Predicting immunotherapy outcomes of melanoma patients before treatment by intestinal microbiota.

Abstract

e21512 Background: To find the appropriate patients that can benefit from immunotherapy is very important in precision medicine. Biomarkers such as PD-L1, TMB and MSI attracted a lot of attentions in immunotherapy. Emerging biomarkers such as intestinal microbiota and others also showed promising predictive/prognosis values in the clinical studies. The purpose of this study was to evaluate new prediction models based on intestinal microbiota data from the melanoma patients before treatment. Methods: This study enrolled 41 microbiome 16S patients from published datasets. Firstly, survival analysis was performed on the response group (R) and non-response group (NR). Species in all the samples were included to perform univariate cox regression analysis, and the differences of intestinal microbiota between R and NR groups were evaluated by alpha and beta diversities. The specific differential microbiota was obtained by Student's t-test and LEfSe and further screened by the best performance of the five machine learning methods. To find the minimum set of the microbiota species, logistic regression and ROC curve were used. Analyses of KO data included PCA, differential and KEGG enrichment analysis. Significantly enriched pathways were scored by ssGSEA. The spearman correlations between ssGSEA scores and final microbiota species were calculated. Results: The non-response (NR) group had shorter PFS. Univariate cox regression analysis on all the samples identified 36 species which might be associated with prognosis. Alpha and beta diversities showed the R and NR groups were significantly different. Student's t-test identified 52 differential species and LEfSe identified 72 OTUs. Most of the microbiome enriched in R group belonged to firmicutes, while those enriched in NR group belonged to firmicutes, bacteroides and proteobacteria. We then constructed five machine learning models to predict PD1 immunotherapy responses of the melanoma patients and finally identified 25 species. Logistic regression and ROC curve identified the minimum set of the microbiota species including methanobrevibacter smithii, schaalia odontolytica, cuneatibacter caecimuris. A total of 119 differentially expressed KO genes were obtained by differential analysis, and those genes were mainly enriched in 25 pathways. Spearman correlation analysis found that those three microbial species were mainly associated with 12 pathways which might influence the classification of the two groups. Conclusions: In this study we identified three microbial species which might contribute to the classification of the R and NR patient groups. The models based on intestinal microbiota data can predict immunotherapy outcomes of melanoma patients.