Detection of hypermethylated genes as markers for cervical screening in women living with HIV

Abstract Introduction To evaluate the performance of hypermethylation analysis of ASCL1,LHX8 and ST6GALNAC5 in physician‐taken cervical scrapes for detection of cervical cancer and cervical intraepithelial neoplasia (CIN) grade 3 in women living with HIV (WLHIV) in South Africa. Methods Samples from a prospective observational cohort study were used for these analyses. Two cohorts were included: a cohort of WLHIV who were invited for cervical screening (n = 321) and a gynaecologic outpatient cohort of women referred for evaluation of abnormal cytology or biopsy proven cervical cancer (n = 108, 60% HIV seropositive). Cervical scrapes collected from all subjects were analysed for hypermethylation of ASCL1,LHX8 and ST6GALNAC5 by multiplex quantitative methylation specific PCR (qMSP). Histology endpoints were available for all study subjects. Results Hypermethylation levels of ASCL1,LHX8 and ST6GALNAC5 increased with severity of cervical disease. The performance for detection of CIN3 or worse (CIN3+) as assessed by the area under the receiver operating characteristic (ROC) curves (AUC) was good for ASCL1 and LHX8 (AUC 0.79 and 0.81 respectively), and moderate for ST6GALNAC5 (AUC 0.71). At a threshold corresponding to 75% specificity, CIN3+ sensitivity was 72.1% for ASCL1 and 73.8% for LHX8 and all samples from women with cervical cancer scored positive for these two markers. Conclusions Hypermethylation analysis of ASCL1 or LHX8 in cervical scrape material of WLHIV detects all cervical carcinomas with an acceptable sensitivity and good specificity for CIN3+, warranting further exploration of these methylation markers as a stand‐alone test for cervical screening in low‐resource settings.


| INTRODUCTION
Women living with human immunodeficiency virus (WLHIV) have an increased risk for the development of cervical cancer and its precursor lesions, classified as cervical intraepithelial neoplasia (CIN) grade 1 to 3 [1][2][3]. Compared to HIV uninfected women, WLHIV develop cervical cancer at a younger age and are more likely to die of the disease [4][5][6]. Both HIV and cervical cancer have a disproportionally high burden in low-and middleincome countries (LMIC): more than 95% of global HIV infections and more than 85% of all cervical cancer cases occur in less developed regions [7,8]. Here, cervical cancer is a leading cause of cancer-related death in women, partly caused by the high incidence of HIV, but also caused by the absence of effective cervical screening programmes and limited access to healthcare [8]. Implementation of screening in low-resource settings is challenging and the development of screening methods that are suitable for this setting is warranted.
Cervical cancer and CIN are caused by a persistent infection with high-risk human papillomavirus (hrHPV) [9,10]. Primary hrHPV testing is currently the preferred method for cervical screening, irrespective of resource settings or HIVprevalence [11]. However as most infections are self-limiting and do not cause cervical lesions, hrHPV testing has limited specificity [12,13], particularly in WLHIV [14][15][16]. Therefore, subsequent triage testing of hrHPV positive women is needed to distinguish women with underlying high-grade cervical disease from women with transient infections [17,18]. Available triage tests for LMIC recommended by the World Health Organization (WHO) include cytology, partial hrHPV genotyping or visual inspection with acetic acid (VIA) [19]. Major limitations of these two-step approaches include the risk of loss to follow-up and their requirement of technical capabilities and healthcare infrastructure, specifically relevant in lowresource settings [20]. In addition, in a setting with a high HPV prevalence, triage strategies require a large number of tests. A single and objective point-of-care test with a high sensitivity and specificity for CIN3 and cervical cancer in both HIV seropositive and HIV seronegative women overcomes these limitations and would be most effective to improve cervical screening in LMIC.
A candidate primary test to identify women at risk for clinically meaningful cervical disease is hypermethylation analysis of promoter regions of host cell genes involved in cervical carcinogenesis [21][22][23]. Hypermethylation of gene promotor regions results in gene silencing and represents an essential step for cervical cancer development [21,24]. Assays detecting hypermethylation are objective and can be applied on various specimen types, including self-collected cervical material [25][26][27]. Multiple genes have been identified as possible targets for cervical precancer and cancer detection, but few have been evaluated in WLHIV [14,[28][29][30]. We previously showed that methylation analysis of CADM1, MAL and miR124-2 genes in cervical scrapes from hrHPV positive WLHIV is an acceptable triage tool, which detects all cervical carcinomas and the majority of CIN3 [14]. However, specificity of this marker panel when evaluated as a primary screening tool, without prior hrHPV testing, was limited. Therefore, additional markers that can be used for cervical screening in WLHIV in lowresource settings without prior hrHPV testing should be evaluated.
In a recent genome-wide DNA methylation profiling study, three hypermethylated genes, Achaete-scute Family bHLH Transcription Factor 1 (ASCL1), LIM Homeobox 8 (LHX8) and ST6 N-Acetylgalactosaminide Alpha-2,6-Sialyltransferase 5 (ST6GALNAC5), were identified as promising triage markers in hrHPV positive women for cervical cancer and CIN3 [31]. The present study evaluates the performance of these methylation markers for the detection of CIN3 and cervical cancer in WLHIV irrespective of their HPV status.

