The Changing Face of Water: A Dynamic Reflection of Antibiotic Resistance Across Landscapes

Little is known about the role of surface water in the propagation of antibiotic resistance (AR), or the relationship between AR and water quality declines. While healthcare and agricultural sectors are considered the main contributors to AR dissemination, few studies have been conducted in their absence. Using linear models and Bayesian kriging, we evaluate AR among Escherichia coli water isolates collected bimonthly from the Chobe River in Northern Botswana (n = 1997, n = 414 water samples; July 2011–May 2012) in relation to water quality dynamics (E. coli, fecal coliform, and total suspended solids), land use, season, and AR in wildlife and humans within this system. No commercial agricultural or large medical facilities exist within this region. Here, we identify widespread AR in surface water, with land use and season significant predicators of AR levels. Mean AR was significantly higher in the wet season than the dry season (p = 0.003), and highest in the urban landscape (2.15, SD = 0.098) than the protected landscape (1.39, SD = 0.051). In-water E. coli concentrations were significantly positively associated with mean AR in the wet season (p < 0.001) but not in the dry season (p = 0.110), with TSS negatively associated with mean AR across seasons (p = 0.016 and p = 0.029), identifying temporal and spatial relationships between water quality variables and AR. Importantly, when human, water, and wildlife isolates were examined, similar AR profiles were identified (p < 0.001). Our results suggest that direct human inputs are sufficient for extensive dispersal of AR into the environment, with landscape features, season, and water quality variables influencing AR dynamics. Focused and expensive efforts to minimize pollution from agricultural sources, while important, may only provide incremental benefits to the management of AR across complex landscapes. Controlling direct human AR inputs into the environment remains a critical and pressing challenge.


INTRODUCTION
Antibiotic resistance (AR) is recognized as a critical global health emergency, with expected increases in deaths worldwide from diseases that were once treatable (World Health Organization, 2014). Antibiotics are routinely used in human and veterinary medicine to treat or prevent infectious disease. They are also often used in agricultural production systems where they can suppress bacterial diseases as well as promote the growth of livestock and crops. However, decades of misuse and overconsumption of antibiotics have exerted enormous selective pressures on bacterial communities, driving the evolution of resistance (Davies, 2007;Zhang et al., 2009). The wide range of antibiotics and the facility with which bacteria evolve and acquire resistance genes results in AR profiles that can range from resistance to a single antibiotic to multidrug resistance. An increasing diversity of AR bacteria and genes have been observed in both clinical and nonclinical settings, including soil, insects, domestic and wild animals (Allen et al., 2010;Pesapane et al., 2013;Jobbins and Alexander, 2015), and even humans in remote locations (Bartoloni et al., 2009).
Across environments, surface water remains a critical connecting resource. AR bacteria can be transmitted into surface water directly from hospital and municipal wastewater effluents, or indirectly from fecal runoff and infiltration from agricultural and animal husbandry practices (Zhang et al., 2009;Pruden et al., 2012;Rizzo et al., 2013). High levels of AR are routinely found in freshwater systems around urban wastewater and agricultural effluent inflow points (Schwartz et al., 2003;Lima-Bittencourt et al., 2007;Knapp et al., 2012;Pruden et al., 2012;Czekalski et al., 2014;Hsu et al., 2015). Once AR bacteria enter waterways, they may be ingested by animals or humans, or may transfer resistance genes to other bacteria in the environment (Bouki et al., 2013). These processes can occur across complex river flood plain habitats, surface water, sediments and animal communities, but our understanding of these processes is critically limited. While the healthcare and agricultural sectors are often identified as the main contributors to AR bacteria dissemination (Van Den Bogaard and Stobberingh, 2000;World Health Organization, 2001), few studies have been conducted in populated regions where these enterprises are absent, potentially obscuring the role of direct human inputs in AR dissemination. The potential relationship between water quality declines [increases in fecal microbial communities and total suspended solids (TSS)] and AR dynamics is also unknown. Using our study system in Northern Botswana, we evaluate the spatial and temporal dynamics of AR movement in the only permanent surface water in the region, the Chobe River, in relation to season, land use, water quality dynamics, and AR accumulation in associated human and wildlife populations. This human populated system provides an ideal location to evaluate these dynamics as commercial livestock production and large health facilities influences are absent.

