Colorectal cancer detected by liquid biopsy 2 years prior to clinical diagnosis in the HUNT study

Colorectal cancer (CRC) is common worldwide, with high incidence in Norway. Prognosis strongly depends on stage at diagnosis, with 5-year survival >90% for Stage I and ~10% for Stage IV. Although established screening tools (FIT, gFOBT, colonoscopy, CT colonography) reduce incidence and mortality, participation rates are suboptimal due to invasiveness and discomfort. Blood-based biomarkers, including circulating tumor DNA (ctDNA), offer minimally invasive alternatives. DNA methylation of promoter CpG islands in tumor suppressor and related genes is an early event in CRC tumorigenesis and can be detected in plasma. Prior studies often included selected cases at diagnosis and may not generalize to population screening. This study asks whether methylated ctDNA markers can detect CRC up to two years prior to clinical diagnosis in an unselected population-based cohort, resembling a true screening setting.

Multiple methylated promoter regions (~70) have been proposed as CRC biomarkers in blood or stool. Commercial tests (e.g., Epi proColon 2.0, ColoSure, Cologuard/multitarget stool DNA) achieve varying sensitivity/specificity, but sensitivity for advanced adenomas remains limited in average-risk populations. Panels combining methylated ctDNA markers may improve performance, with some reports of sensitivities and specificities above 90% in selected cohorts. ctDNA has been detected years before cancer diagnosis in smaller prospective studies and single-marker assays have shown high detection in selected groups. However, previous work often used cases at diagnosis or controls not representative of screening populations; methylation status can be influenced by age, sex, BMI, and smoking, necessitating adjustment.

Design: Nested case–control study within the Trøndelag Health Study (HUNT3, 2006–2008). Cases were participants later diagnosed with colorectal adenocarcinoma (ICD-10 C18–C20, appendix cancers excluded) identified via linkage to the Cancer Registry of Norway. From HUNT3, 245 subjects were diagnosed <54 months after sampling; the primary analysis defined the outcome as CRC diagnosis within 24 months after sampling. Controls were matched 1:1 to cases by sex and age (±1 year) and had no CRC diagnosis in the registry through 12/31/2017; other cancers were not exclusionary. Participants and samples: Plasma from 212 individuals (106 cases diagnosed within 24 months; 106 matched controls) was analyzed. Blood was drawn at HUNT3, transported at 4°C, processed (centrifuged at 6°C, 10 min, 2500×g) and EDTA plasma aliquoted within 24 h, then stored at −80°C. For analysis, samples were de-identified and randomly processed at Albany University Hospital. Biomarkers and laboratory analysis: A panel of methylation regions across 20 genes was selected from systematic reviews and key publications. Two targeted markers per gene were designed; primer/probe details are in supplements. Cell-free DNA was extracted from 200 µL plasma using easyMAG/NucleoSpin, eluted in 2×25 µL. Sodium bisulfite conversion was followed by multiplex PCR (Agena EpiTyper) to assess methylation. M-values were used for QC and calibration, including correction for PCR efficiency; commercially methylated control DNA was included in each run. Variables: Outcome: CRC diagnosis within 24 months (binary). Predictors: methylated/not methylated for each marker; age, sex, BMI, smoking (pack-years), diabetes, hemoglobin (Hb), CEA; TNM and AJCC stage, tumor site from records/registry. Missing pack-years (n=16) and BMI (n=11) were imputed with cohort medians. Statistical analysis: Data split by randomization into a test set (70%, n=143) for marker/panel selection and a validation set (30%, n=69) for independent evaluation. For each marker, the Ct cutoff maximizing the Youden index in the test set defined methylation positivity (Ct ≤ cutoff positive; Ct > cutoff or >40 or missing negative). Group comparisons used chi-square and t tests. Logistic regression assessed univariate and multivariable associations (adjusted for age, sex, BMI, smoking). ROC analyses computed AUC, sensitivity, specificity, PPV, NPV, and accuracy. A panel (HUNT-CRC) required at least two of eight specified markers to be positive; adjusted analyses accounted for sex, age, BMI, and smoking.

