Population Dynamics of Salmonella enterica within Beef Cattle Cohorts Followed from Single-Dose Metaphylactic Antibiotic Treatment until Slaughter

Salmonella is a leading bacterial foodborne pathogen, causing a significant number of human infections and deaths every year in the United States. Macrolides and 3rd-generation cephalosporins play critical roles in the treatment of human salmonellosis. Use of these antibiotics in beef cattle can select for resistant bacteria that may enter the food chain or spread from the farm via manure. There is a lack of longitudinal research concerning the long-term effects of metaphylactic antibiotic administration. Here, we assessed Salmonella population dynamics during the feeding period until slaughter following single-dose antibiotic treatment. We found no long-term effects of antibiotic use early in the cattle-feeding period on Salmonella prevalence and antimicrobial resistance at slaughter. We identified the pens in which cattle were housed as the factor that contributed most to Salmonella serotypes being shared; importantly, the dominant strain in each pen changed repeatedly over the entire feeding period.

ABSTRACT

Antibiotic use in cattle can select for multidrug-resistant Salmonella enterica, which is considered a serious threat by the U.S. Centers for Disease Control and Prevention. A randomized controlled longitudinal field trial was designed to determine the long-term effects of a single dose of ceftiofur or tulathromycin on Salmonella population characteristics in cattle feces and peripheral lymph nodes and on hides. A total of 134 beef cattle from two sources were divided among 12 pens, with cattle in each of the 3-pen blocks receiving a single dose of either ceftiofur or tulathromycin or neither (control) on day 0. Fecal samples were collected before treatment (day 0) and repeatedly following treatment until slaughter (day 99+). Hide and lymph node samples were collected at slaughter age. Salmonella prevalence, phenotypic antimicrobial resistance, serotype, and phylogenetic relationships were examined. Multilevel mixed logistic regression models indicated no significant effects (P ≥ 0.218) of metaphylactic antibiotics on the prevalence of Salmonella across sample types. However, there was a significant time effect observed, with prevalence increasing from spring through the midsummer months (P < 0.0001) in feces. The majority of Salmonella isolates were pansusceptible to a panel of 14 antibiotics both before and after treatment. Highly prevalent Salmonella serotypes were Salmonella enterica serovar Montevideo, Salmonella enterica serovar Anatum, Salmonella enterica serovar Cerro, and Salmonella enterica serovar Lubbock across all sample types. Strong pen and cattle source serotype clustering effects were observed among Salmonella isolates originating from fecal, lymph node, and hide samples; however, the potential role of Salmonella isolates from the pen environment prior to animal placement was not assessed in this study.

IMPORTANCE Salmonella is a leading bacterial foodborne pathogen, causing a significant number of human infections and deaths every year in the United States. Macrolides and 3rd-generation cephalosporins play critical roles in the treatment of human salmonellosis. Use of these antibiotics in beef cattle can select for resistant bacteria that may enter the food chain or spread from the farm via manure. There is a lack of longitudinal research concerning the long-term effects of metaphylactic antibiotic administration. Here, we assessed Salmonella population dynamics during the feeding period until slaughter following single-dose antibiotic treatment. We found no long-term effects of antibiotic use early in the cattle-feeding period on Salmonella prevalence and antimicrobial resistance at slaughter. We identified the pens in which cattle were housed as the factor that contributed most to Salmonella serotypes being shared; importantly, the dominant strain in each pen changed repeatedly over the entire feeding period.

INTRODUCTION

Each year in the United States approximately 9.4 million foodborne illnesses are estimated to occur (1). Of these, Salmonella is estimated to cause 1.2 million infections, 23,000 hospitalizations, and 450 deaths every year (1, 2). Based on data from the U.S. Centers for Disease Control and Prevention (CDC) National Outbreak Reporting System, 1,833 Salmonella-related outbreaks were calculated to have occurred from 2009 to 2017 in the United States. Of these cases, 1,186 (64.7%) were attributed to consumption of contaminated food; in contrast, the remaining 35.3% of cases were attributed to direct animal contact, environmental sources, person-to-person contact, or various unknown reasons (publicly accessible national outbreak data were accessed at https://wwwn.cdc.gov/norsdashboard/ in July 2019). Most food-related human cases of salmonellosis are attributed—in order of importance—to ingestion of contaminated seeded vegetables, eggs, poultry, and beef products (3, 4).

Cattle are usually exposed to Salmonella by ingestion of contaminated feed and water (5). Salmonella carriers shedding the bacteria through their fecal waste into the environment cause persistent Salmonella contamination in beef feedlots (6). Salmonella can be found in the digestive tract and lymph nodes of healthy cattle, as well as in the feedlot environment (7, 8). Salmonella organisms are typically introduced into cattle carcasses at slaughter from aerosols generated from hides through the skinning process or via direct fecal contamination (9–11). Carcass contamination may carry through to final beef products. Moreover, cattle lymph nodes harboring Salmonella embedded in fat tissue also can become incorporated in ground meat products during the fat-trimming process (7, 11, 12).

Salmonella mostly causes self-limiting infections in humans. However, invasive Salmonella can migrate from the intestinal lumen via the bloodstream or lymphatics to other body sites, resulting in hospitalization, along with the need for antibiotic treatment. Antibiotics used for the treatment of invasive and extraintestinal Salmonella infections in humans include 3rd-generation cephalosporins (e.g., ceftriaxone), fluoroquinolones (e.g., ciprofloxacin), and macrolides (e.g., azithromycin) (13, 14). Fluoroquinolone use is restricted in children and pregnant women due to adverse side effects of the antibiotic (15, 16); therefore, ceftriaxone and azithromycin are often the primary choices for empirical therapy of pediatric and obstetric cases.

There is growing global concern about antimicrobial resistance among enteric pathogens due to an observed increase of cephalosporin- and fluoroquinolone-resistant Salmonella in humans (2, 17, 18). In addition, plasmid-mediated reduced susceptibility to fluoroquinolones is also an emerging problem (19, 20). The CDC classified multidrug-resistant (MDR) Salmonella as a serious public health threat in a report published in 2013 (2). According to that report, MDR Salmonella strains were conservatively estimated to cause 100,000 infections and 40 deaths annually in the United States. In the same threat report, increasing ceftriaxone and ciprofloxacin resistance was reported in Salmonella isolates recovered from human cases. Based on data obtained from human Salmonella isolates generated by the National Antimicrobial Resistance Monitoring System (NARMS) since 1996, levels of ceftriaxone resistance found in human Salmonella isolates gradually increased from 0.2% to up to 4.4% by 2003. Since 2003, these levels have not decreased below the level of 2.4% (2). Similarly, the number of isolates exhibiting resistance or reduced susceptibility to ciprofloxacin had increased from below 0.5% to 3% by 2011 (2). The NARMS Human Isolates Surveillance Report published in 2015 also highlighted an increasing prevalence of azithromycin resistance among human Salmonella isolates. According to that report, azithromycin resistance was observed in the range of 0.05 to 0.23% of isolates since 2011. However, it had reached the highest observed resistance level (0.34%) by 2015 (21).

