1. Introduction
High-grade serous ovarian cancer (HGSOC) is the most common (75%) histological type found in advanced-stage epithelial ovarian cancer (EOC) [
1,
2]. The different histological subtypes of ovarian cancer, such as HGSOC, low-grade serous, endometrioid and clear cell are characterized by differences in their molecular profile. HGSOC is characterized by mutations in
TP53 in >90% of cases and by mutations in
BRCA1 or
BRCA2 in around 20% of cases. Other genomic alterations include copy number variations. Most patients are treated with chemotherapy and debulking surgery as the standard of care, but many patients develop disease recurrence and eventually die from the disease. Clearly, there is a need for improvement.
One of the most successful strategies has been the use of poly-ADP ribose-polymerase (PARP) inhibitors (PARPi), especially for patients with a somatic or germline BRCA mutation. Since the approval of the first PARPi in 2014, initially approved as a maintenance therapy after the response to chemotherapy for patients with a BRCA mutation, these drugs have expanded to encompass first-line maintenance treatment in EOC patients with or without a BRCA mutation. In parallel, major efforts are ongoing to select which patients will respond best to PARPi therapy.
A few studies have investigated blood cell-free DNA (cfDNA) from ovarian cancer patients and linked cfDNA alterations to PARPi resistance. Most of these studies observed genomic changes [
3,
4,
5,
6,
7,
8,
9,
10]. These studies showed
BRCA1/2 reversion mutations in cfDNA from PARPi-resistant patients [
5,
8,
9]. Moreover, an
MRE11 p.K446R mutation was also frequently found in cfDNA from patients resistant to Olaparib [
6], and one study showed in cell line models that this mutation resulted in the reduced accumulation of cellular DNA damage [
3]. Only one study analyzed cfDNA methylation in PARPi-treated patients and showed that
HOXA9 cfDNA methylation was related to poor outcomes in patients receiving PARPi [
11].
Our hypothesis is that cfDNA epigenetic and genomic markers can help to predict resistance to PARPi in patients with a first recurrence of EOC. This study aims to simultaneously evaluate cfDNA genomic and methylation alterations in EOC patients and relate these changes to PARPi resistance. Therefore, we evaluated blood cfDNA before and after the PARPi maintenance therapy of 31 HGSOC patients with and without therapy resistance using several genome-wide Next-Generation Sequencing (NGS) assays.
3. Results
The main goal of this study was to characterize the genomic and epigenetic hallmarks of PARPi therapy resistance as well as disease progression in HGSOC patients. The design of this study is shown in
Figure 1. Briefly, genome-wide NGS was performed on 31 HGSOC patients, namely on blood plasma cfDNA from PARPi-resistant (
n = 15) and -sensitive patients (
n = 16) (
Figure 1A). All patients received platin-based chemotherapy prior to PARPi therapy (
Figure 1B). Eleven patients displayed a
BRCA1/2 wildtype in the germline and tumor DNA, equally distributed amongst the two PARPi response groups (Supplemental
Table S1). HBDs and the tissue and treatment-naïve blood of ten HGSOC-patients were also used and analyzed as reference sets. To detect genomic alterations, pre- and post-treatment blood was evaluated by means of mFastSeq, while post-treatment blood was also analyzed by means of shWGS and exome-seq. MeD-seq was performed on all samples from HGSOC patients and HBDs to determine DMRs. These DMRs were established in pre- and post-treatment blood relative to HBDs and were used to define three different epigenetic PARPi hallmarks. Unique DMRs from pre-and post-treatment blood representing methylation changes during disease progression and PARPi treatment were defined as dynamic tumor evolution hallmarks (DTE). Unique DMRs from PARPi-resistant and -sensitive patients in pre-treatment samples taken before disease progression and PARPi-treatment were defined as PARPi predictive hallmarks (PP). Unique DMRs from PARPi-resistant and-sensitive patients in post-treatment samples taken after disease progression and/or PARPi-treatment were defined as PARPi response hallmarks (PR). The study is described in more detail below.
