Unveiling the epigenomic mechanisms of acquired platinum‐resistance in high‐grade serous ovarian cancer

Resistance to platinum‐based chemotherapy is the major cause of death from high‐grade serous ovarian cancer (HGSOC). We hypothesise that detection of specific DNA methylation changes may predict platinum resistance in HGSOC. Using a publicly available “discovery” dataset we examined epigenomic and transcriptomic alterations between primary platinum‐sensitive (n = 32) and recurrent acquired drug resistant HGSOC (n = 28) and identified several genes involved in immune and chemoresistance‐related pathways. Validation via high‐resolution melt analysis of these findings, in cell lines and HGSOC tumours, demonstrated the most consistent changes were observed in three of the genes: APOBEC3A, NKAPL and PDCD1. Plasma samples from an independent HGSOC cohort (n = 17) were analysed using droplet digital PCR. Hypermethylation of NKAPL was detected in 46% and hypomethylation of APOBEC3A in 69% of plasma samples taken from women with relapsed HGSOC (n = 13), with no alterations identified in disease‐free patients (n = 4). Following these results, and using a CRISPR‐Cas9 approach, we were also able to demonstrate that in vitro NKAPL promoter demethylation increased platinum sensitivity by 15%. Overall, this study demonstrates the importance of aberrant methylation, especially of the NKAPL gene, in acquired platinum resistance in HGSOC.

samples from an independent HGSOC cohort (n = 17) were analysed using droplet digital PCR. Hypermethylation of NKAPL was detected in 46% and hypomethylation of APOBEC3A in 69% of plasma samples taken from women with relapsed HGSOC (n = 13), with no alterations identified in disease-free patients (n = 4). Following these results, and using a CRISPR-Cas9 approach, we were also able to demonstrate that in vitro NKAPL promoter demethylation increased platinum sensitivity by 15%.
Overall, this study demonstrates the importance of aberrant methylation, especially of the NKAPL gene, in acquired platinum resistance in HGSOC.

K E Y W O R D S
acquired drug resistance, DNA methylation, high-grade serous ovarian cancer, liquid biopsy, targeted epigenetic editing What's new?
Despite an initially favourable response, most patients with high-grade serous ovarian cancer (HGSOC) develop resistance to standard platinum-based chemotherapy. Predicting platinum resistance via monitoring tools such as epigenetic markers could improve therapeutic approaches and outcomes in HGSOC. Here, to discover promising markers, the authors explored novel mechanisms of platinum resistance in HGSOC. Circulating tumour DNA from HGSOC patients exhibited dynamic changes in methylation patterns specifically in three genes, PDCD1, NKAPL and APOBEC3A. The changes were associated with therapeutic response, with NKAPL potentially modulating chemoresistance. The findings identify methylation patterns as a promising early predictor of platinum resistance in HGSOC.

| INTRODUCTION
Epithelial ovarian cancer (EOC) is the most lethal of all gynaecological malignancies, with over 200 000 deaths in 2020. 1 High-grade serous ovarian cancer (HGSOC) accounts for $70% of all EOC cases and is characterised by late-stage diagnosis and an aggressive aetiology. 2 The standard therapeutic approach is cytoreductive-surgery, combined with platinum-based chemotherapy. 3 Whilst patients typically have an initial favourable response, >80% of women relapse, and eventually succumb to platinum resistant disease, the most important contributor to the high mortality associated with HGSOC. 4 Several mechanisms involved in the development of platinum resistant HGSOC have been proposed, including epigenetic alterations, [5][6][7] with DNA methylation and concurrent transcriptional changes associated with therapeutic response reduction in HGSOC. [8][9][10] Examples include the hypermethylation of the MLH1 promoter, a member of the mismatch repair system, which has previously been detected in plasma and is associated with poor survival following chemotherapy, 8 and BRCA methylation, which has been associated with increased survival rates and prolonged response to chemotherapy. 11,12 In addition, at present, there is no accurate method to predict drug resistance in HGSOC. The advent of liquid biopsies, in particular the use of circulating tumour DNA (ctDNA), has now shown promising results as a potential minimally invasive, sensitive and accurate monitoring tool in multiple cancer types. 13,14 Hence, the feasibility of incorporating DNA methylation markers, in combination with liquid biopsies, into patient care should be studied in the HGSOC setting, as it could offer a reliable and sensitive avenue to monitor therapeutic response.
In this study, we used a publicly available dataset to study epigenomic dysregulation in acquired drug resistant HGSOC and employed a longitudinal assessment of methylation dynamics in plasma samples from HGSOC patients to explore the potential of using epigenetic markers as monitoring tools of acquired resistance in this setting.
Finally, functional analysis of site-specific epigenetic editing of two genes (APOBEC3A and NKAPL) was carried out to investigate a potential causative effect between DNA methylation and platinum response in vitro.

