Effective methylation triage of HPV positive women with abnormal cytology in a middle‐income country

The S5‐methylation test, an alternative to cytology and HPV16/18 genotyping to triage high‐risk HPV‐positive (hrHPV+) women, has not been widely validated in low‐middle‐income countries (LMICs). We compared S5 to HPV16/18 and cytology to detect cervical intraepithelial neoplasia Grade 2 or worse (CIN2+) and CIN3+ in hrHPV+ women selected from a randomized pragmatic trial of 2661 Colombian women with an earlier‐borderline abnormal cytology. We included all hrHPV+ CIN2 and CIN3+ cases (n = 183) age matched to 183

positive Colombian women. Furthermore, S5 triage had comparable sensitivity and significantly fewer false positives than cytology and HPV16/18 combination. In 2018, the World Health Organization issued a call for action to eliminate cervical cancer by a comprehensive approach that includes increasing HPV vaccine coverage and screening of women aged more than 30 years with hrHPV testing followed by treatment of hrHPVpositive (hrHPV+) women that in the visual inspection are suspicious of cervical cancer precursor lesions. 3 hrHPV DNA testing has greater sensitivity and negative predictive value for detecting cervical intraepithelial neoplasia Grade 2 or worse (CIN2+) and CIN3+ (more reproducible and definitive surrogate endpoint of cervical cancer risk) than cytology, 4 which permits longer screening intervals and more costeffective prevention programs. 5 Because HPV-based screening provides 60% to 70% greater protection against invasive cervical carcinomas compared to cytology, 6 it is expected that HPV-based screening would result in a lower incidence of and mortality from cervical cancer. hrHPV testing is the most effective approach for reducing cervical cancer mortality especially in LMICs, where improvement of quality of cytology remains a challenge and the number of life-time screening visits is low. 5 However, hrHPV testing has low specificity, which can increase referrals to colposcopy, leading to more anxiety and overtreatment of women with nonprogressive disease. 7 Approaches to stratify (triage) hrHPV+ women include conventional (Pap smear), liquid-based cytology (LBC), without or with adjunctive p16/Ki67 double staining (herein p16/Ki67 cytology) and HPV16/18 genotyping. Reassurance provided by cytology against false negatives is low 8 and a high proportion of women hrHPV+ and cytology negative at screening need further follow-up. p16/Ki67 has both higher sensitivity and specificity than cytology testing for triage of HPV-positive women and negative results have greater reassurance against CIN2+ than negative cytological results, 9 but as with cytology, good results depend on expert visual interpretation of well-preserved morphological specimens from LBC, which is an expensive and subjective expertise not widely available in LMICs. PCR-based HPV16/18 genotyping is a robust, operator independent assay without extra requirements for sample preservation; however, it reaches a sensitivity of only 50% to 60% 10 and misses all the CIN2+ in women positive for the other 11 hrHPV types.
Deregulated expression of hrHPV E6 and E7 genes can induce uncontrolled cell cycle progression and favors the development of persistent hrHPV infection and precancer. This process coincides with a decline in activity of the late capsid genes promoter (L1 and L2), which shows a higher methylation level in hrHPV+ women diagnosed with CIN2+. 11 In addition, changes in the levels of methylation of CpG sites in promoters or introns of host-cell genes such as EPB41L3, JAM3, TERT, CADM1, MAL, mir124 and FAM19A4 are also associated with higher risk of cervical cancer and its precursor lesions. 12

What's new?
For cervical cancer screening, testing HPV16/18 types has lower specificity than cytology, but a high false positive rate.
Here, the authors evaluated the S5 classifier test, which is based on DNA methylation. They found that for triaging women who test positive for high risk HPV, S5 had better sensitivity and fewer false positives than cytology plus HPV16/18. This paper represents the first time that S5 methylation has been tested head-to-head with cytology and HPV16/18 testing in a LMIC. As affordable methylation tests become available, this strategy may prove useful for triage in low resource settings.
Although earlier studies in developed countries have shown that the performance of the S5 classifier to detect CIN2+ is superior to that of HPV16/18 genotyping and similar to complex triage strategies such as a combination of repeated LBC and HPV genotyping, 15 there are few studies validating S5 in LMIC settings. 16

| Selection of methylation sub-study participants
Cases were women identified after the end of the ASC-US-COL trial as women who had hrHPV+ test results at baseline (recruitment visit) and with a colposcopy-directed biopsy diagnosis of CIN2, CIN3 or carcinoma in situ, adenocarcinoma (ADC) or squamous cell carcinoma (SCC) at any time during the 2-year follow-up. Controls were randomly selected regardless of arm allocation from women who were hrHPV+ at baseline, had a biopsy with a diagnosis of less than CIN2 (<CIN2) during the follow-up confirming that they were at low risk of cervical cancer and with enough remainder of archived baseline samples in specimen transport medium (STM; QIAGEN) for further testing. Controls were individually matched to cases by age and time to diagnosis (±12 months). As shown in the flowchart (Figure 1  using the percentage of individual CpG sites methylated as described previously. 13

