Antimicrobial-Resistant Escherichia coli from Environmental Waters in Northern Colorado

Waterborne Escherichia coli are a major reservoir of antimicrobial resistance (AMR), including but not limited to extended-spectrum beta-lactamase (ESBL) and Klebsiella pneumoniae carbapenemase (KPC) mechanisms. This study quantified and described ESBL- and KPC-producing E. coli in Northern Colorado from sewer water, surface water, and influent and effluent wastewater treatment sources. Total detected bacteria and E. coli abundances, and the percentages that contain ESBL and/or KPC, were compared between water sources. Seventy E. coli isolates from the various waters had drug resistance validated with a panel of 17 antibiotics using a broth microdilution assay. The diverse drug resistance observed across E. coli isolates was further documented by polymerase chain reaction of common ESBL genes and functional relatedness by PhenePlate assay-generated dendrograms (n=70). The total E. coli abundance decreased through the water treatment process as expected, yet the percentages of E. coli harboring ESBL resistance were increased (1.70%) in surface water. Whole-genome sequencing analysis was completed for 185 AMR genes in wastewater E. coli isolates and confirmed the presence of diverse AMR gene classes (e.g., beta-lactams and efflux pumps) in isolate genomes. This study completed surveillance of AMR patterns in E. coli that reside in environmental water systems and suggests a role for integrating both phenotypic and genotypic profiling beyond ESBL and KPC mechanisms. AMR screening via multiple approaches may assist in the prevention of drug-resistant E. coli spread from waters to animals and humans.

Escherichia coli is abundant in water systems [15] and is a concerning reservoir for AMR in these locations [12,16,17]. A survey of AMR E. coli in the Netherlands showed 17.1% of ESBL E. coli isolated from river water and wastewater were reported as pathogenic, and of those pathogenic strains, approximately 84% exhibited resistance in up to three drug classes including beta-lactams, tetracyclines, and aminoglycosides [18]. In an estuary in Portugal, isolated E. coli was phylogenetically distributed into commensal and pathogenic groups, and bacteria in both groups were resistant to lastresort antibiotics, including carbapenems [19]. Given that antibiotics are conventionally used to treat E. coli infections in people and animals, improved AMR surveillance and a better understanding of how resistance spreads in environmental waters are warranted [20,21].
Two major types of beta-lactam resistance include microbial production of extended-spectrum beta-lactamase (ESBL) and Klebsiella pneumoniae carbapenemase (KPC) enzymes [22]. ese enzymes work via hydrolysis of the betalactam ring (ESBL) or via interference with antibiotic binding to penicillin-binding proteins (KPC) [23]. e Centers for Disease Control and Prevention classifies ESBL as a serious threat and KPC as an urgent threat to public health [24]. In the USA alone, ESBL resistance among Enterobacteriaceae, which includes E. coli, was reported to have caused approximately 1,700 deaths with medical costs per infection in excess of $40,000 US dollars [24]. KPC was attributed to 600 deaths in 2013 [24], and carbapenem drugs are considered a last-resort antibiotic for severe bacterial infections [25]. Furthermore, E. coli that express ESBL and/or KPC genes have been frequently shown to harbor resistance to other antibiotic classes [11,26]. A thorough application of functional screening methods may be needed to comprehensively characterize the AMR phenotypes of waterborne E. coli.
ere are multiple genes contributing to ESBL and KPC resistance, including blaCTX-M, blaOXA, blaSHV and blaTEM, and blaKPC, blaVIM, blaIMP, and blaNDM [27], all of which can be transferred to other species horizontally on plasmids [28]. Considering the potential of horizontal AMR gene transfer into reservoirs of human pathogens, increased attention is being given to the identification of AMR genes in bacteria isolated from environmental waters. e objective of this study was to screen environmental water samples for AMR E. coli, to understand the phenotypic and genotypic resistance profiles of E. coli isolates, and to examine clonal relatedness of AMR E. coli strains across water systems. It was hypothesized that KPC and ESBL screening of environmental E. coli would be predictive of additional multidrug resistance mechanisms in these isolates, and that genotypic and phenotypic analysis would reveal patterns in drug resistance based on date and location of the water sources sampled. e cross-validation of the methodologies reported herein illustrates the depth by which multiple analysis platforms can be integrated to establish AMR profiles of E. coli across environmental waters.

