DNA methylation testing with S5 for triage of high‐risk HPV positive women

Abstract Methylation of host and viral genes is promising for triage of women with high‐risk human papillomavirus infections (hrHPV). Using a population‐based sample of hrHPV positive women with cervical biopsies within 12 months after cervical screening, the clinical value of the S5 methylation classifier (S5), HPV genotyping and cytology were compared as potential triage tests, for outcomes of cervical intraepithelial neoplasia (CIN) grade 3 or greater (CIN3+), CIN2+ and CIN2, and the area under the curve (AUC) calculated. S5 scores increased with histopathology severity (P trend < .001). For CIN3+, the AUC was 0.780 suggesting S5 provides good discrimination between <CIN3 and CIN3+. AUCs were significant for all pairwise comparisons of <CIN2, CIN2 and CIN3+ (P < .001). The positive predictive value (PPV) of HPV16/18 genotyping for women with any abnormal cytology was greater than S5 (25.36% vs 20.87%, P = .005) for CIN3+, while sensitivity was substantially greater for S5 (83.33% vs 59.28%, P < .001). Restricting to women with abnormal cytology, but excluding those with high‐grade cytology, both S5 and HPV16/18 provided CIN3+ PPVs high enough to recommend colposcopy. Triage with S5 also appeared useful for hrHPV positive women negative for HPV16/18 (CIN3+ PPV: 7.33%, sensitivity: 57.52%). S5 provided increased sensitivity for CIN3+ compared to HPV16/18 genotyping for hrHPV positive women, overall and when restricted to women with abnormal cytology, suggesting S5 may improve colposcopy referral. S5 also has the ability to distinguish between <CIN2, CIN2 and CIN3+, a finding of importance for managing CIN2, given the complexity and uncertainty associated with this diagnosis.


| INTRODUCTION
It is well established that human papillomavirus (HPV) testing is a more sensitive screening test than cytology. It allows earlier diagnosis of high-grade disease, and is more effective at preventing invasive cervical cancers. [1][2][3] However, HPV testing is less specific than cytology and most HPV positive women have transient infections which will regress naturally. 4 Thus many colposcopy referrals and associated cervical excisional treatments are unnecessary and could be reduced with better triage tests. Conversely, triage also produces decisions about which women with HPV infection and/or abnormal cytology do not need referral to an expert clinician, which can result in loss to follow-up and undetected cancers.
Various triage strategies have been suggested, but currently no optimal approach has been identified. Previously, cytology was the primary screening method in many countries, with cytology results of high-grade squamous intraepithelial lesions or greater (HSIL+) or atypical squamous cells, cannot exclude HSIL (ASC-H) recommended for referral to immediate colposcopy. 5 However, for women with less severe cytological abnormalities, including low-grade squamous intraepithelial lesions (LSIL) or high-risk HPV (hrHPV) positive atypical squamous cells of unknown significance (ASC-US), the best management approach remains uncertain. Referral of all women with LSIL or hrHPV positive ASC-US cytology is not efficient; the proportion of women found to have cervical intraepithelial neoplasia (CIN) grade 2 or greater (CIN2+) at colposcopy has varied significantly, with generally only a small proportion of colposcopies showing detectable high-grade disease. [6][7][8] Triage tests for women with hrHPV infections that have greater accuracy, reliability and reproducibility remain an important challenge.
Cytology, HPV genotyping, HPV viral load, immunocytochemistry using p16 ink4a alone or p16/ki-67 dual-staining, as well as DNA methylation, have been suggested as potential triage tests for hrHPV positive women. [9][10][11][12][13][14] High levels of DNA methylation have been shown to be associated with persistent HPV infection, more severe precancerous lesions and an increased risk of cervical cancer. [15][16][17][18][19][20] Measuring DNA methylation of host and viral genes at specific CpG sites has emerged as a promising approach for distinguishing between potentially progressive CIN2/3 lesions and those likely to regress. 18,[21][22][23][24] In a recent meta-analysis including 43 studies of 16 336 women, DNA methylation of multiple human and viral genes, especially the HPV L1 and L2 gene regions of HPV16, were found to be significantly increased in women with both CIN2+ and CIN3+ biopsies compared to those with ≤CIN1, and provided an increased sensitivity and similar specificity, compared to ASC-US or worse cytology (ASC-US+). 25 Molecular triage tests are advantageous as they are less subjective and can be performed on multiple specimen types, including vaginal self-samples 26 and urine, 27 as morphologically intact cells are not required. However, to-date no single gene, human or viral, has shown high enough sensitivity to be the sole triage marker. Identifying an optimum panel of markers remains a key area of interest. We have previously developed a DNA methylation classifier (S5) based on target regions of the human gene EPB41L3, and HPV late gene regions (L1, L2) of HPV16, HPV18, HPV31 and HPV33. [19][20][21]28 S5 has shown promise for the accurate detection of CIN2+, CIN3+ and cancers in many studies worldwide including Canada, China, Colombia, Finland, Mexico and the United Kingdom. 15,16,20,[29][30][31] Here we conducted the first population-based study in the United States evaluating the clinical value of the S5 DNA methylation classifier, compared to liquid based cytology (LBC) or HPV genotyping, for triage of hrHPV positive women.

