Next Article in Journal
Oleic Acid and Palmitic Acid from Bacteroides thetaiotaomicron and Lactobacillus johnsonii Exhibit Anti-Inflammatory and Antifungal Properties
Next Article in Special Issue
Mathematical Models for Cholera Dynamics—A Review
Previous Article in Journal
The Metabolic Potential of Endophytic Actinobacteria Associated with Medicinal Plant Thymus roseus as a Plant-Growth Stimulator
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Microorganisms
Volume 10
Issue 9
10.3390/microorganisms10091804
microorganisms-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Sources and Drivers of ARGs in Urban Streams in Atlanta, Georgia, USA

by
Robert A. Sowah
1,2,
Marirosa Molina
3,*,
Ourania Georgacopoulos
4,
Blake Snyder
2 and
Mike Cyterski
5
1
Oak Ridge Institute for Science and Education, Oak Ridge, TN 37831, USA
2
U.S. EPA, Laboratory of Services & Applied Sciences Division, 980 College Station Rd., Athens, GA 30605, USA
3
U.S. EPA, Office of Research and Development, Center for Environmental Measurement and Modeling, 109 T. W. Alexander Drive, Durham, NC 27709, USA
4
Student Services Contractor to the U.S. EPA, Office of Research and Development, Center for Environmental Measurement and Modeling, 960 College Station Rd., Athens, GA 30605, USA
5
U.S. EPA, Office of Research and Development, Center for Environmental Measurement and Modeling, 960 College Station Rd., Athens, GA 30605, USA
*
Author to whom correspondence should be addressed.
Microorganisms 2022, 10(9), 1804; https://0-doi-org.brum.beds.ac.uk/10.3390/microorganisms10091804
Submission received: 12 July 2022 / Revised: 1 September 2022 / Accepted: 1 September 2022 / Published: 8 September 2022
(This article belongs to the Special Issue Foodborne and Waterborne Pathogens)
Browse Figures
Review Reports Versions Notes

Abstract

:
The spread of antibiotic resistance genes (ARGs) in the aquatic environment is an emerging concern in the interest of protecting public health. Stemming the environmental dissemination of ARGs will require a better understanding of the sources and drivers of ARGs in the water environment. In this study, we used direct measurement of sewage-associated molecular markers, the class 1 integron gene, standard water quality parameters, and watershed characteristics to evaluate the sources and drivers of ARGs in an urban watershed impacted by a gradient of human activities. Quantitative polymerase chain reaction (qPCR) was used to quantify the abundance of the sewage-associated HF183, the E. coli fecal indicator, class 1 integron gene (int1), and the ARGs sulI, sulII, tetW, tetM, ampC, and blaSHV in stream water samples collected from the Proctor Creek watershed in Atlanta, Georgia. Our findings show that ARGs were widely distributed, with detection frequencies of 96% (sulI and sulII), 82% (tetW and tetM), and 49% (ampC and blaSHV). All the ARGs were positively and significantly correlated (r > 0.5) with the HF183 and E. coli markers. Non-linear machine learning models developed using generalized boosting show that more than 70% of the variation in ARG loads in the watershed could be explained by fecal source loading, with other factors such as class 1 integron, which is associated with acquired antibiotic resistance, and environmental factors contributing < 30% to ARG variation. These results suggest that input from fecal sources is a more critical driver of ARG dissemination than environmental stressors or horizontal gene transfer in aquatic environments highly impacted by anthropogenic pollution. Finally, our results provide local watershed managers and stakeholders with information to mitigate the burden of ARGs and fecal bacteria in urban streams.

1. Introduction

Antibiotic resistance genes (ARGs) and genetic determinants are increasingly being recognized as contaminants in the aquatic environment [1,2,3,4,5]. The U.S. Environmental Protection Agency (U.S. EPA), in its 5-year review of the 2012 recreational water quality criteria, cites ARGs and antimicrobial resistant bacteria (ARB) as emerging concerns in the interest of protecting public health [6]. The presence of ARGs and ARB in the aquatic environment is gaining attention as a critical water quality and public health concern due to research linking environmental and clinical resistance [3,7]. It has also been suggested in recent studies that the rise in community-acquired ARB infections can be attributed to the emergence and spread of ARGs and ARB in the environment [8,9,10,11]. The development of methods and tools to understand the contribution of the environment to the spread of ARGs and ARB is a critical step to addressing the global antibiotic resistance menace [12,13] and supports the One Health approach, which recognizes the connection between the health of people, animals, and the environment.
Although some ARGs occur naturally in the environment, the rapid evolution and spread of new ARGs have been attributed to human activities including the widespread use or misuse of antibiotics for human and animal purposes [14,15,16]. In general, three main factors are recognized for driving the spread of ARGs in the aquatic environment. These include wastewater/sewage discharges from human or animal sources, selection pressure exerted by antibiotics, metals, and other contaminants in the water environment, and horizontal gene transfer between microorganisms in the environment. Much work has been done to characterize ARG loads at the source level, however, questions remain about the contributions of these sources/drivers to ARG loads in each system.
The role of the aquatic environment in the spread of ARGs and ARB cannot be overemphasized, as the water environment provides a pathway for contact with resistance determinants and pathogens through recreational water use, the consumption of crop products irrigated with contaminated water, and the consumption of meat products from animals that have come into contact with contaminated surface waters [16,17,18]. Urban streams, by nature of the impact of anthropogenic activities, are subject to multiple stressors that can promote the emergence and spread of resistance determinants in the environment. Traditionally, urban waters have been known to be impacted by nutrients, pharmaceuticals, metals, and pathogens through multiple stressors such as wastewater discharges, stormwater run-off from large areas of impervious cover, and old and/or failing sewage infrastructure that often results in sewage overflows to the aquatic environment. The combined effect of these anthropogenic sources has been known to increase the loading of ARGs in urban streams, above what is naturally observed in the environment [5,15,19]. Accordingly, Sanderson et al. [1] opined that these polluted waters have become reservoirs of ARGs and a conduit for the dissemination of antibiotic resistance genes and determinants to new environments. The aquatic environment, moreover, provides an interface for the exchange of ARGs between pathogens and environmental bacteria, which can lead to the development of new resistance mechanisms [20].
Recent studies have shown that the direct measurement of sewage-associated microbial molecular markers can be used to assess the sources and drivers of ARGs in the aquatic environment. Ahmed et al. [21] demonstrated the efficacy of this approach using the sewage-associated HF183 Bacteroidales marker and the crAssphage bacteriophage marker to identify the sources of ARGs in storm drain outfalls off the Florida coast. Metadata analysis by Karkman et al. [22] used the crAssphage marker to understand the connection between fecal pollution and ARG patterns in sewage-polluted environments. The authors observed that, in sewage-polluted environments, the presence of ARGs is largely explained by fecal pollution, whilst the effect of factors such as selective pressure was only evident in environments polluted by very high levels of antibiotics from manufacturing.
Although these studies provide insight into the dynamics of ARGs in the aquatic environment, the narrow scope of these studies, in terms of the sources considered (storm drain outfalls) or the indicators used, do not allow for a detailed analysis of sources and drivers of ARG abundance and spread in the environment. This is particularly important in urban watersheds such as the Proctor Creek watershed in Atlanta, GA, USA, which has a long history of water quality impairment from multiple sources. Anthropogenic impacts in this watershed stem from sanitary sewer overflows, industrial activities, and runoff from highly developed urban centers and associated impervious areas [23,24,25]. A study by the U.S. EPA showed that despite the progress made in the watershed, including the decommissioning of combined sewer systems, the Proctor Creek watershed still suffers from elevated concentrations of fecal contaminants and nutrients in streams [6]. The goal of this study was to demonstrate a watershed-level approach to identifying the sources and drivers of ARGs in the Proctor Creek watershed. In this study, we investigated the role of human fecal pollution, water quality parameters, and watershed characteristics in the abundance and spread of ARGs in urban streams. Predictive models were developed to quantify the influence of these factors on ARG concentrations in the aquatic environment.

