Urine- and Plasma-Based Detection of Minimal Residual Disease in Localized Bladder Cancer Patients

Abstract

Standard-of-care for treating muscle-invasive bladder cancer (MIBC) typically involves radical cystectomy, which is a morbid procedure. Nonoperative treatment including chemoradiation is a standard-of-care alternative, but it requires frequent invasive monitoring to assess for response and recurrence. We sought to develop a noninvasive liquid biopsy solution by applying CAPP-Seq to urine and plasma samples from localized bladder cancer patients obtained pre-operatively, which we hypothesized would correlate with pathologic complete response (pCR) assessed by surgery.We acquired urine and plasma samples from 40 localized bladder cancer patients with an indication for radical cystectomy. CAPP-Seq variant-calling was performed noninvasively without tumor mutational knowledge using a 145 kb panel consisting of 49 consensus driver genes. Urine tumor DNA (utDNA) and plasma circulating tumor DNA (ctDNA) per patient were represented by the non-silent driver mutation with the highest variant allele fraction in each respective fluid compartment after removing germline variants. Minimal residual disease (MRD) detection for each biomarker was defined using the optimal threshold from a receiver operating characteristic (ROC) curve for classifying pCR. utDNA MRD was compared with ctDNA MRD using a multivariate logistic regression, controlling for age, gender, and nodal disease. Leave-one-out cross validation was performed to compare the predictive accuracy of each model.The difference in median utDNA between patients with residual disease detected in their cystectomy specimen (n = 25) and those who achieved a pCR (n = 15) was 4.8% (4.8% vs. 0%; P = 0.002). In comparison, the difference in median ctDNA between these two groups was only 0.4% (0.8% vs. 0.4%; P = 0.03). Consistent with this result, utDNA was a superior classifier of pCR than ctDNA based on ROC analysis, with an area under the curve of 0.79 vs. 0.70, respectively. Using MRD detection as a categorical predictor in a multivariable logistic regression, negative utDNA MRD corresponded with 21.6x greater odds of pCR (P < 0.001) than positive utDNA MRD, while negative ctDNA MRD corresponded with 8.1x greater odds of pCR than positive ctDNA MRD (P = 0.007). After cross validation, the utDNA MRD model yielded a highly significant accuracy of 83% with sensitivity of 84% and specificity of 80% (P = 0.005). The ctDNA MRD model yielded a lower and statistically insignificant accuracy of 68% with sensitivity of 72% and specificity of 60% (P = 0.32).Urine-based MRD detection for localized bladder cancer significantly predicted pCR in the cystectomy specimen with a greater effect size, degree of accuracy, and sensitivity and specificity than plasma ctDNA MRD. In the future, utDNA MRD could potentially help identify candidates for bladder-sparing chemoradiation in a personalized and precise way.