In order to accurately explore the epidemiology of antibiotic resistance and related risk factors, it is important to determine the effects of antibiotics that are extensively used in food-producing animals. Antibiotics are widely used for treatment, metaphylaxis (control), and prophylaxis (prevention) purposes in beef cattle. Metaphylaxis is commonly used to control infectious bacterial diseases in groups of cattle during an outbreak, for example, respiratory diseases caused by bacterial pathogens that reach epidemic rates soon after arrival, processing, and placement of cattle in feedlot pens. Bovine respiratory disease (BRD) (also known as shipping fever) is the most common health problem of beef cattle, affecting an estimated 16.2% of cattle arriving at U.S. feedlots (22–24); as a result, 59% of cattle receive metaphylaxis to reduce the incidence, severity, and consequences of BRD in feedlots. Cattle receiving metaphylactic treatments upon arrival remain in the same feedlot for 3 to 10 months—depending on starting weight—until they reach the desired slaughter weight (U.S. Department of Agriculture Economic Research Service [USDA-ERS], Cattle and Beef Sector at a Glance; https://www.ers.usda.gov/topics/animal-products/cattle-beef/sector-at-a-glance/ [accessed August 2019]).

Tulathromycin and ceftiofur are antibiotics that are often prescribed by consulting veterinarians to control BRD in beef cattle (22, 25, 26). Tulathromycin, an extended-duration azalide macrolide in the same subclass as azithromycin, was approved for use in beef cattle for BRD in 2005. Ceftiofur crystalline-free acid (CCFA), a long-acting 3rd-generation cephalosporin formulation with a molecular structure similar to that of ceftriaxone, was approved for BRD infections in both dairy and beef cattle in 2003. Azithromycin and ceftriaxone are macrolides and 3rd-generation cephalosporins, respectively, antibiotic classes that continued to be listed as highest-priority critically important antimicrobials for human medicine by the World Health Organization (WHO) as of 2017 (27).

In the past 2 decades, numerous studies have been conducted to better understand the role of antibiotic use in both direct and indirect selection of resistant bacteria in food animals (28–36). A subset of these studies suggest use of antibiotics to control BRD in beef cattle may significantly or transiently increase the selection pressure for resistant bacterial populations during short periods (28–32, 34). Evidence from previous research suggests that antibiotic treatment in beef cattle has an effect on both Escherichia coli and Salmonella (30–32). In previous work from our group at the same feedlot location, a 26-day randomized and controlled field trial was performed to test the effects of injectable CCFA and oral chlortetracycline treatment on E. coli populations in beef cattle (30). The results showed that ceftiofur invoked a moderate selective pressure for MDR E. coli by reducing the proportion of susceptible strains compared to the resistant E. coli population. Ohta et al. further explored the dynamics of enteric Salmonella in the same controlled trial and found there was an expansion of MDR Salmonella related to a single-dose long-acting ceftiofur treatment on day 4; importantly, it remained elevated until the end of the study (day 26) (32). According to Ohta et al., the MDR phenotype was found only in Salmonella enterica serovar Reading. That study is the only published longitudinal controlled field study that has measured direct effects of ceftiofur on Salmonella populations for up to a month following antibiotic treatment. The dynamics and persistence of MDR Salmonella between 1 month posttreatment and slaughter (typically, >90 days) remain unknown.

Even though there are published studies focusing on the effect of metaphylactic ceftiofur treatment on fecal E. coli (30, 31) and the resistome in cattle feces (33, 34), studies measuring the effects of ceftiofur use on Salmonella specifically are very limited (32). Especially limited are those that follow animals from treatment early in the feeding period until slaughter. In addition, there is a lack of research measuring the effects of tulathromycin treatment on enteric bacterial populations in cattle. Using shotgun metagenomic sequencing, Doster et al. published the only study to have provided a comparison of the fecal resistome and microbiome among tulathromycin-treated and nontreated beef cattle before treatment (day 1) and after treatment (day 11) early in the feeding period (35). Their results did not show any effect of tulathromycin within the fecal resistome and microbiome after 11 days.

Clearly, there is a need for randomized and controlled longitudinal studies measuring longer-term effects of ceftiofur and tulathromycin on Salmonella populations and their antibiotic resistance patterns. Further work is needed focusing on Salmonella, not only in feces, but also in the lymph nodes and on hide surfaces of cattle, all of which are potential contamination sources of cattle origin Salmonella at slaughter. The objective of the present study was to explore the longer-term effects of antibiotic use on Salmonella population characteristics, that is, until slaughter age, following >90 days on feed (Fig. 1). More specifically, we aimed to compare Salmonella prevalences, antibiotic resistance profiles, serotype distributions, and genotypic relatedness among feces, lymph nodes, and hides in pens of animals administered ceftiofur, tulathromycin, or neither (Table 1).

FIG 1.

Timeline of the study. Antibiotic administration (blue), fecal sample collection (red), and lymph node and hide swab collection at slaughter age (green) are shown.

TABLE 1.

Descriptive field trial data for cattle origin, block, pen, treatment, and slaughter day

Source	Block	Pen identifier	No. of cattle	Pen level treatment	Slaughter day
1	1	7	12	Tulathromycin	134
1	1	8	11	Control	134
1	1	9	12	Ceftiofur	134
2	2	51	11	Tulathromycin	141
2	2	52	11	Control	141
2	2	53	11	Ceftiofur	141
2	3	54	11	Control	120
2	3	55	11	Ceftiofur	120
2	3	56	11	Tulathromycin	120
2	4	57	11	Ceftiofur	99
2	4	58	11	Tulathromycin	99
2	4	59	11	Control	99

RESULTS

Descriptive statistics.

A total of 1,155 samples were received. Of these, 799 were fecal samples, 224 were lymph nodes, and 132 were hide samples. One fecal sample from day 0, due to a steer escaping the chute, and 36 lymph nodes at the first slaughter day, due to a lack of training in sample collection and identification of lymph nodes among responsible personnel at the slaughter plant, were missing. One steer from a ceftiofur-treated pen (pen 55) and another steer from a tulathromycin-treated pen (pen 51) were removed from the study due to BRD and a foot injury, respectively, after day 28 and were lost to follow-up. In total, 132 cattle completed the study.

(i) Enrichment results. Salmonella prevalences in fecal samples were estimated at 43.6% (58/133) on day 0, 20.1% (27/134) on day 7, 20.1% (27/134) on day 14, 41.0% (55/134) on day 28, 57.5% (76/132) on day 56, and 80.3% (106/132) on terminal day (aggregated into ∼day 112). Lymph node prevalence at the carcass level was 84.2% (96/114 [20 cattle lymph nodes were missing]), whereas overall lymph node prevalence was 75.4% (169/224). The hide prevalence was 84.8% (112/132) across all the pens. Among control and tulathromycin- and ceftiofur-treated animals, terminal fecal prevalences were distributed as 66.6%, 92.3%, and 87.1%; lymph node prevalences were distributed as 77.0%, 88.2%, and 90.6%; and hide prevalences were distributed as 77.7%, 87.1%, and 92.3%, respectively (Table 2).