Environmental Water Sampling Locations and Collection
Procedures. Water samples were collected from Larimer County, Colorado, from May 2016 to April 2017. All water samples were collected with autoclaved Pyrex ® wide mouth storage bottles (quantity of 1 liter). Following collection, all samples were immediately placed on ice and stored in a lightsensitive container until analysis, which occurred at least one hour post-collection. Remaining sample was then stored at 4°C. A map of sample collection locations is provided in supplementary information ( Figure S1). Seventeen water samples were collected and classified as follows: sewer water, wastewater treatment plant (WWTP) influent, WWTP effluent, and surface water. Sewer water and WWTP influent were grouped as wastewater, and WWTP effluent and surface water were grouped as ambient water. Sewer water (n � 6) was collected from five manholes downstream from a human hospital and also from area residents, businesses, and a university. WWTP-influent (n � 3) and WWTP-effluent (n � 2) samples were collected using the Hawk Composite Sampler (Hach, Loveland, CO) at the Drake Water Reclamation Facility, a conventional activated sludge treatment plant. WWTP-influent samples were collected before water treatment, and WWTP-effluent samples were collected after sulfur dioxide and chlorine treatment and prior to discharge into the Cache la Poudre River. Surface water (n � 6) was collected from five sources along the Cache La Poudre River both upstream and downstream from the WWTP-effluent discharge, with one site being upstream of the city limits. Storage bottles were submerged midstream in surface water for river samples and from covered manholes for sewer water samples. WWTP-influent and WWTP-effluent samples were collected midflow from the Hawk Composite Sampler.

Microbiological Analysis of Total Detected Bacteria and Suspected E. coli from Environmental Water Samples.
Seventeen independent sampling events between May 2016 and April 2017 were screened for total detected bacterial and E. coli abundance, as well as for ESBL and KPC resistance. Water samples were processed within 24 hours of acquisition. Samples were serially diluted in sterilized phosphatebuffered saline (1× PBS, pH 7.4, VWR, Radnor, PA) at room temperature (23°C) and subjected to membrane filtration through cellulose nitrate filters (0.45 µm pore size; Millipore Sigma, Burlington, MA) for bacterial enumeration. With each sample, filtration was performed in triplicate at three different dilutions and membrane filters were placed on three different chromogenic agar types: CHROMagar Orientation ™ ; CHROMagar ESBL ™ , which screens for ESBLproducing isolates; and CHROMagar KPC ™ (DRG Diagnostics, Springfield, NJ), which screens for KPCproducing isolates. All agars, broths, and other liquid reagents were incubated for 24 hours and checked for sterility before use, and agar plates were stored along with inoculated plates to monitor for any contamination occurring during the filtering process. Additionally, contamination in sample dilutions was controlled for using sterilized and filtered phosphate-buffered saline (PBS) that was plated, incubated, and assessed for colony growth along with inoculated plates. Although CHROMagar KPC plates have higher sensitivity than KPC-screening agars [29,30], bacterial growth on these plates is not an absolute conformation of KPC production [31,32]. is is because resistance to carbapenems can be mediated by both KPC enzymes and efflux pumps [33], and consequently, a positive result will be referred to as "suspected KPC." CHROMagar Orientation quantified total detected bacterial abundance and total E. coli. CHROMagar ESBL quantified ESBL-producing total detected bacteria and ESBL E. coli, and CHROMagar KPC quantified suspected KPC-producing total detected bacteria and E. coli. Plates were incubated at 37°C at 2-5% CO 2 for 24-72 hours, until countable colonies were observed. Total colony-forming units (CFUs) per plate within the range of 15-300 were included for descriptive analysis. As per manufacturer instructions, species identifications were made using CFU color and morphology, where E. coli colonies were pink with a faint halo [34,35]. E. coli counts were included in the abundance calculation if they were counted on a plate with at least 15 (lowest countable detection limit) and at most 300 colonies (highest countable detection limit). To compare total detected AMR bacteria and total AMR E. coli abundances between sampling locations, relative percentages of ESBL and suspected KPC resistance were calculated by dividing the number of ESBL or suspected KPC-positive E. coli or total detected bacteria by E. coli or total detected bacteria in each water source.

