Patients and study design

The study included treatment-naïve patients with asymptomatic and symptomatic CRC from two independent cohorts, Cohort#1 and Cohort#2. Cohort#1 also included non-cancer controls. Plasma from all study subjects was retrospectively analysed for ctDNA blinded to the individual’s outcome (Supplementary Fig. S1).

Plasma from Cohort#1 subjects (215 patients with asymptomatic CRC, 117 patients with symptomatic CRC, and 804 non-cancer controls) was analysed using a tumour-agnostic ctDNA detection method querying three CRC-specific methylation markers by droplet digital PCR (ddPCR). Cohort#1 patients and non-cancer controls were recruited from 2014–2020 at the Surgical Departments of Aarhus, Randers, Horsens, Herning, Viborg, Bispebjerg, Hvidovre, Hillerød, Herlev, and Slagelse Hospitals. The symptomatic patients were all diagnosed with CRC after referral for diagnostic work up due to symptoms. The asymptomatic patients and the non-cancer controls were participants in the Danish CRC-screening programme. The asymptomatic patients were all diagnosed by colonoscopy after a positive faecal immunochemical screening test (FIT). The FIT test was scored as ‘positive’ at >100 ng haemoglobin/mL buffer corresponding to ≥20 µg haemoglobin/g faeces. The non-cancer controls included both FIT-negative individuals (n = 305) and FIT-positive (n = 499) individuals where no cancer was detected upon colonoscopy. The Danish CRC screening programme encourages individuals to participate only if they are healthy and asymptomatic. If screening invitees experience CRC-related symptoms the programme guidelines urge them to contact a physician directly. Thus, most patients with CRC diagnosed through screening are asymptomatic and will throughout the study be referred to as ‘asymptomatic’. All healthcare services and screening participation is free of charge in Denmark. This limits the risk of inclusion bias from socio-economic status among both asymptomatic and symptomatic CRC patients.

To ensure that Cohort#1 results were not biased by a difference in methylation patterns between tumours from patients with asymptomatic and symptomatic CRC and to validate the Cohort#1 findings, an independent cohort of 1,090 asymptomatic (n = 368) and symptomatic (n = 722) patients with CRC (Cohort#2) was analysed using an orthogonal tumour-informed mutation-based method. Cohort#2 patients were consecutively recruited and had treatment-naïve pre-operative plasma collected during 2018–2021 as part of the clinical trial IMPROVE (ClinicalTrials.gov NCT03637686).

After surgery, patients were offered follow-up according to the Danish national guidelines [18] recommending computed tomography (CT) scans at 12 and 36 months after surgery. For some patients (n = 133) no CT scan results were available. This was due to: (1) the follow-up shorter than 12 months (n = 63), (2) the patient died before month 12 (n = 13), (3) the patient was not offered follow-up due to old age (>80 years) (n = 53), or (4) the patient declined any follow-up (n = 4). Since our main aim of the study was to investigate differences in ctDNA detection complete follow-up was not an inclusion criterion. Patients without follow-up were excluded from recurrence analysis (see “Statistical analysis” section).

Due to the study’s explorative design, no sample size estimation (power analysis) was performed.

Blood collection and processing

The same standard operating procedure was used for collection and processing of all blood samples included in this study (Supplementary Appendix 1). In brief, blood was collected in BD Vacutainer K2 EDTA tubes (Becton Dickinson) and processed within 2 h from venipuncture. A two-step plasma isolation procedure using double centrifugation at 3000 g for 10 min at 21 °C was applied to avoid contamination from hematopoietic cells. Isolated plasma was stored in cryotubes (Techno Plastic Products AG) at −80 °C until the time of circulating cell-free DNA (cfDNA) extraction.

cfDNA isolation and quantification

cfDNA was extracted from 8 mL plasma on the QIAsymphony robot using the QIAamp Circulating Nucleic Acids kit (Qiagen) or manually using the QIAamp Circulating Nucleic Acid Kit (Qiagen) following manufacturer’s instructions. cfDNA was quantified by ddPCR (Bio-Rad Laboratories, Hercules, CA, USA) as previously described [19,20,21]. cfDNA purification efficiency and contamination with DNA from lysed lymphocytes were estimated by ddPCR [19, 22]. For details, see Supplementary Appendix 1.

