Cfdna fragmentomes for non-invasive early detection of HCC

The National Cancer Institute awarded The Johns Hopkins University $645,435 on September 2, 2026, under the Cancer Detection and Diagnosis Research program (CFDA 93.394) to develop a multimodal blood-based classifier for early hepatocellular carcinoma (HCC) detection.

The research integrates genome-wide methods—fragmentomics (DELFI), repeat landscapes (ARTEMIS), methylomics, mutational signatures (GEMINI), and plasma proteomics—into a single machine learning classifier to improve HCC surveillance, particularly for AFP-negative tumors. The classifier will be tested using plasma samples from two cohorts: the DTECT Liver Study from Vietnam, which includes HCC patients and at-risk controls, and the STOP-HCC Study spanning Vietnam and Saudi Arabia, involving high-risk patients under semi-annual surveillance. The research aims to develop a more generalizable HCC surveillance test applicable to U.S. populations, including those affected by hepatitis B virus, hepatitis C virus, and nonalcoholic fatty liver disease.

Specific aims include examining genome-wide changes in cell-free DNA and tissue, developing the multimodal plasma classifier, and validating its performance in an independent cohort. Preliminary data indicate that genome-wide cell-free DNA fragmentation profiles and repeat landscapes can distinguish HCC patients from controls with high sensitivity and specificity.

The award period runs from September 2, 2026, through August 31, 2031, with performance based in Baltimore, Maryland.

Abstract

Hepatocellular carcinoma (HCC) is a leading cause of cancer mortality, with over 60% of patients diagnosed at advanced stages, resulting in a 5-year survival rate below 20%. Early detection significantly improves survival rates, but current surveillance tools like alpha-fetoprotein (AFP) and ultrasound are inadequate, especially for AFP-negative tumors. This study aims to develop a multimodal blood-based classifier for early HCC detection by integrating genome-wide methods we have created, including fragmentomics (DELFI), repeat landscapes (ARTEMIS), methylomics, mutational signatures (GEMINI), and plasma proteomics into a single machine learning classifier. The proposed approach will be tested using plasma samples from two cohorts: the dtect liver study from vietnam includes HCC patients and at-risk controls, while the stop-HCC study from vietnam and saudi arabia involves high-risk patients under semi-annual surveillance. Vietnam and saudi arabia have large populations with chronic liver disease, allowing timely accrual of early-stage and AFP-negative HCC cases that would be impractical and more costly to achieve in the US, while representing disease etiologies relevant to US patients. Importantly, these collaborations directly benefit the US by enabling development and validation of a more generalizable HCC surveillance test for US populations, including those affected by HBV, HCV, and nafld-associated liver disease. Specific aims include examining genome-wide changes in cfdna and tissue, developing a multimodal plasma classifier, and validating its performance in an independent cohort. Our preliminary data show that genome-wide cfdna fragmentation profiles (DELFI) and repeat landscapes (ARTEMIS) can distinguish HCC patients from controls with high sensitivity and specificity. The integration of these features with methylomics and proteomics is expected to enhance early detection, particularly for AFP negative tumors. The study also explores the potential of genome-wide mutational signatures (GEMINI) and cfdna fragment end positions as surrogate measures of DNA methylation. This approach has broad clinical applications in cancer screening, early diagnosis, and disease monitoring, potentially reducing cancer morbidity and mortality, especially in underserved communities. By leveraging established international surveillance cohorts while performing advanced assay development and computational analyses in the US, this project will accelerate a US-led, scalable approach to HCC surveillance. The success of this approach could transform HCC surveillance, making IT more sensitive, accessible, and scalable across populations.