E. coli Purification, Isolation, and Species Confirmation.
A spread plate-based screening of all water samples was used to collect E. coli isolates for more in-depth drug susceptibility and clonal relatedness analysis. E. coli isolates from each water sample were selected from 1 of the 3 CHRO-Magar plates (Orientation, ESBL, or KPC) based on the chromogenic appearance as pink colonies with a faint halo [34,35]. Single colonies were isolated to MacConkey agar (Difco, BD Biosciences: Franklin Lakes, NJ) and incubated for 24-72 hours at 2-5% CO 2 , until visible colonies were observed. As per manufacturer instructions, isolates that appeared bright pink on MacConkey agar were defined as E.
coli and were further isolated onto tryptic soy agar (Remel ™ , Lenexa, KS) plates and incubated overnight at 37°C. Single colonies were selected and inoculated into sterile tryptic soy broth (Remel ™ , Lenexa, KS) for 24-72 hours, until bacterial cell growth was observed. Blank tryptic soy broth was incubated alongside isolated cultures as a negative control, and any growth in noninoculated media blanks was addressed by repeating the assay by selecting new E. coli colonies from the original agar plate and inoculating them into new sterile media.
Potential E. coli isolates in tryptic soy broth were species-confirmed using matrix-assisted laser desorption ionization time of flight mass spectrometry (MALDI-TOF) at the Colorado State University Proteomics and Metabolomics Facility (CSU-PMF). Briefly, isolates were dissolved in 1 µL of 70% formic acid on a Biotyper plate (Bruker Daltonics Inc© Billerica, MA), and the dried spots were mixed with 1 µL of alpha-Cyano-4-hydroxycinnamic acid (HCCA) and analyzed in triplicate using the VITEK-MS ™ (Biomerieux, USA). A 0.5 µL Bacterial Test Standard (Bruker Daltonics Inc© Billerica, MA) was added as an internal control. Protein identities were assigned using a CSU-PMF internal library, where a 70% match of total mass-spectral peaks (score of >2.0) identified an isolate as E. coli. Seventy total isolates were confirmed as E. coli, including ESBL-producing (n � 12), ESBL + suspected KPC-producing (n � 35), and neither ESBL-nor KPCproducing (n � 23) E. coli.

DNA Isolation and PCR Amplification of ESBL and KPC
Genes from E. coli Isolates. To further establish E. coli ESBL resistance profiles, polymerase chain reaction (PCR) was performed on the 70 confirmed E. coli isolates for 3 genes (blaOXA, blaCTX-M, and blaTEM) representative of ESBL phenotypes. e following primer sequences were used during DNA amplification, where "F" indicates the sequence of the forward primer, "R" indicates the sequence of the reverse primer, and "bp" indicates the number of base pairs in the primer: E. coli isolates were lysed in 100 µL of Millipore water at 100°C for 1 hour using a Bio-Rad T100 ™ thermocycler (Bio-Rad Laboratories, Inc, California). Amplification was carried out by 2 µL DNA, 10 pmol of each primer, and 12.5 µl Emerald Amp ® GT PCR Master Mix (Takara Bio Inc., Clontech, Japan) under previously described conditions [36]. e PCR conditions were as follows: 15 minutes of denaturation at 95°C (1 cycle), 30 seconds of denaturation at 94°C, 90 seconds of annealing at 62°C, and 1 minute of polymerization at 72°C (34 cycles), with a final extension at 72°C for 10 minutes. PCR products were analyzed on a 1.5% agarose gel (Bio-Rad, Hercules, CA) and visualized using 1 µL of ethidium bromide ( ermo Fisher, Lafayette, CO). PCR-grade water or sterile autoclaved 1× PBS was used as a negative control. Bacterial isolates with confirmed presence of blaOXA, blaCTX-M, and blaTEM were included as positive controls.

Biochemical Analysis to Establish Clonal
Relatedness across E. coli Isolates. E. coli profiled with PCR were screened for clonal relatedness using a PhP-RE plate (PhP-RE, PhPlate AB, Microplate Techniques, Stockholm, Sweden) following the manufacturer's instructions [37]. e system consisted of a 96-well plate coated with 11 carbon sources: cellobiose, lactose, rhamnose, deoxyribose, sucrose, sorbose, tagatose, D-arabitol, raffinose, gal-Lacton, and ornithine. ree hundred µL of media containing a pH indicator (bromothymol blue 0.11% w/v) and proteose peptone (Sigma-Aldrich: St. Louis, MO) was combined with ∼1 mg of each bacterial isolate. After 1.5 hours, 12 µL of inoculum was transferred into each substrate well. One row of substrate was not inoculated and served as a negative control. As an internal check for reproducibility, three replicate isolate pairs were grown on separate 96-well plates and were determined to have correlation values of 0.94 and above. Plates were incubated at 37°C, covered with light-sensitive material, and read at 620 nm on a BioTek Cytation ™ 3 (BioTek Instruments: Winooski, VT) at 8 h, 24 h, and 48 h. Clonal relatedness was estimated using PhenePlate ™ software (PhPlate AB), which examined Journal of Environmental and Public Health variability in absorbance across each substrate and presented relatedness as a dendrogram. A cutoff value of <97.5% similarity in functional profiles defined isolates as clonally distinct from each other.

