Clinical Evaluation of INNO-LiPA HPV Genotyping EXTRA II Assay Using the VALGENT Framework

In this diagnostic test validation study, we assessed the clinical accuracy and HPV genotyping performance of the INNO-LiPA HPV Genotyping Extra II (INNO-LiPA) within the VALGENT-3 framework. VALGENT is designed to assess the analytical and clinical performance of HPV tests with genotyping capacity. The VALGENT-3 panel comprised 1300 consecutive cervical cell specimens enriched with 300 samples with abnormal cytology obtained from women attending the Slovenian cervical cancer screening programme. The INNO-LiPA allows type-specific detection of 32 HPV types; however, for the clinical accuracy assessment, we considered it as high-risk (hr)HPV positive when at least one of the following HPV types was present: HPV16, HPV18, HPV31, HPV33, HPV35, HPV39, HPV45, HPV51, HPV52, HPV56, HPV58, HPV59, and HPV68. Clinical accuracy for detection of cervical intraepithelial neoplasia grade 2 or worse (CIN2+) was compared between INNO-LiPA and Hybrid Capture 2 (HC2), which is a standard comparator test for HPV tests used in cervical cancer screening. In addition, hrHPV and type-specific detection HPV types were compared between INNO-LiPA and Linear Array HPV Genotyping Test (Linear Array). The prevalence of hrHPV determined by INNO-LiPA was 17.1% (95% CI, 15.0–19.2%) in the screening population. HrHPV testing with INNO-LiPA had a sensitivity for CIN2+ of 96.9% (95% CI, 92.1–99.1%) which was non-inferior to HC2 (relative sensitivity of 1.01; 95% CI, 0.97–1.04; pn.inf = 0.0002) and a specificity for ≤CIN1 of 85.3% (95% CI, 83.2–87.3%) which was inferior to HC2 (relative specificity of 0.95; 95% CI, 0.93–0.97; pn.inf = 0.9998). Genotyping agreement between INNO-LiPA and Linear Array was excellent for hrHPV, HPV16, HPV18, HPV35, HPV45, HPV58 and HPV59, but good or fair for other HPV types. To conclude, INNO-LiPA demonstrated non-inferior clinical sensitivity but lower specificity compared to HC2 in addition to excellent concordance compared to Linear Array for hrHPV and some genotypes.


Introduction
Over 200 human papillomavirus (HPV) types have been identified and classified based on their nucleotide sequences, with new HPV types being characterized at an increasing rate [1]. Among them, 12 high-risk HPV (hrHPV) types (HPV16, HPV18, HPV31, HPV33, HPV35, HPV39, HPV45, HPV51, HPV52, HPV56, HPV58 and HPV59) are causally linked with cervical cancer and their immediate of any of the 13 hrHPV types by INNO-LiPA. The overall and type-specific prevalence of 13 hrHPV types in the total study population determined by INNO-LiPA is summarized in Table 1. The hrHPV prevalence was 15.2% in women with NILM (negative for intraepithelial lesion or malignancy) and increased to 42.8%, 69.0% and 86.0% in women diagnosed with ASC-US (atypical squamous cells of undetermined significance), LSIL (low-grade squamous intraepithelial lesion) and HSIL(high-grade squamous intraepithelial), respectively. The risk ratio (RR) of HSIL compared to women with NILM was highest (RR > 8.00) in women infected with in HPV16, HPV33, HPV18 and HPV45. Table 1. Overall prevalence of hrHPV (aggregate of 13 types) and of individual hrHPV types detected by INNO-LiPA in the total study population according to baseline cytology.

