2574 Background: Clinical and preclinical experiments suggest that the gut microbiome can affect outcome in cancer patients treated with immune checkpoint inhibitors (ICI). Most data to date has been in melanoma, so the relationship of the gut microbiome with treatment outcome in other cancers is poorly understood. Here, we evaluated the microbiome composition in correlation to ICI response in patients with metastatic lung, urothelial, or renal cancer, as well as metastatic melanoma. Methods: Fecal microbiome samples were obtained from patients with metastatic melanoma, lung, urothelial, or renal cancer immediately before ICI therapy was initiated. Bacterial genomic DNA was isolated and profiled by whole metagenome sequencing. Sequence data were analyzed using a custom implementation of MetaPhlAn2. Response to ICI was defined as partial or complete response or remaining on therapy for more than 6 months. Results: Samples were prospectively collected from 94 patients, including metastatic melanoma (n = 17), lung (n = 44), urothelial (n = 23), or renal cancer (n = 10). Treatment included anti-PD(L)1 monotherapy (n = 51), anti-PD1 + anti-CTLA4 combination therapy (n = 17), or a combination of anti-PD1 and chemotherapy (n = 26). Clinical response was observed in 58% of patients, including partial or complete response (45%) and on treatment for more than 6 months (55%, with 31% on treatment for more than 1 year). Although the variance in the composition of pretreatment microbiome samples did not explain response alone (R vs NR, PERMANOVA, p = 0.273), a significant portion of the variance in microbiome composition was explained by the interaction of cancer type and outcome (PERMANOVA, p = 0.014), suggesting a cancer-specific microbiome relationship. Notably, there was some similarity in the signature of NR across three cancer types (lung, urothelial and melanoma). One sample in this NR cluster was from a patient whose metastatic NSCLC was nonresponsive to pembrolizumab and carboplatin/pemetrexed. This microbiome sample was evaluated in vivo using subcutaneous MC38 and CT26 tumor models in germ-free mice. In contrast to mice colonized with stool from a healthy donor, mice colonized with stool from this patient yielded a nonresponsive result upon treatment with anti-PD1 or anti-PD-L1 in combination with anti-CTLA4. Conclusions: Analysis of the fecal microbiome composition from patients with metastatic lung, urothelial, renal cancer, and melanoma identified a cancer-specific signature of R and NR to ICI. Across three cancer types, a consistent signature of NR was identified and corroborated experimentally in preclinical models.