Cell-free DNA in plasma and ascites as a biomarker of bevacizumab response- a translational research sub-study of the REZOLVE (ANZGOG-1101) clinical trial

Highlights

• •Ascites has more abundant and tumour enriched cfDNA than plasma.
• •Tumour-derived cfDNA proportion in plasma was positively correlated with CA125 level.
• •Tumour-derived cfDNA proportion over 75 % in ascites was associated with shorter time to paracentesis (or death).
• •Tumour-derived cfDNA detection in plasma was associated with shorter survival.
• •Plasma and ascites provide an ideal template to monitor biomarkers in clinical trials with the study of cell-free DNA.

Abstract

Objective

To investigate cell-free DNA (cfDNA) in plasma and ascites and its association with clinical outcomes (paracentesis-free interval, overall survival) and CA125 level in participants with advanced ovarian cancer, treated with palliative intraperitoneal bevacizumab to delay re-accumulation of ascites.

Methods

cfDNA was extracted from 0.3 to 1 mL samples from 20/24 participants of the REZOLVE trial. Standard and methylation-specific PCRs were performed to measure 3 biomarkers: total cfDNA (Alu), tumour-derived cfDNA (ctDNA, methylated IFFO1 promoter) and endothelium-derived cfDNA (ec-cfDNA, unmethylated CDH5 promoter). Values were correlated to clinical outcomes.

Results

cfDNA was detected in all samples, with higher yield in ascites (mean 669 ng/mL) than plasma (mean 75 ng/mL, p < 0.0001). Ascites had a higher ctDNA proportion than plasma (74 % vs. 20 %, p < 0.0001) and plasma had a higher ec-cfDNA proportion than ascites (24 % vs. 16 %, p < 0.002). High ctDNA proportion (>75 %) in ascites was associated with a significantly shorter paracentesis-free interval (median interval 47.5 versus 84 days, hazard ratio (HR) 2.21, 95 % confidence interval (CI) 0.85 to 5.73, p = 0.039) and ctDNA presence in plasma was unfavourable for survival (median survival 56 versus 242 days, HR 3.21, 95 % CI 1.15 to 9.00, p = 0.008). A significant positive correlation was observed between ctDNA proportion in plasma and CA125 level (p = 0.012). No significant difference in total cfDNA, ctDNA nor ec-cfDNA was observed between participants who were responders versus non-responders.

Conclusion

Sufficient cfDNA was detected in both plasma and ascites to study three biomarkers. These samples can provide useful information and should be considered in the design of future ovarian cancer trials.

Introduction

The REZOLVE clinical trial, reported by Sjoquist et al. [1], investigated the effect of administering bevacizumab via the intraperitoneal (IP) route to reduce the accumulation rate of refractory ascites in patients with platinum-resistant ovarian cancer, deemed not suitable for further chemotherapy. Plasma and ascites were collected from participants for translational research (TR). Ovarian cancer-related ascites has recently been found to contain cell-free DNA (cfDNA), a large proportion of which is tumour derived (ctDNA) [2,3]. While ctDNA from plasma is of increasing interest in cancer research and has potential implications for treatment, little is known of its clinical significance in ascites [4]. This TR study sought to investigate cfDNA in plasma and ascites from the REZOLVE trial and to relate the findings to clinical outcomes, particularly where relevant to the effect of bevacizumab treatment.

Ascites is a common manifestation of ovarian cancer and contributes significantly to morbidity; causing fatigue, dyspnoea, pain, discomfort and anorexia, among other symptoms [5,6]. The management of ascites commonly includes intermittent abdominal paracentesis to palliate symptoms [6]. The primary aim of the REZOLVE trial was to evaluate the efficacy of IP bevacizumab to delay the accumulation of ascites and thereby the time to paracentesis in patients with chemotherapy-resistant ovarian cancer [1].

The trial involved IP administration of bevacizumab, a monoclonal antibody targeting VEGF-A, traditionally used in ovarian cancer treatment as a single-agent or combination anti-angiogenic therapy in various settings [7], [8], [9], [10]. Along with its involvement in angiogenesis, causing proliferation and migration of endothelial cells, VEGF-A promotes vascular permeability [11]. VEGF-A is commonly highly expressed in ovarian cancer and the subsequent increased vascular permeability promotes fluid leakage into the peritoneal cavity, where it is retained in high volumes, causing ascites [12]. By inhibiting this effect of VEGF-A, bevacizumab has been found to reduce ascites development in several clinical trials, including the REZOLVE trial [1,[13], [14], [15]]. REZOLVE aimed to reduce bevacizumab dosage, frequency and toxicity by substituting intravenous administration with IP injection (administered immediately following paracentesis), with the rationale being that this would prolong bevacizumab action and facilitate local targeting. A reduction in ascites accumulation of 4.29-fold, measured by time between paracenteses was reported [1].

