Study design

The CURE (Clinical Utility of circulating Tumor DNA in Gastro-Esophageal Cancer) study is a prospective, observational cohort study investigating the correlation between the presence of ctDNA in plasma and clinical outcomes of patients with resectable gastric and GEJ AC [22]. The Ethics Committee of the Capital Region Denmark approved the collection and use of biological samples [H-19076846]. Informed consent was obtained from all participating patients. The Danish Data Protection Agency [P-2019-701] approved the study.

Sample size estimation

A postoperative ctDNA positivity rate of 15% and hazard ratio (HR) of 5.4 for recurrence in ctDNA-positive patients [15, 23] were assumed. Additionally, an expected median recurrence-free survival (RFS) of 30 months as reported by Al-Batran [6] was applied. Based on these assumptions, a sample size of 52 patients with complete analyses of postoperative samples was calculated using an online tool from UCSF [24]. With this sample size, a difference in RFS could be detected with an 80% power and a 5% risk of type I error [25].

Patient population

Patients were recruited prospectively at the Department of Oncology at the Copenhagen University Hospital–Rigshospitalet, Denmark. Eligible patients were 18 years or older, diagnosed with resectable gastroesophageal carcinoma, and scheduled for perioperative chemotherapy. Four cycles of chemotherapy before surgery and four cycles of chemotherapy after surgery were planned, consisting of docetaxel, oxaliplatin, and fluorouracil/leucovorin (FLOT), in accordance with national guidelines [26, 27]. Following postoperative chemotherapy, patients were scheduled for routine clinical follow-up assessments at every 3–6 months for a duration of 2 years, as per the national guidelines [26, 27].

Plasma samples

Blood samples were planned for collection at baseline, after one cycle of chemotherapy, after preoperative chemotherapy, and 4–6 weeks after surgery. Whole blood was collected in Streck Cell-Free DNA BCT tubes and processed within 36 h of collection. The tubes underwent centrifugation at 2250G for 10 min to separate plasma. Subsequently, the plasma was subject to another centrifugation step at 16,000G for 10 min to remove platelets and any remaining cell debris. The resulting plasma supernatant was carefully transferred to 15 mL tubes (Corning CentriStar) and stored at –80 °C until further processing. Prior to extracting cfDNA, plasma samples were thawed at room temperature and subjected to an additional centrifugation at 3000G for 10 min to separate any potential sediment from the plasma. The cfDNA extraction was performed immediately after thawing, using the QIAamp Circulating Nucleic Acid Kit (Qiagen), following the manufacturer's instructions, with a median of 8 mL (interquartile range (IQR): 7.25–8.75 mL) of plasma. The cfDNA was eluted in a total volume of 60 µL in LoBind 96-well plates (Eppendorf) and then stored at -20 °C until use.

Sodium bisulfite conversion

Sodium bisulfite conversion was performed as previously described [19, 20]. Prior to sodium bisulfite conversion, the cfDNA eluates were dried using vacuum at 30 °C (speedVac, Concentrator plus 5350; Eppendorf AG) and resuspended in 20 µL nuclease-free water. All 20 µL of cfDNA was utilized as input for bisulfite treatment. The EZ-96 DNA Methylation-DirectTM MagPrep kit (Zymo Research) was employed for bisulfite conversion of all samples as per the manufacturer’s instructions, with adjusted volumes of reagents (60 µL CT conversion reagent, 280 µL M-Binding Buffer, 5 µL MagBinding Beads, 185 µL M-Wash Buffer, 93 µL M-Desulphonation Buffer, and 22 µL M-Elution Buffer). Positive and negative controls, including fully methylated human control DNA (Nordic Biosite) and fully unmethylated control DNA (Nordic Biosite), were incorporated in each conversion batch. Subsequently, bisulfite-converted cfDNA was analyzed using droplet digital PCR (ddPCR) within 24 h after the completion of bisulfite conversion.