TABLE 2.

Overall and treatment level Salmonella prevalence distribution by sample day (feces only) and sample type

Treatment	% prevalence (no. of positive samples/total no.)
	Fecal						Lymph nodea	Hide
	Day 0	Day 7	Day 14	Day 28	Day 56	Terminal
Total	43.6 (58/133)	20.1 (27/134)	20.1 (27/134)	41 (55/134)	57.5 (76/132)	80.3 (106/132)	84.2 (96/114)	84.8 (112/132)
Ceftiofur	32.5 (13/40)	15 (6/40)	17.5 (7/40)	35 (14/40)	66.6 (26/39)	87.1 (34/39)	90.6 (29/32)	92.3 (36/39)
Control	54.7 (29/53)	16.6 (9/54)	22.2 (12/54)	42.5 (23/54)	42.5 (23/54)	66.6 (36/54)	77.0 (37/48)	77.7 (42/54)
Tulathromycin	40 (16/40)	30 (12/40)	20 (8/40)	45 (18/40)	69.2 (27/39)	92.3 (36/39)	88.2 (30/34)	87.1 (34/39)

^aAnimal level frequencies are presented for the lymph node data. The overall lymph node prevalence was 75.4% (169/224).

(ii) Phenotypic antibiotic susceptibility. Among 630 Salmonella isolates, the majority were pansusceptible (79.0%). The remaining isolates presented with either single (20.4%) or double (0.4%) phenotypic antibiotic resistance. Most of the isolates resistant to a single antibiotic (n = 108) exhibited resistance to tetracycline, while 21 other isolates were resistant to streptomycin. Three isolates were resistant to both tetracycline and streptomycin. The isolates tested in this study did not shown any phenotypic resistance to the remaining 12 antibiotics tested, including azithromycin, ceftriaxone, and ceftiofur (Table 3).

TABLE 3.

Percentage distribution of MIC values for 630 Salmonella isolates tested against 14 antibiotics

^aThe MIC of the first antibiotic in a combination is listed.

^bNARMS breakpoints were used for classification.

^cR, resistance.

^dA one-sided 97.5% confidence interval was used when the prevalence estimate was zero.

^eThe shaded areas represent the out-of-dilution range of the Sensititre plate. Resistant values are indicated in red, and breakpoints for resistance classification fall between the black and red numbers. The numbers in the shaded areas are right-censored MICs.

(iii) Quality metrics of sequencing data. Whole-genome sequencing (WGS) was performed on all the fecal isolates (n = 191) recovered on days 0 and 7 and the terminal day; one lymph node isolate from each animal (n = 96); and all hide isolates (n = 112) in order to explore the population dynamics of Salmonella before treatment, immediately after treatment, and at slaughter age. Run parameters were evaluated after genome assembly of the 399 isolates. The average contig number was 43 (range, 17 to 95), the N₅₀ value was on average 423,770 bp (range, 101,429 to 966,986 bp), and the average depth of coverage was 54× (range, 28× to 190×). The average genome length of the Salmonella isolates was 4,775,057 bp (range, 4,541,106 to 5,072,387 bp).

(iv) Serotypes and sequence types (STs). Seven serotypes were identified during the study from the three sampling periods. The serotypes found in the study were Salmonella enterica serovar Lubbock (n = 136), Salmonella enterica serovar Anatum (n = 113), Salmonella enterica serovar Montevideo (n = 68), Salmonella enterica serovar Cerro (n = 64), Salmonella enterica serovar Kentucky (n = 11), Salmonella enterica serovar Newport (n = 6), and Salmonella enterica serovar Norwich (n = 1). Of these, seven serotypes (S. Anatum, S. Cerro, S. Lubbock, S. Montevideo, S. Kentucky, S. Newport, and S. Norwich) were observed in the fecal samples. Five different serotypes (S. Anatum, S. Cerro, S. Lubbock, S. Montevideo, and S. Newport) were observed in the lymph nodes, and four different serotypes (S. Anatum, S. Cerro, S. Lubbock, and S. Montevideo) were observed among the hide samples (Fig. 2). S. Kentucky was observed only in day 0 (n = 7) and day 7 (n = 4) fecal samples and only in pens where the cattle originated from a single source (source 1). S. Cerro was observed only at slaughter age, though it occurred across multiple types of samples. Six S. Newport isolates were found in cattle placed in one particular pen (pen 53); of those, five isolates were located in lymph nodes and one was located in the final fecal sample from one of the cattle. S. Norwich was isolated once from a terminal fecal sample. Almost half of the S. Montevideo isolates recovered in this study were found in day 0 fecal samples. In contrast, S. Anatum was prevalent across all sample days and sample types (Fig. 3). The antigenic profiles of the serotypes found in the study are presented in Table 4.

FIG 2.

Serotype distribution of Salmonella isolates by day and sample type for all cattle enrolled in the study. There were a total of 399 positive Salmonella isolates, 246 negative Salmonella isolates, and 25 missing samples.

FIG 3.

Distribution of serotypes among individual animals, pens, and sample types and by day. ID, unique animal identifier; 0, day 0; 7, day 7; TER, terminal fecal isolates; LYM, lymph node isolates; HD, hide isolates. Each block of pens (three pens on each corner) was harvested on the same day: block 1 (pens 7 to 9) on day 134, block 2 (pens 51 to 53) on day 141, block 3 (pens 54 to 56) on day 120, and block 4 (pens 57 to 59) on day 99. Block 1 was from source 1; the remaining blocks were from source 2. The numbers in the upper left corners are pen identifiers. Pen treatments are shown at the top of each pen. The geographic locations of pens and treatments are shown schematically in the middle.

TABLE 4.

Antigenic formulas, serotypes, sequence types, and antibiotic resistance phenotypes of sequenced isolates from fecal, lymph node, and hide samples

Antigenic formula	Serotype	ST	No. with antibiotic resistance phenotype
			STRa	TETb	TET-STR	Pansusceptible
3,10:e,h:1,6	Anatum	64	7	0	0	106
18:z4,z23:-	Cerro	367	0	1	0	63
7:g,m,s:e,n,z15	Lubbock	413	6	2	0	128
7:g,m,s:-	Montevideo	138	1	63	2	2
8:i:z6	Kentucky	152	1	0	0	10
8:e,h:1,2	Newport	118	0	0	0	6
7:e,h:1,6	Norwich	2119	0	0	0	1

^aSTR, streptomycin.

^bTET, tetracycline.

Analysis of multilocus sequence typing (MLST) data showed that a single ST corresponding to each serotype was found. They were identified as (i) ST 64 (S. Anatum), (ii) ST 138 (S. Montevideo), (iii) ST 367 (S. Cerro), (iv) ST 413 (S. Lubbock), (v) ST 118 (S. Newport), (vi) ST 2119 (S. Norwich), and (vii) ST 152 (S. Kentucky) (Table 4).

