Plasma cell-free DNA as a sensitive biomarker for multi-cancer detection and immunotherapy outcomes prediction

Abstract

BackgroundCell-free DNA (cfDNA) has shown promise in detecting various cancers, but the diagnostic performance of cfDNA end motifs for multiple cancer types requires verification. This study aimed to assess the utility of cfDNA end motifs for multi-cancer detection.MethodsThis study included 206 participants: 106 individuals with cancer, representing 20 cancer types, and 100 healthy individuals. The participants were divided into training and testing cohorts. All plasma cfDNA samples were profiled by whole-genome sequencing. A random forest model was constructed using cfDNA 4 bp-end-motif profiles to predict cancer in the training cohort, and its performance was evaluated in the testing cohort. Additionally, a separate random forest model was developed to predict immunotherapy responses.ResultsIn the training cohort, the model based on 4 bp-end-motif profiles achieved an AUC of 0.962 (95% CI 0.936–0.987). The AUC in the testing cohort was 0.983 (95% CI 0.960–1.000). The model also maintained excellent predictive ability in different tumor sub-cohorts, including lung cancer (AUC 0.918, 95% CI 0.862–0.974), gastrointestinal cancer (AUC 0.966, 95% CI 0.938–0.993), and other cancer cohort (AUC 0.859, 95% CI 0.776–0.942). Moreover, the model utilizing 4 bp-end-motif profiles exhibited sensitivity in identifying responders to immunotherapy (AUC 0.784, 95% CI 0.609–0.960).ConclusionThe model based on 4 bp-end-motif profiles demonstrates superior sensitivity in multi-cancer detection. Detection of 4 bp-end-motif profiles may serve as potential predictive biomarkers for cancer immunotherapy.