2. Materials and Methods

2.1. Study Area

The study was conducted in the Proctor Creek watershed located within the City of Atlanta, GA, USA (Figure 1). The Proctor Creek watershed drains the west side of the city of Atlanta, its headwaters starting near the city center and flowing northwest for approximately 15 km into the Chattahoochee River (a major water body for recreation, agricultural, and fishing purposes in the tri-states of Georgia, Florida, and Alabama). The watershed drains an area of ~41 km2. The area around the headwaters is highly impervious, with most of the streams either piped underground or channelized aboveground [6]. The headwaters receive urban runoff from the downtown Atlanta metropolitan area and the rest of the watershed is impacted by multiple types of point and nonpoint sources of pollution including combined sewer overflow (CSO) facilities, railroads, a large freight yard along the northern boundary of the watershed, landfills, automotive salvage yards, and illegal trash dumps [6].

2.2. Sample Collection

Grab surface water samples were collected in 1 L sterile Nalgene bottles on a bi-weekly basis from 12 sampling locations distributed across the Proctor Creek watershed. The sampling points consist of 6 sites that are located on the mainstem of the creek and 6 tributary sites (Figure 1). Samples were kept on ice during transport to the U.S. EPA’s Office of Research and Development (ORD) laboratory in Athens, GA, USA and processed within 6 h of sample collection following guidelines in the National Field Manual for the Collection of Water Quality Data [26]. Samples (n = 287) were collected over a 1-year period from May 2016 to April 2017. During sampling, in situ water quality parameters, including dissolved oxygen (DO), specific conductance (SPC), pH, and temperature, were measured using a multiparameter probe (YSI, Yellow Springs, OH, USA).

2.3. Water Quality Analysis

Water samples were analyzed for culturable Escherichia coli (E. coli) using the Colilert-defined substrate method (IDEXX Laboratories Inc., Westbrook, ME, USA). Samples were diluted 10-fold and 100-fold in 100 mL volumes and run in duplicate to obtain average E. coli concentrations in MPN/100 mL. Additionally, turbidity measurements were performed on water samples in the laboratory using a Turbidimeter (LaMotte Company, Chestertown, MD, USA).
Fifty-milliliter samples were filtered in quadruplicate on 0.45 µm polycarbonate filter membranes (Millipore Sigma, Burlington, MA, USA) to concentrate microbial DNA for molecular analysis. Filters with microbial residue were placed into 2 mL microcentrifuge tubes and stored in a freezer at −80 °C until further processing. Two filters were subsequently extracted using the DNeasy PowerLyzer PowerSoil kit (Qiagen, Germantown, MD, USA) and recommended protocol with some modifications (see supplemental material for details). The extracted DNA was then stored at −80 °C until further analysis. The two remaining filters were kept in storage as a backup. The extracted DNA was analyzed for fecal indicator markers EPA EC 23S and HF183/BacR287, which will be referred to as E. coli marker and HF183, respectively, going forward. Other genetic markers that were analyzed include the 16S rRNA, which measures microbial community abundance; class 1 integron gene (int1), which is associated with acquired antibiotic resistance; and the sulfonamide resistance genes (sulI, sulII), tetracycline resistance genes (tetW, tetM), and beta-lactam resistance genes (ampC and blaSHV). Among all the ARGs monitored in the aquatic environment in previous studies, the sulfonamide and tetracycline resistance genes are the most abundant and widely distributed, followed by beta-lactam genes [27]. All the quantitative polymerase chain reaction (qPCR) assays were run on a QuantStudio 3 instrument. All qPCR assays in this study followed reaction conditions described in previous studies unless stated otherwise (see supplemental material for reaction conditions). Taqman qPCR assays were run for all markers except for the sulI, ampC, and blaSHV markers which were run using SYBR Green chemistry. The Taqman Universal Master Mix (2X) was used in Taqman assays whilst the PowerUp SYBR Green Master Mix (2X) was used for SYBR assays in a final reaction volume of 20 µL. Marker quantification was performed using standard curves that were generated during each qPCR run. The supplemental material provides additional details on the qPCR procedures and the quality control protocols implemented.

2.4. Data Analysis

The qPCR and culturable E. coli data were log-transformed for statistical analysis. Samples that were non-detectable for culturable E. coli were given a log-transformed value of 0 (i.e., the detection limit of 1 MPN/100 mL). Molecular markers that were below the lower limit of quantification (LLOQ), and thus undetected in this study, were imputed using regression on order statistics (ROS), as described in Sowah et al. [28]. The LLOQ for qPCR assays was determined as the lowest copy number of the qPCR standards that was accurately measured in all assays [29,30,31,32]. qPCR data in gene copies (GC) per ml were normalized with the copies of the 16S rRNA gene to yield the relative abundance of genetic markers in samples. All data analysis was performed in the R program [33] with significance determined at the 5% significance level. The relationships between ARGs and fecal indicators, integrase gene, standard water quality parameters, and antecedent rainfall were examined using the non-parametric Spearman correlation analysis. Principal component analysis (PCA) was performed to create an index score variable that is an optimally weighted combination of the six ARGs (sulI, sulII, tetM, tetW, ampC, and blaSHV) measured in this study. The rationale for this approach was to reduce the number of response variables (ARGs) into one or two principal components that explain most of the variability in the data and can be used as response variables in predictive modeling. The prcomp function in the VEGAN package in R was used for PCA. The first two principal components (referred to as PC1 and PC2) were used as the response variables in our predictive models.