E. coli Isolates Tested for Antibiotic Susceptibility to 17
Antibiotics.
irty-four E. coli isolates were further screened for antimicrobial resistance using a broth microdilution assay at the Colorado State University Veterinary Teaching Hospital-Diagnostic Medicine Center (VTH-DMC). e isolates were from sewer water (n � 17), WWTP influent (n � 10), WWTP effluent (n � 2), and surface water (n � 5), and were ESBL-producing (n � 10), suspected ESBL + KPCproducing (n � 1), or neither KPC-nor ESBL-producing (n � 23) E. coli. Given the higher abundance of ESBL/KPCpositive E. coli detected in wastewater samples during membrane filtration, this analysis preferentially screened E. coli isolates from those water sources. e 17 antibiotics tested included amoxicillin/clavulanic acid, ampicillin, cefalexin, cefovecin, ceftiofur, chloramphenicol, enrofloxacin, gentamicin, imipenem, marbofloxacin, nitrofurantoin, piperacillin, tetracycline, tobramycin, trimethoprim/sulfamethoxazole, amikacin, and cefpodoxime. Minimum inhibitory concentrations (MIC) were calculated and categorized as susceptible, intermediate, or resistant based on standards set by the Clinical and Laboratory Standards Institute (CLSI) guidelines [38]. For each drug, percent resistance was calculated by dividing the number of resistant isolates by the total number of isolates screened with that drug.

Whole-Genome Sequencing of E. coli Isolates.
Whole-genome sequencing was performed on a subset of E. coli isolated from WWTP-influent and sewer water samples. Twenty-five E. coli isolates were grown at 37°C at 2-5% CO 2 for 24-48 hours in sterile tryptic soy broth (TSB), centrifuged at 4000× g for 10 minutes (Beckman Coulter Allegra X-14R, Indianapolis, IN, USA), and reconstituted in approximately 250 µL of TSB. DNA was extracted using a DNeasy PowerSoil Kit (Qiagen, Valencia, CA) following manufacturer protocols and stored at −20°C until quality checking and quantification with a NanoDrop 2000 ( ermo Scientific, Lafayette, CO). Incubated sterile TSB media and sterile DNA extraction media were used as negative controls.
Whole-genome sequencing of DNA extracts was performed at the Animal Disease Research and Diagnostic Laboratory at South Dakota State University in Brookings, SD, using previously described methods [39]. Briefly, 0.3 ng/ µL of DNA from each isolate was processed using a Nextera XT DNA Sample Prep Kit (Illumina Inc., San Diego, CA), pooled together, and sequenced on an Illumina Miseq platform using a 2 × 250 paired end approach with V2 chemistry (Illumina Inc., San Diego, CA). Raw sequencing files were demultiplexed and converted to FASTQ files using Casava version 1.8.2 (Illumina Inc., San Diego, CA). e CLC Genomics workbench version 9.4 (Qiagen bioinformatics, Valencia, CA) trimmed and assembled sequences, and for each isolate, AMR genes were BLAST searched against 185 distinct gene sequences from ResFinder [40] and the Comprehensive Antibiotic Resistance Database [41]. Gene identities were made based on a minimum 85% sample gene identity match over 50% of the gene sequence length, as compared with the database entry of the gene.

Statistical Analysis
Total bacteria and total E. coli abundances were compared across the four water sample types using a Kruskal-Wallis nonparametric analysis of variance test with a Dunn's posttest to adjust for multiple comparisons. Ambient and wastewater were compared using a nonparametric Mann-Whitney test. A p-value of p < 0.05 was defined as a statistically significant difference between water sources, and these values are reported alongside source abundance fold differences in the results below.

Total Detected Bacterial and E. coli Abundances through
Wastewater Treatment Process. Total detected bacterial and E. coli abundances across environmental waters tested in 17 independent sampling events are depicted in Table 1. e negative control plates included for each filtered sample did not contain microbial growth. Filter plates from each source that had CFU counts between 15 and 300 and were included in source abundance. Of 460 total filter plates counted, a total of 216 plates were excluded from analysis, including 34 that exceeded a count of 300. To estimate relative bacterial density, total detected bacterial abundance and E. coli abundance were compared across sewer water, WWTP-influent, WWTP-effluent, and surface water locations. Total detected bacteria ranged from 2.1 + 04 CFU/100 mL in surface water to 3.4E + 08 CFU/ 100 mL in WWTP influent. Among wastewater, WWTP influent had 1.7-fold greater total detected bacterial abundance than sewer water (3.4E + 08 CFU/100 mL and 2.1E + 08 CFU/100 mL, respectively), though this difference was not significantly significant. Elevated levels of bacteria in WWTP-influent samples could be due to the convergence of different sewer water sources into a common WWTP ( Figure S1). Wastewater (sewer and WWTP influent) had 3221-fold higher (p < 0.001) total detected bacterial abundance than ambient water (WWTP effluent and surface water). Decreases in total detected bacterial abundance after wastewater treatment have been previously reported in other microbial wastewater treatment sampling investigations [42][43][44], suggesting that this studied wastewater treatment plant functions in a similar manner to previously examined plants.
As a percentage of total detected bacteria, E. coli was the highest in sewer water at 37%, detected at lower levels in WTTP influent (22%) and surface water (14%), and was lowest in WWTP effluent (2.4%) ( Table 1). E. coli abundance followed similar trends to total detected bacterial abundance, with the greatest total E. coli found in wastewater (averaging 1.5E + 08 CFU/100 mL) than in ambient water (averaging 3.3E + 03 CFU/100 mL). e observed elevations in E. coli in sewer water and WWTP influent could be due to the proximity of fecal contamination from humans and animals, as reported previously [20,21,[45][46][47][48].

