Antibiotic resistance among Escherichia coli and Salmonella isolated from dairy cattle feces in Texas

In several developing countries, studies on antimicrobial resistance among bacteria from food animals are rare mostly because of under-resourced laboratories. The objective of this study was to develop and field-test a low cost protocol to estimate the isolate- and sample-level prevalence of resistance to critically important antibiotics among Escherichia coli and Salmonella isolated from dairy cattle feces. Using a predesigned protocol, fecal samples were collected to isolate non-type-specific E. coli and Salmonella using selective media without antibiotic supplements. Besides, samples were screened for E. coli and Salmonella isolates not susceptible to third-generation cephalosporins and quinolones using selective media supplemented with cefotaxime (1.0 μg/mL) and ciprofloxacine (0.5 μg/mL), respectively. All bacterial isolates were further tested for antibiotic susceptibility using disk diffusion. Bacterial isolates not susceptible to third-generation cephalosporins were tested for extended spectrum beta-lactamase (ESBL) phenotype using the combination disk test. Molecular methods were performed on selected bacterial isolates to identify and distinguish genetic determinants associated with the observed phenotypes. Among 85 non-type-specific E. coli isolated from MacConkey agar without antibiotics, the isolate-level prevalence of resistance to tetracycline was the highest (8.2%). Among 37 E. coli recovered from MacConkey agar with cefotaxime, 56.8% were resistant ceftriaxone. Among 22 E. coli isolates recovered from MacConkey agar with ciprofloxacin, 77.3% and 54.5% were resistant to nalidixic acid and ciprofloxacin, respectively. Sixteen Salmonella were isolated and only one demonstrated any resistance (i.e., single resistance to streptomycin). Among E. coli isolates not susceptible to ceftriaxone, an AmpC phenotype was more common than an ESBL phenotype (29 versus 10 isolates, respectively). Whole genome sequencing showed that phenotypic profiles of antibiotic resistance detected were generally substantiated by genotypic profiles. The tested protocol is suited to detecting and estimating prevalence of antimicrobial resistance in bacteria isolated from food animal feces in resource-limited laboratories in the developing world.


Introduction
Monitoring of the emergence, spread, and changes in levels of antimicrobial resistant bacteria along the food chain is needed to inform and guide integrated strategies for combating antimicrobial resistance [1]. In most cases, surveillance systems of antibiotic resistant bacteria in food animals target pathogenic bacteria, such as Salmonella and Campylobacter as well as indicator bacteria, such as E. coli and Enterococcus spp. After their isolation from samples, genus/ species confirmation, and subtyping when necessary, the bacteria of interest are tested for susceptibility to a select number of antibiotics using a standard phenotyping method of choice. Even though the broth microdilution method is preferred in many surveillance systems and in research [2][3][4], other less technology intensive methods of antibiotic susceptibility testing, such as disk diffusion, may also be used [1,5]. After phenotypic antibiotic susceptibility testing, genetic characterization can be done through the detection of targeted genes or through whole genome sequencing (WGS). Additionally, the selective culture and detection of bacteria with rare antimicrobial resistance (AMR) mechanisms is highly recommended [1].
In many developed countries, AMR surveillance systems are replacing phenotypic antimicrobial susceptibility testing with WGS [6,7]. In many developing countries, however, surveillance systems are still not well established and the startup costs associated with equipment acquisition and maintenance can be prohibitive in these countries [8]. One of the main reasons impeding the establishment of AMR surveillance systems in these countries is the lack of sufficient resources needed for establishment, and then sustainment of the surveillance systems [8,9]. An inexpensive and reliable protocol that can generate sufficient high quality and reproducible information of the burden of antibiotic resistance among isolated bacteria would be one of the most helpful solutions for establishment of an AMR surveillance system in situation where resources are limited. Studies have demonstrated that the disk diffusion method is a cost effective method that can generate results comparable to other phenotypic methods of antibiotic susceptibility testing, such as the broth microdilution or the agar dilution methods [10]. This method can be used efficiently to establish phenotypic antimicrobial resistance profiles and to detect mechanisms of resistance such as the production of extended spectrum beta-lactamases (ESBLs) among bacterial isolates.
We designed a protocol that uses the disk diffusion method to determine the isolate-level prevalence of resistant to various antibiotics and the sample-level prevalence of any bacteria not susceptible to third-generation cephalosporin or quinolone antibiotics. Additionally, the protocol described herein was designed to estimate the proportion of bacteria resistant to third-generation cephalosporins that produce either ESBL or AmpC enzymes. Afterward, the developed protocol was field-tested on fecal samples collected from dairy cattle to estimate isolate-level and sample-level prevalence of AMR among E. coli and isolate-level and sample-level prevalence of AMR among Salmonella.

