Human Papillomavirus 16, 18, 31 and 45 viral load, integration and methylation status stratified by cervical disease stage

Persistent infection with oncogenic Human Papillomavirus (HPV) is associated with the development of cervical cancer with each genotype differing in their relative contribution to the prevalence of cervical disease. HPV DNA testing offers improved sensitivity over cytology testing alone but is accompanied by a generally low specificity. Potential molecular markers of cervical disease include type-specific viral load (VL), integration of HPV DNA into the host genome and methylation of the HPV genome. The aim of this study was to evaluate the relationship between HPV type-specific viral load, integration and methylation status and cervical disease stage in samples harboring HPV16, HPV18, HPV31 or HPV45. Samples singly infected with HPV16 (n = 226), HPV18 (n = 32), HPV31 (n = 75) or HPV45 (n = 29) were selected from a cohort of 4,719 women attending cervical screening in England. Viral load and integration status were determined by real-time PCR while 3’L1-URR methylation status was determined by pyrosequencing or sequencing of multiple clones derived from each sample. Viral load could differentiate between normal and abnormal cytology with a sensitivity of 75% and a specificity of 80% (odds ratio [OR] 12.4, 95% CI 6.2–26.1; p < 0.001) with some variation between genotypes. Viral integration was poorly associated with cervical disease. Few samples had fully integrated genomes and these could be found throughout the course of disease. Overall, integration status could distinguish between normal and abnormal cytology with a sensitivity of 72% and a specificity of 50% (OR 2.6, 95% CI 1.0–6.8; p = 0.054). Methylation levels were able to differentiate normal and low grade cytology from high grade cytology with a sensitivity of 64% and a specificity of 82% (OR 8.2, 95% CI 3.8–18.0; p < 0.001). However, methylation varied widely between genotypes with HPV18 and HPV45 exhibiting a broader degree and higher magnitude of methylated CpG sites than HPV16 and HPV31. This study lends support for HPV viral load and CpG methylation status, but not integration status, to be considered as potential biomarkers of cervical disease.


Background
Persistent infection with oncogenic genotypes of genital Human Papillomavirus (HPV) is associated with the development of cervical cancer, a significant cause of morbidity and mortality of women worldwide [1]. Precancerous cervical disease is classified by cytological (low [LSIL] or high grade [HSIL] squamous intraepithelial lesions) and histological stages (cervical intraepithelial neoplasia [CIN] grades 1 to 3). There are about a dozen HPV types associated with the development of cervical cancer [2], differing in their relative contributions to the prevalence of cervical disease [3]. Testing for the presence of oncogenic HPV DNA offers improved sensitivity, though lower specificity, than cytology alone [4] while the next generation of molecular tests, including those with limited genotyping capability, may improve upon this [5]. Other potential molecular markers of cervical disease include type-specific viral load (VL), integration of HPV DNA into the host genome and methylation of the HPV genome. An improved understanding of the role of these potential molecular markers in cervical disease development may shed some light on HPV pathogenesis and may be helpful to guide future cervical cancer screening or treatment algorithms.
HPV DNA VL, usually estimated as the amount of HPV genome copies per cell, has been variably associated with cervical disease. Some studies were able to use HPV16 VL to differentiate between high grade (HSIL, CIN2+) and low grade (LSIL, CIN1) disease [6][7][8], between cervical cancer and lower grades of disease [9][10][11], or between any grade of cervical disease and normal samples [12]. Other studies could not find any association between HPV16 VL and cervical disease [13,14]. There are few studies examining any potential link between VL and cervical disease for other HPV types with some finding a positive association between some stages of disease for HPV18 [8,9], HPV31 [8], HPV33 [8], and HPV52 [9] while others have not [9,[14][15][16][17].
HPV16 integration status has been able to distinguish between HSIL and LSIL samples [6], between cervical cancer samples and those of a lower grade of disease [9,10], or in some studies a strong positive correlation with increasing disease severity has been found [18]. In other studies no relationship between cervical disease grade and HPV16 integration status was apparent [11][12][13]19]. For HPV18 there are fewer studies overall with some finding an association with disease [9,18] and some not [16]. One study [18], found a strong positive correlation with increasing disease severity for HPV31, HPV33 and HPV45 while not for HPV52 [9,15] and HPV58 [9,17] in others.
Mixed infections are common throughout the course of cervical disease [3]. Few of these studies have explicitly used, or separately analyzed, samples harboring a single HPV type wherein the association between the HPV type under evaluation and cervical disease can be made with some confidence. Within these limited number of studies, the VL of samples harboring single infections has been associated with disease severity in some [7] but not in other studies [13,14,17]. For integration-based studies that explicitly mentioned the use of single infection samples, one study found an association between HPV16 integration status and disease [7] while another did not [13]. The only study explicitly to examine methylation levels in single infections (HPV18 and HPV31) demonstrated that some CpG sites exhibited higher methylation levels in CIN3 cases harboring single infections than in CIN3 cases with multiple infections with these types [29].
In this study, we evaluate the DNA viral load, integration and CpG methylation status of women singly infected with HPV16, HPV18, HPV31 or HPV45 in order better to understand the potential role for these molecular markers in cervical disease.

