Performance of CADM1, MAL and miR124-2 methylation as triage markers for early detection of cervical cancer in self-collected and clinician-collected samples: an exploratory observational study in Papua New Guinea

Objective WHO recommends human papillomavirus (HPV) testing for cervical screening, with triage of high-risk HPV (hrHPV) positive women. However, there are limitations to effective triage for low-resource, high-burden settings, such as Papua New Guinea. In this exploratory study, we assessed the performance of host methylation as triage tools for predicting high-grade squamous intraepithelial lesions (HSIL) in self-collected and clinician-collected samples. Design Exploratory observational study. Setting Provincial hospital, same-day cervical screen-and-treat trial, Papua New Guinea. Participants 44 hrHPV+women, with paired self/clinician-collected samples (4 squamous cell carcinomas (SCC), 19 HSIL, 4 low-grade squamous intraepithelial lesions, 17 normal). Primary and secondary outcome measures Methylation levels of CADM1, MAL and miR124-2 analysed by methylation-specific PCRs against the clinical endpoint of HSIL or SCC (HSIL+) measured using liquid-based-cytology/p16-Ki67 stain. Results In clinician-collected samples, MAL and miR124-2 methylation levels were significantly higher with increasing grade of disease (p=0.0046 and p<0.0015, respectively). miR124-2 was the best predictor of HSIL (area under the curve, AUC 0.819) while MAL of SCC (AUC 0.856). In self-collected samples, MAL best predicted HSIL (AUC 0.595) while miR124-2 SCC (AUC 0.812). Combined miR124-2/MAL methylation yielded sensitivity and specificity for HSIL+ of 90.5% (95% CI 69.6% to 98.8%) and 70% (95% CI 45.7% to 88.1%), respectively, in clinician-collected samples, and 81.8% (95% CI 59.7% to 94.8%) and 47.6% (95% CI 25.7% to 70.2%), respectively, in self-collected samples. miR124-2/MAL plus HPV16/HPV18 improved sensitivity for HSIL+ (95.2%, 95% CI 76.2% to 99.9%) but decreased specificity (55.0%, 95% CI 31.5% to 76.9%). Conclusion miR124-2/MAL methylation is a potential triage strategy for the detection of HSIL/SCC in low-income and middle-income country.


INTRODUCTION
Cervical cancer is the fourth most frequently diagnosed cancer and the fourth leading cause of cancer-related death in women, with an estimated 604 000 new cases and 342 000 deaths worldwide in 2020. 1 Age-standardised rates are disproportionally higher in