| Clinical cohorts
Tumour data used in a "discovery cohort" were acquired from platinum-sensitive (n = 32; mean age = 58.1 years) and acquiredresistant HGSOC individuals (n = 23, with a total of 28 samples; mean age = 60.6 years), as previously reported (Table S1). 5 A number of primary resistant (n = 49) and control samples (n = 7), from the same dataset, were also used for comparison purposes (Table S2). Clinical criteria for including patients/samples in each group can be found in Data S1.
A "prospective longitudinal cohort" comprised of pre-platinum matched primary and metastatic tumour tissue and peripheral blood plasma samples (baseline; BL) from 17 patients (mean age = 55.8 years) attending the Mater Misericordiae University Hospital (Dublin, Ireland). Subsequent peripheral blood samples were collected after systemic chemotherapy (6 cycles of carboplatin and paclitaxel) was completed, at 6-monthly intervals (follow-up, FU) (Table S3). A one-time peripheral blood sample was also collected from healthy volunteers (n = 20, mean age = 52.65 years) (Table S3).

| Differential methylation and gene expression analysis
HM450K raw methylation profiling data used for the discovery dataset was downloaded from the Gene Expression Omnibus (GEO; GSE65821) and processed using minfi, as previously described 15,16 (Data S1). Pre-processed data was downloaded for the gene expression dataset, with samples having been sequenced as paired-end 100-bp on a HiSeq2000 instrument. 5 Differentially methylated probes (DMPs) and regions (DMRs) were obtained by comparing methylation patterns between primary sensitive and relapsed resistant HGSOC samples, using the limma 17 and DMRcate 18 packages, respectively. Age was included as a covariate in the differential methylation analysis model (as well as the gene expression model). Individual probes were considered significantly differentially methylated if FDR P ≤ .05. Selection of DMRs required a mean absolute β-value difference of ≥0.1 between groups, FDR P ≤ .05, and a minimum of ≥3 hyper-or hypo-methylated DMPs within 1 kb of each other.
Investigation of differentially expressed genes (DEGs) was carried out using the limma-trend function, in which the mean-variance of samples is incorporated into the empirical Bayes procedure. Logtransformed and normalised counts were provided as input for this model, and genes were considered differentially expressed between groups if there was an absolute log 2 fold change ≥1 and FDR P ≤ .05.
Genes were annotated using Ensembl (hg19). Overlap of both datasets identified genes that were concurrently differentially methylated and expressed (DMEGs).
Gene set enrichment analysis was conducted using missMethyl 19 and clusterProfiler 20 packages, for the DMRs and DEGs datasets (respectively). In both instances, enrichment was assessed among the Kyoto encyclopedia of genes and genomes (KEGG) pathways. Pathways were considered significantly enriched if FDR P ≤ .05.   (Table S4).

| PCR and Sanger sequencing
Amplification of APOBEC3A, NKAPL and PDCD1 was carried out with 30-50 ng of BC DNA and EpiMark Hot Start Taq DNA polymerase (New England BioLabs), using the HRM primers (Table S4). The mutational status of TP53 was assessed in primary and metastatic tumours, using previously described primer sets, 23 and Platinum II Hot-Start PCR Master Mix (Invitrogen). All PCR-amplified products were sequenced at Eurofins Genomics. Sequencing analysis was done using the Unipro Gene software (v38.1), and methylation percentage at each CpG site was calculated as follows: [peak height cytosine/(peak height cytosine + peak height of thymidine)]*100.