Total and E. coli ESBL and KPC Abundance Varied across
Environmental Water Sources. Each water sample was screened for total ESBL and KPC resistance to estimate levels in sampled bacteria. e abundances (absolute counts) and relative percentages (resistant E. coli or total detected bacteria divided by total E. coli or total detected bacteria) of ESBL and suspected KPC bacteria are depicted by water source in Table 1 and Figure 1. ESBL and suspected KPC resistance were present in all water sources tested for total detected bacteria. Sewer water (2.5E + 06 CFU/100 mL) and WWTP influent (3.5E + 06 CFU/100 mL) together had a 2146-fold higher (p < 0.0001) ESBL-positive bacterial abundance than the ambient sources, which were 1.4E + 03 CFU/ 100 mL (WWTP-effluent) and 1.4E + 03 CFU/100 mL (surface water). Suspected KPC bacterial abundance in wastewater was 8847-fold higher (p < 0.0001) (3.4E + 06 CFU/ 100 mL WWTP-influent, 3.2E + 06 CFU/100 mL sewer water) than that in ambient water (5.8E + 02 CFU/100 mL surface water, 1.7E + 02 CFU/100 mL WWTP-effluent). e relative percentage of ESBL in total detected bacteria ranged from 0.93% (WWTP-effluent) to 6.6% (surface water) and the relative percentage of suspected KPC ranged from 0.11% (WWTP-effluent) to 2.7% (surface water) (Figure 1(a)).
Abundance of ESBL E. coli in wastewater (3.1E + 05 CFU/ 100 mL sewer water, 2.3E + 05 CFU/100 mL WWTP-influent) was 23318-fold higher (p < 0.01) than that in ambient water (1.0E + 01 CFU/100 mL WWTP-effluent, 5.0E + 01 CFU/ 100 mL surface water). Similarly, abundance of suspected KPC E. coli in wastewater (2.1E + 05 CFU/100 mL sewer water, 7.8E + 04 CFU/100 mL WWTP-influent) was higher, though not statistically significant than suspected KPC E. coli abundance in ambient water (E. coli in WWTP effluent and surface water were below the limit of detection). Relative percentages of ESBL-positive E. coli ranged from 0.28% in WWTP effluent to 1.7% in surface water, and suspected KPCpositive E. coli was below detectable limits in WWTP effluent and surface water and greatest in sewer water (0.28%) ( Figure 1(b)). E. coli represented 8.9% of total ESBL and 4.2% of total suspected KPC abundance of all bacteria screened.
Abundance of drug-resistant total detected bacteria and E. coli decreased between wastewater and ambient water samples and specifically between WWTP-influent and WWTP-effluent samples. e relative percentage of KPC also decreased between WWTP-influent and WWTPeffluent sources in both E. coli and total detected bacteria. It is noteworthy that other wastewater treatment studies have reported increased relative percentage of KPC-producing bacteria following wastewater treatment [49][50][51]. e sludge found in many WWTPs is thought to be an antibiotic resistance gene reservoir [52][53][54], and pressures put onto bacteria during the treatment process can include contact with residual antibiotics [55], driving the transfer and development of resistance. Consequently, the factors contributing to the reduced abundances of suspected KPC E. coli between WWTP influent and WWTP effluent in this plant warrant investigation for application to other water treatment systems.
Surface water had the greatest relative percentage of ESBL E. coli and the greatest relative percentage of ESBL and suspected KPC-positive total detected bacteria. e proximity of surface water sampling sites to human recreational and agricultural inputs may explain this trend. Supporting this explanation, multiple studies have reported a transfer of ESBL and KPC-expressing bacteria from people and animals into water systems [11,56,57].

