Host-linked virome assembly and turnover predict bacterial community structure in wastewater treatment systems

Viruses play crucial roles in bacterial ecology and evolution through virus-host interactions; however, their distribution, assembly mechanisms, and temporal turnover remain underexplored in engineered ecosystems. In the present study, we used activated sludge (AS) and anaerobic treatment (AT) reactors from four full-scale industrial textile wastewater treatment plants as model ecosystems, integrating metagenomics, macroecological modeling, and deep learning to characterize viral structure, dynamics, and host interactions. A total of 1046 and 1386 high-quality viral operational taxonomic units were recovered from AS and AT systems, respectively, and most were affiliated with Caudoviricetes. Viral composition and genetic microdiversity were highly plant-specific and shaped by environmental selection and host interactions. Lognormal species abundance distributions and deviations from neutral expectations indicated deterministic assembly. Virulent viruses exhibited faster temporal turnover than temperate viruses. Viral co-occurrence networks showed strong plant-specific modularity and greater temporal stability than bacterial networks, suggesting that they play a stabilizing role in community dynamics. Tight virus-host abundance coupling and gene-level signatures of host-linked selection indicated ongoing coevolutionary interactions. A deep learning model accurately predicted bacterial community dynamics from viral composition at both the taxon and sample levels, highlighting the ecological relevance of viral signatures. Together, these findings reveal dynamic, plant-specific viromes tightly coupled to bacterial communities and highlight viral signatures as potential indicators for monitoring engineered ecosystems. Incorporating viral ecology into microbial management could enhance the stability, resilience, and functional performance of engineered ecosystems.