Methods

Study population

The Ludwig-McGill study cohort study of HPV natural history recruited women at maternal health clinics in São Paulo, Brazil, between 1993 and 1997. Details of the study design have been published elsewhere.20 Eligible women had to be between 18-60 years old, permanent residents of São Paulo, not currently pregnant, with an intact uterus, and with no history of treatment for cervical disease. During the first year, women were followed-up every 4 months; at each visit they provided cervical specimens for HPV genotyping and underwent an interview administered by study nurses collecting data on sociodemographic information, reproductive health, sexual behaviors, smoking, and diet. Women provided signed informed consent to study participation. The study was approved by the ethical review boards of all participating institutions: McGill University, Montreal, Canada; University of Toronto, Ontario, Canada; and the Ludwig Institute for Cancer Research and the Hospital Maternidade Vila Nova Cachoeirinha, both in São Paulo, Brazil.
The current analysis is restricted to a random subset of 505 women sampled from the full cohort who had a follow-up visit within 10 days of the 1-year follow-up date, had valid samples (either β-globin or HPV-positive samples) at the baseline and 1-year visits, and who had complete questionnaire and HPV genotyping data. Our previous publication details the selection process, baseline characteristics, and the betapapillomavirus prevalence at baseline (1st visit) and 1-year (4th visit) in this subset of women.19 For the current study, we extended genotyping to the 4 month (2nd visit) and 8-month (3rd visit) samples collected from these women in order to measure betapapillomavirus incidence rates over the first year of follow-up and increase the number of observations for analyses.

HPV genotyping

Exfoliated cervical cells were digested with 100 µg/ml proteinase K for 3-18 hours at 55°C, and DNA was obtained by spin column chromatography. Samples were tested for alphapapillomaviruses by polymerase chain reaction (PCR) amplification using MY09/11 or PGMY09/11 generic primers, and genotyped using hybridization with HPV type-specific oligonucleotide probes or restriction fragment length polymorphism analysis. This methodology allowed the identification of potentially more than 40 genital alphapapillomavirus genus types from the following species: α-1: HPVs 32, 42; α-3: HPVs 61, 62, 72, 81, 83, 84, 89; α-4: HPV57; α-5: HPVs 26, 51, 69, 82; α-6: HPVs 53, 56, 66; α-7: HPVs 18, 39, 45, 59, 68, 70; α-8: HPV40; α-9: HPVs 16, 31, 33, 35, 52, 58, 67; α-10: HPVs 6, 11, 44; α-11: HPVs 34, 73; α-13: HPV54; and α-14: HPV71.21
The presence of betapapillomaviruses was assessed by a type-specific, multiplex genotyping PCR assay followed by genotyping via a bead-based Luminex technology.22 This assay distinguishes 43 betapapillomavirus genus types from the following species: β-1: HPVs 5, 8, 12, 14, 19, 20, 21, 24, 25, 36, 47, 93, 98, 99, 105, 118 124, 143; β-2: HPVs 9, 15, 17, 22, 23, 37, 38, 80, 100, 104, 107, 110, 111, 113, 120, 122, 145, 151; β-3: HPVs 49, 75, 76, 115; β-4: HPV92; and β5: HPVs 96, 150. In this assay, results are expressed as the median fluorescence intensity (MFI) of at least 100 beads per bead set. For each probe, the MFI values obtained when no PCR product was added to the hybridization mixture were considered the background values. Different bead preparations impact MFI values; because the samples from visits 2 & 3 were tested using a different preparation of beads than the samples from visits 1 & 4, a different MFI cutoff value was used for these visits. Based on MFI values from positive and negative controls reads, cutoffs were computed by adding 20 MFI to 1.1X the median background values in samples obtained in visits 1 and 4;20 and by adding 50 MFI to 1.1X the median background values for samples obtained in visits 2 and 3, with the exception of HPV100 for which the cutoff was computed by adding 100 MFI to 1.1X the median background values for visits 2 and 3.

Statistical analysis

All analyses were type-specific, with HPV type as the unit of observation. The prevalences of betapapillomaviruses were summed across the four visits to derive a time-averaged measure of prevalence over the first year of the study. The incidence rates of betapapillomaviruses were calculated in women who were negative for that specific betapapillomavirus type at the previous visit, with the denominator being the time between visits. We pooled results by summing observations over all HPV types.
To assess whether betapapillomaviruses and alphapapillomaviruses occur more frequently together than expected, we used permutation tests. Permutation tests work by performing rearrangements of the observed data without replacement (permutations).23 The permutations allow deriving an expected distribution of betapapillomaviruses and alphapapillomaviruses under the assumption that infections are independently distributed across individuals. Because there was evidence of within-genus clustering, we implemented two versions of the permutation test (Figure 1). The first test assumed full independence of all HPV types; in this case, the results for each individual HPV type are permutated across participants, with each HPV type resampled separately. This first test will give the expected distribution of type-specific detections assuming all HPV types are fully independent, both across and within genera. The second test assumed independence between papillomavirus genera but not within genus; in this case, the set of results from all HPV types within a genus from the same person are permutated across participants, with each genus resampled separately. This second test will give the expected distribution of type-specific detections assuming that alphapapillomaviruses are independent from betapapillomaviruses and vice-versa, but accounting for the tendency of HPV types from the same genus to cluster together. We performed 2500 permutation resamples for both tests. We divided the observed number of co-detections by the mean number of co-detections in the permutation resamples to obtain observed over expected values. The 95% confidence intervals were obtained using the 2.5-97.5 percentiles of the permutation resample distribution.
To assess the cross-sectional association between alphapapillomavirus and betapapillomavirus prevalence, we calculated odds ratios (OR) using random effects logistic regression models, with a woman-level random intercept. To assess the prospective association between alphapapillomavirus and betapapillomavirus incidence, we calculated hazard ratios (HR) using Cox proportional hazard regression models, using the robust sandwich estimate of the covariance to account for multiple HPV types per woman.24 All models were fitted separately for alphapapillomavirus and betapapillomavirus types and used the HPV type as the unit of analysis. To assess interactions between genera, the models included as predictors whether the woman was positive for any HPV type of a different genus at either the same visit sample (logistic regressions) or the previous visit sample (Cox regressions). Multivariable models were adjusted for age and sexual behaviors reported since the previous visit. Sexual behavior variables were included as a priori predictors because our previous publication had found that lifetime number of sex partners might be inversely associated with betapapillomavirus prevalence.19