Methodological differences between studies confound one-size-fits-all approaches to managing surface waterways for food and water safety

Abstract

Even though differences in methodology (e.g., sample volume and detection method) have been shown to affect observed microbial water quality, multiple sampling and laboratory protocols continue to be used for water quality monitoring. Research is needed to determine how these differences impact the comparability of findings to generate best management practices and the ability to perform meta-analyses. This study addresses this knowledge gap by compiling and analyzing a data set representing 2,429,990 unique data points on at least one microbial water quality target (e.g., Salmonella presence and Escherichia coli concentration). Variance partitioning analysis was used to quantify the variance in likelihood of detecting each pathogenic target that was uniquely and jointly attributable to non-methodological versus methodological factors. The strength of the association between microbial water quality and select methodological and non-methodological factors was quantified using conditional forest and regression analysis. Fecal indicator bacteria concentrations were more strongly associated with non-methodological factors than methodological factors based on conditional forest analysis. Variance partitioning analysis could not disentangle non-methodological and methodological signals for pathogenic Escherichia coli, Salmonella, and Listeria. This suggests our current perceptions of foodborne pathogen ecology in water systems are confounded by methodological differences between studies. For example, 31% of total variance in likelihood of Salmonella detection was explained by methodological and/or non-methodological factors, 18% was jointly attributable to both methodological and non-methodological factors. Only 13% of total variance was uniquely attributable to non-methodological factors for Salmonella, highlighting the need for standardization of methods for microbiological water quality testing for comparison across studies.IMPORTANCEThe microbial ecology of water is already complex, without the added complications of methodological differences between studies. This study highlights the difficulty in comparing water quality data from projects that used different sampling or laboratory methods. These findings have direct implications for end users as there is no clear way to generalize findings in order to characterize broad-scale ecological phenomenon and develop science-based guidance. To best support development of risk assessments and guidance for monitoring and managing waters, data collection and methods need to be standardized across studies. A minimum set of data attributes that all studies should collect and report in a standardized way is needed. Given the diversity of methods used within applied and environmental microbiology, similar studies are needed for other microbiology subfields to ensure that guidance and policy are based on a robust interpretation of the literature.

Keywords: Listeria; Salmonella; methods comparison; produce safety; shiga toxin Escherichia coli; water quality.

Fig 1

Schematic representation showing data exclusion due to data quality and compatibility issues. *Numerous samples were tested for more than one microbial target. GPS, Global Positioning System.

Fig 2

Location of the 100,410 unique sampling site locations represented by the data set compiled here. GPS coordinates were modified slightly to ensure confidentiality. (The maps were created in R using the ggplot2 and sf packages.)

Fig 3

Variance in the likelihood of detecting (A) Salmonella, (B) Listeria monocytogenes, (C) and pathogenic E. coli that is jointly versus uniquely attributable to non-methodological (e.g., sampling site, season, water type, waterway, and year) and methodological (e.g., culture versus molecular-based detection, sample type, and volume) matrices.

Fig 4

Results of conditional forest analysis that identified methodological and spatiotemporal factors associated with detection of (A) Salmonella, (B) L. monocytogenes and (C) pathogenic E. coli in water. The outcome of these forests was the residuals of a regression analysis that modeled likelihood of target pathogen detection as a function of two nested random effects (site and waterway). The y-axis shows the features ranked from highest to lowest variable importance. Variable importance is a unitless relative measure; thus, the importance of one variable should only be compared to another variable in the same plot, not between.

Fig 5

Impact of sample volume on probability of detection of (A) Salmonella, (B) L. monocytogenes, and (C) pathogenic E. coli according to generalized linear mixed models implemented with fixed effects of sample volume and season and random effects of site nested in waterway nested in state. No grab samples were tested for pathogen data in volumes greater than 10 L, and Moore swab volume was set to 10 L.