The effect of tylosin on antimicrobial resistance in beef cattle enteric bacteria: a systematic review and meta-analysis

Abstract

Background:

Tylosin is a commonly used in-feed antimicrobial and is approved in several countries to reduce the incidence of liver abscesses in beef cattle. Macrolides are critically important antimicrobials in human health and used to treat some foodborne bacterial diseases, such as Campylobacter jejuni and Salmonella. Feeding tylosin could select for resistant enteric bacteria in cattle, which could contaminate beef products at slaughter and potentially cause foodborne illness. We conducted a systematic review and meta-analysis to evaluate the impact of feeding tylosin to cattle on phenotypic and genotypic resistance in several potential zoonotic enteric bacteria: Enterococcus species, Escherichia coli, Salmonella enterica subspecies enterica, and Campylobacter species. This review was registered with PROSPERO (#CRD42018085949).

Results:

Eleven databases were searched for primary research studies that fed tylosin at approved doses to feedlot cattle and tested bacteria of interest for phenotypic or genotypic resistance. We screened 1,626 citations and identified 13 studies that met the inclusion criteria. Enterococcus species were tested in seven studies, Escherichia coli was isolated in five studies, three studies reported on Salmonella, and two studies reported on Campylobacter species. Most studies relied on phenotypic antimicrobial susceptibility testing and seven also reported resistance gene testing. A random-effects meta-analyses of erythromycin-resistant enterococci from four studies had significant residual heterogeneity. Only two studies were available for a meta-analysis of tylosin-resistant enterococci. A semi-quantitative analysis demonstrated an increase in macrolide-resistant enterococci after long durations of tylosin administration (>100 days). Semi-quantitative analyses of other bacteria-antimicrobial combinations revealed mixed results, but many comparisons found no effect of tylosin administration. However, about half of these no-effect comparisons did not record the cumulative days of tylosin administration or the time since the last dose.

Conclusions:

When fed at approved dosages for typical durations, tylosin increases the proportion of macrolide-resistant enterococci in the cattle gastrointestinal tract, which could pose a zoonotic risk to human beef consumers. Feeding tylosin for short durations may mitigate the impact on macrolide-resistant enterococci and further studies are encouraged to determine the effect of minimizing or eliminating tylosin use in beef cattle. There may also be an impact on other bacteria and other antimicrobial resistances but additional details or data are needed to strengthen these comparisons. We encourage authors of antimicrobial-resistance studies to follow reporting guidelines and publish details of all comparisons to strengthen future meta-analyses.

Introduction

Effective antimicrobials are essential for treating disease in human and animal populations but are also used to prevent and control disease and have been used to increase growth in food animals. Each use of an antimicrobial is an opportunity to select for antimicrobial-resistant bacteria, which annually cause two million infections and 23,000 deaths in the United States (Centers for Disease Control and Prevention, 2013). Antimicrobial resistance, in bacteria, viruses and parasites, causes about 700,000 deaths worldwide and is predicted to cause 10 million deaths annually by 2050 (O’Neill, 2016). Resistant bacteria circulate among humans, animals and the environment (Berendonk et al., 2015), so efforts to reduce antimicrobial use in all sectors is required to combat antimicrobial resistance. As animal agriculture expands in developing countries and the global human population increases, antimicrobial use in food animals is expected to increase by 67% from 63,000 tons in 2010 to 105,000 tons in 2030 (Van Boeckel et al., 2015). Efforts to reduce antimicrobial use in food animals are expanding, with at least 53% of OIE member countries prohibiting the use of antimicrobials for growth promotion (Moulin et al., 2016) and the United States recently eliminating growth promotion label claims for medically-important antimicrobials (US Food and Drug Administration, 2012, 2013b).

