Human Papillomavirus Biology and Cervical Neoplasia: Implications for Diagnostic Criteria and Testing Open Access

Cervical cancer is the best example of a common human malignancy with a proven infectious etiology.1 The data linking human papillomavirus (HPV) infection with the epidemiology and pathogenesis of cervical neoplasia is overwhelmingly clear and convincing. While there are probably no absolutes in medicine, the fact that HPVs are etiologically involved in causing cervical cancer of virtually all histologic types comes very close. It should be highly gratifying to pathologists that much of this knowledge is derived from the careful correlation of molecular studies applied to archival pathology specimens.

All papillomaviruses have a nonenveloped icosahedral capsid that is 55 nm in diameter and that contains a double-stranded circular DNA genome of about 7900 nucleotide pairs.2,3 Human papillomaviruses are strictly species specific and epitheliotropic. They cause localized hyperproliferations manifested most commonly as skin warts or as mucosal condylomata. More than 100 types of HPVs have been defined on the basis of DNA sequence heterology.4,5 Despite this heterogeneity of viral type, comparison of DNA sequences has revealed a remarkable conservation in HPV genomic organization. All HPV genomes are collinear and have analogously situated open reading frames (ORFs), all encoded on 1 DNA strand. Comparative studies have allowed the assignment of ORFs into 2 regions encoding, nominally, early (E) and late (L) functions and a third upstream (of the E region) regulatory region (URR). Studies using deletion or site-directed mutagenesis have allowed the assignment of specific functions to the various viral ORFS. Briefly, proteins encoded by ORFs E6, E7, and E5 are responsible for transformation; E1 specifies 2 or more proteins required for regulated episomal DNA replication; E2 encodes 2 or more proteins that positively or negatively regulate viral transcription; and L1 and L2 encode the major and minor capsid proteins, respectively. The function of the E4 protein is yet to be determined, but it is the predominant product of both early and late stages of productive viral infection and may modulate structural changes in the cytoplasm of infected cells.

Squamous cell carcinomas of the cervix and their precursors have the epidemiologic characteristics of a sexually transmitted disease.6,7 In epidemiologic terms, almost 90% of the risk for cervical neoplasia is attributable to HPV. When HPV is accounted for, most other traditional risk factors become statistically insignificant. Infection with HPV precedes the development of pathology. Persistent HPV infection is the major risk factor for progression to neoplasia.8,9 Koilocytotic atypia, the morphologic hallmark of HPV cytopathic effect, is the earliest cytologic manifestation of cervical intraepithelial neoplasia.10,11 This low-grade squamous intraepithelial lesion is the most common definite abnormality in cytologically screened populations, being present in approximately 5% of Papanicolaou tests. High-grade intraepithelial lesions are invariably HPV positive with a more restricted type spectrum reflective of the types found in invasive carcinoma. Essentially, all invasive cervical cancers are HPV-associated.12,13 Human papillomavirus types 16, 18, 31, and 45 alone are present in almost 80% of cervical cancers. In a definitive multicenter international study in which both the pathology and methods of HPV detection and typing were highly controlled, more than 94% of approximately 1000 tumors from all over the world were HPV positive.12 Recent follow-up reports have proven that the few negative tumors in this study were actually false negatives, and that repeat or new samples from those negative tumors were HPV positive.13 Hence, molecular methods applied to paraffin-embedded material can demonstrate HPV genetic material in nearly 100% of premalignant and malignant lesions of the uterine cervix.

Approximately 30 HPV types have been found in the anogenital tract, with two thirds of the HPV-associated anogenital neoplasms involving HPV types 6, 11, 16, and 18.14 Types 6 and 11 cause predominantly benign exophytic genital warts (condylomata acuminata) and are only rarely associated with high-grade squamous intraepithelial lesions (HSILs) or invasive squamous cancers. Human papillomavirus 16 is the most prevalent virus to infect the uterine cervix, and it is closely associated with the entire range of intraepithelial and invasive squamous neoplasia as well as less frequently with epithelial neoplasms of nonsquamous type.15 While accounting for fewer cervical infections, HPV-18 is relatively more frequently associated with adenocarcinomas and small cell neuroendocrine cancers of the cervix and somewhat less frequently with invasive squamous cancer.16 The absolute prevalence of some of the other mucosotropic types (eg, HPVs 31, 33, 35, 39, 42, 43, 44, 45, 51, 52, 56, and others) may be underestimated because they have not been generally available for large-scale screening or because the detection methods assign group types rather than individual types. However, in contrast to earlier concepts, it is clear from recent studies that so-called high-risk types account for almost 90% of cervical infections.17 

