Infection with high-risk human papillomaviruses (HR-HPVs) is a necessary but insufficient cause of cervical carcinoma, with additional viral and host genetic events required to drive cells to the malignant phenotype 1, 2. Most cervical cancers are squamous cell carcinomas (SCCs), which develop in a multi-step fashion from precursor lesions classified as low-grade or high-grade squamous intraepithelial lesions (L-SILs and H-SILs, respectively) in the two-tier Bethesda system 3. L-SILs generally represent productive HPV infections with a relatively low risk of progression to invasive disease, whereas H-SILs can be viewed as abortive infections in which the late events in the HPV life cycle are not supported. It is currently not possible to distinguish between rare SILs destined to persist or progress and the vast majority that will regress 4, 5. A more objective approach to doing so would circumvent the unnecessary clinical follow-up of women with non-progressive lesions. In order to achieve this, we require a greater understanding of the viral and host cellular changes that induce progression of SILs, as well as their temporal associations. Persistent infection with HR-HPV types, such as HPV16, 18, 45, 31, and 33, associated with ineffective viral clearance is a major risk factor for cervical cancer development 6, 7. Currently, an area of very active research interest is HR-HPV integration into the host genome.
The significance of HR-HPV integration in cervical carcinogenesis
In the normal viral life cycle, HPV genomes exist in a circular (or ‘episomal’) state and are thought to be retained in basal cells of the squamous epithelium at approximately 50–100 copies per cell 8, 9. Episomal expression of the viral oncogenes E6 and E7 is tightly regulated, with high-level expression only in suprabasal post-mitotic cells 10, 11. Here, the viral oncogenes induce unscheduled re-entry into S-phase of the cell cycle, with activation of the host replication machinery needed for amplification of viral genomes prior to virion synthesis 12. In these ‘productive’ infections, expression of the viral oncogenes fails to pose a carcinogenic threat, since it occurs within a compartment containing cells destined to be lost from the cervical squamous epithelium (which undergoes constant renewal). Cervical neoplastic progression requires both spatial and quantitative deregulation of this tight transcriptional control, such that viral oncogene expression occurs at a high level throughout the epithelium 10, 11, 13. Oncogene expression in basal epithelial cells - which are retained for long periods within the cervical epithelium and thus represent the target cells of carcinogenic events—inhibits differentiation and induces high-level chromosomal instability that can drive progression towards the malignant phenotype 14–19.
In the vast majority of cervical carcinomas, deregulated viral oncogene expression (and host genomic mutation) is seen in cells in which truncated viral genomes are integrated into host DNA. Such cells have a proliferative advantage in vitro, explaining their outgrowth in a mixed population 20. Integration is not a normal part of the HR-HPV life cycle and is characterized by deletion of viral genes that are essential for synthesis of an infectious virus. Thus, it represents a by-product of viral infection that may confer a selective advantage to the host cell without any apparent advantage to the virus. Although variable portions of the HR-HPV genome are deleted in viral integrants, consistent features are seen (see Figure 1); particularly loss of the viral E2 gene, which can inhibit transcription from the integrated viral promoter 21–24. In general, integration leads to increased expression and stability of transcripts encoding the E6 and E7 proteins, which bind and disrupt the function of a number of key cellular proteins such as p53 and pRb 1, 15. Such effects are restricted to high-risk HPV types, providing a biological explanation for the difference in cancer risk associated with high-risk (HR) and low-risk (LR) HPV types. Furthermore, given that hTERT expression is inhibited by E2 25 and activated by E6 26, HR-HPV integration is also an efficient way to activate telomerase and, through cooperative effects with E7, to immortalize epithelial cells 27, 28. Thus, HR-HPV integration is a ‘one-shot’ mechanism for cellular immortalization, deregulated proliferation, and increased genomic instability—cellular hallmarks that contribute to development of the malignant phenotype 29.
