.6
Sensivity
0.9 0.8
0.0
C
1.0
0.8
Sensivity
Sensivity
A
0.5 0.4 0.3 0.2 0.1 0.0
0.0
0.2
0.4
0.6
0.8
1.0
1-specificity Fig. 2. ROC curves showing separation of samples by DNA methylation at different combinations of CpGs. The panels show separation of cases: (A) with cancer from those without cancer; (B) with severe dyskaryosis or cancer from those with normal cytology; (C) with severe cytology from those with normal cytology. Classifiers were based on mean methylation for the regions or individual CpGs indicated in the figure.
histology reports to confirm presence of Cervical Intraepithelial Neoplasia (CIN). However, in Wales, 84% of women referred for moderate and severe dyskaryosis show CIN2 or worse [16]. It is therefore highly likely that the majority of samples with severe dyskaryosis will be associated with clinically significant disease (i.e. CIN2+). The mean age differed significantly between the three groups (normal cytology: 26.1 years, severe dyskaryosis: 31.4 years and cancer: 51.1 years), and so increases in methylation could be associated with age. However, within each group, there was no significant correlation between age and methylation (Spearman 2 tailed P > 0.05). Furthermore, the duration of the infection may be more relevant than the actual patient age. Confounding by age could be partly controlled for by inclusion of age matched controls, but this would be an imperfect control as the duration of infection in the controls would not be known. This is consistent with a previous report that observed a correlation between methylation and age in women with CIN2-3, but not in age matched controls [17]. The modest sample size limits the generalisability of the results and they require replication in larger longitudinal sample sets linked to histological data. The data in Fig. 1 demonstrate that for the LCR and E2 CpGs assayed, there were only minor increases in methylation levels between normal and severely dyskaryotic cytology samples, however there were major increases (from 5–10% to 20–30%) in CpGs at 5600 and 5609 in the L1/L2 region. The largest AUCs were consistently achieved using classifiers based on combinations of CpGs and this analysis highlights the potential of the L1/L2 CpGs
at 5600 and 5609 to discriminate between normal samples and those likely to have come from women with cervical neoplasia. The assay is quantitative and hence cut off can be refined to provide optimal sensitivity and specificity. Since the assay may potentially be applied in cervical screening, sensitivity is a paramount consideration. In the current study (with severe dyskaryosis as the endpoint) 100% sensitivity could be achieved whilst retaining high (66.7%) specificity (albeit in a highly selected sample with an over-representation of dyskaryotic samples relative to the screening population). Our data are consistent with an investigation of the entire methylome of HPV16 in 18 samples representing normal cervix, CIN and cervical carcinoma, which suggested several regions of the HPV genome (especially L1 and L2) were progressively methylated with increasing disease severity [14]. Similar results were reported in a genome wide assessment of HPV DNA methylation in a Central American cohort which identified CpGs, located primarily in the L1, L2 and E2/E4 regions, where methylation differed significantly between disease grades [3]. Of three significant L1 CpGs, one (5609) was also highlighted in the current study. In an expanded follow-up study in the same cohort, a strong association was again observed between increased methylation in the L1/L2 regions and CIN and cancer [17]. Three of the CpGs assayed in the current study (nt 3432, 3435 and 5609) also demonstrated AUC values of >0.750 for separation of CIN2+ from HPV cleared controls in the follow-up study by Mirabello et al. [17] Although the two studies differ in design and outcome measures, these specific CpGs produced
Please cite this article in press as: Bryant D, et al. Quantitative measurement of Human Papillomavirus type 16 L1/L2 DNA methylation correlates with cervical disease grade. J Clin Virol (2013), http://dx.doi.org/10.1016/j.jcv.2013.10.029