In 2011, the macrolide tylosin was one of the most commonly used antimicrobials in U.S. feedlot cattle, with 71.2% of all U.S. feedlot cattle receiving tylosin, typically for the prevention of liver abscesses (USDA, 2013). In other countries, tylosin is one of the most commonly used growth-promoting antimicrobials (Moulin et al., 2016). Tylosin was first developed, and previously approved in the U.S, for increasing weight gain in swine and chickens (Truow Nutrition; Simpson, 1969). Feeding trials in the 1960’s demonstrated that tylosin could also increase weight gain in cattle (Simpson, 1969), likely by reducing the severity of liver abscesses (Brown et al., 1975). The Food and Drug Administration “judicious use” policy implemented in 2018 resulted in the withdrawal of the growth promotion label claims in the United States but tylosin is approved in cattle for continuous use at 60 to 90 mg/head/day to reduce the incidence of liver abscesses (Elanco US Inc). Liver abscesses are a sequelae to ruminal acidosis, which occurs when feeding high amounts of fermentable carbohydrates (Owens et al., 1998; Amachawadi and Nagaraja, 2016), and are the leading cause of liver condemnation (Nagaraja and Chengappa, 1998). Dietary management and tylosin are the predominant methods for controlling ruminal acidosis and liver abscesses (Amachawadi and Nagaraja, 2016). Diets higher in roughage with fewer processed grains, reduced starch content, supplemented with bases or buffers (e.g. bicarbonate), and longer diet adaption periods prevent ruminal acidosis and subsequent liver abscesses (Owens et al., 1998). However, some dietary modifications may result in reduced weight gain and economic losses (Owens et al., 1998). Tylosin is hypothesized to prevent liver abscesses by suppressing the growth of Fusobacterium necrophorum in the rumen (Nagaraja and Chengappa, 1998; Nagaraja et al., 1999; Amachawadi and Nagaraja, 2016) and may reduce the risk of ruminal acidosis by suppressing lactate-producing microbes (Nagaraja et al., 1997). Feeding tylosin reduces the risk of liver abscesses from 30% to 8% (Wileman et al., 2009). Alternative methods to control liver abscesses, including vaccines against F. necrophorum, probiotics and essential oils, are largely ineffective (Fox et al., 2009; Meyer et al., 2009; Amachawadi and Nagaraja, 2016; Huebner et al., 2019). Relatively few studies have been done on oral tylosin pharmacokinetics compared to parenteral tylosin pharmacokinetics, however oral tylosin reportedly has poor bioavailability in cattle (EMA, 1997; Lewicki, 2006; Jacek Lewicki et al., 2009) and can be recovered from manure of treated cattle (Amarakoon et al., 2016). Therefore, the gastrointestinal microbiome, including potential food-borne pathogens such as Salmonella, Campylobacter, and Escherichia coli, is exposed to tylosin after oral administration, which can result in the selection of macrolide-resistant bacteria. The risk of ruminal acidosis and liver abscesses in the absence of tylosin treatment, plus subsequent economic and welfare losses, must also be considered in the context of the risk for macrolide-resistant bacteria during and after tylosin administration.

Although tylosin is used only in veterinary medicine, macrolide antimicrobials as a class are considered the highest priority of critically important antimicrobials by the World Health Organization because of their use for treating Campylobacter infections in humans (World Health Organization, 2016). In addition, the World Health Organization strongly recommends that medically important antimicrobials should not be used for preventing infectious diseases or growth promotion in food animals (World Health Organization, 2017). While resistance to all clinical macrolides, plus lincosamides and streptograminB, is commonly conferred by ribosomal methylation (erm genes) or enzymatic inactivation (ere genes, mph genes) (Weisblum, 1995; Cattoir and Leclercq, 2017), the use of tylosin in animals can select for bacteria that are resistant to all macrolides. In addition, macrolide resistance genes can be linked to other resistance genes on mobile genetic elements or on chromosomes, resulting in co-selection of multiple resistances from the use of one antimicrobial class (Hasman and Aarestrup, 2002). The prohibition of tylosin as a growth promoter in swine in Switzerland was associated with decreased enterococci resistance to macrolides, lincosamides and tetracycline (Boerlin et al., 2001). Similarly, in Denmark, swine-associated enterococci retained glycopeptide-resistance until tylosin use was banned as a growth promotor because of a plasmid-mediated genetic linkage between macrolide (ermB gene) and glycopeptide (vanA gene) resistance (Aarestrup et al., 2001).