Study Site
This study was conducted in the Chobe District in Northern Botswana (Figure 1). The Chobe River is a dryland flood pulse system and one of only three permanent sources of surface water in Botswana. This river system supplies potable water to the resident populations living across nine villages spanning urban to rural landscapes within a biodiversity hot spot. Municipal water is piped directly from the Chobe River at a water plant located downstream from the Chobe National Park (Figure 1). The region experiences pronounced seasonal hydrologic variability with distinct wet (November-March) and dry seasons (April-October) with variable rainfall and periods of drought that appear to occur on a 10-year cycle (Alexander and Blackburn, 2013;Fox and Alexander, 2015). In this system, rainfall and the flood pulse provide the dominant ecological drivers with the peak flood occurring at the beginning of the dry season Fox and Alexander, 2015). These seasonal drivers influence significant biomass fluxes in the river flood plain system, resulting in large densities of wildlife populations congregating at the only permanent surface water resource in the dry season and dispersing in the wet season to access water in ephemeral pans (Alexander et al., 2012). Shared water resources, such as the dryland Chobe River, are under enormous pressure to sustain growing human and wildlife populations.

Water Quality and Microbiological Assessments
Water grab samples (1 L) were collected bimonthly from transect sites (n = 28 sampling points; Figure 1) located at 1 km intervals along the Chobe River between July 2011 and April 2012. Water quality assessments were conducted to determine E. coli (CFU/100 ml), fecal coliform (FC, CFU/100 ml) and TSS (mg/L) concentrations in the Chobe River following methodology previously described in Fox and Alexander (2015). Briefly, water grab samples were collected and processed within 24 h following United States Environmental Protection Agency (USEPA) methods 1103.1 and 1604 for E. coli enumeration (USEPA, 2000;Oshiro, 2002) and method 160.2 for TSS (USEPA, 1983). TSS samples were vacuum filtered (Millipore AP-40; Thermo Fisher) and dried at 103-105 • C for an hour and weighed, and this drying cycle was repeated until a constant weight was obtained. To enumerate and isolate E. coli, water samples were vacuum filtered through sterile gridded nitrocellulose membrane filters (0.45 µm pore size; Thermo Fisher Scientific, Waltham, Massachusetts, USA), then aseptically transferred to RAPID' E.coli2 agar plates (BIORAD, Hercules, California, USA) and incubated at 37 • C for 24 h prior to being enumerated. Up to six putative E. coli colonies were picked from each water sample and grown overnight in tryptic soy broth. DNA was extracted from all samples using the standard boiling method involving detergent/heat lysis followed by ethanol precipitation. DNA samples were screened by PCR FIGURE 1 | The study site is located in Northern Botswana (red map inset) in sub-Saharan Africa, and spans three different land use types (Town, Mixed Use, and Park, shown in red, yellow, and green, respectively). Transect points for water sample collection are noted in red and are located at 1 km intervals along the Chobe River. Tourist lodges (stars) and the water intake point in Town (blue hexagon) are noted.
to confirm presence of the E. coli malB locus, as previously described for water and wildlife samples (Pesapane et al., 2013) and originally described in Candrian et al. (1991). We note that these criteria will generate rare false positives arising from Klebsiella and Enterobacter species, which are ubiquitous in water environments and are detected in mammal feces (Leclerc et al., 2001). Members of both genera are positive for betagalactosidase, thus test positive for lactose fermentation on MacConkey agar, like E. coli. As well, some isolates from each genus are positive for beta-glucoronidase (Maheux et al., 2015), allowing them to present like E. coli on differential colorimetric assays such as RAPID' E.coli2 agar. Both Klebsiella and Enterobacter encode the malB locus, though Enterobacter differs sufficiently at the DNA sequence level that stringent PCR conditions are unlikely to yield false positive results. A recent study of AR in surface water identified highly multidrug resistant isolates of Klebsiella and Enterobacter, but at significantly lower numbers than E. coli (Zarfel et al., 2017), confirming that Klebsiella and Enterobacter can contribute to the environmental pool of AR genes, but their presence in surface water is generally unlinked to fecal contamination and these genera are unlikely to pose health risks to humans and other animals (Leclerc et al., 2001).

