2606 Background: A subset of cancer patients could benefit long-termly from immune checkpoints inhibitors (ICIs), which was called smearing effect. Gut microbiota was widely found to be associated with the initial ICIs efficacy and immune-related adverse event (irAE). This study aimed to identify the baseline gut microbiota characteristics of long-benefit patients and explore the possible underlying mechanisms. Methods: Lung cancer patients initially treated with anti-PD-1/PD-L1 therapy were consecutively enrolled from May 2020 to September 2022. All patients were classified into Benefit group, Resistance group and Severe irAE group based on the long-term prognosis. The baseline gut microbiota characteristics were identified by metagenome sequencing. MelonnPan was used to predict the abundances of 79 metabolites. The sequences of autoantigen, lung cancer antigen and anticancer peptides (ACPs) from IEDB and CancerPPD databases were aligned against the scaftigs of fecal samples. Results: 54 patients were enrolled with 24 in Benefit group, 13 in Severe irAE group and 17 in Resistance group. The microorganism composition was significantly different between Benefit group and Resistance group ( P=0.03), while no difference between Benefit group and Severe irAE group ( P=0.93). The co-occurrence network showed the Benefit group patients had the most compact microbiota network. Metabolites prediction analyses showed the short chain fatty acids propionate ( P=0.005) and butyrate/isobutyrate ( P=0.03) were significantly enriched in Benefit group than Resistance group, wherein the acetyl-CoA pathway ( P=0.04) accounted for the largest proportion of butyrate-production ability difference. The beneficial species had more positive correlations with butyrate-producing enzymes ( P<0.001). The microbiota of irAE patients had abundant aligned sequences with antigen epitopes of multiple autoimmune diseases. The total counts of systemic lupus erythematosus-related epitopes were significantly more enriched in patients with immune-related colitis ( P=0.04), especially FDNGRRGRPVTGP ( P=0.02) and IDNGRRGRPITGP ( P=0.02). Likewise, the microbiota of patients with good ICIs efficacy tended to have more abundant cancer-related antigen counts ( P=0.07) and ACPs counts ( P=0.06). Strain-level analysis revealed Escherichia coli and Faecalibacterium prausnitziishowed a centralized trend of genomic variations based on different clinical phenotypes. Conclusions: Long-term benefit patients had converging characteristics of gut microbiota, which coevoluted a compact microbial community with high butyrate-producing ability. The microbiota molecular mimicry of autoimmune and tumor antigen might contribute to irAE and ICIs efficacy. The strain-level genomic variations of specific species may also play a role in clinical phenotypes.