STRENGTHS AND LIMITATIONS OF THIS STUDY
⇒ The assessment of methylation markers in selfcollected vaginal samples recollected in a lowmiddle-income country allows the detection of high-grade squamous intraepithelial lesion (HSIL+)  and open the way to full molecular screening in these settings.⇒ A direct comparison of self-collected samples and paired clinician-collected samples was possible, thus identifying different performance and optimal cut-off values by sample type.⇒ Study design allowed exploration of the clinical performance of extended genotyping (human papillomavirus 16/18/31/33/45/52/58) combined with host methylation analysis for the detection of HSIL.⇒ A small sample size led to broad CI.⇒ Due to limited specialist staff and infrastructure, liquid-based cytology was the diagnostic reference standard rather than a histological endpoint.
3][4][5] Infection with high-risk human papillomavirus types (hrHPV) may cause cervical abnormalities graded histologically as cervical intraepithelial neoplasia (CIN, grades 1-3) or cancer (principally squamous cell carcinoma (SCC) or adenocarcinoma), or graded cytologically (using exfoliated cells from the cervix) as low-grade squamous intraepithelial lesion (LSIL), high-grade squamous intraepithelial lesion (HSIL) or SCC. 6o achieve cervical cancer elimination, 5 many countries are transitioning their cervical screening programmes to HPV nucleic acid-based testing, which is more sensitive than cytology for detecting underlying high-grade disease.][8][9][10][11] Another important advantage is that nucleic acid amplification testing and subsequent treatment can be performed at point-of-care, greatly simplifying programme delivery, particularly in low-resource settings.These advances led the WHO, in 2021, to endorse HPV screen-and-treat guidelines for women in LMIC. 6n important limitation of HPV testing is that it suffers from a lower specificity than cytology, leading to concern that the treatment of all hrHPV+ women would result in substantial overtreatment. 12Hence, a second (triage) test is recommended to identify hrHPV+ women at highest risk of having underlying disease requiring treatment.The triage technologies most used are cytology, partial genotyping for HPV16 or 18, and in many low-resource settings, visual inspection using acetic acid or Lugol's iodine. 12However, these triage approaches present challenges for sensitivity and/or specificity in LMIC settings. 8 12olecular triage tests are advantageous over current methods as they are less subjective and could be performed on multiple specimen types, including selfsamples.In particular, aberrant host gene DNA methylation patterns in promoter regions of some genes have shown promise as a triage marker in cervical screening in high-income countries.Aberrant methylation is pivotal in cervical carcinogenesis; it causes changes in gene expression, faulty condensation and chromosomal instability. 13 14Increased DNA methylation has been associated with hrHPV persistence 13 and correlates with increasing severity of cervical disease. 14 15DNA methylation markers such as CADM1, MAL and miR124-2 have shown good performance for the detection of CIN2 or worse (CIN2+) among hrHPV+women. 14 16 17However, data on their performance have been based on studies conducted in high-income settings and among highly screened populations. 18There are currently no data on the performance of these markers among LMIC populations by using self-collected vaginal samples, nor in settings where HPV-based screening and same-day treatment strategies are used.
Papua New Guinea has a high burden of cervical cancer, with age-standardised incidence and mortality rates of 29.2 and 19.1 per 100 000, respectively, making it the second most common cancer among Papua New Guinean women. 3A trial study (HPV self-collect, test and same-day-treat, HPV-STAT) performed on women from this country evaluated the clinical performance, treatment completion rates, adverse events profile and acceptability of a fully integrated screening strategy, comprising point-of-care HPV test of self-collected specimens and same-day thermal ablation. 19This trial supported the introduction of HPV screening and same-day treatment for the control of cervical cancer in LMIC. 19In the current study, we explored the performance of CADM1, MAL and miR124-2 DNA methylation markers alone and in combination with HPV16/18 and extended genotyping (HPV16/18/31/33/45/52/58) as a triage strategy for the detection of HSIL+ among HPV+women that participated in this cervical screening trial.A direct comparison was also performed for each methylation marker, between paired self-collected and clinician-collected samples.

MATERIALS AND METHODS Study population and design of the trial
The exploratory study was conducted as a substudy of HPV-STAT a prospective, single-arm intervention trial among women attending cervical screening services at two clinical sites in Papua New Guinea.The trial is registered with ISRCTN, ISRCTN13476702 (https://www.isrctn.com/editorial/retrieveFile/3ed9173e-cce5-4158-a586-577775f0cbdd/35731).Study design, recruitment and protocols have been described. 19Briefly, between 5 June 2018 and 6 January 2020, 4285 women aged 30-59 years attending clinics for routine cervical screening at Modilon Hospital (Madang Province) and Mt Hagen General Hospital (Western Highlands Province), who meet the inclusion criteria were invited to participate.Women who were pregnant or who had given birth in the past 6 weeks, or women who had a history of cervical cancer or hysterectomy, were excluded. 19Those women able to provide written informed consent were eligible to participate, enrolled sequentially and collected a vaginal specimen using a cytobrush ('Just for Me', Preventative Oncology International, Cleveland Heights, Ohio, USA), which was placed into a 20 mL ThinPrep vial (Hologic, Marlborough, Massachusetts, USA). 19int-of-care HPV testing and follow-up assessment For self-collected vaginal specimens, a 1 mL aliquot was tested for the presence of oncogenic HPV genotypes using the Xpert HPV Test (GeneXpert; Cepheid, Sunnyvale, California, USA) as per the manufacturer's instructions.Results were reported as 'HPV16', 'HPV18/45' and 'other HPV' (a summary result for HPV31, 33, 35, 39, 51, 52, 56, 58, 59, 66 and 68) and were provided to women before midday to allow sufficient time for same-day counselling, pelvic examination and treatment/referral as indicated. 19All women with a negative HPV test were informed of their results and advised to return to the clinic for HPV-based screening in 5 years.
hrHPV+women provided a cervical specimen collected by a clinician using a Cervex-Brush Combi (Rovers Medical Devices, Oss, The Netherlands), placed in a 20 mL ThinPrep vial and stored at 4°C prior to shipment to the Australian Centre for the Prevention of Cervical Cancer in Melbourne, Australia for liquid-based cytology (LBC) and p16/Ki67 dual stain cytology.A 15% random sample of HPV-negative women were also asked to provide a clinician-collected cervical specimen for LBC, as above.

