Pollutant-ARG associations show contrasting hydrometeorological context dependence at aggregate and resistance-class...

Riverine antibiotic resistance genes (ARGs) are commonly interpreted in relation to pollutant concentrations, but these associations may also depend on hydrometeorological context and the resolution at which ARG responses are analyzed. We integrated ARG and pollutant datasets from the Pearl and Xiangjiang rivers with campaign-window precipitation, air temperature and streamflow data to compare pollutant-state, hydrometeorological and combined explanations of total and class-specific ARG variation. At the total-ARG level, pollutant-state gradients captured most of the coherent explainable variation, whereas hydrometeorological variables alone contributed relatively little. At the resistance-class level, supported model assignments were concentrated mainly in pollutant-state and combined models, with little support for hydrometeorology-only models. No interaction between pollutant-state variables and hydrometeorological conditions was retained for total ARG abundance after FDR correction in either basin. In contrast, multiple FDR-retained interactions were identified among resistance classes, with streamflow-related modifiers occurring more frequently than meteorological modifiers in the primary analysis. However, none of these class-level interactions met all evaluated robustness criteria, and their number and identity were sensitive to response specification. These findings should therefore be interpreted as exploratory signals rather than evidence of causal effect modification. These patterns indicate that hydrometeorological context was associated with variation in pollutant-ARG relationships at finer response resolution. Taken together, the contrasting patterns between aggregate and class-specific responses suggest that aggregation may provide less information about heterogeneous context dependence, although the underlying mechanisms remain unresolved. Overall, pollutant-state gradients characterized the aggregate ARG signal, whereas hydrometeorological context was more apparent in class-specific associations.