| Flow cytometry
Chemotherapy-naïve primary tumour and metastatic tissue samples from a total of seven HGSOC patients were labelled with fluorescent antibodies against CD3, CD4, CD8 and PD1 markers (BD Biosciences or BioLegend). Dead cells were excluded by live/dead staining using Zombie Aqua or Zombie Near IR (BioLegend). CD4+ and CD8+ T cells were identified by gating on live CD3+ lymphocytes. Stained cells were analysed on a FACS Canto I, FACS Canto II (BD Biosciences) or Aurora (CytekBio) cytometer. Flow cytometry data were analysed using FlowJo software (BD Biosciences).

| Droplet digital PCR
APOBEC3A, NKAPL and PDCD1 methylation status was assessed in BL and FU plasma samples from patients and age-matched healthy volunteers. The presence of specific TP53 mutations was confirmed in plasma and PBMCs, for the six cancer patients for whom a mutation was found during sanger sequencing analysis. Methylation assays were composed of primers/probe set for the methylation target and primers/5 0 -HEX probe set for ACTB. TP53 mutation assays comprised a single pair of primers, along with two probes discriminating between wild type and mutant templates (Table S4). Droplet digital PCR (ddPCR) reactions were performed using the QX200 droplet generator and reader (Bio-Rad). QuantaSoft Analysis Pro software (Bio-Rad, v.1.0.596) was used for analysis. Limit of blank and limit of detection were calculated for each assay (Data S1). Only wells with >10 000 droplets were considered, and thresholds for positivity were set manually and used to calculate predicted ratio (methylation assays) and mutant allele fraction (MAF; TP53 assays). Classification of markers as hyper-or hypomethylated at FU was established as a 0.05 ratio increase/decrease, respectively, observed at any FU timepoint.

| Statistical analysis
Statistical analysis was performed using GraphPad software (v.9.3.1) and P ≤ .05 was considered statistically significant. Normal distribution assessment was carried out before any statistical inference by Shapiro-Wilk's test. Correlation analysis was evaluated by Pearson's test. Comparisons between two groups were carried out with a Mann-Whitney test, while one-way ANOVA/Kruskal-Wallis tests were used when more than two groups were compared (depending on normal distribution test results). A two-way ANOVA was carried out to analyse differences in cell viability after epigenetic editing in cell lines. Adjustment for multiple testing was performed for comparisons between groups using the Benjamini-Hochberg FDR method.

| In silico analysis reveals association between dysregulated methylation patterns and response to platinum therapy in HGSOC patients
To investigate the relationship between DNA methylation and acquired drug resistance in HGSOC, we compared methylation and gene expression patterns in a discovery dataset, comprising primary platinum sensitive (n = 32) and relapsed platinum resistant samples (n = 28) (Table S1). 5  hypomethylated and upregulated ( Figure 1C and Table S5). Interestingly, the methylation patterns of these genes were only significantly different between primary sensitive and relapsed resistant samples, with no difference between primary sensitive and primary resistant samples ( Figure S2).
Focusing on the 12 patients with paired primary and relapsed samples (Table S1), seven of the eight genes demonstrated a negative correlation between methylation and gene expression, statistically significant for NKAPL, C1QTNF4, THY1 and APOBEC3A ( Figure 1D and Figure S3). Additionally, we found a trend towards shorter time to relapse associated with higher numbers of these eight dysregulated DMEGs, except for one patient who demonstrated an exceptional response of over 3.5 years (AOCS-088) ( Figure 1E).