Clinical Performance of the INNO-LiPA
The accuracy data for the INNO-LiPA and HC2 for the outcomes CIN2+, CIN3+ and ≤CIN1 are shown in Table 2   The relative sensitivity of INNO-LiPA compared to HC2 was 1.01 (95% CI, 0.97-1.04; p mcn = 0.6547; p n.inf = 0.0002) for CIN2+ and 1.01 (95% CI, 0.97-1.06; p mcn = 0.5637; p n.inf = 0.001) for CIN3+. The relative specificity of INNO-LiPA for ≤CIN1 was 0.95 (95% CI, 0.93-0.97; p mcn = 0.0000; p n.inf = 0.0000). Similar results were obtained if the analysis was restricted to women ≥30 years (Table 3). or specificity of the INNO-LiPA assay are not significantly different from that of the HC2. b p for the test for non-inferiority. A sensitivity threshold of at least 90% and a specificity threshold of at least 98% relative to that of the HC2 were applied in a non-inferiority score test. p n.inf < 0.05 means that the sensitivity or specificity of the INNO-LiPa is not significantly lower than that of the HC2.

Genotyping Agreement between INNO-LiPA and Linear Array
In the total study population, concordance between INNO-LiPA and Linear Array was assessed at type-specific level and overall for 13 hrHPV types (Table 4). Overall concordance of the two assays for 13 hrHPV types was 93.0% and the corresponding κ value was 0.805 (95% CI, 0.757-0.0854), indicating excellent agreement between INNO-LiPA and Linear Array. Similarly, the level of agreement was also excellent for detection of HPV16, HPV18, HPV35, HPV45, HPV58 and HPV59. However, for the identification of other individual types, level of agreement ranged from good to poor between the two assays (Table 4). In addition, INNO-LiPA detected more positive cases than Linear Array for all individual types common to both assays.

Discussion
The INNO-LiPA provides full genotyping capability for 32 HPV types. For the purpose of this study, INNO-LiPA was considered positive if at least one of the 13 hrHPV types targeted by HC2 was detected.
To the best of our knowledge, the INNO-LiPA has not been validated previously according to the international guidelines for evaluation of new HPV tests in primary cervical cancer screening settings. Here, we present the first study to evaluate the clinical performance of the INNO-LiPA compared to HC2 using samples from the VALGENT-3 panel. In the whole study population, the INNO-LiPA showed a sensitivity for the detection of CIN2+ and CIN3+ of 97% and 99%, respectively, which was similar to HC2. However, the clinical specificity for ≤CIN1 was only 85%, which was 5% (95% CI, 3-7%) lower than the comparator test.
INNO-LiPA is a SPF10 PCR that targets a short highly conserved region in the L1 gene [21,26]. The small size of the amplicon makes the test analytically very sensitive. However, at the same time, discrimination of the individual types is challenging and complex and it is therefore not so surprising that the clinical specificity is lower compared to HPV tests targeting longer DNA sequences [27]. The small size of the amplicon makes INNO-LiPA particularly useful for testing of archived cell preparations or formaline-fixed-paraffin-embedded tissue blocks stored over long periods where parts of the viral genome can be fragmented [28].
INNO-LiPA provides for each HPV type a qualitative output, which is translated into a positive/negative result. The appreciation of presence or absence of blue lines is not quantifiable. Therefore, adaptation of the cut-off, which may allow a more optimal balance between clinical sensitivity and specificity, is in case of INNO-LiPA not possible.
Excellent analytical agreement between INNO-LiPA and Linear Array was observed for 13 hrHPV types overall, HPV16, HPV18, HPV35, HPV45, HPV58 and HPV59. In addition, INNO-LiPA detected more positive cases than Linear Array for all individual HPV types that are common to both assays, with a positivity rate of hrHPV determined by INNO-LiPA statistically significantly higher than that determined by Linear Array (p mcn < 0.001) in the total study population. The significant difference of positivity rate for hrHPV with these two genotyping methods is consistent with the tests' clinical performances with regard to relative specificity compared to HC2.
In conclusion, in our study, INNO-LiPA exhibits lower clinical specificity; however, this is most likely due to detection of HPV infections with low concentrations and therefore can play an important role in evaluation of viral infection outcomes of vaccination trials and in monitoring the impact of HPV vaccination. Moreover, INNO-LiPA may be useful in epidemiological studies to investigate the prevalence and distribution of HPV types and in studies of the natural history of HPV infection at the type-specific level.