Cytological assessment
LBC was carried out in accordance with established laboratory procedures at the Australian Centre for the Prevention of Cervical Cancer.Slides were independently assessed by two cytologists blinded to HPV test results.Where both cytologists agreed on a diagnosis of HSIL or HSIL+, a final diagnosis was recorded.If the assessment differed, dual p16/Ki-67 dual stain was carried out by using CINTec PLUS Cytology (Roche Diagnostics, Pleasanton, USA) to make a final diagnosis. 19The primary clinical endpoint was then HSIL or SCC (HSIL+) using LBC/p16-Ki67 stain.LBC was accepted as a reference standard rather than histology, which is the gold reference in high-resource settings.It was not feasible from a staffing and logistical perspective to provide colposcopy or to collect cervical biopsies for histological examination in Papua New Guinea.

Participants in the exploratory study and design
The substudy included 44 paired cervical and vaginal samples from women participating in the trial at Mt Hagen General Hospital (Mount Hagen, Western Highlands Province).This included all 23 hrHPV+HSIL+ (19 HSIL and 4 SCC) cases identified on LBC by the end of 2018 and 21 randomly selected hrHPV+normal/LSIL samples (17 normal LBC and 4 LSIL).Sample size selection was driven by a published study by Li et al that showed through simulation studies and real data from two studies downloaded from the NCBI Gene Expression Omnibus that at least 12 specimens in each group is needed to detect truly differential DNA methylation with enough power (≥ 80%), reproducible data and consistency when using different statistical methods. 20Molecular biologists and technicians were blinded to point-of-care HPV and clinical diagnosis.
DNA extraction and HPV-specific typing DNA was extracted as described previously, quantitated by Qubit Fluorometer (Life Technologies, California, USA) and assessed for integrity by quantitative PCR amplification of a 260 base-pair product of the human beta-globin gene. 21In addition, DNA extracted from cervical cell line SiHa (1-2 copies of HPV16 per cell, American Type Culture Collection (ATCC) Cat# HTB-35, RRID:CVCL_0032; ATCC, Manassas, Virginia, USA; 100 ng) was used as positive control for methylation analysis.This DNA was confirmed to have only HPV16 by HPV typing for 28 HPV types (Anyplex II HPV28 Detection, Seegene, Seoul, South Korea).
Bisulphite DNA modification and quantitative methylationspecific PCR for CADM1, MAL and miR124 genes DNA from the paired samples (1-100 ng) and control (SiHa, 100 ng) were bisulphite treated using Methylamp DNA modification Kit (Epigentek, Brooklyn, New York, USA) as per the manufacturer's instructions.Modified DNA was eluted in 40 µL of the methylamp elution buffer.Samples without the adequate concentration were excluded from the bisulphite modification and methylation analysis.
Quantitative methylation-specific PCR (qMSP) targeting CpG sites in promoter regions of CADM1 (promoter region M9), MAL (promoter region M1) and miR124-2 (promoter region 2) was performed as previously described, 14 with minor modifications.Due to the use of a different platform (Light Cycler 480 II, Roche), validations of the qPCRs were performed utilising duplicated dilution series of SiHa cell line (intra and inter assays of reproducibility) and by using a training panel of samples as reported previously. 21As a reference, a qMSP for the housekeeping gene b-actin (ACTB) was performed.Samples with Ct values for ACTB of >34 were considered ACTB negative as this indicated poor sample quality due to insufficient DNA or inadequate bisulphite conversion.Target DNA methylation levels were normalised to the reference gene as described previously. 21