Sample collection
Using a convenience sampling scheme, we collected 85 freshly voided fecal samples from dairy cattle of different age groups at a dairy farm located near Lubbock, Texas. Fecal samples were aseptically collected into sterilized polypropylene specimen containers then kept on wet ice and transported to a microbiology laboratory at Texas Tech University. Fecal samples were processed using a protocol available at Protocols.io [dx.doi.org/10.17504/protocols.io. brjkm4kw].
As fecal samples were collected from the pen-floor, there was no interaction with vertebrate animals, consequently an approval from an Institutional Animal Care and Use Committee wasn't needed.

Isolation of bacteria from fecal samples
In the laboratory, 10g of each fecal sample was weighted in a 710mL Whirl Pak 1 bag (Whirl-Pak, Madison, Wisconsin) and 90mL of buffered peptone water (Becton Dickinson, New Jersey, United States) was added. The mixture was placed in a commercial stomacher for 2 minutes at 230 rpm. Thereafter, the mixture was incubated at 42˚C overnight prior to isolation of E. coli and Salmonella.
Isolation and identification of Escherichia coli. From each overnight non-selective enrichment, a 10μL loopful was streaked onto MacConkey agar (MAC, Hardy Diagnostics, California, United States) to isolate non-type-specific E. coli(NTS E. coli), meanwhile, another 10 μL loopful was streaked onto MAC supplemented with 1.0 μg/mL of cefotaxime (MAC +CTX) to screen for E. coli resistant to third-generation cephalosporins (3GCr E. coli). An additional 10 μL loopful was streaked onto MAC supplemented with 0.5μg/mL of ciprofloxacin (MAC+CIP) to screen for E. coli not susceptible to quinolones (Qr E. coli). All three types of MacConkey agar plate types were incubated at 37˚C overnight. Following the incubation, agar plates were inspected to identify growth of colonies with typical morphology of E. coli (i.e., pink, convex, circular and dry colonies with a surrounding pink zone). From each type of MacConkey agar plate, one typical colony was selected and re-streaked onto a similar Mac-Conkey agar plate type for isolation of pure colonies. All well isolated presumptive E. coli were tested for indole production using an indole spot test (Hardy Diagnostics, California, United States) and were confirmed as E. coli by detection of the wecA gene using a real time polymerase chain reaction (RT-PCR).
Isolation and identification of Salmonella. One mL of each overnight non-selective enrichment was transferred into 9mL of Rappaport-Vassiliadis Salmonella broth (Hardy Diagnostics, California, United States) and another 1mL was transferred into 9mL of tetrathionate broth (Hardy Diagnostics, California, United States). Both inoculated broths were incubated at 42˚C overnight. After incubation, a 10μL loopful of each broth was streaked onto brilliant green sulfa agar (BGS, Becton Dickson, New Jersey, United States) and onto xylose lysine deoxycholate agar (XLD, Hardy Diagnostics, California, United States) to isolate Salmonella. In addition, 10 μL loopful of each broth was streaked onto BGS and onto XLD agar plates, each supplemented with 1μg/mL of cefotaxime (BGS+CTX and XLD+CTX) to screen for Salmonella resistant to third-generation cephalosporins. Another 10μL loopful of each broth was streaked onto BGS and XLD agar plates both supplemented with 0.5μg/mL of ciprofloxacin (BGS+CIP and XLD+CIP) to screen for Salmonella not susceptible to quinolones. All agar plates were incubated at 37˚C overnight. Following incubation, agar plates were inspected for growth of colonies with morphology typical of Salmonella (i.e. pink, circular, dry, convex colonies on BGS; black, circular convex colonies on XLD). From each type of agar plate, a single typical colony was selected to be re-streaked onto the same type of agar plate for isolation of pure colonies. All presumptive Salmonella were tested for production of H 2 S gas, dextrose fermentation and decarboxylation reaction using lysine iron agar (Hardy Diagnostics, California, United States) and were confirmed to be Salmonella by detection of the ttrC gene using RT-PCR.