Cross-tabulations of serotypes by individual phenotypic antibiotic resistance data showed strong associations. Among S. Montevideo isolates, 95.5% exhibited tetracycline resistance; in contrast, the remaining six serotypes exhibited either complete susceptibility to tetracycline (S. Anatum, S. Kentucky, S. Newport, and S. Norwich) or a limited number of resistant isolates (i.e., one S. Cerro isolate and two S. Lubbock isolates). Phenotypic streptomycin resistance was found in seven S. Anatum isolates, six S. Lubbock isolates, one S. Montevideo isolate, and one S. Kentucky isolate (Table 4).

In contingency table analyses, unadjusted likelihood ratio chi-square tests showed significant associations among the 12 pens with the detected serotypes among hide samples. The unadjusted crude likelihood ratio chi-square statistic was 168.9, with 33 degrees of freedom (P < 0.0001), for the pen. The unadjusted crude likelihood ratio chi-square statistic was 32.2, with 6 degrees of freedom (P < 0.0001), for pen level treatments (Table 5).

TABLE 5.

Hide level unadjusted frequencies and prevalence proportions of Salmonella serotypes by pen and pen level treatments found at slaughter age^a

^aFrequencies of serotypes are indicated with a gray shading scale. Serotypes with higher frequencies are indicated by increasingly darker shading.

(v) Phylogenetic analysis. A complete genome of cattle origin S. Anatum (GenBank accession no. CP014621.2), isolated in the United States, was found to be closest to the genome of the S. Anatum strains found in this study. Prophage regions marked as “intact” or “questionable” were considered for further masking. A total of four prophage regions were detected in the strain, and prophage regions were masked with the letter “N” from the reference (see Table S1 in the supplemental material). A generalized time reversible (GTR)-plus-G4 (gamma parameter) model was selected for tree inference using IQ-tree (see below). The reference strain described above was used to root the inferred tree. The bootstrap support values from 0.8 to 1.0 were included in circles in the tree with the size scaled proportionately to the values. A cladogram illustration was generated, with unique colors representing each pen subcoded with a preferred scale of colors relative to each block/cattle origin across serotypes as follows: pens (n = 12) were colored individually; blocks (n = 4) were colored in shades of blue, pink/purple, red/orange, and green; and the source (n = 2) was distinguished by shades of blue (source 1) versus shades of the remaining block colors (source 2) (Fig. 4).

FIG 4.

Cladogram representing the pen level overall population structure of Salmonella serotypes. The maximum-likelihood cladogram was generated using a GTR-plus-G4 model. The serotype colors are represented in the cladogram branches and outer circle. The colors in the inner circle represent the pen distribution. Block 1, pens with shades of blue; block 2, pens with shades of pink/purple; block 3, pens with shades of red/orange; block 4, pens with shades of green; source 1, shades of blue; source 2, shades of pink, red, or green.

Multilevel mixed-effects logistic regression analysis.

Variance-covariance/correlation matrices and structures were examined and preset as exchangeable for pen and as unstructured for repeats within individual animals for random effects. Likelihood ratio-based tests of the fixed effects of treatment and day (period; feces only) and their interactions in mixed models are presented in Table S2 in the supplemental material. Predictive marginal estimates obtained from the multilevel mixed-effects logistic regression are summarized as marginal means and 95% confidence intervals (CI) in Table S3 in the supplemental material for fecal, lymph node, and hide observations. Specific post hoc multiple comparisons adjusted using the Bonferroni method are presented for contrasts of day and treatment for feces in Table S4 and treatments in Table S5 in the supplemental material.

(i) Fecal samples. Likelihood ratio-based tests showed that the treatment group did not result in statistically significant differences (P = 0.825) among treatment and control groups for fecal prevalence of Salmonella. However, there was a significant (P < 0.0001) day effect observed across all treatment groups in fecal prevalence of Salmonella (see Table S2). According to pairwise adjusted comparisons of marginal results, Salmonella prevalence in antibiotic-treated groups showed similarity across all treatment groups for any given day (see Table S5). However, the day had a significant effect on the prevalence within each treatment group, as follows: a significant (P = 0.001 and P = 0.018, respectively) prevalence decrease was observed from day 0 to day 7 and day 14 in the control group. Later, for the same group, recovery to the baseline value was observed by day 28 and day 56. By day 112, prevalence was significantly increased compared to day 7 and day 14 (P < 0.0001 for both). Prevalences of Salmonella in the tulathromycin- and ceftiofur-treated groups were not different from baseline values until day 56. A significant increase was observed at day 56 compared to day 7 (P = 0.006 and P < 0.0001, respectively, for tulathromycin and ceftiofur) and day 14 (P < 0.0001 for both antibiotics). This increase continued through days 56 and 122 (P < 0.0001 for both), and by the end of the study, Salmonella prevalences in the ceftiofur- and tulathromycin-treated groups were significantly higher than at earlier days (Fig. 5). Pairwise comparisons of marginal contrasts can be found in Table S4.

FIG 5.

Marginal mean modeled predictions of Salmonella fecal prevalence by day and treatment. The error bars represent 95% confidence intervals.

(ii) Lymph node and hide samples. Overall, likelihood ratio-based tests did not reveal any statistically significant differences in Salmonella prevalences among treatment groups for lymph node and hide samples (P = 0.297 and P = 0.218, respectively) (see Table S2). Salmonella prevalence in the control group was not significantly different than that in the tulathromycin and ceftiofur groups in the lymph nodes (P = 0.344 and P = 1.000, respectively) (see Table S5). Similarly, Salmonella prevalence in the control was not significantly different from that in the tulathromycin and ceftiofur groups for the hide samples (P = 1.000 and P = 0.212, respectively). A detailed summary of pairwise contrasts of predicted margin comparisons is provided in Table S5, while predictive marginal estimates can be found in Table S3.

(iii) Intraclass correlation coefficients. Pen level intraclass correlation coefficients (ICC) for Salmonella prevalences were reported as 0.21 (95% CI, 0.12 to 0.35) for feces, 0.13 (95% CI, 0.02 to 0.44) for lymph nodes, and 0.30 (95% CI, 0.06 to 0.73) for hide samples. Animal level ICC for Salmonella prevalence were 0.21 (95% CI, 0.12 to 0.35) for temporally dependent fecal samples within animals and 0.64 (95% CI, 0.37 to 0.84) for contemporaneously sampled bilateral subiliac lymph nodes within animals.

DISCUSSION

Our findings suggest that neither ceftiofur nor tulathromycin, when used for control of BRD early in the feeding period, affects feedlot origin Salmonella prevalence and antibiotic resistance profiles in beef cattle at slaughter, whether in fecal, lymph node, or hide samples. Importantly, this inference is limited to situations where MDR strains are absent during the feeding period.