Viral integration
The integration status of HPV16 [13], HPV18 [16], HPV31 and HPV45 was assessed using a ratio of the E2 gene copies over the E6 gene copies per reaction. Two estimates were made using an amino-terminal (Nt) and a carboxy-terminal (Ct) E2 PCR (Additional file 1: Table S1). PCR amplification was carried out on the ABI 7500 Fast PCR platform using Platinum SYBR green qPCR SuperMix-UDG (Life Technologies). Discriminatory power was determined using genome (representing episomal DNA) and E6 only (representing integrated DNA) plasmids in a background of C33A cells to simulate integration proportions of 0, 20, 50, 80, and 100% for each type. Linearity of these calibrators was good (median r 2 0.998 [IQR 0.994 -0.998]). The lower limit of the 99% confidence interval (CI) for E2/E6 ratios obtained using the full length plasmid (0% integration) could be differentiated from the upper 99% CI of the E2/E6 ratios for samples containing a simulated integration level of 20% (p ≤ 0.01). This empirical threshold was used to differentiate between samples bearing fully episomal or mixed forms of the virus [7,11]. If the E2/E6 ratio for both Nt and Ct E2 fragments was above the respective empirical threshold, the sample was considered as bearing only episomal forms of the virus. If one or both fragments were not amplified, the sample was considered to bear fully integrated virus, otherwise the sample was designated as containing mixed forms of the virus.

Statistical analysis
The Mann Whitney U test and the test for trend were used to evaluate differences between two groups and three independent groups, respectively. Kruskal-Wallis was used to test for differences between multiple groups. The Wilcoxon paired sign rank test was used to test for differences between two groups of paired data.
Receiver operating characteristic (ROC) analyses were used to evaluate whether a parameter could differentiate between cases (HG with or without LG cytology samples) and controls (normal with or without LG cytology samples). The area under the curve (AUC) is a measure of how well a parameter could differentiate between these two groups (cases and controls), where no differentiation yields a value of 0.500 (equality) and scores of 0.800 or above are considered strong. The crossing point for the ROC curve was derived using the maximum Youden index (J) yielding the optimum balance of sensitivity and specificity and the resulting threshold of the parameter under study. The Fisher's exact test was used to test for differences in proportions between cases and controls with crude odds ratios (95% CI) also given.
Tests were 2-tailed where appropriate and all tests were carried out using Stata 12.1 (StataCorp, USA).