Methods of detecting HR-HPV integration in cervical neoplasia
Clinical studies have aimed to determine the prevalence of HR-HPV integration in cervical carcinoma and whether integration occurs at a particular stage during SIL progression. Most studies have analysed integration of only HPV16 and HPV18, the virus types most commonly detected in cervical malignancy 30. The frequencies at which viral integrants are detected in carcinomas is consistently high, being seen in 100% of HPV18-positive cases 31–33 and up to 80% of HPV16-positive cases 32–34. However, there is a much higher level of discrepancy in the frequency of viral integration reported in SIL 35, 36, which is variably claimed to be an early event associated with progression of L-SIL to H-SIL 37–46 or a later event associated with progression of H-SIL to SCC 32, 34, 47–52. While these contrasting views are likely to be partly due to technical factors (eg stringency of real-time PCR analysis criteria 50, 52 or method of signal detection in in situ hybridization (ISH) 35), a more important issue concerns fundamental differences in the biological significance of the integrants detected by different techniques. Available methods can be divided into two broad categories: (i) those that detect virus–host fusion transcripts, ie transcriptionally active (and therefore ‘selected’) viral integrants (eg amplification of papillomavirus oncogene transcripts 49 and RNA ISH 53); and (ii) those that detect integrated viral DNA regardless of its transcriptional status (eg Southern blotting 32, 54, quantitative real-time PCR 41, 50, restriction-site PCR 55, and DNA ISH 35, 51, 56, 57). The consistency of integration rates detected in cervical carcinoma, whichever detection method is adopted, is not surprising, since it would be expected that in all malignancies containing integrated HR-HPV the integrant is transcriptionally active and exerting a significant selective pressure. However, this may not be the case in SIL. While 15% of H-SILs and 0% of L-SILs contain transcriptionally active, fully selected viral integrants 34, 49, the frequency of integrated HR-HPV DNA in SILs of both grades is higher 35, 36 and H-SILs containing integrated HR-HPV DNA in the absence of virus–host transcripts have been described 58. In the latter study, the integrants had a disrupted E2 open reading frame (ORF) and were therefore likely to be the ‘selectable’ type, remaining in a silent ‘latent’ state until additional events provide an environment permissive for their transcriptional activity and full selection. Taken together, the available data lead us to conclude that the process of HR-HPV integration and subsequent selection may be two independent, and equally important, events during cervical neoplastic progression. Further clinical studies that address both parameters simultaneously are required to provide a more accurate picture of their temporal association in vivo.
Mechanisms underlying HR-HPV integration and its selection
Requirement for DNA double-strand breaks
The mechanism of HR-HPV integration is not understood. Unlike retroviruses, HPVs do not encode an integrase, and integration plays no role in the normal life cycle. Integration must therefore represent a chance occurrence, presumably rendered more probable by any event that increases the frequency of double-strand breaks (DSBs) in host and viral DNA. Investigation of factors associated with integration events is confounded by the fact that some integrants have a growth advantage over others in a mixed population of cells. A number of studies have overcome this problem by studying the integration frequency of foreign DNA constructs containing antibiotic resistance markers 59–61, enabling selection of any integrant regardless of whether it confers a growth advantage. These studies have indicated that the frequency of integration is increased by the induction of DSBs, either by HR-HPV E6 and E7 or by expression of common fragile sites (CFSs).
The site of HR-HPV integration and potential role of host genes
The sites of integration in host chromosomes have been determined by ISH of metaphase preparations using probes for HR-HPVs 62, or through sequence analysis of PCR-amplified virus–host junction sequences 55, 63 or virus–host fusion transcripts (49). Integration sites are distributed throughout the host genome, with no single region predominating 64. No specific cellular sequence motif has been observed, and integration at the same genomic region in different tumours occurs at different nucleotide sequences. That said, there is a clear predilection for integration into host CFSs 55, 65. In a review of HR-HPV integration sites from multiple studies (some detecting viral DNA; others virus–host fusion transcripts), 38% of 192 integrants were in regions encompassing known CFSs 64. The true percentage is likely to be higher, since a number of studies did not provide sufficient sequence information to define the precise site of integration. An important question is whether CFSs harbour integrants simply because they are more prone to breakage or because they contain host sequences that increase the probability of providing a selective advantage when adjacent to HR-HPV. The latter scenario could be due to a host effect on viral transcription and/or insertional mutagenesis caused by the virus. The demonstration that foreign DNA vectors (selected through multi-drug resistance) integrate at known CFSs at a similar frequency (44%) to HR-HPV integrants in cervical cancers in vivo60 suggests that CFSs offer relative accessibility to HR-HPV DNA, with subsequent selective growth advantage determined by the level at which viral oncogenes are transcribed. Similar observations were made using constructs containing antibiotic resistance genes and the HPV18 URR/E6/E7 region 59. However, both studies were performed in cells in which CFS breaks were chemically induced or occurred at a high level spontaneously. Whether a lower frequency of integration at CFSs would be observed in cells with a ‘normal’ frequency of CFS breaks is unknown. If so, the resultant high proportion of cervical cancers with integrants at CFSs would suggest that sequences at these sites do indeed confer some form of selective advantage. On the other hand, CFS breaks may occur at increased frequency in HR-HPV-infected cells expressing E6 and E7, thus increasing the probability of viral integration at such sites. These issues, as well as a more accurate characterization of CFSs themselves, need to be resolved before the biological significance of CFSs in HR-HPV integration can be determined.