The sampling day had a significant (P < 0.0001) effect on the prevalence of Salmonella, with increasing prevalence associated with a temporal shift from spring through the summer months. This difference likely reflects a seasonal effect on Salmonella populations due to increasing ambient temperatures enhancing Salmonella persistence in the pen environment in March 2016 versus August 2016. However, one of the limitations of this study was a lack of prior environmental samples to determine the contributions of the environment in determining Salmonella dynamics from early in the feeding period until slaughter. Therefore, the observed difference could also be attributed to the aging of the animals during the same period, among other potential confounding factors. The Salmonella population increase across seasons was also reported by the U.S. Department of Agriculture (USDA) National Animal Health Monitoring System (NAHMS) 1999 feedlot Salmonella survey. According to the report, the lowest Salmonella prevalence was observed from January through March, whereas the highest Salmonella prevalence was found from July through September in fecal samples collected from pen floors of 73 randomly selected feedlots in the United States (https://www.aphis.usda.gov/animal_health/nahms/feedlot/downloads/feedlot99/Feedlot99_is_Salmonella.pdf [accessed July 2019]). This observed increase was also supported by Vikram et al., who reported a significant increase (P < 0.01) in Salmonella prevalence in the feces of beef cattle at slaughter in the summer compared to winter and spring (37).

The ICC reveals the magnitude of the effect of clustering within the nested components of variance observed. The ICC can have absolute values between zero and one. While an ICC value of zero suggests no correlation among observations within clusters, positive values suggest a positive correlation among observations within clusters (38). ICC values can play an important role in understanding the ecological and animal aggregating factors (such as pen or repeated observations for an animal) related to the epidemiology of Salmonella in feedlot settings. In our study, 30% of the variance related to Salmonella presence on hides, 21% of the variance in feces, and 13% of the variance in lymph nodes were attributed to pen level variability. This result suggests a larger role is played by the environmental versus animal-related factors, since at the pen level, the Salmonella hide prevalences are more likely to be similar within a pen than are the prevalences in the lymph nodes obtained from each animal. Animal level dependencies were measured only for fecal and lymph node samples. They showed that 64% of the variance in lymph nodes and 21% of the variance in feces observed for Salmonella prevalence were attributed to within-animal dependencies. Lymph node prevalence was significantly influenced by individual animal carcasses being contemporaneously sampled bilaterally at slaughter in the summer, while fecal sample dependencies arose longitudinally, starting in the spring and ending in the summer period. In our study, the overall results show that pen and animal level clusters had the greatest impact on Salmonella prevalence regardless of the assigned treatment groups. To our knowledge, pen level and animal level ICC estimates for Salmonella prevalence have not been previously reported in randomized cohorts of beef cattle. However, Cull et al. reported the ICC of enterohemorrhagic E. coli from different pens and feedlots. Our estimates were similar to the recent cross-sectional study published by Cull et al., measuring the feedlot and pen level ICC of E. coli by collecting cattle fecal samples across eight commercial feedlots in a region near the Texas panhandle. In their study, E. coli ICC ranged from 4% to 8% between cattle feces and feedlots, whereas at the pen level, ICC ranged from 26% to 31% (39).

Our study did not reveal any effect of tulathromycin or ceftiofur antibiotic given as a control for BRD on phenotypic antibiotic resistance in Salmonella isolates from cattle from day 0 until slaughter. In our study, approximately 80% of Salmonella isolates recovered from feces, lymph nodes, and hides from ceftiofur, tulathromycin, and control group cattle were pansusceptible; further, the remaining 20% of Salmonella isolates exhibited only tetracycline and/or streptomycin resistance. Even though tetracycline and streptomycin are not generally used for the treatment of Salmonella cases, the antibiotics are classified as highly important for human medicine by the WHO (27). Based on comparisons of phenotypic antibiotic resistance data and serotypes, we found that phenotypic tetracycline resistance was observed mostly among S. Montevideo isolates found in feces, lymph nodes, and hides throughout the study without any known direct selection pressure caused by the use of tetracycline.

Our findings also demonstrated that serotype and antibiotic resistance patterns were closely associated. Further genetic analysis is required to understand if tetracycline resistance is located on a chromosome or a plasmid among these S. Montevideo isolates. In this study, we found 15 streptomycin-resistant isolates; among those, 7 isolates were S. Anatum, 6 isolates were S. Lubbock, and the 2 remaining isolates were other serotypes. The NARMS breakpoint of streptomycin decreased from ≥64 μg/ml to ≥32 μg/ml in 2014 due to inconsistencies observed in the genotype-phenotype match. In this study, we utilized the most up-to-date MIC breakpoint for streptomycin, which yielded a 6-fold increase in the number of isolates coded as resistant compared to the previous breakpoint value (i.e., prior to 2014). If a bias was involved, it would lead to false-positive results for str genes and other aminoglycoside resistance genes, as previously described by Tyson et al. (40). Genetic elements of resistance need to be further evaluated to address this potential problem of test accuracy.

In our study, isolates were not resistant to azithromycin, ceftiofur, or ceftriaxone either before or after either ceftiofur or tulathromycin treatment. This result was supportive of metagenomic data analyses conducted by Weinroth et al. and Doster et al. but counter to the findings of Alali et al. (33–35). The studies measured the effect of either ceftiofur or tulathromycin treatment on the microbiome and/or resistome of the fecal community of beef cattle; however, none of the studies focused directly on culturable bacteria, let alone Salmonella, as the outcome bacteria of interest.

Alali et al. quantified the bla_CMY-2 gene in feces (collected on days 0, 3, 7, 10, 14, 18, 21, and 28) of beef cattle during a 28-day period, following single (day 0) or multiple (days 0, 6, and 13) administration of two different doses of ceftiofur (4.4 mg/kg of body weight and 6.6 mg/kg) (34). In contrast to our study, they found that administration of ceftiofur for all treatment groups increased the absolute and normalized numbers of bla_CMY-2 genes detected in fecal samples compared to the control group throughout the 28-day period. The observed increase in bla_CMY-2 genes found by Alali et al. was most likely related to phenotypically ceftiofur-resistant E. coli, which was observed in the same cattle study conducted by Lowrance et al. (41). The latter authors evaluated phenotypic antibiotic resistance in E. coli. According to their findings, ceftiofur treatment increased resistant E. coli populations after treatment; however, they observed that the population returned to preadministration resistance levels after a 2-week period. Weinroth et al. also investigated the effects of ceftiofur use on the resistome of cattle feces, focusing on the bla_CMY-2 and bla_CTX-M-24 genes in the fecal microbiome on day 0 and day 26 (33). In contrast to the results of Alali et al., but similar to our results, their results did not indicate any significant changes in ceftriaxone resistance over a 26-day period, though they did not analyze the samples taken during peak antibiotic activity (i.e., days 4 to 12).

Doster et al. found that there was no significant difference in the cattle resistome and microbiome among tulathromycin-treated and control groups; however, their results may be related to their small sample size (n = 15 per treatment group by day). The authors did, however, report a significant increase in the 16S rRNA normalized antibiotic resistance gene abundance and the average relative abundances of microbial taxa between days 1 and 11 (35). Similar to the study of Weinroth et al. (33), Doster et al. ignored the likely peak day of antibiotic effect on enteric populations; importantly, their study did not measure the long-term effects of tulathromycin treatment out to slaughter eligibility. Overall, the results suggest that ceftiofur and tulathromycin may have a transient or no effect on selection of resistance genes or on the microbiome in cattle.