Statistical analysis
The percentage of methylation was calculated as described previously. 22Differences in the median percentage methylation by disease grade (SCC, HSIL, LSIL/normal) were visualised using box and whisker plots and compared using Wilcoxon test (between groups) and Kruskal-Wallis (overall analysis). 20 21Area under the receiver operating characteristic curve (ROC) (AUC) was used to assess the ability of the methylated genes to classify HSIL or SCC, providing two main outcomes: the diagnostic accuracy of the test and the optimal cut-off point value for dichotomisation of the methylation test. 23The optimal cut-off point was calculated by using the maximum sum of sensitivity and specificity. 24Sensitivity (number of correct positives (ie, positive for at least one marker)/number of reference assay positives) and specificity (number of correct negatives/number of reference assay negatives) were determined for the following triage strategies: (1) single Open access gene methylation analysis, (2) combination of methylation markers, (3) HPV 16 genotyping, (4) combination of HPV16/18 genotyping, (5) combination of HPV 16/18/31/33/45/52/58 genotyping, (6) HPV16/18 with single and combined methylation analysis and (7) HPV 16/18/31/33/45/52/58 with single and combined methylation analysis.
Median percentage methylation for each marker was compared for self-collected and clinician-collected vaginal paired specimens using the Wilcoxon signed rank test.The vaginal specimen was considered the test while the cervical specimen acted as the non-reference standard.Agreements between DNA methylation results from cervical and vaginal specimens were determined using the optimal cut-off point established for each gene by calculating positive percent agreement (PPA), negative percent agreement (NPA) and overall per cent agreement (OPA) as per recommendations by the Food and Drug Administration, USA. 25 The results were analysed by using XLSTAT and the statistical platform R studio (V.4.0.1) and programmes ggplots2 (V.3.3.2),ggpubr (V.0.4), pROC (V.1.16.2) and cutpointr (V.1.1.1). 21 24e used the Standards for Reporting of Diagnostic Accuracy Studies (STARD) reporting guidelines for reporting diagnostic accuracy studies. 26tient and public involvement There was no patient or public involvement in study design or conduct.
Using clinician-collected samples, ROC analysis showed that miR124-2 was the best methylation marker to distinguish SCC and HSIL (from normal/LSIL), with an AUC of 0.850 and 0.819, respectively (figure 2A,C).Methylation of MAL showed higher AUC for detecting SCC (0.856) than HSIL (0.700).CADM1 methylation showed poor performance for detecting SCC or HSIL with AUC of 0.700 and 0.607, respectively (figure 2A,C).
In self-collected samples, there was a trend of increasing DNA methylation with disease grade for MAL and miR124-2, though this did not reach significance (p=0.083,p=0.075, respectively) (figure 1, lower centre panel).As found for clinician-collected samples, methylation of CADM1 did not differ significantly with increasing cytology grade (figure 1, lower left panel).
ROC analysis showed that from self-collected samples, miR124-2 was the best methylation marker to distinguish SCC, but not HSIL, from normal/LSIL with an AUC of 0.812 and 0.563, respectively (figure 2B,D).Methylation of MAL or CADM1 showed a lower clinical performance for SCC (AUC of 0.725 and 0.750, respectively) than miR124-2, and a similar performance for HSIL (AUC of 0.595 and 0.515, respectively) (figure 2B,D).

Comparison of levels of methylation between paired selfcollected and clinician-collected samples
There was generally a lower percentage of methylation for each marker in vaginal samples compared with their paired cervical samples (figure 3).However, this only Open access reached statistical significance for miR124-2 in HSIL (p=0.0002)(figure 3).

