Patient characteristics

A total of 53 patients were included in our study (Fig. 1). 19 patients from the retrospective pilot study (ascites at 1° laparoscopy (N = 18) or interval 2° surgery/laparoscopy (N = 1)) and 34 from the prospective study (ascites at 1° laparoscopy (N = 17), ascites at 2° laparoscopy (N = 10) or therapeutic paracentesis N = 7). The high rate of ascitic sample collection at interval surgical exploration in our series is attributable to the fact that some patients were referred to our center for debulking surgery after 3–4 cycles of neoadjuvant chemotherapy.

Fig. 1

Direct ascites was collected for 34/35 of samples obtained at 1° laparoscopy, for 5/11 at 2° laparoscopy and 7/7 at paracentesis. Among patients with visible ascites at laparoscopy (N = 39), ascitic volume was 20-100 ml in 18% (7/39), 100-500 ml in 8% (3/39), 500-1000 ml in 28% (11/39) and > 1000 ml in 46% (18/39). For patients without visible ascites (at 1°, N = 1 or 2° laparoscopy, N = 6), indirect ascites was obtained by peritoneal washings with saline.

Median age was 65 years old (range: 42 – 86) and most patients had FIGO stage III/IV (98%), and high grade ovarian cancer (92%, 49/53) other histologies included low grade ovarian cancer (N = 4). Most ascites and tumor sample were collected at primary laparoscopy (66%) (Table 1).

Table 1 Characteristics of patient at baselineContributive cftDNA detection from ascites

Cell-free DNA was detectable in peritoneal fluid from 49/53 patients (92,5%). Overall, DNA quality was high. Figure 2 illustrates the typical profile of cfDNA extracted from peritoneal fluids in our study featuring a single mononucleosomal peak at a mean length of 177 bp (range: 128 – 204), along with two additional peaks at around 360 and 520 bp. This showcases the high quality of cfDNA extracted from peritoneal fluids with low contamination by high molecular weight DNA (> 700-pb).

Fig. 2

Example of patient’s electropherogram: sizing range of DNA detected in ascites

Additionnaly, DNA yield was very high with a median concentration of total cfDNA of 3700 ng/ml (range 109 – 65 000 ng/ml). The reported cfDNA concentration represents the concentration of cfDNA with a size range of less than 1000 base pairs in the extracted material (from 1 to 4 ml of ascites). For patients with peritoneal lavage the concentration was also high with a median concentration of 1310 ng/ml (range: 1000-2120 ng/ml). In comparison, cfDNA concentration extracted from plasma in patients with any solid tumor usually ranges from 5 to 1500 ng/ml [18]. When considering only direct ascites, cfDNA was detected in 100% of cases (46/46), including ascites obtained after neoadjuvant chemotherapy at 2° laparoscopy. Importantly direct ascites yielded cfDNA regardless of volume present. Among the patients with < 100 ml (N = 7) or 100-500 ml (N = 3), cfDNA was detected in all cases and median concentration of cfDNA for patient with < 500 ml ascite was 3150 ng/ml (range: 20,9 – 15 900 ng/ml). Interestingly, in patients without visible peritoneal free fluid and in whom peritoneal washings were collected, cfDNA was still detected in 42% of these cases (3/7). The characteristics of the 4 patients without detectable cfDNA are summarized in Table 2. All samples were obtained at 2° laparoscopy after a median of 3,5 cycles of neoadjuvant chemotherapy, via peritoneal washings and all patients demonstrated a good clinico-biological response to chemotherapy.

Table 2 Characteristics of patient with cfDNA analysis on peritoneal washingsPathogenic variant detection on cftDNA from peritoneal fluid

Crucially, among the 49 of 53 cases with detectable cfDNA, a pathogenic variant was detected in 96% of peritoneal fluid samples (47/49), thus confirming that when cfDNA was detected it almost invariably contained tumoral DNA (cftDNA) (Fig. 3). For 1 of the 2 patients for whom no pathogenic variant was detected on ascites cftDNA (acftDNA), the NGS CGP also did not identify a pathogenic variant on tumor tissue.

Fig. 3

Oncoprint for CGP testing on cfDNA

The most common pathogenic variant identified in peritoneal fluid samples was TP53 detected in 86% of 49 contributive samples (Fig. 3). Three low grade OC acftDNA harbored a RAS pathogenic variant and one had ATM pathogenic variant. BRCA1 and BRCA2 pathogenic variants were detected in respectively 14% (7/49) and 10% (5/49) of patients including one large BRCA1 rearrangement (deletion of exon 21 to 24) detected on acftDNA, confirming the high quality of the acftDNA to detect such a quantitative event. Finally, the median testing turn-around time was only 21 days (range: 14–36) for NGS on acftDNA.

