|Home | About | Journals | Submit | Contact Us | Français|
We conducted a Linear Array test/retest analysis using cytologic specimens from 198 women. A total of 67.2% of samples had the same human papillomavirus (HPV) types detected in both tests (type-specific positive agreement was 83.3% overall [Kappa = 0.9] and 86.8% for carcinogenic types [Kappa = 0.92]). Discordance was highest with a low hybridization signal strength. Overall, Linear Array was highly reproducible.
Given the strong association between persistent human papillomavirus (HPV) infection and cervical precancer/cancer (1), HPV genotyping is becoming increasingly important for secondary cervical cancer prevention (2–6). Although clinical laboratories use FDA-licensed assays for HPV detection, one of the most common HPV genotyping tests for research use is the Roche Linear Array (LA) HPV genotyping system (7). To evaluate LA reproducibility, we conducted a test/retest analysis using cytologic specimens from the Study to Understand Cervical Cancer Early Endpoints and Determinants (SUCCEED).
As previously described (8), women referred to colposcopy at The Oklahoma University Health Sciences Center (OUHSC) due to abnormal cytological screening results or a biopsy diagnosis of cervical intraepithelial neoplasia (CIN) or invasive carcinoma were enrolled into SUCCEED from November 2003 through September 2007. Exclusions included age of <18 years, pregnancy at visit, previous chemotherapy/radiation, vaginal-only colposcopy, or known HIV positivity. Participants provided written informed consent; the OUHSC and National Cancer Institute institutional review boards approved the study.
We randomly selected 198 of 1,868 women with initial LA results and sufficient material remaining for retesting, including 138 women with multiple HPV infections, 50 with single HPV infections, and 10 with no detectable HPV infection on the initial test; 470 HPV infections were detected in the first and/or second LA tests. Cervical cytologic specimens were collected and processed as previously described (9, 10), and HPV DNA was amplified by PCR using the LA system. An additional 1-ml aliquot from the same sample was processed and tested by LA for the reproducibility analysis. The median time from specimen collection to test 1 was 5.6 months (range, 0.2 to 21.2 months), and the medium time from specimen collection to test 2 was 33.5 months (range, 7.2 to 54.1 months). LA detects 37 HPV types (Table 1). HPV52 is detected only with a mixed probe that simultaneously detects HPV52, 33, 35, and 58, while HPV33, 35, and 58 are also detected by individual probes. Thus, 32 women who were positive for both the mixed probe and an individual probe for HPV33, 35, and/or 58 were excluded from the HPV52 analyses because it is impossible to know if HPV52 was present or absent in these cases.
Each LA genotyping run, including three HPV16-positive controls and one HPV-negative control, was conducted according to the manufacturer's instructions except that 10 μl of eluted patient DNA and 40 μl of water were used in each reaction instead of 50 μl of DNA. Results were reported only if DNA hybridized to both high- and low-concentration β-globin probes (indicating adequate DNA amplification) and were considered positive only for unambiguous, continuous hybridization bands. The intensity of each hybridization band was visually classified as strong, moderate, weak, very weak, or extremely weak as previously described (11), correlating well with quantitative-PCR-based viral load (12).
A total of 133 women (67.2%) had exactly the same HPV types detected in a retest of a second aliquot of their liquid-based cytology specimen, including nine of 10 samples (90.0%) from women who initially tested negative for all HPV types (HPV31 was detected with moderate signal strength on retest in one initially HPV-negative sample), 42 of 50 women (84.0%) with a single HPV type detected, and 82 out of 138 women (59.4%) with multiple types detected (not unexpected, since the more types that are present, the more likely it is that at least one type will be missed). The sociodemographic and sexual behavior characteristics were similar for women with a single HPV type and women with multiple HPV types (data not shown).