Viral load
The distribution of HPV16, HPV18, HPV31 and HPV45 DNA VL by cytology grade is shown in Figure 1A We next wished to evaluate whether VL could differentiate between normal cytology and cervical disease with an appropriate degree of sensitivity and specificity. ROC plots for normal (control) versus LG and HG cytology (cases) samples are shown in Figure 1B while the resulting sensitivity and specificity plots for detection of disease are shown in Figure 2. HPV16 VL was able to distinguish between cases and controls with a sensitivity of 70% and a specificity of 91% (AUC 0.859) at a VL threshold of 1.58 c/c (OR 24.5 [95% CI 7.1 -128.2]; p < 0.001). HPV18 VL was able to distinguish between cases and controls with a sensitivity of 81% and a specificity of 80% (AUC 0.800) at a VL threshold of 0.04 (OR 17.6 [1.2 -904.3]; p = 0.015). HPV31 VL was able to distinguish between cases and controls with a sensitivity of 86% and a specificity of 78% (AUC 0.862) at a VL threshold of 0.22 (OR 20.3 [4.5 -101.8]; p < 0.001). HPV45 VL was able to distinguish between cases and controls with a sensitivity of 80% and specificity of 78% (AUC 0.778) at a VL threshold of 0.51 (OR 12.0 [1.6 -99.0]; p = 0.005). HPV VL overall was able to distinguish between cases and controls with a sensitivity of 75% and These data lend support to the potential for VL to be used to differentiate between normal and cytologically abnormal samples for a range of HPV types. Most other studies either did not include samples displaying normal cytology or histology [7][8][9], or could not make such a determination when they did [6,11,[13][14][15][16][17]. Where ROC or similar analyses have been applied, two studies found that HPV16 VL could differentiate between HSIL and LSIL with a sensitivity of 50% and a specificity of 90% [6,7]. Another study found that VL levels could differentiate between low grade CIN and CIN2+ with improved sensitivity compared with cytology alone, but with an unacceptable impact on specificity [8]. Yet another study was able to differentiate between cervical cancer samples and those exhibiting CIN2 or CIN3 disease using HPV16, HPV18 and HPV52 (but not HPV58) VL levels [9]. It is not clear why there are such conflicting reports on the potential for an association between HPV VL and cervical disease, but the use of single infection samples may have helped in this respect.

Viral integration
A comparison of the physical status of the HPV genome with cytological grade is shown in Figure 3. While a small proportion of samples harbored fully integrated HPV (n = 12 of 163 samples, 7.4%), the majority contained either a mixture of integrated and episomal forms or purely episomal forms. Of the fully integrated genomes (n = 12), samples lacking only the Ct E2 fragment or lacking both the Nt and Ct E2 fragments were more common (n = 4 [33%] and n = 7 [58%], respectively) than was disruption of the Nt E2 fragment alone (n = 1; 8%). Where this has been examined, studies report that disruption is commonly found in the Nt fragment of the E2 gene and entire E2 deletion is rare [13,15,17,35].
ROC analyses for the Nt E2/E6 and Ct E2/E6 ratios for HPV16 showed poor discrimination (AUC 0.430 and 0.415, respectively). The proportion of LG or HG cytology samples containing mixed or fully integrated forms was similar to those of the normal samples for HPV16 (OR 2.4 [95% CI 0.7 -9.1]; p = 0.154) and HPV31 (OR 2.6 [0.3 -18.0]; p = 0.348). For HPV18 and HPV45 there were too few normal samples to perform an appropriate analysis. For all samples, regardless of HPV type, there was a tendency for a slightly higher proportion of samples containing mixed or fully integrated forms in the cases (LG and HG samples) than in the controls (normal samples) as suggested by the OR of 2.6 (95% CI 1.0 -6.8; p = 0.054) accompanied by a sensitivity of 72% but a poor specificity of 50% ( Figure 2).
We used an empirical threshold [7,11] since in reconstitution experiments we could distinguish between simulated 0% and 20% integration levels. However, we also examined the potential for an association between integration status and disease using a range of thresholds (1.0, 0.8 and 0.5) for comparison with other studies [6,9,[15][16][17]19]. Reducing the threshold had the effect of reducing the number of mixed forms leading to a decrease in the sensitivity and increase in the specificity for detection of LG or HG disease, although not reaching significance at any threshold (Additional file 2: Figure S1).
Although some studies could differentiate between cervical disease stages, usually between LG and HG  Figure 2 Sensitivity and specificity plot. Sensitivity and specificity plot with 95% CI shown for viral load (black), integration (blue) and methylation (red) status. Viral load and integration status were used to differentiate between normal and abnormal (LG and HG) cytology, whilst methylation status was used to differentiate normal and LG from HG samples.
disease using integration status [6,7,9], many could not [11][12][13][15][16][17]19]. Despite using samples harboring only single HPV types, where potentially confounding effects of targeting a non-disease associated type in a mixed infection sample should have been reduced, these data suggest that HPV integration status, at least as measured by E2/E6 PCR, may not be appropriately sensitive for utility as a diagnostic tool for cervical disease.
Other assays such as APOT and/or DIPS may have better utility in this regard [18].