| Presence of dynamic methylation changes upon treatment with platinum agents
The platinum-associated methylomic changes were further investi- We also observed methylation differences in these markers between primary and metastatic tumour samples. Interestingly, those differences were also present pre-platinum administration, possibly indicating clonal heterogeneity. DNA methylation differences observed between primary and metastatic samples has been previously covered by several recent studies. [26][27][28] In fact, one of those reports highlights that these differences seen between matched primary and metastatic samples can be the result of clonal divergence from the primary tumour, 28 1 and 1, respectively). Green dots indicate genes that are significantly downregulated and hypermethylated (6 genes; lower right quadrant) and upregulated and hypomethylated (two genes; upper left quadrant) (D) A representative example of one of the eight DMRs in which hypermethylation (shown in green) was found to be inversely correlated with gene expression (shown in blue). This was analysed in a sub-cohort of matched sensitive and acquired-resistant HGSOC tumour samples (n = 12). Missing data are indicated (*) (E) Presence of a higher number of abnormal methylation patterns across the 8-marker panel was associated with shorter time to relapse Abnormal methylation patterns are defined as methylation gain (in PDCD1, THY1, NKAPL, C1QTNF4, DIO3, EOMES) or methylation loss (in APOBEC3A and S100A8) when comparing paired sensitive and acquired drug resistant samples within the same subject (n = 12). Only one patient (indicated by green arrow; outlier) displayed a different pattern, with methylation dysregulation being associated with a longer time to relapse. Spearman correlation was carried out without the outlier patient (n = 11). therapy, and the association with platinum resistance in the discovery dataset, may reflect a reduction in immune cell infiltration, which is associated with platinum resistance. 29

| Longitudinal monitoring of methylation patterns via cfDNA is associated with relapse status in HGSOC patients
Following these results, we set out to evaluate the presence these epigenetic changes in APOPBEC3A, NKAPL and PDCD1 in an independent cohort of HGSOC patients treated with platinum-based chemotherapy. Peripheral blood plasma was obtained from 17 chemo-naïve HGSOC patients at BL, and at 6-monthly intervals after commencing chemotherapy (FU) ( Table S3). The yield of cfDNA was significantly higher in cancer patients than age-matched healthy controls ( Figure 3A), consistent with previous studies. 30 Investigation of the methylation patterns of APOPBEC3A, NKAPL and PDCD1 in cfDNA was assessed by ddPCR ( Figure S8). Overall, PDCD1 and APOBEC3A methylation showed similar levels between cancer patients (BL) and healthy volunteers. However, for NKAPL, there was a trend towards higher levels of DNA methylation in some cancer patients ( Figure 3B).

Comparison of methylation levels between BL and FU timepoints
showed NKAPL hypermethylation in 46% of the relapsed samples (6/13), and APOBEC3A hypomethylation in 69% of relapsed cases (9/13) ( Figure 3C). Notably these methylation patterns in APOBEC3A and NKAPL were not observed in any of the patients that did not relapse (n = 4). PDCD1 was broadly methylated in cfDNA from patients irrespective of their disease status.
We next interpreted the plasma DNA methylation findings in the context of ctDNA. To assess the proportion of ctDNA in each plasma sample, we sequenced all the tumours to identify TP53 mutations, as these are commonly found in HGSOC. TP53 mutations were identified in only 35% of the primary tumours (6/17), which suggest that this is not a viable approach to assess ctDNA in our cohort (Table S6). Nevertheless, we did evaluate tumour proportion in the plasma samples of the patients for whom we found TP53 mutations. ddPCR assays were designed to detect mutational frequencies in their tissue, PBMC and plasma samples ( Figure S9), and as expected no mutations were found in any of the PBMC samples, confirming the somatic nature of these mutations. A total of 14 plasma samples were collected between these six patients with TP53 mutations. Notably, it was possible to observe that ctDNA proportion has a great impact on methylation dynamics. This observation is noticeable in three relapsed patients for which we found no NKAPL hypermethylation at follow-up (OV006, OV027, OV044).
Assessment of ctDNA unveiled that no tumour proportion was detected in any of these patients' follow-up samples, which then affected our ability to detect any NKAPL methylation, given that this signal is specific to tumour DNA ( Figure 3D). Thus, it is imperative that any clinical interpretation of methylation findings considers ctDNA proportion. Although validation in a larger cohort is necessary, our results suggest that NKAPL methylation patterns mimic our previous observations in the discovery cohort, and that these can be monitored using liquid biopsies.  Table S4). Cells were subjected to antibiotic selection (G418), to increase the abundance of transfected cells ( Figure S10).

