Identifying and exploring the presence of antibiotic resistant Escherichia coli in private groundwater sources in southeastern Ontario via machine learning

According to the World Health Organization, antibiotic resistance represents one of the top global public health and development threats. The environment serves as a significant reservoir and transmission pathway for antibiotic-resistant bacteria (ARB), with anthropogenic activities introducing resistant bacteria, resistance genes, and antibiotic residues to soils and aquatic systems, thereby promoting development and dissemination of resistance. Groundwater has recently been highlighted as an important reservoir for antibiotic-resistant bacteria, with over four million Canadian residents currently relying on private groundwater wells (Statistics Canada, 2024). While groundwater is increasingly acknowledged as a reservoir of ARB, the influence of external drivers affecting their occurrence, such as local weather, anthropogenic activity, and hydrogeological setting, remains limited. The present study sought to characterize current rates and combinations of antibiotic resistance among Escherichia coli isolated from private groundwater wells in southeastern Ontario over a two-year period, and identify key environmental drivers and likely mechanisms associated with the presence of resistant E. coli. Principal component analysis was used to reduce the dimensionality of antibiotic susceptibility profiles and identify dominant combinations of resistance. Binary logistic regression and decision tree analyses were subsequently applied to identify explanatory variables and examine multivariate mechanistic associations with resistance profile principal components. A total of 737 isolates from 260 private well samples were phenotypically tested against a minimum of eight antibiotics, with penicillin-resistant isolates tested against 14 (extended-spectrum screening). Findings showed that 64.5% (n = 475) of isolates exhibited resistance to at least one antibiotic, while 63.1% (n = 164) of well samples contained at least one resistant isolate. Principal component analysis revealed two distinct cycles of resistant E. coli, with resistance patterns linked to likely veterinary sources peaking during fall months (Sept. – Nov.) and human-associated resistance peaking during spring (Mar. – May). Most well samples contained a single dominant resistance pattern (64%), over one third (35%) contained both veterinary and human-associated resistance patterns. Monthly antecedent precipitation and season emerged as key explanatory variables via modelling. Seasonal trends seemed to reflect broader patterns in antimicrobial usage and agricultural activities, including periods of increased human antibiotic prescriptions and seasonal on-farm practices. Study findings may inform future antibiotic resistance surveillance efforts by highlighting seasonal and source-specific patterns of resistant bacteria in private groundwater wells.