G Model JCV-2906; No. of Pages 6
ARTICLE IN PRESS D. Bryant et al. / Journal of Clinical Virology xxx (2013) xxx–xxx
similar AUC in both settings (in our study, the AUC for separating severe dyskaryosis samples from normal cytology were: nt 3432 – 0.744 (95% CI: 0.571–0.906), nt 3435 – 0.573 (95% CI: 0.384–0.763) and nt 5609 – 0.858 (95% CI: 0.731–0.985), and in both studies the AUC were improved by using combinations of CpGs. A related study showed a correlation between increased methylation of several CpGs in the L1 and L2 regions and CIN2+ in women with low grade cytological abnormalities, and was able to differentiate
ARTICLE IN PRESS Journal of Clinical Virology xxx (2013) xxx–xxx
Contents lists available at ScienceDirect
Journal of Clinical Virology journal homepage: www.elsevier.com/locate/jcv
Quantitative measurement of Human Papillomavirus type 16 L1/L2 DNA methylation correlates with cervical disease grade Dean Bryant a , Amanda Tristram a , Triantafillos Liloglou b , Samantha Hibbitts a , Alison Fiander a , Ned Powell a,∗ a b
School of Medicine, Cardiff University, UK Department of Molecular and Clinical Cancer Medicine, University of Liverpool, UK
a r t i c l e
i n f o
Article history: Received 8 July 2013 Received in revised form 13 September 2013 Accepted 28 October 2013 Keywords: Human Papillomavirus HPV Cervical cancer Methylation
a b s t r a c t Background: Persistent infection with Human Papillomavirus (HPV) type 16 causes the majority of cervical cancers. Genital HPV infection is very common, but neoplastic progression is uncommon. There is an urgent need to identify biomarkers associated with cervical neoplasia that can be used to triage women who test positive for HPV. Objectives: To assess the ability of quantitative measurement of HPV16 DNA methylation to separate samples of different cytology grades and cervical cancers, and determine which of the assessed regions of the HPV genome and individual CpGs are most informative. Study design: DNA methylation was quantified by pyrosequencing of bisulphite converted DNA from liquid based cytology samples from 17 women with normal cytology and 20 women with severe dyskaryosis, and from fixed tissue from 24 women with cervical cancer. Methylation was assessed in the HPV Long Control Region (LCR), E2 and L1/L2 regions. Results: In cervical cancers, increased HPV DNA methylation was present in all regions. Increased methylation was also observed in severely dyskaryotic relative to normal samples, but only in the E2 and L1/L2 regions. The ability of methylation based classifiers to separate the three classes of material was assessed by ROC curve analyses. The best separation between normal and dyskaryotic samples was achieved by assessment of the L1/L2 CpGs at nucleotide positions 5600 and 5609 (AUC = 0.900, 95% CI: 0.793–1). Conclusions: This study demonstrates the potential of quantification of HPV DNA methylation as a biomarker of cervical neoplasia. An algorithm considering methylation at specific L1/L2 CpGs appeared the most promising model. © 2013 Elsevier B.V. All rights reserved.
1. Background DNA methylation is the covalent addition of a methyl group to the 5 position of cytosine or adenine, and occurs primarily at CpG dinucleotides. Dynamic methylation of DNA is a fundamental epigenetic mechanism that plays an important role in cancer and facilitates interaction between genotype and environment [1]. Cervical cancer is caused by persistent infection with high risk Human Papillomavirus (hrHPV). During carcinogenesis, substantial changes in methylation are observed in both the host cell
Abbreviations: HPV, Human Papillomavirus; LBC, liquid based cytology; CIN, cervical intraepithelial neoplasia; LCR, long control region; ROC, Receiver Operating Characteristic; AUC, Area Under Curve. ∗ Corresponding author at: HPV Research Group, School of Medicine, Cardiff University, Heath Park, Cardiff CF14 4XN, UK. Tel.: +44 02920 744742; fax: +44 2920 744399. E-mail address: [email protected] (N. Powell).
