Background: Circulating tumour DNA (ctDNA) is being explored in the neoadjuvant setting in multiple prospective trials. Dynamic changes may be a surrogate marker of pathological complete response. ABACUS was a multi-centre, single-arm phase II trial investigating two cycles of atezolizumab before cystectomy in patients with muscle-invasive urothelial cancer who were ineligible for or refused neoadjuvant cisplatin-based chemotherapy (NCT02662309). Results of primary and secondary endpoints have been previously published. Here we perform exploratory biomarker analysis using different definitions of ctDNA response and assess correlation with tissue response at time of cystectomy in the ABACUS trial. We also assess how these definitions of response correlate with baseline biomarker expression. Methods: Patients with baseline PDL-1, CD8, TMB and sequential ctDNA measurements (baseline and at cystectomy) were included. ctDNA analysis was performed using the Signatera assay, PD-L1 positivity was defined as ≥5% of immune cell staining, TMB was assessed using the FoundationOne CDx assay and CD8 measurement was performed via immunohistochemistry analyses. Two definitions of ctDNA response were used—ctDNA clearance and a 50% reduction (or greater) in ctDNA variant allele frequency (VAF). Results were correlated with pathological complete response (pCR) rate at time of cystectomy and relapse-free survival. Results: The 2-yr DFS and OS rates in ABACUS were 68% and 77%, respectively (N=95). 40 patients had sequential DNA analysis. 43% (17/40) had pathological complete response and 20% (8/40) experienced relapse. 63% (25/40) patients were ctDNA+ at baseline, with 40% (10/25) having ctDNA response of 50% VAF reduction and 8% (3/40) achieving ctDNA clearance. 30% (3/10) of patients who had VAF reduction experienced relapse and 40% (4/10) achieved pCR. All patients with ctDNA clearance achieved pCR none relapsed. There was no association between VAF reduction of 50% and tissue response (pCR) (p=0.24). Additional analysis showed no association between a ctDNA reduction of 75% (rather than 50%) and pCR (p=0.24). Baseline PDL-1, TMB and not predictive of pathological complete response (p=0.18, p=0.77, p=0.10, respectively) or ctDNA response (p=0.54, p=0.77, p=0.74) to neoadjuvant atezolizumab. There was no difference in relapse rates between ctDNA/PD-L1+ve vs ctDNA+ve/PD-L1-ve patients (16% vs 16%, respectively). Conclusions: ctDNA clearance is rare but appears more accurate than 50% reduction in VAF to predict response/relapse. This is relevant for ongoing neoadjuvant trials planning to use this as an endpoint. Combining immune tissue based biomarkers with ctDNA does not appear to improve biomarker accuracy.