| Study population
Women attending routine cervical screening by cytology or a combination of cytology and HPV testing in New Mexico, with and without cervical biopsies taken within 12 months of a screening cytology were identified at three major laboratories serving New Mexico residents between June 2014 and December 2015 (n = 128 649). Women were aged 17 to 82 years. Although no statewide organised screening program exists, coverage has been shown to be reasonably high. 32 Inclusion criteria included women whose screening cytology immediately preceding their biopsy was hrHPV positive for one or more of 13 hrHPV genotypes (HPV16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59 and 68; N = 4112). Cytology and histology classifications were obtained from the New Mexico HPV Pap Registry (NMHPVPR) and were based on community laboratory results and pathologist diagnoses. Technicians performing methylation were blinded to cytology and histology results.
A stratified sample of LBC specimens was selected to overrepresent women who developed high-grade lesions (CIN2+). In total 798 LBC specimens from hrHPV positive women were selected for S5 DNA methylation. A further 159 LBC specimens from women positive for one of seven hrHPV types (HPV16, 18,31,33,45,52 and 58) and cytology negative who were not biopsied within 12 months were also tested using S5 to further examine specificity. Figure S1 shows the consort diagram and distribution of histologic findings for the population and the sample. To account for the selection bias towards high-grade lesions in the sample, sampling weights were applied so that the results better represent the entire biopsied screening population.

| HPV genotyping
hrHPV positivity was determined using the Linear Array HPV  28 Amplification was performed using the PyroMark PCR kit (QIAGEN, Germany) with 20 ng input of converted DNA in a 25 μL volume following manufacturer's instructions. The PCR products were pyrosequenced using a PyroMarkQ96 ID (QIAGEN, Germany) instrument. All pyrosequencing runs included negative and positive controls of known methylation levels (0%, 50% and 100%) to allow standardised direct comparisons between different primer sets and all runs were replicated twice.

| Statistical analysis
The predefined S5 DNA methylation classifier was calculated as Analyses were re-weighted to represent all women attending routine cervical screening in New Mexico in the study time period who were biopsied within 12 months based on histologic diagnoses ( Figure S1). All estimates, confidence intervals and P-values were based on adjusted analyses to account for re-weighting and McNemar      Figure 1 shows the distribution of S5 scores ( Figure 1A) and HPV genotyping ( Figure 1B)  HPV16 positive compared to all other hrHPV types (P < .001, Figure S2). The S5 score was not significantly related to age (unadjusted P = .96, adjusted for histology P = .68), with a median value of 0.71 for women <30 years, 0.75 for women 30 to 45 years and 0.85 for those aged 46 or more years.
In Figure 2

