Environmental impact of long-term Bt rice cultivation on soil microbiomes: insights from 15-year field monitoring and interpretable machine learning.

Understanding the long-term ecological consequences of genetically modified crops is essential for evaluating ecosystem stability under agricultural intensification. Here, we combined 15 years of continuous field observations with interpretable machine learning to examine how Bacillus thuringiensis (Bt) rice cultivation influences soil bacterial communities. Across all samples, α- and β-diversity metrics showed no significant differences between Bt and non-Bt soils, whereas temporal variation, crop growth stage (before sowing and after harvest), and particularly soil depth emerged as dominant drivers of microbial community structure. Using an Elastic Net model, we identified a set of microbial taxa (including ASV1403 Corynebacteriales, ASV4837 Oscillospirales, and ASV893 Micrococcales) that consistently contributed to the discrimination of Bt and non-Bt soils, despite exhibiting minimal abundance differences in conventional differential analyses. These taxa may reflect subtle but ecologically relevant shifts associated with long-term Bt rice cultivation, although their functional roles require further validation. Overall, our results indicate that Bt rice exerts limited influence on soil bacterial community structure relative to broader temporal and edaphic controls. By integrating long-term ecological monitoring with interpretable machine learning, this study provides a data-driven and transparent framework for environmental risk assessment of genetically modified crops and highlights the importance of spatiotemporal variability in shaping soil microbiome responses under agricultural management. One-sentence summary: Long-term field evidence shows that Bt rice has minimal effects on soil bacterial communities, with temporal and depth-related factors dominating microbial variation.