- Study population: 212 plasma samples (106 CRC cases within 24 months and 106 matched controls) split into test (n=143) and validation (n=69) sets with similar demographics and clinical features; in validation, cases had more smoking exposure than controls (mean 16.0 vs 6.5 pack-years, P<0.05). - Single-marker associations (test set): By chi-square, FLI1, IKZF1, SFRP2, and WNT5A were significantly associated with disease status. In logistic models adjusted for age, sex, BMI, and smoking, significant predictors included FLI1 (OR 3.07, 95% CI 1.12–8.55, P=0.030), IKZF1 (OR 3.54, 95% CI 1.19–10.49, P=0.023), and SFRP2 (OR 2.69, 95% CI 1.11–6.49, P=0.028). VIM could not be modeled due to sparse positives. - Diagnostic performance of single markers (test set): Several markers had moderate AUCs; examples include IKZF1 AUC 0.603 (P=0.034), NPTX3 AUC 0.607 (P=0.027), SEPT9 AUC 0.665. Markers with higher specificity (>70%) reported included BMP3, FLT1, IKZF1, SFRP1, SFRP2, NPTX2, SLCA1, and VIM. - Panel performance (HUNT-CRC, requiring ≥2 of 8 markers positive; adjusted for age, BMI, sex, smoking): • Test set: AUC 0.669 (95% CI 0.580–0.795), sensitivity 43.1% (95% CI 42.7–43.4), specificity 85.9% (95% CI 85.7–86.2), PPV 76.5%, NPV 69.8%, accuracy 64.3%. • Validation set: AUC 0.680 (95% CI 0.553–0.806), sensitivity 47.1% (95% CI 46.5–47.6), specificity 77.1% (95% CI 76.7–77.6), PPV 66.7%, NPV 60.0%, accuracy 62.3%. - Overall: Detection of methylated ctDNA markers up to 2 years before clinical CRC diagnosis is feasible in an unselected population; a simple methylation-specific PCR-based panel yielded moderate sensitivity with relatively high specificity and reproduced performance in an independent validation set.

The study addresses whether methylated ctDNA can serve as an early, minimally invasive CRC screening tool in a population-based setting. Using archived plasma from HUNT participants, several methylation markers, and especially a composite panel, discriminated future CRC cases from controls up to 24 months before diagnosis. Although single-marker performance was limited, combining markers improved overall discrimination. Adjusting for age, sex, BMI, and smoking—factors that influence both methylation and CRC risk—helps ensure robustness. The PCR-based assay is less resource-intensive than sequencing approaches, suggesting practicality for clinical screening or for higher-risk cohorts (e.g., family members), potentially with enhanced sensitivity via repeated testing or larger plasma volumes. While sensitivity remains modest, the high specificity and early detection window indicate ctDNA methylation could complement existing screening methods, increase participation due to minimal invasiveness, and enable earlier, potentially curative interventions.

Methylated ctDNA markers for CRC can be detected up to two years prior to clinical diagnosis in a general, screening-like population. A multi-marker PCR-based panel achieved moderate sensitivity and high specificity, with reproducible performance in an independent validation set. These findings support integrating ctDNA methylation assays into CRC screening strategies to facilitate earlier diagnosis. Future work should include prospective studies, optimization of sample volume and processing, refinement of marker panels, and formal cost-effectiveness analyses to assess clinical implementation.

- Biospecimen handling not optimized for ctDNA at time of collection; potential leukocyte lysis and contamination of cfDNA. - Long-term storage (11–13 years) may have led to ctDNA degradation, potentially underestimating sensitivity. - High false-positive rate in controls at higher Ct values highlights risk of nonspecific detection when running multiple PCRs; stringent Ct cutoffs are needed. - False negatives among cases; limited plasma input volume (≈1.8 mL total) likely reduced sensitivity; larger volumes (e.g., 4–5 mL plasma from 10 mL EDTA blood) could improve detection. - Nested case–control design can introduce bias in diagnostic accuracy studies, though risk minimized by using a population-based cohort without strict inclusion/exclusion criteria. - Possible selection bias between HUNT participants and non-participants cannot be excluded.