Abstract Hepatocellular carcinoma (HCC) is a leading cause of cancer death world-wide. In the US, liver cancer has increased in incidence and mortality due to a growing population from South and Central America, which faces higher risk of disease in part due to aflatoxin exposure. Recently, we showed that the DELFI (DNA evaluation of fragments for early interception) approach using genome-wide cell free DNA (cfDNA) fragmentation profiles and machine learning can be used for detection of liver and other cancers. Having previously demonstrated that a DELFI classifier accurately detects HCC in populations from the US and Hong Kong, we evaluated the classifier in other clinically relevant cohorts worldwide. Here we show that approach generalizes to diverse populations, and that including other genomic and protein features from the same blood draw improves performance. We examined plasma samples from 377 individuals, including 244 individuals with HCC and 133 without cancer, including 85 with cirrhosis. Plasma samples were collected from individuals with or without HCC in a case-control study in Guatemala (n=203) and from a prospective collection in Romania (n=174) and cfDNA was analyzed by whole- genome sequencing at ∼10x coverage. The median DELFI scores using a locked classifier (Foda et al., Cancer Discovery, 2023) were higher in both cohorts for patients with cancer across all stages compared to individuals without cancer, regardless of the presence of cirrhosis. Using the locked model with a threshold that corresponded to 80% specificity in prior work, DELFI detected individuals with HCC with sensitivities of 90% and 69% in the two cohorts at specificities of 92% and 86%, respectively. For early-stage HCC within Milan Criteria for liver transplantation, sensitivities were 85% in the case-control cohort and 64% in the prospective cohort. In these cohorts, the fixed DELFI model outperformed AFP at the clinically used threshold of 20 ng/mL with a sensitivity in the combined cohorts of 78% at 91% specificity, compared to 63% sensitivity at 91% specificity for AFP. A combined approach using either DELFI at a threshold that corresponded to a 90% specificity in our previous study or AFP at the clinical threshold resulted in 82% sensitivity (74% in early stage) at 92% specificity and was superior to estimates of AFP and ultrasound for early-stage disease (63% sensitivity at 84% specificity). Individuals who tested positive with DELFI had a significantly shorter overall survival (p=0.002, log rank test), even amongst individuals at the earliest stage, while AFP alone did not stratify survival for patients with early-stage disease. Single-molecule analyses from low coverage WGS of cfDNA revealed genome-wide mutational profiles that were similar to those of HCC and in individuals from Guatemala that were characteristic of aflatoxin exposure. Overall, this work provides insight into the origins of cfDNA in populations at risk for HCC and validates our genome-wide fragmentome approach for non-invasive cancer detection that may facilitate liver cancer screening. Citation Format: Zachariah H Foda, Daniel Bruhm C Bruhm, Akshaya V Annapragada, Shashikant Koul, Sarah Short, Keerti Boyapati, Adrianna Bartolomucci, Vilmos Adleff, Nicholas A Vulpescu, Hope Orjuela, Andrei Sorop, Razvan Iacob, Liana Gheorghe, Simona Dima, Katherine A McGlynn, Manuel Ramírez-Zea, Jillian Phallen, John Groopman, Robert B Sharpf, Victor E Velculescu. Early detection of liver cancer from diverse populations using cfDNA fragmentome and protein biomarkers [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr A050.
Read full abstract