Pulse-driven and persistent antimicrobial resistance markers in a transboundary Great Lakes connecting channel: pulse-week-stratified water-quality thresholds for One Health surveillance

Abstract

Surface waters in urban watersheds receive episodic inputs of wastewater, runoff, and road-salt residues during spring, yet the contribution of these short hydrological windows to antibiotic resistance gene (ARG) loading remains poorly resolved. Weekly samples were collected from offshore and nearshore sites in the Detroit River, a Great Lakes transboundary connecting channel, from February to December 2025. Five clinically relevant ARGs encoding resistance to carbapenems, methicillin, and colistin, alongside the fecal marker pepper mild mottle virus (PMMoV), were quantified by qPCR and paired with ten conventional water-quality variables. blaNDM, mcr-1, and blaVIM-7 were not detected while blaKPC occurred as discrete pulses. One week (5 May) accounted for 37.2% of annual offshore blaKPC loading, and three weeks in late April to early May accounted for 72.8%, with peak concentrations reaching 5.4 x 103 and 7.2 x 103 copies L{square}{superscript 1}. mecA was detected in nearly all samples without a dominant pulse. PMMoV normalization showed blaKPC did not vary seasonally (Kruskal-Wallis p = 0.198), consistent with diluted wastewater during spring precipitation events rather than an emergent source. mecA/PMMoV varied seasonally (p = 0.003) and was lowest in spring, implicating non-wastewater inputs in summer and autumn. Seven water-quality variables were significantly elevated during blaKPC pulse weeks. PCA distinguished pulse from background conditions, explaining 81.3% of variance. A random forest classifier achieved leave-one-out AUC of 0.917; ROC AUC reached 0.943 for total phosphorus, 0.924 for chloride, and 0.974 for the multivariate model. These results demonstrate that blaKPC and mecA operate through distinct source pathways and that routine water-quality monitoring can flag elevated blaKPC risk without additional sampling infrastructure.