Efficacy of cell-free DNA methylation-based blood test for colorectal cancer screening in high-risk population: a prospective cohort study

Abstract

Background: Although colonoscopy is the standard screening test for colorectal cancer (CRC), its use is limited by a poor compliance rate, the need for extensive bowel preparation, and the risk of complications. As an alternative, an FDA-approved stool-based DNA test, Cologuard, has demonstrated satisfactory detection performance for CRC, but its compliance rate remains suboptimal, primarily attributable to individuals' reluctance to provide stool samples.

Methods: We developed a noninvasive blood-based CRC test, ColonSecure, based on cell-free DNA containing cancer-specific CpG island methylation patterns. We initially screened publicly available datasets for differentially methylated CpG sites in CRC with prediction potential. Subsequently, we performed two sequential bisulfite-free methylation sequencing on blood samples obtained from CRC patients and non-cancer controls. Through rigorous evaluation of each marker and machine learning-assisted feature selection, we identified 149 hypermethylated markers from over 193,000 CpG sites. These markers were then utilized to construct the ColonSecure model, enabling accurate CRC detection.

Results: We validated the efficacy of our cell-free DNA methylation-based blood test for CRC screening with 3493 high-risk individuals identified from 114,136 urban residents. The ColonSecure test identified 89 out of 103 CRC patients diagnosed by the follow-up colonoscopy, outperforming CEA, CRP, and CA19-9 (with a sensitivity of 86.4% compared to 45.6%, 39.8%, and 25.2% for CEA, CRP, and CA19-9 respectively; an AUROC of 0.956 compared to an AUROC of < 0.77 for other methods).

Conclusion: Our observations emphasize the potential of our multiple cfDNA methylation marker-based test for CRC screening in high-risk populations.

Keywords: Cell-free DNA; Colorectal cancer; Hypermethylation; Populational screening; Prospective validation.

Fig. 1

Workflow of three major phases for this study: marker discovery, model construction and prospective validation. During the marker discovery phase, two sequential targeted bisulfite sequencing experiments were conducted to identify potential methylation biomarkers with significant methylation alternation in both tissue and blood samples of CRC patients. The first marker discovery probe panel was synthesized based on methylation markers identified from TCGA/GEO databases and previous studies. The second model training probe panel was designed based on the markers further identified through analysis of clinical blood samples. P: positive group; N: negative group. The box icons, appearing in sequence, represent the databases, literature, selection/filtering criteria, and capture probe panel used. During the model construction phase, targeted methylation sequencing using the model training probe panel was performed on 396 clinical blood samples, which were split into training and test sets for model training and evaluation. To prospectively validate the methylation model, 3493 CRC high-risk participants were selected from urban residents based on inclusion and exclusion criteria

Fig. 2

The construction of a cfDNA methylation-based CRC screening model and validation by a prospective high-risk population cohort. A Hierarchically clustered heatmap demonstrating the cfDNA methylation increment of the selected markers in CRC patients of training set. B ROC plots the of cfDNA methylation-based model and commonly used protein biomarkers for predicting CRC in training set. Random forest was used for training the cfDNA methylation-based model. The ROC curve of the methylation model was plotted based on out-of-bag prediction results. C Hierarchically clustered heatmap showing the methylation alternations of the selected markers in CRC patients of test set. D ROC curves of the methylation model and protein biomarkers for CRC detection in test set. E ROC curves for CRC detection by the methylation model and protein biomarkers in the prospective cohort. F Overall sensitivity and specificity. G Sensitivity of the CRC detection model for early and late-stage CRC. H Positive prediction values for cfDNA methylation-based model and protein markers