Statistical Analysis
Our study site is characterized by diverse land use types, which may influence AR dynamics. In this study, land type was therefore broken down and defined as "Park" for transects within the boundaries of the Chobe National Park, "Town" for transects occurring within the urban center (Kasane), or "Mixed Use" for transects in the growing, peri-urban area of Kazungula that incorporates land used for farming, wildlife corridors, and human settlement. All water-sampling observations were recorded between July 2011 and April 2012 at 15-day intervals. Based on rainfall during our study period, we classified the dry season as 13th July 2011 to 28th November 2011 and the wet season as 16th December 2011 to 25th April 2012. Statistical analyses were carried out using R 3.1.3 (R Core Team, 2014).  Frontiers in Microbiology | www.frontiersin.org

AR, Water Quality Dynamics, and Environmental Predictors
To investigate the association between AR in E. coli samples and environmental predictors, including land use type and season in the sampling location, we used a combination of linear models. Ordinary least squares (OLS) linear regression was used to estimate the effect of environmental predictors on AR isolates in a given sample ( Table 2). Separate OLS regression models were fit for wet and dry season data. Model residuals were assessed for linearity and normal error distribution using quantile-quantile plots, and Levene's test for equal variance, and a natural-log transformation was applied to mean AR and water quality perimeters (E. coli and TSS) to better approximate a normal distribution prior to fitting the regression models. FC was not included in the analysis because it strongly covaried with E. coli. We assessed for spatial autocorrelation (SAC) in model residuals using the Global Moran's I statistic with rowstandardized weights, which was not significant for either the dry or wet season indicating SAC was not influential in the model residuals.
Additional linear regressions were run to determine whether similar trends held when considering the mean number of antibiotics to which an isolate was resistant (ranging from 0 to 10). A generalized linear model (GLM) was used to determine the effect of environmental predictors on the outcome of AR for each isolate. Once non-significant predictors were eliminated, a linear mixed model was used to account for the potential correlation between isolates within a given water sample through the assignment of random intercepts associated with each of the 414 water samples used in the regression.
Spatiotemporal patterns of mean AR and water quality perimeters (E. coli, FC and TSS) were interpolated and mapped using empirical Bayesian kriging (EBK) in ESRI ArcGIS v.10.2 (Environmental Systems Research Institute, Redlands, California). EBK makes predictions through automated subsetting and simulation using a restricted maximum likelihood approach to obtain an optimal empirical semivariogram and calculate weights such that the interpolation mean square error is minimized (Pilz and Spöck, 2008). Each iteration of EBK uses semivariogram models defined by nearby values, rather than being influenced by more distant points, allowing for more accurate predictions even when the data are moderately non-stationary (Krivoruchko, 2012). A standard search radius of 10,000 m was used to ensure a minimum of 10 and a maximum of 15 neighbors were included in the interpolation. Model suitability was assessed based upon the criteria that root-meansquare error (RMSE) was minimized and the root-mean square standardized error was close to one ( Table 3). The resulting maps show predicted mean AR (Figure 3), and the concentrations of E. coli, FC and TSS (Figure 4) in water samples collected from a given location. A heat map of AR, E. coli and TSS isolated from water samples is shown by sampling date and transect ( Figure 5) to illustrate location and timing of water quality declines and AR emergence.