| Study population and procedures
The study population is outlined in Figure 1 and consists of two groups: a cohort of WLHIV and a gynaecological referral population. The total study population included 429 women who were originally included between February 2013 and March 2015 at a gynaecologic outpatient clinic in Steve Biko Academic Hospital or Tshwane District Hospital, Pretoria, South Africa, in a study (Ethical Committee of the University of Pretoria, South Africa protocol numbers 100/2012 and 155/2014) comparing different cervical screening strategies. These women previously had a valid study endpoint and were eligible for inclusion in the present report. All participants were aged 18 years and above, and had not been treated for cervical cancer or precancer in the preceding two years. Nearly all women (99%) in the cohort of WLHIV were on antiretroviral treatment (ART) and their median CD4 + cell count was 514 cells/lL. In total, 60% of women in the referral cohort were HIV seropositive of whom 37% were on ART; their median CD4 + cell count was 342 cells/lL. Highrisk HPV positivity was 42% in the cohort of WLHIV and 95% of women in the referral population. Median age was 41 [interquartile range (IQR): 35 to 46 years] in the cohort of WLHIV and 44 years (IQR: 34 to 51 years) in the referral population. Detailed characteristics, inclusion criteria and study procedures of this study have been described previously [14]. Written informed consent was obtained from all participants.
In short, HIV-infected women visiting the gynaecologic outpatient clinic for cervical screening were included in the cohort of WLHIV. Cervical cells were collected using a Cervex Brush â (Rovers Medical Devices B.V., Oss, the Netherlands) and, after preparation of a conventional slide, stored in Thinprep PreservCyt â solution (Hologic, Marlborough, MA, USA). Colposcopy was performed for all participants and two biopsies from either the most severe cervical lesion or, if no lesion was present, two random biopsies were collected. The referral population included women who visited the gynaecologic outpatient department for evaluation of an abnormal Pap smear [≥high-grade squamous intraepithelial neoplasia (HSIL)] or biopsy-proven cervical cancer. A cervical scrape was also collected from these women and the material was stored in Thinprep PreservCyt solution.
Women with abnormal cytology (≥HSIL) or CIN2 or worse (CIN2 + ) on a cervical biopsy received treatment according to local guidelines [large loop excision of the transformation zone (LLETZ) or clinical cancer staging]. Study endpoints were based on histological diagnosis of either the cervical biopsy or LLETZ specimen (worst histology). Participants with invalid histology results were excluded from the analysis.

| Reference population
Cervical scrapes from a Dutch reference population (n = 265; study endpoints: 196 ≤ CIN1, 30 CIN3, 39 cervical carcinomas; median age 40 years (IQR: 32 to 47 years)) were used only to compare methylation levels between HIV seronegative and HIV seropositive women (manuscript in preparation). This study group was assumed to be HIV seronegative, since HIV incidence rates in the Netherlands are very low (0.1% in women) [32].

