Background Cancer cell–specific variation and circulating tumour DNA (ctDNA) methylation are promising biomarkers for non‐invasive cancer detection and molecular classification. Nevertheless, the applications of ctDNA to the early detection and screening of cancer remain highly challenging due to the scarcity of cancer cell–specific ctDNA, the low signal‐to‐noise ratio of DNA variation, and the lack of non‐locus‐specific DNA methylation technologies. Methods We enrolled three cohorts of breast cancer (BC) patients from two hospitals in China (BC: n = 123; healthy controls: n = 40). We developed a ctDNA whole‐genome bisulfite sequencing technology employing robust trace ctDNA capture from up to 200 μL plasma, mini‐input (1 ng) library preparation, unbiased genome‐wide coverage and comprehensive computational methods. Results A diagnostic signature comprising 15 ctDNA methylation markers exhibited high accuracy in the early (area under the curve [AUC] of 0.967) and advanced (AUC of 0.971) BC stages in multicentre patient cohorts. Furthermore, we revealed a ctDNA methylation signature that discriminates estrogen receptor status (Training set: AUC of 0.984 and Test set: AUC of 0.780). Different cancer types, including hepatocellular carcinoma and lung cancer, could also be well distinguished. Conclusions Our study provides a toolset to generate unbiased whole‐genome ctDNA methylomes with a minimal amount of plasma to develop highly specific and sensitive biomarkers for the early diagnosis and molecular subtyping of cancer.

The changes in circulating tumor DNA (ctDNA) methylation are believed to be early events in breast cancer initiation, which makes them suitable as promising biomarkers for early diagnosis. However, applying ctDNA in breast cancer early diagnosis remains highly challenging due to the contamination of background DNA from blood and low DNA methylation signals. Here, we report an improved way to extract ctDNA, reduce background contamination, and build a whole‐genome bisulfite sequencing (WGBS) library from different stages of breast cancer. We first compared the DNA methylation data of 74 breast cancer patients with those of seven normal controls to screen candidate methylation CpG site biomarkers for breast cancer diagnosis. The obtained 26 candidate ctDNA methylation biomarkers produced high accuracy in breast cancer patients (area under the curve [AUC] = 0.889; sensitivity: 100%; specificity: 75%). Furthermore, we revealed potential ctDNA methylated CpG sites for detecting early‐stage breast cancer (AUC = 0.783; sensitivity: 93.44%; specificity: 50%). In addition, different subtypes of breast cancer could be well distinguished by the ctDNA methylome, which was obtained through our improved ctDNA‐WGBS method. Overall, we identified high specificity and sensitivity breast cancer‐specific methylation CpG site biomarkers, and they will be expected to have the potential to be translated to clinical practice.