Two-Week versus Six-Month Sampling Interval in a Short-Term Natural History Study of Oral HPV Infection in an HIV-Positive Cohort

Background Oral HPV infections detected six-months apart were compared to those detected bi-weekly, in an HIV-positive cohort, during the intervening months to elucidate systematic biases introduced into natural history studies by sampling interval. Methods Fourteen consecutive oral rinse samples were collected every two weeks for six months from an HIV-positive cohort (n = 112) and evaluated for the presence of 37 HPV types. The cumulative probability of type-specific HPV detection at visits 1 through 14 was determined as a function of infection categorized at visits 1 and 14 as persistent, newly detected, cleared or absent. Transition models were used to evaluate the effect of HPV viral load (measured by RT-PCR for HPV 16, 18, 31, 33, 35) on infection persistence. Results The average point prevalence of oral HPV infection was similar at two-week and six-month sampling intervals (45% vs. 47%, p = 0.52), but cumulative prevalence was higher with the former (82% vs. 53%, p<0.001) as was the cumulative prevalence of type-specific infections (9.3% vs 3.8%, p<0.0001). Type-specific infections persistent under a six-month sampling interval had a high probability (0.93, 95%CI 0.83–0.98) of detection at 50% or more of the intervening visits and infections that were absent had a high probability (0.94, 95% CI 0.93–0.95) of no interval detection. The odds of detection at any visit significantly increased for each unit increase in HPV viral load at the previous visit. Conclusions Six-month sampling is appropriate to model factors associated with type-specific oral HPV infection persistence but may misclassify HPV-exposed individuals as unexposed.


Introduction
Oral human papillomavirus (HPV) 16 infection is strongly associated with oropharyngeal squamous cell carcinoma (OPSCC) [1]. Moreover, seropositivity to HPV16 elevates risk for development of OPSCC [2]. Although consistent with a temporal link between oral HPV infection and development of OPSCC, oral HPV natural history studies have not been reported. Such studies will be critical for clarifying potential use of oral HPV detection in OPSCC screening programs.
An important factor for HPV natural history studies is the sampling interval, and six-month sampling intervals are commonly used for cervical studies. We have previously evaluated the feasibility of a six-month sampling interval for oral HPV infection [3]. However, a recent natural history study of cervical HPV infection among adolescent women using a two-week sampling interval revealed short-lived infections (, lasting six-month) were common.
As a consequence, the cumulative prevalence was higher in this study relative to prior studies in similar populations [4].
We performed a short-term natural history study of oral HPV infections to compare the results of a two-week to a six-month sampling interval to identify potential systematic biases introduced into natural history studies by the frequency of sampling. Oral HPV infections refers to infections of the oral cavity and oropharynx. An HIV-positive cohort, known to be at elevated risk for oral HPV infection and related malignancies, was selected as our study population.

Ethics statement
This study was conducted according to the principles expressed in the Declaration of Helsinki. This study was approved by the Institutional Review Board of Johns Hopkins Hospital (application number 03-11-24-04). All patients provided written informed consent for the collection of samples and subsequent analysis.

Study population
A convenience sample of 112 HIV-positive patients from the Johns Hopkins Hospital was recruited during March 2005. Eligibility criteria included known HIV-positive status and willingness to comply with bi-monthly (every two week) visits for six months. Enrollment took into consideration both CD4 count (, or $200 cells per mL) and gender to result in a study population with approximately equal proportions. Subjects received monetary compensation ($160) for participation.

Procedures
Oral rinse samples were collected every two weeks for six consecutive months (visits 1-14) using a 30 second Scope TM mouthwash rinse and gargle [5]. A self-administered questionnaire was completed at enrollment for measurement of baseline characteristics, however interval behavioral data were not ascertained. Venous blood was collected at visits one and 14 for measurement of CD4 count and HIV viral load. A Scope TM sample was collected per 20 rinse samples and processed similarly as a contamination control.