| Targeted demethylation of NKAPL in vitro demonstrates involvement of DNA methylation in an acquired drug resistance EOC model
Sanger sequencing showed little to no methylation differences between the untransfected and plasmid control (no gRNA) cells, which excluded the possibility of any observed demethylation activity being due to the dCas9-TET1CD complex itself ( Figure 4B and Figure S11A).
Although some demethylation was achieved at the APOBEC3A promoter (max $15% difference, mean three CpG sites, gRNA1) ( Figure S11B), this did not translate into any changes in cisplatin sensitivity ( Figure S11C). Similarly, this demethylation did not result in any increase in mRNA expression of APOBEC3A ( Figure S11D). Demethylation was more successful for NKAPL ( Figure 4B). Examining the five CpG sites with highest demethylation, located within the NKAPL promoter, we obtained up to 23.5% methylation reduction in A2780 cells, and 34% in A2780/CP70 cells (mean five CpG sites, gRNA2), with some CpG sites showing differences of up to 50% ( Figure 4C). Moreover, removal of methylation significantly upregulated NKAPL mRNA expression in both A2780 and A2780/CP70 cells ( Figure 4D). As expected, NKAPL demethylation had no impact on A2780 cisplatin sensitivity. However, NKAPL demethylation significantly increased the platinum sensitivity of the resistant A2780/CP70 cell line, across a range of concentrations, with a maximum 15% reduction in viability (5 μM; gRNA2) ( Figure 4E). These findings show that alteration of the NKAPL promoter methylation status translates into a partial reversal of platinum resistance. To rule out the possibility that the changes in response to cisplatin were due to off-target effects of the gRNAs, we examined the expression of off-target genes identified through in silico prediction. No significant changes in gene expression were found ( Figure S12 and Table S7).
To further explore the mechanism behind this potential NKAPL methylation-induced drug resistance, we analysed the expression of several Notch signalling pathway downstream genes, as NKAPL has previously been identified as an inhibitor of this pathway. 31 A total of seven targets were selected, and analysis of the results showed no significant differences in mRNA expression of these targets in the demethylated samples ( Figure 4F), results that are on par with what was found in the "discovery" cohort for the same markers (Table S8).
F I G U R E 4 Legend on next page.