Genomic testing on cftDNA from peritoneal fluids vs tumor samples

As our main objective was to evaluate the performance of HRD testing on peritoneal cftDNA, we focused on HGOC pts (N = 49) and compared the performance of cftDNA to tumor tissue based testing. NGS CGP on DNA from FFPE matching tissue samples identified a pathogenic variant in 90% (44/49) of cases. Median testing turn-around time was longer than for acftDNA at 45 days (range: 14–96). cfDNA NGS on peritoneal samples from HGOC was comparable with mutated cftDNA identified in 88% (43/49). However if the analyses were limited to cftDNA identified in direct ascites, ctDNA detection rate was higher at 98% (45/46). The patients for whom no pathogenic variant was detected on ascites cfDNA (acfDNA), the NGS CGP also did not identify a pathogenic variant on tumor tissue. For the 5 patients with a failed tumor tissue analysis, a TP53 mutation was detected in the matching ascites sample, including one patient for whom acftDNA analysis identified a BRCA2 pathogenic variant. Together these data support that cftDNA from ascites may allow physicians to salvage non-contributive tumor analyses.

Subsequently, an evaluation of sensitivity, specificity, and concordance was conducted to compare acfDNA (cell-free DNA from ascites) with tumor testing for the detection of TP53 pathogenic variants.

This analysis was specifically focused on patients diagnosed with high-grade ovarian carcinoma who exhibited contributive results in both cfDNA and tumor tissue analyses (N=42, 80% of cohort).

Sensitivity and sensibility analyse were perform using the contingency table in Table 3 with GraphPad Prism version 10.0.0. Confidence interval were estimate using the The hybrid Wilson/Brown method Kappa Cohen analysis was conducted using R Studio (v 3.3.0, R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/) with the kappa.cohen function to evaluate concordance for this study. The concordance column in Table 3 correspond to the percentage of agreement between acfDNA and tissue analysis.

Table 3 Assay Positive Predictive Value, Negative Predictive Value, Specificity, Sensitivity, and Concordance of TP53 pathogenic variant on acfDNA compared With Tumor Tissue

The sensitivity,specificity and concordance were 97% (95% IC : 86%-100%), 83% (95% IC : 43%-100%) and 95% (K = 0,81: P <0,001) respectively (Table 3). One discordant patient harbored a TP53 pathogenic variant detected on acftDNA that was not detected on tissue DNA, potentially attributable to low cellularity and inversely one patient had TP53 pathogenic variant detected on tissue DNA but not on acftDNA.

Mean variant allele frequency (VAF) in ascites also compared favorably to tissue. The mean VAF for TP53 pathogenic variant were 54% in acftDNA vs 45% in ttDNA. For BRCA1 pathogenic variant the mean VAF was 67% in acftDNA and 63,3% in ttDNA and for BRCA2 pathogenic variant it was 80% in acftDNA and 84,5% in ttDNA.

However, suprisinsigly,examination of the individual Variant Allele Frequencies (VAFs) of TP53 in tissue and ascites using Pearson correlation analysis revealed no significant correlation between these two sample types (r = 0.19, p = 0.27, see Supplementary Fig. 1). The statistical analyses were conducted using the cor.test function in R Studio (version 3.3.0, R Foundation for Statistical Computing, Vienna, Austria, https://www.R-project.org/).

Genomic instability testing

Tumor-based genomic instability testing (Myriad MyChoice CDx) was performed on all HGOC samples as part of routine care, 3 results were pending at the time of publication. 44 patients had result available with 75% (33/44) yielding a contributive result. Among the 32 patients with successful tumor-based genomic instability testing, 16 (50%) were considered HRD + with a GIS > 42.

Genomic instability using shallow WGS (sWGS HRD) was measured on 18 acftDNA samples (including 4 with failed tumor-based GIS). All 18 patient had direct ascites sample. sWGS HRD was successful for all 18 samples, including the 4 with failed tumor testing resulting in a 100% contributive genomic instability test result and 10/18 acftDNA samples exhibited high genomic instability (LGA > 20) thus confirming the feasibility of performing genomic instability testing on cftDNA from ascites.