2.4.1. Generalized Boosted Models

We used a Generalized Boosted Model (GBM), a machine learning regression/classification method based on constructing a chain of decision trees [34] to determine which independent variables were most influential in predicting PC1 and PC2 scores. The independent variables evaluated include int1, 16S rRNA, culturable E. coli, E. coli marker, HF183, SPC, DO, pH, water temperature, turbidity, antecedent rainfall (24 h, 48 h, and 72 h), and site location. We developed models for PC1 and PC2 separately using the gbm package (version 2.1.5, 11) in R. There is some randomness in the results of the GBM algorithm when fitting a given dataset, so we developed bootstrapped estimates of model performance measures by fitting 500 models for PC1 and PC2. In each of the 500 bootstrap iterations, the dataset was randomly split: 80% of the dataset was randomly placed into a training set, and the remaining 20% was put into a testing set. The gbm algorithm uses multiple hyperparameters that influence the fitting process and final solution. Hyperparameter values were set as follows: Error term = Gaussian, Maximum number of trees = 10,000, Shrinkage = 0.005, interaction depth = 3, Bag fraction = 0.5, Train fraction = 1, Number of minimum observations per node = 5, and cross-validation folds = 10. See the supplemental material for a discussion of hyperparameter selection methods.

2.4.2. Partial Dependence Plots

The gbm package in R allows for the creation of Partial Dependence Plots (PDPs), which graphically summarize the relationship between a single covariate and its influence on the response variable. One option within the gbm package will allow for maximum flexibility in the PDP, permitting it to increase or decrease amidst multiple inflection points over the range of covariate values. The modeler can, however, constrain the covariate/response relationship (via a parameter called “var.monotone”) to be monotonically increasing or decreasing, resulting in easier-to-interpret PDPs, but possibly also affecting the model’s predictive performance. To constrain model complexity (i.e., limit the number of significant covariates), we modeled PC1 using two stages: first, we fit 500 models to PC1 using all the predictors; second, we fit 500 additional simplified models using only those covariates with >3% average influence from the initial stage. Next, we examined the stage-two PDPs to see if their relationship could be adequately described/captured using a monotonically increasing/decreasing function. Finally, a third set of 500 models was run using the “var.monotone” parameter to constrain the covariate/response relationship for certain covariates. We call this the “Refined” PC1 model, and we created aggregate PDPs for each covariate in the Refined PC1 model by using the partial dependence points for each of the 500 model runs from the Refined models. The 2.5th, 50th, and 97.5th percentiles of the response values at each evaluated covariate value were taken from this array of points. These PDPs show how much the PC1 score would change (adjustment value) on the y-axis across a specific covariate’s range on the x-axis. The average PC1 adjustment values (±2 standard errors) were used to evaluate significant differences between the sampling sites.

3. Results

3.1. Concentration of 16S rRNA, ARGs, Fecal Markers, and int1

The concentration of 16S rRNA varied from 2.3 Log10 to 7.6 Log10 GC/mL with an average of 6.0 Log10 GC/mL of water (Figure 2). The concentrations of ARGs, fecal markers, and int1 markers are summarized in Figure 3A. The class 1 integron gene int1 was detected frequently (98%, n = 287) in the watershed. The concentration of int1 ranged from 1.3 to 5.6 Log10 GC/mL with an average concentration of 3.9 Log10 GC/mL. The average concentrations of the E. coli and the HF183 markers were very similar with values of 2.1 Log10 GC/mL and 2.2 Log10 GC/mL, respectively. E. coli was detected in 87% (n = 287) of samples and concentrations varied from 0.71 to 4.0 Log10 GC/mL. The detection frequency of HF183 was 63% (n = 287) with concentration between 0.5 and 5.0 Log10 GC/mL. Among the ARGs, the detection frequency varied from a high of 97% for sulI to a low of 45% for blaSHV (Table 1). A comparison of the detection frequencies between the different ARGs shows that the sulfonamide resistance genes sulI (97%) and sulII (95%) were the most frequent, followed by the tetracycline resistance genes tetM (87%) and tetW (78%), and lastly, the beta-lactam resistance genes ampC (53%) and blaSHV (45%). The concentrations of ARGs also followed a similar trend as the detection frequencies. ARG concentrations were as follows: sulI (1.1–5.4 Log10 GC/mL), sulII (1.1–5.9 Log10 GC/mL), tetM (0.6–4.9 Log10 GC/mL), tetW (0.6–5.2 Log10 GC/mL), ampC (0.3–4.7 Log10 GC/mL), and blaSHV (0.2–4.8 Log10 GC/mL).

3.2. Culturable E. coli and Standard Water Quality Parameters

Details of the distribution of culturable E. coli and water quality parameters including DO, SPC, pH, temperature, and turbidity are summarized in Figure S1 of the supplemental material. The data show that the culturable E. coli concentration varied from 0 to 4.6 Log10 MPN/100 mL with an average of 3.0 Log10 MPN/100 mL on the pooled data. The average temperature, SPC, pH, DO, and turbidity were 18 °C, 270 µS/cm, 7.4, 7.9 mg/L, and 8.1 NTU, respectively.

3.3. Relative Abundance of ARGs, Fecal Markers, pr and int1

We determined the relative abundances of the ARGs, fecal, and int1 markers by normalizing the gene concentrations with the concentration of the 16S rRNA gene. This gives us an idea of the abundance of ARGs, fecal, and int1 genes in relation to the total microbial community abundance in the streams. The relative abundance of sulI (−2.7 Log10 GC/mL), sulII (−2.9 Log10 GC/mL), tetM (−3.7 Log10 GC/mL), tetW (−3.5 Log10 GC/mL), ampC (−4.4 Log10 GC/mL), and blaSHV (−4.6 Log10 GC/mL) are summarized in Figure 3B.

3.4. Correlation of ARGs, Fecal Markers, 16S rRNA, int1, and Water Quality Parameters

Figure 4 shows a correlation plot that depicts the relationships between ARGs, water quality, and environmental parameters measured. Other watershed level characteristics such as the antecedent rainfall prior to sample collection were also evaluated to determine whether there is any association with ARG abundance in the watershed. The strength of the correlation was considered “strong” when the correlation coefficient r > 0.7, “moderate” when 0.4 ≤ r ≤ 0.7, and “weak” when r < 0.4. The same criteria were used to describe the relationships in the negative direction. The relationships between the ARGs were positive and moderate to strong, with r values ranging from 0.4 to 0.86. The relationships were particularly strong within the ARG groups rather than between groups. For example, a strong correlation (r = 0.86) was observed between tetM and tetW, sulI and sulII (r = 0.72), and ampC and blaSHV (r = 0.74). The relationships between the different ARG groups, tet and sul genes (r = 0.54 to 0.68), tet and beta-lactam genes (r = 0.59 to 0.7), and sul and beta-lactam genes (r = 0.40 to 0.63), were positive and varied from moderate to strong.
The fecal markers E. coli and HF183 were also positively correlated with the ARGs, with r ranging from 0.53 to 0.89. Both E. coli and HF183 markers were strongly correlated with the tet genes, moderately correlated with the sul genes, and a moderate and a strong correlation was observed with the beta-lactam genes. The correlation of ARGs with culturable E. coli was strong for the tet genes and moderate for the sul genes and beta-lactam genes. The int1 gene was strongly correlated with the sulI gene (r = 0.77) and moderately correlated with sulII (r = 0.49). Weak correlations were, however, observed between int1 and tet (r = 0.17 to 0.33) and beta-lactam genes (r = 0.14 to 0.17). Moreover, the 16S rRNA gene was moderately correlated with ARGs, with r ranging from 0.4 to 0.5. The relationships between ARGs and water quality parameters including DO, pH, turbidity, temperature, and SPC were weak, whilst antecedent rainfall was generally weakly correlated with ARGs except for tetM, which was moderately correlated (r = 0.40) with 24 h antecedent rainfall.