As there seems to be a correlation between specific HPV types and the potential for clinical malignancy, there also may be an association between the physical state of HPV DNA within the cell and the malignant potential of the associated epithelial proliferation.18,19 In benign HPV-infected lesions, the viral DNAs exist as extrachromosomal plasmids, mostly as monomeric circular molecules. However, in some cancers, HPV DNAs are found as multimeric circular molecules, sometimes with deletions, or frequently integrated into host chromosomes. Viral integration most often disrupts the E2 ORF encoding the transcription regulatory proteins. Loss of these regulatory proteins is thought to be the basis for dysregulation and overexpression of the transforming E6 and E7 ORFs.18 

A molecular model for HPV-induced carcinogenesis has emerged involving the interaction of HPV gene products with what is recognized to be a tightly controlled network of cellular oncogenes and antioncogenes involved in the control of cell proliferation.18,20 Histogenetically, papillomaviruses must infect the reserve, basal, or stem cell population of the cervical transformation zone, cells with the potential to differentiate along squamous, glandular, or neuroendocrine lines that are responsible for epithelial maintenance. In cells committed to squamous differentiation, there is an orderly program of maturation throughout the epithelial thickness both at the morphologic and the molecular level. In normal squamous epithelia, the only cells capable of cell division are the basal or parabasal cells. In morphologically normal, but HPV-infected basal cells, papillomavirus gene expression is inhibited to essentially maintenance levels. Productive HPV gene expression is tightly regulated and permitted only in cells that have begun squamous maturation, with concurrent loss of proliferative capacity. In the immediate suprabasal zone, there is expression of the early regions of the virus, and as the cells differentiate, there is an induction of all viral genes as well as viral DNA synthesis, leading to assembly and production of virions in the cells just beneath the surface. In the cervix, one recognizes such lesions as low-grade squamous intraepithelial lesions, mild dysplasia, or CIN 1, most of which at some point demonstrate koilocytotic atypia. Such lesions usually regress in less than a year, but rarely persist for extended periods. An explanation for some of the diagnostic criteria used by pathologists is implicit in this program of differentiation-linked expression. The nuclear enlargement and hyperchromasia recognized as atypia are direct results of E6/E7-mediated activation of host DNA synthesis. In a low-grade lesion, this is regulated to occur in cells that can no longer divide (ie, the intermediate squamous cells) and is primarily directed at the production of viral DNA. Given the small size of the viral genome, the several thousand copies of the virus present in a productively infected cell clearly cannot account for the 2- to 4-fold nuclear enlargement that is observed. It is a diagnostically fortunate coincidence that ineffective (in the sense of cell division) E6/E7-mediated host DNA synthesis produces the enlarged nuclei and increased nuclear-cytoplasmic ratio that one recognizes as abnormal. If the process is not fully developed or is perhaps regressing, then the cells derived from the surface often have less nuclear abnormality (atypical squamous cells of uncertain significance) than those seen in classic dysplasia. In the fully developed case, they are classified as being derived from a mild dysplasia/low-grade squamous intraepithelial lesion (LSIL). If the cells also have the correct amount and form of the cytokeratin binding protein HPV E4 expressed, then they appear as koilocytes. Koilocytotic atypia, while very often present, does not have to be seen to recognize a low-grade lesion. Every cytologist recognizes cells derived from the upper levels of a mild dysplasia that meet the diagnostic criteria for dysplasia yet do not have the characteristic perinuclear halo termed koilocytosis. Such lesions are just as HPV-associated as those that do have koilocytes, and the differences undoubtedly represent temporal variation within the life cycle of a low-grade lesion.