A number of studies have suggested that insertional mutagenesis may have a role in at least some cervical cancers. Cases have been reported in which HR-HPV integration has occurred within or adjacent to known oncogenes, most commonly within intronic sequences 64–67. The most frequently observed integration site, particularly in cervical cancers positive for HPV18, is in the region of the MYC gene at chromosomal band 8q24 66–69. Recurrent integration (albeit at low frequency) has also been observed at the TERT70 and FANCC66, 67 loci, at chromosomal bands 5p15 and 9q22, respectively. However, very few studies have determined whether integration in these regions has an effect on expression of the candidate host oncogenes and consequently the phenotype of the cell. One recent study showed that HPV18 integration at 8q24 in cervical carcinoma cell lines is associated with increased expression levels of MYC mRNA and protein relative to cells containing integrants at other loci 68. However, as MYC levels in normal cervical keratinocytes cultured in the same conditions appear similar, or higher, to those in the cells with the 8q24 integrant 71, the significance of these results is uncertain. Functional studies in suitable cell lines will be of particular value in determining the significance, if any, of insertional mutagenesis in cervical carcinogenesis.
Transcriptional regulation of integrated HR-HPV by host factors
The implicated existence of selectable but transcriptionally inactive HR-HPV integrants in SIL indicates that important subsequent events are necessary for deregulated transcription from the integrated viral URR. Overcoming host restraints to viral transcription therefore represents an important step during selection of cells containing viral integrants (Figure 2).
Early studies demonstrated that E6 and E7 expression in HR-HPV-infected L-SIL is localized primarily to the upper spinous and granular layers of the epithelium, with little/no detectable expression in proliferative basal layers 10, 11. The possibility of host repression of viral transcription in basal epithelial cells was supported by in vitro work showing that HPV early gene transcription from the native URR is activated in suprabasal layers of keratinocyte raft cultures 12. In keratinocytes retrovirally transduced with the URR of either HPV11 or HPV18, URR activity is detected in monolayer culture but is absent in basal layers of raft cultures 72–74. This suggests that repression of viral oncogene expression occurs in basal cells, at least in part due to trans-acting repressors of the HPV URR, and demonstrates that cellular context is critical when assessing URR transcriptional activity. The identity and precise regulation of these host repressors have not formally been demonstrated for HPV18, although mutational analysis of the HPV11 URR has indicated that the CCAAT/enhancer binding protein (C/EBP) binding sites are critical for basal repression 74. C/EBP binding sites have also been detected in the URR of HPV16 75. Although the URRs of all HPVs are generally very similar, there are also differences, which are assumed to make an important contribution to tissue tropism. Thus, the precise requirements to overcome basal transcriptional repression may be HPV type-specific. Other transcriptional inhibitors, such as YY1 76 and CCAAT displacement protein (CDP/Cut) 77, for which binding sites have been found in the URR of many HR-HPVs, may also play a role in basal repression, as may chromatin remodelling by histone deacetylases 78. Interestingly, CDP/Cut is expressed in proliferating, but not differentiated, keratinocytes 79 and associates with the histone deacetylase HDAC1 80.