3.5. Predictive Modeling with Generalized Boosted Models

We developed predictive models with a generalized boosted machine learning algorithm using the PC1 and PC2 scores as the response variables. The results of the PCA (Figure S2 of Supplementary Materials) show that 82% of the variation in the ARGs could be explained by the first two principal components, with PC1 contributing 70% to the total variation in the data. The performance of the gbm package on training and test data is provided in Table 2. For each of these models, the average R2 values for the testing data were about 12% lower than the average R2 values for the training data, indicating some, but not severe, model overfitting. In addition, the Refined PC1 saw only a small decrease (0.04 R2 units) in performance after removing less influential covariates and constraining some of the remaining covariates. Table 3 shows the average influence of each predictor in the PC1 model, the Refined PC1 model, and the PC2 model.
As stated earlier, the covariates with an average influence >3% in the PC1 model were retained and used to develop the Refined PC1 model. The results from the initial model runs for PC1 and PC2 shows that the E. coli marker, HF183, site, int1, culturable E. coli, and water temperature were the most significant in explaining the variation in ARGs in the watershed, contributing 88% and 66% of the average influence for PC1 and PC2, respectively. The refined PC1 model shows that the E. coli marker and HF183 together explain a significant proportion (66%) of the variability in ARGs. Site was the next influential, followed by the int1 gene, water temperature, culturable E. coli, and 16S rRNA. We present PDPs showing the average influence of the predictors on PC1 (analysis limited to PC1, which explained a significant proportion of the variance in ARGs) in Figure 5. The PDPs for the E. coli marker and the HF183 show that as the marker copies increase, the adjustment to PC1 scores becomes less negative, meaning an increase in ARG concentration. Culturable E. coli, int1, and 16S rRNA all had a positive influence on ARG abundance; water temperature showed an interesting relationship (Figure 5f). PC1 scores were high at low temperatures, then decreased at temperatures from 6.4 °C to ~20 °C, then increased again for temperatures above ~20 °C. Figure 6 shows the average influence of the sampling sites on the PC1 scores and ARGs abundance. The results show that GR had by far the most ARG abundance in the watershed, followed by LST, NAVE and HT, and NAVECSO. The sites with the lowest PC1 scores (and thus ARG abundance) were NW and LC, followed by JJ, KC, and SST.

4. Discussion

The contribution of urban watersheds to the emergence and spread of ARGs is now being fully recognized as part of the One Health approach to combatting antibiotic resistance. Urban streams such as Proctor Creek can act as reservoirs for the emergence and spread of ARGs in the environment. Our results show that ARGs occur frequently in streams in the Proctor Creek watershed over the study period. The high prevalence and abundance of sul genes in this study are comparable to findings from previous studies [21,27,35]. Ahmed et al. [21] reported elevated levels of sul genes in storm drain outfalls in Tampa, FL, USA during dry and wet conditions and concluded that fecal pollution was the most likely source of the ARGs. Sul genes were also found to be highly prevalent and abundant in the Huangpu River watershed in China with average concentrations four and five orders of magnitude higher than tet and beta-lactam genes, respectively [27]. The high prevalence of sulI and sulII genes is not surprising given the widespread use of sulfonamide antibiotics, the high abundance of these genes in wastewater, and their association with mobile genetic elements [27,36].
At the watershed level, ARGs may originate from animal sources, sewage or wastewater discharges, selective pressure due to the presence of selective agents such as antibiotics and metals, or may occur naturally in the environment. Although ARG burden from sources such as wastewater treatment plants and animal feeding operations have been well characterized, the contribution of these sources to ARG loads at the watershed level is not well understood. In this study, we evaluated the association of fecal pollution, class 1 integron, water quality, and watershed parameters on ARGs distribution in the urban aquatic environment. Currently, Proctor Creek is listed as impaired for fecal coliforms on Georgia’s 303(d) list of impaired waters and the high prevalence of fecal indicators observed in this study supports that designation. Along with the high prevalence of fecal indicators, the results also show a widespread distribution of ARGs across the watershed. In general, moderate to strong positive correlations were observed between ARGs and the HF183 and E. coli markers, which suggest a strong link between the levels of human fecal pollution and ARG abundance and distribution at the watershed level. Ahmed et al. [21] observed a similar association between ARG abundance in storm drain outfalls in Tampa, FL and the levels of sewage-associated HF183 and crAssphage markers.
The int1 gene, on the contrary, showed a strong positive correlation with the sulI gene and a moderate positive correlation with sulII, whilst the relationships with the other ARGs were weak. The association of the int1 gene with the sulfonamide genes, particularly the sulI gene, has been well described in previous studies and generally reflects the selective pressure imposed by anthropogenic pollution [37,38]. The observed influence of the sewage-associated HF183 and the integrase gene on the sulI gene suggests that human fecal pollution and class 1 integrons, respectively, are important drivers of sulfonamide resistance genes in Proctor Creek. Interestingly, a weak but positive correlation was observed between the int1 gene and the sewage-associated HF183 marker. The occurrence of class 1 integrons from other sources in the environment in addition to human fecal inputs may have contributed to the weak relationship observed between HF183 and int1 gene [2]. Additionally, the abundance of the 16S rRNA gene showed moderate positive correlations with the sewage-associated HF183 and ARGs, suggesting that increased microbial populations resulting from human fecal inputs are an important driver of ARGs in the urban aquatic environment.
To identify the important drivers of ARGs in the Proctor Creek watershed, generalized boosting models were developed to predict the average influence of water quality variables and watershed characteristics on ARG concentrations in the watershed. Generalized boosting models are powerful algorithms that make no distributional assumptions on the model error terms, use stochastic methods to improve model performance, and are robust to missing values and outliers [39]. Results from the refined boosted models show that the abundance of ARGs in the watershed could be primarily explained by the concentrations of E. coli marker, HF183, and the differences between site locations in the watershed. Similar to the sewage-associated HF183 marker, the E. coli marker has been used in water quality studies as an indicator of fecal pollution due to the high levels of E. coli in the feces of humans and other warm-blooded animals [40]. The levels of HF183 and E. coli marker, when considered together, are a strong indicator of the extent of fecal contamination; the model results support the significant contribution of fecal pollution to the abundance of multiple ARG classes in the urban aquatic environment.
The influence of site is also significant, as the location in the watershed reflects a gradient of human impact from the highly urbanized areas surrounding the headwaters to the low-density residential suburbs around the lower reaches of the watershed. In general, there was a marked increase in ARGs in the highly urbanized areas of the watershed at GR, LST, NAVE, HT, and NAVECSO compared to the lower reaches (NW, LC, JJ, KC, and SST). The combined influence of the fecal indicators and site location on ARG variability in the watershed suggests the significant contribution of sewage/wastewater to ARG loading in the Proctor Creek watershed. Other parameters such as the class 1 integron gene, water temperature, and the bacterial community also contributed to the variability in ARGs at the watershed level, but their effect was insignificant compared to fecal pollution. Additionally, water quality parameters such as turbidity, temperature, antecedent rainfall, and season did not appear to influence ARG concentrations at the watershed level. These findings are in line with a recent metadata study that found that the presence of resistance genes in the environment can largely be explained by fecal pollution, except in environments polluted by very high levels of antibiotics, where selection pressure dominates [22]. As mentioned earlier, the Proctor Creek watershed has a long legacy of fecal pollution from CSOs and sanitary sewer leaks that continues to impact water quality. Efforts have been made to control fecal loading in the watershed by decommissioning a CSO at the GR location and limiting the discharges from the NAVE CSO. However, our results suggest that the GR location is still a hotspot for fecal pollution resulting in increased ARG loading into streams. Further studies are needed in the watershed to assess the sources of fecal pollution at the GR location and to implement the necessary mitigation measures to reduce the impact of fecal pollution on the dissemination of ARGs in streams. Additionally, future work should focus on the contribution of fecal pollution to the spread of antibiotic-resistant bacteria and pathogens in the streams to enable a thorough assessment of the risk to public health.