Cattle hides and feces are two major sources of direct carcass contamination at slaughter, that is, during the hide removal and evisceration processes. Carcass surfaces can also be contaminated indirectly from the slaughterhouse environment. Slaughterhouse interventions, such as exposing carcasses to steam vacuuming, lactic acid, and hot-water washes, are highly efficacious at reducing Salmonella numbers on carcass surfaces (42). However, infected lymph nodes embedded in fatty tissues are likely to be incorporated into ground beef during meat processing, thus contaminating finished retail products and potentially leading to salmonellosis in humans through consumption of undercooked meat. Salmonella organisms present in the lymph nodes are not likely to be responsive to slaughterhouse interventions, which focus on carcass hygiene. This may result in contamination of ground beef products and outbreaks regardless of antibiotic resistance profiles. In our study, animal level Salmonella prevalence in lymph nodes was over 80% at slaughter age. All S. Newport serotypes were recovered from a single pen (pen 53) and mostly in the lymph nodes. These S. Newport isolates were strains that remained pansusceptible throughout the study. This is somewhat unusual for the serotype, especially given its association with MDR S. Newport outbreaks between 2004 and 2013 (43), This finding is important from a public health perspective, since one of the most recent Salmonella outbreaks involved a pansusceptible S. Newport strain contaminating ground beef products that resulted in 403 reported salmonellosis cases, 117 hospitalizations, and the recall of 5,488 tons of beef products (http://www.outbreakdatabase.com/reports/2018-2019_JBS_CDC_Marc_22_2019.pdf [accessed August 2019]). The source of this outbreak strain was traced and found to be associated with beef trim and beef products processed in two Texas slaughterhouses and processing facilities (44). The sources of Salmonella in lymph nodes remain unclear; however, migration of Salmonella through the lymphatics has been shown experimentally to occur via the gastrointestinal tract or transdermal route (45). The epidemiology of Salmonella—specifically S. Newport—in cattle lymph nodes needs to be better evaluated, and preharvest intervention practices need to be developed and utilized to reduce pathogen carriage in final beef products.

In this study, we also identified strong cattle origin/source, pen, and day effects on the distribution of serotypes among hides, lymph nodes, and fecal samples. Overall, the distribution of Salmonella serotypes among fecal, lymph node, and hide samples did not seem to differ, though the last two sample types were recovered only at slaughter and not earlier, as from feces. S. Lubbock was the most prevalent serotype (34.1%), followed by S. Anatum (28.3%) and S. Montevideo (17.0%). In a study by Ohta et al. (32) conducted in 2009 in the same feedlot as the present study, a total of 566 Salmonella isolates were identified. The most prevalent serotype was found to be Salmonella enterica serovar Mbandaka (ST 413), with 37.9% prevalence, followed by 19.1% Salmonella enterica serovar Give (ST 654), 15.2% S. Reading (ST 1628), 13.6% S. Kentucky (ST 198), 13.4% S. Montevideo (ST 138), and 0.7% S. Anatum (ST 64) (32). When common serotypes identified in both studies were observed, S. Montevideo and S. Anatum shared identical STs, whereas S. Kentucky (ST 152) differed. In our study, we did not identify any S. Give, S. Mbandaka, or S. Reading. In contrast, in our study, we identified S. Lubbock (ST 413), S. Cerro (ST 367), S. Newport (ST 118), and S. Norwich (ST 2119). S. Lubbock is a recently named serotype that was first isolated from a cattle peripheral lymph node and reported in 2015 (46). This new serotype is believed to have emerged from S. Mbandaka by acquiring the fliC gene operon from S. Montevideo (46, 47). Interestingly, during the 7-year gap between our study and that of Ohta et al., the shift of the most dominant serotype from S. Mbandaka to S. Lubbock with the same ST (ST 413) is likely explained by host- and environment-related selection pressure on certain serotypes. In the study by Ohta et al. (32), MDR Salmonella isolates were largely S. Reading, which could potentially explain the absence of MDR Salmonella in our study, since we failed to identify any other serotype that exhibited an MDR profile under the selection pressure of ceftiofur treatment. As mentioned, the population dynamics of Salmonella prevalence seen in the present study would be expected to differ greatly in the presence of an MDR serotype, such as that of S. Reading seen by Ohta et al. (32). Therefore, we caution that the findings presented here are restricted to situations in which no resistance to the antibiotics being tested is observed among within-host Salmonella populations.

In the present study, S. Cerro appeared only at slaughter age and was otherwise absent in the early feeding period, perhaps demonstrating the unique temporal dynamics of Salmonella. S. Cerro is known to be a pathogen for at-risk cattle, such as dairy cows; however, beef cattle involved this study did not show any clinical signs of salmonellosis (48). In contrast, S. Kentucky was isolated from only one origin/source of cattle in the early feeding period, and it was not recovered at slaughter age (Fig. 3). In addition, according to the Salmonella pen and serotype distributions demonstrated in Table 5, for a majority of the pens, a single dominant serotype was determined to be prevalent on hide samples from animals in the same pen at slaughter age.

In our study, a single ST was found for each serotype, which suggests the relatively clonal spread of Salmonella among days and sample types and within single pens, pen blocks, and sources of cattle in pens. This conclusion is also supported by the phylogenetic analysis illustrating that the genotypic relatedness of the isolates belonging to serotypes that were recovered from animals from the same pen, block, and origin/source was much closer than that of the isolates that belonged to other groups (Fig. 4).

Clearly, our findings suggest that both origin/source (host) and environmental (pen) factors play very important roles in determining the dominant serotypes and potentially in selection of antibiotic resistance. The ecology of Salmonella within cattle populations is clearly far more complex than a simple fecal-oral mode of transmission, supporting the idea that origin/source and other ambient environmental factors are likely to be involved (49–52). One of the limitations of this study was that prior environmental samples were not collected and analyzed to determine the contributions of preexisting soil microbiota in the feedlot pens, other animals (birds and pests) that had contact with the cattle, feedlot equipment, and bacteriophages in the environment (53, 54). Therefore, the environmental factors that influenced the temporal dynamic selection of dominant serotypes of Salmonella among pens and cattle remain unclear. Further studies are needed to evaluate the effects of bacterial, environmental, and cattle-host-related factors in temporal Salmonella serotype dynamics.

MATERIALS AND METHODS

The animal field trial was approved by the West Texas A&M University/Cooperative Research, Educational and Extension Team Institutional Animal Care and Use Committee (protocol no. 05-09-15), while the microbiological assays were performed under the oversight of the Texas A&M University Institutional Biosafety Committee (IBC2017-049).

Experimental design.

Healthy crossbreed cattle (n = 134) were purchased from two different sources in west Texas. Prior antibiotic exposure information for the cattle was not available. The cattle, at initial body weights of 310 to 370 kg, were transported to the West Texas A&M University research feedlot near Canyon, TX. The cattle were blocked into three pens by source and weight and randomly allocated into a total of 12 pens (4 blocks) to control for potential confounding. Each block contained two pens in which cattle were administered either tulathromycin or ceftiofur, and a third pen contained untreated control cattle. The cattle in a total of four pens received tulathromycin (Draxxin; Zoetis, Kalamazoo, MI) at a therapeutic dose of 2.5 mg/kg injected subcutaneously (s.c.) in the neck. The cattle in another four pens received ceftiofur crystalline-free acid (Excede; Zoetis Inc., Kalamazoo, MI) at 6.6 mg/kg injected s.c. in the posterior aspect of the ear. Finally, the cattle in four pens remained as untreated controls (Table 1). Each pen housed 10 cattle. In addition, one or two control (untreated) cattle were added to each of the 12 pens as sentinels.

