Bistable States of Microbial Communities Driven by Nutrient Loading in the Hyporheic Zone of Effluent-Dominated Rivers: Predicting Taxonomic Composition and Metabolic Functions

Abstract

Effluent discharge from wastewater treatment plants alters nitrogen and sulfur cycling in the hyporheic zone (HZ), potentially shifting microbial communities to alternative stable states. However, these transitions remain poorly understood in such specific subsurface environments. Here, we characterized and predicted multiple stable states of communities in the HZ of representative effluent-dominated rivers by integrating molecular techniques, alternative stable states theory, and machine learning models. The results revealed the existence of bistable states in terms of microbial taxa, functional genes, and metabolic pathways. The potential analysis demonstrated that with increases in nitrogen and sulfur loading, the taxonomic composition shifted from a state with higher diversity and lower stability to one with more prominent interspecific competition. The regime shift in metabolic functions was likely the initial transformation, as it was subsequently followed by alterations in the taxonomic composition. Optimized random forest and XGBoost models combined with network embedding achieved over 90% accuracy in predicting taxonomic composition and metabolic functions, outperforming stand-alone machine learning models. The generated results demonstrated that the accurate description and prediction of microbial responses to anthropogenic disturbances, e.g., effluent discharge, required the joint evaluation of variability in community structure and metabolic function.