| Comparison with other potential triage tests
The diagnostic accuracy of S5 methylation for women with abnormal cytology was compared to cytology and genotyping (with five groups; HPV16, HPV18, HPV16/18, HPV31/33 and other hrHPV) for CIN3+, CIN2+ and CIN2 endpoints ( Table 2). Except for ASC-US+ cytology, which by design occurred in almost all women in this population, sensitivity was substantially higher for S5 than for any other triage test.
PPV and specificity increased with increasing cytological grade (Table 2). This is further illustrated in Figure 2, where the point esti- However their combined sensitivity for CIN3+ was substantially lower than seen for S5 (58.97% vs 83.33%, P < .001, Figure 2).
For women with abnormal cytology, 28.23% were HPV16/18 positive, 12.75% were HPV31/33 positive (and HPV16/18 negative) and 59.02% were positive only for other hrHPV types (   15,20,29,30 The previously validated 0.8 cut-off score provided higher detection rates for all histological outcomes compared to a 1.4 cut-off. Generally, we seek to maximise the detection of cancers and CIN3, and minimise the detection of normal, CIN1 and CIN2; this process requires careful selection of the cut-offs in a population-dependent manner. Populations with higher HPV prevalence and less screened women require higher cut-offs. 35 While high positivity for women with negative and CIN1 diagnoses is undesirable, the sensitivity for CIN2 vs <CIN2 was greater for a 0.8 cut-off, and there was a relatively small reduction in specificity compared to the cut-off at 1.4.

