Analysis of CpG methylation sites and CGI among human papillomavirus DNA genomes

Human papillomavirus (HPV) constitute a group of over 100 different types. HPV infect stratified epithelia, both mucosal and cutaneous and associate with benign and malignant proliferative disorders. Viral types that preferentially infect mucosal epithelia are grouped into either a low-risk group (LR-HPV) not associated with cancer, or into a high-risk group (HR-HPV) whose members are found in almost all cases of cervical cancer [1]. HPV are non-lytic, non-enveloped, icosahedral-shaped viruses with a circular, double-stranded DNA genome of approximately 8.0 kb that is functionally divided into two coding regions (denoted E for early or L for late) and one regulatory region or LCR. The E region includes six major open-reading frames (ORFs) encoding functional proteins (E1, E2 and E4) and oncoproteins (E5, E6 and E7), and the L region encodes the two capsid proteins, L1 and L2. E4 is expressed in both early and late stages of the viral life cycle [2].

HPV gene expression is mainly regulated at the transcriptional and post-transcriptional levels and several studies have suggested that viral DNA methylation may be associated with viral gene expression and cancer progression [3–6].

Most of the work on HPV methylation has been carried out on the two main viral types involved in cervical cancer (types 16 and 18). In both genomes there is a progressive increase in methylation from asymptomatic carriers, through benign lesions and pre-malignant disease, to cancer tumors. Nevertheless, there is heterogeneity of CpG methylation in viral genomes derived from clinical specimens [4–7], and the exact role of viral DNA methylation remains unclear.

Furthermore, based on Epstein-Barr virus studies it has been proposed that viral genome methylation may enable a proportion of infected cells to survive cytotoxic T-cell immune surveillance [8, 9], and studies on adenovirus late viral genes suggest that they are more sensitive to methylation than early ones [10]. In the case of HPV L2 and L1 genes, they have been proposed to be preferentially recognized by the cellular methylation machinery [11, 12].

Methylation is the only known covalent modification of DNA in eukaryotes and plays an important regulatory role in vertebrates by silencing specific genes during development and cell differentiation. Cytosine methylation occurs at the 5'-position of the pyrimidine ring, mainly within a CpG context (m5CpG), although methylation of cytosines in different contexts has recently been described [13, 14] and 5-hydroxymethylcytosine (5 hmC), a novel DNA modification was reported [15].

Usually, the presence of m5CpG in genomic DNA is associated with chromatin condensation and inactivation of gene expression. Nevertheless, it is not clear whether the primary evolutionary role of DNA methylation is transcriptional silencing or a host defense system against endogenous or exogenous parasitic sequence elements [16].

In the genomes of higher eukaryotes, CpG dinucleotides are usually underrepresented, from one third down to 5% of their expected frequency [17–20]. The mechanism proposed to explain this underrepresentation, first recognized in prokaryotic systems [21], is that CpG sites are mutagenic due to the frequent conversion of methylcytosines to thymines through deamination [22].

Similar to their hosts, CpG dinucleotides are underrepresented in the majority of small DNA viruses, although to a lesser extent [23]. It has been proposed that low CpG frequencies may either allow viruses to avoid methylation by host methyltransferases and thus maximize their transcriptional efficiency, or could be a means of reducing CpG mediated immune responses [24].

The general view is that CpG sites in vertebrates, grouped within clusters or islands (CGI), are mostly unmethylated in promoter regions of transcriptionally active genes, and methylated in promoter regions of transcriptionally inactive genes. In contrast, methylation in body genes seems to be evolutionarily conserved and plays an important role in tissue- or cell-specific alternative promoter regulation. Furthermore, tumor cells usually show hypomethylation of the majority of the genome and methylation of CGIs in promoter regions of tumor suppressor genes [14, 25].

According to Gardiner-Garden and Frommer [26] a CGI is defined as a DNA fragment of at least 200 bp containing at least 50% of CpG's, with a CpG Observed/Expected (O/E) ratio of 0.60 or higher and at least a 100 bp gap in between different CGIs. This definition, despite its popularity, has raised many criticisms and various different proposals have been postulated [27–31]. Han et al., [32] recently compared three computational algorithms for CpG island identification and, based on vertebrate gene function and other genomic factors, the Takai and Jones criteria performance was the best. The analysis of 132 CGIs across the entire human chromosome 21 allowed researchers to identify several attributes associated with either methylation-resistance or methylation-sensitivity of CGIs [33].

The majority of the proposals for CGI identification and criteria to analyze methylation predisposition at CpG islands cannot be directly applied to HPV genomes because of their small size. In addition, lack of multiple promoter regions and the scarce data on methylation of different viral types make this scenario difficult.