Antibiotic susceptibility testing
All isolates confirmed as E. coli and Salmonella were tested for susceptibility to a panel of 12 antibiotics using the disk diffusion method in accordance to guidelines of the Clinical Laboratory Standard Institute (CLSI) [11]. The antibiotics and concentration in each disk were amoxicillin-clavulanic acid 20/10μg(AMC), ampicillin 10μg (AMP), azithromycin 15μg (AZI), cefoxitin 30μg (FOX), ceftriaxone 30μg (CRO), chloramphenicol 30μg (CHL), ciprofloxacin 5μg (CIP), colistin 10μg (COL), meropenem 10μg (MER), nalidixic acid 30μg (NAL), streptomycin 10μg (STR), and tetracycline 30μg (TET). Inhibition zone diameters around the antibiotic-impregnated disks were measured in mm and rounded to the closest integer before being compared to the CLSI clinical break points in order to classify each bacterial isolate as resistant, intermediate or susceptible [11]. Because there was no CLSI standard inhibition zone diameters for colistin, these data were interpreted in accordance with a study conducted by Galani and collaborators [12].
Based on previous investigation of antimicrobial resistance in the region, all bacterial isolates not susceptible (i.e., intermediate and resistant) to a third-generation cephalosporin (ceftriaxone) were expected to have a phenotype reflecting ESBL-or AmpC-production. Suspected ESBL-or AmpC beta-lactamase-producing bacteria were discriminated by the combination disk test (CDT) according to CLSI guidelines [11] using a second panel of 12 antibiotic-impregnated disks. In addition to cefotaxime 30μg (CTX), cefotaxime-clavulanic acid 30/ 10μg (CTX-CLA), ceftazidime 30μg (CAZ), and ceftazidime-clavulanic acid 30/10μg (CAZ--CLA), the 4 antibiotic disks required by the CDT method, the second panel of antibiotics included amikacin 30μg (AMK), cefazolin 30μg (CFZ), cefepime 30μg, (FEP), fosfomycin 200μg (FOS), gentamicin 10μg (GEN), imipenem 10μg (IMP), sulfisoxazole 300μg (SSS), and trimethoprim/sulfamethoxazole 1.25/23.75μg (SXT). A bacterial isolate was classified to have a phenotype of ESBL-production when the absolute difference between the inhibition zone diameter around ceftazidime (coded as resistant) versus ceftazidime-clavulanic acid (CLA) and/or around cefotaxime (coded as resistant) versus cefotaxime-clavulanic acid was equal to or greater than 5mm. An isolate was classified to have a phenotype of AmpC beta-lactamase production when the difference between the inhibition zone diameter around ceftazidime (coded as resistant) versus ceftazidime-clavulanic acid and/or around cefotaxime (coded as resistant) versus cefotaxime-clavulanic acid was less than 5mm [11].
E. coli ATCC 25922 and Klebsiella pneumoniae ATCC 700603 were used as quality control strains.

Polymerase chain reaction to detect genes encoding for beta-lactamases
All isolates confirmed as exhibiting either a phenotype suggestive of ESBL or else AmpC betalactamase-production were subjected to DNA extraction using a boiling preparation method. The extracted DNA was used as a template to detect the family of bla CTX-M genes (genes encoding for ESBL production) or bla CMY-2 gene (gene encoding for AmpC beta-lactamase production) by conventional PCR (cPCR), using previously published primers [13].

Whole genome sequencing (WGS)
Following phenotypic AMR characterization, 24 E. coli isolates classified as either ESBL or AmpC phenotypes, resistant to NAL or reduced susceptibility to CIP were selected for WGS. The DNA was extracted using a commercial DNA extraction kit (Qiagen, Venlo, Netherland), libraries were prepared using the Nextera XT DBA library preparation kit (Illumina, California, United States) and the sequencing was performed using an Illumina Miseq (Illumina, California, United States). Generated raw reads (fastq files) were assembled using SPAdes 3.9 on the Center for Genomic Epidemiology platform.