If viral gene expression is so tightly regulated, how do high-grade lesions develop? The sine qua non of high-grade dysplasia is morphologic evidence of basal-like cell proliferation. In these cells, the coordinate link between differentiation and viral early gene expression is lost. How this occurs is unclear, although it certainly must be a rare event(s) given the relative frequency of low- versus high-grade lesions. Potential mechanisms might include viral integration or mutations in HPV E2, such that E2-mediated regulation of E6/E7 expression is lost. In such cases, the viral oncogenes E6 and E7 are inappropriately expressed in a population of cells that retain the capacity to divide, thereby initiating cell proliferation. As this population of cells proliferates, it overtakes the epithelium-producing lesions that are, by definition, characterized by less orderly squamous maturation and basal-like cell overgrowth with evident mitotic activity. Possible promoters of this process could be smoking, other viruses and infectious stimuli, inflammation, or random mutation. The relative infrequency of these effects is biologically manifest by the latency and relative rarity of HSILs versus LSILs. Progression to the proliferative phenotype occurs most frequently, albeit not exclusively, with high-risk viral types, and results in the HSILs, also called moderate squamous dysplasia, severe squamous dysplasia, or squamous carcinoma in situ (CIN 2/3). Thus, the Bethesda System's break between low-grade and high-grade follows in part from the biologic changes manifest between these morphologies. Indeed, from the standpoint of epithelial biology, there is little rationale for separating moderate from severe dysplasia, in that the critical break occurs between mild and moderate dysplasia with the switch to a proliferative as opposed to a differentiated and virally productive phenotype.

In HSILs, the proliferating basaloid cells, driven by E6/E7 overexpression, are at much greater risk for the acquisition of additional genetic errors, clonal selection, etc, perhaps under the influence of the same external mutagens and/or host genetic predisposition, which further promotes the development of the fully malignant phenotype, most often an invasive squamous cell carcinoma. The different subtypes of squamous cancer are probably related to the multistep and somewhat random nature of the process. The proportion of different types may reflect the relative likelihood of different genetic pathways to a “successful” cancer, in part modulated by the microenvironment in which the lesion develops. Hence, early observations that keratinizing cancers are often more ectocervical than large cell nonkeratinizing or small cell malignancies, which tend to originate higher in the endocervical canal, have some contemporary validation.

Given this model for cervical squamous neoplasia, one still needs to account for glandular and small cell neuroendocrine neoplasms. Reserve cells that are already committed to glandular differentiation are, because of a lack of an appropriate differentiation environment, not going to be productive of virions. This is because the productive viral life cycle requires the cellular milieu of orderly squamous differentiation. If this is true, then viral infection in cells committed to glandular differentiation most often results (from the viral standpoint) in an abortive or latent infection of morphologically normal endocervical cells. Rarely, dysregulation of viral early gene expression occurs in these usually nonpermissive cells. This dysregulation leads to hyperproliferative lesions of glandular cells, which pathologists recognize as severe endocervical dysplasia/adenocarcinoma in situ. There is no biological correlate in this model of a low-grade glandular dysplasia. Hence, this description nicely explains the inability of pathologists to reproducibly recognize, either cytologically or histologically, a clinically meaningful lesion less severe than what most call adenocarcinoma in situ.

Human papillomavirus 18 seems to be more successful at inducing neoplastic proliferation in glandular cells than HPV-16. Perhaps this success occurs because HPV-18 has a greater disposition to integrate into the genome, and perhaps because it may have some preference for cells predisposed to other than squamous differentiation. Parenthetically, little if anything is known about the mechanism of HPV type-specific cellular tropism. However, no HPV type can be exclusively trophic for nonsquamous cells, because under the model described, that strain of virus would be eliminated from the population. Depending on the genetic switches that over time accompany virally induced glandular proliferations, the outcome may be an invasive adenocarcinoma, most often endocervical, but less frequently of another type, for example, endometrioid or clear cell. The relative frequencies of the different types of cervical adenocarcinomas again may simply reflect the relative frequency of the different populations committed toward various types of differentiation. Essentially identical arguments can be made for the development of small cell neuroendocrine carcinomas, tumors that are almost always associated with HPV-18 and whose low incidence probably reflects the relative abundance of a susceptible neuroendocrine-committed precursor cell population and the rarity of “successful” viral induction of cell proliferation in such cells.