Droplet digital PCR (ddPCR)

All ddPCR experiments were performed as previously described [19, 20] and are reported in accordance with the dMIQE2020 guideline [Online Resource 1]. The analysis was conducted on a QX200 Droplet Digital PCR system (Bio-Rad) following the manufacturer's specifications. Each analysis comprised positive, negative, and no-template controls. The sample reaction mix consisted of 2–8 µL template DNA, 18 pmol forward primers, 18 pmol reverse primers, 5 pmol probes [Online Resource 2], 11 µL 2X supermix for Probes (Bio-Rad), and nuclease-free water to achieve a final reaction volume of 22 µL. On average, 20978 droplets (IQR: 20230–21456) were generated for each sample using the QX200 AutoDG Droplet Generator (Bio-Rad). Following droplet generation, samples underwent PCR amplification on a S1000 Thermal cycler (Bio-Rad) with the following program: 95 °C for 10 min, 45 cycles of 95 °C for 30 s and 56 °C for one minute, and one final cycle of 98 °C for 10 min. PCR products were stored at 4 °C for up to 12 h before being analyzed on a QX200 reader (Bio-Rad). Data analysis for ddPCR was performed using Quantasoft v1.7 software (Bio-Rad).

DNA quantification and quality control

DNA quantification and quality control was performed as previously described [19, 20]. Prior to bisulfite conversion, assessment of cfDNA purification efficiency and lymphocyte DNA contamination was conducted using ddPCR. To determine purification efficiency, a fixed amount of soybean CPP1 DNA fragments was added to each plasma sample before cfDNA extraction. The purification efficiency was calculated as the percentage recovery of CPP1 fragments after cfDNA extraction, assessed by the CPP1 assay. Additionally, lymphocyte DNA contamination was estimated using a PBC assay, which targeted the VDJ rearranged IGH locus-specific for B cells. The median purification efficiency was 92% (IQR: 85.25–97.68). Quantification of DNA was performed before and after bisulfite conversion using a quantification assay targeting a cytosine-free region on chromosome 1 (CF assay). All assays are listed in Online Resource 2. For each sample, the DNA recovery from bisulfite conversion was calculated as the DNA quantity before bisulfite conversion divided by the DNA quantity after. The median bisulfite recovery of cfDNA samples was 51% (IQR: 44.22–56.51). Samples were excluded for further analysis if they displayed the following criteria: leucocyte DNA contamination, fewer than 10.000 droplets for ddPCR, and non-measurable CF2 signal in ddPCR after bisulfite conversion, to ensure calculation of DNA recovery.

Methylation-specific ddPCR (TriMeth)

ctDNA assessment using TriMeth was performed at Department of Molecular Medicine, Aarhus University Hospital. The analysts were blinded to patient outcome. The TriMeth test comprises methylation-specific ddPCR assays targeting the promoter regions of the genomic regions C9orf50, CLIP4, and KCNQ5 [19, 20] and the CF assay. The TriMeth ddPCR analysis was run in two duplex reactions (C9orf50 + KCNQ5 and CLIP4 + CF), as previously described [19, 20]. Each TriMeth setup included a positive control (5 ng of fully methylated DNA, Nordic Biosite), a negative control (66 ng of fully unmethylated DNA, Nordic Biosite), and a non-template control. For each ddPCR plate, the threshold for separating positive and negative droplets was objectively and automatically defined based on droplet signals in the positive and negative controls, as previously described [19, 20].

Data analysis

A sample was considered TriMeth 'positive' if > 1 positive droplet was detected for at least 2 out of 3 TriMeth markers [19, 20]. For Trimeth (overall), and for each individual marker, we report the number of methylated copies per milliliter plasma [Online Resource 3]. To assess the predictive accuracy of the individual markers and the TriMeth sample calls, Receiver-Operating Characteristics (ROC) analyses were conducted using the R package ROCR [Online Resource 4]. RFS was the primary endpoint. RFS was calculated from the date of inclusion to the occurrence of radiological or clinical recurrence or death resulting from gastric or GEJ cancer. OS was calculated from the date of inclusion to death of any cause. Patients were censored at the end of data collection (1st of July 2023). The association of ctDNA and various prognostic variables with RFS and OS was assessed using cox proportional-hazards regression analysis. A multivariate analysis was conducted, incorporating clinical variables that exhibited statistical significance in the initial univariate analysis. Density plots and boxplots were performed using the R package ‘ggplot’. The Kaplan–Meier method and log-rank test were employed to perform survival analysis. HRs and significance were calculated using the R package 'survival', based on the time elapsed since study inclusion. All p values were based on two-sided testing, and differences were considered significant when p was less than 0.05. All statistical analyses were carried out using R software version 4.3.0.