Ovarian cancer-related ascites has recently been found to contain abundant cfDNA [2,3]. cfDNA, also present in blood plasma and therefore detectable by liquid biopsy, is of growing interest for the detection, diagnosis and monitoring of cancer and for biomarkers of treatment response [4,16,17]. Its heightened presence in blood plasma has been associated with an increased susceptibility to acute inflammatory and thrombotic complications, relevant in ovarian cancer, indicating potential prognostic significance [18,19]. However, little is known of the relationship between cfDNA (and ctDNA) in blood and ascites and any prognostic implications associated with the latter [4].

When cfDNA is dispersed into body fluids upon cell death, it retains the epigenetic signature of its source [20]. This means that cfDNA can be traced to its tissue of origin by assessment of tissue-specific CpG methylation patterns [21]. Assessment of methylation patterns unique to ovarian cancer, such as in the promoter region of the intermediate filament family orphan 1 (IFFO1) gene, indicates where cfDNA is tumour derived (ctDNA) [22]. Additionally, methylation patterns can also distinguish normal tissue types: endothelial cell-derived cfDNA (ec-cfDNA) can be identified through low methylation in the promoter of CDH5 (encoding VE-cadherin) [23,24].

The objective of this TR study was to compare cfDNA in ascites and plasma in terms of total concentration, tumour proportion and endothelial-cell derived proportion, and to relate these to clinical outcomes, including paracentesis-free interval (PFI), survival and CA125 level. These variables were assessed in plasma and ascites fluid collected from REZOLVE participants, using standard and methylation-specific quantitative PCR assays. We hypothesised that the effect of bevacizumab (with its dual implications for tumour cells and endothelial cells) may relate to cfDNA, ctDNA and ec-cfDNA levels in plasma and/or ascites and that these levels may be associated with clinical outcomes.

Methods

REZOLVE cohort

Participants were recruited to the single-armed REZOLVE clinical trial, according to the cohort eligibility criteria described in [1]. Longitudinal plasma and ascites samples were collected from REZOLVE participants at paracenteses (prior to administration of IP bevacizumab). Plasma was also collected at subsequent 3-weekly clinic visits. TR samples were available from 20/24 participants, where collection was possible (Fig. 1A). All participating sites had ethical approval from the relevant bodies and all patients provided informed consent. This TR study obtained additional ethical approval from the UNSW HREC (HC210818). Analysis of TR samples was conducted blinded to clinical outcomes.

Fig. 1.

REZOLVE translational research samples

(A) Participant trajectories as described in Sjoquist et al. 2021, adapted to include translational research sample collection (ascites and plasma). (B) sample processing pipeline. cfDNA, cell-free DNA; qPCR, quantitative polymerase chain reaction; QoL, quality of life.

Ascites and blood processing

Samples were processed at hospital sites within 1 h of collection, according to standard operating procedures for REZOLVE. Ascites was collected in centrifuge tubes, centrifuged at 1500 g at 4 °C for 15 min, the supernatant recovered and centrifuged at the same conditions. Blood collected in EDTA tubes was centrifuged at 1500 g at 4 °C for 15 min, the plasma recovered and centrifuged at the same conditions. Double-spun ascites and plasma were stored at -70 °C or below.

Cell-free DNA extraction

cfDNA was extracted from plasma and ascites, using the QIAamp Circulating Nucleic Acid Kit (Qiagen), as per manufacturer's specifications. Carrier RNA was used to improve yield and DNA was eluted in 2 × 25 µL AVE elution buffer (Qiagen) (50 µL total elution volume). Extractions from longitudinal samples from a single participant were batched. Extraction volumes were determined based on sample availability, with an attempt to maximise volume while standardising extractions within participants (volumes were adjusted to 1 mL with PBS before extraction). Extraction volumes are listed in Supplementary Table 1. Only samples extracted from 0.925 (adjusted) - 1 mL were used to compare cfDNA concentration between participants, to reduce variability introduced by the extraction.