AR Profiles Across Sample Type and Land Use
In order to determine relationships between AR in isolates from surface water and AR isolates found in humans and wildlife sharing the same water source, we conducted a Spearman rank correlation in R. Spearman rank correlation coefficients were calculated by comparing AR in water samples (n = 1997) to AR in E. coli isolates from fecal samples collected from humans (n = 113), non-water-associated animals (n = 382), and waterassociated animals [n = 58, previously published data, see   (Jobbins and Alexander, 2015)]. The relationships between AR profiles by sample type and land use are visualized in Figure 6.

Effects of Land Use and Season on Surface Water AR Dynamics
To explore AR dynamics, we examined the distribution and frequency of resistance phenotypes in E. coli in our sample area over a 10-month period, spanning wet and dry seasons. By land use, the mean level of resistance (number of antibiotics to which an isolate demonstrated resistance) was 1.39 (SD = 0.051) in the Park, 2.15 (SD = 0.098) in Town, and 1.76 (SD = 0.079) in the Mixed land use areas. When examining differences in mean AR, ordinary linear models identified both land use and season as being statistically significant predictors, with the wet season resulting in higher mean AR than the dry season (coeff = 0.215; t = 2.938; p = 0.003). Isolates from the Park exhibited lower levels of AR than Mixed Use (coeff = −0.430; t = −5.184; p < 0.001) and Town, which exhibited the highest levels of AR (coeff = 0.322; t = 3.130; p = 0.002; Figure 2). A linear mixed model was then used to evaluate these relationships accounting for the expected correlation among isolates (n = 1997) distributed among water samples (n = 414). Similar to the results of the OLS, Analysis of Variance (ANOVA) of the most parsimonious model revealed significance of both land use type (F = 18.633) and season (F = 4.144) in the mean number of antibiotics to which an isolate exhibited resistance.

Spatial Patterns of AR, E. coli, and TSS
Using empirical Bayesian kriging, we evaluated the spatiotemporal patterns of AR, E. coli, FC and TSS (Figure 3). A summary of root mean square error (RMSE) associated with semiovariogram models estimated using empirical Bayesian kriging (EBK) are shown in Table 3. The RMSE are the cross validation results for wet and dry season semivariogram models used in EBK predictions. Lower model RMSE and a RMSE standardized close to one indicate improved model performance. A RMSE standardized greater than one indicates underestimation of the variability in our predictions, while < 1 indicates overestimation of the variability.
The spatial and temporal trends of AR, E. coli, FC and TSS were plotted to visualize how concentrations change along the river (Figures 3, 4 (spatial) and Figure 5 (temporal and spatial)). Maps for both wet and dry season mean AR show how bacterial isolates in water grab samples collected from the river in the town of Kasane had the highest resistance, with significantly lower mean AR observed in the Park and Mixed land use (Figure 3). In the wet season, the region of high AR extended further upstream into the Park compared to in the dry season. Numerous storm water drains are also located in the region of elevated AR, but it is unclear how these structures contribute to AR distribution in surface water. Evaluating the spatiotemporal patterns in E. coli, FC and TSS from water quality grab samples showed that the area of the river in the Chobe National Park just upstream of Town transects had the highest concentrations during both the wet and the dry seasons (Figure 4). During the wet season, water samples from transects in Mixed land use also had higher predicted concentrations of E. coli, FC, and TSS.