Data analysis
The prevalence proportion (expressed as percentage) of bacteria resistant to each tested antibiotic was determined and confidence intervals were calculated as 95% binomial proportions representing Wilson intervals using R.3.0. software. The proportion (expressed as percentages) of samples with bacteria not susceptible (i.e., resistant or intermediate) to a third-generation cephalosporin (ceftriaxone) or else to quinolones (nalidixic acid and/or ciprofloxacin) was calculated by dividing the number of samples with non-susceptible bacteria by the total number of samples collected (i.e., only those isolates screened on media with antibiotics were used to calculate the sample-level percentages). Whole genome sequencing data were analyzed using various bioinformatics tools found on the website of the Center for Genomic Epidemiology (https://cge.cbs.dtu.dk/services/), including Resfinder3.0 that detect mobilizable genes and chromosomal mutations conferring antibiotic resistance in bacteria.

Isolated bacteria
In total, NTS E. coli were recovered from all 85 fecal samples. The recovery rate of Salmonella and of either bacterial species (i.e., E. coli or Salmonella enterica) presumptively resistant to third-generation cephalosporins or else resistant/reduced susceptibility to quinolones was lower (Table 1).

Antibiotic resistance among isolated bacteria
Antibiotic susceptibility testing of isolated bacteria showed that resistance to antibiotics was rare among NTS E. coli isolated on MAC when compared to presumptive 3GCr E. coli screened on MAC+CTX or else presumptive Qr E. coli screened on MAC+CIP. Isolate-level prevalence of resistance to tetracycline (8.2%) was the highest among NTS E. coli isolates while resistance to cefoxitin, colistin, meropenem, ceftriaxone, and ciprofloxacin were completely absent among these bacterial isolates (Table 2).
Among the 37 presumptive 3GCr E. coli screened on MAC+CTX, all isolates were resistant to ampicillin (100%), 21 isolates were resistant to ceftriaxone (55.8%), and none of the isolates was resistant to meropenem (Table 3). In total, 36 out of the 37 presumptive 3GCr E. coli were not susceptible (i.e., either resistant or intermediate) to a third-generation cephalosporin (CRO). These isolates were from 36 out of the 85 collected samples. The sample-level prevalence of E. coli non-susceptible to third-generation cephalosporin was calculated to be 42.3% (36/85) with a 95% confidence interval of 32.4% -53.0%.
Most presumptive Qr E. coli screened on MAC+CIP were resistant to tetracycline (90.9%), meanwhile 17 isolates were resistant to nalidixic acid (77.3%), 12 isolates were resistant to ciprofloxacin (54.5%), and none of the isolates was resistant to colistin or meropenem ( Among Salmonella isolated on culture media without antibiotics, resistance to antibiotics among Salmonella isolated on culture media without antibiotics was very low yielding a single isolate that was resistant to streptomycin only. The sole Salmonella isolated on XLD+CTX was confirmed to be resistant to ceftriaxone.

Phenotypic and genotypic detection of ESBL-and AmpC-producing bacteria
The sole Salmonella isolated from XLD+CTX plates, exhibited the phenotype of an AmpC beta-lactamase producer; later, this was confirmed by the presence of bla CMY-2 gene using cPCR.   A total of 40 E. coli isolates were found to be resistant or intermediate to ceftriaxone. Thirty-six of them were isolated from MAC+CTX and four were isolated from MAC+CIP plates. The second panel of antibiotics showed that these isolates were resistant to at least one of the third-generation cephalosporins tested; furthermore, all of them were resistant to cefazolin. Only a very few of these E. coli isolates were resistant to sulfisoxazole and trimethoprim/ sulfamethoxazole ( Table 5).
The phenotypic combination disk test detected 19 E. coli isolates with a phenotype indicating ESBL-production, 20 isolates with a phenotype indicating AmpC beta-lactamase-production and a single isolate that was not confirmed to be resistant to either cefotaxime or ceftazidime. On the other hand, molecular cPCR detected genes encoding for ESBL production (bla CTX-M-1 or bla CTX-M-9 family genes) in 10 isolates (25.0%), the gene encoding for AmpC beta-lactamase production (i.e., bla CMY-2 ) in 29 isolates (72.5%) and none of these genes in one isolate (2.5%).

