Impact of “Raised without Antibiotics” Beef Cattle Production Practices on Occurrences of Antimicrobial Resistance

ABSTRACT

The specific antimicrobial resistance (AMR) decreases that can be expected from reducing antimicrobial (AM) use in U.S. beef production have not been defined. To address this data gap, feces were recovered from 36 lots of “raised without antibiotics” (RWA) and 36 lots of “conventional” (CONV) beef cattle. Samples (n = 719) were collected during harvest and distributed over a year. AMR was assessed by (i) the culture of six AM-resistant bacteria (ARB), (ii) quantitative PCR (qPCR) for 10 AMR genes (ARGs), (iii) a qPCR array of 84 ARGs, and (iv) metagenomic sequencing. Generally, AMR levels were similar, but some were higher in CONV beef cattle. The prevalence of third-generation cephalosporin-resistant (3GC^r) Escherichia coli was marginally different between production systems (CONV, 47.5%; RWA, 34.8%; P = 0.04), but the seasonal effect (summer, 92.8%; winter, 48.3%; P < 0.01) was greater. Erythromycin-resistant (ERY^r) Enterococcus sp. concentrations significantly differed between production systems (CONV, 1.91 log₁₀ CFU/g; RWA, 0.73 log₁₀ CFU/g; P < 0.01). Levels of aadA1, ant(6)-I, bla_ACI, erm(A), erm(B), erm(C), erm(F), erm(Q), tet(A), tet(B), tet(M), and tet(X) ARGs were higher (P < 0.05) in the CONV system. Aggregate abundances of all 43 ARGs detected by metagenomic sequencing and the aggregate abundances of ARGs in the aminoglycoside, β-lactam, macrolide-lincosamide-streptogramin B (MLS), and tetracycline AM classes did not differ (log₂ fold change < 1.0) between CONV and RWA systems. These results suggest that further reductions of AM use in U.S. beef cattle production may not yield significant AMR reductions beyond MLS and tetracycline resistance.

IMPORTANCE The majority of antimicrobial (AM) use in the United States is for food-animal production, leading to concerns that typical AM use patterns during “conventional” (CONV) beef cattle production in the United States contribute broadly to antimicrobial resistance (AMR) occurrence. In the present study, levels of AMR were generally similar between CONV and “raised without antibiotics” (RWA) cattle. Only a limited number of modest AMR increases was observed in CONV cattle, primarily involving macrolide-lincosamide-streptogramin B (MLS) and tetracycline resistance. Macrolides (tylosin) and tetracyclines (chlortetracycline) are administered in-feed for relatively long durations to reduce liver abscesses. To ensure judicious AM use, the animal health, economic, and AMR impacts of shorter duration in-feed administration of these AMs should be examined. However, given the modest AMR reductions observed, further reductions of AM use in U.S. beef cattle production may not yield significant AMR reductions beyond MLS and tetracycline resistance.

INTRODUCTION

The contribution of antimicrobial (AM) use in food-animal production agriculture to bacterial antimicrobial resistance (AMR) is an important question (1–3). In the United States, food-animal production agriculture accounts for approximately 80% of all AM use (by mass), although 40% of this use is ionophores, which do not impact AMR (4). Food-animal AM use is perceived to be an important contributor to AM-resistant infections in humans, although there is substantial debate regarding the relative contributions of each food-animal commodity and human medical uses (5–7). Food-animal production without AMs has been offered as a practice to reduce AM-resistant human infections, but the effect is difficult to predict because AM-resistant bacteria (ARB) are present in nearly all environments, and “background” AMR levels (AMR occurrence without anthropogenic impacts) cannot be defined for most environments (8–10).

Defining the impact of AM use in food-animal agriculture on human health is hindered by the lack of comprehensive risk assessments due to data gaps. Traditionally, AMR is studied using culture-dependent methods, especially, culture of Escherichia coli. In the context of beef cattle production, a previous study determined the susceptibilities to 16 AMs of 8,882 fecal E. coli isolates from beef cattle produced with and without AMs. Statistically different levels of resistance were found for only 4 AMs: chloramphenicol, streptomycin, sulfamethoxazole, and tetracycline. For chloramphenicol, streptomycin, and sulfamethoxazole, the percentage prevalence differences were considered small (<4% lower in cattle produced without AMs), whereas for tetracycline, the difference was 9% lower in cattle produced without AMs (11).

Most of the food-animal production microbiome is unculturable, and numerous AMR genes (ARGs) are harbored on mobile genetic elements that are potentially capable of horizontal gene transfer between diverse bacterial species. Thus, culture-independent methods are required for comprehensive AMR analysis. Quantitative PCR (qPCR) and gene arrays can be used to detect specific ARGs regardless of bacterial host, but the ARGs assessed are determined a priori. The alignment of metagenomic sequences to comprehensive ARG sequence databases may ameliorate these biases, but, as ARGs usually represent less than 1% of bacterial metagenomic sequences, sensitivity is a concern (12–15). The inability to determine the bacterial host and phenotype associated with ARGs is a common drawback of culture-independent methods (qPCR, gene array, and metagenomic sequencing).

From February 2014 to January 2015, we had access to a large plant that processed beef cattle from “raised without antibiotics” (RWA) and “conventional” (CONV) production systems. This presented an opportunity to provide a more comprehensive assessment of AMR present in the feces of RWA and CONV U.S. beef cattle, although detailed AM treatment records for the CONV cattle could not be obtained. The RWA cattle were produced without AMs regardless of administration route, purpose, or classification. CONV cattle were produced with no restrictions on AM use other than meeting regulatory compliance requirements. Feces were recovered from the colons of 10 animals/lot for 36 lots of CONV cattle and 36 lots of RWA cattle over a 12-month period. Feces were recovered from colons to avoid cross-contamination associated with harvest (16, 17).