Transcriptional inhibition of integrated HR-HPV has also been demonstrated in vivo. When HR-HPV-immortalized (but non-tumourigenic) keratinocytes that expressed viral early genes in monolayer culture were xenografted onto immunocompromised mice, inhibition of early gene expression was observed throughout the epithelium 81–83. In contrast, tumou- rigenic segregants retained the ability to express viral oncogenes following grafting (including basal layers), suggesting that mechanisms of viral transcriptional silencing were no longer functional in these cells. These observations gave rise to the cellular interfering factor concept 1 and subsequent experiments suggested a role for macrophage-derived cytokines 84. Indeed, viral gene expression in HR-HPV-immortalized keratinocytes in vitro can be inhibited by tumour necrosis factor alpha (TNFα), transforming growth factor beta, and interleukin 1, and resistance to such repression is closely associated with tumourigenicity in vivo84–88. In the case of TNFα, inhibition of viral gene expression is induced by alterations of the AP-1 transcription factor complex 89, which, when in the correct conformation, can activate viral gene expression 90. The AP-1 complex is unaffected by TNFα in tumourigenic cells and somatic cell hybrid studies have indicated that resistance to exogenous factors such as TNFα is conferred by the presence of mutation in the host genome 83, 84, 89.
Deregulated oncogene expression from HR-HPV integrants therefore appears to require loss of basal repression of viral gene transcription and insensitivity to inhibitory cytokines. Whether such mechanisms also apply to episomal HR-HPV largely remains to be determined. Transient transfection experiments have shown that inhibition of the HR-HPV URR in plasmid form in monolayer cultures can be induced by treatment with inflammatory cytokines 85. The low level of viral oncogene expression in basal layers of productively infected L-SILs would also be consistent with inhibitory mechanisms affecting episomes 10, 11. However, differences in the transcriptional regulation of episomal and integrated HR-HPV sequences have been described and attributed either to cis-acting host sequences that modulate HR-HPV integrants but not episomes 91, differences in chromatin structure at the URR 92, or to differences in function of sequence motifs in the URR, such as the matrix attachment region 93.
Regulation of integrated HR-HPV by HR-HPV episomes
An additional important and largely overlooked consideration is that HR-HPV integration must occur in a cell containing a background of episomal virus. This is important, as the transcriptional regulatory effects of E2 appear to depend on the physical state of HR-HPV. Whereas overexpression of the HPV16 E2 protein has no apparent effect on transcription from episomal HPV16, it strongly inhibits expression from integrated HPV16 92. Moreover, in cell lines generated from cervical carcinomas, the HR-HPV integrants that were fully selected in vivo remain responsive to the transcriptional inactivation effects of E2 23, 24. We therefore hypothesized that within a cell containing a mixture of episomal and integrated HR-HPV, E2 expressed from episomes at physiological levels could inhibit expression from the co-existent integrant in trans and that overcoming this inhibition would represent another important event required for the selection of integrated HR-HPV. Given that available techniques cannot accurately detect intracellular mixtures of episomes and integrants together with their transcriptional status, it is not possible to identify this state in vivo. Nevertheless, SILs containing non-expressing integrants and transcriptionally active episomes do appear to exist 58, although information regarding the spatial relationship in vivo is not yet available.
Using the unique W12 in vitro model of cervical neoplastic progression, we observed that selection of cells containing integrated HPV16 from which viral oncogenes were expressed at high level was consistently preceded by spontaneous rapid loss of transcriptionally active episomes that expressed E2 94. Moreover, rapid episomal loss was closely linked with activation of antiviral response genes inducible by type I interferon (itself able to induce rapid episome clearance in W12 95), for which we are currently investigating the initial trigger (Figure 3). W12 has also recently indicated that selectable HPV16 integrants can exist in a latent state in an apparently episome-only population and that selection does not occur until loss of inhibitory episomes is initiated 95. It will be of great interest to determine whether episome loss is also required for selection of integrated HR-HPV in other in vitro models of early cervical neoplastic progression 9, 96, 97. Given the rapidity of spontaneous episome loss in W12 in vitro, it may prove difficult to demonstrate this process in vivo. Nevertheless, the concept is supported by the observations that most cervical carcinomas containing integrated HR-HPV have little or no episomal DNA 32, 37, 47–49 and that in the cases that do contain both integrated and episomal HR-HPV DNA, ISH shows clonal expansion of regions containing only integrated virus adjacent to regions containing only episomes 51. These new insights indicate that the assumption that cervical neoplastic progression occurs through integrant-only cells outgrowing episome-only cells is likely to be oversimplified. Indeed, without subsequent episome loss, HR-HPV integration is unlikely to confer a selective advantage.