Comparison of AR Profiles in Water, Animals and Humans Across Land Use
Mean AR in isolates from human fecal samples (n = 113) was compared to mean AR in E. coli isolates from fecal samples collected from non-water-associated animals (n = 382), waterassociated animals [n = 58, previously published data, see (Jobbins and Alexander, 2015)], and water samples (n = 1997). A similar rank order of resistance to antibiotics was observed across sample types and land use (Figure 6), with high Spearman rank correlation coefficients between resistance in water across land uses (R = 1, p < 0.0001) indicating that the river is well-mixed and relatively homogeneous along this stretch, and between human isolates and water isolates (R = 0.8, p < 0.001). Antibiotic resistance in water-associated animals demonstrated the strongest positive association with water AR profiles in the Park (R = 0.88, p < 0.0001), followed by Mixed (R = 0.82, p < 0.001) then Town (R = 0.7, p < 0.05). Non-water-associated wildlife AR values in the Town and Park land areas were highly correlated with that observed in water (R = 0.96 and 0.94, p < 0.001), while non-water-associated wildlife in the Mixed land type was not associated with water AR profiles (R = 0.57, p < 0.05).

DISCUSSION
In this study, we identify widespread movement of AR in surface water across land use in the absence of commercial livestock production systems and large medical facilities. This suggests that direct human inputs together with other environmental factors may be sufficient for the extensive dispersal of AR observed in this ecosystem across protected and unprotected landscapes. Season and land use were both important predictors of mean AR in surface water with greater AR levels observed in the wet season than in the dry season (p = 0.003). Our results suggest FIGURE 2 | Mean water Escherichia coli isolates exhibiting antibiotic resistance (AR), corresponding total E. coli counts and number of E. coli isolates obtained during water quality assessments within each land use type and sampling period.  that even in the absence of agricultural inputs, human source AR movement may be sufficient to create widespread AR dispersal across surface water even in protected areas upstream of urban centers.

Urban and Peri-Urban Landscapes
The urban landscape (Town) had the highest level of AR E. coli across land types (Figure 2, µ = 2.145, SD = 0.098). The area with the highest levels of AR was located along a segment of the river in Town associated with a peak density of low-income residential properties (264 residences/ha). In this area, an estimated 69% of households use traditional, selfbuilt pit latrines, instead of flush toilets (Alexander and Godrej, 2015). Pit latrines have been previously linked to water quality declines through contamination of groundwater (Jacks et al., 1999;Ahaneku and Adeoye, 2014). Our work suggests that these impacts may be even greater in areas where health-seeking behaviors lead to the use of pit latrines as wastewater receptacles to avoid contaminating residential living areas. Disposal of greywater into the pit latrine, particularly where the ratio of households to pit latrine is high, resulting in hydraulic overloading of the pit latrine may lead to the leaching of human fecal waste and AR bacteria (commensal and pathogenic) into the surrounding soil and groundwater with subsurface movement into nearby surface water resources (Alexander and Godrej, 2015). In the wet season, rising water tables may inundate previously unsaturated soils and further flush AR bacteria from environmental reservoirs, resulting in the elevated levels of AR E. coli seen in the Chobe River during this time period. Spatial patterns of AR concentration also appeared to be associated with storm water drain locations suggesting that the flushing of these drains during heavy rainfall may also contribute to water quality declines observed during the wet season (Hsu et al., 2015) (Figures 3, 4).
is reminiscent of pit systems and lagoons in animal agriculture, whereby fecal indicator bacteria, AR bacteria, and ARGs leach or flow into groundwater and river systems (Chee-Sanford et al., 2001;Sapkota et al., 2007;Li et al., 2014). In these situations, AR bacteria have been discovered in high concentrations in rural groundwater (60% of coliforms tested), with up to 33% of isolates resistant to five or more antibiotics (Mckeon et al., 1995). Movement through ground water can be significant with, for example, ARGs from a swine production facility using pit systems moving in ground water more than 250 m from the point of origin (Chee-Sanford et al., 2001).
The Mixed Use landscape had lower AR E. coli levels than in the urban land use, despite being located downstream. These differences may be due to a number of factors. Firstly, the number of low-income residential properties is significantly less (100 residences/ha) than in Town (Alexander and Godrej, 2015), and these residences are located on elevated land, farther away from the river. The groundwater table is much lower than in Town relative to the region of more concentrated residential areas. Secondly, the use of pit latrines to discard graywater is less prevalent in this region than in Town (Alexander and Godrej, 2015), potentially reducing the leaching of AR bacteria. The presence of the wastewater treatment plant in the Mixed Use land area may also contribute to the overall levels of AR, E. coli, and FC in the Chobe River in this region through leaching and run-off. Although levels were lower than in the urban land use, the concentrations of E. coli, FC and TSS in the mixed land use increased significantly during the wet season.

