High-content imaging driven algal phenotypes enable precise multilevel discrimination of heavy metal stress

Conventional water quality monitoring based on fixed chemical thresholds often fails to capture the integrated biological effects of pollutant mixtures and sublethal stress. To address this limitation, we developed a high-content imaging (HCI)-based pipeline that converts phenotypic changes in algal cells into multidimensional quantitative features for toxicity assessment. A heavy metal stress model was established by exposing Chlorella vulgaris to copper, nickel, and zinc across concentrations ranging from regulatory standards to toxic levels. An integrated HCI analysis workflow was applied to simultaneously capture brightfield morphology, nuclear fluorescence, and chloroplast autofluorescence at single-cell resolution. The multidimensional phenotypic features detected significant sublethal alterations at markedly lower exposure levels than those detected by conventional optical density measurements, and Mantel tests revealed consistent phenotypic response patterns across the examined metals. A Phenotypic Toxicity Screening (PTS) framework constructed from core image-derived features enabled classification into No, Low, and High toxicity categories with high sensitivity and specificity. A multiclass random forest model was further trained on fused features, classifying samples into five concentration-dependent stress categories with high discriminative performance (AUC > 0.956), outperforming models based on single phenotypic endpoints. Furthermore, the framework generalized to binary and ternary metal co-exposure scenarios without retraining, achieving high sensitivity for mixture stress detection. This work establishes an image-driven, phenotype-based pipeline for water quality assessment. By translating subtle cellular signatures into quantitative phenomic fingerprints, our approach provides a scalable, biologically grounded foundation for transitioning from chemical compliance assessment toward an automated, high-throughput, and effect-based monitoring system.