Quantitative PCR (qPCR)

Total cfDNA was measured by qPCR targeting repetitive Alu elements [25] using a CFX384 Touch Real-Time PCR Detection System (BioRad). Each qPCR reaction contained 0.2 μM of each primer, 0.2 mM of each dNTP (New England Biolabs), 2.5 μM SYTO9 (Thermo Fisher Scientific), 3 mM MgCl₂ (Invitrogen), 2 μL reaction buffer (Invitrogen), and 0.12 μL of Platinum Taq Polymerase (Invitrogen). Each sample was tested in triplicate. Primer sequences were as follows: forward, 5′-CCTGAGGTCAGGAGTTCGAG-3′; reverse, 5′-CCCGAGTAGCTGGGATTACA-3′, amplifying a 115 base-pair Alu sequence (amplicon referred to as Alu115). This assay is sensitive down to 0.01 pg of DNA [25]. DNA was diluted to include 0.1 µL of elution per qPCR reaction and compared against a standard curve (400 pg - 0.64 pg), of commercial genomic DNA (Roche). Samples with DNA yield greater than 2 standard deviations from the cohort mean of the respective sample type were considered outliers and excluded from cfDNA concentration analyses.

DNA fragmentation/size analysis

cfDNA fragmentation was assessed by comparison of Alu115 PCR output to a second qPCR assay, targeting a 247 base-pair sequence of Alu elements (Alu247), which straddles the Alu115 sequence. The Alu247 assay was conducted with identical conditions as for Alu115, with forward, 5′-GTGGCTCACGCCTGTTAATC-3′; and reverse, 5′- CAGGCTGGAGTGCAGTGG-3′, primers. Size ratio was calculated as previously [25]. Size ratios approaching 1 indicate high DNA integrity, with a lower dominance of 167 base-pair cfDNA fragments.

Methylation-specific qPCR (MSqPCR)

40 µL purified cfDNA was used as input for bisulfite conversion, using the Epitect Fast Bisulfite Conversion Kit (Qiagen), as per manufacturers specifications. 20 min 60 °C incubation hold steps were used to ensure full conversion. Converted cfDNA was eluted in 70 µL elution buffer and stored at -20 °C until use. After thawing, DNA was diluted with sterile water (Baxter) at a ratio of 4 µL DNA : 6 µL water. 10 µL of this mix was used as template for each MSqPCR reaction.

To amplify the IFFO1 promoter, primer sets targeting the same region of either methylated (IFFO1Me) or unmethylated (IFFO1UM) DNA were used in reactions as follows: IFFO1Me - 0.2 μM of each primer (Forward – 5′GTTGTAGAGAGCGCGGATTT3’; Reverse – 5′CCCGAATAAAATAAACGTCCA3’, 61 base-pair amplicon), 0.2 mM of each dNTP (New England Biolabs), 2.5 μM SYTO9 (Thermo Fisher Scientific), 2.5 mM MgCl₂ (Invitrogen), 2 μL reaction buffer (Invitrogen), and 0.12 μL of Platinum Taq Polymerase (Invitrogen), 62 °C annealing; IFFO1UM - 0.2 μM of each primer (Forward – 5′GTTGTAGAGAGTGTGGATTT3’; Reverse – 5′CCCAAATAAAATAAACATCCA3’, 61 base-pair amplicon), 0.2 mM of each dNTP (New England Biolabs), 2.5 μM SYTO9 (Thermo Fisher Scientific), 2 mM MgCl₂ (Invitrogen), 2 μL reaction buffer (Invitrogen), and 0.12 μL of Platinum Taq Polymerase (Invitrogen), 58 °C annealing.

MSqPCR reactions amplifying unmethylated CDH5 promoter (CDH5UM) were as follows: - 0.2 μM of each primer (Forward – 5′TGTGTTTAAGATGGGAGGGTTT3’; Reverse – 5′AACCCAACATACCCTCAAAAA3’, 96 base-pair amplicon), 0.2 mM of each dNTP (New England Biolabs), 2.5 μM SYTO9 (Thermo Fisher Scientific), 2.5 mM MgCl₂ (Invitrogen), 2 μL reaction buffer (Invitrogen), and 0.15 μL of Platinum Taq Polymerase (Invitrogen), 62 °C annealing.

Cancer antigen 125 (CA125) monitoring

As per clinical protocol, participants’ CA125 was monitored at local laboratories. These were performed at regular time points throughout the trial – within 7 days prior to first on-study drainage, within 72 h prior to drainage, whilst on treatment following 1st drainage and 2nd drainage (if done), and within 7 days after end of treatment.

