Early detection of hepatocellular carcinoma using red blood cell DNA signatures.

Abstract

601 Background: Liver cancer is the third leading cause of cancer-related mortality worldwide, with hepatocellular carcinoma (HCC) accounting for ~90% of cases. Although early detection offers the potential for curative interventions, conventional screening approaches, such as serum alpha-fetoprotein (AFP) testing combined with abdominal ultrasound, exhibit limited sensitivity. Cytoplasmic DNA fragments, such as micronuclei and mitochondrial DNA, have recently been implicated in cancer development and represent promising non-invasive biomarkers. Although mature red blood cells (RBCs) lack nuclei, they retain residual DNA following enucleation, which provides a unique system to investigate cytoplasmic genomic instability. Leveraging this property, we developed a method to detect residual DNA in RBCs (rbcDNA) and demonstrated its association with tumor-induced genomic instability, supporting its potential utility as a biomarker for cancer detection (Sun et al., PMID: 40341742). In this study, we further demonstrate that HCC-specific rbcDNA features have substantial diagnostic value for hepatocellular carcinoma. Methods: A total of 452 participants were enrolled across two batches. rbcDNA was isolated from 1-2 mL of peripheral blood and subjected to whole-genome sequencing. The first batch included 219 non-HCC controls (including individuals with hepatitis, cirrhosis, and no detectable abnormalities underwent abdominal ultrasound) and 136 HCC patients, randomly divided into discovery (70%) and internal test (30%) sets. Machine learning algorithms were applied to identify HCC-associated rbcDNA features and to construct a diagnostic model, which was subsequently blindly validated in both the internal test cohort and an independent cohort (N=97). Results: Leveraging HCC-associated rbcDNA features, the model achieved an AUC of 96% in the internal test cohort, with 90% sensitivity at 98% specificity. Comparable performance was observed in the independent cohort, yielding an AUC of 95% and 93% sensitivity at the same specificity threshold. Notably, the model detected 75% of AFP-negative HCC cases at 98.7% specificity, and when combined with AFP sensitivity improved to 86%. Conclusions: This pilot study demonstrates that rbcDNA is a promising biomarker for the early detection of HCC, exhibiting high sensitivity and specificity. Incorporating rbcDNA profiling into routine screening protocols could further enhance early HCC detection and provide clinical benefits, with further validation in larger cohorts.