81 Background: Checkpoint inhibition is a promising approach for cancer treatment with durable responses and survival benefits. Currently, immune-checkpoint inhibitors are delivered at maximally tolerated doses, resulting in significant toxicities and cost burden. There is significant need to develop more efficacious and safer immune checkpoint inhibitor. Towards this goal, we have developed a next-generation humanized PD-L1 antibody with pH-dependent antigen binding and evaluated its biological activities in vitro and in vivo. Methods: Hybridoma approach was employed to generate anti-PD-L1 antibodies. A cell-based receptor-ligand blocking assay was employed to identify neutralizing antibodies. Antibodies with pH-dependent antigen binding property were identified and characterized further with those without using PD-L1/PD-1 binding assay, mixed lymphocyte reaction assay, and in vivo tumor growth inhibition assay in syngeneic MC38 tumor model expressing human PD-L1 protein. Results: A panel of neutralizing antibodies blocking PD-L1/PD-L1 binding was isolated. Among them we identified antibodies with or without pH-dependent binding property. While the in vitro activities of these antibodies are similar, their tumor growth inhibition activities differ significantly. The in vivo activity of 23F11, an antibody with ph-dependent binding, is maximized at 1mg/kg in MC38/hPD-L1 model. At 1 mg/kg, 23F11 induced complete tumor regression in 7/12 of the mice treated, more potent than antibodies without pH-dependent binding. Imaging study revealed that significantly higher drug concentration in the tumor for 23F11 and correlated with extended period of target occupancy. Furthermore 23F11 treatment is correlated with sustained up-regulation of plasma mIL-18, a potential pharmacodynamics response marker. Conclusions: We have identified antibodies with pH-dependent PD-L1 binding property. These antibodies can maintain high antibody level in PD-L1 expressing tumor and have potent tumor inhibition activity at low dose. Current work will evaluate whether humanized 23F11 can be at least as efficacious as benchmark antibodies yet with improved safety profile in patients.