Multiple methods were employed to ensure comprehensive assessment. First, all fecal samples were cultured for AM-resistant E. coli, Salmonella enterica, and Enterococcus spp. Second, metagenomic DNA (mgDNA) was isolated from each sample. The mgDNA samples were pooled by lot, and the abundances of 10 ARGs were determined by qPCR. Third, the lot-pooled mgDNAs were used to determine the prevalences of 84 ARGs using a commercially available ARG qPCR array. Fourth, the lot-pooled mgDNAs were sequenced and compared to the MEGARes database to determine ARG relative abundance (18). Finally, bacterial phylogenetic analyses were performed by sequencing the PCR-amplified 16S rRNA gene sequences and by metagenomic sequencing.

RESULTS

Antimicrobial-resistant E. coli.

Generic E. coli (defined as all E. coli regardless of susceptibility) were present at quantifiable concentrations (≥2.30 log₁₀ CFU/g) in all 719 fecal samples. The mean generic E. coli concentrations (Fig. 1A) differed significantly by season (P < 0.01), as higher levels occurred during summer (6.32 log₁₀ CFU/g) and fall (6.31 log₁₀ CFU/g) than during spring (5.97 log₁₀ CFU/g) and winter (5.82 log₁₀ CFU/g). The mean generic E. coli concentrations did not differ between production systems (CONV, 6.03 log₁₀ CFU/g; RWA, 6.18 log₁₀ CFU/g; P = 0.15). Tetracycline-resistant (TET^r) E. coli were present in all samples, and TET^r E. coli was quantified in 97.4% of the samples. Similar to those of generic E. coli, TET^r E. coli mean concentrations differed by season (P < 0.01), as higher levels occurred during summer (5.36 log₁₀ CFU/g) and fall (5.41 log₁₀ CFU/g) than during winter (4.38 log₁₀ CFU/g) (Fig. 1B). TET^r E. coli mean concentrations did not differ between production systems (CONV, 5.22 log₁₀ CFU/g; RWA, 4.98 log₁₀ CFU/g; P = 0.08). Generic and TET^r E. coli concentrations were strongly correlated (Spearman's ρ = 0.72, P < 0.01) (Fig. 1).

FIG 1.

(A) Monthly mean concentrations of generic (regardless of antimicrobial susceptibility) Escherichia coli. (B) Monthly mean concentrations of tetracycline-resistant (TET^r) E. coli. Red circles indicate values for conventionally produced (CONV) beef cattle. Green squares indicate values for beef cattle raised without antibiotics (RWA). Each data point represents the mean concentration from 30 samples (except March RWA, which represents 29 samples); error bars indicate 95% confidence intervals.

Sulfamethoxazole-trimethoprim-resistant (COT^r) E. coli and third-generation cephalosporin-resistant (3GC^r) E. coli detection rates (here referred to as prevalences) were compared, since quantifiable concentrations were present in only 19.8% and 2.8% of the samples, respectively. Interestingly, the highest COT^r E. coli prevalences occurred during summer (92.8%), followed by fall (78.9%), winter (48.3%), and spring (44.0%) (Fig. 2A). COT^r E. coli prevalence was affected by season (P < 0.01) but not by production system (CONV, 71.4%; RWA, 60.6%; P = 0.07). Similar to that of COT^r E. coli, 3GC^r E. coli prevalence was highest during summer (73.9%), followed by fall (55.0%), spring (22.9%), and winter (12.8%). Although 3GC^r E. coli prevalence was affected by production system (CONV, 47.5%; RWA, 34.8%; P = 0.04), the seasonal effect was much greater (P < 0.01) (Fig. 2B).

FIG 2.

(A) Monthly mean sulfamethoxazole-trimethoprim-resistant (COT^r) Escherichia coli prevalences. (B) Monthly mean third-generation cephalosporin-resistant (3GC^r) E. coli prevalences. Red circles indicate values for conventionally produced (CONV) beef cattle. Green squares indicate values for beef cattle raised without antibiotics (RWA). Each data point represents the mean percent prevalence from 3 lots.

Antimicrobial-resistant Salmonella enterica.

Overall, generic Salmonella prevalence was 13.8%, and quantifiable concentrations (≥2.30 log₁₀ CFU/g) were present in only 1.7% of samples. Generic Salmonella prevalence did not differ by production system (CONV, 11.7%; RWA, 15.9%; P = 0.42) (see Fig. S1 in the supplemental material). Generic Salmonella prevalence differed by season (P < 0.01), as higher levels occurred during summer (38.3%) than during fall (9.4%), winter (5.6%), and spring (1.7%). Only one sample (a CONV sample) was positive for 3GC^r Salmonella. Nalidixic acid-resistant Salmonella was not detected from any sample.

Antimicrobial-resistant Enterococcus spp.

The prevalence of generic Enterococcus spp. was 99.7%, and quantifiable concentrations were present in 76.1% of the samples. Erythromycin-resistant (ERY^r) Enterococcus prevalence was 88.4% and was quantified from 35.0% of the samples. Both production system and season (P ≤ 0.01) affected generic Enterococcus (Fig. 3A) and ERY^r Enterococcus (Fig. 3B) concentrations. The mean concentration of generic Enterococcus was lower (P = 0.01) in CONV (2.87 log₁₀ CFU/g) than in RWA (3.38 log₁₀ CFU/g) systems, whereas the mean concentration of ERY^r Enterococcus was higher (P < 0.01) in CONV (1.91 log₁₀ CFU/g) than in RWA (0.73 log₁₀ CFU/g) systems. Thus, the ERY^r proportion of generic Enterococcus was larger in CONV than RWA systems.