| DISCUSSION
The link between DNA methylation and the emergence of platinum resistance in HGSOC has been widely reported over the years, with several studies identifying key pathways potentially involved in HGSOC progression to a platinum-resistant state. [32][33][34] Our goal was to expand on the current knowledge regarding epigenomic regulation of therapeutic response in HGSOC and to detect these changes in plasma. Using a publicly available data resource, 5 we identified 100 genomic regions that were differentially methylated between primary sensitive and acquired-resistant HGSOC and subsequently identified eight genes where methylation was found to negatively correlate with gene expression. We identified a trend towards an inverse correlation between the time to relapse and the proportion of methylation changes in the matched cohort, which hints at a potential prognostic role for these markers. Most of the acquired-resistant samples were from ascites ($79%). The use of this sample type for research and diagnostic purposes is a common occurrence in the EOC setting, as secondary cytoreductive surgery is seldom performed in relapse patients. Although ascitic fluid contains a range of tumour and non-tumour cellular sources, 35 in this instance the fluid was processed specifically to isolate tumour cells, allowing the comparison with tissue samples and eliminating any other sources that might otherwise hinder our analysis. Nevertheless, there is a possibility, albeit small, that our results could be due to differences between sample types, and as such, further validation is warranted.
Cell line models of platinum resistance did not recapitulate methylation changes identified in the discovery cohort. This is not entirely surprising, mostly due to some of our markers being associated with immune-related pathways (ie, PDCD1, C1QTNF4, EOMES). However, we observed methylation changes in three of the genes in response to platinum-based chemotherapy: two hypermethylated (PDCD1 and NKAPL) and one hypomethylated (APOBEC3A). The role of PDCD1, a gene encoding the immune checkpoint programmed cell death 1 (PD-1), has been widely studied in several cancers due to its central role in immune checkpoint inhibitor therapies (ICI). 36 In fact, hypermethylation of PDCD1 has been shown to potentially predict response to ICIs in head and neck cancer, and it might be a good prognostic marker in low grade gliomas. 37,38 However, the role of this gene's epigenetic regulation in HGSOC is still unknown. In our study, we observed DNA methylation changes between primary and metastatic samples, both before and after administration of platinum therapy. Pre-platinum methylation differences between these two types of samples, which translate into less PDCD1 expression in metastatic samples, could potentially explain the poor response to ICIs observed in HGSOC patients, since most of these studies were conducted on the primary tumour tissue. Further studies investigating the metastatic tissue profile of HGSOC patients prior to ICI administration should be carried out. Our results thus suggest that PDCD1 methylation could potentially have a dual function, both as an indicator of chemotherapeutic response, and a potential predictive marker for ICI therapeutic response. Compared with PDCD1, information surrounding NKAPL and its involvement in the cancer setting is very limited. Although methylation of NKAPL has been reported to be associated with poor prognosis in both breast cancer and hepatocellular carcinoma, 39,40 it is its essential role in spermatogenesis that has gathered the most scientific attention. 31 Interestingly, Okuda et al 41 reported that NKAPL acts as a transcriptional suppressor of the Notch signalling pathway, which in turn negatively affects spermatogenesis. More recently, this link was also observed in a study related to osteosarcoma. This suggests that NKAPL methylation association with chemoresistance in HGSOC might be due to aberrantly elevated Notch signalling, a feature that has been reported in this setting. 42 The third marker, APOBEC3A, has been widely studied in the context of therapeutic resistance. Responsible for encoding one of the members of the cytosine deaminase family of proteins, APOBEC3A is best known for its involvement in innate immunity inhibition of viral infection. 43 However, abnormal expression of APOBEC proteins has been reported in a number of cancers, 44,45 and was identified as a major mediator of a specific type of mutagenesis, often denominated as the APOBEC-mutational signature. 46   We also investigated the possibility of using the identified markers to monitor HGSOC during therapeutic administration via blood. Our initial longitudinal assessment showed that all three markers exhibited dynamic changes (hypermethylation for PDCD1 and NKAPL; hypomethylation for APOBEC3A) in most relapsed patients, while NKAPL and APOBEC3A also demonstrated the opposite patterns in non-relapse patients, alluding to the potential of using methylation dynamics to distinguish between these two groups of patients. However, while genomic mutations are commonly tumour-specific, methylation patterns are known to be cell-type specific. 