Patients with metastatic urothelial carcinoma have few treatment options after failure of platinum-based chemotherapy. In this trial, we assessed treatment with atezolizumab, an engineered humanised immunoglobulin G1 monoclonal antibody that binds selectively to programmed death ligand 1 (PD-L1), in this patient population.For this multicentre, single-arm, two-cohort, phase 2 trial, patients (aged ≥18 years) with inoperable locally advanced or metastatic urothelial carcinoma whose disease had progressed after previous platinum-based chemotherapy were enrolled from 70 major academic medical centres and community oncology practices in Europe and North America. Key inclusion criteria for enrolment were Eastern Cooperative Oncology Group performance status of 0 or 1, measurable disease defined by Response Evaluation Criteria In Solid Tumors version 1.1 (RECIST v1.1), adequate haematological and end-organ function, and no autoimmune disease or active infections. Formalin-fixed paraffin-embedded tumour specimens with sufficient viable tumour content were needed from all patients before enrolment. Patients received treatment with intravenous atezolizumab (1200 mg, given every 3 weeks). PD-L1 expression on tumour-infiltrating immune cells (ICs) was assessed prospectively by immunohistochemistry. The co-primary endpoints were the independent review facility-assessed objective response rate according to RECIST v1.1 and the investigator-assessed objective response rate according to immune-modified RECIST, analysed by intention to treat. A hierarchical testing procedure was used to assess whether the objective response rate was significantly higher than the historical control rate of 10% at an α level of 0·05. This study is registered with ClinicalTrials.gov, number NCT02108652.Between May 13, 2014, and Nov 19, 2014, 486 patients were screened and 315 patients were enrolled into the study. Of these patients, 310 received atezolizumab treatment (five enrolled patients later did not meet eligibility criteria and were not dosed with study drug). The PD-L1 expression status on infiltrating immune cells (ICs) in the tumour microenvironment was defined by the percentage of PD-L1-positive immune cells: IC0 (<1%), IC1 (≥1% but <5%), and IC2/3 (≥5%). The primary analysis (data cutoff May 5, 2015) showed that compared with a historical control overall response rate of 10%, treatment with atezolizumab resulted in a significantly improved RECIST v1.1 objective response rate for each prespecified immune cell group (IC2/3: 27% [95% CI 19-37], p<0·0001; IC1/2/3: 18% [13-24], p=0·0004) and in all patients (15% [11-20], p=0·0058). With longer follow-up (data cutoff Sept 14, 2015), by independent review, objective response rates were 26% (95% CI 18-36) in the IC2/3 group, 18% (13-24) in the IC1/2/3 group, and 15% (11-19) overall in all 310 patients. With a median follow-up of 11·7 months (95% CI 11·4-12·2), ongoing responses were recorded in 38 (84%) of 45 responders. Exploratory analyses showed The Cancer Genome Atlas (TCGA) subtypes and mutation load to be independently predictive for response to atezolizumab. Grade 3-4 treatment-related adverse events, of which fatigue was the most common (five patients [2%]), occurred in 50 (16%) of 310 treated patients. Grade 3-4 immune-mediated adverse events occurred in 15 (5%) of 310 treated patients, with pneumonitis, increased aspartate aminotransferase, increased alanine aminotransferase, rash, and dyspnoea being the most common. No treatment-related deaths occurred during the study.Atezolizumab showed durable activity and good tolerability in this patient population. Increased levels of PD-L1 expression on immune cells were associated with increased response. This report is the first to show the association of TCGA subtypes with response to immune checkpoint inhibition and to show the importance of mutation load as a biomarker of response to this class of agents in advanced urothelial carcinoma.F Hoffmann-La Roche Ltd.