一种设计的活菌治疗剂在预测模型中恢复应答者样微生物组特征并改善免疫检查点抑制剂(ICI)应答

Background: The gut microbiome has been shown to predict response to ICI, but associations of individual species have not been statistically robust or reproducible. Fecal microbial transplant (FMT) has been reported to improve response probability in melanoma and kidney cancer, but drug development has focused on very simple microbial consortia. Methods: We sought to replicate the therapeutic benefits of FMT for improving response to ICIs in a model, and design therapeutic microbial consortia that when added to a patient's gut microbiome could improve response rates to ICIs. We trained a machine learning model to predict response to ICI, based on gut microbiome species profiles from baseline samples in patients starting ICI therapy, and used it to evaluate candidate therapeutic consortia intended to improve response to ICI. We trained a machine learning model to predict responders (R) vs. non-responders (NR) to ICI, based on baseline gut microbiome species profiles from patients starting ICI therapy (n=418). Parameters were fixed and called NR vs. R model. We built therapeutic consortia designed around donor FMT and ICI response-related species and mechanisms. ICI patients can be NR for many reasons, so we next focused on patients where the baseline gut microbiome had good predictive value for IO response (n=257) and asked whether FMT and our therapeutic consortia would improve response probabilities. Results: Publicly-available data from 418 NSCLC patients was analyzed. The median age was 65 years (range 24-92), 260 men, 63% were on at least 2nd line therapy. 19.4% received antibiotics near the start of ICI, and 118 of 271 had tumors with PD-L1 <50%. Our model had mean AUC 0.62 for predicting response in hold-out test sets (patients never seen by the model), and discriminated OS curves (p<0.0001). Focusing on patients where the gut microbiome model had good predictive value (n=257), we showed that in agreement with clinical experience, healthy donor FMT improves response rates in our model when added to baseline patient microbiomes and outcomes are exposure-dependent (i.e.modeling increasing fractions of FMT species engrafting and replacing increasing fractions of patient baseline microbiomes). We showed that our candidate therapeutic consortia improve response rates when added to baseline patient microbiomes, and outcomes are again exposure-dependent. Conclusions: Our candidate therapeutic consortia have promise for improving response rates in patients on ICI therapy. We isolated bacteria to culture these consortia, which we are developing to combine with ICI. Conclusions: Our candidate therapeutic consortia have promise for improving response rates in patients on ICI therapy. We isolated bacteria to culture these consortia, which we are developing to combine with ICI.