Feedlot and laboratory staff were blinded to the treatment status of pens and cattle. The cattle were individually identified and assigned to the pens for the duration of the study. Any animal requiring antibiotic therapy due to illness was removed from its pen and from the study. The cattle were fed diets consistent with standards in the beef-finishing industry. Major components of the diet consisted of wet corn gluten, chopped corn stalks, corn oil, steamed-flaked corn, and mineral supplements without antibiotics throughout the feeding period. Two automated watering bowls were available in each pen.

The single-dose antibiotics were administered according to the product label instructions (day 0) in early March 2016. Fecal samples were longitudinally collected throughout the feeding period, including at slaughter age, whereas lymph nodes and hide swabs were collected only one time from each animal at slaughter age. Sample collection and processing were performed according to previously published methods as follows. Fecal samples were collected, via direct palpation of the rectum using a sterile obstetric glove, from individual cattle before treatment (day 0) and after treatment on days 7, 14, 28, and 56 (Fig. 1). After day 99, the cattle from each block of three pens were sent to slaughter as a group on days 99, 120, 134, and 141, based on finishing weights. One day before slaughter, a final fecal sample (referred as terminal day) and a single 1-m² ventral hide swab were collected from individual cattle in late July and August 2016. The hide swabs were collected using sterile sponges (Whirl-Pak Speci-Sponge environmental sampling bag; Nasco, Fort Atkinson, WI) premoistened with 25 ml Butterfield’s phosphate buffer (Remel, Lenexa, KS). Following slaughter (i.e., hide and offal removal), bilateral subiliac peripheral lymph nodes were collected from each carcass that had passed both ante- and postmortem inspections conducted by federal inspectors. The lymph nodes, embedded in fat tissue, were individually placed in prelabeled 2.5-gal slider bags.

All the samples were placed in coolers with ice packs in order to maintain the samples at 4°C and shipped overnight to the laboratory at Texas A&M University, College Station, TX, for processing. The fecal samples were stored with sterile glycerol (at a ratio of 1:1) in 5-ml polypropylene tubes and preserved at −80°C until the time of microbiological processing. The lymph nodes and hide swabs were processed fresh upon arrival at the laboratory.

Salmonella isolation and confirmatory tests.

Previously published methods were used to isolate Salmonella from fecal samples, lymph nodes, and hide swabs (8, 32, 55–57). Feces were thawed and homogenized. A 0.5-g aliquot of the sample was suspended in 5 ml of tryptic soy broth (TSB) (Bacto; Becton, Dickinson, Sparks, MD). Individual lymph nodes were placed in a sterile petri dish, and the surrounding fascia and fat were trimmed aseptically with sterile scissors, taking care not to cause a rupture to the lymphoid capsular tissue. The trimmed lymph nodes were surface sterilized by parboiling in water for 3 to 6 s. The lymph nodes were placed in sterile plastic stomacher filter bags (Seward, Norfolk, United Kingdom) and pulverized with a rubber mallet. Next, 80 ml of TSB was added to the bags containing the pulverized lymph nodes, and the suspensions were homogenized at 230 rpm for 2 min using a stomacher (Circulator 400; Seward, Norfolk, United Kingdom). The hide swab sponges were suspended in 75 ml of TSB and homogenized with the stomacher at 230 rpm for 2 min.

All three TSB sample suspension types were incubated at 42°C for 3 h. Following initial incubation, 1 ml of suspension was transferred into 9 ml tetrathionate broth (Difco, Becton, Dickinson, Sparks, MD) containing 180 μl iodine solution (Remel, Lenexa, KS) and further incubated at 37°C for 24 h. Next, 100 μl of the bacterial suspension of tetrathionate broth was transferred into 10 ml Rappaport-Vassiliadis R10 broth (Difco, Becton, Dickinson, Sparks, MD) and incubated at 42°C for 18 h. Then, a 50-μl aliquot was spiral plated onto brilliant green agar (BGA) (Difco, Becton, Dickinson, Sparks, MD) via an Eddy Jet 2 spiral plater (Neutec Group Inc., Farmingdale, NY). The plates were further incubated at 37°C for 18 h.

Two round, pink colonies from each plate were streaked onto blood agar (tryptic soy agar with 5% sheep blood; Remel, Lenexa, KS) and incubated at 37°C for 18 h. Isolate colonies were subjected to a serum agglutination test using Salmonella O antiserum Poly A-I+Vi (factors 1 to 16, 19, 22 to 25, 34, and Vi; Difco, Becton, Dickinson, Sparks, MD). In addition, the matrix-assisted laser desorption ionization–time of flight mass spectrometry (MALDI-TOF MS) method was applied to confirm isolates as Salmonella. A quality control (bacterial test standard; Bruker Daltonics) and a negative control (matrix only) were included in each plate. A Microflex LT instrument (Bruker Daltonics GmbH, Leipzig, Germany) was used to measure the protein mass. Spectra were analyzed using FlexControl v.3.4 and MBT Compass software (Bruker Daltonics GmbH, Leipzig, Germany). Genus and species were identified using the main spectrum (BDAL MSP [Bruker Daltonics main spectrum profile]) library. Salmonella isolates scoring ≥2.0 were confirmed as Salmonella. Isolates confirmed by both confirmation assays were considered to be Salmonella and were carried to the next step in the analysis. Positive Salmonella isolates were preserved on CryoCare beads (Scientific Device Laboratory, Des Plaines, IL) at −80°C for further characterization.

Phenotypic antibiotic susceptibility testing.

The MICs of 14 antibiotics representing a total of nine classes of antibiotics (amoxicillin-clavulanic acid, ampicillin, azithromycin, cefoxitin, ceftiofur, ceftriaxone, chloramphenicol, ciprofloxacin, gentamicin, nalidixic acid, streptomycin, sulfamethoxazole, tetracycline, and trimethoprim-sulfamethoxazole) were determined for individual isolates using the broth microdilution method with the Sensititre system (Trek, Thermo Scientific Microbiology, Oakwood Village, OH) on NARMS Gram-negative CMV3AGNF (Thermo Fisher Scientific, Waltham, MA) plates. A detailed protocol in our laboratory has previously been described by Ohta et al. (32). The MIC for each antibiotic and each isolate was determined following the manufacturer’s protocol and interpreted as resistant, intermediate, or susceptible based on Clinical and Laboratory Standards Institute (CLSI) MIC breakpoint values where applicable (58). MIC breakpoint criteria for azithromycin and streptomycin have not been established for Salmonella enterica by the CLSI; therefore, NARMS consensus breakpoints were used for azithromycin (≥32 μg/ml) and streptomycin (≥32 μg/ml) (21; https://www.cdc.gov/narms/antibiotics-tested.html [accessed July 2019]).