Statistical analysis

Data was analysed using GraphPad Prism 9.2.0 software and SAS v9.4. cfDNA, ctDNA and ec-cfDNA values were correlated with time to paracentesis pre- and post-bevacizumab treatment (the primary outcome of the REZOLVE trial) as well as overall survival, using log-rank tests. For this analysis, cut-off values of cfDNA concentration and ctDNA/ec-cfDNA proportion were chosen to split the cohort approximately in half, without dividing closely positioned values. Relationships between CA125 and biomarker levels (collected within 7 days of each other, over up to 4 visits per participant) were modelled by General Estimating Equations with a compound symmetry correlation and identity link function. Non-parametric analysis, including Mann-Whitney, Kruskal Wallis (used when comparing two or more than two groups, respectively) and Spearman's correlation co-efficient were used to compare plasma and ascites, due to the small sample size. Where making comparisons between plasma and ascites, (ie. for Fig. 1 and Supplementary Fig. 1) Bonferroni's correction for multiple comparisons was made and p-values were considered significant where below 0.004. Elsewhere, p-values were considered significant where below 0.05, with the aim to identify hypotheses for testing in a larger cohort.

Results

Cohort

Of the 24 REZOLVE participants, baseline ascites and plasma samples, prior to first IP bevacizumab administration were available from 19 participants (Fig. 1). Subsequent serial ascites samples were available from 8 participants and plasma was available from 12 participants. IP bevacizumab was administered subsequent to collection of all ascites samples.

cfDNA from small volumes of plasma and ascites retains integrity after storage for up to 8.5 years

cfDNA extracted from plasma and ascites was assessed for fragmentation with Agilent Bioanalyzer and with qPCR assays, targeting nested DNA regions with different sized amplicons.

Bioanalyzer analysis of cfDNA size found that samples either had a distinct peak at ∼170 bp (low molecular weight, LMW), a DNA smear between ∼1000 and >10,000 bp (high molecular weight, HMW) or both (Supplementary Table 1). Alu size ratio, with values approaching 1 indicating less DNA fragmentation, was significantly higher in ascites (average 0.83 ± 0.21(SD)) than plasma (average 0.43 ± 0.30(SD), p < 0.0001, Supplementary Fig. 1A). We found samples grouped by the three size profiles identified by Bioanalzyer to have significantly different Alu size ratios (p < 0.0001, Supplementary Fig. 1B). No relationship was observed between sample yield and fragmentation (Supplementary Fig. 1C). Additionally, no correlation was observed between time in storage and DNA yield from 1 mL of plasma nor ascites (Supplementary Fig. 1D), nor between storage time and fragmentation (Supplementary Fig. 1E).

cfDNA and ctDNA are more abundant in ascites than in plasma, but ec-cfDNA is proportionately more abundant in plasma than ascites

Total cell-free DNA

With a qPCR assay targeting a 115-base pair repetitive DNA sequence, cfDNA was detected in all samples. We found significantly more cfDNA in ascites (mean, 668.5 ng/mL; median 432.2; range 15.7–2962.5; n = 17 from 13 participants), compared to plasma (mean 63.6; median 21.1; range 9.7–633.0; n = 24 from 11 participants; p < 0.0001) when considering only samples extracted from ∼1 mL fluid (Fig. 2A). We did not find a significant correlation between cfDNA concentrations from matched plasma and ascites samples taken at the same timepoint (Fig. 2B).

Fig. 2.

Total, endothelium-derived and tumour derived cell-free DNA in plasma and ascites

(A) Significantly higher total cfDNA yield in samples extracted from 1 mL of plasma than ascites, by Mann-Whitney test, represented with base-10 log scale. (B) No relationship in cfDNA yield (ng/mL) in matched ascites and plasma from single timepoint, by Spearman's correlation coefficient. (C) cfDNA (ng/mL) in all samples as determined by Alu115 measurement and IFFO1Me+IFFO1UnMe. (D) ctDNA concentration significantly differs in 1 mL plasma and ascites, determined by methylated IFFO1 detection, by Mann-Witney test, represented with base-10 log scale. (E) ctDNA as a proportion of total cfDNA significantly differs in plasma and ascites by Mann-Whitney test. (F) No correlation between the proportion of ctDNA in plasma and ascites, by Spearman's correlation coefficient. (G) ec-cfDNA concentration in 1 mL fluid (indicated by CDH5UM qPCR) was higher in ascites than plasma, however, (H) ec-cfDNA as a proportion of total DNA (indicated by IFFO1Me + IFFO1UM) was higher in plasma, by Mann-Whitney test. (I) No correlation between the proportion of ec-cfDNA in matched plasma and ascites from a single time point, by Spearman's correlation coefficient. No correlation between plasma concentration of cfDNA (J), ctDNA (K) or ec-cfDNA (L) and respective total in ascites (determined by concentration x volume) at matched timepoints, by Spearman's correlation coefficient. All plots use base-10 log scale. ns, not significant; cfDNA, cell-free DNA; ec-cfDNA, endothelial-cell derived cell-free DNA; ctDNA, circulating tumour DNA; ** p < 0.01; **** p < 0.0001.