DISCUSSION
In this study, we explored CADM1, MAL and miR124-2 DNA methylation using clinician-collected and self-collected HPV+samples from Papua New Guinea, a setting with a high burden of cervical disease.In clinician-collected cervical samples, DNA methylation of MAL and miR124-2 was significantly higher in HSIL/cancer compared with normal/LSIL samples while the methylation of CADM1 Open access failed distinguish between cytology grades.Methylation of miR124-2 showed the highest AUC to distinguish HSIL from normal/LSIL and the second-highest performance to predict SCC, with AUC values above 0.819.Exploration of diagnostic performance of miR124-2 in combination with MAL for the detection of HSIL+, showed an excellent performance (sensitivity 90.5%, specificity 70%).Similar findings were obtained with self-collected samples, though their performance was reduced.Overall, miR124-2 in combination with MAL DNA methylation performed well in the detection of HSIL+ in women from LMIC.Addition of HPV16/18 or extensive genotyping improved sensitivity at the expense of lower specificity and higher referrals to immediate treatment.miR124-2/ MAL methylation markers are suitable for further evaluation against histology endpoints, and subsequently in LMIC-based field studies for triage of hrHPV+women to identify high-grade disease.
8][29] miR124-2 methylation alone or in combination with CADM1 and MAL has high sensitivity and specificity for the detection of CIN3+ in HPV+ clinician-collected cervical specimens in women from high-income countries (sensitivity of 68.0%-94.7%,specificity of 50.7%-78.9%). 30ew studies have been performed in LMIC using clinician-collected samples.An analysis of the same markers in HPV+ and HIV+ women from Kenya found the AUC of miR124-2 methylation for CIN2+ and CIN3+ were 0.730 and 0.810, respectively, 31 marginally lower to those obtained in the current exploratory study.However, these results are not directly comparable to ours as their lesion grade was based on histological diagnosis and HIV+ participants.In the current study, we were unable to stratify by HIV status, but assume a low HIV prevalence as the HIV prevalence in the Papua New Guinean general population is <1%.Two other studies performed in South Africa and Thailand (upper-middle-income countries) also showed good performance of some of these markers, alone or in combination, for the detection of HSIL+. 32 33ADM1 was not a good diagnostic marker, this is consistent with the findings of two other studies conducted in high-income countries. 34 35o research has been conducted looking methylation in self-collected samples from LMIC.Diagnostic performance of miR124-2 for detection of HSIL+ in self-collected samples showed a sensitivity of 77.3% and specificity of 47.6% while combining miR124-2 with MAL increased the sensitivity to 81.8%, showing promise for self-sampling applications.While the sensitivity/specificity was slightly lower than clinician-collected samples, this may be offset by higher screening rates achieved by this more acceptable method of collection. 117][38][39] This wide variation may be explained by variable methodology (self-sampling access collection devices, methylation detection assays, genes analysed, thresholds to define methylation positivity, study populations, disease endpoints (cytological HSIL or histological CIN2/3+), etc). Methylation analysis of miR124-2/MAL combined with HPV16/18 genotyping had a sensitivity for CIN3+ of 77.6% and specificity of 54.8%. 40A recent prospective cohort study in an upper-middle-income country analysing methylation markers ZNF671/ASTN1/ITGA4/RXFP3/SOX17/ DLX1 showed a sensitivity and specificity of 96.6% and 58.3%, respectively, to detect CIN2+ when combining Open access methylation and HPV16/18. 42In our study, combining HPV16/18 detection with miR124-2/MAL methylation improved sensitivity (increasing from 90.5% to 95.2%), but reduced specificity (decreasing from 70.0% to 55.0%).
In self-collected samples, the sensitivity was similar to clinician-collected samples (95.5%) but the specificity was lower (33.3%).Combining extended genotyping (HPV16/18/31/33/45/52/58) with miR124-2/MAL methylation had similar sensitivity to miR124/MAL/ HPV16/18 testing but showed lower specificity.Regardless of study population, HPV prevalence, sample type, marker panel and age, the combination of host cell DNA methylation analysis with partial or extended genotyping could serve as a potential triage strategy for hrHPV+ women, at the expense of a higher referral to treatment.
Self-collected samples generally had lower levels of methylation than paired clinician-collected samples, likely due to lower diseased/normal cell ratios.For example, values for the GynTect assay (detecting six methylation markers) were significantly lower for self-collected than clinician-collected samples and the overall concordance was moderate (kappa 0.394). 43Further discovery work to identify markers suitable for self-collection should consider using self-collected samples from LMIC.Verlaat et al performed a chip-based genome DNA methylation screen on self-samples from a high-income country and identified a three-gene methylation classifier (ASCL1, LHX8 and ST6GALNAC5) that was able to identify CIN3 in cervical self-samples with an ROC of 0.88, which is superior to currently available methods. 44imitations of this study include the small sample size which led to broad CIs; however, it aimed to provide estimated effects (point estimates with CIs) for all measures of interest which are valuable for planning subsequent studies in LMIC.It was not feasible to conduct cervical biopsies in this setting due to limited specialist staff and infrastructure; hence LBC HSIL+ was the diagnostic reference standard rather than a histological endpoint and this may have impacted estimates of sensitivity/specificity for methylation markers.Since HPV testing is more sensitive than cytology at detecting HSIL+, then the selection of cytology as a reference standard will result in an overestimation of the sensitivity of the HPV test and a decrease Open access in the specificity, compared with results using histology as a reference standard.However, to avoid disease misclassification by cytology all slides were assessed by two independent experienced cytologists, and dual p16/ Ki-67 immunostaining was performed to resolve disagreements.A triage test based on methylation markers could be highly suited to regions without access to colposcopy, where the combination of HPV detection, methylation analysis and cytology could confirm HSIL+ in the absence of colposcopy examination.Additional information is required to understand the long-term safety of methylation markers in screening programmes, including longitudinal data from a low-resource setting and analysis of methylation in self-collected samples from the general population and HIV+ women.

