Recruitment of cases and controls

An ongoing case-control study on lifestyle factors, infection and cancer began in 1995 in the main public referral hospitals for cancer treatment in greater Johannesburg, South Africa. Infectious agents already tested include HIV [13], human herpes virus 8[14], and human herpesviruses 1 to 6 [15]. All newly diagnosed black African adults with cancer are eligible for inclusion. Patients are asked whether they have previously had a cancer and, where possible, this is verified by checking records. In about 80% of cases the diagnosis and primary site were confirmed by histology, haematology, or cytology [13]. During the time period for this study (1995 – 2000) selected patients from the same hospitals with a new diagnosis of cardiovascular disease were also included. Following written or witnessed oral informed consent, patients were interviewed by trained nurses using a two-page demographic and lifestyle questionnaire[16]. Interviews were conducted in the main vernacular languages. Blood specimens were collected prior to initiation of treatment. Sera were separated, aliquotted and stored at minus 30°C.

Sera from subjects were batch tested for HIV-1 by the South African National Health Laboratory Service Serology Laboratory using standard serological methods described previously [13]. The current analysis used questionnaire data and serum samples from 3757 HIV seronegative subjects in order to determine the relative importance of HPV-16 in relation to risk of a number of cancer types. Cases were subjects with cancer of the cervix (n = 946), other anogenital organs (61 females, 19 males, comprising 10 anal, 51 vulval, 6 vaginal and 13 penile cancers), cancers of the oral cavity and pharynx (21 females, 81 males, comprising 20 cancers of the tongue, 25 of the floor of mouth, 8 of the gum, 10 of the parotid/salivary glands, 6 of the tonsil, and 11 of the oro-naso pharynx, and 22 cancers in other regions of the oral cavity), cancer of the oesophagus (126 female, 243 males) and cancer of the prostate (205 males). The comparison group consisted of 2,055 age and sex-matched patients randomly selected from the same data base, diagnosed at the same hospitals, but with a vascular disease or with a cancer type thought to be unrelated to HPV infection. These comprised 1382 female and 673 male subjects with breast cancer (506 females), colorectal cancer (60 females, 51 males), endocrine cancer (19 females, 9 males), Hodgkin lymphoma (37 females, 37 males), leukaemia (104 females, 103 males), liver cancer (9 females, 13 males), lung cancer (34 females, 99 males), myeloma (53 females, 65 males), non-Hodgkin lymphoma (27 females, 40 males), stomach cancer (27 females, 36 males), other minor cancer types (132 females, 127 males) and with cardiovascular disease (374 females, 93 males).

HPV-16 serology

Serum aliquots from all participants were air freighted in dry ice to the University of Cape Town, Division of Medical Virology for anti-HPV-16 IgG testing by ELISA. The ELISA was a modification of that described by Studentsov and colleagues [17, 18], using polymer solutions (Sigma) for blocking (polyvinyl alcohol, PVA) and secondary antibody (polyvinylpyrrolidone, PVP) enhancement. This enhanced ELISA uses serum at a 1:100 dilution instead of a previous 1:20 dilution [19]. Plates were coated with 100 μl of 0.05 μg virus-like particles (VLP) to HPV-16, (MedImmune Inc) in phosphate buffer solution (PBS) overnight at 4°C. After washing twice in PBS, plates were blocked with 200 μl 0.5% PVA in PBS (0.5% PVA) at room temperature. After washing six times in PBS, serum diluted 1:100 in 0.5% PVA was added and allowed to incubate at 37°C for one hour. The subsequent procedure used 0.8% PVP and 0.5% PVA in PBS for the dilution of the horse radish peroxidase (HRP) conjugated anti-human IgG with six PBS washes between this and other procedures [19].

Minimising laboratory variation

To minimise laboratory variation and case-control bias, serum aliquots from the cases and the comparison group were randomly allocated on each ELISA plate by five age groups (25–34, 35–44, 45–54, 55–64, and 65–74 years) and by case-comparison status. Specimens from males and females were run on separate plates. The laboratory staff were 'blinded' with regard to the distribution of the aliquots on the plates. In addition, for quality control purposes, ten adult consenting volunteer donors provided anonymous blood samples to use as plate controls. These donors were tested for anti-HPV-16 IgG antibody responses prior to the commencement of this study, and the sera of three donors who were found to have low, medium and high anti- HPV-16 IgG optical density values were added on wells on each of the 111 ELISA plates. As a secondary precaution, pooled sera of children collected for other purposes were added on wells of each ELISA plate to represent an additional reference group. Children's anti-HPV IgG titres are usually low, and therefore could represent a true negative (uninfected) group [2]. Intra-class correlation coefficients were estimated in the volunteer and children's sera to measure the plate-to-plate assay variability.