FIG 3.

(A) Monthly mean concentrations of generic (regardless of antimicrobial susceptibility) Enterococcus spp. (B) Monthly mean concentrations of erythromycin-resistant (ERY^r) Enterococcus spp. Red circles indicate values for conventionally produced (CONV) beef cattle. Green squares indicate values for beef cattle raised without antibiotics (RWA). Each data point represents the mean concentration from 30 samples (except March RWA, which represents 29 samples); error bars indicate 95% confidence intervals.

Antimicrobial resistance gene abundance as determined by qPCR.

For the 10 genes examined by qPCR, the criteria for significantly different abundance were log₂ fold change ≥1.0 and a P value < 0.05. Season did not have a significant effect on any of the ARG abundances. The abundances of the three tetracycline ARGs examined, tet(A), tet(B), and tet(M), were all higher (P ≤ 0.01) in CONV than in RWA systems, with log₂ fold changes of 1.3, 1.9, and 1.9, respectively (Fig. 4). The abundance of the macrolide-lincosamide-streptogramin B (MLS) ARG examined, erm(B), was 2.6 log₂-fold higher (P < 0.01) in CONV than in RWA systems (Fig. 4). Two aminoglycoside ARGs were examined, namely, aac(6′)-Ie-aph(2″)-Ia and aadA1. The aac(6′)-Ie-aph(2″)-Ia ARG was not detected. The aadA1 ARG was 3.8 log₂-fold higher (P < 0.01) in CONV than in RWA systems. Four β-lactam ARGs were examined, namely, bla_CMY-2, bla_CTX-M, bla_KPC-2, and mecA. The mecA ARG was not detected. The log₂ fold changes in the abundances of bla_CMY-2, bla_CTX-M, and bla_KPC-2 between CONV and RWA systems were <1.0 (Fig. 4).

FIG 4.

16S rRNA-normalized abundances of eight antimicrobial resistance genes. Red circles indicate values for conventionally produced (CONV) beef cattle. Green squares indicate values for beef cattle raised without antibiotics (RWA). Yellow bars and whiskers indicate geometric means and 95% confidence intervals, respectively. For each antimicrobial resistance gene, n = 72.

Prevalences of 84 antimicrobial resistance genes as determined by qPCR array.

Of the 84 ARGs on the array, only 29 were detected. Of the 29 detected ARGs, only 7 were present in >4 lots (see Table S1), including the aminoglycoside ARG aadA1, the MLS ARGs erm(A), erm(B), erm(C), and mef(A), and the tetracycline ARGs tet(A) and tet(B). The prevalences of the following ARGs were higher (P ≤ 0.01) for CONV than for RWA systems: aadA1 (CONV, 91.7%; RWA, 19.4%), erm(A) (CONV, 41.7%; RWA, 2.8%), erm(C) (CONV, 52.8%; RWA, 2.8%), and tet(B) (CONV, 97.2%; RWA, 75.0%).

Antimicrobial resistance gene abundances as determined by metagenomic sequencing.

Metagenomic DNAs were sequenced to a geometric mean depth of 7.60 × 10⁷ reads/lot. Following trimming and filtering, the geometric mean depth was 6.85 × 10⁷ reads/lot. A comparison of these sequences to the MEGARes database of >4,000 curated ARG sequences resulted in a geometric mean aligned reads/lot of 9.02 × 10⁴, indicating that approximately 0.1% of the total sequencing reads corresponded to ARG sequences in MEGARes. Overall, 175 ARG sequences were identified and classified into 95 ARGs, 27 resistance mechanisms, and 12 AM classes (see Table S2). ARG databases, including the MEGARes database, contain “core structural” genes that are present in their bacterial hosts' “core” genomes (i.e., all or nearly all members of the species harbor the gene). These core structural genes include, for example, the mdtABCD operon, which is annotated as a multidrug efflux pump that is present in many Enterobacteriaceae species and only confers reduced susceptibility to certain compounds (novobiocin and deoxycholate) when overexpressed (19–21). Of the 95 ARGs identified, 52 were core structural genes that did not possess a specific mutation(s) associated with reduced susceptibility (see Table S3). These 52 core structural ARGs were excluded from further analyses.

The remaining 43 ARGs belong to 14 resistance mechanisms and 7 AM classes (Table 1). Aggregated by AM class, the most abundant resistances are for tetracyclines, MLS, β-lactams, and aminoglycosides (Fig. 5A). The ARGs in these four AM classes were detected in every lot. ARGs in the fluoroquinolones, rifampin, and phenicols AM classes were detected in 36.1%, 11.1%, and 6.9% of lots, respectively. Neither the total ARG abundance (sum of the abundances of the 43 ARGs) nor any of the AM class abundances met the criteria for differential abundance between production systems, i.e., ≥1.0 log₂-fold change and P < 0.05.

TABLE 1.