| Methylation analysis
For methylation analysis, DNA previously isolated from cervical scrapes was bisulphite-converted using the EZ DNA Methylation Kit (Zymo Research, Irvine, CA, USA). Multiplex quantitative methylation-specific PCR (qMSP) for ASCL1, LHX8 and ST6GALNAC5 was performed as described previously using 50 ng of bisulphite-converted DNA, EpiTect MethyLite Master Mix (Qiagen, Hilden, Germany) and 100 to 300 nmol/ L of each primer and fluorescent dye-labelled probe [31]. Housekeeping gene b-actin (ACTB) was used as a reference to assure sample quality and successful bisulphite conversion. A plasmid containing the amplicon sequences of all targets and ACTB was used as a calibrator. Cycle threshold (Ct) values were measured at a fixed fluorescence threshold. Methylation values of all targets were normalized to the reference gene and the calibrator using the comparative Ct method (2 ÀΔΔCt 9 100) to obtain ΔΔCt ratios [33]. Analyses were done on the ViiA 7 real-time PCR system (Applied Biosystems, Foster City, CA, USA). All samples with ACTB Ct ratios >30 were considered invalid and were excluded from the analysis (n = 3).

| Statistical analysis
The Kruskal-Wallis omnibus test was performed on each methylated gene to assess differences in methylation levels among disease categories. Following a significant result from the omnibus test, post-hoc testing was then performed using Mann-Whitney U. Bonferroni correction was subsequently used to correct p-values for multiple testing. Log10-transformed Ct ratios were visualized in boxplots for the cohort of WLHIV and the referral population together. To assess the effect of HIV status on methylation levels of ASCL1, LHX8 and ST6GALNAC5, differences in methylation levels between HIV seropositive and HIV seronegative women within cases [CIN3 or worse (CIN3 + )] and controls [CIN1 or less (≤CIN1)] were calculated using Mann-Whitney U. For this analysis, the Dutch reference population was used to enrich the group of HIV seronegative samples.
To assess the performance of each individual methylation marker to distinguish cases from controls in cervical scrapes from women from the cohort of WLHIV, univariable logistic regression analysis was performed on the square root transformed Ct ratios. Histologically classified CIN2 is a heterogeneous group of cervical disease which can be either the results of a productive or transforming HPV infection [22]. Therefore, samples from women with CIN2 were excluded from this analysis.
Then, the clinical performance of each individual marker to detect CIN3 + in the cohort of WLHIV was evaluated by the leave-one-out cross-validation approach. With this approach, predicted probabilities were calculated for each sample, representing the risk for an underlying CIN3 + . Receiver operating characteristic (ROC) curves from the cross-validated predicted probabilities were used to visualize the performance of the logistic regression models and were evaluated by area under the ROC curve (AUC). Based on the ROC curves, fixed thresholds for predicted probabilities corresponding to 75% and 80% specificity were chosen and corresponding CIN3 + sensitivities were calculated. In addition, positivity rates for CIN2 were calculated.
To further evaluate the clinical performance for detection of CIN3 and cancer, the logistic regression models described above were applied in the referral population. Positivity rates per disease category were calculated for each methylation marker using the fixed thresholds. All

| RESULTS
Four hundred and twenty-six women from the South African study cohorts were included in this report. An overview of the histology endpoints is given in Figure 1.

| Methylation levels across disease categories and HIV status
Differences across cervical disease categories in methylation levels of ASCL1, LHX8 and ST6GALNAC5 were evaluated in the two South African study cohorts combined. As shown in To assess a potential influence of HIV status on methylation levels of ASCL1, LHX8 and ST6GALNAC5, the Ct ratios of each marker were stratified by HIV status and compared within cases (CIN3 + ) and controls (≤CIN1) using the Dutch reference population. Five women from the South African population with unknown HIV status were excluded from this analysis. Median age of HIV seropositive women and HIV seronegative  Figure 3B) and ST6GALNAC5 ( Figure 3C) were significantly higher in HIV seropositive women with ≤CIN1, compared to HIV seronegative women with ≤CIN1 (p < 0.001). In women with CIN3 + methylation levels were comparable between HIV seropositive and HIV seronegative women (p > 0.05).