CONCLUSIONS
DNA methylation of miR124-2/MAL alone or in combination with HPV16/18 is very promising as a triage test for the prediction of HSIL/SCC among women with oncogenic HPV infections.Self-collected samples showed reduced specificity compared with clinician-collected samples, indicating a need to identify additional biomarkers for this sample type.Future evaluation of these results against histological endpoints, and subsequently in different populations may lead to better tests to manage HPV+ women in LMIC.

Figure 1
Figure 1 Percentage of DNA methylation of gene CADM1, MAL and miR124-2 according to cytological grade in oncogenic HPV positive clinician-collected cervical samples (cervical) in the top panel and self-collected vaginal samples (vaginal) in the bottom panel.Whiskers correspond to the 1st and 3rd quartiles (the 25th and 75th percentiles).Overall significance by Kruskal-Wallis is indicated.HPV, human papillomavirus; HSIL, high-grade squamous intraepithelial lesions; LSIL, low-grade squamous intraepithelial lesions; SCC, squamous cell carcinomas.

Figure 2
Figure 2 Receiver operating characteristic (ROC) curve values of the performance of DNA methylation of CADM1, MAL and miR124-2 for distinguishing HSIL from LSIL/normal, and SCC from LSIL/normal, stratified by type; clinician-collected cervical samples (A, C) or self-collected vaginal samples (B, D).HSIL, high-grade squamous intraepithelial lesions; LSIL, low-grade squamous intraepithelial lesions; SCC, squamous cell carcinomas.

Table 1
Performance of host DNA methylation markers for detection of HSIL+ Shaded rows show markers or a combination of markers with the best diagnostic performance and those with the maximal sensitivity for detection of HSIL+.*Positive for at least one of the indicated host genes.HSIL, high-grade squamous intraepithelial lesion.

Table 2
Performance of HPV genotyping combined with methylation markers for the detection of HSIL+