CpG methylation
The methylation status of CpG sites within the 3'L1-URR fragment of HPV16 (n = 79 samples) and HPV18 (n = 18) and of sites within the 3'L1 fragment of HPV31 (n = 50) and HPV45 (n = 18) are shown in Figure 4. There were multiple CpG sites within the 3'L1-URR fragment of HPV16 that displayed low levels of methylation across all grades of cytology. On an individual CpG site basis, there were no sites whose methylation levels were able to distinguish normal samples from LG and HG samples (p > 0.05). However, there were three sites able to separate HG samples from LG and normal samples: one site in L1 (bp 7089; p = 0.011) and two in the URR (7268, p = 0.021 and 7431a, p = 0.007) ( Figure 4A). The average methylation of these three sites could distinguish normal and LG (controls) from HG (cases) cytology samples with a sensitivity of 67% and a   Figure 3 Integration status distribution according to disease grade and genotype. Number and percentage of normal, LG or HG cytology samples exhibiting episomal (white boxes), mixed (grey) or fully integrated (black) HPV genomes for each HPV type. specificity of 86% (AUC 0.768) at a threshold of 5.3% (OR 12.8 [95%CI 3.7 -49.6]; p < 0.001) (Figure 2). CpG site methylation in the 3'L1-URR fragment of HPV18 was similarly low in the normal and LG samples, but considerably higher in some of the HG samples, typically those classified as being severely dyskaryotic ( Figure 4B). HG cytology samples could be distinguished from normal and LG samples with a sensitivity of 78% and a specificity of 89% (AUC 0.901) at an average methylation threshold of 6.7% across the nine HG disease-associated sites (OR 28.0 [95%CI 1.5 -1413.3; p = 0.015) (Figure 2).
Of the sites in the short 3'L1 fragment of HPV31, methylation of the CpG site 6950 ( Figure 4C) was weakly able to distinguish normal and LG samples from HG samples with a sensitivity of 50% and a specificity of 77% (AUC 0.656) at a threshold of 20% (OR 3.3 [95% CI 0.8 -13.3]; p = 0.07) (Figure 2). For HPV45 there were nine sites associated with HG disease ( Figure 4D) such that normal and LG samples could be distinguished from HG samples with a sensitivity of 75% and a specificity of 100% (AUC 0.839) at an average methylation threshold of 16.1% (p = 0.005) (Figure 2).
Overall, normal and LG samples could be distinguished from HG samples with a sensitivity of 64% and a specificity of 82% (AUC 0.746) at an average methylation level of 5.3% (OR 8.2 [95% CI 3.8 -18.0; p < 0.001) ( Figure 2). However, given the wide disparity in breadth and magnitude of methylated CpG sites between HPV genotypes, it may not be helpful to make such a generalization and perhaps highlights the need to treat each genotype individually.
These observations corroborate and extend previous reports on HPV16 [20,21,25] and HPV18 [25,29,30] demonstrating that while some CpG sites in HPV16 may be weakly associated with increasing cervical disease severity, HPV18-associated high grade cervical disease is characterized by broad CpG site methylation of a high magnitude. Two recent studies suggest that the methylation status of certain key HPV16 CpG sites may be able to differentiate between cervical cancer and low grade disease [22,36] and may even have the utility to be predictive [22]. Data on the methylation profiles of HPV31 and HPV45 are limited [29,37]. One study demonstrated differences between CIN3 cases and asymptomatic controls across a range of sites [29], while the other reported the proportion, but not the magnitude, of methylated sites in the HPV genome and a cellular gene increased with increasing disease severity [37]. The present data appear to suggest that this highly methylated state probably only occurs late in disease. In addition, these data corroborate the finding that there is an apparent species-specific association in HPV genome methylation [29], given the quite different methylation profiles of HPV types within the alpha-9 (HPV16, HPV31) and alpha-7 (HPV18, HPV45) species groups. These data support the observation that samples harboring fully integrated HPV DNA tend to have lower VL than samples containing episomal or mixed forms [6,7,10,15] in contrast to other studies [16,17,38].