Whole genome sequencing of select E. coli isolates
Whole genome sequencing of 24 E. coli isolates showed that the gene tet(A) encoding for a tetracycline efflux pump, was present in 70.8% of sequenced isolates. In general, all the detected genes were in accordance with the phenotypic antibiotic resistance observed in each of the isolate tested. Some exception included the presence of the aph (3')-Ia gene that purportedly confers resistance to aminoglycosides, though in our case, in isolates phenotypically susceptible to these antibiotics. The mutation gyrA[87:D-Y] that confers resistance to quinolone antibiotics was observed in one isolate exhibiting no phenotypic resistance to either nalidixic acid or ciprofloxacin. The presence of the mef(B) gene that encodes for resistance to macrolides was detected in one isolate susceptible to azithromycin. Finally, the mutation pmrB[161: V-G] that confers resistance to polymixins was observed in one isolate that was phenotypically susceptible to colistin (Fig 1).
Finally, WGS revealed that E. coli isolates resistant to nalidixic acid typically had only a single point mutation in the gyrA gene or else harbored a plasmid-mediated quinolone-resistance gene (qnr). All isolates resistant to both nalidixic acid and ciprofloxacin had a mutation in both gyrA and parE or parC genes of the quinolone resistance determining region (QRDR).
All sequencing data are accessible at the NCBI Sequence Reads Archive through the accession number PRJNA690823.

Discussion
This study was conducted to field-test a cost-effective and highly valid protocol that could be used to determine the status of antibiotic resistance among E. coli and Salmonella isolated from food-producing animals, especially where laboratory resources are limited. The protocol used in this study was inspired by different guidelines for antimicrobial resistance detection, including guidelines from the European Food Safety Authority(EFSA) [14], the U.S National Antimicrobial Resistance Monitoring System (NARMS) [2], the European Union Reference Laboratory for Antimicrobial Resistance (EURL-AR) [15], and the Danish Integrated Antimicrobial Resistance Monitoring and Research Program (DANMAP) [16]. Unlike the above mentioned guidelines, this protocol used the disk diffusion method instead of broth or agar dilution methods because disk diffusion is recognized as a simple and low-cost method when compared to other antibiotic susceptibility testing methods [17]. The disk diffusion method is a cost effective method known to generate reliable results that can be related to results from other phenotypic methods of antibiotic susceptibility testing (AST) [18,19]. In addition to its low cost, this method is not as complicated to perform as other AST methods and remains one of the most popular manual techniques for antibiotic susceptibility testing [10,18,20]. To increase our confidence in recommending the protocol, results of the phenotypic methods used herein were thereafter cross-referenced and validated using by results from relevant molecular methods. In regard to results obtained in this study, all NTS E. coli and Salmonella isolated on bacterial culture media without supplemented antibiotics were largely susceptible to all antibiotics tested; meanwhile, bacteria isolated on culture media with antibiotics supplemented at subbreakpoint levels tended to be resistant to more than three antibiotics. This observation provided evidence that when a bacterium acquires resistance to one antibiotic it tends also to be resistant to other antibiotics. In fact, in most of the cases, different genes encoding antibiotic resistance are known to be co-located on transmissible genetic elements such as plasmids. When a resistance plasmid is transferred to a previously susceptible bacterium, multidrug resistance can be transferred in a single conjugation event [21].
Furthermore, percentages of samples that generated presumptive 3GCr E. coli (43.5%) and presumptive 3GCr Salmonella (1.2%) were lower than the percentages of samples with presumptive 3GCr E. coli (89.1%) and presumptive 3GCr Salmonella (10.9%) reported in three beef feed lots in Nebraska [22]. In addition, similar to the study in Nebraska [22], we also noted that the number of 3GCr E. coli isolates phenotypically or genotypically confirmed to be AmpC-producers was higher than the number of isolates confirmed to be ESBLs-producers. The only identified 3GCr Salmonella was confirmed to be an AmpC beta-lactamase producer. In the U.S, resistance to third-generation cephalosporins among Salmonella from food animals has historically been largely due to the gene bla CMY-2 encoding for AmpC beta-lactamase production [23].
In the present study, nine 3GCr E. coli had a phenotype typical of ESBL-production (according to the CDT) while the cPCR showed that these isolates carried bla CMY-2 , a gene encoding for AmpC beta-lactamase production, instead of genes encoding for an ESBL. The combination disk test is reported to be an accurate phenotypic method widely used to detect ESBL-producing Enterobacteriaceae [24]. Despite its widespread use and solid reputation, the sensitivity and specificity of this method to detect ESBL-producing Enterobacteriaceae is not always 100% [25]. In fact, different authors have reported a number false positive ESBL-producing Enterobacteriaceae by the CDT in their investigations [26][27][28][29].
To mitigate misclassification of bacterial isolates as ESBL-producers using phenotypic methods, some of these authors have suggested modifications of standard methods, such as the CDT, in order to increase their efficacy in discriminating ESBL-producing bacteria from AmpC beta-lactamase-producing bacteria [26,27]. Along this line, our study illustrated that when the combination disk test is used to identify ESBLs-producing E. coli, some precautions should be taken as the test may produce several false positive E. coli producing ESBLs, especially when the results are close to 5 mm decision point. A close look at the 9 E. coli isolates falsely classified as ESBL-producers showed that the difference between inhibition zone diameters around cefotaxime with clavulanic acid and around cefotaxime was less than 5mm for all the isolates. In contrast, the difference between inhibition zone diameters around ceftazidime with clavulanic acid and around ceftazidime was equal to 5mm or slightly higher than 5mm. The later observation led to the conclusion of classifying these same isolates as ESBL-producers. Importantly, 8 of the 9 isolates were not susceptible to cefoxitin (a second-generation cephalosporin (cephamycin) antibiotic). Based on all these observations made in this study, we came up with the following rule of thumb: false positive ESBL-producing E. coli can be detected by looking at the increase in inhibition zone diameters around CAZ-CLA versus CAZ and around CTX-CLA versus CTX. False positive ESBL-producing E. coli can be identified as isolates for which inhibition zone diameters around ceftazidime is increased by 5mm or slightly higher (6mm) due to clavulanic acid while the increase of the inhibition zone diameter around cefotaxime caused by clavulanic acid is less than 5mm. When false positives are identified, we recommend using the information generated by CTX with and without CLA to classify an isolate as ESBL-or AmpC-producer. In addition, a look at bacterial isolates' susceptibility to cefoxitin (a cephamycin) or amoxicillin-clavulanic acid maybe helpful to conclude that an isolate is not an ESBL-producer but might be an AmpC-producer. It is reported that AmpC-producing Enterobacteriaceae are resistant to cephamycins such as cefoxitin while ESBL-producers are susceptible to cephamycins [30,31].
In general, WGS showed that phenotypic antibiotic resistance observed in E. coli isolates was a good indicator of the presence of genetic antibiotic resistance determinants, with only few exceptions. The observed discordance may be explained by the fact that the presence of a resistance gene doesn't always mean its expression. In addition, some genes are cryptic or even misclassified as to their primary purpose in bacterial host function. Results of WGS also showed that all isolates resistant to both NAL and CIP had mutations in both genes of the QRDR while isolates resistant to NAL alone had a single mutation in one of the target genes in QRDR, else or carried a plasmid mediated quinolone-resistance gene (i.e. qnr gene). In fact, mutations in the QRDR have been identified as the common genetic determinant conferring a higher level of resistance to quinolone antibiotics while plasmid mediated quinolone-resistance genes confer moderate resistance [32,33]. Several studies have proven that the detection of genetic determinants of antibiotic resistance by WGS accurately predicts antibiotic resistance phenotypic behavior of bacterial isolates [6,34]. High level of agreement between WGS and phenotypic antibiotic susceptibility testing have been reported [35,36].
In conclusion, this study established that bacteria resistant to antibiotics were present in dairy cattle on the study farm but at a low level. The addition of an antibiotic to the culture medium of isolation helped in detecting and later characterizing the few antibiotic resistant bacteria. The developed protocol can help to establish percentages of indicator E. coli resistant to various antibiotics, including critically important antibiotics for human medicine, at a relatively low cost and with high reliability, even in the developing world. Furthermore, cPCR and WGS both supported our phenotypic findings at a high level which increased our confidence in recommending the tested protocol for its use to establish status of AMR in food animals; specifically, when laboratory facilities are limited and financial and other resources are scarce.