Clinical Specimens
The VALGENT framework is designed to assess the comparative analytical and clinical performance of HPV tests that offer limited to full genotyping capability [29]. VALGENT is iterative, using specimens collected in different countries. The first two VALGENT panels have been completed, using samples collected from Belgium [30][31][32] and Scotland [33][34][35][36]. The third study panel (VALGENT-3) was collated in Slovenia as previously described [11,23,24]. Briefly, 1300 consecutive cervical samples were collected from women who participated in the organised Slovenian national cervical cancer screening program between December 2009 and August 2010 (screening population). The study panel was enriched with 300 cytologically abnormal specimens collected between January 2014 and May 2015 (enrichment population). As required in the VALGENT protocol [29], the enrichment population included 100 women with atypical squamous cervical cells of undetermined significance (ASC-US), 100 women with low-grade squamous intraepithelial lesion (LSIL) and 100 women with high-grade squamous intraepithelial lesion (HSIL).
The sample collection, aliquoting procedure and storage details have been described in detail in previous VALGENT-3 manuscripts [23,24]. Briefly, in July 2016, Ghent University (Ghent Belgium) received 1600 samples of extracted DNA. DNA was extracted from original samples collected into ThinPrep PreservCyt solution (Hologic, Marlborough, MA, USA). Prior to DNA extraction, 1 mL aliquot of original ThinPrep sample was centrifuged at 13,000× g for 15 m with supernatant being discarded and cellular pellet resuspended in 200 µL PBS buffer. DNA extraction was performed using QIAamp MinElute Media Kit (Qiagen, Hilden, Germany) following the manufacturer's instructions.
Bound DNA was finally eluted with 50 µL of ATE buffer and stored at −70 • C prior to further testing. According to the manufacturer's instructions, we have used 10 µL of extracted DNA for INNO-LiPA testing. Similarly, from the second aliquot of original ThinPrep DNA was extracted for Linear Array testing, where 50 µL of extracted DNA was used for further testing.

HC2
HC2 detects 13 hrHPV types (HPV16, HPV18, HPV31, HPV33, HPV35, HPV39, HPV45, HPV51, HPV52, HPV56, HPV58, HPV59, and HPV68) and is accepted as a standard comparator test for the clinical validation of hrHPV DNA assays that may be used for primary cervical cancer screening [15]. For the purpose of the present study, hrHPV positivity for INNO-LiPA was defined as the presence of one or more of the 13 hrHPV targeted by HC2.

Clinical Outcome and INNO-LiPA Performance Assessment
As described in previous VALGENT-3 reports [23,24], cytological assessment and referral of patients with abnormal cytology results to colposcopy were done according to the Slovenian national screening guidelines [37], which are in agreement with European guidelines [38]. Colposcopy-directed punch biopsies were obtained from suspicious areas for final histopathological assessment.
Women with histologically confirmed CIN2+ results were considered as diseased subjects. Due to the fact that women with normal cytological results of negative for intraepithelial lesion or malignancy (NILM) were not referred to colposcopy verification in our study, we considered them as subjects without disease only if they had two or more consecutive NILM cytological results (at enrolment and at subsequent screening between 12 to 48 months later). We used this group of women to compute the clinical specificity for ≤CIN1.
The clinical sensitivity and specificity of the INNO-LiPA for CIN2+ and CIN3+ were calculated. We compared the clinical accuracy of INNO-LiPA to HC2 for CIN2+ and CIN3+, using non-inferiority statistics with a relative sensitivity threshold of 90% and a relative specificity threshold of 98% [39]. The McNemar statistic was used in order to compare the differences between matched proportions [40]. For both statistics, the level of significance was set at 0.05. All analyses were performed using STATA version 14 (Manufacturer, College Station, TX, USA).
Separate and consensus genotyping agreement for the types common to INNO-LiPA and Linear Array was assessed using κ [41] and McNemar statistics [40]. κ values from 0.0 to 0.20, 0.21 to 0.40, 0.41 to 0.60, 0.61 to 0.80 and 0.81 to 1.0 indicate poor, fair, moderate, good and excellent level of agreement between two assays. A McNemar p-value of <0.05 indicates significant discordance between genotyping results determined by the two assays.