Levels of transcription from integrants
As cervical cancers usually contain one or a few integration sites, it would be predicted that high-level DNA amplification of these regions would further increase the level of E6 and E7 expression and confer a stronger growth advantage. Amplification of integrated HR-HPV DNA has indeed been observed 98, with a number of cell lines containing hundreds of copies of viral genomes 92, 99. However, a recent study in which viral transcripts were detected by highly sensitive ISH protocols demonstrated that within cells containing multiple integrated copies of HR-HPV, there is selection of cells containing only one or very few transcriptional centres, as a result of methylation of other URRs 53, 100. Where head-to-tail concatamers of integrated HR-HPV are present, these events may be selected due to silencing of full-length E2 transcripts such that viral expression is restricted to the virus–host junction where E2 is disrupted 53. However, the process may also reflect host restriction of E6 and E7 expression, since stable transfection of a construct expressing HPV18 E6 and E7 into cells already containing a transcriptionally active HPV18 integrant leads to silencing of one of the transcription centres 53. This implies that the amount of viral oncogene expression required to confer a selective advantage may vary during neoplastic progression. High-level expression may be important in the early stages, in order to establish an environment in which additional host genomic mutations can be acquired at a greater rate. However, once the genome has acquired a certain number of changes and cells have a selective advantage, continued high levels of E6 and E7 (and consequent genomic instability) may be deleterious, so that a further selective advantage is gained by cells in which the number of transcriptional centres is reduced.
Temporal association between events leading to progression of SIL
It is important to establish the position of HR-HPV integration in the timeline of neoplastic progression and its functional contribution to carcinogenesis. In order to address these issues accurately, separate consideration needs to be given to the biologically distinct states of transcriptionally inactive ‘latent’ integrants and fully selected integrants.
ISH analysis has shown that a key difference between L-SIL and H-SIL is the spatial pattern of viral oncogene expression—whereas basal oncogene expression is repressed in L-SIL, expression is consistently observed throughout the full thickness of H-SILs 10, 11, 13. As only a small proportion of H-SILs contain transcriptionally active viral integrants 34, 49, spatial deregulation of episomal oncogene expression most likely precedes full selection of viral integrants in H-SILs. Moreover, such deregulation would be likely to increase the probability of new integration events 61. This scenario does not exclude the possibility that viral integration occurs prior to spatial deregulation of oncogene expression, as may occur, for example, if an individual is exposed to risk factors that increase the probability of DNA DSBs. The resulting integrants would be predicted to remain latent until subsequent events eliminate transcriptional inhibition. We believe that loss of inhibitory E2-expressing episomes is likely to be a key feature of full integrant selection, which occurs subsequent to and independent of spatial deregulation of episomal oncogene expression. This implies that transcriptionally active viral integrants contribute to cervical carcinogenesis through quantitative deregulation of E6 and E7 in progressive H-SIL, with consequent high-level genomic instability and acquisition of mutations likely to drive malignant progression 101.
Inclusion of episome loss in models of cervical carcinogenesis allows re-evaluation of the significance of HR-HPV integration. If episome clearance is induced in a population of cells that has undergone spatial deregulation of episomal viral oncogene expression, E6 and E7 expression would not be sustained and loss of the anti-apoptotic effects of the viral oncoproteins would induce widespread apoptosis of episome-only cells 95. In this scenario, the only cells capable of sustaining E6 and E7 expression would be those containing latent integrants, which would be de-repressed as a consequence of episome loss. Thus, the primary reason for integrant selection may be the ability of integrants to sustain E6 and E7 expression and prevent apoptosis. Further clonal expansion would be due to selection of integrants that express the highest levels of E6 and E7. In models of carcinogenesis where episome clearance is induced (likely to be the majority of cases), HR-HPV integration would therefore be a critical and prerequisite event for neoplastic progression, since absence of latent integrants would result in lesion regression.
