Environmental drivers of high-risk antibiotic resistance genes propagation in the plastisphere unveiled by meta-analysis and interpretable machine learning.

The plastisphere serves as an expanding reservoir and dissemination vector for antibiotic resistance genes (ARGs), yet the environmental driving factors on high-risk ARG dynamics within this niche remain poorly understood. Herein, a multi-effect meta-analysis was conducted to quantify the influence of environmental factors on high-risk ARGs within the plastisphere. Relative to ambient waters, significant enrichment was observed for ARGs targeting macrolide (ermC, +114.47%), quinolone (aac(6')-Ib, + 89.62%), sulfonamide (sul1, +57.16%), quinolone (qnrS, +33.01%), and class I integrons (intI1, +48.61%). Among microplastics, polyethylene and polypropylene exhibited selective ARGs enrichment, exceeding concentration levels of ambient waters by 80.63% and 71.89%, respectively. Mantel and binning analyses quantified contributions of 13 environmental factors to 9 ARG genotypes and intI1. Additionally, molecular fingerprints (n = 92) obtained via RDKit revealed the contributions of microplastics' physicochemical properties. Independent explainable machine learning (ML) models developed using the H2O-AutoML platform for sul1 and intI1 achieved high predictive accuracy. Shapley Additive Explanations (SHAP) analysis identified near-equal associations between intI1 risk and aquatic parameters (49.3%) versus microplastics (MPs) characteristics (50.3%), whereas sul1 risk was predominantly influenced by MPs (91.2%). This study enhances the understanding of the dynamics of ARGs in the plastic cycle and provides a methodological reference for the subsequent development of a predictive framework for rapid, large-scale monitoring of high-risk ARGs in aquatic environments.