Circulating tumour DNA

ctDNA abundance, determined by abundance of IFFO1 promoter methylation, as seen in ovarian cancer [22], was measured by IFFO1Me qPCR assay. In cfDNA extracted from ∼1 mL fluid, ascites contained significantly more ctDNA (mean, 462.7 ng/mL; median, 240.1 ng/mL; range, 7.6–3294.6, n = 16 from 12 participants) than plasma (mean, 3.4 ng/mL; median, 1.16 ng/mL; range, 0.0–18.8, n = 24 from 10 participants) (Fig. 2D). Methylated IFFO1 was detected in plasma and/or ascites from each participant at least 1 timepoint; in 30/32 (94 %) ascites samples and 37/56 (68 %) plasma samples (Table 1). The two ascites samples in which IFFO1Me was not detected (both from the same participant) also did not have detectable IFFO1UM, indicating this was likely due to low DNA input. Of samples with cfDNA input of > 400 pg per qPCR reaction, where methylated DNA contributing > 5 % could be reliably detected by the assay (sensitive to 20 pg per reaction), 100 % of 28 ascites samples and 71 % of 41 plasma samples contained IFFO1 methylation. ctDNA was detected in plasma from as low as 20.6 µL equivalent loading volume and in ascites from the minimum equivalent volume loaded (22.9 µL). ctDNA proportion was expressed as the amount of methylated IFFO1 DNA (detected by the IFFO1Me primers) as a percentage of the total DNA (detected by the IFFO1Me plus the IFFO1UM primer sets). A strong correlation between cfDNA concentration as determined by total IFFO1 and Alu115 was observed (Fig. 2C). We found ascites to have a higher tumour proportion than plasma (mean 73.5 ± 26.3 %(SD) vs. 19.7 ± 24.1 %(SD); p < 0.0001) (Fig. 2E). We did not find a significant correlation between the tumour proportion in matched plasma and ascites samples taken at the same time point (Fig. 2F).

Table 1.

qPCR of cfDNA biomarkers in all plasma and ascites samples.

Sample type	Alu115 positive (%) (range ng/mL)	IFFO1Me positive (%) (range ng/mL)	CDH5UM positive (%) (range ng/mL)
Ascites n = 32 from 20 participants	32 (100 %) (1.02–2317.11)	30 (94 %) (0.75–12,566.47)	26 (81 %) (0.97–1523.67)
Plasma n = 56 from 19 participants	56 (100 %) (5.55–243.13)	37 (66 %) (0.03–18.81)	49 (86 %) (0.03–86.38)

Endothelium-derived cfDNA

ec-cfDNA was measured by a qPCR assay targeting the unmethylated promoter of CDH5 (uniquely unmethylated in endothelial cells) [23], and its proportion was determined as a percentage of total IFFO1. Plasma contained a higher proportion of ec-cfDNA than ascites (mean 16.2 ± 31.7(SD) vs. 24.4 ± 24.5(SD), p = 0.002)(Fig. 2H). However, in cfDNA extracted from ∼1 mL fluid, ascites contained significantly more ec-cfDNA (mean, 68.1 ng/mL; median, 22.7; range 0–441.4; n = 17 from 12 participants) than plasma (mean, 13.2 ng/mL; median, 0.8; range, 0–209.6; n = 25 from 11 participants p = 0.002) in terms of absolute concentration (Fig. 2G), consistent with our findings of more total cfDNA in ascites than in plasma. We did not find a significant correlation between the ec-cfDNA proportion in matched plasma and ascites samples taken at the same time point (Fig. 2I).

As ascites can accumulate in great volumes, potentially diluting cfDNA, we also considered the theoretical amounts in the total ascites fluid volume. No relationship between ascites volume and cfDNA, ctDNA or ec-cfDNA was observed (Supplementary Fig. 2). Additionally, there was no correlation in cfDNA between the total in ascites and the concentration in plasma per millilitre, (Fig. 2J) or ctDNA (Fig. 2K) or ec-cfDNA (Fig. 2L).

Lower ctDNA in ascites at baseline is suggestive of improved PFI after bevacizumab treatment

No significant relationship was observed in ascites between the responders (defined by PFI of >42 days after bevacizumab infusion) and non-responders, in terms of baseline cfDNA, ctDNA or ec-cfDNA concentration (Fig. 3. A–C). Log-rank test of time to paracentesis (or death, if no post-enrolment paracentesis was carried out) found that high ctDNA percentage in ascites (>75 %) was associated with shorter time to paracentesis (median interval 47.5 versus 84 days, hazard ratio (HR) 2.21, 95 % confidence interval (CI) 0.85 to 5.73, p = 0.039), but no other relationship was observed (Fig. 3. D–F). Similarly, no relationship between responders and non-responders was observed in plasma, nor was an association with time to paracentesis (or death) observed (Fig. 3. G–L).