Laboratory analysis
Oral exfoliated cells were separated by centrifugation, washed, and re-suspended with phosphate buffered saline and stored at 280uC.
DNA was purified by use of a modified protocol for Puregene DNA Purification (Gentra Systems, Minneapolis, MN) [5]. The presence of HPV genomic DNA of 37 HPV types (Low-risk types 6,11,38,40 [6]) was detected by PGMY09/11 L1 consensus primer PCR [7], and type-specified by hybridization to a prototype line blot (Roche Molecular Systems, Inc., Alameda, California). Samples positive for b-globin were evaluable, scored as negative or positive, and the HPV type(s) detected reported for positive samples.
HPV viral load for types 16, 18, 31, 33, and 35, chosen because of their association with OPSCC [8], was measured for all samples collected from individuals ever positive by line blot hybridization for the corresponding HPV type(s). Type-specific, Taq-Man, realtime PCR (Applied Biosystems, Foster City, CA) assays were performed as previously described [9,10]. Each sample, including the standards, was assayed in duplicate, and a duplicate mean $ one copy of viral DNA was positive. To adjust oral HPV viral load to the number of oral epithelial cells present per PCR-reaction (see below), cell number was estimated by use of a TaqMan real-time PCR targeting a single copy human gene on chromosome 7, Human Endogenous Retrovirus 3 (ERV-3) [11,12].
HIV viral load was measured by use of the Roche AMPLI-COR TM HIV-1 Monitor Test, version 1.5 and reported from ,400 to 750,000 copies per mL. CD4 cell counts were measured by use of the Beckton Dickenson BD TriTest CD4 FITC/CD8 PE/CD3 PerCP Reagent and reported as the number of CD4 + cells per microliter of blood (absolute count).

Statistical analysis
Descriptive statistics were used to report demographic, biological and behavioral characteristics of the study population and to compare subjects who completed the study, defined as having attended visits one to 14 with a maximum of one interval missed visit, to those who were lost to follow-up. HPV infection status was determined at visits one through 14 and evaluated both in terms of individuals and for type-specific infections. In analysis by individual, a subject was considered infected if one or more of the 37 HPV types evaluated were detected at each visit (point prevalence) or detected in any of the samples collected over the course of the study (cumulative prevalence). In analysis by infection, presence of each of the 37 type-specific HPV infections was evaluated for each individual at each visit (point prevalence) or in any of the samples collected over the course of the study (cumulative prevalence).
Point and cumulative prevalence of oral HPV (and exact binomial 95% confidence intervals (CI)) were calculated for visits one through 14. Average point prevalence represented the weighted average of point prevalence for each visit, using the number of individuals or infections at a given visit as the weight.
McNemar's test was used to compare the two-week and six-month cumulative prevalence estimates.
The primary goal was to evaluate HPV infection patterns over time, specifically to compare two-week to six-month sampling intervals. These analyses were restricted to the 91 individuals who completed the study. Under the six-month sampling interval, typespecific HPV infections were categorized as ''persistent'' (+/+) if present at both visit one and 14, ''newly detected'' (2/+) if absent at visit one and present at visit 14, ''cleared'' (+/2) if present at visit one and absent at visit 14 or ''absent'' (2/2) if not detected at either visits one or 14. These categories were distinct from those for the two-week sampling intervals in which infections were categorized as ''persistent'' if present in two or more consecutive visits, ''transient'' if detected without persistence, and ''absent'' if never detected. A persistent infection was considered ''cleared'' if not detected for at least two consecutive visits. An infection was defined as having ''recurred'' if it was detected at any visit after clearance.
To evaluate the relationship between HPV viral load and infection persistence, the adjusted HPV viral load for types 16, 18, 31, 33, and 35 was plotted across visits. The adjusted HPV viral load was the logarithmically converted ratio of (10 5 ) HPV copy number (as measured by real-time PCR) divided by the number of cells in the sample (as measured by ERV-3 real-time PCR). A logistic transition model was used to evaluate the relationship between the magnitude of the adjusted viral load at one visit and a positive (i.e. non-zero) adjusted viral load at the next visit.

Study population
The characteristics of the study population are shown in Table 1. Ninety-one of 112 (81%) individuals completed the sixmonth study. Of these, 74 completed all 14 visits and 17 missed a single visit. The median number of visits for the 21 individuals lost to follow-up was 10 (interquartile range: 4-12). There were no significant differences between patients who completed the study and those lost to follow-up (Table 1).

HPV detection
In total, 1,434 of 1,568 (91%) scheduled visits were completed. Almost all (1,432 of 1,434 (99.9%)) oral rinse samples were bglobin positive. All controls for specimen collection and processing were appropriately negative for HPV and b-globin.

