Non-invasive early detection of gastric cancer using red blood cell DNA genomic signature: A multi-center validation study.

Abstract

452 Background: Early detection of gastric cancer (GC) is critical for improving survival but remains challenged by the invasiveness of endoscopy and the limited sensitivity of current blood-based assays. DNA fragments retained in mature red blood cells (rbcDNA) represent a form of cytoplasmic DNA arising from genomic instability during erythropoiesis. Emerging evidence suggests that solid tumors can remotely perturb hematopoietic cells, leading to altered rbcDNA patterns. Unlike cell-free DNA, which relies on direct tumor shedding, rbcDNA captures the systemic effects of tumor-induced genomic stress, enabling signal detection even in patients with low tumor burden. This study reports the first multi-center validation of an rbcDNA-based assay for early GC detection. Methods: A total of 865 participants were enrolled, including 434 patients with stage I–III GC (~60% stage I) and 431 non-GC controls with superficial gastritis, atrophic gastritis, or intestinal metaplasia, and H. pylori status was available for all participants. rbcDNA was extracted from 1–2 mL of peripheral blood and subjected to shallow whole-genome sequencing (~2×). Among them, 544 participants were randomly assigned to a discovery ( N = 435) and an internal test set ( N = 109) for model development. GC-associated rbcDNA features were defined as genomic regions whose normalized rbcDNA read counts differed significantly between GC patients and controls in the discovery cohort. These features were summarized by principal component analysis and used to train an ensemble machine learning classifier. The locked model was evaluated across three independent cohorts from medical centers in Zhejiang, Henan, and Shandong, China. Results: The rbcDNA classifier achieved an AUC of 0.93 in the discovery set and 0.95 in the internal test set, with 83% overall sensitivity and 79% sensitivity for stage I GC at 90% specificity. Performance was stable across sex, age, H. pylori status, tumor size, and histological subtypes. Independent validation across the three external cohorts demonstrated consistent performance, with AUCs of 0.90–0.95 and sensitivities of 83%, 90%, and 86%, respectively. Across all external cohorts, sensitivities were 85% for stage I and 86% for stage II–III GC using the same specificity threshold. Population-scale simulations suggested that rbcDNA testing in high-risk individuals could nearly double early GC detection compared to standard endoscopy, with higher adherence and lower false-negative rates. Conclusions: rbcDNA offers a novel, non-invasive strategy for early GC detection, with robust and consistent performance across multiple centers, supporting its potential for broad clinical application and large-scale screening.