1978P - Accurate detection of urothelial carcinoma by whole-genome methylation profiling of urinary cell-free DNA

Background

Urothelial carcinoma (UC) presents a diagnostic challenge due to the lack of non-invasive, accurate tests for distinguishing between malignant and non-cancer cases. Recent studies have suggested that urinary cell-free DNA (ucfDNA) methylation patterns hold promise as potential biomarkers for cancer detection. This study aims to utilize the epigenetic profiles of ucfDNA to enhance the detection accuracy on tumor diagnosis.

Methods

Urine samples were collected from 121 patients with pathologically diagnosed malignant UC and with benign lesions. The ucfDNA was extracted from urine supernatant and subjected to PredicineEPIC^TM approach, a genome-wide methylation assay capable of identifying differentially methylated fragments and tissue-of-origin probability. The XGBoost machine learning algorithm was implemented to establish a cancer/non-cancer classifier. We utilized leave-one-out cross-validation to mitigate sample size constraints and prevent overfitting, with the classifier's performance assessed in the validation set.

Results

We estimated the differentially methylated fragments (DMF) by sample, which showed the significantly higher score of malignant groups (mean = 428, 95% CI 260-596) than the benign group (mean = 7, 95% CI 2-12), Wilcoxon rank sum test P-value = 4*10^-11. Based on the methylated fragment profiles, we then performed tissue-of-origin deconvolution to evaluate the bladder epithelial cell-originated DNA fragments. The results showed the proportion of bladder cell sources in the ucfDNA of UC patients (mean = 7.1, 95% CI 6.1-8.0) was significantly higher compared to benign patients (mean = 3.3, 95% CI 2.6-4.0), Wilcoxon rank sum test P-value = 6*10^-6. The integrated features used by classifier achieved 89% accuracy in distinguish between two groups on validation cohort, demonstrating improved tumor detection performance.

Conclusions

Our study underscores an approach that differentiates malignant from benign lesions using ucfDNA fragment methylation profiles, enhancing the potential and precision of non-invasive liquid biopsy methods for urothelial carcinoma detection.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

H. Dong, H. Tang: Financial Interests, Personal, Full or part-time Employment: Huidu (Shanghai) Medical Technology Co., Ltd. P. Du, S. Jia: Financial Interests, Personal, Ownership Interest: Predicine, Inc. G. Bonora: Financial Interests, Personal, Full or part-time Employment: Predicine, Inc. All other authors have declared no conflicts of interest.