# 5-hydroxymethylcytosine signature in plasma extracellular vesicle DNA as a diagnostic molecular biomarker for precancerous lesions of gastric cancer

## TL;DR

This study identifies a non-invasive blood test using DNA modifications to detect early signs of gastric cancer, offering a promising diagnostic tool.

## Contribution

The study introduces a novel diagnostic model using 5-hydroxymethylcytosine patterns in plasma extracellular vesicle DNA for detecting gastric cancer precursors.

## Key findings

- A diagnostic model with nine differentially hydroxymethylated regions achieved 95.45% sensitivity and 81.82% specificity for detecting precancerous lesions.
- The model showed strong correlation with clinical pathological indicators of precancerous gastric cancer.
- The method outperforms existing assays in accuracy and robustness across different batches.

## Abstract

Aim: Precancerous lesions of gastric cancer (PLGC) represent a critical window for prevention. Developing non-invasive tools that can reliably detect these lesions is therefore a prerequisite for lowering gastric-cancer incidence. Recent work has highlighted the diagnostic promise of plasma extracellular vesicle DNAs (evDNAs) and the 5-hydroxymethylcytosine (5hmC)-Seal epigenomic platform. Here we profiled genome-wide 5hmC patterns in circulating evDNA to discover biomarkers and build a classification model.

Methods: We performed whole-genome 5hmC-Seal on plasma evDNAs from 67 PLGC patients and 67 healthy individuals. By identifying trend-expressed differentially hydroxymethylated regions (DhMRs), we used machine learning algorithms to screen for diagnostic biomarkers of PLGC and established a corresponding diagnostic model.

Results: We ultimately constructed a diagnostic model comprising nine DhMRs. In the test set, the area under the curve (AUC) value was 0.963, with an accuracy of 0.886, sensitivity of 95.45%, and specificity of 81.82%. These results indicate that DhMRs in evDNA can serve as diagnostic biomarkers for PLGC, with good diagnostic capability and reliability. Correlation analysis showed a strong association between the DhMRs in the diagnostic model and clinical pathological indicators of PLGC.

Conclusion: We developed a non-invasive diagnostic model for PLGC by profiling 5hmC in plasma evDNA. In both accuracy and inter-batch robustness, it surpasses all previously reported assays. Our findings establish plasma-evDNA 5hmC profiling as a reliable, minimally invasive strategy for the early detection and precise diagnosis of gastric precancerous lesions, and provide a new translational and clinical framework for future work.