3.1. General Patient and Blood Characteristics
The clinicopathological characteristics for the 31 HGSOC patients were compared between PARPi-resistant (
n = 15) and -sensitive patients (
n = 16) (
Table 1). None of the parameters were significantly different between the two patient subsets, except for the PARPi response, which was used to categorize patients into the two PARP response groups. In addition, cfDNA yields per mL of plasma were compared between blood collection timepoints (before and after treatment) and patient subsets. No significant differences were seen in cfDNA yields between pre- and post-treatment blood (median: 7.5 ng/mL and 8.9 ng/mL; Student
t-test,
p = 0.45). The cfDNA yields between resistant and sensitive patients were comparable in pre-treatment blood (median: 7.3 and 7.6 ng/mL) and increased in post-treatment blood (median: 11.0 and 8.0 ng/mL), but were not different at both timepoints between the two patient subsets (Student
t-test,
p = 0.41 and
p = 0.18), respectively.
3.2. Genomic Hallmarks in Blood cfDNA from PARPi-Resistant and -Sensitive Patients
The mFast-Seq analyses to detect circulating tumor DNA (ctDNA) via aneuploidy only revealed GWZ-scores below 5 in all pre-treatment samples, irrespective of PARPi response (
Figure 2). Blood with a GWZ-score above 5 has been reported to have a high ctDNA load [
12], and thus these findings indicate no or low levels of ctDNA in the pre-treatment samples. In contrast, six post-treatment samples displayed high GWZ-scores above 5, all except one from PARPi-resistant patients. IchorCNA analyses of shWGS and Exome-seq data from post-treatment blood found tumor fractions above 10% in seven and eleven samples, respectively (
Figure 2). Again, high tumor fractions were more often seen in PARPi-resistant (median tumor fractions of 9% and 7%) than -sensitive patients (median tumor fractions 7% and 5%) and confirmed most of the post-treatment mFast-Seq findings (Supplemental
Figure S1).
Next, mutational signatures defined by SigMiner were evaluated in post-treatment exome-seq samples. Analyses were focused on 52 mutational signatures (43 SBSs and 9 IDs) with a known mutational process (
Figure 3A) and identified 9 SBS signatures and one ID signature in ovarian PARPi samples which were not seen in HBDs (
Figure 3B,C). The observed mutational processes were linked to HR and MMR deficiency (SBS3, SBS14, SBS26, SBS44), chemotherapy (SBS17B (5-FU-related), SBS25, SBS35 (platinum)), AID activity (SBS84), duocarmycin exposure (SBS90), and colibactin exposure (ID18). The identification of SBS25, SBS84, SBS90, and ID18 was unexpected, while all other SBS signatures can be linked to the systemic therapy that the ovarian cancer patients in our cohort received. All SBS signatures except for SBS26 were seen in at least one PARPi-resistant patient, while only five (SBS17B, SBS25, SBS26, SBS44, and SBS84) were observed in PARPi-sensitive patients. SBS26 was seen in four PARPi-sensitive samples but not in PARPi-resistant samples (
p = 0.04), and alongside SBS14 and SBS44 is one of the seven signatures associated with defective DNA mismatch repair. The SBS26-positive patients displayed a
BRCA1/2 wildtype for the germline and tumor (
n = 2), a germline wild type (
n = 1), or had a germline
BRCA2 mutation (
n = 1) and were treated with niraparib (
n = 3) or olaparib (
n = 1). The SBS3 signature, which was strongly associated with germline and somatic
BRCA1/2 mutations and
BRCA1 methylation in other studies, was seen in our study in only two samples; however, both were obtained from
BRCA1/2 germline and tumor wild-type patients. The identification of chemotherapy-linked SBS35 makes sense since the patients in this cohort received platinum-based chemotherapy before PARPi maintenance therapy.
