Meta-analysis and interpretable machine learning characterize ARG and MGE co-enrichment under microplastic exposure during anaerobic digestion.

Waste activated sludge is an important sink for microplastics (MPs), and anaerobic digestion (AD), the mainstream process for sludge treatment, may provide conditions that facilitate MP-associated enrichment and dissemination of antibiotic resistance genes (ARGs) and mobile genetic elements (MGEs), posing potential resistance-related risks. However, how MP physical attributes and AD operating conditions are jointly associated with ARG and MGE dynamics remains unclear. Here, we integrated evidence using dependency-aware three-level random-effects meta-analysis and XGBoost-SHAP association analysis, with reactor mode tested as a moderator and generalizability assessed by study-grouped and leave-one-study-out validation. MP exposure was associated with 57.5% and 35.0% increases in ARGs and MGEs, respectively, while responses were broadly consistent across batch and semi-continuous reactors. Tetracycline, β-lactam, and sulfonamide ARGs showed enrichment, whereas MGE responses were concentrated in integron-related elements, particularly intI1. SHAP analysis associated MP shape and aging status more strongly with ARG enrichment and particle size more closely with MGE responses, while incubation duration acted as a shared factor. Joint interaction analysis identified shape- and aging-centred combinations as predominantly ARG-associated, duration-temperature interactions as predominantly MGE-associated, and particle size and duration as links between the two response structures. The design separates pooled-effect estimation and reactor-related moderation from nonlinear association mapping while accounting for dependent observations across studies. Together, the findings support a staged associative framework linking plastisphere establishment, ARG host enrichment, and MGE-associated mobilization, providing a basis for coordinated monitoring and management of MP-related resistance risks in AD systems.