Because in our view, understanding DNA methylation may benefit from our knowledge of CpG distribution along different viral genomes, we have analyzed genomic and regional proportions of CpG sites (PCS) and their arrangement into CGIs, in most human papillomavirus types sequenced to date. To our knowledge, there has been no comprehensive analysis of CGIs in virus genomes. We propose hypotheses about their functional roles, considering HPVs as both a single group or as subgroups in function of their preferential tropism (cutaneous and mucosal) and their associated risk to induce cancer (mucosal low- and high-risk).

To our knowledge, this is the first systematic analysis of CGIs, as well as of CpG site distribution performed in 92 HPV types. This analysis provides the main structure underlying the methylation phenomenon, guiding the identification of interesting genomic regions for future epigenetic experimental studies.

As anticipated [23], we found significant differences between CpG observed and expected values among all sequenced HPV genomes. Nevertheless, contrary to the notorious underrepresentation of CpG sites among eukaryotic genomes where they account for only one third down to 5% of the expected values; viral genomes have an underrepresentation of only around 50% of the expected CpG values. According to the co-evolution hypothesis, it would be expected that both human and HPV genomes exhibit similar CpG underrepresentation values. The great differences found could be related to other relevant phenomena, such as the species-specific codon usage and the genomic base composition. In fact a relationship between codon use and base composition with C+G content in HPVs was previously found [34, 35].

In this context, it seems necessary to mention that HPVs are small sized viral DNA genomes that do not evolve rapidly, in contrast to other viruses. It has been calculated that approximately 200,000 years are needed for 17 bp to change in these genomes [36].

In a biological context, a significant correlation between CpG observed and expected values could reflect global common functional constrictions shared by most HPV types. At the same time, the AsPCS inter-regional lack of correlations point to local functional constriction divergences, particularly in the E6 and LCR regions (Table 2). Nevertheless, more studies of different HPV types are necessary to confirm it.

Interestingly, L2 and L1 ORFs show a parallel CpG behavior. Both of them possess relatively low PCS arranged mostly in CGI. The L2 region is sixth place in average PCS value and fourth place in island number, and L1 is eighth place in average PCS value and sixth place in island number. It is known that in low grade cervical lesions associated to HPV 16, the viral genome is episomal, hypomethylated, and the L1 and L2 ORFs are expressed in order to generate viral capsids. In contrast, most high grade cervical lesions show viral DNA integrated, hypermethylated, and the L1 and L2 ORFs are not expressed. In vertebrates, methylation in body genes plays an important role in tissue- or cell-specific alternative promoter regulation, such that orphan CGIs are used for alternative promoter identification [37]. Even though this is a risky idea, it is tempting to speculate that viral capsid coding ORFs behave similarly and that when the cell hypermethylates them, it is possible that the L1/L2 region plays a role as an alternative promoter for viral genes and, when integrated, for some cellular genes.

Additionally, viral oncogenes also show a parallel CpG behavior, but in an opposite manner as the capsid coding genes. Both of them possess relatively high PCS values associated to few CGI. E7 occupies the second place according to average regional PCS values and the fifth place according to the average island number, while the E6 region occupies the fifth place according to regional PCS and the eighth place according to the average island number. This means CpG sites are widely distributed and suggest both regions could be sensitive to methylation, especially when the HPVs integrate in to the cell genome.

The E4 ORF seems to be a special case because of its multiple roles during the viral cycle. It is involved in diverse cell and viral functions, especially in diminishing keratinocyte integrity [38], cell cycle arrest at G2 [39], HPV DNA replication [40] and expression of late proteins [41], possibly through several proteolysis products from the E4 protein [42]. In fact, the E4 ORF is expressed throughout the viral cycle [2]. Its functions and continuous expression could impound hard functional restrictions to sequence mutations, as well as the need for a mechanism involved in avoiding methylation. The maintenance of CpG sites in CGIs could be a kind of compromise to meet sequence code restrictions with the necessary expression of the E4 protein.

In addition to the underrepresentation of CpG sites in HPV genomes, single nucleotide proportions point to T and A increased frequencies at the expense of G and C frequencies, reinforcing the C → T mutation hypothesis as a result of CpG deamination.

Nevertheless, the E4 region does not share the same behavior. Amazingly, even when A and G proportions are similar to those from the whole genome, C frequencies seem to be increased at the expense of Ts. This fact, added to the high number of CpG sites, mostly arranged into CGIs, leads to the question of whether there is any mechanism to avoid C → T mutations, resulting from deamination, that affects the HPV and specially the viral E4 region. In fact, there exist repair enzymes that remove deaminated bases; specifically, thymine DNA glycosylase (TDG) and methyl-CpG binding domain protein 4 (MBD4) which are able to remove either uracil or thymine from G·U and G·T miss-pairings where follow-on base excision repair enzymes restore a G·C pair [43, 44].