and Human Papillomavirus (HPV) genomes. Quantification of these changes may facilitate both diagnosis and prognostication [2,3]. Due to its high sensitivity and potential applicability to selfcollected samples, testing for hrHPV is likely to replace cytology as the primary screening method to prevent cervical cancer [4,5]. However HPV testing has significantly lower specificity than cytology, and therefore triage testing is required to prevent excessive referrals for colposcopy. Initially, cytology is likely to be used [6] but a molecular test that would allow reflex testing of HPV positive samples would potentially streamline workflows and increase efficiency. Unlike cytology, a molecular test could potentially be applicable to self-collected samples. Quantitative measurement of HPV DNA methylation shows significant promise as a simple test for triage of HPV positive women. Limited prospective data shows that in women infected with HPV16, the level of methylation of specific CpGs increases gradually with viral persistence and is highest in women with high grade neoplasia [3]. Cross-sectional studies have also demonstrated a trend for increased methylation with disease progression, but the results
1386-6532/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jcv.2013.10.029
Please cite this article in press as: Bryant D, et al. Quantitative measurement of Human Papillomavirus type 16 L1/L2 DNA methylation correlates with cervical disease grade. J Clin Virol (2013), http://dx.doi.org/10.1016/j.jcv.2013.10.029
G Model JCV-2906; No. of Pages 6
ARTICLE IN PRESS D. Bryant et al. / Journal of Clinical Virology xxx (2013) xxx–xxx
2
of these studies vary depending on the specific CpGs investigated, the assay used, and the type of material examined [7–10]. The aims of the current study were to develop and validate pyrosequencing assays to quantitatively assess HPV16 DNA methylation in the regions identified as potentially significant in previous studies and apply these assays to clinically relevant Liquid Based Cytology (LBC) samples. DNA from cervical cancers was assessed to determine the degree of association of HPV methylation with disease grade and analyse patterns of methylation across early and late stages of the disease. The study also aimed to identify the regions and CpGs of the HPV genome, at which DNA methylation level was most associated with disease grade. 2. Objectives The primary aim was to assess the ability of quantitative measurement of HPV16 DNA methylation, by pyrosequencing of bisulphite converted DNA, to separate samples of different cytology grades and cervical cancers. The secondary aim was to determine which of the assessed CpGs and regions of the HPV genome correlated best with disease grade. 3. Study design 3.1. Samples and DNA extraction Methylation of HPV DNA was investigated in 40 LBC samples and 27 cervical cancers. Twenty LBC samples had normal cytology and 20 severe dyskaryosis. DNA was extracted by overnight digestion in 50 mM Tris with 1 mg/ml Proteinase K followed by heat inactivation (100 ◦ C, 5 min) and centrifugation. HPV typing was performed using GP5+/6+ PCR ELISA [11,12]. For the cervical cancers, DNA was extracted using the Qiagen DNeasy Blood and Tissue kit (Qiagen, Hildesheim, Germany) without xylene washes [13]. Research Ethics Committee approval was obtained for use of all clinical material. 3.2. Sodium bisulphite treatment and pyrosequencing For each sample, sodium bisulphite treated DNA (500 ng) was prepared using the EZ DNA Methylation Kit (Zymo Research Corporation, CA, USA) according to the manufacturer’s instructions. Assays were designed to target two regions of the HPV16 LCR, as published studies suggested that methylation of the LCR may influence HPV gene expression; and to the E2 and L1/L2 overlap regions which have previously been identified as potentially clinically significant [14]. Assays were designed using PyroMark Assay Design SW 2.0 (Qiagen, Hilden, Germany). PCRs were performed using the primers described in Table 1 using ZymoTaq Premix hot start Taq (Zymo Research Corporation, CA, USA). The pyrosequencing reactions were conducted using PyroMark Gold Q96 Reagents and PyroMark Q96 ID Instrument (Qiagen, Hilden, Germany). 3.3. Statistical analyses All statistical analyses were performed using PASW statistics 18 and Minitab 16 statistical software. ROC curves and AUC were calculated for mean values of combinations of CpGs, representing regions of the HPV genome or selected CpGs. Kruskal–Wallis tests were performed to compare median methylation among disease groups for each HPV region, and each CpG tested. To account for multiple comparisons, P-values were corrected using the Benjamini–Hochberg FDR method to ensure that the expected proportion of falsely positive associations remained below 0.05 [15].
