The present invention provides therapeutic and diagnostic methods and compositions for cancer, for example, lung cancer (e.g., NSCLC), bladder cancer (e.g., UC), kidney cancer (e.g., RCC), breast cancer (e.g., TNBC), or melanoma. The invention provides methods of treating cancer (e.g., lung cancer (e.g., NSCLC), bladder cancer (e.g., UC), kidney cancer (e.g., RCC), breast cancer (e.g., TNBC), or melanoma), methods of determining whether a patient suffering from cancer (e.g., lung cancer (e.g., NSCLC), bladder cancer (e.g., UC), kidney cancer (e.g., RCC), breast cancer (e.g., TNBC), or melanoma) is likely to respond to treatment comprising a PD-L1 axis binding antagonist, methods of predicting responsiveness of a patient suffering from cancer (e.g., lung cancer (e.g., NSCLC), bladder cancer (e.g., UC), kidney cancer (e.g., RCC), breast cancer (e.g., TNBC), or melanoma) to treatment comprising a PD-L1 axis binding antagonist, and methods of selecting a therapy for a patient suffering from cancer (e.g., lung cancer (e.g., NSCLC), bladder cancer (e.g., UC), kidney cancer (e.g., RCC), breast cancer (e.g., TNBC), or melanoma), based on a tissue tumor mutational burden (tTMB) score, which reflects somatic mutation levels of genes in a tumor tissue sample obtained from the patient, alone or in combination with PD-L1 expression levels (e.g., PD-L1 expression levels in tumor or tumor-infiltrating immune cells in a tumor sample (tumor area) obtained from the patient).