5. Conclusions

Understanding the contribution of different sources to the spread of ARGs in the aquatic environment is an important first step in curtailing the dissemination of resistance genes in the environment. In this study, we identified the influence of fecal pollution and quantified the contribution of sewage-polluted environments to the abundance and spread of ARGs in urban streams. Our results confirm that the abundance and distribution of ARGs in the Proctor Creek watershed are largely explained by the levels of human fecal pollution in this stream. Other factors, such as the levels of the class 1 integron gene, microbial community abundance, and water temperature, played a lesser role in explaining the variability of ARGs in this stream. The evidence from this study supports the theory that in aquatic environments highly impacted by sewage pollution, the input from fecal sources is a more critical ARG source/driver than other factors such as environmental parameters (i.e., temperature or seasonal patterns) and watershed stressors. Finally, the results from this study provide local watershed managers and stakeholders with the information needed to reduce the burden of ARGs and fecal bacteria in urban streams. Future research should examine the relationships between fecal pollution, ARB, and pathogens to identify specific health risks associated with the presence of ARGs in the aquatic environment.

Supplementary Materials

The following supporting information can be downloaded at: https://0-www-mdpi-com.brum.beds.ac.uk/article/10.3390/microorganisms10091804/s1. Table S1: Primers and Probes used in this study; Figure S1: Distribution of standard water quality parameters; Figure S2: Principal component plot showing the loading of ARGs and the influence on first two principal components, PC1 and PC2. The color gradient shows the quality of representation of the ARGs on the principal components. References [41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57] are cited in the supplementary materials.

Author Contributions

Formal analysis, R.A.S. and M.C.; Investigation, M.M., O.G. and B.S.; Project administration, M.M.; Writing—original draft, R.A.S.; Writing—review & editing, M.M. and M.C. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by US Environmental Protection Agency, Office of Research and Development.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data supporting this study can be found at: https://catalog.data.gov/dataset/epa-sciencehub (accessed on 31 August 2022).

