Letter to the Editor: clinical utility of urine DNA for noninvasive detection and minimal residual disease monitoring in urothelial carcinoma

Abstract

Current methods for the early detection and minimal residual disease (MRD) monitoring of urothelial carcinoma (UC) are invasive and/or possess suboptimal sensitivity. We developed an efficient workflow named urine tumor DNA multidimensional bioinformatic predictor (utLIFE). Using UC-specific mutations and large copy number variations, the utLIFE-UC model was developed on a bladder cancer cohort (n = 150) and validated in The Cancer Genome Atlas (TCGA) bladder cancer cohort (n = 674) and an upper tract urothelial carcinoma (UTUC) cohort (n = 22). The utLIFE-UC model could discriminate 92.8% of UCs with 96.0% specificity and was robustly validated in the BLCA_TCGA and UTUC cohorts. Furthermore, compared to cytology, utLIFE-UC improved the sensitivity of bladder cancer detection (p < 0.01). In the MRD cohort, utLIFE-UC could distinguish 100% of patients with residual disease, showing superior sensitivity compared to cytology (p < 0.01) and fluorescence in situ hybridization (FISH, p < 0.05). This study shows that utLIFE-UC can be used to detect UC with high sensitivity and specificity in patients with early-stage cancer or MRD. The utLIFE-UC is a cost-effective, rapid, high-throughput, noninvasive, and promising approach that may reduce the burden of cystoscopy and blind surgery.

Keywords: Early detection; MRD; Urine DNA; Urothelial carcinoma; utLIFE.

Fig. 1

The liquid biopsy testing of urine markers in the early detection of UC. A Schematic illustration of utLIFE-UC algorithm. Urine samples were collected from UC patients, as well as healthy controls. The ucfDNA was then extracted from the urine supernatant samples and subject to target sequencing, and the uexDNA was extracted from the urine sediment samples and subject to 1 × WGS. Mutation and large CNV features were extracted, and a base model was constructed. The DNA features were then calculated into a large matrix, which was subsequently trained by three ML methods, RF, LR, and SVM. The SVM method was chosen as the utLIFE-UC algorithm to be validated in independent cohorts. B ROC curve of the utLIFE-UC in the training set. C ROC curve of the utLIFE-UC in the test set. D The landscape of utLIFE-UC and cytology detection results in NMIBC and MIBC. E Diagnostic sensitivity of utLIFE-UC compared to cytology in the training cohort (Fisher’s exact test; **p < 0.01). F ROC curve of the independent TCGA validation cohort. G ROC curve of the independent UTUC validation cohort

Fig. 2

utLIFE-UC MRD analysis in patients with localized bladder cancer. A Line chart of the utLIFE-UC score for 2 groups: patients with pCR and patients with non-pCR (student’s t test; *p < 0.05, **p < 0.01). B, C Stacked bar plots showing the proportions of each group with positive or negative utLIFE-UC scores of the training set (B) and the validation set (C). D The landscape of the utLIFE-UC MRD model, cytology, and FISH. E Diagnostic sensitivity and NPV of the utLIFE-UC MRD model compared to cytology or FISH (Fisher’s exact test; *p < 0.05, **p < 0.01)