EACR26-0091
Esophageal cancer (EC) remains one of the most lethal malignancies worldwide. Current diagnostic modalities are invasive and suboptimal for early detection. Moreover, accurate noninvasive tools for distinguishing EC subtypes and precancerous lesions are urgently needed. Circulating cell-free mitochondrial DNA (ccf-mtDNA) fragmentomics has emerged as a promising biomarker source due to its abundance and distinct fragmentation patterns. This multi-center study aimed to develop and validate machine learning models based on plasma ccf-mtDNA fragmentomics for early EC detection and pathological subtyping.
This multi-center study enrolled 1,180 participants, including healthy controls (HC), patients with benign esophageal diseases (BED), precancerous lesions, and EC patients. Plasma samples were collected, and ccf-mtDNA was extracted and subjected to fragmentomic analysis.An EC Detection (ED) model and a pathological subtyping prediction (PSP) model. Key clinical variables were integrated to enhance subtyping accuracy.
The ED model achieved AUCs of 0.9706 (EC vs. non-EC) in the training cohort and maintained high performance in internal validation (AUC = 0.9762), external validation (AUC = 0.9748), and a Chinese minority (AUC = 0.9668) cohort for distingushing EC from non-EC. At the optimal cut-off of 0.5167, the sensitivity for EC patients at stage I was 92.16% in the combined validation cohort. Strikingly, the model correctly identified 93.10% of cases where preoperative biopsy had underestimated early EC. Notably, fragmentomic profiles revealed a molecular continuum from from BED to low-grade intraepithelial neoplasia (LGIN), high-grade intraepithelial neoplasia (HGIN), and EC, enabling the model to distinguish HGIN from non-EC with an AUC of 0.9614. Furthermore, the PSP model differentiated esophageal squamous cell carcinoma (ESCC) from esophageal adenocarcinoma (EAC) with an accuracy of 89.71%, which improved to 90.74% after integrating key clinical variables. These findings establish ccf-mtDNA fragmentomics as a powerful tool for early EC detection and accurate tumor subtyping.
In summary, ccf-mtDNA fragmentomics provides a robust, non-invasive platform for early EC detection, precancerous risk stratification, and histological subtyping, demonstrating strong potential to transform clinical management.
We thank Yuqing Yang, Zhaoquan Su, Xiumin Ma, Zhiyun Gong, Renquan Lu for their contributions to this study. This work was supported by the National Natural Science Foundation of China (grant no. 82330073, 82402731), the Science & Technology Co-ordination and Innovation Project of Shaanxi Province, China (grant no. 2024SF-ZDCYL-02-02), Clinical Research Project of Air Force Medical University (2023LC2324 and 2023LC2315).