The concepts described in the previous section have definite implications for the utility of HPV testing in diagnosis. Indeed, HPV testing will assume increasing importance in clinical practice, especially on correlated cytology samples.21 However, one must be cautious when evaluating and implementing new technologies.22,23 Biologic rationale and technical analytic variables, such as sensitivity and specificity, while obviously important, do not prove clinical utility and validity. Controlled clinical trials of candidate new tests provide the clearest proof of utility, especially in an era of cost containment.21,22 As virtually all cancers are HPV-associated, so too should all real precursors be HPV positive. Indeed, the linear relationship between diagnostic certainty on cervical cytology and HPV infection as assessed by hybrid capture testing provides an excellent means for both diagnostic adjudication and quality control.24,25 The 2 morphologic settings in which diagnostic difficulty on biopsy most often arises is the diagnosis of LSIL and distinguishing it from normal, as well as the differential diagnosis of immature metaplasia from a true high-grade lesion. Based on data from the National Cancer Institute–sponsored ASCUS-LSIL Triage Study (ALTS) trial, both are real diagnostic problems; diagnostic reproducibility even among experts is moderate at best.26 An easy-to-apply, sensitive, and specific HPV assay to adjudicate these interpretations could potentially be of great patient benefit.

However, how to assay for HPV is problematic, particularly on biopsies. Functionally, the most common HPV assay is best suited for solubilized cytology specimens, allowing only for a correlated study between cytology and histopathology. In situ hybridization would be seemingly ideal, but the large numbers of potential HPV types that it is necessary to detect for a clinically robust test makes developing such an assay quite problematic.27 Lowering the stringency and using probe cocktails may have untoward effects on both sensitivity and specificity. Polymerase chain reaction–based assays do work and are improving, but issues of both sensitivity and specificity can still plague assays on wax.28,29 Sensitivity in this case is an issue of HPV type spectrum, as well as of analytic quantity.

What about testing for specific HPV types? Assuming we did have an adequate assay, does knowing an individual type as opposed to a “risk group” help the clinical decision process? Original concepts of low-risk viruses being associated with low-grade lesions and high-risk viruses being correlated with high-grade lesions are simplistic and inaccurate. The first analysis from the ALTS enrollment database was a correlative analysis of HPV prevalence in patients with LSIL cytology.17 Of 642 women with an LSIL diagnosis who had analyzable HPV test results, 82.9% of the women (95% confidence interval = 79.7%–85.7%) had a positive result for the high-risk probe mix used in this assay. This high frequency of high-risk HPV positivity was confirmed by independent polymerase chain reaction assays on a subset of 210 of these patients with a very high concordance. Because of this finding, the potential for this HPV assay to effectively triage a population of women with an LSIL diagnosis is obviously limited. This study supports a growing consensus that the earlier reports suggesting that low-grade dysplasias were highly associated with low-risk HPV viral types were incorrect. Most of the mucosotropic viral infections that occur in the uterine cervix are high-risk viral types. Furthermore, most of these high-risk viral infections produce only low-grade lesions that are transient and not productive of high-grade dysplasia. Thus, the term high-risk or oncogenic HPV is a relative misnomer.15 

Could other HPV correlates be useful in selecting out the patients at risk for high-grade precursors? As stated, much is known about the molecular pathogenesis of cervical neoplasia, and selected markers in these pathways might prove to be discriminating.30 For example, expression of the E6 and E7 viral oncogenes is thought to be essential for lesion development, yet these oncogenes do not effectively segregate low- from high-grade lesions, although detection of these RNAs might more effectively predict lesion presence, that is, higher clinical specificity than HPV DNA alone. Integration of HPV genomes into the host DNA might predict for the presence of a high-grade lesion. Conceptually, this also has some attraction because there are differences in the molecular host interaction of at least the prototype viruses HPV-16 versus HPV-18. The viral genome of HPV-18 is virtually always present in a purely integrated form that could theoretically cause higher-level dysregulation of the oncogenic E6 and E7 regions of this virus compared with that of HPV-16. There is also a stronger association of HPV-18 with adenocarcinoma and especially small cell carcinoma, both histologic types associated with poorer prognosis in many, but not all, studies. However, integration assays are somewhat complex, especially given the heterogeneity of potential HPV types, as well the random nature of HPV integration sites.

In summary, HPV testing of some sort could have high specificity in segregating true lesions from their mimics, but which assay and how it should be applied, especially at the time of biopsy diagnosis, is still an evolving, and at times complex, problem.