| Statistical methods
The analysis was based on a prespecified analytical plan. The primary hypothesis was that using the hrHPV+ baseline samples, S5 at the standard predefined cut-point 0.8 13 can distinguish between <CIN2 and CIN2+ (includes CIN2, CIN3 and cancer) and also have a very high sensitivity for CIN3+. We also compared the sensitivity of S5 with the other tests at a specificity of 76% that corresponds to using a positivity threshold of 3.1. We used this threshold for S5 because it gave the same specificity as cytology, the currently rec-      Figure 3 shows similar patterns of methylation levels increasing by lesion severity for EPB4IL3 ( Figure 3A, Cuzick trend test χ 2 = 23.47, P < .001) and HPV16L1 individually ( Figure 3B, Cuzick trend test χ 2 = 25.4, P < .001).

| Performance of the S5 classifier to detect CIN2+ or CIN3+
The receiver operating characteristic curve showed that S5 had an AUC of 0.70 (95% CI 0.64-0.75, P < .0001) for CIN2+ and of 0.72 (95% CI 0.65-0.80, P < .0001) for CIN3+ (Figure 4). Cross tabulation of the number of cases or controls with negative or positive test results and the comparison of sensitivities and specificities of the tests to detect CIN2+ or CIN3+ are shown in Tables 2 and 3, 26 Methylation also exhibited higher specificity than cytology (at an ASC-US threshold) and higher sensitivity than HPV16/18 to detect CIN2+ and CIN3+ among hrHPV+. Because of the heterogeneity of assays and targets to estimate the level of methylation, and because some studies did not have histopathology verification, there is a need of further confirmation of these conclusions. In contrast to cell-based methods, DNA methylation can be reflexed using the same cervical exfoliates used for HPV testing and the molecular methods offer the possibility for self-sampling. These characteristics make a methylation-based test a very good candidate for immediate triage and treatment where necessary when self-sampling is used as the primary test. However, there are very few studies validating methylation markers in women from LMICs, 25,26 which are the countries with the highest cervical cancer incidence and mortality rates.
Here, we have evaluated the S5 classifier, a multigene methylation test, which has very well validated procedures and parameters for estimating the methylation levels and that has shown good performance in the United Kingdom, 14 Canada 15 and Mexico. 16   higher sensitivity is crucial to identify at-risk women in fewer screening visits and decreasing the use of resources to follow-up women with low risk of disease.
In our study, cytology had a specificity of 75%, but was the test with by far the highest false negative rate. We recognize that the sensitivity of our cytology was much lower than the performance of this test seen in specialist centers, but several studies have demonstrated that the sensitivity in LMICs including some Latin American countries may be as low as 30%. [31][32][33][34] It is worth noting that despite many efforts to improve cytology quality and sensitivity, difficulties in quality control and delays in diagnosis still prevail. 5 It is recognized that borderline and mild cytology have low reproducibility. Furthermore, since we matched controls to cases by age and hrHPV status, the distribution of cytology grades of controls may be biased. Thus, our results must be interpreted with the knowledge that ASC-US diagnoses and cytology grades distribution may not be comparable to other clinical settings. The appropriate sensitivity/specificity combination (and the corresponding decision threshold) of the methylation tests for the application of triaging hrHPV+ women has not been defined. In one meta-analysis the pooled sensitivity and specificity for CIN2+ was estimated irrespective of threshold used to define methylation positivity. 25 In the other, the threshold corresponded to a specificity of 70%. 26 We used the prespecified 0.8 cut-point that corresponded to a sensitivity of 90% (95% CI 87-92) and a specificity of 49% (95% CI 46-52) in a previous study with S5, 13 and conducted a post hoc analysis setting the specificity at 76% that corresponded at 3.1 cut-point. Future work is planned to assess the performance of S5 as a triage for HPV-positive tests to determine appropriate cutoffs within a screening population. Also, further work is needed to determine the performance of the S5 methylation assay in vaccinated populations.
Despite the differences in study designs and the proportion of CIN2+ and CIN3+ cases included, the consistency of our results are remarkably similar to what has been observed previously. 13 The S5 methylation test accurately identified women with higher risk of cervical high-grade disease and cancer among those who were hrHPV+.
Even further, our study demonstrated that S5 outperforms both cytology and HPV16/18 CIN2+ and CIN3+ detection in hrHPV+ women. [14][15][16] Currently S5 DNA methylation test is labor intensive and costly. The recent developments of affordable and scalable nextgeneration sequencing assays 35

DATA AVAILABILITY STATEMENT
The dataset excluding personal identifiers will be available to proper academic parties on request from the corresponding author in accordance with the data sharing policies of Queen Mary University of London (UK).

ETHICS STATEMENT
All women from ASC-US-COL trial signed informed consent for use of samples and data for future studies. The ethics committees of Sede