Researchers applied a machine learning classifier to plasma samples from 377 individuals in Guatemala and Romania to analyze genome-wide patterns of cell-free DNA (cfDNA) fragmentation. The classifier detected hepatocellular carcinoma (HCC) across disease stages and underlying causes of liver disease and outperformed alpha-fetoprotein (AFP). Combining the classifier with AFP and clinical risk factors further improved sensitivity, including for early-stage disease. Study findings were published in Cell Press Blue.

“Our earlier studies showed that fragmentome analyses could detect liver cancer and, more recently, chronic liver diseases that increase cancer risk,” said co-senior study author Victor Velculescu, MD, PhD, the Cancer Genetics and Epigenetics Professor, Co-Director of the Cancer Genetics and Epigenetics Program at Johns Hopkins. “This study demonstrates that the approach works with high performance across different patient populations while revealing the biological signals in the bloodstream that make this type of detection possible.”

Study and Model Methods

Previously, the research team developed the machine learning–based liquid biopsy platform DELFI (DNA Evaluation of Fragments for Early Interception), which was used for analyzing cfDNA fragmentation profiles for the detection of lung and liver cancers. In a prior study published in Cancer Discovery, a DELFI model detected HCC with 88% sensitivity at 98% specificity in an average-risk population and 85% sensitivity at 80% specificity among high-risk individuals, with results extending to cohorts in the United States, Europe, and Hong Kong.

The researchers also found that the fragmentomics approach could be used to detect liver fibrosis and cirrhosis.

In this study, the researchers used the locked DELFI model to identify HCC in high-risk independent populations across different stages and etiologies. The two cohorts consisted of patients with and without HCC from Guatemala and Romania to represent different disease etiologies.

They then built a tissue-of-origin algorithm for identifying methylation, which they called MethID, on the cfDNA fragmentomes to deconvolute methylation changes and identify disease-specific changes that came from liver, vascular, or immune cells.

Key Findings

The researchers found that cfDNA fragmentome characteristics such as chromatin, genomic, methylome, mutational, and repeat element profiles were changed in patients with HCC of different etiologies.

The AI classifier outperformed AFP at the clinical threshold of 20 ng/mL in detecting HCC across all stages. The validation cohort included 244 patients with HCC and 133 individuals without cancer, including 86 with cirrhosis. In this cohort, AFP detected 62% of HCCs, including 49% of early-stage cancers, at 93% specificity. Comparatively, DELFI detected 70% of all cancers, and 56% of early-stage HCCs, at 92% specificity.

When cfDNA fragmentomes, AFP, and clinical risk were all combined in one AI model, the model demonstrated higher sensitivity for detecting early- and late-stage HCC in the two geographic cohorts than conventional approaches to disease detection. The stacked model identified 89% of HCCs, including 79% of early-stage cancers, at 83% specificity.

“Importantly, our results show that the combination of DELFI and AFP outperforms AFP alone, the current standard biomarker for HCC detection,” the study authors wrote.

Beyond detection, the fragmentome captured features of tumor biology. Patients with HCC showed characteristic chromosomal alterations, cancer-associated changes in repetitive DNA elements, and mutation signatures that were both shared across populations and specific to each one. In the Guatemala cohort, the analysis detected a mutation pattern associated with aflatoxin exposure. The researchers described this as the first genome-wide identification of the aflatoxin signature in circulating DNA.

MethID showed increased hepatocyte-derived cfDNA in patients with HCC. It also revealed contributions from vascular and immune cells, suggesting that the blood test picks up the body’s response to the tumor as well as the tumor itself.

“These DNA fragments contain much more information than whether cancer is present,” said co-senior study author Zachariah Foda, MD, PhD, Assistant Professor of Medicine at the Johns Hopkins University School of Medicine. “It tells us where these fragments originate and how they change during cancer development, allowing us to better understand the biology of the disease and improve our ability to detect it.”

Going forward, the researchers plan to prospectively validate and refine multimodal liquid biopsy approaches using fragmentomics to improve the early detection of HCC.

DISCLOSURES: This research was supported by the Dr. Miriam and Sheldon G. Adelson Medical Research Foundation, the Commonwealth Foundation, the Cole Foundation, a research grant from Delfi Diagnostics, National Institutes of Health grants, and a Department of Defense grant. Drs. Phallen, Adleff, and Scharpf are co-founders of Delfi Diagnostics, and Adleff, Scharpf, and Foda are consultants. Drs. Phallen, Scharpf, and Foda are inventors on patent applications submitted by Johns Hopkins University related to cfDNA and cancer detection that have been licensed to Delfi Diagnostics. Dr. Velculescu is a founder of Delfi Diagnostics and Artemyx, serves on the board of directors and owns stock of both organizations. Johns Hopkins University owns equity in Delfi Diagnostics. Dr. Velculescu divested his equity in Personal Genome Diagnostics (PGDx) to LabCorp in February 2022. He is an inventor on patent applications submitted by Johns Hopkins University related to cancer genomic and cell-free DNA analyses that have been licensed to one or more entities, including Delfi Diagnostics, Artemyx, LabCorp, Qiagen, Sysmex, Agios, Genzyme, Esoterix, Ventana, and ManaT Bio. Under the terms of these license agreements, the university and inventors are entitled to fees and royalty distributions. Velculescu is an adviser to Viron Therapeutics and Epitope. Drs. Annapragada, Phallen, Scharpf, and Foda are co-founders of Artemyx and hold equity in Artemyx. For data and code availability, visit cell.com.

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.