Quantitative Assessment of Nanoplastic Toxicity Risks Across Aquatic Trophic Levels with Data-Driven Models and Exposure Experiments.

Emerging contaminants such as nanoplastics (NPs) pose significant risks to aquatic ecosystems. However, the key features governing NPs toxicity and the variations in their effects across different trophic levels remain poorly understood. Herein, we integrated machine learning with meta-analysis to comprehensively evaluate NPs toxicity towards aquatic organisms under different conditions, based on 944 data points extracted from 60 studies for meta-analysis and 661 data points from 44 studies for machine learning modeling. Random Forest (RF) demonstrated the highest predictive accuracy (R2 = 0.881) for NPs toxicity, identifying surface charge of NPs as the most influential factor determining NPs toxicity. NPs exposure significantly inhibited the survival, growth, photosynthetic efficiency, and reproduction in aquatic organisms, with effect sizes of ‒73.4%, ‒44.4%, ‒36.6%, and ‒23.9%, respectively. Notably, the inhibition on reproduction reached a maximum of 185.2% at the exposure concentration of 10 mg/L NPs. Compared to fish and plants, invertebrates were more sensitive to NPs exposure due to thin epidermis and high specific surface areas. Exposure experiments on H. cumingii revealed that NPs accumulated primarily in the gills under short-term exposure, causing a 362.3% increase in ROS levels. In contrast, long-term exposure resulted in NPs accumulation in the intestine of H. cumingii, accompanied by a 571.2% rise in intestinal ROS. These experimental findings were highly consistent with predictions from data-driven models, with R2 values of 0.804 for growth and 0.754 for oxidative stress, respectively. Our results underscore the importance of quantifying key NPs characteristics and organismal indicators for designing more rational toxicological studies and objectively assessing the environmental risks of emerging contaminants.