Phenotype and Genotype Profiles Moderately Separate E.
coli Isolates by Source. Previous studies have inspected ESBL and KPC gene reservoirs to understand their spread through environmental waters [12,58,59], yet additional research is needed to elucidate patterns in E. coli resistance based on sampling location and date. 70 E. coli isolates with confirmed identity by MALDI-TOF were examined using a PhenePlate assay, which uses patterns in carbon source metabolism to describe clonal relatedness across bacterial isolates. e dendrogram of clonal relatedness for these selected E. coli isolates is shown in Figure 2. PhenePlate analysis generated a dendrogram showing 3 major branches of clonal relatedness and was able to separate 38 functionally distinct isolates. Branch 1 contained isolates exclusively from wastewater sources and branches 2 and 3 included bacteria from both waste water and ambient water locations. PhenePlate analysis was previously applied to separate E. coli by site, whereby this approach exhibited varying degrees of sensitivity [60,61]. Each branch in Figure 2 contained isolates collected from multiple sampling dates, though sample dates do show a small degree of clustering.
ese results suggest functional differences in carbon source metabolism alone are not sensitive enough to clonally distinguish environmental E. coli by collection site and date.
In an attempt to further distinguish the E. coli, each isolate examined with PhP-RE was profiled for the presence of three commonly expressed genes conferring ESBL and KPC resistance, blaOXA, blaCTX-M, and blaTEM through PCR (expression profiles across isolates are shown in Figure 2). All PCR-negative controls were determined to be free of contamination, and all positive controls displayed the expected band size patterns (data not shown). High prevalences of blaOXA, blaCTX-M, and blaTEM have been reported in clinically resistant strains of E. coli in Colorado [62]. Dendrogram branching patterns show blaCTX-M and blaTEM across all sample types and collection dates. Other studies in Turkey and Austria, which examined AMR in waterborne E. coli before and after wastewater treatment, similarly observed that blaCTX-M was the most abundant ESBL-associated gene among environmental E. coli [63,64]. e presence of blaCTX-M was independent of water source and was not indicative of clonal relatedness among isolates [65,66]. e apparent prevalence of blaCTX-M across 3 branch points and water sources may also be explained by the different variants of this gene, including blaCTX-M-1, blaCTX-M-2, blaCTX-M-8, blaCTX-M-9, and blaCTX-M-25, which are present in environmental, human, and animal sources [67], but were not distinguished during PCR analysis herein. blaOXA was only detected in wastewater isolates collected during April 2017, and thus merits further attention as a marker that could distinguish antimicrobial E. coli between wastewater and ambient water sources. is is supported by previous investigations, where blaOXA presence was found to differentiate beta-lactam-resistant E. coli across different water sources [36,68]. A greater understanding of E. coli ESBL resistance distinctions could be valuable in WWTP design and could help to direct wastewater treatment to continue elimination of blaOXA resistance from bacterial populations. Furthermore, despite some separation of ambient and wastewater samples, the integrated PhenePlate and PCR analysis was not sensitive enough to distinguish wastewater E. coli from each other, suggesting that genotypic and phenotypic variables other than carbon source metabolism and presence of common beta-lactam genes are needed to better characterize clonal relationships.

Broth Microdilution-Expanded E. coli-Resistant Phenotypes from Environmental Waters.
To cross-validate the ESBL and suspected KPC-resistant phenotypes detected during spread plating of E. coli, and to understand if screening for additional classes of AMR could better distinguish isolates, a broth microdilution assay was performed on a subset of the MALDI-TOF-confirmed E. coli isolates. Given the higher abundance of ESBL and suspected KPC E. coli detected in wastewater compared with ambient water, sewer water (n � 17) and WWTP-influent (n � 10) samples were preferentially screened in this assay. Prior to broth microdilution testing, all of the negative control samples incubated along with the inoculated E. coli isolates were determined to be free of bacterial growth. Results from broth microdilution assay are shown in Table 2, which displays percent drug resistance of E. coli by sample source. Table S1 provides a breakdown of resistance profiles for each isolate examined.
Diversity of drug resistance was seen between and within water sources (    Figure 2: Clonal relatedness of E. coli isolates (n � 70) from environmental sources using a functional biochemical analysis and PCR verification of beta-lactam resistance gene presence. Identity "ID" levels are to the left of the dotted line, which represents a sensitivity of 0.975. Branches to the right of this line represent isolates that are too similar to distinguish. Bold numbers on branches (1, 2, 3) refer to three major clusters of isolates identifiable from the Pheneplate assay. Isolate numbers represent the unique identifiers assigned to each E. coli isolate throughout this study. WWTP � water treatment plant. Date � isolation date. ESBL � extended-spectrum beta-lactamase and KPC: Klebsiella pneumoniae carbapenamase. OXA, CTM-M, and TEM were the three beta-lactamase genes assessed with PCR, where "+": gene present and "neg": gene absent.
patterns across isolates may reflect multiple, independent horizontal gene transfers along isolate lineages [69]. is reasoning is supported in past studies that reported an escalation in the rate of horizontal gene transfer among E. coli over the last few decades, including genes responsible for ESBL and other resistance mechanisms [70,71]. Broth microdilution is the CLSI-approved and conventional method used to screen for AMR in medical settings [72]. Broth microdilution results supported that screening E. coli for resistance to multiple drug classes can improve the capacity to distinguish closely related isolates from each other. e successful application of this assay to environmental samples collected herein emphasizes its utility as part of a harmonized surveillance approach for identifying shared relatedness of AMR in people, animals, and the environment.