To summarize the genomic hallmarks, aneuploid cfDNA and high tumor fractions were predominantly seen in post-treatment blood from PARPi-resistant patients, while the SBS26 signature for defective DNA mismatch repair was observed in PARPi-sensitive patients.
3.3. Epigenetic Hallmarks
3.3.1. Detection of Ovarian Cancer DMRs
To find the epigenetic hallmarks for EOC and PARPi resistance, we first defined the TSS regions that are differentially methylated between disease and healthy samples using DeSeq2. For this, generated MeD-seq data from the samples included in the study were used to find DMRs with an adjusted
p-value < 0.05 across comparisons between cancer sample subsets and the HBDs as the reference subset (
Figure 4A). These analyses revealed that each cancer subset showed unique DNA methylation alterations, characterized by varying proportions of hypo- and hypermethylated DMRs. The highest numbers of DMRs were found in tissue and PARPi-resistant blood, while treatment-naïve blood had the lowest number of DMRs. All subsets more frequently had hyper- than hypomethylated DMRs.
3.3.2. Defining PARPi-Related Epigenetic Hallmarks
The above detected disease-related DMRs were compared within four cancer subclasses to define PARPi-specific epigenetic hallmarks (
Figure 4B). The first analysis compared the pre- and post-PARPi treatment subsets to find methylation changes over time during PARPi treatment and disease progression and was defined as DTE. This comparison resulted in 710 and 599 DMRs only found at the start or after treatment, respectively, while 694 DMRs were seen in both subsets. The second and third subclass compared resistant and sensitive patients to define DMRs related to PARPi therapy response in pre- and post-treatment subsets, respectively. The first PARPi-response subclass (called PP) used pre-treatment subsets, prior to disease progression and treatment outcome, thus selecting predictive DMRs. The second PARPi response subclass (called PR) used post-treatment subsets and detected DMRs at disease progression (or at treatment termination) after PARPi therapy. This latter comparison provides information on tumor biology beyond therapy failure. The comparisons in the pre- and post-treatment subclasses resulted in 1481 and 542 DMRs which were only seen in resistant patients, while 61 and 562 DMRs were exclusively seen in sensitive patients. The overlap within these two subclasses resulted in 104 and 162 DMRs. Finally, the fourth subclass was defined as disease/tumor and compared the DMRs found in treatment-naïve blood at diagnosis with the DMRs detected in matched tumor tissue. This resulted in 30 and 11,353 unique DMRs for blood and tissue, respectively, and 29 matching DMRs found in both. This finding suggests that only half of the disease-related DMRs in the blood originates from tumor cells. The DMRs from the tissue subset were, for this reason, used to select tumor-related DMRs in the PAPRi blood subsets and subclasses (
Figure 4B).
From all the tumor-related unique DMRs, three different epigenetic PARPi therapy hallmarks were defined: the DTE hallmark, composed of 352 DMRs, and the PP and PR hallmarks, composed of 304 and 247 DMRs, respectively. The PP and PR hallmarks had more tumor-specific DMRs in resistant than in responding patients in both pre- and post-treatment samples (274 vs. 30 DMRs, 190 vs. 57 DMRs, Χ
2-test
p < 0.001) (
Figure 4B). The three epigenetic PARPi hallmarks had overlapping DMRs in addition to hallmark-specific DMRs for DTE (155 DMRs), PP (126 DMRs), and PR (82 DMRs) (
Figure 4C). All DMRs from the pre-PARPi subset were also analyzed with Cox proportional hazard regression analyses to find DMR profiles associated with PARPi TTF (
p < 0.05) as a continuous variable instead of the categorical variables, resistant versus sensitive, applied in DeSeq2 (
Figure 4D). The number of TTF-associated DMRs was determined in all subsets and epigenetic hallmarks. As expected, the PP hallmarks had more TTF-related DMRs and a higher fraction (40 DMRs, 13%) than the DTE and PR hallmarks (25 and 17 DMRs, 6% and 7%) (Χ
2-tests
p < 0.03).