Table 1 Primer sequences and CpGs sequenced. LCR (Enhancer) Sense primer Antisense primer Sequencing primer Sequenced CpGs Amplified DNA
BTN-ATTGTATTATGTGTAATTATTGAA CCAAAAATATATACCTAACAAC CCAAAAATATATACCTAACAAC nt 7691, 7679, 7673 104 bp (nt 7611–7714)
LCR (Promoter) Sense primer Antisense primer Sequencing primer Sequenced CpGs Amplified DNA
GTAAAATTGTATATGGGTGT BTN-TAAAATATCTACTTTTATACTAACC TAATTTATGTATAAAATTAAGG nt 31, 37, 43, 52, 58 156 bp (nt 7832–83)
E2 Sense primer Antisense primer Sequencing primer Sequenced CpGs Amplified DNA
GTGAAATTATTAGGTAGTATTTGG BTN-CAACAACTTAATAATATAACAAAAA GTGAAATTATTAGGTAGTA nt 3411, 3414, 3416, 3432, 3435, 3447, 3461, 3472 148 bp (nt 3378–3525)
L1/L2 Sense primer Antisense primer Sequencing primer Sequenced CpGs Amplified DNA
BTN-TTATTGTTGATGTAGGTGATTT CCCAATAACCTCACTAAACAACC TAACCTCACTAAACAACCAA nt 5600, 5606, 5609, 5615 118 bp (nt 5551–5668)
BTN refers to the position of a biotin label. All nucleotide positions are relative to NC001526.1
Receiver Operating Characteristic (ROC) analysis was used to examine the relationship between the sensitivity and specificity using different cut-off values, and to investigate which features of the methylation data were potentially the most informative. ROC analyses were performed in several conformations, first, as non-cancer versus cancer, then as normal cytology versus severe dyskaryosis and cancer, and finally as normal cytology versus severe dyskaryosis. DNA methylation was considered as: mean methylation of all CpGs tested; mean methylation of each region; and three combinations of the four individual CpGs with the greatest differences in mean DNA methylation between the normal cytology and severe dyskaryosis groups. The performance of each combination of CpGs was assessed by comparing the Area Under Curve (AUC). 4. Results The pyrosequencing assays performed well with both LBC and FFPE material. The assay includes multiple controls (bisulphite conversion, complete extension of product, positive and negative template controls, background noise, reference sequence errors, etc.) and 6 samples that produced low quality data were excluded. The final analysis included 17 samples with normal cytology, 20 samples with severe dyskaryosis and 24 cancers (19 squamous cell carcinomas, 3 adenocarcinomas, 1 adenosquamous carcinoma and 1 carcinosarcoma). The mean ages for the three groups were 26.1 years for normal cytology, 31.4 years for dyskaryotic cytology, and 51.1 years for cancer. Within each disease group, there was no discernible relationship between patient age and HPV DNA methylation. DNA methylation varied among CpGs and regions, and also with severity of disease (Fig. 1). The HPV genome was most methylated in DNA derived from cervical cancers and least methylated in DNA derived from samples with normal cytology. Median DNA methylation was significantly different among the disease grades for all four regions tested (Kruskal–Wallis; P < 0.001 in all instances). Similarly, median DNA methylation of each CpG was significantly different among disease grades (P < 0.002). Samples with severe dyskaryosis had levels of methylation that were typically intermediate between
Please cite this article in press as: Bryant D, et al. Quantitative measurement of Human Papillomavirus type 16 L1/L2 DNA methylation correlates with cervical disease grade. J Clin Virol (2013), http://dx.doi.org/10.1016/j.jcv.2013.10.029
ARTICLE IN PRESS
G Model JCV-2906; No. of Pages 6
D. Bryant et al. / Journal of Clinical Virology xxx (2013) xxx–xxx
3
Table 2 AUC values for ROC analyses. Classifier
Area Under Curve (95% CI)
All CpGs E2 L1/L2 LCR 5600, 5609 3411, 3432 5600, 5609, 3411, 3432
Outcome 1: Cancer Outcome 2: Normal cytology and severe dyskaryosis
Outcome 1: Cancer or severe dyskaryosis Outcome 2: normal cytology
Outcome 1: Severe dyskaryosis Outcome 2: normal cytology
0.979 (0.944–1.000) 0.983 (0.956–1.000) 0.945 (0.878–1.000) 0.935 (0.867–1.000) 0.940 (0.861–1.000) 0.988a (0.966–1.000) 0.983 (0.955–1.000)
0.893 (0.800–0.985) 0.803 (0.680–0.926) 0.922 (0.839–1.000) 0.530 (0.367–0.693) 0.933 (0.860–1.000) 0.876 (0.779–0.973) 0.934a (0.866–1.000)
0.828 (0.689–0.968) 0.685 (0.508–0.862) 0.882 (0.763–1.000) 0.295 (0.121–0.469) 0.900a (0.793–1.000) 0.802 (0.657–0.947) 0.895 (0.790–1.000)
AUC and asymptotic 95% confidence intervals are shown for the potential classifiers. a Highest point estimate.