| Performance of primary methylation marker analysis
To determine the clinical performance of ASCL1, LHX8 and ST6GALNAC5 to distinguish cases (n = 61) from controls (n = 224), logistic regression analysis was performed in the South African cohort of WLHIV. All three methylation markers significantly distinguished cases from controls (p < 0.001).
Subsequently, the clinical performance of ASCL1, LHX8 and ST6GALNAC5 to detect CIN3 + was evaluated by leave-one-out cross-validation approach. ASCL1 and LHX8 showed a good clinical performance, visualized by ROC curves and quantified by AUCs being 0.79 for ASCL1 ( Figure 4A) and 0.81 for LHX8 ( Figure 4B). ST6GALNAC5 showed a moderate performance with an AUC of 0.71 ( Figure 4C). Based on these ROCs, fixed thresholds corresponding to a specificity of 75% and 80% were chosen. At fixed thresholds corresponding to 75% specificity, the sensitivity for CIN3 + was 72.1% (95% confidence interval (CI) 59. 2

| Methylation positivity in a gynaecological referral population
To further evaluate the performance of ASCL1, LHX8 and ST6GALNAC5 to detect CIN3 and cervical cancer, we calculated the positivity rates in the referral population for each   In the referral population and cohort of WLHIV combined, we analysed the proportion of samples testing positive for none, one, two or three of the markers ASCL1, LHX8 and ST6GALNAC5 within each disease category, at both 75% and 80% specificity thresholds. We found the proportion of samples testing positive for two or three markers to increase with severity of cervical disease at 75% threshold ( Figure 5). Similar results were found at the 80% thresholds.