Forty-three antimicrobial resistance genes aligned to metagenomic sequences

Antimicrobial class	Resistance mechanism	ARG(s)
Aminoglycoside	Aminoglycoside N-acetyltransferase	sat
	Aminoglycoside O-nucleotidyltransferase	ant(6)-I, ant(9)-I
	Aminoglycoside O-phosphotransferase	aph(3′)-I, aph(6)-I
β-lactam	Class A β-lactamase	blaACI, blaCblA, blaCTX-M, blaROB, cfxA
	Class D β-lactamase	blaOXA
Fluoroquinolone	Fluoroquinolone-resistant DNA topoisomerase	gyrA, gyrB, parC, parE
MLSa	23S rRNA methyltransferase	erm(A), erm(B), erm(C), erm(F), erm(G), erm(Q), erm(R), erm(T), erm(X)
	Lincosamide nucleotidyltransferase	lnu(C)
	Macrolide efflux pump	mef(A), mel, msr(D)
Phenicol	Chloramphenicol acetyltransferase	cat
Rifampin	Rifampin-resistant β-subunit of RNA polymerase	rpoB
Tetracycline	Tetracycline inactivation enzyme	tet(X)
	Tetracycline resistance efflux pump	tet(A), tet(B), tet(C), tet(D), tet(L), tet(40)
	Tetracycline ribosomal protection protein	tet(M), tet(O), tet(Q), tet(W), tet(32), tet(44)

^aMLS, macrolide-lincosamide-streptogramin B.

FIG 5.

Cumulative sum scaling (CSS)-normalized abundances of antimicrobial resistance genes (ARGs) in metagenomic sequences. Red circles indicate values for conventionally produced (CONV) beef cattle. Green squares indicate values for beef cattle produced without antibiotics (RWA). Yellow bars and whiskers indicate geometric means and 95% confidence intervals, respectively. (A) ARG abundances aggregated by antimicrobial class. (B) First to 11th most abundant ARGs. (C) Twelfth to 17th most abundant ARGs. (D) Eighteenth to 21st most abundant ARGs. For each ARG or each antimicrobial class, n = 72. All, sum of 43 detected ARGs; Tet, tetracyclines; MLS, macrolides-lincosamides-streptogramin B; β-lac, β-lactams; Amgly, aminoglycosides; Fq, fluoroquinolones; Rif, rifampin; Phncl, phenicols.

The 11 most abundant ARGs comprised 8 tetracycline, 2 MLS, and 1 β-lactamase AM class ARGs. The abundances of these 11 ARGs did not differ between CONV and RWA systems (Fig. 5B). The 12th to 17th most abundant ARGs were more abundant (P ≤ 0.01) in the CONV system (Fig. 5C). The ant(6)-I ARG (1.0 log₂-fold higher in the CONV system) encodes an aminoglycoside O-nucleotidyltransferase associated with streptomycin resistance. The bla_ACI ARG (1.4 log₂-fold higher in the CONV system) encodes a class A β-lactamase. The erm(F) and erm(Q) ARGs (2.8 log₂- and 1.7 log₂-fold higher, respectively, in the CONV system) encode 23S rRNA methyltransferases associated with MLS resistance. The tet(M) ARG (2.5 log₂-fold higher in the CONV system) encodes a ribosomal protection protein associated with tetracycline resistance (22). The tet(X) ARG (2.8 log₂-fold higher in the CONV system) encodes a monooxygenase that inactivates tetracycline class AMs. The abundances of the 18th to 21st most abundant ARGs did not differ between production systems (Fig. 5D). All other ARGs were detected in <50% of lots; thus, the abundances of these ARGs were not compared.

Of the 14 resistance mechanisms detected (Table 1), only three were more abundant (P < 0.01) in the CONV system, including 23S rRNA methyltransferases (1.5 log₂-fold higher in the CONV system), aminoglycoside O-nucleotidyltransferases (1.0 log₂-fold higher in the CONV system), and tetracycline inactivation enzymes (2.8 log₂-fold higher in the CONV system). The 23S rRNA methyltransferase resistance mechanism was represented by two ARGs with higher abundances in the CONV system, namely, erm(Q) and erm(F). The aminoglycoside O-nucleotidyltransferases resistance mechanism was represented by the ant(6)-I ARG, which had higher abundance in the CONV system. The tetracycline inactivation enzymes resistance mechanism was represented by only one ARG, tet(X). The tet(M) ARG was more abundant in the CONV system, but this did not result in a higher abundance of ribosomal protection proteins, its resistance mechanism, since the other tet genes in this mechanism (O, Q, W, 32, and 44) were more abundant than tet(M) and did not differ between production systems. Similarly, the bla_ACI ARG was higher in the CONV system, but the aggregate abundance of class A β-lactamase ARGs did not differ between production systems, because the ARG cfxA was more abundant than bla_ACI and did not differ between production systems (Fig. 5B).

Bacterial phylogenetic analysis using 16S rRNA sequences and metagenomic sequences.

Phylogenetic analyses were performed using 16S rRNA sequences and metagenomic sequences. The overall bacterial community structures in feces from CONV and RWA cattle were similar. Detailed analyses are provided in Text S1.

DISCUSSION

In general, the AMR occurrences and microbiomes of feces from CONV and RWA cattle were similar. Perhaps the broadest measures of AMR occurrence in this study were the aggregated metagenomic sequencing abundances of ARGs by AM class and all 43 ARGs detected (Fig. 5A). None of these aggregated abundances differed between CONV and RWA systems. Most of the differences in AMR occurrence involved specific MLS or tetracycline ARGs or ARB (Table 2). Both tylosin ([TYL] a macrolide class AM) and chlortetracycline ([CTC] a tetracycline class AM) are FDA-approved for in-feed administration to beef cattle for the prevention of liver abscesses. Treatment records for the CONV cattle in this study were unobtainable, but an extensive USDA survey indicates that approximately 71% and 20% of U.S. beef cattle receive in-feed TYL and CTC, respectively (23). Informal discussions with beef cattle producers in the region the processing plant serves support this approximation. Additionally, the producers indicated that in-feed AM administration for the prevention of liver abscesses usually occurs until harvest, because there is no legally mandated withdrawal period. Thus, we conclude that the higher levels of MLS and tetracycline resistance observed in CONV cattle are plausibly related to this practice. We hypothesize that limiting in-feed CTC and TYL administrations for the prevention of liver abscesses to their most effective period will effectively mitigate AMR impacts. This hypothesis is based on our previous research in beef cattle demonstrating that AMR impacts are transient for one-time ceftiofur injections to treat bacterial infections and for 5-day in-feed CTC administrations to prevent respiratory disease (24, 25).

