Longitudinal gut microbiome dynamics during immunotherapy identify microbial features of clinical benefit in advanced primary liver cancer

Background The gut microbiome has been linked to immune checkpoint inhibitor (ICI) outcomes, but the temporal dynamics of microbial communities during treatment remain poorly characterised. Objective To characterise gut microbiome trajectories during ICI therapy and evaluate whether on-treatment microbial states improve the identification and generalisability of pretreatment biomarker signatures. Design We performed a large prospective longitudinal shotgun metagenomic study of 315 patients with advanced primary liver cancer receiving ICI-based therapy, profiling 777 serial stool metagenomes collected at baseline and at approximately 3-month intervals on treatment. Responders (durable clinical benefit ≥6 months) contributed extended follow-up beyond 18 months. On-treatment windows were used for feature discovery; baseline-trained models were evaluated across all nine public ICI cohorts (n=1204). Results Responders showed higher baseline alpha diversity and distinct community structure. Longitudinal profiling revealed marked ecological remodelling during therapy in both response groups, characterised by reduced network connectivity, increased modularity and strong time point specificity of discriminatory species. Using the on-treatment contrast at ~6 months (T2) as a discovery window, we identified a 16-species panel. A baseline model built from this panel outperformed models based on baseline-only feature discovery and generalised across nine public ICI studies. The resulting gut microbiome-derived immunotherapy outcome score stratified overall and progression-free survival in the discovery cohort (HRs 0.49 and 0.44) and across multiple external datasets, including stable-disease subsets. Conclusion The gut microbiome undergoes structured ecological remodelling during ICI therapy; on-treatment longitudinal windows improve pretreatment signature portability and support microbiome-guided stratification in immuno-oncology.