| DISCUSSION
This study reports on primary DNA methylation analysis of host cell genes ASCL1, LHX8 or ST6GALNAC5 for the detection of cervical cancer and CIN3 in cervical scrapes from WLHIV. In this screening cohort of HIV-infected women we showed a good CIN3 + performance for ASCL1 and LHX8 (AUC 0.79 and 0.81 respectively) and a moderate performance for ST6GALNAC5 (AUC 0.71). At a fixed specificity of 75%, the CIN3 + sensitivity of ASCL1 and LHX8 was good (72.1% and 73.8% respectively), but the sensitivity of ST6GAL-NAC5 was low (55.7%). In line with previously described methylation makers, these markers were highly accurate for the detection of cervical cancer: all carcinomas in this study tested positive for single markers ASCL1 or LHX8, and 89.9% tested positive for ST6GALNAC5 [21,22,34]. This indicates that   single methylation marker analysis of ASCL1 or LHX8 would be an interesting primary cervical screening tool in WLHIV in low-resource settings, as it detects the majority of CIN3 lesions that need treatment and provides a high reassurance against cervical cancer. Previous studies described a relationship between HIV and hypermethylation of host cell genes, resulting from an upregulation of DNA methyltransferase expression and activity in HIVinfected cells [35,36]. This may explain the rather low CIN3 + specificity of 49.6% detected in our previous study analysing the methylation marker panel CADM1, MAL, and miR124-2 [14]. Also in this study, the methylation levels for ASCL1, LHX8 and ST6GALNAC5 were relatively high in control samples of HIV-infected women compared to HIV-uninfected women. In contrast to our previous publication, we were able to define hypermethylation positivity thresholds for optimal performance for CIN3 + detection in WLHIV and accomplished a relatively high CIN3 + specificity, combined with a good sensitivity.
In a setting with limited resources, it is important that women with the highest risk for cervical cancer receive suitable treatment. Early ART initiation and sustained adherence seem to reduce the risk for cervical cancer and its precursor lesions in WLHIV [37], however future studies need to identify which HIV-infected individuals can be screened less frequently [38]. Previous studies have described that methylation marker analysis is specifically sensitive for cervical cancer and cervical lesions caused by a long-standing (>5 years) persistent hrHPV infection, so called advanced transforming lesions with a high short-term progression risk. [22,34,39,40] The increase in methylation levels and marker positivity of ASCL1, LHX8 and ST6GALNAC5 with severity of the underlying lesion, the high sensitivities in an gynaecological outpatient setting and the extremely high methylation levels in cervical carcinomas described in this study, are in line with this concept. This suggests that methylation analysis of these genes identifies lesions with a cancer-like methylation profile that are in need of treatment.
The simplest approach in low-resource settings would be to treat all hrHPV positive women, as it yields a high sensitivity [12,41]. Our previous data confirmed the high sensitivity (83.6%) for primary hrHPV testing in WLHIV, but at moderate specificity (67.7%). If such a strategy is implemented without triage testing, a large number of hrHPV positive women (~50%) would receive unnecessary treatment, due to the transient nature of most HPV infections [14]. Hence, any triage strategy would require a large amount of secondary tests in a setting with a high HPV prevalence. Subsequent partial hrHPV16/18 genotyping is feasible as it carries the advantages of a molecular test, but with the disadvantage of missing about 30% of non-HPV16/18 related cervical cancers. Stratification of hrHPV positive women by methylation marker analysis has previously been shown to overcome this issue since it consistently detects all carcinomas [14,30]. This study shows that primary methylation analysis of ASCL1 or LHX8 has the same benefit, but without the need for additional triage testing.
An objective and reproducible "see-and-treat" strategy for cervical screening in LMIC, which particularly reduces loss to follow-up, is within reach as methylation assays can be further developed into point-of-care tests. Visual inspection with acetic acid, often combined with visual inspection with Lugol's iodine (VILI), is currently recommended by the WHO guidelines for a see-and-treat protocol in LMIC [19]. Although this approach is cost-effective and reasonably safe, the subjectivity of the diagnosis influenced by the healthcare provider's experience and environmental conditions, limits the reliability of the technique and leads to under and overtreatment [41][42][43][44][45]. Methylation marker analysis of ASCL1 or LHX8 on the other hand, is objective, and is applicable on both cervical scrapes and self-collected cervical-vaginal material [26,27,46]. Although promising, methylation assays are still relatively costly and labour-intensive, and need further implementation studies. A standardized, easy and robust high-throughput workflow is needed before implementation in cervical screening can be realised.
Limitations of this study include the cross-sectional set up as we are still awaiting clinical follow-up data. Second, to compare hypermethylation levels between HIV seropositive and seronegative women, we enriched the small study group of seronegatives with a Dutch reference population and possible population effects cannot be excluded [47]. To differentiate between an HIV effect and a population effect, these comparisons should be repeated with an African HIV uninfected control group. Third, we developed suitable thresholds for scoring methylation marker positivity in the same population as used for the performance analyses of these marker thresholds. A leave-one-out cross-validation approach was used to facilitate these performance analyses. Accordingly, the applicability of ASCL1 or LHX8 as primary cervical screening tool requires further validation of this panel in an independent cohort.
This study shows that hypermethylation analysis of ASCL1 or LHX8 is a promising method for cervical screening in WLHIV, as well as in an outpatient population. The high sensitivity and high specificity for cervical cancer and CIN3 of these methylation assays, plus their applicability to self-collected cervical-vaginal material, make the test a promising screening tool for LMIC and warrants further investigation and development.  All authors had full access to all of the data in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis and believe that the manuscript represents honest work. CJLMM affirms that the manuscript is an honest, accurate and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.

A C K N O W L E D G E M E N T S
We gratefully acknowledge all the women participating in the described study trial. Also, we are grateful for the active cooperation of the teams of the clinics of the department of Obstetrics and Gynaecology of the Steve Biko Academic Hospital and the HIV clinic at the Tshwane District Hospital, in particular Erika Breytenbach for her contribution in patient inclusions. Furthermore, we thank all the research staff members and technicians of the Department of Pathology of the VU Medical Center.