Acknowledgments

This document has been reviewed by the U.S. Environmental Protection Agency, Office of Research and Development, and approved for publication. Any mention of trade names, products, or services does not imply an endorsement by the U.S. Government or the U.S. Environmental Protection Agency. The views expressed in this article are those of the author(s) and do not necessarily represent the views or policies of the U.S. Environmental Protection Agency.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Sanderson, H.; Fricker, C.; Brown, R.S.; Majury, A.; Liss, S.N. Antibiotic resistance genes as an emerging environmental contaminant. Environ. Rev. 2016, 24, 205–218. [Google Scholar] [CrossRef]
  2. Stalder, T.; Barraud, O.; Jové, T.; Casellas, M.; Gaschet, M.; Dagot, C.; Ploy, M.-C. Quantitative and qualitative impact of hospital effluent on dissemination of the integron pool. ISME J. 2014, 8, 768–777. [Google Scholar] [CrossRef]
  3. Pruden, A.; Larsson, D.J.; Amézquita, A.; Collignon, P.; Brandt, K.K.; Graham, D.W.; Snape, J.R. Management options for reducing the release of antibiotics and antibiotic resistance genes to the environment. Environ. Health Perspect. 2013, 121, 878. [Google Scholar] [CrossRef]
  4. Engemann, C.A.; Adams, L.; Knapp, C.W.; Graham, D.W. Disappearance of oxytetracycline resistance genes in aquatic systems. FEMS Microbiol. Lett. 2006, 263, 176. [Google Scholar] [CrossRef]
  5. Pruden, A.; Pei, R.; Storteboom, H.; Carlson, K.H. Antibiotic resistance genes as emerging contaminants: Studies in northern Colorado. Environ. Sci. Technol. 2006, 40, 7445–7450. [Google Scholar] [CrossRef]
  6. U.S. Environmental Protection Agency. In Proctor Creek Watershed Monitoring Final Summary Report; #15-0425 USEPA-Region 4 Science and Ecosystem Support Division: Athens, GA, USA, 2018.
  7. Rhodes, G.; Huys, G.; Swings, J.; Mcgann, P.; Hiney, M.; Smith, P.; Pickup, R.W. Distribution of oxytetracycline resistance plasmids between aeromonads in hospital and aquaculture environments: Implication of Tn1721 in dissemination of the tetracycline resistance determinant Tet A. Appl. Environ. Microbiol. 2000, 66, 3883–3890. [Google Scholar] [CrossRef]
  8. Cillóniz, C.; Dominedò, C.; Torres, A. Multidrug resistant gram-negative bacteria in community-acquired pneumonia. Crit. Care 2019, 23, 79. [Google Scholar] [CrossRef]
  9. Kassakian, S.Z.; Mermel, L.A. Changing epidemiology of infections due to extended spectrum beta-lactamase producing bacteria. Antimicrob. Resist. Infect. Control. 2014, 3, 9. [Google Scholar] [CrossRef]
  10. Carrillo-Marquez, M.A.; Hulten, K.G.; Hammerman, W.; Lamberth, L.; Mason, E.O.; Kaplan, S.L. Staphylococcus aureus pneumonia in children in the era of community-acquired methicillin-resistance at Texas Children’s Hospital. Pediatr. Infect. Dis. J. 2011, 30, 545–550. [Google Scholar] [CrossRef]
  11. Skov, R.; Jensen, K. Community-associated meticillin-resistant Staphylococcus aureus as a cause of hospital-acquired infections. J. Hosp. Infect. 2009, 73, 364–370. [Google Scholar] [CrossRef]
  12. Holmes, A.H.; Moore, L.S.; Sundsfjord, A.; Steinbakk, M.; Regmi, S.; Karkey, A.; Piddock, L.J. Understanding the mechanisms and drivers of antimicrobial resistance. Lancet 2016, 387, 176–187. [Google Scholar] [CrossRef]
  13. Larsson, D.J.; Andremont, A.; Bengtsson-Palme, J.; Brandt, K.K.; de Roda Husman, A.M.; Fagerstedt, P.; Kuroda, M. Critical knowledge gaps and research needs related to the environmental dimensions of antibiotic resistance. Environ. Int. 2018, 117, 132–138. [Google Scholar] [CrossRef] [PubMed]
  14. McArthur, J.; Fletcher, D.; Tuckfield, R.C.; Baker-Austin, C. Patterns of Multi-Antibiotic-Resistant Escherichia Coli from Streams with No History of Antimicrobial Inputs. Microb. Ecol. 2016, 72, 840–850. [Google Scholar] [CrossRef]
  15. Finley, R.L.; Collignon, P.; Larsson, D.J.; McEwen, S.A.; Li, X.-Z.; Gaze, W.H.; Topp, E. The scourge of antibiotic resistance: The important role of the environment. Clin. Infect. Dis. 2013, 57, 704–710. [Google Scholar] [CrossRef]
  16. Khachatourians, G.G. Agricultural use of antibiotics and the evolution and transfer of antibiotic-resistant bacteria. Can. Med. Assoc. J. 1998, 159, 1129–1136. [Google Scholar]
  17. Kümmerer, K. Antibiotics in the aquatic environment—A review—Part II. Chemosphere 2009, 75, 435–441. [Google Scholar] [CrossRef]
  18. Perreten, V.; Schwarz, F.; Cresta, L.; Boeglin, M.; Dasen, G.; Teuber, M. Antibiotic resistance spread in food. Nature 1997, 389, 801–802. [Google Scholar] [CrossRef]
  19. Chen, B.; Liang, X.; Huang, X.; Zhang, T.; Li, X. Differentiating anthropogenic impacts on ARGs in the Pearl River Estuary by using suitable gene indicators. Water Res. 2013, 47, 2811–2820. [Google Scholar] [CrossRef]
  20. Baquero, F.; Martínez, J.-L.; Cantón, R. Antibiotics and antibiotic resistance in water environments. Curr. Opin. Biotechnol. 2008, 19, 260–265. [Google Scholar] [CrossRef]
  21. Ahmed, W.; Zhang, Q.; Lobos, A.; Senkbeil, J.; Sadowsky, M.J.; Harwood, V.J.; Ishii, S. Precipitation influences pathogenic bacteria and antibiotic resistance gene abundance in storm drain outfalls in coastal sub-tropical waters. Environ. Int. 2018, 116, 308–318. [Google Scholar] [CrossRef]
  22. Karkman, A.; Pärnänen, K.; Larsson, D.J. Fecal pollution can explain antibiotic resistance gene abundances in anthropogenically impacted environments. Nat. Commun. 2019, 10, 80. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  23. Lawrence, S.J.; LaFontaine, J.H. Occurrence of Organic Wastewater-Indicator Compounds in Urban Streams of the Atlanta Area, Georgia, 2003–2006; U.S. Geological Survey Scientific Investigations Report: Reston, VA, USA, 2010; p. 113. [Google Scholar]
  24. Horowitz, A.J. Monitoring suspended sediments and associated chemical constituents in urban environments: Lessons from the city of Atlanta, Georgia, USA water quality monitoring program. J. Soils Sediments 2009, 9, 342–363. [Google Scholar] [CrossRef]
  25. Horowitz, A.J.; Elrick, K.A.; Smith, J.J. Monitoring urban impacts on suspended sediment, trace element, and nutrient fluxes within the City of Atlanta, Georgia, USA: Program design, methodological considerations, and initial results. Hydrol. Process. 2008, 22, 1473–1496. [Google Scholar] [CrossRef]
  26. U.S. Geological Survey. Collection of Water Samples. In Techniques of Water-Resources Investigations; U.S. Geological Survey: Reston, VA, USA, 2006; Chapter A4; p. 231. [Google Scholar]
  27. Jiang, L.; Hu, X.; Xu, T.; Zhang, H.; Sheng, D.; Yin, D. Prevalence of antibiotic resistance genes and their relationship with antibiotics in the Huangpu River and the drinking water sources, Shanghai, China. Sci. Total Environ. 2013, 458, 267–272. [Google Scholar] [CrossRef]
  28. Sowah, R.A.; Habteselassie, M.Y.; Radcliffe, D.E.; Bauske, E.; Risse, M. Isolating the impact of septic systems on fecal pollution in streams of suburban watersheds in Georgia, United States. Water Res. 2017, 108, 330–338. [Google Scholar] [CrossRef]