Although we believe that the most common sequ- ence of events in cervical carcinogenesis is the integrant-driven route shown in Figure 4, there are likely to be exceptions. As cis-acting sequences can regulate the effects of trans-acting molecules on viral integrants 91, host sequences could affect the action of trans-acting host repressors or viral E2. In the latter case, integration events would be insensitive to E2-mediated transcriptional inhibition and would not require episome loss for selection. Indeed, rare cervical carcinoma cell lines do contain HR-HPV integrants that are relatively insensitive to the effects of E2 91. These phenomena may explain at least some cases of rapid lesion progression observed clinically 41.
In addition, an alternative route to cervical carcinogenesis is demonstrated by the fact that approximately 12.5% of cervical carcinomas contain viral transcripts from only episomal HR-HPV 49, with a higher proportion in HPV16-positive malignancies 34. In these cases, HR-HPV episomes are likely to have undergone quantitative deregulation of viral oncogene expression via alternative routes, which also confer resistance to innate host mechanisms of episome clearance; for example, mutations of sites in the viral URR that bind host transcriptional repressors such as YY1 102. Given that E2 is a telomerase repressor 25 and is presumably still expressed in episome-only carcinomas, the route to immortalization in such cases may be different. The relative extent of E2 deregulation is of particular interest, since high-level E2 expression can induce apoptosis in a p53-independent manner 103. Furthermore, there may be an important role for the HR-HPV E5 oncogene 1, the ORF for which is downstream of E2 and usually deleted in viral integrants selected in vivo. Further study of episome-driven carcinogenesis, using clinical material and suitable in vitro models, is therefore clearly warranted.
Clinical implications and future directions
An important question is whether this emerging view of HR-HPV integration can be exploited to improve the clinical management of cervical neoplasia. Given that clearance of viral episomes can be induced through activation of an antiviral response 95, 104, it is unsurprising that there is interest in treating cervical SIL with exogenous type I interferon 105, 106, or with imidazoquinolone compounds that induce an antiviral response through activation of toll-like receptor 7 107. In a number of small clinical trials, treatment of HR-HPV-infected cervical SIL with type I interferons (IFNs) produced mixed results, with some cases of progression 105, 108. Recent insights can provide an explanation for this observation 95, 104. If a SIL contains rare cells harbouring latent ‘selectable’ HR-HPV integrants (as well as inhibitory episomes), episome loss driven by IFN treatment could accelerate clinical progression, rather than regression. Thus, we believe that treatment of HR-HPV-associated cervical lesions by activating an antiviral response must be undertaken with caution and careful follow-up.
An indication of lesions at increased risk of progression would be provided by a test able to detect integrated HR-HPV, ideally using a method that distinguishes between latent and fully selected transcriptionally active integrants to indicate level of risk. The most practical approach may be to find surrogate markers, which would also circumvent difficulties caused by the existence of multiple HR-HPV types, each of which would require a unique probe/primer sequence. While it is currently difficult to predict potential indicators of latent integrants, the presence of selected integrants may be identified using markers of the cellular effects of E6 and E7 109, 110. Although such markers are unlikely to distinguish between deregulated integrants and deregulated episomes, this would not be disadvantageous as deregulation of HR-HPV oncogene expression (by any mechanism) would predict lesions with a high probability of progression.
While HR-HPV integration is a key event in the development of most cervical cancers, it is not just the process of integration, but also subsequent events leading to selection of integrants that are required for progression of cervical SIL. In many cases, one of these events is likely to be loss of inhibitory E2-expressing episomes. It is ironic that activation of innate immune mechanisms that would usually protect against viral infection may have an important role in causing disease progression. Methods used to detect HR-HPV integration in clinical samples must be chosen carefully if the biological significance of a positive test result is to be interpreted correctly. Further refinement of such techniques and detection of surrogate markers of viral integration have the potential to improve objective management of cervical SIL and, feasibly, HR-HPV-associated neoplasia at other anatomical sites. The recent insights described in this review may also be of benefit to our understanding of the biology of human neoplasms associated with integration of other DNA tumour viruses.
We thank the Medical Research Council and Cancer Research UK for funding our research, some of which is described in this review.