Point and cumulative prevalence of oral HPV
First, we determined the proportion of the study population that had an oral HPV infection. The point prevalence for oral HPV infection measured at two-week sampling intervals remained relatively stable (range, 32-54%, Figure 1A). Average point prevalence for the two-week and six-month sampling intervals was similar (45%, versus 47%, P = 0.52) ( Table 2 and Figure 1A). However, cumulative oral HPV prevalence, a measure of cumulative oral HPV exposure over the course of the study, was significantly higher with a two-week than a six-month sampling interval (82.1 versus 53%, P,0.0001, Figure 1B and Table 2).
We then evaluated type-specific oral HPV infections among the entire study population. We considered all possible infections, i.e. 37 infections from visits 1-14 (a total of 1, 434 evaluable samples) in 112 individuals ( Table 2). The average point prevalence was similar for the two-week and six-month sampling intervals (3.0% versus 2.9%, P = 0.59). However, cumulative prevalence of detection of 37 type-specific infections was significantly higher with a two-week than a six-month sampling interval (9.3% (n = 386 infections) versus 3.8 (n = 158 infections), P,0.0001). The average point prevalence for each of the 37 type-specific infections ranged from 2.1-3.9% at the two-week sampling intervals (Table 3, data only shown for HPV types 16,18,31,33,35).

Patterns of HPV infection
The patterns of HPV infection were evaluated for both twoweek and six-month sampling intervals among the 91 individuals who completed visits 1 and 14 ( Figure 2). Among these 91 individuals, we evaluated a total of 3,365 possible infections, i.e. 37 HPV types among 91 individuals, minus two samples missing data on one HPV type or (37691) 22 = 3,365.
Six-month sampling results. At visit one, 88 HPV infections were detected in 41 of 91 (45%) participants. At the final visit, 69 (78%) were categorized as persistent and 19 (22%) were categorized as cleared (Figure 2), and 56 infections were newly detected. The remaining 3,221 infections (96% of all possible infections) were absent at visit one and 14 ( Figure 2).
Two-week sampling results. As would be expected, the number of infections detected with a two-week sampling interval was higher than for a six-month sampling interval ( Figure 2). In total, 338 of 3,365 (10%) possible infections were detected. By the two-week sampling definition, 194 were transient, 144 were persistent, and 59 (41%) of these persistent infections were categorized as cleared. Of the infections that cleared, 42 (71%) recurred, 27 (46%) of which again met criteria for persistence. The median percentage of visits an infection was detectable was 14% (range, 7-100%), and 83 infections (25%, CI: 20-30%) were detectable for the majority (.50%) of visits. Only 24 of 144 persistent infections under a two-week definition were detectable at all subject visits.
Comparison of sampling schemas. We then compared the HPV detection results obtained with a two-week to a six-month sampling interval ( We evaluated the probability that an infection was detectable for a certain percentage of the two-week visits stratified by their categorization under the six-month sampling interval as persistent, cleared, newly detected or absent (Figure 3). Oral HPV infections defined as persistent under the six-month sampling interval had a high probability (0.93, 95% CI: 0.83, 0.98) of being detected at more than 50% of the two-week visits. Similarly, there was very low probability (0.002, 95% CI: 0.00, 0.004) that oral HPV infections that were absent at both visits 1 and 14 were detected at more than 50% of intervening visits. Indeed, if a type-specific HPV infection was not detected at either visit 1 or 14, the probability that the infection was not detected with a two-week sampling interval was 0.94 (95% CI: 0.93, 0.95). Probabilities for cleared and newly detected infections were largely indistinguishable. The likelihood that a cleared or newly detected infection by the six-month sampling interval definition was detectable at 50% or more of the two-week visits was equivalent (0.26 and 0.25, respectively).

Factors affecting persistence
We evaluated the relationship between HPV viral load and persistence or clearance of five high-risk HPV types (HPV16, 18, 31, 33, or 35). The pattern of viral load over the course of the study for those individuals with at least one positive line blot assay for a specific high-risk HPV type is shown in Figure 4. Upon visual inspection, infections present across multiple visits appeared to have a higher viral load (Figure 4, darker intensity bars) than those inconsistently present over time (Figure 4, lighter intensity bars).
This visual impression was statistically confirmed in a transition analysis that evaluated whether the odds of a positive oral HPV test at a given visit was affected by the HPV viral load at the prior visit (Table 4). For each HPV type evaluated, the odds of a positive test were significantly increased among individuals who tested positive at the previous visit. Additionally, there was a significant increase in the odds of a HPV-positive test for each unit increase in adjusted HPV copy number from the previous visit (OR range for HPV 16, 18, 33, 35: 1.26-1.51, P-values range: ,0.0001-0.0001).  Similar transition analyses focusing on line blot analysis results for the six-month interval sampling (visits 1 and 14) and baseline characteristics (age, gender, smoking, CD4 count, viral load, and HPV serostatus) did not show any significant associations (data not shown).