Determination of cut-off values for negative, medium and high antibody levels

The ELISA optical density (OD) cut-off for positivity was 0.45, determined by calculating the mean plus 4 standard deviations (SD) of the absorbance values obtained from children's serum samples tested by VLP-16 IgG ELISA [20]. OD values between 0.45 and 0.767 were considered medium antibody levels and OD values greater than 0.767 were considered as high. The value of 0.767 was the median OD value in the seropositive comparison group. Due to small numbers, cancers of the male and female anogenital organs (aside from cervix) were only divided into HPV-16 IgG seropositives (OD ≥ 0.45) and seronegatives (OD <0.45).

Statistical methods

Case-comparison selection

This study is part of an ongoing cancer case control study run in the two major tertiary hospitals in Johannesburg, where the degree of diagnostic accuracy is high. Controls (comparison group) were selected from the pool of incident cancer patients and patients admitted for the fist time with vascular disease. In this study the controls were selected by excluding subjects whose cancer type was thought associated with the exposure of interest, e.g. those with squamous cell skin cancers and laryngeal cancers. We also excluded subjects with endometrial and ovarian cancer to avoid misclassification of the genital cancers of interest. Whilst a number of co-factors are thought to be important in the aetiology of the cancer types of interest (e.g. smoking), these were not part of the main hypothesis and so smoking related cancers were retained in the comparison group.

HPV-16 serology

Virus-like particles (VLPs) engineered from various HPV types and used in enzyme-linked immunoassay (ELISA) have been validated as type specific antigens to detect serological markers of past and present HPV infection [23] Studentsov et al[18] indicated a twofold increase in sensitivity and specificity of the serological detection of anti-HPV-16 IgG antibodies using a polymer enhanced HPV-16 VLP ELISA protocol compared to previous HPV16 VLP ELISAs. As well as the improved detection of anti-HPV-16 IgG antibodies, the polymer-based ELISA also detects cross-reactive responses from closely related types to HPV-16 of alpha species A9 (containing HPV types HPV -31, -33, -35, -52 and -58) [24]. Williamson et al [25] found that in Cape Town, South Africa, the prevalence of HPV genotypes in cervical cancers was 46% for HPV-16, 2% for HPV-18, 6% for HPV-31, and 6% for HPV-33. Pegoraro et al [26] examined cervical tumour tissue from Zulu speaking black women in Durban, South Africa and found prevalences of 47% for HPV-16, 14% for HPV-18 and 25% for other HPV types. In another South African study of women with cervical or pre-invasive cancer [27], the prevalence of oncogenic genotypes in cervical biopsies was: HPV-16: 47%, HPV-18: 14%, HPV-33: 10%, HPV-35: 2%, HPV-52: 2% and HPV-58: 1%. However, the prevalence of types other than HPV-16 is expected to be low as responses to species A9 types other than HPV-16 would be to minor VLP epitopes [28] and so the antibody response in this study would be predominantly due to HPV-16.

Viral persistency and high viral loads are associated with an increase in the risk of cervical and anogenital cancers[29–31]. Anti-HPV IgG VLP ELISA optical density (which is proportional to the amount of antibody present) was found to be a reliable marker of past exposure to the HPV type tested and an indirect marker of high HPV viral loads measured using PCR[18].

In this study, a number of measures were introduced to minimise laboratory variation and bias. These included: i) using individual sera from a group of local volunteers to assess the reliability of the assay at low, moderate and high anti-HPV 16 IgG optical density levels; ii) adding a group of children's sera as a cross check for negative control values; iii) randomly allocating cases and the comparison groups on each plate to minimise case control bias; iv) 'blinding' the laboratory so that the assays were performed without knowledge of the disease or demographic status of the subjects; and v) testing cancers that are known to be associated with HPV (cervix, anogenital) and those with an uncertain association (oral cavity and pharynx, oesophageal, prostate) using the same assay, under the same laboratory conditions.

Whilst there is no gold standard to evaluate whether the anti-HPV-16 IgG optical density cut-off levels truly reflect infection by HPV 16, a cut-off of 4 SD above that found among healthy children in a previous study (>0.45) seemed appropriate for the purposes of this study [19]. This is similar to the approach used by Wang et al [32], who used 5SD as an appropriate cut off point as a proxy of persistent HPV infection. Using tertiles of anti-HPV IgG optical density level yielded similar results. Approximately 60% of the subjects in the comparison group (63% in females, 58% in males) had values above the cut off level.