- Top of page
- Material and methods
- Supporting Information
Determining the causal attribution of human papillomavirus (HPV) genotypes to cervical disease is important to estimate the effect of HPV vaccination and to establish a type spectrum for HPV-based screening. We analyzed the prevalence of HPV infections and their attribution to cervical disease in a population of 1,670 women referred to colposcopy for abnormal cytology at the University of Oklahoma. HPV genotyping was performed from cytology specimens using the Linear Array assay that detects 37 HPV genotypes. We used different methods of type attribution to revised cervical disease categories. We found very high prevalence of multiple HPV infections with up to 14 genotypes detected in single specimens. In all disease categories except for cancers, there was a significant trend of having more infections at a younger age. We did not see type interactions in multiple genotype infections. HPV16 was the most frequent genotype at all disease categories. Based on different attribution strategies, the attribution of vaccine genotypes (6, 11, 16, 18) ranged from 50.5 to 67.3% in cancers (n = 107), from 25.6 to 74.8% in CIN3 (n = 305), from 15.2 to 52.2% in CIN2 (n = 427), and from 6.6 to 26.0% in <CIN2 (n = 708). In the HSIL cytology group (n = 651), attribution ranged from 26.1 to 64.7%. The attribution of vaccine types to HSIL was substantially higher compared to the lower cytology categories. The potential range of HPV genotype attribution is wide at the disease categories <CIN2 to CIN3. Genotyping from cervical lesions and analyzing viral oncogene expression can improve estimates of HPV genotype attribution. © 2009 UICC
Carcinogenic types of human papillomaviruses (HPV) are the causal agents of cervical precancer and cancer. More than 100 HPV genotypes have been identified to date. Among 30 types that infect the genitourinary mucosa, approximately 15 are carcinogenic and are highly associated with the development of cervical cancer.1 Two genotypes, HPV16 and HPV18, account for ∼65% of cancer and its immediate tissue precursor, cervical intraepithelial neoplasia 3 (CIN 3); however, over 12 other types account for the remaining third of cases.2
Most carcinogenic HPV infections, especially among women less than 30 years of age, are clinically occult and spontaneously regress.3 Routine cytologic screening detects a minority of such infections as equivocal (atypical squamous cells, ASC) or mildly abnormal (low-grade squamous intraepithelial lesions, LSIL) Pap tests. Referral of such women to colposcopy and biopsy may result in histologic diagnoses of CIN 1 or 2. As would be expected, data suggest that the percentage of lesions stratified by severity varies with the associated HPV type.4 Types other than HPV 16 and 18 account for a higher percentage of cytologic ASC or LSIL or histologic =<CIN2, many of which spontaneously regress.
Defining the contribution of individual HPV genotypes to each grade of severity of cervical neoplasia is important for multiple purposes. Such data are needed to estimate how much cervical disease will be prevented by current vaccination programs,5 and to determine which genotypes should be targeted by the next generation of vaccines. Similarly, it is important to consider which genotypes should be included in future HPV screening assays. Moreover, as vaccines and type-specific HPV screening are applied, it will be important to monitor changes in disease patterns caused by other HPV genotypes, especially in the context of understanding cross-protection (i.e., prevention of infections and lesions caused by HPV genotypes closely related to those included in vaccines) and “unmasking” of putative carcinogenic types (i.e., lesions caused by nonvaccine genotypes that previously remained clinically elusive possibly because of faster progression and treatment of concurrent HPV16/18 associated lesions).6
However, identifying the causal genotype for each grade of cervical neoplasia is complicated because multiple HPV genotypes often co-exist within the cervical epithelium. Multiple cervical lesions within an individual patient may be caused by different genotypes and a precancerous lesion caused by a specific carcinogenic genotype can be surrounded by transient HPV infections.7
Attribution of HPV genotypes to cervical disease is also complicated by frequent misclassification of cervical disease, related either to interpretation problems during histological examination8 or, even more importantly, to misidentification of the worst lesion during colposcopy and biopsy.9 On the basis of the analysis of combinations of cytology, histology and HPV genotyping results in the Study to Understand Cervical Cancer Early Endpoints and Determinants (SUCCEED), we recently proposed a revised disease grading algorithm that reflects the functional categories of cervical disease progression and can correct some of the misclassification errors related to biopsy placement.10
In the present analysis, we evaluate the distribution of 37 HPV genotypes in cervical cytology specimens obtained from 1,670 women at a single US institution comprising the disease continuum, from HPV infection to invasive cancer. We demonstrate the challenges of attributing HPV genotypes to disease categories and show the potential range of attribution based on different algorithms.
- Top of page
- Material and methods
- Supporting Information
Twenty years ago, cross-sectional and case-control studies established the etiological link between major carcinogenic HPV types and cervical cancer.20–22 Although the most frequent carcinogenic types have been refined for cervical cancers since then, the carcinogenicity of rare HPV genotypes is still being studied.1 Vaccination against HPV 16 and 18 will likely change the associations between different grades of cytologically or histologically defined lesions, the causative HPV types, and the biological potential and clinical threat posed by such disease states.23, 24 Accordingly, estimating the attribution of carcinogenic HPV genotypes to cancer and CIN 3 is an important step in considering the application of HPV genotyping in screening and management and in developing multivalent prophylactic vaccines. However, developing these estimates has been challenging because of the frequent occurrence of multiple type infections, imprecision in defining disease states, differences in detection between cytologic and histologic specimens and other factors.