TABLE 2.

Antimicrobial resistances significantly higher in feces from conventional (CONV) than from “raised without antibiotics” cattle

Antimicrobial class	Antimicrobial-resistant bacteria or antimicrobial resistance gene	Method	Amount higher in CONV cattle	P value
Aminoglycoside	aadA1	Array	72.3% more prevalent	<0.01
	aadA1	qPCR	3.8 log2-fold more abundant	<0.01
	ant(6)-I	Metagenomic sequencing	1.0 log2-fold more abundant	<0.01
β-Lactam	3GCr E. coli	Culture	12.7% more prevalent	0.04
	blaACI	Metagenomic sequencing	1.4 log2-fold more abundant	<0.01
MLS	ERYr Enterococcus	Culture	1.18 log10 CFU/g	<0.01
	erm(A)	Array	38.9% more prevalent	<0.01
	erm(B)	qPCR	2.6 log2-fold more abundant	<0.01
	erm(C)	Array	50.0% more prevalent	<0.01
	erm(F)	Metagenomic sequencing	1.7 log2-fold more abundant	<0.01
	erm(Q)	Metagenomic sequencing	2.8 log2-fold more abundant	<0.01
Tetracycline	tet(A)	qPCR	1.3 log2-fold more abundant	0.01
	tet(B)	Array	22.2% more prevalent	0.01
	tet(B)	qPCR	1.9 log2-fold more abundant	<0.01
	tet(M)	Metagenomic sequencing	2.5 log2-fold more abundant	0.01
	tet(M)	qPCR	1.9 log2-fold more abundant	<0.01
	tet(X)	Metagenomic sequencing	2.8 log2-fold more abundant	<0.01

Third-generation cephalosporins (3GCs) are critically important to human medicine, and resistance complicates treatment of human E. coli infections (26, 27). CTC use has been hypothesized to increase 3GC^r E. coli because 3GC resistance in E. coli is often conferred by a bla_CMY-2 gene contained in an IncA/C plasmid that also contains a tet(A) gene conferring tetracycline resistance (28–30). While 3GC^r E. coli prevalence and bla_ACI abundance were higher in the CONV system (Table 2), these β-lactam resistance differences were judged “small” and unlikely to be biologically relevant. The CONV and RWA system bla_ACI abundance distributions largely overlapped, and the more abundant β-lactam ARG cfxA did not differ between CONV and RWA systems (Fig. 5B). For 3GC^r E. coli prevalence, the effect of season (P < 0.01) was larger than that of the production system (P = 0.04) (Fig. 2B). 3CG^r E. coli was detected in only 41.2% of samples and quantified (concentration, ≥2.30 log₁₀ CFU/g) in 2.8% of samples. The resulting 3CG^r E. coli estimated mean concentrations were −0.64 log₁₀ CFU/g for CONV and −1.05 log₁₀ CFU/g for RWA systems. The coselection of 3CG^r E. coli by CTC use is unlikely, since these 3CG^r E. coli estimated mean concentrations were several orders of magnitude lower than the mean generic E. coli (CONV, 6.03 log₁₀ CFU/g; RWA, 6.18 log₁₀ CFU/g) and TET^r E. coli (CONV, 5.22 log₁₀ CFU/g; RWA, 4.98 log₁₀ CFU/g) concentrations. Similarly, 3GC^r E. coli concentrations were several orders of magnitude lower than generic and TET^r E. coli concentrations in a study recently published by our research group demonstrating that 5-day in-feed administration of CTC did not increase levels of 3GC^r E. coli (25).

The aminoglycoside resistance genes aadA1 and ant(6)-I were higher in the CONV system (Table 2). Both aadA1 and ant(6)-I are associated with resistance to “lower-level” aminoglycosides (streptomycin) and are broadly distributed (31). This result cannot be easily attributed to AM use practices, since aminoglycoside use during U.S. beef cattle production is extremely low (23). Coresistance could be a factor; aadA1 is known to reside with other ARGs on mobile genetic elements, including integrons (32).

Comparisons of culture-independent methods for the detection of ARGs were not a goal of this study. Nevertheless, qPCR was a more sensitive method than metagenomic sequencing and arrays for the detection of ARGs (Table 3). The low proportion (approximately 0.1%) of ARG sequences in the metagenome constrains the sensitivity of metagenomic sequencing for detecting ARGs. Other published metagenomic sequencing projects report similar proportions of ARGs in metagenomes (12, 33). Indeed, an examination of 10,181 metagenomic surveys for ARGs led Fitzpatrick and Walsh (33) to pose the question, “If we do not detect the ARG, is it in fact not there or just below the limits of our techniques?” We now provide a well-supported (Table 3) response, “ARGs not detected by metagenomic sequencing may be present.”

TABLE 3.

Antimicrobial resistance gene detection by qPCR, qPCR array, and metagenomic sequencing

ARG	% detection (n = 72)
	qPCR	qPCR array	Metagenomic sequencing
aadA1	100.0	55.6	0.0
aac(6′)-Ie-aph(2″)-Ia	0.0	0.0	0.0
blaCMY-2	100.0	5.6	0.0
blaCTX-M	100.0	0.0	27.2
blaKPC-2	100.0	0.0	0.0
erm(B)	100.0	88.9	44.4
mecA	0.0	0.0	0.0
tet(A)	100.0	93.1	98.6
tet(B)	100.0	86.1	100.0
tet(M)	100.0	NAa	84.7

^aNA, not applicable, tet(M) was not present on the array.