  29. Ahmed, W.; Harwood, V.; Gyawali, P.; Sidhu, J.; Toze, S. Comparison of Concentration Methods for Quantitative Detection of Sewage-Associated Viral Markers in Environmental Waters. Appl. Environ. Microbiol. 2015, 81, 2042–2049. [Google Scholar] [CrossRef]
  30. Kildare, B.J.; Leutenegger, C.M.; McSwain, B.S.; Bambic, D.G.; Rajal, V.B.; Wuertz, S. 16S rRNA-based assays for quantitative detection of universal, human-, cow-, and dog-specific fecal Bacteroidales: A Bayesian approach. Water Res. 2007, 41, 3701–3715. [Google Scholar] [CrossRef]
  31. Okabe, S.; Okayama, N.; Savichtcheva, O.; Ito, T. Quantification of host-specific Bacteroides–Prevotella 16S rRNA genetic markers for assessment of fecal pollution in freshwater. Appl. Microbiol. Biotechnol. 2007, 74, 890–901. [Google Scholar] [CrossRef]
  32. Reischer, G.H.; Kasper, D.C.; Steinborn, R.; Mach, R.L.; Farnleitner, A.H. Quantitative PCR method for sensitive detection of ruminant fecal pollution in freshwater and evaluation of this method in alpine karstic regions. Appl. Environ. Microbiol. 2006, 72, 5610–5614. [Google Scholar] [CrossRef]
  33. R Core Team. R: A language and environment for statistical computing. In R Foundation for Statistical Computing; R Core Team: Vienna, Austria, 2019; Available online: http://www.R-project.org/ (accessed on 31 August 2022).
  34. Friedman, J.H. Greedy function approximation: A gradient boosting machine. Ann. Stat. 1999, 29, 1189–1232. [Google Scholar] [CrossRef]
  35. Storteboom, H.; Arabi, M.; Davis, J.G.; Crimi, B.; Pruden, A. Tracking antibiotic resistance genes in the South Platte River basin using molecular signatures of urban, agricultural, and pristine sources. Environ. Sci. Technol. 2010, 44, 7397–7404. [Google Scholar] [CrossRef]
  36. Stoll, C.; Sidhu, J.; Tiehm, A.; Toze, S. Prevalence of clinically relevant antibiotic resistance genes in surface water samples collected from Germany and Australia. Environ. Sci. Technol. 2012, 46, 9716–9726. [Google Scholar] [CrossRef]
  37. Gillings, M.R.; Gaze, W.H.; Pruden, A.; Smalla, K.; Tiedje, J.M.; Zhu, Y.-G. Using the class 1 integron-integrase gene as a proxy for anthropogenic pollution. ISME J. 2015, 9, 1269. [Google Scholar] [CrossRef]
  38. Koczura, R.; Mokracka, J.; Taraszewska, A.; Łopacinska, N. Abundance of Class 1 Integron-Integrase and Sulfonamide Resistance Genes in River Water and Sediment Is Affected by Anthropogenic Pressure and Environmental Factors. Microb. Ecol. 2016, 72, 909–916. [Google Scholar] [CrossRef]
  39. De’ath, G.; Fabricius, K.E. Classification and regression trees: A powerful yet simple technique for ecological data analysis. Ecology 2000, 81, 3178–3192. [Google Scholar] [CrossRef]
  40. Harwood, V.J.; Staley, C.; Badgley, B.D.; Borges, K.; Korajkic, A. Microbial source tracking markers for detection of fecal contamination in environmental waters: Relationships between pathogens and human health outcomes. FEMS Microbiol. Rev. 2014, 38, 1–40. [Google Scholar] [CrossRef]
  41. Bradshaw, J.K.; Snyder, B.J.; Oladeinde, A.; Spidle, D.; Berrang, M.E.; Meinersmann, R.J.; Molina, M. Characterizing relationships among fecal indicator bacteria, microbial source tracking markers, and associated waterborne pathogen occurrence in stream water and sediments in a mixed land use watershed. Water Res. 2016, 101, 498–509. [Google Scholar] [CrossRef]
  42. Oladeinde, A.; Bohrmann, T.; Wong, K.; Purucker, S.; Bradshaw, K.; Brown, R.; Molina, M. Decay of fecal indicator bacterial populations and bovine-associated source-tracking markers in freshly deposited cow pats. Appl. Environ. Microbiol. 2014, 80, 110–118. [Google Scholar] [CrossRef]
  43. Keely, S.P.; Brinkman, N.E.; Wheaton, E.A.; Jahne, M.A.; Siefring, S.D.; Varma, M.; Hill, R.A.; Leibowitz, S.G.; Martin, R.W.; Garland, J.L.; et al. Geospatial Patterns of Antimicrobial Resistance Genes in the US EPA National Rivers and Streams Assessment Survey. Environ. Sci. Technol. 2022. [Google Scholar] [CrossRef]
  44. Field, K.G.; Samadpour, M. Fecal source tracking, the indicator paradigm, and managing water quality. Water Res. 2007, 41, 3517–3538. [Google Scholar] [CrossRef]
  45. Opel, K.L.; Chung, D.; McCord, B.R. A study of PCR inhibition mechanisms using real time PCR. J. Forensic Sci. 2010, 55, 25–33. [Google Scholar] [CrossRef] [PubMed]
  46. Greenwell, B.; Boehmke, B.; Cunningham, J.; GBM Developers. gbm: Generalized Boosted Regression Models. R package version 2.1.5. 2019. Available online: https://cran.r-project.org/package=gbm (accessed on 31 August 2022).
  47. Pei, R.; Kim, S.-C.; Carlson, K.H.; Pruden, A. Effect of River Landscape on the sediment concentrations of antibiotics and corresponding antibiotic resistance genes (ARG). Water Res. 2006, 40, 2427–2435. [Google Scholar] [CrossRef] [PubMed]
  48. Calero-Cáceres, W.; Melgarejo, A.; Colomer-Lluch, M.; Stoll, C.; Lucena, F.; Jofre, J.; Muniesa, M. Sludge as a potential important source of antibiotic resistance genes in both the bacterial and bacteriophage fractions. Environ. Sci. Technol. 2014, 48, 7602–7611. [Google Scholar] [CrossRef]
  49. Walsh, F.; Ingenfeld, A.; Zampicolli, M.; Hilber-Bodmer, M.; Frey, J.E.; Duffy, B. Real-time PCR methods for quantitative monitoring of streptomycin and tetracycline resistance genes in agricultural ecosystems. J. Microbiol. Methods 2011, 86, 150–155. [Google Scholar] [CrossRef]
  50. Peak, N.; Knapp, C.W.; Yang, R.K.; Hanfelt, M.M.; Smith, M.S.; Aga, D.S.; Graham, D.W. Abundance of six tetracycline resistance genes in wastewater lagoons at cattle feedlots with different antibiotic use strategies. Environ. Microbiol. 2007, 9, 143–151. [Google Scholar] [CrossRef] [PubMed]
  51. Fernando, D.M.; Tun, H.M.; Poole, J.; Patidar, R.; Li, R.; Mi, R.; Farenhorst, A. Detection of Antibiotic Resistance Genes in Source and Drinking Water Samples from a First Nations Community in Canada. Appl. Environ. Microbiol. 2016, 82, 4767–4775. [Google Scholar] [CrossRef]
  52. Xi, C.; Zhang, Y.; Marrs, C.F.; Ye, W.; Simon, C.; Foxman, B.; Nriagu, J. Prevalence of antibiotic resistance in drinking water treatment and distribution systems. Appl. Environ. Microbiol. 2009, 75, 5714–5718. [Google Scholar] [CrossRef]
  53. Barraud, O.; Baclet, M.-C.; Denis, F.; Ploy, M.-C. Quantitative multiplex real-time PCR for detecting class 1, 2 and 3 integrons. J. Antimicrob. Chemother. 2010, 65, 1642–1645. [Google Scholar] [CrossRef]
  54. Green, H.C.; Haugland, R.A.; Varma, M.; Millen, H.T.; Borchardt, M.A.; Field, K.G.; Kelty, C.A. Improved HF183 quantitative real-time PCR assay for characterization of human fecal pollution in ambient surface water samples. Appl. Environ. Microbiol. 2014, 80, 3086–3094. [Google Scholar] [CrossRef]
  55. Chern, E.; Siefring, S.; Paar, J.; Doolittle, M.; Haugland, R. Comparison of quantitative PCR assays for Escherichia coli targeting ribosomal RNA and single copy genes. Lett. Appl. Microbiol. 2011, 52, 298–306. [Google Scholar] [CrossRef]