| Combining triage markers
Importantly, our results support that S5 may be able to distinguish between CIN2 and CIN3/AIS/cancer, which has also been reported by our group previously. 16 While many studies have shown very high sensitivity and specificity for cancer using S5, there were too few cancers cases in our sample (n = 8) to make any strong inferences.
In our study, for women with any abnormal cytology, the S5 classifier had substantially greater sensitivity for CIN3+ than HPV16/18 detection (83.33% vs 59.28%), and only a slightly lower PPV (20.87% vs 25.36%), consistent with the lower specificity, although the PPV for CIN3+ was well above the 5% level deemed to be appropriate for referral to colposcopy in the United States. 36  F I G U R E 2 ROC curve for the S5 classifier for CIN3+ vs <CIN3, CIN2+ vs <CIN2 and CIN2 vs <CIN2 endpoints restricted to biopsied hrHPV positive women with abnormal cytology, re-weighted to represent hrHPV positive biopsied women (n = 3797). See Table S1 for AUC values for other comparisons. Point estimates are also given for HSIL+ or ASC-H cytology, HPV16 and HPV18 genotyping (separately) and S5 at 0.8 and 1.4 cut-offs. Receiver operator characteristic (ROC) curves showing the diagnostic ability of S5 methylation for outcomes CIN3+, CIN2+ and CIN2 separately. Sensitivity and specificity are also plotted for HSIL+ or ASC-H cytology, HPV16 and HPV18 genotyping (separately) and S5 at 0.8 and 1.4 cut-offs. AUC is calculated for outcomes CIN3+, CIN2+ and CIN2. HPV16/18 genotyping (84% vs 58%), and similar specificity for <CIN2 (65% vs 71%). 20 Other methylation markers have also been evaluated as triage tools in cervical screening and most have also shown increased sensitivity and specificity compared to HPV genotyping or cytology triage. [37][38][39][40] However, due to differences in gene panels used, and study designs, direct comparison of methylation classifiers is complicated. In a subset of samples from the POBASCAM trial, methylation of the promotor regions of CADM1 and MAL achieved a sensitivity of 84.2% and specificity of 52.5%, which was very similar to that achieved by considering either abnormal cytology or HPV16/18 positivity as grounds for referral (sensitivity 84.2%, specificity 54.0%). 37 In a worldwide study in women with invasive cervical cancer, the effectiveness of the QIAsure methylation assay (QIAGEN, Germany) for the FAM19A4/miR124-2 genes showed very high sensitivity for cervical cancer, 26,41 with nearly all carcinomas, including rare histological types and hrHPV negative carcinomas identified. 42 This provides promise that women negative for methylation markers have a very low chance of having cervical cancer. Another DNA methylation biomarker panel, the GynTect test (Oncgnostics, Germany) 43 has shown modest CIN3+ sensitivity (64.8%) and good specificity (94.6%). 44 In our study, S5 provided good discrimination between CIN3+ and <CIN3 biopsies, consistent with findings from previous studies. 15,20 In a UK based screening study of hrHPV positive women, the AUC for CIN3+ was 0.84 20 and in the HPV FOCAL trial it was 0.83. 15 We observed even greater discrimination between CIN3+ and lowgrade (CIN1) or negative biopsies.
The biological and clinical meaning of a CIN2 diagnosis remains not well understood, and the value of detecting CIN2 is uncertain. 45 Most CIN2 cases regress naturally making it difficult to decide on the best course of action, especially in younger women. 4 CIN2 diagnoses also suffer from a lack of reproducibility. 46 We found that there was a significant difference in AUCs for CIN2 and CIN3+ endpoints, supporting that CIN2+ as a clinical endpoint is a composite of different clinical entities and is potentially misleading. The performance of S5 to detect CIN2 was low compared to CIN2+ indicating the latter is strongly influenced by CIN3+. Using S5 methylation as a triage test may have an advantage over genotyping or cytology in that it may have the ability to better differentiate between progressive vs regressive CIN2, where optimal management may be different. 16 This is an important area for future research.
There is clear evidence, seen here and elsewhere, that women with HSIL+ and ASC-H cytology should be referred to colposcopy.
However, for women with lesser cytologic abnormalities (ASC-US and LSIL), appropriate management is less clear. In this population, the PPV for CIN3+ was high enough for women positive for either HPV16/18 (13.15%), HPV31/33 (8.22%) or S5 methylation (10.40%) that immediate referral to colposcopy would be recommended based on the previously suggested 5% threshold. 36 However, while the PPV for HPV16 or HPV18 positivity was similar to that for S5 positivity, T A B L E 3 Positive predictive values (PPV) for combinations of HPV genotyping and S5 methylation restricted to biopsied women with abnormal cytology excluding high-grade cytology (HSIL+, AGC and ASC-H), for CIN3+, CIN2+ and CIN2 endpoints, re-weighted to represent hrHPV positive biopsied women (n = 3014)  48 Further studies are needed for hrHPV positive women with negative cytology and to follow-up women with abnormal cytology or hrHPV who did not attend colposcopy within 12 months. Including more cases of cancer would also have been of value. Linear Array was the genotyping assay used in the NMHPVPR for research purposes. Although Linear Array is not a Food and Drug Administration (FDA) approved assay, it has shown good correlation with gold-standard methods, 49,50 and strong concordance with clinical HPV data in our study when available (Table S2). Nevertheless, confirmation of these results in another screening population where the HPV test was performed by an assay approved by the FDA is desirable.

| CONCLUSIONS
While HPV vaccination is likely to help reduce the incidence of HPV related disease, the full benefits of this are still decades away. Even in favourable circumstances, vaccination coverage is rarely above 70%, so cervical screening will still be needed. However, disease rates and HPV16/18 infections will be lower in vaccinated populations, so highly specific triage tests become even more important. In this regard it is encouraging to note that S5 testing of HPV positive women who were not positive for HPV16/18 had a relatively good PPV and sensitivity for CIN3+. Methylation of many different genes and panels has shown promise as triage tests for cervical pre-cancer, especially in hrHPV positive women. S5 methylation provided a similar PPV and significantly greater sensitivity for CIN3+ than HPV genotyping or cytology triage, safely enabling a reduction in the number of unnecessary colposcopy referrals. The S5 classifier measured from a LBC specimen has also shown an ability to better distinguish between <CIN2, CIN2 and CIN3+ in subsequent biopsies, a finding of importance for managing CIN2, given the complexity and uncertainty associated with this diagnosis.

AUTHOR CONTRIBUTIONS
The work reported in the article has been performed by the authors, unless clearly specified in the text. Attila T Lorincz, Jack Cuzick,