In January 2017, the U.S. Food and Drug Administration (FDA) limited the in-feed use of AMs important for human medicine to approved applications for disease treatment, prevention, or control with a written Veterinary Feed Directive specific to the targeted animal population (34). Nevertheless, the contributions to AMR of AM uses during food-animal production relative to other AM uses (such as human clinical uses) remain controversial. We acknowledge that interpretations of these data regarding human and environmental health impacts are complicated by the lack of risk models, risk assessments, and importance rankings of ARB and ARGs (2, 5).

Consideration of this study's results along with other available information provides several lines of evidence that cast doubt on the likelihood of broad detrimental impacts from typical AM use during U.S. beef production on human or environmental health relative to those from other sources (i.e., “driving resistance”). First, this study demonstrates that the majority of the AMR differences between U.S. CONV and RWA beef production involve resistance to macrolides or tetracyclines (Table 2). Second, the final meat products produced from CONV and RWA beef cattle may not have different AMR levels, since sanitary processing interventions effectively remove ARB and ARGs on carcasses (12, 35) and a significant proportion of the microbiome present on meat products is attributed to processing and distribution plant microbial flora rather than fecal and production microbial flora (36–40). Studies comparing AMR occurrences between CONV and RWA U.S. retail beef products are few, rely on culture methods, and have found similar levels of resistance (41). Research is under way in our laboratories using the methods described in this report to address this data gap. Third, RWA beef cattle require approximately 7 more weeks than CONV cattle to reach market weight (11), and thus the AMR differences observed in this study may not alter the environmental AMR impact if the additional manure produced by RWA cattle is considered. Fourth, our group recently demonstrated that effluent from a human wastewater treatment plant contained more ARGs and more ARGs associated with β-lactam resistance, including carbapenem resistance, than cattle and swine wastes (42). Finally, the largest differences between production systems were for aadA1, erm(B), erm(Q), tet(M), tet(X), and ERY^r Enterococcus (Table 2), all of which could be considered “ubiquitous,” since they are frequently identified in multiple environments, particularly human feces, fecal-impacted environments, and human-inhabited environments (14, 42–45).

Continued research to reduce AM use in U.S. beef cattle production is important. The majority of the AMR differences identified by this study involved MLS and tetracycline AM class resistances, and AMs in both of these classes (TYL and CTC) are routinely included in beef cattle feed until harvest (Table 2). Therefore, to support judicious AM use, the animal health, AMR, and economic impacts of reduced TYL and CTC in-feed administration should be comprehensively studied. Equally, this study suggests that typical AM use during U.S. beef cattle production contributes few additional resistances beyond MLS and tetracyclines. Thus, additional AM use restrictions on U.S. beef cattle production should not be expected to yield major (i.e., consistently detectable) resistance reductions across all AM classes, including classes critically important to human medicine.

MATERIALS AND METHODS

Sample collection.

Seventy-two lots of cattle were sampled. Samples were obtained during monthly visits to a commercial beef processing plant from February 2014 to January 2015. The goal each month was to obtain 60 fecal samples: 10 from each of three lots of RWA cattle and 10 from each of three lots of CONV cattle. A lot was defined as a group of cattle from a production site that were harvested together. For each month, the six lots represented six different production sites. Overall, 359 and 360 fecal samples were obtained from RWA and CONV cattle, respectively. For one of the March 2014 lots of RWA cattle, feces could only be recovered from 9 colons. Samples were otherwise equally distributed by month and production system among the 72 lots. Feces were harvested from colons after evisceration by making an incision in the colon and then removing up to 30 g of fecal material into a sealable plastic bag. Gloves and scissors were changed between samples. Samples were transported to the laboratory in coolers with ice packs and held at 4°C until processing.

Antimicrobial use.

RWA cattle were produced without using AMs, regardless of the AM administration route (in-feed, in-water, subcutaneous injection, etc.), purpose (production, preventative, and therapeutic), or importance of the AM to human medicine (certain classes of AMs, e.g., ionophores, are classified by the FDA as “not currently important to human medicine”). For CONV cattle, no data regarding AM use were available.

Bacterial culture.

Detailed descriptions of the bacterial culture methods are provided in Text S2 in the supplemental material. Briefly, for each sample, 10 g of feces was transferred to a filter barrier bag and suspended by adding 90 ml of phosphate-buffered tryptic soy broth (TSB-PO). Aliquots were plated on media specific for the following bacteria: generic E. coli, TET^r E. coli, COT^r E. coli, 3GC^r E. coli, generic Salmonella, 3GC^r Salmonella, nalidixic acid-resistant (NAL^r) Salmonella, generic Enterococcus, and ERY^r Enterococcus. TSB-PO suspensions were incubated overnight and subcultured into secondary enrichment media specific for generic E. coli, TET^r E. coli, COT^r E. coli, 3GC^r E. coli, generic Salmonella, and generic Enterococcus. Following an overnight incubation, the secondary enrichments were struck onto media specific for the following bacteria: generic E. coli, TET^r E. coli, COT^r E. coli, 3GC^r E. coli, generic Salmonella, 3GC^r Salmonella, NAL^r Salmonella, generic Enterococcus, and ERY^r Enterococcus. From each sample, two presumptive colonies were confirmed by PCR for species-specific genes. TET^r E. coli were cultured beginning in July.