3.4. Exploring Epigenetic Hallmarks and DMR Linkages to Known Tumor Driver and HRD Genes
The expression profiles of the different epigenetic PARPi hallmarks were evaluated by means of hierarchical cluster analyses using Heatmapper and cluster metrics average linkage and Spearman rank correlation (
Figure 5). The DMRs from the DTE and PP hallmarks (
Figure 5A,B) resulted each in two distinct clusters of samples. The DTE hallmark DMRs did not display a specific enrichment of samples within each cluster, either for pre- versus post-treatment or for sensitive versus resistant. In contrast, the PP hallmark DMRs for pre-treatment samples displayed predominant clusters of samples from resistant (left cluster) and sensitive (right cluster) patients. The DMRs from the PR hallmarks (
Figure 5C) resulted in three clusters of post-treatment samples. Samples from resistant patients were grouped into two distinct clusters (left and middle cluster, 4 and 10 samples, respectively), while samples from sensitive patients were found in one distinct cluster (right cluster, 8 samples), while the remaining samples were divided between the other two clusters. Finally, 40 DMRs from the PP hallmarks which were significantly associated with TTF were used to cluster the pre-treatment samples. This also resulted in three clusters, with the left and right clusters mainly containing samples from sensitive (left cluster: seven out of nine samples) and resistant (right cluster: five out of nine samples) patients, respectively, while the middle cluster contained the remaining samples from each group of patients. The epigenetic PARPi hallmark DMRs were also linked to published gene lists of 722 tumor driver (TD) genes and 58 HRD genes to see whether these hallmarks contained already-known cancer- and PARPi-relevant genes. This analysis identified three TD genes that were seen in the DTE hallmarks (
SOX2,
POU2AF1, and
PTK6), while two TD genes were found in the PP hallmarks (
VHL,
JAK3) and in the PR hallmarks (
EZH2,
SOX2). Only one HRD gene (
RAD51C) was detected in the DTE hallmark, while none were detected in the other two epigenetic hallmarks.
3.5. Functional Insights from DMR Signatures by Pathway Analysis
Subsequently, pathway analysis was conducted in Metascape using custom settings to detect enriched functional categories, processes, protein–protein interactions, and transcription factor targets within the three epigenetic PARPi hallmarks. Metascape could not link all DMRs to genes, and, therefore only used 213, 194, and 155 unique genes as the input from the DTE, PP and PR hallmark DMRs (
Figure 6).
Publicly available databases were explored for functional categories and processes and enriched entities were defined and compared between the epigenetic hallmarks. The top 20 most significant enriched categories are presented in
Figure 6A. DTE was enriched for 10 functional entities, PP for 11 entities, and PR for 13 entities. Unique functional categories were observed for DTE (two entities), for PP (three entities), and for PR (four entities). Four enriched entities were related to H3K27me3, and were especially seen in PP and PR. On the other hand, three entities were linked to immune cells (M9337, M5591, and M8543) and were observed in DTE and PR. Next, nine Gene Ontology processes were enriched, with the immune system process only seen in DTE and biological regulation and growth only seen in PR. The PP hallmarks had three processes which were all also seen in DTE. Protein–protein interaction analyses identified in the most significant interaction node three proteins unique for each of the DTE and PP hallmarks and one protein for the PR hallmarks (
Figure 6B). Finally, enrichment analyses of transcription factor targets demonstrated the largest number of targets in the PP hallmarks (13 targets), while DTE and PR had only eight and four targets (
Figure 4). The
TST1 target (M19088) was seen in all epigenetic hallmarks and contains the motif NNKGAATTAVAVTDN within 4 kb around the transcription starting site of the
POU3F1,
STAT, and
NFKB targets, which were enriched in the PP hallmark, while
GATA targets were seen in the DTE hallmark, and
MYC and
ZNF targets were observed in the PR hallmark. All of these findings indicate that the three epigenetic PARPi hallmarks have subtle differences in their functional and biological processes, in their protein–protein interactions, and in their transcription factor targets.