We present a comprehensive analysis of HPV genotypes in SUCCEED, a large cross-sectional study of women referred to colposcopy due to recently abnormal cytology or histology in the United States. The current analysis includes 1,670 women spanning all categories of cervical squamous neoplasia, with a high proportion of cervical cancers and precancers. As we analyzed a population referred mainly due to abnormal Pap results, women with transient HPV infections that do not cause cytological abnormalities are underrepresented in our population. In contrast, due to the large catchment area of the OUHSC dysplasia clinic, we have very good representation of women with productive HPV infections, precancers, and cancers from an urban population from Oklahoma City and a rural population from the state of Oklahoma.
We demonstrate a strong association between patient age and number of HPV genotypes for all disease categories from <CIN2 to CIN3; however, disease category itself showed no association with number of genotypes. Prevalence and natural history studies on HPV infection have demonstrated that the peak age of HPV infections is between 20 and 25 years, related to the highest level of sexual activity and associated risk of exposure to HPV infections.25–30 In agreement with that, the average age of women with 2 or more HPV infections in our study was 30 or younger in the respective disease categories (except for women with cancer).
We note that our analysis of observed and expected type combinations is exploratory. As stratification into double, triple and quadruple combinations of the 37 genotypes led to large numbers of potential combinations with extremely few expected and observed events, we combined all cases with 2 or more infections. Previous studies have used different approaches, e.g., including cases with single infections, limiting the analysis to the most frequent types, or analyzing combinations of clades rather than single types.31–34 Although some previous studies have shown specific clustering of different genotypes,31, 34 we only observed 2 more frequent than expected combinations of closely related types that suggest cross-hybridization of the genotyping probes rather than biologic clustering.
Although HPV genotyping data from cancers are the most important parameters for public health purposes, currently, the attribution of the vaccine types HPV16 and HPV18 (and, to a lesser extent, that of HPV6 and HPV11) to the overall disease burden is being used extensively to make assumptions on HPV vaccine efficiency.5, 35 Similarly, when designing HPV screening assays, ultimate clinical sensitivity for the detection of precancers by inclusion of types rarely associated with cancer has to be weighed against the potentially dramatic loss of specificity when the respective type is frequent in low-grade disease (e.g., HPV53).36
In our study, only about a third of the CIN2 and CIN3 cases had infections with single HPV genotypes. Even after restriction to carcinogenic types, only 50% of CIN2 and CIN3 lesions could be attributed to single carcinogenic type infections (Supporting Information Table 1). In the remaining cases, assumptions underlying attribution matter, but there are no accepted rules on how to attribute causative HPV genotypes without further functional data.
We present a range of potential attributions for each genotype, based on the frequency in single infections and counting all lesions containing the respective genotype. Due to the high proportion of multiple infections observed, this range can be very wide. In CIN3, HPV16 was found in 25% of the cases as a single infection, but was present in 73% of the overall cases. We can provide more precise estimates of type-specific causality for cancers since most cancers contain only one carcinogenic HPV type. The percentage of cancers caused by each genotype is very similar for single-type, all-type, proportional or hierarchical attribution (e.g., the range of potential attribution of HPV16 varied only from 41 to 55%). A notable difference in HPV genotype attribution to cancers in our study was an underrepresentation of HPV31 compared to previous studies from North America.2
We have used 2 different approaches to attribute HPV genotypes to cervical disease categories: In proportional attribution, a fraction of each case is attributed to every genotype in a multiple infection, while in hierarchical attribution a case is completely attributed to the most frequent type. Thus, the hierarchical attribution favors the more frequent types, especially HPV16, while the proportional attribution is more likely to attribute some cases to types without carcinogenic potential. Ideally, to use data on truly causal HPV infections, the frequency of types in single infections should constitute the underlying hierarchy/proportions (as previously described by19). However, the low number of single genotype infections in our study prohibited us from using the single type frequencies. The high number of multiple infections observed in our study can be attributed to the use of a sensitive assay that is capable of detecting 37 HPV genotypes simultaneously. In most previous studies, HPV typing covered fewer types and consequently, these studies have described lower frequencies of multiple infections. We also acknowledge that there may be minor errors in our genotyping results leading to additional variation in HPV attribution.
Within the constraints of our analysis, we observe that the percentage of vaccine types in <CIN2 is substantially lower (ranges from 6.6 to 26.0%) than in CIN3 (ranges from 25.6 to 74.8%). Although at least 26.1%, but up to 64.7% of HSIL cytology might be eliminated by the quadrivalent vaccine, probably much less than a third of ASCUS and LSIL will be prevented. This confirms the notion that cytological screening will suffer substantially in a vaccinated population, since the relative excess of low-grade lesions will increase further.23, 24 In our attribution of vaccine types to disease categories, we have not accounted for potential partial cross-protection against closely related types, especially HPV45 and HPV31, that has been previously demonstrated in vaccine trials.37 On the basis of our data, cross-protection would not reduce cases by more than 4–6% in the CIN3 and HSIL categories, and even less in cancers and the lower disease categories.
Our HPV genotyping data were derived from sampling the complete cervical surface which could involve multiple lesions each with their own independent causative HPV genotype. In theory, it is possible to determine the causal type in multiple infections by analyzing DNA from isolated lesions, HPV RNA expression patterns, oncoprotein expression and viral integration. However, it has been demonstrated that it is challenging to obtain lesion-specific genotypes from histological specimens,7, 38 and other more specific assays are currently not available in reliable high-throughput formats.
In our molecular studies in SUCCEED, further determination of causal HPV genotype attribution will require analyzing additional markers such as viral oncogene expression and integration. Only based on these additional data can we move forward with improved estimates of genotype attribution, validate the attribution models described here and gain a better understanding of the potential effects of HPV vaccination on cervical disease, and which genotypes to include in new generation HPV detection assays.