Whole-Genome Sequencing for Isolate Resistance Profiles and the Spatial and Temporal Relationship of Resistance
Genes. While the integrated PhenePlate and PCR analysis supported differences in gene presence between wastewater and ambient samples, it was not sensitive to distinguish individual wastewater isolates from each other. Given the large variability in the antibiotic resistance profiles of wastewater E. coli observed during broth microdilution, whole-genome sequencing was performed on a subset (n � 21) of wastewater isolates. e objective of whole-genome sequencing was to evaluate if deeper sequencing could reveal spatial and temporal differences in resistance patterns among these closely related E. coli. All blank media and DNA extraction reagents were confirmed to be free of contamination (data not shown), supporting that the downstream analysis was sensitive to differences across E. coli isolates. e results depicted in Figure 3 categorizes E. coli by sample type, date, and gene classes. A comprehensive table including each gene profiled is included in Table S2.
Of the 185 genes screened, 154 were identified across all isolates and included 66 that conferred resistance to betalactams, 18 to aminoglycosides, 7 to fluoroquinolones, 1 to chloramphenicol, 7 to peptides, 2 to sulfonamides, 2 to diaminopyrimidines, 12 to efflux pump regulators, 36 to efflux pumps, and 9 miscellaneous resistance genes involved in the inactivation and alternation of bacterial enzymes. is genetic profiling cross-validated and expanded the information available for E. coli isolates with AMR profiles established through ESBL/KPC spread plating and PCR.
Whole-genome sequencing revealed differences across wastewater E. coli isolates, most notably in beta-lactam genes that were dependent on water sample source and collection date. Across water samples, 51 beta-lactam resistance genes (broken down as 29 sewer water and 22 WWTP influent) distinguished sewer water and WWTP-influent isolates.
ese included multiple variants of tem and ges genes unique to sewer water isolates and cmy and ctx-m genes unique to WWTP-influent isolates. When examining sewer water E. coli across sampling dates, 43 beta-lactam genes were unique to one sampling date, including 37 among 6/29/16 isolates, 1 among isolates dated 11/7/16, and 5 among isolates dated 4/ 12/17, and included multiple tem, ges, and ctx-m gene variants. For WWTP-influent E. coli, 25 beta-lactam resistance genes were specific to sample collection dates. ere were 10 cmy variants in samples dated 9/7/16, 6 ctx-m variants in samples dated 11/7/16, and 9 genes, primarily tem  [73][74][75], emphasizing that they should be closely monitored for spread into the environment. ese results suggest the potential connection between AMR gene presence and time of year and support that temporal screening of AMR genes should be considered in future surveillance programs.
Furthermore, whole-genome sequencing provided additional insight into whether E. coli that grew on KPC plates were KPC-producing isolates. Although no carbapenemaseencoding genes, including ndm-1, oxa-48, and vim family genes [76][77][78], were detected in isolates, E. coli harbored multiple efflux pump genes. Efflux pump genes (35 sewer water and 31 WWTP influent) and 12 efflux pump regulator genes (12 sewer water and 12 WWTP-influent) were detected across E. coli. Efflux pumps represent one mechanism of resistance to carbapenem drugs [27] and provide supporting rational for why isolates may have tested positive for KPC despite the lack of carbapenemase-encoding genes. It is possible that these E. coli isolates contain KPC gene variants that are not currently recognized by standard AMR gene databases, and therefore highlights the need for expanded KPC resistance gene indexing.
is study provides a framework for analysis by which new investigations can enhance the statistical power of studies to characterize AMR in other geographic locations, and with potential utility across other matrices for testing, such as human and animal gastrointestinal tracts, agricultural systems, and food production systems. Screening samples for AMR via standardized detection and monitoring will improve needed comparisons between surveillance programs [79].