Measurements of cell proliferation and abnormalities of nuclear DNA content have been extensively evaluated in many tumor systems as adjunctive prognostic and diagnostic markers.31–34 Historically, some have demonstrated a strong correlation between aneuploidy and dysplasia, yet ploidy analysis is very time-consuming and problematic on small biopsies. More recently, assays of cell proliferation and indices of cell death/apoptosis have also come under scrutiny.35–38 For cervical neoplasia, these indices may seem highly likely to be relevant, since the papillomavirus E6 and E7 protein's interactions with the retinoblastoma and p53 pathways, respectively, place them squarely at the center of cellular regulation of both proliferation and apoptosis. Induction of cell proliferation, either S-phase measurement or immunohistochemical indices like Ki-67 or proliferating cell nuclear antigen, have been the most widely reported correlate of cervical preneoplasia, and the switch from a virally productive to cellular proliferative phenotype seems well correlated with the progression to high-grade lesions. Thus, markers like Ki-67 or cyclin E may be useful in the problematic biopsy, when the differential diagnosis is between a reactive process like immature metaplasia or gland regeneration and a high-grade cancer precursor.39 

Direct analyses of genotypic and phenotypic associations with development of neoplasia are highly prevalent in the literature.40–45 Immunohistochemical studies looking for overexpression or underexpression of epidermal growth factor, HER-2/neu, carcinoembryonic antigen, MN, and others all have some biologic rationale, but at times the relationships are not well developed or represent second-level correlates (eg, MN). Often the reports on these markers are directly conflicting, and the reasons for these discrepancies are most often in the details of study design. Sources of variability between these studies run the gamut from technical and reagent variability to differences in populations studied to differences in the type of statistical analysis and elements included in the multivariate (if there is one) model. Similarly, some genotypic studies have been performed including c-myc, c-Ha-ras, p53, and others. In some cases, such as p53, HER-2/neu, cyclin D1, and RB, the results of genetic versus immunohistochemical studies may or may not agree, again often due to technical variables. While some data appear promising based on reasonable analyses, most are unconfirmed or not replicated and hence for practical purposes cannot be used for clinical decision making.

To reiterate, studies have shown that the accuracy of the cervical biopsy is low, especially the predictive value of low-grade changes. Interpretive reproducibility is at best moderate, and pathologists make variable trade-offs between sensitivity and specificity.26 High sensitivity and specificity for the detection of high-grade precancer (CIN 2/3) are required for the appropriate treatment of patients. An objective biomarker that is both sensitive and specific would allow unambiguous identification of truly dysplastic cells. Many experimental and epidemiological studies reveal that expression of the high-risk HPV oncogenes E6 and E7 in cervical epithelia is required to induce and maintain neoplastic growth of cervical epithelia. There is increased expression of the E6 and E7 genes in high-grade precancer. However, attempts to monitor expression of viral oncogenes in routine clinical samples have been limited by the great variety of the high-risk HPV types, the low expression level of the viral oncogene products, and the lack of sufficiently specific technologies and reagents to detect them. To overcome these limitations, a cellular surrogate marker, that is, a gene expressed by the host cell in response to the expression of viral oncogenes, but not expressed in normal nontransformed cells, could be a better test.

Recent reports demonstrate that increased expression of the high-risk HPV oncogenes in cervical dysplasia results in highly specific overexpression of p16^INK4a in cervical dysplasia and cervical cancer cells, which can easily be detected by certain specific monoclonal antibodies.37,46 Almost 100% of high-grade cervical lesions and invasive cancers have preliminarily been shown to express very high levels of p16^INK4a, whereas nondysplastic cervical epithelia did not stain positive for p16^INK4a using the same antibodies. These observations suggested that p16^INK4a immunostaining might allow precise identification of dysplastic cervical epithelial cells and provide a diagnostic test that was sensitive, specific, and reproducible for high-grade and progressive precancer. Some authors have also sought to combine p16 with Ki-67 in an immunohistochemical panel approach to problematic cases. Perhaps some of these more promising markers will be submitted for formal large-scale clinical trials. Such an effort could potentially lead to further revision of our diagnostic concepts.

Accurate diagnosis that directs appropriate management triage is the goal of cervical biopsy. Accurate diagnosis requires consistent application of diagnostic criteria. While better education on criteria and simpler clinically relevant classifications can help improve diagnostic accuracy and reproducibility, the numerous biologic variables inherent in this process mean that pathologists will always be faced with common problems related to subjective morphologic interpretation. In some of these cases, selected diagnostic adjuncts may prove useful and may even help refine further what is or is not a significant lesion. Among the candidates, some form of HPV test, measures of cell proliferation, and perhaps the cell cycle–linked marker p16 appear most promising for making the transition to routine clinical practice.