Keystone Microbial Taxa Identified by Deep Learning Reveal Mechanisms of Phosphorus Stoichiometric Homeostasis in Submerged Macrophytes Under Different Hydrodynamic States

Phosphorus (P) pollution in aquatic ecosystems triggers eutrophication, disrupting ecological processes. Although phytoremediation using submerged macrophytes is promising, its efficacy depends on plant-microbe interactions and stoichiometric homeostasis. A significant knowledge gap exists regarding the assembly and impact of key microbial communities on stoichiometric homeostasis under fluctuating environmental conditions, hindering the optimization of phytoremediation strategies. Given that hydrodynamic fluctuations are a primary source of environmental variability in aquatic systems, this study explored the intricate relationships among stoichiometric homeostasis, microbial community structure, and ecosystem stability, with a specific focus on their impact on rhizosphere P metabolism in Vallisneria natans and Myriophyllum spicatum under different hydrodynamic states. A Deep Learning-based Keystoneness Taxa Identification (DLKTI) framework was developed to identify key microbial taxa. Microbial community stability analysis preceded key taxa determination to enhance result reliability and ecological relevance based on the premise that distinct states provide a more dependable baseline for attributing observed changes to specific perturbations rather than to inherent fluctuations. These findings indicate that the key taxa identified by the DLKTI framework adequately characterized the overall ecological features of the microbial community (average ρ = 0.39, p < 0.05). Moreover, including microbial pools and diversity indices of the screened key microbial taxa improved the explanatory power for submerged macrophyte traits (5% and 6%, respectively) and rhizosphere oxidative stress responses (25% and 4%, respectively). Partial least squares path modeling demonstrated the crucial role of stoichiometric homeostasis for P in ecosystem functioning (path coefficient of inhibition of phytoplankton growth = 0.58, p < 0.001). The findings elucidating plant-microbe interaction patterns under different hydrodynamic states allow for the development of targeted interventions to enhance rhizosphere P metabolism, thereby increasing the efficiency of phytoremediation for eutrophication management and aquatic ecosystem restoration.