Metagenomic DNA isolation.

mgDNA was extracted from 250 mg of fecal samples using a PowerSoil DNA isolation kit (MoBio Laboratories, Inc., Carlsbad, CA) according to the manufacturer's instructions and quantified using a NanoDrop 2000 spectrophotometer (Thermo Scientific, Waltham, MA). Quality mgDNA in sufficient quantities for the purposes of this study was recovered from all 360 CONV samples and 357 of the 359 RWA samples. This resulted in 10 mgDNA samples from 10 fecal samples for all 72 lots except two: a March 2014 RWA lot that had 8 mgDNA samples and another March 2014 RWA lot that had 9 mgDNA samples.

qPCR.

Detailed descriptions of the qPCR methods are provided in Text S2. Briefly, equal amounts of DNA (200 ng) from each sample were pooled by lot. Then, 40 ng of DNA was amplified using 2.5 μM forward and reverse primers (Table 4) and 10 μl of Fast SYBR green master mix (ABI) in a 20-μl reaction. Standard curves for each primer pair were generated using DNA extracted from bacterial strains harboring the target genes (Table 5). Thermal cycling was performed on an ABI 7500 Fast real-time PCR system with the following conditions: 95°C for 30 s, followed by 40 cycles of 95°C for 3 s and 60°C for 30 s. Three replicate runs were performed.

TABLE 4.

Oligonucleotide primers used for qPCR

ARG	Primer sequence		Reference(s)
	Forward	Reverse
aac(6′)-Ie-aph(2″)-Ia	ACATGAATTACACGAGGGCA	ATCGCCGTCTAGTTCTGCTG	This study
aadA1	GCGAGCTTTGATCAACGACC	ATGTCATTGCGCTGCCATTC	This study
blaCMY-2	TTCTCCGGGACAACTTGACG	GCATCTCCCAGCCTAATCCC	This study
blaCTX-M	GATACCACTTCACCTCGGGC	CCTTAGGTTGAGGCTGGGTG	This study
blaKPC-2	AATGAGCTGCACAGTGGGAA	ATCGCCGTCTAGTTCTGCTG	This study
erm(B)	TCACCGAACACTAGGGTTGC	CTGTGGTATGGCGGGTAAGT	This study
mecA	GGCATTGTAGCTAGCCATTCC	TGGCAGACAAATTGGGTGGT	This study
tet(A)	CGGCAATCATTCCGAGCATG	ATTCTGCATTCACTCGCCCA	This study
tet(B)	TGGTGGTGGGATCGCTTTAC	AATGGGCCAATAACACCGGT	This study
tet(M)	GTGCCGCCAAATCCTTTCTG	GCATCCGAAAATCTGCTGGG	This study
16S rRNA	GTGYCAGCMGCCGCGGTAA	GGACTACNVGGGTWTCTAAT	63, 64

TABLE 5.

Sources of DNA for qPCR standard curve generation

ARG	Strain or plasmid	Source
aac(6′)-Ie-aph(2″)-Ia	Enterococcus faecalis ATCC 51299	ATCCa
aadA1	pSkunk3-BLA	Addgene, Cambridge, MA
blaCMY-2	Escherichia coli J79	Schmidt laboratory collection, beef cattle isolate
blaCTX-M	E. coli J82	Schmidt laboratory collection, beef cattle isolate
blaKPC-2	Klebsiella pneumoniae (Schroeter) Trevisan ATCC BAA-1898	ATCC
erm(B)	E. coli 52185	Gift from Dayna Harhay, USMARC
mecA	Staphylococcus aureus ATCC 43300	ATCC
tet(A)	E. coli J106	Schmidt laboratory collection, beef cattle isolate
tet(B)	E. coli J108	Schmidt laboratory collection, beef cattle isolate
tet(M)	Enterococcus sp. J78	Schmidt laboratory collection, beef cattle isolate

^aATCC, American Type Culture Collection.

16S rRNA sequencing and phylogenetic analysis.

Detailed descriptions of the 16S rRNA sequencing and phylogenetic analysis methods are provided in Text S2. Briefly, equal amounts of DNA (200 ng) from each sample were pooled by lot. Amplicon libraries were prepared by PCR amplification of the V1 to V3 region of the 16S rRNA gene as previously described (46). Amplification was carried out at 94°C for 90 s, followed by 25 cycles of 94°C for 30 s, 58°C for 30 s, and 68°C for 45 s, and a final extension at 68°C for 5 min. The amplicon libraries were sequenced on a MiSeq desktop sequencer (Illumina, San Diego, CA) using the 2 × 300-bp v3 600-cycle kit (Illumina). The 16S rRNA sequences were analyzed using tools in the Quantitative Insights Into Microbial Ecology (QIIME) package 1.9.1 (47) on MacQIIME (http://www.wernerlab.org/software/macqiime). Nonmetric dimensional analysis was conducted in the “Vegan” package in R version 3.2.4 (48). Operational taxonomic units (OTUs) conserved across all the samples were determined using the core microbiome analysis pipeline. Linear discriminant analysis effect size (LefSe) with a logarithmic LDA score cutoff value of 1.3 on the Galaxy server hosted at the Huttenhower lab, Harvard University (http://huttenhower.sph.harvard.edu/galaxy), was used to determine differentially abundant OTUs between the CONV and RWA systems (49, 50).

qPCR array detection of 84 ARGs.