Fig. 3.

cfDNA biomarkers in plasma and ascites in baseline and response to bevacizumab

(A–C) comparison of cfDNA concentration, ctDNA proportion and ec-cfDNA proportion in ascites in participants with post-bevacizumab PFI of less or greater than 42 days, by Mann-Whitney test. (D-E) Log-rank analysis of time to first event after initial bevacizumab treatment, being either paracentesis or death (if no further paracentesis preceded), saw better prognosis with lower ctDNA proportion in ascites. (F–I) comparison of cfDNA concentration, ctDNA proportion and ec-cfDNA proportion in plasma in participants with post-bevacizumab PFI of less or greater than 42 days, by Mann-Whitney test. (J–L) Log-rank analysis of time to first event after initial bevacizumab treatment, being either paracentesis or death, in plasma. PFI; paracentesis-free interval; cfDNA, cell-free DNA; ec-cfDNA, endothelial-cell derived cell-free DNA; ctDNA, circulating tumour DNA; * p < 0.05.

Serial monitoring of biomarkers in cfDNA and comparison with CA125

Serial CA125 measurements were taken throughout the REZOLVE trial. When comparing cfDNA, ctDNA and ec-cfDNA in plasma and ascites collected throughout the trial with matched CA125 values measured within 7 days, we found a weak positive correlation between CA125 and ctDNA in plasma (a unit increase in CA125 increased ctDNA by 0.007 %, 95 % CI: 0.002 to 0.012, p = 0.012)) and no other significant relationship (Fig. 4. A–F).

Fig. 4.

Serial monitoring of bevacizumab response

CA125 levels recorded within 7 days of ascites (A-C) or plasma (D-F) collection over up to 4 visits. Each participant was assigned a colour, maintained across charts A-F. (G) cfDNA concentration (Alu115 ng/mL), (H) ctDNA proportion (IFFO1Me%), (I) ec-cfDNA proportion (CDH5UnMe%) and (J) CA125 levels in plasma samples at baseline and 3 weeks after bevacizumab treatment in participants whose paracentesis was >42 days (responders) or <42 days (non-responders). All plots use base-10 log scale. cfDNA, cell-free DNA; ec-cfDNA, endothelial-cell derived cell-free DNA; ctDNA, circulating tumour DNA; * p < 0.05.

In 5 participants, plasma samples were collected 3 weeks post baseline bevacizumab administration. We compared cfDNA, ctDNA, ec-cfDNA and CA125 in these samples with their baseline equivalent to monitor the effect of bevacizumab. Though the sample size was too small to warrant statistical analysis, we observed that both non-responders had stable, undetectable ctDNA and stable CA125 before and after bevacizumab and a steep increase in ec-cfDNA (Fig. 4. H–J). No clear patterns were observed in cfDNA (Fig. 4. G).

In participants with multiple samples of plasma and/or ascites, we saw no distinct pattern of change over time in CA125, cfDNA, ctDNA or ec-cfDNA in responders (Supplementary Fig. 3) or non-responders (Supplementary Fig. 4).

ctDNA detection in plasma has unfavourable prognosis for overall survival

No relationship between overall survival and baseline levels of cfDNA, ctDNA or ec-cfDNA in ascites by log-rank test was observed (Fig. 5. A–C). This was also true of plasma cfDNA and ec-cfDNA, however, detection of ctDNA in plasma at baseline was unfavourable for survival (median survival 56 versus 242 days, HR 3.21, 95 % CI 1.15 to 9.00, p = 0.008)(Fig. 5. D–F). No relationship between CA125 level and overall survival was observed (Fig. 5. G)

Fig. 5.

Overall survival and cfDNA biomarkers in plasma and ascites

No relationship with overall survival by log-rank test in (A) cfDNA concentration (Alu115), (B) ctDNA (IFFO1ME) proportion or (C) ec-cfDNA (CDH5UnMe) proportion in ascites. No relationship with overall survival by log-rank test in (D) cfDNA concentration (Alu115) or (F) ec-cfDNA (CDH5UnMe) proportion in plasma, but (E) favourable probability of survival with ctDNA proportion of 0 (p = 0.005). (F) No relationship between CA125 level and overall survival. cfDNA, cell-free DNA; ec-cfDNA, endothelial-cell derived cell-free DNA; ctDNA, circulating tumour DNA; OS, overall survival;** p < 0.01.

