揭示黄芪微生物组的多维异质性：用于生物标志物发现的集成多组学和机器学习框架

Astragalus (primarily represented by A. mongholicus and A. membranaceus) is a widely used functional food whose quality is closely associated with geographical origin, a concept known as geoherbalism. However, the relationship between its microbiota and bioactive constituents across different plant species, production regions, cultivation patterns, and chemotypes remains poorly understood. In this study, we characterized the endophytic microbial communities of Astragalus roots from 61 samples collected across six major cultivation regions in four Chinese provinces using high-throughput amplicon sequencing (targeting the 16S rRNA and ITS regions). We further integrated publicly available rhizosphere metagenomic (PRJNA1119562) and root transcriptomic (PRJNA1119057) datasets to explore microbial functional potential and host-microbe interactions. Machine learning models were developed to predict the levels of key bioactive ingredients (Astragaloside IV [AS-IV] and Calycosin-7-O-β-D-glucoside [CG]), and candidate biomarkers were validated by qPCR. The composition and diversity of endophytic microbiota varied significantly with plant species, geographical origin, cultivation pattern (cultivated, imitated-wild, and wild), and chemotype. Specific taxa, including Penicillium, Coprinopsis, and Pseudomonas, were enriched in high-quality samples, and species-specific microbial signatures were also observed. Rhizosphere metagenomic profiling revealed enhanced energy and amino acid metabolism in low-quality samples, suggesting potential competition for key substrates. Transcriptomic analysis identified differentially expressed genes involved in secondary metabolism, some of which correlated with specific rhizosphere microbes. Lactarius and Alistipes were identified as candidate microbial biomarkers through machine learning and further validated by qPCR assays targeting their corresponding ITS and 16S rRNA gene sequences, respectively, achieving accuracies of 85% and 95% in internal and external validation. However, the relatively limited validation cohorts may not fully represent the geographic and cultivation diversity of Astragalus, and larger-scale validation using samples covering broader geographic origins and cultivation conditions is required to further confirm the robustness and generalizability of these microbial biomarkers. Overall, this study provides a comprehensive characterization of the Astragalus microbiome, highlighting its strong associations with plant species, geoherbalism, cultivation practices, and bioactive accumulation. The validated biomarkers offer promising tools for quality assessment and support microbiome-based strategies for improving the functional quality of Astragalus as a food ingredient.