When grouped by tropism, the cutaneous HPV group, shows the greatest CGI average size (446 bp), followed by the LR-HPV group with an intermediate CGI average size (363 bp) and the HR-HPV group shows the smallest average CGI size (272 bp). Additionally, even though CGIn from both cutaneous and mucosal HPV types share similar values (3 CGIs, average), it seems there is a trend toward smaller CGIn in HR- compared to LR- HPVs (1.6 and 3.93, respectively). These data raise questions about the possible role of CGIs on the interaction between virus and cells. At the same time, this suggests a possible differential susceptibility to methylation in function of either HPV tropism or associated risk. Nevertheless, the only experimental studies on methylomes have been carried out on HPV types 16 and 18, where methylation changes and high heterogeneity were observed along the viral infection phases [3–8]. In summary, those inter-group differences could be related to different viral infection epigenetic strategies.

Finally, it should be interesting to examine virus targeted genes in the human genome and see what their CGI and CpG features are. We hope that our analysis of CpG and CGI distribution among HPV genomes can provide elements for rational experimental approaches related to biological and evolutionary processes of these viruses and their hosts.

First, we want to point out that this work is based solely on sequence feature analysis, so the results reported here may need experimental corroboration. In this study, we found that there are both significant differences and correlations between CpG observed and expected values among all sequenced HPV genomes, and that viral genomes have an underrepresentation of only around 50% compared to eukaryotic genomes where underrepresentation is from 30 down to 5%.

The regional analysis of CpG site distribution shows that correlations between complete genome PCS are significant, even though there is a lack of correlation between several genomic region pairs, especially those involving LCR and E6. Furthermore, we found that L2 and L1 ORFs exhibit an opposite behavior to oncogenes E6 and E7; the first pair possessing relatively low numbers of CpG sites which are associated to CGI while the oncogenes possess a relatively high number of CpG sites which are not associated to CGI. We would point out that E4 is the region possessing the highest content of CpGs in every HPV type where an E4 is identified.

From the CGI analysis, the E2/E4 is the only region with at least one CGI in every HPV, and the main difference between this region and the complete genome is that in the E4 region the frequency of Cs seems increased at the expense of Ts. The mucosal HR-HPVs possess the shortest CGI sizes, followed by the mucosal LR-HPVs and finally by the cutaneous viral subgroup. At the same time, the mucosal HR-HPVs show a trend toward lower CGI numbers, followed by the cutaneous viral subgroup and finally by the mucosal LR-HPVs.

Finally, we hope that our analysis of CpG and CGI distribution among HPV genomes can provide support for rational experimental approaches related to the biological and evolutionary processes of these viruses and their hosts.

Our CpG analysis comprises two phases: the first one is the comparison of CpG site distribution among HPV types as a single group, and in the second phase we identify possible similarities or differences in CpG site distribution associated to the preferential tropism and/or the risk of cancer caused by the mucosal viral subgroups. The CpG distribution refers to CpG proportions (PCS) and either CGI sizes or numbers.

Output Data

In order to identify the CpG sites we used PISMA, a computational tool we designed, which is available upon request to SCG and/or RAS. PISMA allows the location and identification of motifs ranging between 2 to 10 bases, in up to 10 kb DNA sequences. In addition, this tool will count the number of motifs in defined regions. We have identified the individual CpG sites per region from all HPV sequences, and based on those data from PISMA, we calculated the PCS as the number of CpG sites divided by total number of nucleotides in the sequence. Identification of putative CpG islands/clusters was carried out using CpG island Explorer 2.0, publically available in http://www.uscnorris.com/cpgislands2/cpg.aspx[28]; parameters used in CpG island identification were those from Gardiner-Garden and Frommer [26]: CpG 50% or more, ObsCpG/ExpCpG ratio 0.60 or higher, at least 200 bp size and 100 bp gap. In Figure 3 results obtained for HPV 16 are shown.

PCS Arc-sin transformations were calculated in order to properly use parametric statistical procedures that are not suitable for proportion data because of the binomial nature of the biological process (CpG occurrence). In summary, the distribution of proportions is often skewed, and the arcsine transformation often makes that distribution more normal.

Finally, we calculated the CpG expected values for each viral genome in order to compare them with the CpG observed values.

SCG is an associate researcher and RAS is a professor at the National University of Mexico; AGC is a researcher at the National University of Mexico and at the National Cancer Institute; VMG is a professor at the Metropolitan University; RM is an associate researcher at the National Cancer Institute; MMS is post doctoral fellow at the AGC lab; GTDC and MAH are students guided by RAS.