Discussion

Though cfDNA is well studied in blood, few studies have directly compared cfDNA in blood and ascites, particularly in participants with advanced ovarian cancer. In our novel application of TR biomarkers, we utilised small-volume clinical samples of plasma and ascites, collected from participants enrolled in a phase 2 trial. We investigated three biomarkers - cfDNA concentration, cfDNA tumour proportion and cfDNA endothelial cell proportion - as potential biomarkers of bevacizumab activity.

We report significantly higher levels of cfDNA in ascites than in matched plasma. This aligns with similar research in lung adenocarcinoma, for example, but is reported for the first time in advanced ovarian cancer, to the best of our knowledge [26,27]. We report a high tumour proportion in cfDNA from ascites, consistent with our previous findings, with an average of 74 % tumour content in ascites, compared to 20 % in plasma [3]. Though ctDNA proportion was higher in ascites than plasma, ec-cfDNA was lower, indicating proximity as a major contributor to cfDNA tissue-of-origin composition, as previously postulated [4]. We saw IFFO1 methylation in plasma and/or ascites in every participant (where >1 plasma and ascites samples were collected per participant, IFFO1 methylation was seen in both plasma and ascites from every participant at some point). This is consistent with Campan et al. (2011), finding IFFO1 methylation in 100 % of ovarian cancers (n = 16)[22]. IFFO1 methylation detection was possible from the equivalent of less than 23 µL plasma input (maximum input <46 µL).

An interesting finding was that there is no correlation between cfDNA concentration, ctDNA proportion nor ec-cfDNA proportion in matched plasma and ascites samples, even when accounting for ascites volume. cfDNA and ctDNA concentrations in plasma have been associated with tumour burden in multiple studies, so we expected larger tumours to contribute more cfDNA/ctDNA to both blood and ascites [28], [29], [30]. The absence of a relationship may be explained by different cfDNA release and clearance pathways between blood and ascites [31]. An additional explanation may be presence of extraperitoneal lesions contributing ctDNA to blood plasma in some cases.

We found that ctDNA detection in plasma was associated with overall survival, yet there was no relationship between ctDNA proportion in ascites and survival. Previous research has found that when tumours are localised to the peritoneal cavity, ctDNA is dilute or not detectable in plasma [32], [33], [34]. Higher ctDNA presence in plasma may therefore be indicative of extraperitoneal disease spread, a poor prognostic indicator [35]. A limitation of this study is that we did not have access to CT scans to assess this. As heightened plasma cfDNA has been suggested to increase the risk of thrombotic events by activating the clotting cascade [36], we considered a possible causative link to mortality. However, as no thrombotic events were reported in the trial and the link to prognosis was specific to the ctDNA component, we deem this unlikely.

A study objective was to assess differences in total cfDNA, ctDNA or ec-cfDNA at baseline between REZOLVE participants deemed responders/non-responders to IP bevacizumab. We did not identify any significant difference in these parameters between groups. However, the arbitrary 6-week PFI cut off value chosen in REZOLVE to discriminate the two groups split some participants with very similar PFIs, so investigation in a larger cohort remains of interest. To better elucidate any relationships between the studied biomarkers and a response to bevacizumab, future trials should involve more regular plasma sampling post-bevacizumab treatment. This would be particularly worthwhile for ec-cfDNA, which increased dramatically in the two non-responders in whom we were able to perform 3-weekly longitudinal plasma tracking. Alternatively, VEGF-A levels in ascites fluid, potentially in ratio with plasma levels, is another biomarker worth investigating as an indicator of bevacizumab response.

When looking at the relationships between clinical outcomes and the biomarkers tests, we found lower ascites ctDNA proportion at baseline to be associated with a longer PFI. This indicates ctDNA as a potential prognostic indicator of refractory ascites. The cut-off value of <75 % for ‘low’ ctDNA among this group (inclusive only of people with refractory ascites) leaves scope for ctDNA proportion as a potential predictor of recurrent ascites and suitability of bevacizumab treatment. This should be investigated in a larger, more diverse cohort.

An important secondary finding of this work was an association between ctDNA concentration in plasma and CA125 level. We also noted that ctDNA presence/absence in plasma was a better indicator of overall survival than CA125 level. ctDNA has been found to outperform CA125 in predicting recurrence or detecting residual disease post-surgery in a number of studies [37], [38], [39], [40]. For example, in a recent study, ctDNA (detected through personalised PCR assays targeting identified somatic clonal variants in tumour tissue) predicted progression an average of nine months earlier than a rise in CA125 (n = 8) [38] . This is the second reported finding of a correlation between ctDNA in plasma and CA125, with Han et al. (2020) reporting a Spearman correlation of r = 0.658 between mutant TP53 variant allele frequency in plasma versus CA125, in individuals with late-stage epithelial ovarian cancer [2].

