In an open-label, single-arm, single-center, phase 2 trial, patients will receive pembrolizumab 200 mg intravenously (IV) on D1 and nab paclitaxel at the dose of 125 mg/m2 IV on D1 and D8. Cycles are repeated every 3 weeks until PD or onset of unacceptable toxicity. Key inclusion criteria are: predominant UC, failure of maximum 2 platinum-based CT regimens for metastatic disease (2nd-to-3rd line only). Neoadjuvant/adjuvant CT is counted if relapse occurred 6 months of the last CT cycle. Response is evaluated by RECIST criteria v.1.1 every 2 cycles. PD-L1 expression will be prospectively assessed on both immune cells (IC) and tumor cells at a centralized laboratory (National Cancer Institute Milano). Combined positivity score (CPS) for PD-L1 assessment will be used, as previously reported, and the 10% cutoff will be adopted for the analyses. The primary endpoint of the study is the progression-free survival (PFS). The target is to detect an improvement in the median PFS from 3.0 months (H0) to 5.0 months (H1). To achieve 90% power with a one-sided non-parametric test at the 10% significance level, we estimated that 64 patients must be accrued over 18 months, with follow-up duration of 12 months. PFS will be also analyzed according to the PD-L1 expression. Should the above investigation suggest that the treatment benefit is stronger in patients with CPS 10%, there is the option to expand this cohort up to a maximum of 50 patients. As such, we estimate 85% power to detect the target improvement in PFS. The decision of cohort expansion will rely on predictive power (PP) calculation: a PP 30% will be regarded as promising. Translational analyses will include multiparametric flow cytometry of blood samples, gene expression (RNA-seq, Illumina HiSeq) and mutation profiling of tumor samples (Ion Torrent Personal Genome Machine). These profiles will be matched with response to treatment and PFS/overall survival.