Discussion
Our data indicate a very low probability of interval detection of a type-specific oral HPV infection if the type was not detected at a six-month sampling interval. Furthermore, there was a very high probability that a type-specific infection categorized as persistent at a six-month sampling interval was detectable for the majority of the intervening two-week visits. A six-month sampling interval therefore appears appropriate for natural history studies of oral HPV infection that model factors associated with type-specific persistence.
Cervical HPV natural history studies serve as a model for development of analogous studies at other anatomic sites (e.g. oral and anal). The clinical algorithm for cervical cancer screening predated natural history studies of cervical HPV infection and predetermined the commonly used six-month sampling interval. Shorter sampling intervals were not examined prior to implementation of large, well-designed landmark natural history studies. The expense, complexity and burden of serial anogenital sampling likely prohibited more frequent sampling. Consequently, cervical infections that would be detected at a shorter interval are generally believed to not be pathologically significant, while persistence of infection at six-month intervals significantly elevates risk for cervical dysplasia [13] and thus has been established as a surrogate for this disease outcome [14]. Given the growing recognition of oral HPV infection as a risk factor for oral cancers, we elected to examine a priori the potential biases introduced by sampling interval, rather than to assume the cervical standard of a sixmonth sampling interval.
Our initial studies of the natural history of oral HPV infection have focused on a study population with high prevalence, HIVinfected individuals. We acknowledge that the behavior of infections in this population likely differs from what would be observed in an immunocompetent population. For example, transient detection of oral HPV infection was commonly observed with two-week sampling, and the underlying reasons for this observation are unclear. This may be attributable to technical issues, such as variability in factors affecting oral rinse sampling over time, or fluctuation of HPV viral load above and below the lower limit of detection of the Roche line blot, which is known to vary by type. Consistent with such fluctuation around the limit of assay sensitivity is our finding that the majority of cleared infections under the two-week definition subsequently recurred and met criteria for persistence. Although laboratory contamina- tion of the specimen could also be an explanation, we did not observe any evidence of contamination in control samples. Alternatively, in an immunosuppressed population the high frequency of transient, low viral load infection may represent intermittent reactivation of infections due to immunosuppression. Transient detection may also represent exposure in the absence of subsequent establishment of infection. We did not have the interval sexual behavior data which may have allowed us to explore this possibility.
The biological significance of transient HPV infections in the oral cavity, oropharynxand cervix is unclear. While the majority of cervical HPV infections are transient infections, persistent infections are significantly more likely to progress to both premalignant and malignant lesions [13]. However, even transient cervical HPV infections (defined as an HPV infection detected at only one visit within a 3 to 12 month period) have been associated with an approximate 5-fold (OR 5.5 95%CI 1.4-21.9) increased risk of squamous intraepithelial lesions in comparison to no detection [15]. High HPV viral load is also associated with elevated risk for incident cervical intraepithelial neoplasia 2 (CIN II) or more invasive pathology [16]. Our two-week interval analysis suggests that infections with a high HPV viral load were more likely to persist than low viral load infections, but an important limitation to this study is that associations with disease outcomes in the oral cavity were not possible.
Our data lend support to the use of a six-month sampling interval for studies designed to evaluate factors associated with type-specific infection persistence. However, misclassification of  transient oral HPV infections as absent would result in an overestimation of the median time to clearance of incident oral HPV infections. A further limitation is that HPV viral load was measured only among individuals with at least one positive line blot assay of the corresponding HPV type. Low copy number oral HPV infections below the lower limit of sensitivity of the line blot would not be selected. As these low copy number infections would be expected to have a higher clearance rate than high copy number HPV infection, we may have underestimate the association between viral load and persistence.
Oral HPV prevalence in the study population appeared to be a reasonable approximation of the proportion infected over time. However, it is important to note the potential effect of missing data, which may have caused us to underestimate cumulative prevalence. The difference in cumulative prevalence estimates at two-week and six-month sampling intervals indicate that a sixmonth interval sampling may significantly underestimate the proportion of the study population that has been ''exposed'' and misclassify individuals as ''unexposed''. Therefore, a six-month sampling interval may bias point estimates for factors that increase risk of acquiring an oral HPV infection toward the null.
Despite these limitations, this is the first short term natural history study of oral HPV infection and the first to report results on more than two consecutive visits. Based upon this data, a sixmonth sampling interval is currently being used in our natural history studies of oral HPV infection to evaluate factors affecting type-specific oral HPV persistence.