DNA extraction and whole-genome sequencing.

Fecal isolates (from days 0 and 7 and the terminal day), lymph node isolates (one randomly selected from each animal), and all the hide isolates were subjected to WGS analysis. The QIAcube HT platform (Qiagen, Valencia, CA) and a QIAamp 96 DNA QIAcube HT kit (Qiagen, Valencia, CA) were used to extract single-colony bacterial DNA according to the protocol described by Ohta et al. (32). DNA purity was assessed at a 260/280-nm ratio of absorbance on a Fluostar Omega multimode microplate reader (BMG Labtech, Cary, NC). The DNA quantity was confirmed by fluorometric methods using a Quant-IT PicoGreen double-stranded DNA (dsDNA) assay kit (Thermo Fisher Scientific, Waltham, MA) on the Fluostar Omega multimode microplate reader and a Qubit 1X dsDNA HS assay kit in a Qubit 4 fluorometer (Life Technologies, Carlsbad, CA) following the manufacturers’ instructions. Illumina Nextera XT and Flex library preparation kits (Illumina, San Diego, CA) were used to generate libraries according to the manufacturer’s instructions. The fragment sizes of the libraries were evaluated using the Standard Sensitivity NGS fragment analysis kit and Fragment Analyzer automated CE system (Advanced Analytical, Des Moines, IA). Library quantities were validated using fluorometric methods as described above. Fragment smear analysis and concentrations of individual libraries were assessed based on the manufacturer’s standard. Good-quality libraries with an average of 500- to 600-bp-long fragments were pooled and denatured. The pooled libraries were run on the Illumina MiSeq platform (Illumina, San Diego, CA, USA) using Illumina MiSeq reagent v2 chemistry with paired-end 2-by-250-bp reads or v3 chemistry with paired-end 2-by-300-bp reads (59).

Bioinformatic analysis.

Web-based and command line tools on the Texas A&M University High-Performance Research Computer (HPRC) were used for bioinformatic analyses. Bad-quality reads (reads below a Phred quality score of 33, the default value [75]) and Nextera library adapters were removed using Trimmomatic v.0.36 (60). FastQC v.0.11.7 (61) was used to assess the raw-read quality, and reports were aggregated using MultiQC v.1.5 (62). Genome assembly was performed using SPAdes v.3.11.1 (63), and the assembly quality was assessed using the Quality Assessment Tool for Genome Assemblies (64). Assemblies of more than 200 contigs and less than 28× depth of coverage were considered bad quality and resequenced.

(i) Serotyping and sequence typing. SeqSero v.1.0 was used for Salmonella serotyping from raw sequencing reads (65). The SeqSero database is curated from somatic O antigen and flagellar H1 (fliC) and H2 (fljB) antigen-encoding genes and corresponds to current antigenic profiles recognized in the Kauffmann-White-LeMinor Scheme (76). STs were determined based on sequence matches found in the legacy MLST database based on alleles of seven housekeeping genes (aroC, dnaN, hemD, hisD, purE, sucA, and thrA) using raw sequencing reads with SRST2 (Short Read Sequence Typing for Bacterial Pathogens) v.0.2.0 (66) and the S. enterica MLST database curated from the Public Databases for the Molecular Typing and Microbial Genome Diversity Platform (http://www.pubmlst.org [accessed July 2019]).

(ii) Phylogenetic analysis. Phylogenetic analyses were conducted to explore patterns within pen, block, and origin/source among serotypes using core genome single-nucleotide polymorphism (SNP) analysis. A complete reference genome from one of the most prevalent serotypes was selected using the Similar Genome Finder tool within PATRIC (the Pathosystems Resource Integration Center) for the phylogenetic analysis, keeping the threshold values at the default (50 maximum hits, 1 P value, and 0.05 Mash/MinHash distance [estimating the distance based on rate of sequence mutation]) (67, 68). The prophage regions of the reference strain were detected using PHASTER (Phage Search Tool Enhanced Release) (69), and these regions were further masked using BEDTools (70). A total of 399 Salmonella genomes were aligned to the reference strain using core genome SNP analysis with ParSNP v.1.2 (71). Extended multi-FASTA files obtained from the alignments were converted to multi-FASTA files using HarvestTools v.1.2 (71). The Model-test NG v.0.1.5 tool was utilized to determine the nucleotide substitution model to generate the maximum-likelihood tree using the multi-FASTA file (72). The phylogenetic tree was inferred using the selected model with IQ-tree v.1.6.10, including bootstrap values with 1,000 iterations (73). Later, the phylogenetic tree was visualized and graphics were generated using interactive Tree of Life (iTOL) (74) to explore within-serotype genotypic relatedness by pen, block, and cattle origin/source.

Statistical analysis. (i) Descriptive statistics.

The data types obtained from our study were as follows: (i) the number of samples with Salmonella following enrichment (binary; aggregated into prevalence), (ii) phenotypic antibiotic resistance (binary; CLSI or NARMS ordered interpretive criteria reclassified as resistant or susceptible, including the intermediate class), and (iii) serotype (multinomial categorical) data. For descriptive statistics, the data were initially cross-tabulated across sampling days and treatments. Unadjusted phenotypic antibiotic resistance data and confidence intervals were examined using the Clopper-Pearson exact method (77). Graphical visualizations and tables were generated using Microsoft Excel 2016 (Microsoft Corporation, Redmond, WA). For statistical analyses, slaughter samples were collapsed to a single period (approximately day 112) for all sample types. Univariate analyses were performed using the global unadjusted likelihood ratio chi-square test to address the pen and treatment effects on serotype presence in hide samples.

(ii) Multivariable regression analyses. Stata v.15.1 (StataCorp LLC, College Station, TX) was used for multivariable analyses of Salmonella prevalence.

Multilevel mixed-effects logistic regression analyses were performed separately for fecal, lymph node, and hide samples. Prior to the regression analysis, correlation structures were examined, and pen level dependencies were determined to be exchangeable for all sample types; in contrast, individual animal dependencies were set as unstructured for feces (n = 6 per animal) and lymph nodes (n = 2 per animal). For fecal samples, fixed effects of sampling day and treatment and their interaction terms were forced into all the models. Lymph node- and hide-related models utilized treatment as the sole fixed effect. Individual animal identifiers and pen identifiers were considered in models as random-effect variables. Day 0 and control treatments were considered baseline values. Likelihood ratio tests for each fixed effect or interaction were assessed stepwise from the full model to the intercept-only model. Day and treatment predictive marginal means and 95% CI (proportions) were computed for each final model. Predictive margins were plotted or visualized with bar graphs. Multiple pairwise comparisons of average marginal effects were performed and reported as adjusted for multiple comparisons with Bonferroni’s method for each contrast of fixed effect and interactions. Residual intraclass correlation coefficients at the pen and individual animal cluster levels were determined following regression analysis.

Accession number(s).

Sequencing data from this project can be found under NCBI BioProject accession number PRJNA521731.