Tumor fraction and copy number burden from urinary cell-free tumor DNA (utDNA) to predict minimal residual disease prior to repeat-transurethral resection in high-risk non-muscle invasive bladder cancer (HR-NMIBC).

4600 Background: Genomic alterations detected in urinary cell-free tumor DNA (utDNA) are concordant with those from tissue samples in patients with urothelial carcinoma. In this prospective study, we demonstrate that utDNA alterations and copy number variation can be used to detect minimal residual disease (MRD) prior to standard-of-care repeat transurethral resection of bladder tumors (rTURBT) in patients with high risk non-muscle invasive bladder cancer (HR-NMIBC). Methods: A total of 53 patients with high-risk NMIBC were enrolled prior to rTURBT. PredicineWES+ whole exome sequencing with boosted 600+ oncogene coverage was performed on the index TURBT (iTURBT) tissue. Urine samples were collected immediately prior to rTURBT (preUR). Low Pass Whole Genome Sequencing (LP-WGS) was used to detect utDNA and estimate copy number burden (CNB). Tumor fraction (tf) detected in utDNA was measured using PredicineBEACON personalized ultra-deep sequencing MRD probes designed from somatic variants detected in iTURBT tissue. MRD probes leverage up to 60 patient-specific alterations and a fixed 500 gene hotspot panel. Analysis included receiver operator curves (ROC) to establish CNB and tf efficacy in detecting residual cancer, based on the pathological findings from rTURBT. Specific thresholds for test positivity were set from ROC analysis and used to evaluate performance characteristics of CNB and tf separately and then in step-wise model. Results: In this HR-NMIBC cohort, residual disease was detected in 36/53 pts (68%) on pathologic evaluation. Using the CNB score from preUR samples, residual disease was detected with an overall AUC of 0.86. Setting the CNB threshold >6.1 yielded a positive likelihood ratio of 11.7 and negative likelihood ratio of 0.37 for detecting MRD. For patients with personalized urinary tf (n=34), a receiver operator curve demonstrated an area under the curve (AUC) of 0.89. Median urinary tf was 3.5% for patients with MRD vs. =6.8% and CNB >6.1. When combined into a step-wise model, the two assays together led to improved performance with 85.7% sensitivity, 100.0% specificity and an overall accuracy of 91.4%. Conclusions: Minimal residual disease can be detected using utDNA prior to rTURBT with high accuracy, making this a promising urinary biomarker. If validated, the combined approach using CNB and tf from utDNA for the detection of MRD can be used to predict patients in need of maximal resection prior to starting intravesical therapy.