The INSPECTOR study: enhanced feasibility for clinical translation of a multi‐cancer early detection method based on enzyme‐assisted high signal‐to‐noise ratio sequencing of methylated circulating tumor DNA
Abstract
Background
Blood‐based cell‐free DNA (cfDNA) methylation testing has emerged as a promising approach for multi‐cancer early detection (MCED), holding the potential to improve cancer survival rates. However, traditional bisulfite‐based methods often encounter sensitivity limitations in detecting early‐stage malignancies or certain cancer types. In the INSPECTOR study, we developed a MCED and cancer signal origin (CSO) system specifically designed for early‐stage or hard‐to‐detect cancers, including those of the lung, breast, colorectum, liver, esophagus, stomach, pancreas, and ovary.
Methods
We established a comprehensive methylation marker discovery database (
n
= 6,342) by integrating public datasets (
n
= 4,699) and in‐house samples (
n
= 1,643), all processed using human TET (hTET) enzyme‐assisted whole‐methylome sequencing (GM‐seq). This enabled the design of a targeted panel encompassing 155,362 methylated CpG sites. Leveraging hTET‐assisted high‐depth next‐generation sequencing (NGS), our blood test achieved a median unique depth of 1,093×. Multicenter case‐control cohorts, including various pathological subtypes, were used for training, validation, and independent validation of MCED and CSO models, and to verify the clinical feasibility.
Results
Clinical validation was conducted across multi‐center case‐control cohorts, including 1,071 participants in the training set, 581 in the validation set, and 824 in the independent validation set. The MCED assay demonstrated robust performance with a specificity of 99.1% and sensitivity of 83.2% in the training set, 99.0% and 81.8% in the validation set, and comparable results in the independent validation set (99.0% specificity, 81.9% sensitivity). Notably, sensitivity reached 65.5% for stage I cancers, 79.7% for stage II, and 71.3% for stages I‐II combined. The sensitivities for different cancer types were as follows: esophageal (79.2%), gastric (76.1%), colorectal (86.2%), pancreatic (66.7%), liver (100.0%), lung (72.9%), breast (88.9%), and ovarian (87.9%). The CSO model exhibited strong accuracy, with top‐1 cancer origin prediction rates of 87.9% (validation) and 87.4% (independent validation), rising to 95.1% and 94.5% for top‐2 predictions, respectively. For stage I cancers specifically, the top‐1 accuracy was 85.5%.
Conclusions
These findings underscore the efficacy of the hTET‐assisted cfDNA methylation sequencing system across diverse cancer types, particularly in early stages. Enzyme‐assisted NGS test of methylated cfDNA thus enhances the clinical utility of non‐invasive blood‐based screening.