The epigenetic hallmark DMRs were also evaluated for CHPs (
Figure 7). In total, 40 DMRs were linked to one or more of these CHPs. This analysis showed that at least three DMRs from the epigenetic hallmarks were linked either to DNA repair,
E2F targets, inflammatory and interferon γ response,
KRAS signaling and
MYC targets, oxidative phosphorylation, or
TNFA NFKB signaling (
Figure 7A). Interestingly, the interferon γ response pathway was not seen in the PP hallmark, whereas all other pathways were only detected in the resistant samples of this hallmark. The CHP-linked DMRs were predominantly hypermethylated, i.e., fold changes above zero, in the different epigenetic hallmarks when compared to HBDs (
Figure 7B).
Significant fold changes (Blue bars in
Figure 7B) were found between pre- and post-treatment samples (DTE: 5DMRs) and between sensitive and resistant patients in pre- (PP: 15DMRs) and post-treatment samples (PR: 5DMRs). Finally, another 40 DMRs from the epigenetic PP hallmarks that were significantly associated with TTF were investigated for their fold changes (
Figure 7C). Remarkably, DMRs hypomethylated compared to HBDs were associated most strongly with worse TTF (
MYO18B,
TUB-AS1, and
MFRP) or with beneficial TTF (
LA16c-444G7.1,
RP4-647J21.1,
RP1-72A23.1, and
KB-1732A1.1).
4. Discussion
The examination of epigenetic and genomic characteristics in HGSOC patients undergoing PARP inhibitor maintenance therapy provides valuable insights into treatment response and disease progression. Our study delved into (dynamic) DNA genomic and methylation changes across disease conditions and treatment responses, revealing patterns indicative of the intricate regulatory mechanisms at play. Our simultaneous genome-wide genomic and epigenetic NGS analyses revealed that PARPi-resistant patients compared to HBDs and PARPi-sensitive and treatment-naïve EOC patients predominantly displayed cfDNA hypermethylation in pre-treatment blood, while aneuploid cfDNA and high tumor fractions were merely observed in post-treatment blood.
The general workflow for our genome-wide cfDNA analyses started with the mFast-SeqS and MeD-seq analyses of all blood samples. These analyses were initially extended with shWGS combined with exome-seq only to samples with aneuploid cfDNA, but were ultimately completed in all post-treatment blood samples to reveal genomic differences at disease progression between PARPi-resistant and -sensitive patients, independent of tumor load.
Our cheap and rapid mFast-SeqS analyses of both pre- and post-treatment blood showed only aneuploid cfDNA at disease progression from mainly PARPi-resistant patients. The lack of aneuploid cfDNA in pre-treatment blood was unexpected because cancer patients with late-stage disease often have high amounts of aneuploid cfDNA in their blood at baseline, as our previous studies in metastatic cancer patients have demonstrated [
13,
18,
19]. However, the EOC patients in our PARPi cohort received (platinum-based) chemotherapy just before the start of PARPi maintenance therapy, which might explain the observed low tumor load in the pre-treatment blood. Moreover, the mFast-SeqS results at disease progression were independently confirmed by ichorCNA analyses of shWGS and exome-seq data, demonstrating higher tumor fractions in PARPi-resistant compared to PARPi-sensitive patients.