AR in Protected Areas Upstream of Human Populations
Direct human influences do not explain the levels of AR seen in the Park, which is located upstream from known sources of human fecal waste. The Park land use is characterized by extensive floodplains where high E. coli, FC and TSS were identified (Figure 4). We previously found a strong association between floodplain habitats and high E. coli levels in the river, with water quality declines significantly associated with wildlife fecal counts (Fox and Alexander, 2015). Significant levels of AR in the wildlife population has previously been identified (Pesapane et al., 2013;Jobbins and Alexander, 2015) with AR levels in E. coli significantly higher in synanthropic wildlife (i.e., banded mongoose) carnivores (i.e., leopard), and waterassociated species (i.e., hippopotamus and otter) (Jobbins and Alexander, 2015). These data suggest life history strategy may be key to understanding exposure and transmission dynamics (Jobbins and Alexander, 2015).
The flood pulse may also play a central role in AR dynamics. Slow-moving floodwaters travel thousands of kilometers along the Zambezi and Cuando/Chobe Rivers, arriving at the start of the dry season and eventually backing up behind the Mambova fault line, which lies several meters higher than the surrounding floodplain. In some years, floodwaters can inundate a massive area covering several thousand square kilometers, maintaining high water levels well into the dry season, when all other natural surface water sources have evaporated. Flooding may result in the mobilization of AR bacteria out of existing biofilms in floodplain sediments and groundwater reservoirs into the water column, resulting in the significant spike in AR E. coli levels during the wet season. At freshwater beaches, fecal coliforms (enterococci) and E. coli abundances in the top 20 cm of wet sand have been known to exceed the amount in the water column by a factor of 38 (Wheeler Alm et al., 2003). Furthermore, AR elements have been discovered in floodplain sediment down to a depth greater than 80 cm, indicating that these elements may persist for 30-40 years (Tamtam et al., 2011). The spike in waterborne AR E. coli levels observed in October coincides with minimum annual water levels in the Chobe River system and may be related to concentration of AR bacteria as water levels drop. Another potential contributing factor to increased AR levels observed at the end of the dry season may be the significant increase in wildlife densities that occur as the dry season advances and water resources dry up in the interior. Wildlife access to water is limited due to developments along the riverfront, concentrating wildlife use into smaller areas focused in the National Park with consequent increases in fecal inputs and water quality declines in the segment upriver of the water intake for the population (Fox and Alexander, 2015).

Water Quality and AR Dynamics
Water quality variables and surface water AR are significantly influenced by season and land use. In our study system, water quality declines are known to be influenced by wildlife distribution in the dry season, and rainfall and river hydrology in the wet season (Fox and Alexander, 2015). Here, in-river E. coli concentrations had a significant positive association with mean AR in the wet season but not in the dry season ( Table 2). Increases in E. coli are known to be associated with rainfall and river height (Alexander et al., 2018), with overland movement and runoff of fecal waste likely contributing to water quality declines (increases in E. coli and FC) and AR levels in the wet season. Across all land uses and both seasons, Town exhibited the best water quality perimeters overall (lowest E. coli, FC and TSS concentrations; Figure 4), yet the highest mean AR levels (Figures 2, 3). TSS was negatively associated with mean AR in both the wet and dry seasons ( Table 2). To our knowledge this is the first study integrating water quality studies and AR across landscapes. Our data suggests important complexity exists signaling the significance of human-environmental couplings on AR movement and accumulation.