cancer and normal cytology. Differences in methylation between adjacent CpGs seemed to be conserved in several areas (i.e. nt 3411–3432 and 5600–5615). There also appeared to be regional differences in DNA methylation, with the E2 region becoming heavily methylated in cancers and the L1/L2 region becoming progressively more methylated with increasing disease severity. It was also notable that whilst the LCR was not heavily methylated in most cancers, the promoter was typically more methylated than the enhancer. ROC analysis demonstrated that methylation of all the regions assayed, either individually or in combination, differentiated well between samples from patients with cancer and those from women showing normal or severely dyskaryotic cytology (Fig. 2A). The AUC was greatest for mean methylation of CpGs 3411 and 34,332 (AUC = 0.988, 95% CI: 0.966–1), but none of the regions performed substantially better than any other (Table 2). ROC analysis to distinguish cancer or severe dyskaryosis from normal cytology (Fig. 2B) showed that good separation was achieved by analysis of CpGs in the E2 and L1/L2 regions, but not of CpGs in the LCR region. Comparing AUC showed that greatest discrimination was achieved using mean methylation at nucleotides (nt) 3411, 3432, 5600 and 5609 (AUC = 0.934, 95% CI: 0.866–1), although other combinations of areas/CpGs were similarly discriminatory. The same trends were observed in ROC analysis with only severe dyskaryosis and normal cytology as the outcomes (Fig. 2C). DNA methylation of the L1/L2 region appeared to discriminate best, whilst assessment of the LCR region was uninformative. Comparison of AUC (Table 2)
80 Cancer 70 Severe dyskaryosis DNA methylation %
60 Normal 50 40 30
showed mean methylation at L1/L2 nucleotides 5600 and 5609 as providing best separation overall (AUC = 0.900, 95% CI: 0.793–1) but again, other combinations produced similarly high AUC. AUC for all individual CpGs assayed are shown in Supplementary Table 1. Supplementary material related to this article can be found, in the online version, at http://dx.doi.org/10.1016/j.jcv.2013.10.029. ROC curves are used to assess the performance of a biomarker at various cut off points, with the intended use of the biomarker informing the appropriate compromise between sensitivity and specificity. HPV DNA methylation is proposed as a triage for HPV positive samples in primary screening, hence the ROC curve comparing normal cytology and severe dyskaryosis LBC samples is most relevant (Fig. 2C). The two classifiers with the highest AUC (0.900 and 0.895) both achieved 100% sensitivity with 66.7% specificity (Table 3). 5. Discussion This study demonstrates that HPV DNA methylation can be successfully assessed by pyrosequencing of bisulphite treated DNA extracted from routine LBC samples. Most importantly, it showed that HPV DNA methylation increased with severity of disease, especially in the L1/L2 and E2 regions. The potential application of this technique is for triage of HPV positive LBC samples; and therefore evidence that it can be applied to routine LBC samples is a significant strength. The major advantages of pyrosequencing methylation analysis are: (i) delivery of fully quantitative results from a heterogeneous mix of DNAs; (ii) assessment of multiple CpGs within a single assay; and (iii) inclusion of multiple rigorous quality controls, which are extremely important in in vitro diagnostic (IVD) tests. Furthermore, the assay was highly reproducible (assessed in multiple runs with HPV16 positive CaSki cell line DNA; Supplementary Fig. 1). Supplementary material related to this article can be found, in the online version, at http://dx.doi.org/10.1016/j.jcv.2013.10.029. The absence of a confirmed histological endpoint was a limitation; LBC samples were collected during an anonymous HPV prevalence study [11] and it was not possible to link these with
20
LCR (Enhancer)
LCR (Promoter)
E2
5615
5609
5606
3472 5600
3447 3461
3435
3416 3432
3414
58 3411
52
43
37
7691 31
0
7673 7979
10
L1/L2
Table 3 Sensitivity and specificity to separate samples with severe dyskaryosis from those with normal cytology. Classifier (CpGs)
Sensitivity (%)
Specificity (%)
Cut off (%)
5600 and 5609
100 80 60
66.7 86.7 93.3
9.8 15.9 19.5
3411, 3432, 5600 and 5609
100 80 60
66.7 73.3 93.3
5.2 8.8 12.4
CpG and region Fig. 1. Mean level of HPV DNA methylation at each CpG. Cancer = triangles, severe dyskaryosis = squares and normal cytology = diamonds. Error bars indicate standard error. Data points have been linked to aid interpretation; no relationship is implied.
Classifiers were based on mean methylation for the CpGs shown.
Please cite this article in press as: Bryant D, et al. Quantitative measurement of Human Papillomavirus type 16 L1/L2 DNA methylation correlates with cervical disease grade. J Clin Virol (2013), http://dx.doi.org/10.1016/j.jcv.2013.10.029
ARTICLE IN PRESS
G Model JCV-2906; No. of Pages 6
D. Bryant et al. / Journal of Clinical Virology xxx (2013) xxx–xxx
4
1.0
B
0.9
0.7
0.7
0.6
0.6
0.5
0.5
0.4 0.3
0.4 0.3
0.2
0.2
0.1
0.1 0.0
0.2
0.4
0.6
0.8
1.0
1-specificity
0.0
0.0
0.2
0.4
0.6
0.8
1.0
1-specificity
1.0 0.9
All CpGs E2 L1/L2 LCR 5600, 5609 3411, 3432 5600, 5609, 3411, 3432
0.8 0.7 0.6
Sensivity
0.9 0.8
0.0
C
1.0
0.8
Sensivity
Sensivity
A
0.5 0.4 0.3 0.2 0.1 0.0
0.0
0.2
0.4
0.6
0.8
1.0
1-specificity Fig. 2. ROC curves showing separation of samples by DNA methylation at different combinations of CpGs. The panels show separation of cases: (A) with cancer from those without cancer; (B) with severe dyskaryosis or cancer from those with normal cytology; (C) with severe cytology from those with normal cytology. Classifiers were based on mean methylation for the regions or individual CpGs indicated in the figure.