48,49 Moreover, a great proportion of cfDNA comes from non-tumour sources, thus assessment of the cfDNA methylome without prior knowledge of tumour's content proportion can lead to erroneous conclusions. In our study, the mutational status of TP53 was assessed to uncover information regarding ctDNA proportions, with 35% of the patients displaying mutations in this gene. TP53 mutations are usually observed in >90% of HGSOC cases, 50 thus the lower frequency observed in our cohort might be due to the lower detection sensitivity of the technique used to assess these mutations (ie, sanger sequencing) or the possible lower tumour purity in some of the tissue samples. Additionally, only point mutations in hotspot locations (exons 4-9) were investigated. While an in-depth analysis in context of ctDNA was not possible, due to the limited number of samples with tumour content, we were able to conclude that methylation changes were affected by this occurrence, and as such cfDNA methylation data analysis should always be accompanied by ctDNA assessment. Notably, we also found that PDCD1 exhibited a hypermethylated background in normal cfDNA samples, which implies that the distinction of a PDCD1 hypermethylated-tumour state might not be possible, diminishing the potential utility of this marker as a blood-based predictive tool.
Most studies examining the relationship between epigenetic alterations and platinum resistance used demethylating agents, which leads to genome-wide demethylation of CpG sites, making it challenging to understand the causal effect of epimutations. In this study, we address this issue by using CRISPR technology to perform site-specific demethylation. 24 Demethylation of the NKAPL promoter increased platinum response of a resistant cell line by 15%. To the best of our knowledge, this is the first report of targeted methylation editing in EOC, thus unequivocally showing a direct relationship between promoter methylation-associated silencing of a specific gene and partial reversal of acquired chemoresistance. The relatively modest reduction could potentially be augmented by using different gRNAs or increasing the transfection efficiency. However, this might also be associated with the multi-factorial nature of drug resistance in cancer, thus implying that NKAPL editing alone is not capable of inducing a higher degree of sensitivity. Nevertheless, these results confirm the potential role of this target in a methylation-induced resistant state. Additionally, investigation of several key components of the Notch signalling pathway, following NKAPL demethylation, revealed no gene expression changes in any of the evaluated downstream targets. This might indicate that the previously reported NKAPL repressor activity of the Notch signalling is context-specific. Further analysis will need to be conducted to fully understand how NKAPL methylation changes affect therapeutic response in EOC. Of note, due to the hypothesis that PDCD1 methylation changes might be associated with immune cells, we decided not to investigate the effect of methylation editing of this gene, as cell line models lack an immune environment. This relationship should be further studied using either co-culture or patientderived organoids/explants to fully understand changes in PDCD1 methylation in the context of HGSOC platinum therapy.
Our study has three main limitations. Firstly, the effects of NKAPL and APOBEC3A methylation editing were studied using A2780 and A2780/CP70 cells. Despite being one of the most commonly used cell line models for EOC, their histopathology origin has recently been proven not to be HGSOC. 51 Validation of our results in three HGSOC cell lines (UWB1.289, COV318 and OVCAR-4), using the same methodology applied to A2780 cells, was unsuccessful (reduced cell viability after transfection-data not shown). This was likely due to toxicity in response to the transfection agent, and further optimization is necessary to recapitulate the results observed in A2780 cells. Furthermore, A2780/CP70 cells are derived from the parental cell line by chronic exposure to cisplatin, a condition that could be considered non-physiological due to patients being commonly exposed to a brief three cycle regimen. Studying the effect of platinum treatment in patient-derived HGSOC organoid/explant models is thus warranted, as it could provide a more realistic overview of real-time platinum exposure in these patients. Secondly, PDCD1 and APOBEC3A ddPCR results showed methylation ratios above the expected. As this did not happen in tissue or control samples, we hypothesise that may be due to cfDNA fragmentation. Although this does not affect our results, as all plasma samples were run with the same assays and are, therefore, comparable, redesigning the housekeeping gene for a no-CpG-region within the target gene might offer a more accurate reading. Finally, although our longitudinal cohort was composed of 17 patients, ctDNA was only assessed in six, hence further validation of our results in a larger cohort is needed to fully understand the predictive potential of these markers in HGSOC.

| CONCLUSIONS
In summary, we found that significantly dysregulated methylation patterns are associated with acquired drug resistance in HGSOC, with three specific genes (PDCD1, NKAPL and APOBEC3A) showing marked dynamic changes in response to treatment with platinum drugs after validation. Finally, using a targeted epigenetic editing approach, alongside a longitudinal plasma methylation assessment, we pinpointed NKAPL as a potential modulator of HGSOC chemoresistance.
Although validation in larger cohorts is warranted, our results highlight the importance of aberrant DNA methylation in HGSOC progression and platinum resistance.

AUTHOR CONTRIBUTIONS
The work reported in the paper has been performed by the authors, unless clearly specified in the text. Conceptualization: Antoinette S. Perry, Donal J. Brennan; Methodology: Romina Silva, Bruce Moran,