New insights into nitrogen removal potential in urban river by revealing the importance of microbial community succession on suspended particulate matter.

The importance of suspended particulate matter (SPM) in nitrogen removal from aquatic environments has been acknowledged in recent years by recognizing the role of attached microbes. However, the succession of attached microbes on suspended particles and their role in nitrogen removal under specific surface microenvironment are still unknown. In this study, the causation among characteristics of SPM, composition and diversity of particle-attached microbial communities, and abundances of nitrogen-related genes in urban rivers was firstly quantitatively established by combing spectroscopy, 16 S rRNA amplicon sequencing, absolute gene quantification and supervised integrated machine learning. SPM in urban rivers, coated with organic layers, was mainly composed of silt and clay (87.59-96.87%) with D50 (medium particle size) of 8.636-30.130 μm. In terms of material composition of SPM, primary mineral was quartz and the four most abundant elements were O, Si, C, Al. The principal functional groups on SPM were hydroxyl and amide. Furthermore, samples with low, medium and high levels of ammoxidation potential were classified into three groups, among which significant differences of microbial communities were found. Samples were also separated into three groups with low, medium and high levels of denitrification potential and significant differences occurred among groups. The particle size, content of functional groups and concentration of SPM were identified as the most significant factors related with microbial communities, playing an important role in succession of particle-attached microbes. In addition, the path model revealed the significantly positive effect of organic matter and particle size on the microbial communities and potential nitrogen removal. The content of hydroxyl and temperature were identified as the most effective predicting factors for ammoxidation potential and denitrification potential respectively by Random Forests Regression models, which had good predictive performances for potential of ammoxidation (R = 0.71) and denitrification (R = 0.61). These results provide a basis for quickly assessing the ability of nitrogen removal in urban rivers.