  56. Suzuki, M.T.; Taylor, L.T.; DeLong, E.F. Quantitative analysis of small-subunit rRNA genes in mixed microbial populations via 5′-nuclease assays. Appl. Environ. Microbiol. 2000, 66, 4605–4614. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  57. Haugland, R.A.; Siefring, S.C.; Wymer, L.J.; Brenner, K.P.; Dufour, A.P. Comparison of Enterococcus measurements in freshwater at two recreational beaches by quantitative polymerase chain reaction and membrane filter culture analysis. Water Res. 2005, 39, 559–568. [Google Scholar] [CrossRef] [PubMed]
Microorganisms 10 01804 g001 550
Figure 1. Proctor Creek watershed showing sampling points and land-use distribution.
Figure 1. Proctor Creek watershed showing sampling points and land-use distribution.
Microorganisms 10 01804 g001
Microorganisms 10 01804 g002 550
Figure 2. Abundance of 16S rRNA gene copies in surface water samples across the sampling points. The boxplots show the 25th quartile, 75th quartile, and median lines. The diamond shape represents the mean observations for each site, whilst outliers are represented by the solid points.
Figure 2. Abundance of 16S rRNA gene copies in surface water samples across the sampling points. The boxplots show the 25th quartile, 75th quartile, and median lines. The diamond shape represents the mean observations for each site, whilst outliers are represented by the solid points.
Microorganisms 10 01804 g002
Microorganisms 10 01804 g003 550
Figure 3. Concentration (A) and relative abundance (B) of ARGs, fecal markers, and int1 gene for the pooled dataset. Relative abundance was calculated as the concentration of ARGs normalized to the concentration of the 16S rRNA gene.
Figure 3. Concentration (A) and relative abundance (B) of ARGs, fecal markers, and int1 gene for the pooled dataset. Relative abundance was calculated as the concentration of ARGs normalized to the concentration of the 16S rRNA gene.
Microorganisms 10 01804 g003
Microorganisms 10 01804 g004 550
Figure 4. Spearman rank correlations among ARGs, fecal markers, int1 gene, 16S rRNA, standard water quality parameters, and antecedent rainfall. The correlation r is represented by the area of the pie.
Figure 4. Spearman rank correlations among ARGs, fecal markers, int1 gene, 16S rRNA, standard water quality parameters, and antecedent rainfall. The correlation r is represented by the area of the pie.
Microorganisms 10 01804 g004
Microorganisms 10 01804 g005 550
Figure 5. Partial dependence plots showing the response of PC1 scores to changes in predictor variables E. coli marker (a), HF183 (b), culturable E. coli (c), int1 (d), 16S (e) and water temperature (f) in the Refined PC1 model. Dotted line shows the 95% confidence band based on 500 runs of the Refined PC1 model.
Figure 5. Partial dependence plots showing the response of PC1 scores to changes in predictor variables E. coli marker (a), HF183 (b), culturable E. coli (c), int1 (d), 16S (e) and water temperature (f) in the Refined PC1 model. Dotted line shows the 95% confidence band based on 500 runs of the Refined PC1 model.
Microorganisms 10 01804 g005
Microorganisms 10 01804 g006 550
Figure 6. Partial Dependence Plot showing the influence of sampling locations on ARG abundance, as represented on the y-axis by the PC1 scores. Sampling locations are arranged following a gradient of high to low human fecal input from GR to LC. Different letters on boxplots represent significant differences whilst similar letters show no differences.
Figure 6. Partial Dependence Plot showing the influence of sampling locations on ARG abundance, as represented on the y-axis by the PC1 scores. Sampling locations are arranged following a gradient of high to low human fecal input from GR to LC. Different letters on boxplots represent significant differences whilst similar letters show no differences.
Microorganisms 10 01804 g006
Table
Table 1. Detection frequencies (%) of ARGs, class 1 integron, and fecal markers.
Table 1. Detection frequencies (%) of ARGs, class 1 integron, and fecal markers.
SitesulIsulIItetM tetW ampCblaSHV int1 E. coliHF183
Burbank (BB)96921001009256969696
Greensferry (GR)100961001001009610096100
North Ave. (NAVE)1001001001008484100100100
Lindsay Street (LST)1001001009640361007680
North Ave. CSO (NAVECSO)968879463833966738
Grove Park (GP)10096889624201009260
Hortense (HT)100100847640441009264
Kerry Circle (KC)1001008472724410010068
Spring St. (SST)100100807636321008856
James Jackson (JJ)10096766840321008840
Lillian Cooper (LC)71717950362186710
Northwest (NW)10096725240401008452
Mean979587785345988763
Table
Table 2. Model performance for training and testing data for each of the modeled datasets, with standard deviations in parentheses.
Table 2. Model performance for training and testing data for each of the modeled datasets, with standard deviations in parentheses.
ModelR2 TrainingR2 Testing
PC10.99 (0.01)0.86 (0.03)
Refined PC10.95 (0.01)0.85 (0.04)
PC20.99 (0.02)0.49 (0.10)
Table
Table 3. Average influence of the predictor variables for the PC1, Refined PC1, and PC2 models based on 500 GBM runs for each. Standard deviations are in parentheses. The 7 predictors above the dotted line were retained and included in the Refined PC1 model.
Table 3. Average influence of the predictor variables for the PC1, Refined PC1, and PC2 models based on 500 GBM runs for each. Standard deviations are in parentheses. The 7 predictors above the dotted line were retained and included in the Refined PC1 model.
ParameterAve Influence,
PC1 (%)		ConstraintAve Influence,
Refined PC1 (%)		Ave Influence,
PC2 (%)
E. coli Marker34.0 (3.8)+36.8 (3.7)7.9 (1.3)
HF18327.4 (3.5)+29.5 (3.5)6.3 (0.8)
Site11.0 (1.0)None12.1 (1.2)15.2 (1.0)
int16.9 (1.1)+6.7 (1.1)21.7 (2.0)
Culturable E. coli5.1 (1.3)+5.4 (1.6)6.5 (1.0)
Water Temperature3.2 (0.4)None6.2 (0.8)8.3 (0.9)
16S3.0 (0.4)+3.2 (0.6)3.5 (0.5)
Season2.2 (0.3) 3.5 (0.6)
Turbidity1.7 (0.3) 4.4 (0.6)
Dissolved Oxygen1.5 (0.2) 5.1 (1.0)
pH1.1 (0.2) 2.8 (0.5)
Specific Conductance1.0 (0.2) 4.7 (0.7)
48 h Rainfall0.9 (0.2) 5.3 (1.3)
24 h Rainfall0.6 (0.1) 3.3 (0.6)
72 h Rainfall0.6 (0.1) 1.4 (0.4)
+ Response variable constrained to be monotonically increasing.
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Sowah, R.A.; Molina, M.; Georgacopoulos, O.; Snyder, B.; Cyterski, M. Sources and Drivers of ARGs in Urban Streams in Atlanta, Georgia, USA. Microorganisms 2022, 10, 1804. https://0-doi-org.brum.beds.ac.uk/10.3390/microorganisms10091804

AMA Style

Sowah RA, Molina M, Georgacopoulos O, Snyder B, Cyterski M. Sources and Drivers of ARGs in Urban Streams in Atlanta, Georgia, USA. Microorganisms. 2022; 10(9):1804. https://0-doi-org.brum.beds.ac.uk/10.3390/microorganisms10091804

Chicago/Turabian Style

Sowah, Robert A., Marirosa Molina, Ourania Georgacopoulos, Blake Snyder, and Mike Cyterski. 2022. "Sources and Drivers of ARGs in Urban Streams in Atlanta, Georgia, USA" Microorganisms 10, no. 9: 1804. https://0-doi-org.brum.beds.ac.uk/10.3390/microorganisms10091804

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop