Quantitative dynamics of Salmonella and E. coli in feces of feedlot cattle treated with ceftiofur and chlortetracycline

Antibiotic use in beef cattle is a risk factor for the expansion of antimicrobial-resistant Salmonella populations. However, actual changes in the quantity of Salmonella in cattle feces following antibiotic use have not been investigated. Previously, we observed an overall reduction in Salmonella prevalence in cattle feces associated with both ceftiofur crystalline-free acid (CCFA) and chlortetracycline (CTC) use; however, during the same time frame the prevalence of multidrug-resistant Salmonella increased. The purpose of this analysis was to quantify the dynamics of Salmonella using colony counting (via a spiral-plating method) and hydrolysis probe-based qPCR (TaqMan® qPCR). Additionally, we quantified antibiotic-resistant Salmonella by plating to agar containing antibiotics at Clinical & Laboratory Standards Institute breakpoint concentrations. Cattle were randomly assigned to 4 treatment groups across 16 pens in 2 replicates consisting of 88 cattle each. Fecal samples from Days 0, 4, 8, 14, 20, and 26 were subjected to quantification assays. Duplicate qPCR assays targeting the Salmonella invA gene were performed on total community DNA for 1,040 samples. Diluted fecal samples were spiral plated on plain Brilliant Green Agar (BGA) and BGA with ceftriaxone (4 μg/ml) or tetracycline (16 μg/ml). For comparison purposes, indicator non-type-specific (NTS) E. coli were also quantified by direct spiral plating. Quantity of NTS E. coli and Salmonella significantly decreased immediately following CCFA treatment. CTC treatment further decreased the quantity of Salmonella but not NTS E. coli. Effects of antibiotics on the imputed log10 quantity of Salmonella were analyzed via a multi-level mixed linear regression model. The invA gene copies decreased with CCFA treatment by approximately 2 log10 gene copies/g feces and remained low following additional CTC treatment. The quantities of tetracycline or ceftriaxone-resistant Salmonella were approximately 4 log10 CFU/g feces; however, most of the samples were under the quantification limit. The results of this study demonstrate that antibiotic use decreases the overall quantity of Salmonella in cattle feces in the short term; however, the overall quantities of antimicrobial-resistant NTS E. coli and Salmonella tend to remain at a constant level throughout.

Introduction Salmonellosis is one of the most common foodborne diseases in the United States [1]. Detection of Salmonella in retail ground beef has been low (0.4-2%) in reports from the National Antimicrobial Resistance Monitoring System (NARMS) from 2002 to 2015 [2]. However, retail ground beef is one of the primary sources of foodborne Salmonella infection for humans in the United States and multiple outbreaks have been reported due to consumption of undercooked ground beef [2,3]. The finishing period for cattle occurs in the feedlot, and this is where Salmonella can readily spread via the fecal-oral route or the environment (e.g., feed and flies) among infected and uninfected cattle residing in the same and adjacent pens [4,5]. Most often, cattle are asymptomatic carriers and no clinical signs are shown.
Salmonella contamination of meat mainly occurs at the slaughterhouse through aerosolization of fecal contaminated hides onto carcasses and from fat-shrouded lymph nodes containing Salmonella being incorporated into batches of ground beef [6][7][8][9][10][11]. Several studies have shown higher prevalence of Salmonella on hides than in feces at slaughter; therefore, hides and lymph-nodes are likely sources of carcass and ground beef product contamination, respectively [9,12,13]. Contamination of meat with antibiotic-resistant Salmonella poses a serious public health threat due to there being few treatment options for high-risk populations such as infants, children, pregnant women, and immunocompromised persons [14].
In lieu of fluoroquinolones, which may cause musculoskeletal disorders, cephalosporins are the choice of treatment for the high-risk populations listed above. Approximately 80% of cephalosporins that are sold for use in food animals are used in cattle production [15]. Since Salmonella is not considered an adulterant in raw meat by the United States Department of Agriculture, it is crucial to decrease both overall Salmonella carriage and unnecessary antibiotic use at the feedlot. This will ultimately reduce hide and carcass contamination of Salmonella overall, and especially of antibiotic-resistant Salmonella at slaughter. Ideally, such reductions will extend to lymph nodes as well. The effects of antibiotic use on the quantities of antibiotic-resistant Salmonella have not been studied systematically in feedlot cattle.
Both chlortetracycline (CTC) and the third-generation cephalosporin, ceftiofur, are antibiotics used for the treatment and control of bovine respiratory diseases (BRD) in feedlot cattle. In our previous work, we demonstrated that the prevalence of Salmonella decreased immediately during and following ceftiofur and CTC treatment; however, the proportion of multidrug resistant (MDR) Salmonella also increased [16]. One-time treatment of cattle with ceftiofur had a continuous effect on elevating the prevalence of MDR Salmonella populations until the study terminated after 26 days. The CTC treatment had and even stronger effect on increasing the proportion of MDR Salmonella than did the ceftiofur treatment. From a quantitative perspective, however, it remained unclear whether the depleted susceptible Salmonella populations were replaced by resistant Salmonella, or the quantity of resistant Salmonella stayed the same. In the current study, we have quantified both the susceptible and resistant Salmonella populations in order to clarify the effects of antibiotics.
The effects of antibiotic use on the quantity of E. coli in cattle have previously been studied [17][18][19][20] E. coli concentrations transiently [17]. However, the change in the quantity of antibiotic-resistant E. coli has not been clearly demonstrated. Data on absolute and relative changes in resistant bacteria following antibiotic treatments are necessary for quantitative risk assessment analysis for foodborne antibiotic-resistant bacteria [21]. Therefore, in the current study, we quantified E. coli in addition to Salmonella. As a continuation from our previous research concerning the prevalence of Salmonella (i.e., absence/presence and serotype), the goal of the present study was to explore the quantitative population dynamics of resistant and susceptible bacteria following antibiotic treatment in cattle. The quantity of Salmonella present in a sample-versus its mere presence or absence-is especially relevant for risk assessments as the risk of human foodborne salmonellosis is dose-dependent [22]. Tetracycline and ceftiofur/ceftriaxone were chosen as the antibiotics of primary interest, since resistance to tetracycline is common among both E. coli and Salmonella strains in feedlot cattle and often presents as co-resistance to cephalosporins [23][24][25][26].

Experimental design and sample collection
The experimental design was the same as described in previous work by this group [16,26]. Briefly, we conducted a randomized controlled longitudinal field trial during two sequential 26-day replicates in an experimental feedlot at West Texas A&M University in Canyon, Texas, USA. A high prevalence of Salmonella in cattle is commonly observed in this area [16]. The first replicate was started in early August and a second replicate in the middle of September of 2009. The cattle were purchased from a single operation and were shipped directly to the experimental feed yard one month before the trial started. It is unknown if the cattle had previously been assembled at the source. The animals were yearling steers that were predominantly of the Angus breed and were fed diets typical of regional feedlots; that is, a flaked-corned based diet with added roughage, protein, vitamins and minerals. If any cattle became sick and required antimicrobial treatments, they were excluded from the study. In each replicate, 88 steers were assigned into 8 pens (n = 11 cattle) to distribute the body weights among the pens evenly in a two-by-two factorial design with four treatment regimens, as described previously [16,26]. Four pens were assigned into each treatment group among two replicates. Across both replicates, all 11 steers received 6.6 mg/kg of CCFA (EXCEDE 1 , Zoetis Animal Health, Florham Park, NJ) treatment subcutaneously at the base of the ear in 8 pens (described as "All-CCFA & CTC" and "All-CCFA/no CTC" in this article; group metaphylaxis model), and in the remaining 8 pens, a single steer treated with CCFA on day 0 was co-housed (mixed) with 10 non-treated steers (S1 Fig). A single steer was mixed in the pen to reproduce a real feedlot situation, such that one steer is treated with CCFA for BRD and send back to its original pen. Repeated within each of the two replicates, four of the pens assigned CCFA treatments received 22 mg/kg CTC (Aureomycin 1 , chlortetracycline complex equivalent to 220.5 g/kg of chlortetracycline, Alpharma, Bridgewater, NJ). The CTC was top-dressed in feed for five consecutive days over three time periods with a one-day interval in between. The CTC feeding occurred during the period from day 4 until day 20 (S1 Fig: "All-CCFA & CTC", "1-CCFA & CTC"). The remaining 8 pens in each replicate did not receive CTC (S1 Fig: "All-CCFA/no CTC" and "1-CCFA/no CTC"). Fecal samples were collected every other day per rectum and samples from days 0, 4,8,14,20, and 26 were tested [16,26]. Fecal samples from Day 0 were collected before the treatment with CCFA and considered as baseline. Samples for culture were mixed with glycerol at a 1:1 ratio and preserved at −80˚C; samples for qPCR analysis were similarly preserved but without glycerol.
The sample size of this study was calculated based on the expected quantities of E. coli and not with Salmonella, since the original study was designed to investigate antibiotic resistant E. coli in the cattle population, where it could be estimated as 100%. Salmonella prevalence were estimated to be much smaller than that; unexpectedly, the prevalence was much higher at 70%. This meant 30% of counts were below the limit of detection which posed problems for the variance estimates and assumptions of normal residuals and was part of the motivation for the imputation approach, given we did not believe the true quantity was zero for all negative samples. That said, with a log10 transformed mean of 6 and 4 for E. coli and Salmonella, respectively, and accounting for pen-and animal-level dependencies (intracluster coefficients) of .05 and .10 for log10 CFU, respectively, we estimated a sample size per group of 16 (cluster adjusted to 40) would provide a power of 0.8 and 95% confidence for a 0.5 log10 difference. This calculation outcome does not depend on the mean of 4 versus 3.5 (Salmonella) or 6 versus 5.5 (E. coli). The

Quantification of Salmonella and NTS E. coli by colony counting
Fecal samples (500 mg) from days 0, 4, 8, 14, 20, and 26 of replicates 1 and 2 were diluted in 4.5 ml of phosphate buffered saline (PBS) at a 1:9 ratio. Diluted fecal samples were plated on BGA, BGA with 16 μg/ml of tetracycline (BGA-tet), and BGA with 4 μg/ml of ceftriaxone (BGA-cef) using an Eddy Jet 1 2 spiral plater (Neutec Group Inc, Farmingdale, NY) with the E-Mode 50 μl setting and incubated at 37˚C for 18 hours. The concentrations of antibiotics were chosen following CLSI breakpoints for tetracycline and ceftriaxone, respectively [27]. Presumptive Salmonella colonies were counted by an automated Flash & Go 1 colony counter the next day following the manufacturer's instructions (Neutec Group Inc, Farmingdale, NY).
Fecal samples also were diluted and spiral plated on MacConkey agar (MAC), MAC with tetracycline (16 μg/ml), ceftiofur (8 μg/ml), and both drugs (16 and 8 μg/ml, respectively) for NTS E. coli quantification. The concentration of ceftiofur was chosen following NARMS consensus breakpoints [28]. Previous work had shown that 99.9% of isolates grown on MAC agar from cattle feces were confirmed as NTS E. coli [29]. The plates were incubated for 18-24 hrs, colonies were counted and then back-calculated through dilution factors to CFU/g feces.

Hydrolysis probe (TaqMan 1 probe) quantitative real-time PCR (probe-qPCR) for Salmonella
Total community DNA extraction for qPCR. Total community DNA was extracted from 200mg of non-glycerol-diluted feces from days 0, 4, 8, 14, 20, and 26 of replicate 1 and 2 by the QIAamp DNA Stool Mini Kit ™ (Qiagen, Valencia, CA) in the QIAcube robot ™ (Qiagen, Valencia, CA) following the manufacturer's instructions as described previously [30]. The quality and quantity of the community DNA was estimated via the NanoDrop 1 ND-1000 UV-Vis Spectrophotometer (NanoDrop Technologies, Wilmington, DE) at wavelengths of 260 and 280 nm. DNA samples were stored at -20˚C for further genotypic analysis.
Standard curve generation. The genomic DNA extracted from Salmonella enterica subsp. enterica serovar Typhimurium ATCC 1 700720™ (ATCC 1 , Manassas, Virginia) using the QIAcube ™ robot (Qiagen, Valencia, CA) was used as a template for standard curve generation [31]. The DNA sample was serially diluted, and each standard curve reaction contained 3 μl of diluted genomic DNA, in which the final copy numbers were 1x10 6 , 1x10 5 , 1x10 4 , 1x10 3 , 1x10 2 , and 1x10 1 . Gene copy numbers were estimated following the calculation described previously [32]. The total size of the Salmonella Typhimurium ATCC 700720 strain is 4,857,450bp (GenBank: AE006468.2). The molar mass per base pair was set as 650 (g/mol)/ bp. Efficiency was calculated following the equation: Primers/probes and reactions setup. Standard curve generation and the primers and probes targeting the invA gene, carried by all Salmonella and necessary for invasion of intestinal epithelial cells [33,34], were adapted and modified from Gonzalez-Escalona et al. (Table 1) [35]. FAM was used for the 5'-Reporter dye ( Table 1). The 16s rRNA data were utilized from a previous study (Table 1) [30,36]. Primer specificity was confirmed by primerBLAST, which returned only Salmonella enterica subsp. enterica. The number of invA gene copies in Salmonella serotypes found in the study population was previously reported to be a single gene copy per genome [16,31]. Quantities were standardized via the 16s rRNA quantity in the community DNA as described previously [30].
Each reaction was composed of 10 μl of Brilliant III Ultra-Fast QPCR Master Mix with Low ROX (Agilent Technologies, Santa Clara, California), 4.6 μl of nuclease-free water, 0.4 μl of probe (0.2 μM), 1 μl each of forward and reverse primers (0.5 μM), and 3 μl of total community DNA. Thermocycler conditions for the reaction were 95˚C for 3 min (activation of master mix), and 40 cycles of 95˚C for 5 sec (denaturing), 60˚C for 10 sec. (annealing/extension) on AriaMx Real-Time PCR system (Agilent Technologies, Santa Clara, California) [31,35,37]. Data collection for fluorescence was conducted at the annealing/extension step of each cycle. The reaction plates were set up manually and each sample was run in duplicate. The assays were subjected to qPCR performance analysis following the MIQE guidelines for qPCR [38].

Probe-qPCR data analysis
After each run, qPCR data were analyzed using the AriaMx ver. 1.0 software (Agilent, Santa Clara, CA). Gene copy numbers were back-calculated to gene copies per gram of wet feces by Broad range primers to generate template for the standard curve [30] 2 Narrow range primers for sample quantification [30] https://doi.org/10.1371/journal.pone.0225697.t001 Effects of antibiotics on the quantity of taking feces loss and dilutions into account at each step of community DNA extraction. Cq derived gene copy numbers were multiplied by 583.33 for the complete back calculation considering the sample lost during DNA extraction process. Both non-standardized invA gene copies per gram wet feces and the estimates obtained by standardizing with the 16S rRNA gene were calculated and analyzed.

Statistical analysis
Analysis of qPCR and colony count data using multilevel mixed-linear regression. All standardized, non-standardized, and colony count derived quantities (CFU/g feces wet matter) were transformed to log base 10 for use as dependent variables in multiple imputation and mixed-linear regression models performed with Stata/IC 14.2 (StataCorp LLC, College Station, TX). Counts from direct plating on plain agar and antibiotic-containing agar for both Salmonella and NTS E. coli were log 10 transformed and used for dependent variables in multilevel mixed-linear regression. CCFA pen mixing (binary), CTC (binary), and days (integer) were treated as independent fixed factor variables in a 3-way full factorial model. Replicate, pen, and animal ID were potential clustering variables and included as random effects. qPCR derived log 10 transformed outcomes were rounded to the nearest whole number and converted from continuous numeric to integer data for the purpose of analysis; further, since many of the qPCR reactions were below the limit-of-quantification (LOQ) these were marked as zero. These counts were set as dependent variables, and CCFA (binary), CTC (binary), and days (integer) were included as independent fixed factor variables specified in a 3-way full factorial model. 1-CCFA mixing (0), No-CTC treatment (0), and day 0 were set as the baseline. The same clustering variables as above were used as inflation variables. The log 10 transformed counts were used for linear regression analyses of both qPCR and spiral plating.
Imputation of missing values. When the estimated quantity of the invA gene was under the LOQ they were recorded as missing values. The assumption was that samples containing Salmonella at concentrations less than the LOQ were unlikely to truly harbor zero bacteria. Originally, missing values were given a value of 1 and log 10 transformed to 0 for visualization purposes. The missing values were explored by imputation procedures in Stata/IC 14.2 with linear regression, truncated regression, and interval-censored regression to provide a censored continuous variable [39,40]. Each imputation was conducted with the full factorial model including trial replicate as a fixed effect, and a 3-way full factorial model using CCFA, CTC, and day as independent variables. The upper limit of missing values was set as 3.6 log 10 (mean and median of the invA gene copy numbers in the samples with one-missing qPCR value) for those with missing observations in truncated regression and interval-censored regression. The lower limit was set to -1. The imputed values were diagnosed for fit with the observed data [40]. A representative imputed value set was chosen for multilevel mixed-linear regression to assess the effects of antibiotic treatments by day on the quantity of Salmonella represented by invA gene copies per gram of feces. In Stata, estimation following multiple imputation can be performed; however, no programmed methods are provided to predict margins from the estimation model. Margins of treatment effects on log 10 invA gene copies per gram feces and those standardized with 16s were predicted by multilevel mixed linear regression.

Summary statistics of direct spiral plating and qPCR
A total of 386 (37.1%) samples were quantifiable for Salmonella using either spiral plating, qPCR, or both among all tested samples (S2(a) Fig). More samples (n = 324) were quantifiable by qPCR than direct plating (n = 252) (S2(a) Fig). Most of the samples that grew on antibiotic-containing agars (e.g., BGA-tet, and BGA-cef) had growth on BGA-plain; unexpectedly, 5 of these samples did not grow on BGA-plain (S2(b) Fig).
The LOQ was 2.6 log 10 per gram feces for direct spiral plating and 2.78 log 10 per gram feces for qPCR ( Table 2). The 16s rRNA gene was quantifiable in all samples.
Linear regression for log 10 invA gene copies based on the log 10 CFU counts was strongly correlated (p < 0.0001) (Fig 1). The mean quantity was higher in qPCR than on BGA-plain, and the difference between those was 0.96 log 10 (data points ranged from -1.13 to 3.46). Zero values in qPCR were imputed in the later analyses, since it was suspected that many of them were below LOQ, but not actually zero.

Quantitative dynamics of Salmonella in cattle treated with CCFA and CTC
Salmonella counts on plain BGA agar modeled using multi-level mixed linear regression are presented in Fig 2. CCFA treatment (All-CCFA group) significantly decreased the quantity of Salmonella at Day 4 (p = 0.037) and Day 8 (p < 0.0001). Without additional CTC treatments, the Salmonella quantity recovered to initial levels by Day 14 (All-CCFA / No CTC). With additional CTC treatments following CCFA, the quantity of total Salmonella remained signifi-   Salmonella; however, absent the earlier treatment with CCFA the quantity recovered near to the initial level by Day 26 (Fig 2). In

Quantitative dynamics of NTS E. coli in cattle treated with CCFA and CTC
Quantities (CFU) of NTS E. coli were also modeled using multilevel mixed linear regression and are presented in Fig 3. The NTS E. coli quantity decreased following CCFA treatment on Day 4 (Fig 3-left). In contrast to Salmonella, CTC treatment appeared to have no effect on the CFU counts of NTS E. coli. That said, the counts of tetracycline resistant E. coli were significantly increased by the CTC treatment (P<0.05). The differences in the counts between overall E. coli and tetracycline-resistant E. coli were reduced by the CTC treatment (Fig 3-bottom, blue and red line). Ceftiofur-resistant E. coli increased following the CCFA treatment. Even in the 1-CCFA & CTC groups, ceftiofur-resistant E. coli increased by Day 8 (Fig 3-right bottom).
Multilevel mixed-linear regression model of invA gene with representative imputed values. Among 1,040 samples tested by qPCR in duplicate (2,080 reactions), 571 reactions (27.5%) were detected as harboring the invA gene. The quantities of invA gene copies were standardized with the 16s rRNA gene by taking their ratio; subsequently, these were log 10 transformed (Table 3, S4 Fig).
Observations with missing values in 1,509 (72.5%) wells (i.e., those with Cq values > 40) from qPCR runs (shown in a bar as 0 in Fig 4(a)) were imputed using interval censored regression. The smallest imputed quantity after 20 imputation runs was as low as -1.95 log 10 and the largest was 3.59 log 10 (Fig 4(b), yellow bars). Marginal estimates of treatment effects on log 10  Effects of antibiotics on the quantity of Salmonella and E. coli in feedlot cattle feces invA gene copies per gram feces (Fig 5(a)) and those standardized with 16s rRNA (Fig 5(b)) were predicted by multilevel mixed linear regression. A significant decrease in imputed and observed gene copies was predicted for CTC treatment, especially on Day 8, where the animals were treated with only CTC (red) and showed a dramatic decrease. With CCFA treatment on Day 0, gene copies were also significantly decreased (orange). A slight increase in gene copies is observed following CTC treatment on Day 8 (orange) in the CCFA group. In the animals treated with CCFA on Day 0 (Green), but no CTC on Day8, only a slight decrease of gene copies was observed.

Discussion
We demonstrated in our previous work that the overall prevalence of Salmonella (presence/ absence, and by serotype) was greatly decreased by the two antibiotic treatments (CCFA and CTC) while the prevalence of MDR Salmonella increased in its place [27]. Here, we have demonstrated that the quantity of overall and MDR Salmonella follow a similar pattern to the Effects of antibiotics on the quantity of Salmonella and E. coli in feedlot cattle feces prevalence dynamics. The risk of foodborne illness for pathogens such as Salmonella has been shown to be dose-dependent [22]. Thus, reduced quantity may also reduce risk of transmission through fecal contamination of either the environment or food. Resistant Salmonella became the dominant population following the antibiotic treatments, which also corresponds with our previous work [27]. Among populations of E. coli, tetracycline-resistant bacteria and ceftiofurresistant bacteria behaved differently, where tetracycline-resistant E. coli expanded its quantity especially after CTC treatments, but ceftiofur-resistant E. coli did not increase in the presence of CTC, either subsequent to CCFA or alone.

Effects of antibiotics on the quantity of Salmonella
Although previous mouse studies have shown that antibiotic treatments cause microbiota disruptions and increase Salmonella colonization [41], in the present study the quantity of Salmonella before (Day 0) and after (Day 26) were similar comparing antibiotic treated to untreated cattle. In our study, roughly 15 cattle increased their Salmonella fecal quantity following the antibiotic treatments. Especially with CCFA treatment, roughly 5 cattle fecal CFU counts increased by 10 times through Day 26, though the sample size remains too small to derive conclusions. Not only antibiotics but also other factors such as feed contamination, high ambient temperatures, the manure-laden pen environment, and biting insects could affect Salmonella infection and quantities. Among the Salmonella that were detected in our study, resistant Salmonella have an advantage during and immediately following antibiotic exposure; as proof, tetracycline and ceftriaxone-resistant Salmonella were detected more frequently from those cattle and pens that were treated with CCFA and CTC.
It was expected that the antibiotic-resistant Salmonella population would increase in quantity following antibiotic treatment. However, no significant increases in quantity were observed after the antibiotic treatments (Fig 2). The quantities of tetracycline-and ceftriaxoneresistant Salmonella were relatively constant at around 4 log 10 CFU/g feces. However, as the study period passed, more animals with growth on BGA-tet and BGA-cef were observed; indeed, this is very consistent with earlier prevalence results [27]. Much of the resistant Salmonella population appears to have resided below the LOQ before treatment, then increased to more than the LOQ, especially following CTC treatment.

Comparison of Salmonella and NTS E. coli quantities
The quantity of NTS E. coli was suppressed by CCFA treatment but not by CTC treatment. On the other hand, the quantity of tetracycline-resistant E. coli increased almost to a quantity similar to total NTS E. coli following the CTC treatment. This suggests that tetracycline-resistant E. coli expanded and replaced the antibiotic-susceptible E. coli during the extended period of CTC treatment; therefore, the overall quantity of total NTS E. coli appeared unaffected by CTC. In contrast, for Salmonella the quantity was influenced by both CCFA and CTC treatment. CCFA treatment decreased the quantity and this was further suppressed by the CTC treatment. While tetracycline resistant E. coli dramatically replaced susceptible E. coli, the overall Salmonella population remained low after the CCFA treatment and was not replaced to the extent of E. coli. We can explain such differences between Salmonella and E. coli behavior by the vast differences of the quantity and genetic diversity and resistance pattern of the E. coli versus Salmonella populations. In previous work by Kanwar et al., which reported on the resistant E. coli population in the same samples, E. coli carrying tet(A) and tet(B) genes were commonly detected by PCR. E. coli with the tet(A) gene also were often carrying the bla CMY-2 gene, while E. coli detected with the tet(B) gene were less likely to be co-resistant with the bla CMY-2 gene [26]. NTS E. coli are known as a highly prevalent indicator bacteria and have more diverse genotypes as shown by PFGE and antibiotic resistance types in comparison to Salmonella serotypes, of which we identified only six [42].
Although we have not investigated the tetracycline resistance genes of the antibiotic-resistant E. coli that replaced the susceptible E. coli, E. coli with single tetracycline resistance may have a fitness advantage under CTC selection pressure. The resistance patterns of Salmonella detected from our prevalence study were more clonal, and most of the resistant isolates were either pan-susceptible or else of the MDR ampicillin-chloramphenicol-streptomycin-sulfisoxazol-tetracycline-ceftiofur (ACSSuT-Cef) phenotype [16]. In our previous work, prevalence of MDR Salmonella was 100% in the fecal samples that were tested after the CTC treatment on Day 14 and Day 20 [16]. It is likely that the MDR Salmonella were pre-selected by CCFA treatment and replaced the susceptible population with the subsequent CTC treatment. However, the quantity of Salmonella did not increase as dramatically as was seen in the E. coli. Even though the MDR Salmonella survived the antibiotic pressure, the fitness cost of resistance may not allow them to expand as dramatically.
Finally, we used ceftriaxone at the CLSI human clinical breakpoint of 4 μg/ml for spiral plating of Salmonella while we used the NARMS consensus breakpoint of 8 μg/ml for E. coli. We do not believe this has substantively affected the inference or comparisons that can be made from the data. Ceftriaxone is a common drug-of-choice for empirical therapy of human cases of salmonellosis. Ceftiofur is an animal-only drug and our interest was in exploring the dynamics of the indicator E. coli in response to that drug. The MIC patterns in in vitro assays, such as broth microdilution, for ceftiofur and ceftriaxone are virtually identical [26].

Quantification by qPCR vs. direct spiral plating
Although invA gene copies and CFU of Salmonella per gram of feces are not directly comparable measurements, the LOQ of qPCR and direct plating to BGA was similar, around 2.6 log 10 gene copies and CFU per gram of feces, respectively. However, among those samples detected positive by both qPCR and direct plating, the quantity of Salmonella was estimated at roughly 10 times higher in qPCR compared with direct plating. Because qPCR detects both viable and also non-viable Salmonella that might be killed by the antibiotics, it is possible that the quantity differences between qPCR and direct plating were simply due to the DNA arising from nonviable Salmonella. It is not likely that other bacterial species were detected with this invA gene unique to Salmonella [35].
In qPCR, multiple steps affect the quantity calculation, including efficiency of DNA extraction, type of template used for creating the standard curve, original fecal sample condition, and back calculation of gene copies to per gram of feces. We generated the standard curves from the whole-genome of Salmonella Typhimurium isolates, which had great reproducibility throughout the 30 individual qPCR 96-well plate runs. The total community DNA used in this study was extracted after the fecal samples were collected, having been stored at -80˚C for up to two years, and with the extracted DNA subsequently stored in the freezer at -20˚C for up to five years. On the other hand, the fecal samples that were used for the direct plating quantification had been stored at -80˚C with 50% glycerol for five years, which could weaken or degrade the bacteria and therefore decrease the counts. Based on our previous work on Salmonella prevalence, roughly 60% of samples that were detected with Salmonella through an enhanced enrichment process were quantifiable by either qPCR or direct spiral plating.

Imputation of missing values
The qPCR results from our study contained many missing values (72.9% of the total wells failed to react using qPCR). This is problematic for data analysis due to bias produced because incomplete data will not be included in the analysis. In addition, multiple linear regression assumptions include normality of residuals which is problematic with zero-inflation of observations combined with normally distributed observed counts. To overcome this, we used multiple imputation for the analysis of treatment effects on the count of invA gene copies.
It has been shown that an appropriate imputation leads to less bias and better prediction of the model [43]. In our qPCR results, an interval censored regression model was chosen between linear regression and truncated regression because we expected that most of the values were missing due to being below the LOQ. Other methods impute values within the range of observed values which was clearly inappropriate for the situation in this study. Several possibilities for values below the LOQ can be considered; 1) cattle were not infected with Salmonella (true negative), 2) susceptible Salmonella were killed by antibiotic treatments or else starting quantities of viable Salmonella were very low, therefore Ct values were higher than 40 (censored data), 3) failure of qPCR runs due to PCR inhibitors, or 4) other reasons. Except for the first reason, the negative results can be interpreted as false negatives. It is highly probable that most samples belong to scenario 2), because, in the prevalence study, only 4 animals were not detected with any Salmonella across the 6 separate sampling days (representing up to 24 / 1040 samples) [16]. The model that was created after imputation was similar to that without imputation, but with predicted values much closer to the observed data (Fig 5). This shows that imputing the missing value can led to a less biased model. While few studies utilizing imputation methods to estimate the missing values in qPCR data have been reported, more established methods are clearly needed going forward [30,[43][44][45].
Until now, the risk of antibiotic use on the quantity of antibiotic-resistant Salmonella has not been well documented. In the current study, even though antibiotics led to a transient decrease in Salmonella quantity, more steers were shedding a quantifiable number of tetracycline-and ceftriaxone-resistant Salmonella by the end of the study. Additionally, among indicator E. coli, the susceptible population that was greatly reduced by antibiotics was quickly replaced by a tetracycline-resistant population. Antibiotic-resistant Salmonella contamination of meat products can pose substantial public health risks; therefore, further longer-term explorations of the risk of antibiotic use on the quantity of Salmonella up until the age of slaughter are warranted.
Supporting information S1 Fig. Experimental design. Both replicates are combined. Treatment groups (boxes) represent a 2 � 2 factorial design of All-CCFA versus 1-CCFA and CTC versus no CTC. There were four pens per combined treatment group. Treatment with CCFA occurred on day 0 after the first fecal sample was taken, and was given to either all steers or else one steer in a pen. CTC was provided in three sequential 5-day pulses with a single day in between. Both treatment regimens were on label. Samples were collected every other day; however, only samples shown above (Days 0, 4, 8, 14, 20 and 26) were tested for Salmonella quantity.