Defining mutational signatures in cfDNA is challenging using shWGS with a low read depth coverage and due to there being much more germline DNA and lower tumor fractions in blood cfDNA compared to tumor tissue genomic DNA. Therefore, in our signature analyses, we used exome-seq with higher read depth coverages (median: 182 reads) and only selected variants which were not identified in the HBDs, were detected with at least 10 mutant reads, and with VAFs below 45% and between 55% and 70%, all to eliminate germline SNPs and to enrich for tumor-specific variants as much as possible. Subsequently, we examined signatures that were not identified in the HBDs to establish tumor biology-related signatures in the cancer blood samples. Using this approach, our analyses predominantly showed defective DNA mismatch repair- and chemotherapy-linked signatures in blood samples at disease progression. These mutational signatures make sense in our cohort of patients because they received platinum-based chemotherapy followed by PARPi maintenance therapy for their first recurrence. To summarize the genomic hallmarks, aneuploid cfDNA and high tumor fractions were seen in post-treatment blood mainly from PARPi-resistant patients, while the SBS26 signature for defective DNA mismatch repair was observed in PARPi-sensitive patients. These findings seem to be contrasting and unexpected, but can be explained due to the rapid disease progression in resistant patients, resulting in higher ctDNA loads. DNA mismatch repair has been described in ovarian cancer patients [
20], but the association with PARPi therapy response, as seen in our samples, needs to be explored further in future studies.
A comprehensive analysis was performed on MeD-seq data to provide detailed insights into methylation dynamics across different contexts, shedding light on disease progression, therapy response, and predictive markers [
21]. For this, all cancer blood and tissue samples were compared to blood from HBDs. We observed that tissue samples exhibited a more balanced distribution of hypermethylation and hypomethylation among DMRs, in contrast with the consistent hypermethylation observed in blood samples across different time points. These observations emphasize the importance of considering tissue heterogeneity in epigenomic research. Interestingly, pre- and post-resistant PARPi samples predominantly displayed hypermethylated DMRs, suggesting a potential association between hypermethylation and disease progression and/or drug resistance. Further investigations into the molecular mechanisms underlying this association are warranted.
Our research extensively explored methylation profiles across three distinct subclasses of sample subsets and defined different epigenetic PARPi hallmarks. Unique DMRs in pre- and post-treatment samples represented longitudinal changes in methylation due to disease progression and PARPi response. Therefore, these DMRs were defined as epigenetic DTE hallmarks. Next, DMRs found before disease progression and PARPi treatment in pre-treatment samples only seen in resistant or in sensitive patients were defined as epigenetic PARPi predictive (PP) hallmarks. Finally, DMRs found after disease progression and PARPi treatment in post-treatment samples of either resistant or sensitive patients were defined as epigenetic PARPi response (PR) hallmarks. Thus, these epigenetic hallmarks represent different stages of disease progression during PARPi maintenance therapy, i.e., before (PP), during (DTE), and after (PR) progression and/or treatment. Further (functional) studies are needed to establish which hallmarks and DMRs are associated with treatment and which are associated with disease progression.
The identified epigenetic PARPi hallmarks contained hundreds of DMRs; however, only a few DMRs were reported as tumor driver genes (
SOX2,
POU2AF1,
PTK6,
VHL,
JAK3, and
EZH2) or as HRD genes (
RAD51C). These genes have already been extensively investigated and linked to (ovarian) tumor progression, and some also to resistance for chemotherapy and PARPi. In particular,
RAD51C methylation has been reported to be associated with PARPi resistance in HGSOC patients [
22] and in cell line models [
23]. Although we expected more tumor driver and HRD genes in the hallmarks, such as
BRCA1/2, we will further investigate the other genes for their role in PARPi resistance.
Our pathway analyses of the epigenetic hallmarks indicated that immune-related processes might be involved in PARPi therapy resistance. The identified transcription factor targets and CHPs might further help to pinpoint specific DMRs that play a role in PARPi response. We already identified several DMRs with significantly altered methylation levels between subsets from the different epigenetic hallmarks or that were associated strongly with TTF after PARPi treatment. Further research will be focused on these DMRs by validating our findings and evaluating in greater detail the most interesting DMRs for their relationship with disease progression and PARPi resistance in clinical samples but also in cell line models.