Human Use of the River and AR Circulation
Classified as a "biodiversity hotspot, " the abundance of wildlife in the Chobe region attracts more than 20,000 tourists annually. The concentrated recreational use of the Chobe River in the Chobe National Park and high density of boats may contribute to the spread of AR into this protected area. This boat activity, combined with low water levels in the dry season, may result in the increased re-suspension of river sediment reservoirs of AR E. coli. In addition, human effluent from boats entering the park may enter the system if the effluent is not disposed of properly or if existing holding tanks over flow, potentially providing a mechanism for additional AR inputs into the system. Numerous studies have shown that tourists can aid in the contraction and distribution of pathogens and AR microbes worldwide (reviewed in Okeke and Edelman, 2001). Indeed, AR E. coli isolates were found in fecal samples from tourists visiting Mexico without ever having consumed prophylactic antibiotics prior to traveling (Murray et al., 1990).
Antibiotic Profiles Across the Human, Wildlife, and Environmental Interface AR profiles of E. coli from surface water, humans and wildlife populations from the Chobe region were highly similar ( Figure 5). These data indicate the need for caution in ascribing AR threats as arising dominantly from agricultural inputs (House, 2015) and highlights the need to better understand the manner in which direct human-environmental couplings may contribute to resistance dissemination across the environment. Previous work utilizing whole-genome sequencing has revealed that there is extensive sharing of AR genes between human, animal and environmental microbial communities (Pehrsson et al., 2016). This sharing between bacteria and microbial communities is assisted by horizontal gene transfer via mobile genetic elements and AR clusters (Pehrsson et al., 2016), increasing the likelihood that these elements may reach pathogenic organisms.
The ingestion of surface water harboring AR microbes has been identified as a likely route of AR movement in naïve wildlife populations (Mariano et al., 2009;Jobbins and Alexander, 2015). In South Africa, impala drinking from rivers containing AR were shown to be 19 times more likely to have AR microbes than impala drinking from uncontaminated rivers (Mariano et al., 2009). However, in our study site, impala, and other water-dependent herbivores (buffalo, kudu, etc.) showed little to no AR, signifying that ingesting water alone was insufficient for transmission of resistance. Instead, waterassociated wildlife (hippopotamus, otters, etc.) had higher levels of AR resistance, indicating that the duration spent in the water, and/or consumption of water-dwelling vegetation or associated sediments may also play a role (Jobbins and Alexander, 2015).
The dissemination of AR in the environment is a key component of the growing global health threat, however, surveillance of environmental antimicrobial hot spots is absent from action plans. Perennial surface water affords ecosystem level surveillance, allowing early detection and control of AR before spreading into clinically important pathogens. Indeed, water quality declines represent the penultimate expression of system degradation. It is clear from our study that surveillance focused on ARG in human and animal pathogens may have limited effectiveness. This surveillance approach simply alerts us to the ultimate failure in controlling AR among pathogens that threaten human health.

CONCLUSION
Our study provides clear indication that humans have the potential to directly influence the movement and accumulation of AR across protected and unprotected landscapes in the absence of commercial livestock production and extensive health facilities. Focused and expensive efforts to minimize pollution from agricultural sources, while important, may only have incremental benefits to the management of antimicrobial resistance across complex landscapes. AR control efforts might benefit from sentinel surveillance systems that identify ARG movement and accumulation in the environment before invasion into pathogenic microbial communities. Controlling direct human AR input into the environment remains a critical challenge.

AUTHOR CONTRIBUTIONS
CS conducted the laboratory work, interpreted the data, and drafted and revised the manuscript. ED conducted the statistical analyses, aided in data interpretation and wrote the method and results sections corresponding to the statistical analyses. JF conducted the spatial analysis and wrote the corresponding sections in the manuscript. AC interpreted the data and assisted in manuscript revision. KA developed the study concept and design, interpreted the data and conducted critical revision of the manuscript, and secured funding for the study.

FUNDING
Funding for this study was provided by the National Science Foundation's Dynamics of Coupled Natural and Human Systems Award, CNH #1518486.