Positive agreement (the number of women positive on both test 1 and test 2 divided by the sum of women positive on either test) was similar among women with just one HPV type on initial analysis and women with multiple HPV types for the most common HPV types, HPV16 (92.6% and 88.9% for single and multiple infections, respectively) and HPV18 (88.9% and 100.0% for single and multiple infections, respectively). For less common HPV types, there was more variation, likely due to random error given the small numbers of infections (see Table S1 in the supplemental material). For all 198 women combined, positive agreement for carcinogenic HPV types, as defined by the International Agency for Research on Cancer (IARC) (13), ranged from 68.0% for HPV51 to 100.0% for HPV33 and 56 (Table 1). For noncarcinogenic HPV types, combined positive agreement ranged from 0.0% for HPV26, 64, and 71 to 100.0% for HPV11, 62, 69, 72, 73, and IS39 (also called 82v).
The overall type-specific positive agreement for individual specific HPV genotypes was 83.3%, with a Kappa of 0.90 (95% confidence interval [CI], 0.88 to 0.92) (Table 2). Type-specific positive agreement for carcinogenic HPV types was 86.8%, while for noncarcinogenic HPV types, it was 77.9% (Kappa values of 0.92 [95% CI, 0.90 to 0.95] and 0.87 [95% CI, 0.83 to 0.91], respectively). HPV16 and 18 had 91.1% type-specific positive agreement (Kappa = 0.94 [95% CI, 0.90 to 0.98]). The α9 clade (including HPV16) had the highest positive agreement (90.2%; Kappa = 0.94 [95% CI, 0.91 to 0.97]). At the woman (non-type-specific) level, 194 women were concordant for HPV status (positive for any HPV versus negative for all HPV types; positive agreement = 97.9%; Kappa = 0.81 [95% CI, 0.62 to 0.99]).
Most changes from positive to negative results and vice versa between the first and second tests occurred when individual HPV hybridization signal intensity was ranked at or below weak. The majority of HPV types with at least moderate signals on the initial LA test had >80% positive agreement, whereas positive agreement was usually ≤50% when HPV type-specific signals were at or below weak on the initial test (Table 1). For carcinogenic HPV infections, positive agreement tended to be slightly more variable for HPV infections detected in women with <CIN2 than for infections in women with ≥CIN2, but more infections were detected in women with <CIN2 than in women with ≥CIN2 (data not shown).
Previous studies that have compared LA to other HPV genotyping tests have generally found that LA performs well, especially in the context of multiple-HPV genotype infections (14–17). In a previous evaluation of LA and line blot assay (14), the positive agreement for carcinogenic HPV was 80% (Kappa = 0.76). Similarly, positive agreement for carcinogenic HPV on LA compared to SPF10 genotyping was 79.3% (18). Thus, the positive agreement for carcinogenic HPV types with repeat LA testing (86.8%; Kappa = 0.92) is as good as or better than positive agreement between LA and other genotyping tests.
Another epidemiologic study of LA reproducibility found 98.2% agreement for non-type-specific HPV (sample positive for HPV versus negative for all HPV types) (19), similar to the 97.9% agreement in our study. The percentage of samples with exact agreement of HPV types between tests was higher in the previous study (83.0%) than in the current study (67.2%), a finding which is unsurprising given that our study oversampled women with multiple HPV infections (increasing the risk of disagreement). Consistent with our findings, discrepancies in the previous study generally occurred when the signal was very weak. Importantly, the previous study retested residual DNA extracts while we used separate aliquots and extracted DNA twice, allowing us to evaluate differences in results from separate samples collected at the same time from the same woman.
In conclusion, this study provides evidence that type-specific results from LA are reproducible, especially for vaccine-related types and other carcinogenic HPV types. Discrepancies most often occur when the signal strength is low. These results have two major implications. First, they provide reassurance regarding extant results from LA, one of the most common HPV genotyping tests used in epidemiological studies, and second, they suggest that LA can be used to reliably test for individual HPV types in future studies.
This research was supported by General Funds from the Intramural Research Program of the National Institutes of Health, National Cancer Institute, Division of Cancer Epidemiology and Genetics.
We thank the laboratory personnel of the Surgical Pathology and Cytopathology Laboratories of The University of Oklahoma Medical Center for their conscientious attention to specimen processing and Pap test interpretation.
Mark Schiffman receives HPV testing from Roche, including Linear Array and Cobas testing, for collaborative research at no cost.
Published ahead of print 28 November 2012
Supplemental material for this article may be found at http://dx.doi.org/10.1128/JCM.03036-12.