Much of the strength of this research lies in the novel contributions to our technical understanding of cfDNA in plasma and ascites, by application of effective assays. We demonstrated the successful extraction of cfDNA from plasma and ascites samples stored for up to 8.5 years, despite previous research reporting a decline in cfDNA yield from plasma over its storage duration [41]. We demonstrated feasibility of detection of cfDNA in 100 % of plasma and ascites samples, from volumes as low as 0.6 µL. By Alu247/Alu115 size ratio assay [25], we observed a lower dominance of DNA of mononucleosomal length in ascites, aligning with the profiles observed by Bioanalyzer, where high molecular weight peaks were observed in 20 of 25 ascites samples (80 %). This supports our previous work where we identified a strong ladder pattern as a trait of ascites derived cfDNA and where high molecular weight DNA, apparently derived from extracellular vesicles, was seen in 39 % of ascites samples (n = 18) [3]. However, this may also be a result of some carryover of genomic DNA to the cell-free ascites fluid, due to decentralised sample preparation, a limitation of the present study.

We were able to concurrently quantitate total cfDNA and ctDNA proportion with a new assay, targeting both methylated and unmethylated CpG sites in the IFFO1 promoter, showing significantly similar results to the established Alu cfDNA concentration assay [25]. ec-cfDNA proportion, detected by targeting the unmethylated CDH5 promoter, could then reliably be calculated as a proportion of the total bisulfite converted DNA (likely more representative than the Alu assay). As only small volumes of plasma were available, our workflow was quite restricted: in order to seed a satisfactory volume of DNA to methylation-specific PCRs, with replicate reactions in each, we could only amplify three targets, leaving CDH5UM to be quantitated against total IFFO1 instead of its methylated counterpart. We were also unable to perform replicate extractions to validate each individual cfDNA concentration value. The incomplete collection of serial samples (not uncommon to clinical trials) was an additional limitation of the work.

For future studies, larger and replicate volumes collected may allow more reliable estimation of ec-cfDNA in plasma and ascites samples, as well as analysis of more targets. A larger scale, follow-on study to REZOLVE will allow more confidence in findings and better compensation for missing samples.

In summary, even small volumes of plasma and ascites collected in a phase-2 trial were informative and enabled translational research. By investigating three biomarkers- cfDNA, ctDNA and ec-cfDNA- we provided insight into biological phenomena, clinical outcomes and technical considerations for research. Though further research is needed to verify suggested associations with clinical outcomes after IP bevacizumab, this research argues for the value of cfDNA as a biomarker and plasma and ascites as biospecimens for translational research.

Conclusion

Sufficient cfDNA was detected in small volumes of both plasma and ascites to perform qPCR for three different biomarkers. Higher total cfDNA and proportionate ctDNA was detected in ascites, and higher proportion of ec-cfDNA was detected in plasma. A positive correlation between ctDNA in plasma and CA125 levels was observed. ctDNA in ascites and plasma was found to have prognostic implications for paracentesis-free and overall survival, respectively. Biospecimen collection as part of a follow-up trial will allow further investigation of cfDNA, ec-cfDNA and ctDNA in plasma and ascites as biomarkers of bevacizumab action in the palliation of ascites. Furthermore, this research calls for embedding plasma and ascites collection for research in future ovarian cancer clinical trials.

CRediT authorship contribution statement

Bonnita Werner: Conceptualization, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing. Katrin M Sjoquist: Funding acquisition, Resources, Writing – review & editing. David Espinoza: Data curation, Writing – review & editing. Sonia Yip: Funding acquisition, Resources, Writing – review & editing. Garry Chang: Resources, Writing – review & editing. Michelle M Cummins: Resources, Writing – review & editing. Linda Mileshkin: Funding acquisition, Resources, Writing – review & editing. Sumitra Ananda: Resources, Writing – review & editing. Catherine Shannon: Resources, Writing – review & editing. Michael Friedlander: Funding acquisition, Resources, Writing – review & editing. Kristina Warton: Conceptualization, Methodology, Project administration, Supervision, Writing – review & editing. Caroline E. Ford: Conceptualization, Funding acquisition, Project administration, Supervision, Writing – review & editing.

Declaration of competing interest

KW declares potential financial conflict of interest due to stock ownership in the following companies that are developing cell-free DNA based clinical assays: Guardant Heath; Exact Sciences; EpiGenomics AG. All other authors declare no competing financial interests.

Appendix. Supplementary materials

S2. Biomarkers and ascites volumes

S3. Longitudinal analysis in responders

S4. Longitudinal analysis in non-responders

Table S1. Sample characteristics