Association between molecular parameters
The VL in samples exhibiting one or more methylated CpG sites in the 3'L1 fragment of HPV16 (
tion and methylation status and current cervical disease stage with regard to a single infecting HPV genotype, wherein the association with disease can be assumed with some confidence. There are potential shortcomings to this study. First, samples containing a single infecting genotype were identified using a generic PCR and genotyping test [33]. While this is the most common approach used and the test used arguably one of the more robust [39], such a determination is not without its problems including the potential for masking within mixed infections [40]. The selection of single infection samples is, we believe, a significant improvement on the use of unselected samples, wherein the potential impact of such masking is likely to be far higher. Nevertheless, the possibility that a minority of samples in this study contained low levels of one or more other HPV types cannot be ruled out. Second, although these samples constitute a highly selected panel and a minority sample type collected during routine screening, they do nevertheless permit the evaluation of the potential for these markers to be used, with a limited number of other confounding factors. Finally, although cytology samples are the primary sample type collected during cervical screening and are readily amenable to such testing, the use of cytology samples alone to improve the definition of cervical disease stage is problematic given the often discontinuous relationship between cytology and histology stage designations [1].
In summary, even under these optimized conditions, the sensitivities and specificities of many of these potential molecular markers were lower than the median sensitivity (98%) and specificity (86%) obtained in a recent evaluation of commercial HPV DNA tests for identifying CIN2+ cases during cytological screening [5]. There were a few individual measurements that have the potential for use in triage, characterized by a similar or higher specificity than required for initial screening. HPV VL appears to have an overall specificity of ca. 80% (range 76 -91%) to differentiate normal cytology from abnormal cytology. Conversely, HPV methylation appears to be able to differentiate HG cytology from normal and LG cytology with a specificity of 77 -100%. Both of these measures, therefore, have potential for use in screening and/or triage, but the utility of HPV genome methylation status may be improved if the disease-associated genotype-specific CpG site methylation patterns can be appropriately exploited.

Additional files
Additional file 1: Table S1. Primer and Probes.
Additional file 2: Figure S1. Impact of Varying Integration Thresholds. Top panels depict the number and percentage of samples exhibiting episomal, mixed or fully integrated genomes arsing from the use of a range of E2/E6 thresholds. Bottom panels depict the sensitivity and specificity plots for integration status being able to differentiate between normal and abnormal cytology (LG and HG) at a range of E2/E6 thresholds. p values refer to the proportion of mixed or fully integrated samples compared to episomal samples when using all samples regardless of HPV type (Fisher's exact test).