DNA (10 μl) from individual samples in each lot was pooled. Microbial DNA qPCR arrays (BAID-1901Z; Qiagen, Valencia, CA) were used to detect ARGs according to the manufacturer's instructions. For each array plate, the 25-μl reaction volume consisted of 2× microbial qPCR master mix (Qiagen, Valencia, CA) and 5 ng of metagenomic DNA. Plates were incubated in an Applied Biosystems 7500 Fast real-time PCR system (Life Technologies, Grand Island, NY). The amplification conditions were 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 2 min. A maximum cycle limit of 34 cycles was used to determine if an AMR gene was present in a sample. Samples that did not meet the threshold for detection for individual genes by 34 cycles were considered to not harbor those genes. The frequencies of detected ARGs between CONV and RWA systems were compared by Fisher's exact test. P values of <0.05 were considered significant.

Metagenomic sequencing.

Equal amounts of DNA (200 ng) from each sample were pooled by lot. Libraries were prepared using the Illumina TruSeq DNA sample preparation kit (Illumina, Inc., San Diego, CA) according to the manufacturer's instructions. Next-generation sequencing was performed on the Illumina NextSeq 500 platform (Illumina, Inc.). Raw sequencing reads were trimmed using Trimmomatic version 0.32 (51) to remove Illumina adapters, low-quality sequences (Phred score <15 in a sliding window of 4 nucleotides), and short reads (<36 nucleotides long). Bovine DNA was removed by mapping trimmed sequence reads to the reference genomes of Bos indicus (52) and Bos taurus (53) using Burrows-Wheeler Aligner (BWA) version 0.6.2 (54).

Identification of ARGs in metagenomic sequences.

To identify ARG sequences, reads were aligned to MEGARes (18), a custom nonredundant database, using BWA. This resistance database contains 3,762 unique resistance genes as a result of combining the ResFinder (55), ARG-ANNOT (56), and CARD (57) databases and the National Center for Biotechnology Information (NCBI) Lahey Clinic β-lactamase archive. The MEGARes database facilitates abundance comparisons, since each ARG sequence is annotated into a three-level hierarchy. From narrowest to broadest, these hierarchical levels are “group,” “resistance mechanism,” and “AM class.” For example, for the group tet(Q), the resistance mechanism is “tetracycline ribosomal protection protein,” and the AM class is “tetracyclines.” Since “group” is roughly equivalent to ARG in the traditional nomenclature, “group” was referred to as ARG in this report.

SAMtools (58) and a custom-developed parsing program were used to generate statistics from the BWA mapping output. Up to 4% (∼5 bases) of the individual read length was allowed to have mismatches with the mapping location in the reference gene to qualify as a “hit” between the read and a gene in the database. To decrease the number of false-positive results, only sequences with a read coverage >80% over the entire reference gene were considered for downstream descriptive and statistical analyses (12).

Genes known to cause resistance as a result of single nucleotide polymorphisms (SNPs) in core structural genes were evaluated by visualizing the BWA alignments with Tablet (59) and visually confirming that the reads aligned with 100% peptide homology for the SNP conferring the resistance. The genes identified in our samples and included in this postprocessing verification step were gyrA, gyrB, parC, parE, and rpoB. ARG sequence abundances were normalized using a cumulative sum scaling-factor approach accounting for various sequence depths across samples (60). The ARG abundances ranged from 1.23 × 10⁰ to 5.92 × 10⁴. An abundance value of 1.00 × 10⁰ was assigned to samples when an ARG, resistance mechanism, or AM class was absent. ARG, resistance mechanism, and AM class geometric means were calculated for each production system and log₂ transformed. Only ARGs, resistance mechanisms, and AM classes with production system log₂ fold changes ≥1.0 or ≤−1.0 were further analyzed. For these ARGs, comparisons of abundances were evaluated by two-way analysis of variance (ANOVA) using the standard least squares program of JMP version 12.0.1 (SAS Institute Inc.) with production system, season, and the production system by season interaction as model effects. P values of <0.05 were considered significant.

Phylogenetic analysis of metagenomic sequence.

Taxonomic labels were assigned to trimmed and nonbovine sequences using Kraken version 0.10.6-beta, which utilizes the National Center for Biotechnology Information (NCBI) reference nucleotide database (RefSeq) to classify bacteria at different taxonomic levels (61). To increase the sensitivity of the taxonomic classification, the optional script kraken-filter was run with a threshold of 0.20, which moves assignments up to higher levels of the taxonomic trees to avoid the overclassification of reads at lower taxonomic levels (62). Data were analyzed using the multivariate ordination technique nonmetric multidimensional scaling (NMDS) using Euclidean distances and the analysis of similarity (ANOSIM) included in the Vegan package version 2.2-2 to determine whether the overall microbiomes differed significantly between production systems. NMDS plots with a goodness of fit (stress) of less than 0.2 were considered not random with little risk of misinterpretation. In addition to the P value, ANOSIM calculates an R value ranging from 0 to 1 that indicates the extent to which the microbiomes differ between groups. For the purpose of this study, R values ranging from 0.00 to 0.25, 0.26 to 0.50, 0.51 to 0.75, and 0.76 to 1.00 were interpreted as weak, mild, strong, and very strong group separation, respectively. A zero-inflated Gaussian distribution mixture model native to the metagenomeSeq R package (60) was used to analyze differential abundances in features of the microbiomes. Cumulative sum scaling-normalized counts were aggregated to the phylum, class, order, family, and genus levels. Statistical inference for each feature in microbial count tables was performed after log₂ transformation followed by a multiple-comparison Benjamini-Hochberg adjustment using a critical α of 0.05. Richness and Shannon's diversity indices were calculated using the Vegan package (48), and a comparison between groups was performed using generalized linear models.

Accession number(s).

The raw sequencing data have been deposited at the National Center for Biotechnology Information's BioProject database under accession number SRP102405.