Methylation-based ctDNA detection

In Cohort#1, cfDNA was sodium bisulfite converted prior to ctDNA analysis using the EZ-96 DNA Methylation-Direct™ MagPrep kit (Zymo Research) either manually or automated on a Zephyr robot (for details, see Supplementary Appendix 1). ctDNA was quantified by a methylation-specific multiplex ddPCR test, targeting the promoter regions of C9orf50, CLIP4, and KCNQ5, and a cytosine-free quantification assay (the CF assay), described in Supplementary Appendix 1. Details regarding marker selection, development of the methylation assays, assay optimisation, and test and validation of assay performance (sensitivity and specificity) in case-control plasma cohorts are thoroughly described in our previous work [19, 20].

Mutation-based ctDNA detection

In Cohort#2, ctDNA was quantified using a tumour-informed mutation-based ddPCR strategy as described by Henriksen et al. [23, 24]. In brief, whole exome sequencing (WES) of paired tumour and normal samples was performed and mutational clonality and multiplicity were assessed as previously described [25]. For each patient, a single clonal somatic mutation was selected and a ddPCR assay targeting this was designed and validated before the assay was applied to patient plasma. For details regarding the selection of mutational targets, assay development, and test and validation of assay sensitivity and specificity, please refer to Supplementary Appendix 1 and our previous work [23,24,25].

Droplet digital PCR analysis

All ddPCR experiments were conducted on the Droplet Digital PCR System (Bio-Rad) according to manufacturer’s instructions and are reported in agreement with the guideline for reporting on Quantitative Digital PCR Experiments (dMIQE) (Supplementary Table S1) [26]. Raw fluorescence intensity data for all individual droplets in each well was extracted using Quantasoft (v1.7.4; Bio-Rad).

For methylation assays, thresholds for positive and negative samples were set as previously described [19, 20]. A sample was classified as ‘ctDNA positive’ if at least two of three methylation markers showed a positive signal, otherwise, the sample was classified as ‘ctDNA negative’. The R code for defining the plate-wise thresholds for positive and negative droplets and for calculating the concentration of methylated DNA is available at GitHub [27].

For the mutation-based assays, the detection of ctDNA was done according to a previously established method using the ddPCR-calling tool CASTLE available at GitHub [23]. Patient plasma samples with a CASTLE p-value below 0.01 were called ‘ctDNA positive’. For details see Supplementary Appendix 1.

Statistical analysis

Subgroups were compared using Fisher’s exact test for discrete data of small sample size and Pearson’s Chi-squared test for multiple subgroups. For continuous data, a comparison of unmatched groups was performed using Wilcoxon rank sum test. Univariable and Multivariable logistic regressions were used to estimate odds ratios (OR) and 95% intervals of confidence (95% CI) to evaluate the association of clinical variables on ctDNA detection (outcome variable). The univariable regression included variables expected to affect ctDNA shedding, according to Table 1: tumour size, depth of invasion (pT), lymph node status (pN), distant metastasis status (pM), age, and location (confounder variables). The Multivariable regression included the variables significant in the univariable regression.

Table 1 Characteristics Cohort#1.

Recurrence detection was assessed from the day of surgery and until: radiological recurrence (event), death (competing event for cumulative incidence functions of recurrence and for competing risk regression of recurrence), or until end of follow-up (censoring). Patients with less than 12 months of follow-up and no CT scan results were excluded from recurrence analyses. Cumulative incidence functions (CIF) of recurrence were constructed using the Aalen-Johansen estimator. Subdistributional hazard ratios (sHR) with 95%CI for recurrence were estimated using Fine&Gray regression and adjusted for pTNM and tumour size. P-values < 0.05 were considered significant (two-sided). All data analysis and statistical analyses were performed using R software versions 4.0.3 and 4.1.1.