Conclusions
is study illustrates the landscape of E. coli AMR across various water sources in Northern Colorado and utilizes an array of methodologies for AMR validation. Total detected bacteria and E. coli abundances were shown to decrease through the water reclamation process, validating the effective clearance of microbes from human water sources by the wastewater treatment plant. e decreased E. coli abundance in ambient water was coupled with an increased relative percentage of ESBL/suspected KPC resistance in total detected bacteria and ESBL E. coli in surface water. e prevalence of ESBL/KPC-resistant E. coli in surface water warrants deeper evaluation of potential antimicrobial inputs to rivers.
Broth microdilution further characterized drug resistance in isolates across water sources and validated E. coli ESBL/ KPC phenotypes. Identification of beta-lactam resistance in E. coli that initially tested negative for ESBL/suspected KPC resistance supports that a multi-platform screening approach may be necessary for accurate characterization of AMR landscapes in environmental water sources. e increased prevalence of multidrug-resistant isolates in WWTP influent as compared with sewer water illustrates how converging water sources can increase levels of drug-resistant in organisms entering wastewater treatment plants.
An integrated PhenePlate/PCR analysis was applied to understand the relationships between water types and diverse E. coli AMR profiles with respect to sampling location and date. PhenePlate analysis alone could not distinguish wastewater and ambient water E. coli and suggests that this single measurement of carbon source metabolism is not sensitive to distinguish some environmental water E. coli from each other. However, the coupling of PCR for common beta-lactam resistance genes, such as blaOXA, does support its use as a potential integrated test to distinguish waste and ambient water E. coli.
Whole-genome sequencing applied to wastewater E. coli validated the resistance phenotypes and genotypes observed in the ESBL/suspected KPC screening, broth microdilution, and PCR analyses, and was effective at separating isolates both by sample type (sewage versus WWTP-influent) and by collection date. Sequencing results supported that resistance to beta-lactam and aminoglycoside antimicrobials are major contributors to these geographic and temporal differences. Additionally, the clustering of isolates by source and collection date supports the integrity of bacterial samples and the unlikelihood of sample contamination. ese results warrant investigation into the development of applications and more sensitive screenings of beta-lactams and aminoglycosides that can both predict and understand how antimicrobial resistance landscapes can change within and across environments over time, as well as from waste and ambient water sources. e sample sizes used to conduct research on these field isolates were small with respect to seasonality, yet the crossvalidation of the methodologies reported herein illustrates the depth by which multiple analysis platforms could be integrated to establish AMR profiles of E. coli across environmental water types. is study provides a framework by which investigators can standardize analyses, in larger, statistically powered studies, and specifically to characterize AMR landscapes across geographically diverse locations. e integrated use of these assays also has relevance to other matrices including human and animal gastrointestinal tracts, agriculture, and food production systems.
Data Availability e whole-genome sequences of all 23 E. coli isolates were submitted to the NCBI Sequence Read Archive (NCBI SRA). However, only 22 of them have been assigned with SRA numbers. Isolate NCBI SRA identification numbers are presented in Supplementary Dataset 1.

Disclosure
Hannah B. Haberecht and Nora Jean Nealon are co-first authors.

Conflicts of Interest
e authors declare that they have no conflicts of interest.

Authors' Contributions
EPR, SV, and JS conceived the overall study design and secured funding support. LM and FS organized and coordinated water collection schemes and assisted in sample collection from the Drake Water Reclamation Facility and Poudre River. HBH, CR, JRG, AVH, and RCO performed field sampling in Northern Colorado waters and bacteria isolations. SV, HBH, JRG, AVH, HMI, and RCO performed data analysis and results interpretations for data in Tables 1  and 2 and Figures 1 and 2. Whole-genome sequencing and data analysis of isolates in Figure 3 were completed by HBH, NJN, LA, and JS. HBH, NJN, JRG, AVH, RCO, SV, and EPR wrote the manuscript. HBH and NJN contributed equally to the preparation of this manuscript. Acknowledgments e authors thank Megan Dietz and Adriana Romero for their contributions in project and protocol coordination for field sampling and Colette Worcester for technical laboratory support. is research was supported by awards and funding from the Colorado State University One Health Institute and the CSU Water Center and the National Institutes of Health (T32) Predoctoral Training Fellowship (#5T32OD012201-05). Figure S1: a map of Fort Collins, Colorado, and the surrounding area with sampling locations of environmental waters. Table S1: AMR profiles of the individual E. coli isolates examined with broth microdilution.