histology reports to confirm presence of Cervical Intraepithelial Neoplasia (CIN). However, in Wales, 84% of women referred for moderate and severe dyskaryosis show CIN2 or worse [16]. It is therefore highly likely that the majority of samples with severe dyskaryosis will be associated with clinically significant disease (i.e. CIN2+). The mean age differed significantly between the three groups (normal cytology: 26.1 years, severe dyskaryosis: 31.4 years and cancer: 51.1 years), and so increases in methylation could be associated with age. However, within each group, there was no significant correlation between age and methylation (Spearman 2 tailed P > 0.05). Furthermore, the duration of the infection may be more relevant than the actual patient age. Confounding by age could be partly controlled for by inclusion of age matched controls, but this would be an imperfect control as the duration of infection in the controls would not be known. This is consistent with a previous report that observed a correlation between methylation and age in women with CIN2-3, but not in age matched controls [17]. The modest sample size limits the generalisability of the results and they require replication in larger longitudinal sample sets linked to histological data. The data in Fig. 1 demonstrate that for the LCR and E2 CpGs assayed, there were only minor increases in methylation levels between normal and severely dyskaryotic cytology samples, however there were major increases (from 5–10% to 20–30%) in CpGs at 5600 and 5609 in the L1/L2 region. The largest AUCs were consistently achieved using classifiers based on combinations of CpGs and this analysis highlights the potential of the L1/L2 CpGs
at 5600 and 5609 to discriminate between normal samples and those likely to have come from women with cervical neoplasia. The assay is quantitative and hence cut off can be refined to provide optimal sensitivity and specificity. Since the assay may potentially be applied in cervical screening, sensitivity is a paramount consideration. In the current study (with severe dyskaryosis as the endpoint) 100% sensitivity could be achieved whilst retaining high (66.7%) specificity (albeit in a highly selected sample with an over-representation of dyskaryotic samples relative to the screening population). Our data are consistent with an investigation of the entire methylome of HPV16 in 18 samples representing normal cervix, CIN and cervical carcinoma, which suggested several regions of the HPV genome (especially L1 and L2) were progressively methylated with increasing disease severity [14]. Similar results were reported in a genome wide assessment of HPV DNA methylation in a Central American cohort which identified CpGs, located primarily in the L1, L2 and E2/E4 regions, where methylation differed significantly between disease grades [3]. Of three significant L1 CpGs, one (5609) was also highlighted in the current study. In an expanded follow-up study in the same cohort, a strong association was again observed between increased methylation in the L1/L2 regions and CIN and cancer [17]. Three of the CpGs assayed in the current study (nt 3432, 3435 and 5609) also demonstrated AUC values of >0.750 for separation of CIN2+ from HPV cleared controls in the follow-up study by Mirabello et al. [17] Although the two studies differ in design and outcome measures, these specific CpGs produced
Please cite this article in press as: Bryant D, et al. Quantitative measurement of Human Papillomavirus type 16 L1/L2 DNA methylation correlates with cervical disease grade. J Clin Virol (2013), http://dx.doi.org/10.1016/j.jcv.2013.10.029
G Model JCV-2906; No. of Pages 6
ARTICLE IN PRESS D. Bryant et al. / Journal of Clinical Virology xxx (2013) xxx–xxx
similar AUC in both settings (in our study, the AUC for separating severe dyskaryosis samples from normal cytology were: nt 3432 – 0.744 (95% CI: 0.571–0.906), nt 3435 – 0.573 (95% CI: 0.384–0.763) and nt 5609 – 0.858 (95% CI: 0.731–0.985), and in both studies the AUC were improved by using combinations of CpGs. A related study showed a correlation between increased methylation of several CpGs in the L1 and L2 regions and CIN2+ in women with low grade cytological abnormalities, and was able to differentiate