Resistant bacteria in food animals can impact humans via direct infection from animal contact, indirect infection from contaminated animal products or via the environment, or transfer of resistance genes from animal-associated bacteria to human pathogens (Chang et al., 2015). It is easier to observe infections caused by direct or indirect animal contact than to estimate the frequency of resistance gene transfer (US Food and Drug Administration, 2013a) within animal or human hosts. Meat products are responsible for about 1.5 million foodborne illnesses in the U.S. annually (Painter et al., 2013), with Campylobacter and Salmonella species (two commonly animal-associated bacteria) responsible for a majority of foodborne illnesses (Scallan et al., 2011). The potential for resistance gene transfer can be reflected in the prevalence of resistance among indicator bacteria, E. coli for gram negative bacteria and Enterococcus species for gram positive bacteria (Chang et al., 2015; Karp et al., 2017). We expect to see a larger effect of tylosin on resistance in Enterococcus compared to E. coli, Campylobacter, and Salmonella because of intrinsic-resistance to older macrolides in gram-negative bacteria (Leclercq and Courvalin, 1991).

We conducted a systematic review and meta-analysis to evaluate antimicrobial resistance of beef cattle enteric bacteria associated with tylosin use. We used a PICOS framework (Richardson et al., 1995) to define our research question and eligibility criteria. We restricted the population of interest to feedlot cattle in countries that approve tylosin for oral administration. Tylosin alone and tylosin combination products (e.g. monensin and tylosin) were permitted interventions; a no-antimicrobial comparison group was preferred but not required. The outcomes examined included phenotypic (e.g. minimum inhibitory concentrations) and genetic (e.g. PCR) resistance in four bacteria: Enterococcus species, Escherichia coli, Campylobacter species, and Salmonella species. All study designs were permitted. We hypothesized that tylosin administration increases the prevalence or likelihood of antimicrobial resistance at the bacteria level and animal level compared to cattle that were not treated or compared to the prevalence prior to treatment.

Materials and methods

Database searches

A protocol for this systematic review was registered January 26, 2018 in PROSPERO (University of York), identifier CRD42018085949. This systematic review was conducted following the PRISMA (Preferred Reporting for Items for Systematic Reviews and Meta-analyses) checklist for standards for systematic reviews (Moher et al., 2009). The review team was composed of six people with combined expertise in antimicrobial resistance in agricultural animals and library and information resources.

Eleven databases were searched on February 7, 2018. No date limits were applied. Databases searched include Medline (PubMed, 1946-present), CAB Abstracts (Clarivate Analytics, 1910-present), Web of Science Core Collection (Clarivate Analytics, 1900-present), VetMed Resource (1972 – present) (Centre for Agriculture and Biosciences International), Scopus (1970-present), Agricola (Ebsco, 1970-present), Embase (Ovid, 1980-present), ProQuest Dissertations and Theses Global (1743-present), DART-Europe E-Theses Portal (unknown date coverage) (DART-Europe), National Library of Australia’s Trove Service (unknown date coverage) (National Library of Australia), and Theses Canada Networked Digital Library of Theses and Dissertations (unknown date coverage) (Library and Archives Canada). The search was updated on August 9, 2018. Additional articles were identified via searching cited references of included studies using Web of Science and Google Scholar.

The search strategy included terms on the topics of cattle, tylosin, drug resistance and bacteria, consistent with the eligibility criteria of: tylosin use (intervention) in feedlot cattle (population) and antimicrobial resistance in potential zoonotic bacteria (outcome). Only studies published in English were included. The search for Medline (PubMed) and Web of Science Core Collection can be seen in Appendix A.

Study screening

Studies were imported then deduplicated in Covidence (Veritas Health Innovation), which was also used for screening. Two reviewers (CC, GL) independently screened studies at the title and abstract level. Four questions were used to determine whether the study met the eligibility criteria:

1. Does the manuscript describe a primary research study (observational or experimental as opposed to a review)?

2. Does the manuscript describe the use of oral tylosin or tylosin combination products in feedlot cattle?

3. Does the manuscript indicate the research took place in a country where the use of oral tylosin products in feedlot cattle is permitted (US, Canada, Mexico, Brazil, Australia)?

4. Does the manuscript include the outcome of phenotypic antimicrobial resistance (minimum inhibitory concentration or selective plating) or genetic resistance (any specific antimicrobial resistance genes) in the bacteria of interest (Enterococcus, Campylobacter, Escherichia coli, Salmonella)?

Studies that did not meet all eligibility criteria were excluded. Studies that passed title/abstract screening were reviewed at the full text level using the same inclusion criteria. If any of the four eligibility questions could not be definitively answered from a title or abstract screening, the study was passed to full text review as long none of the other criteria were violated. Reproducibility using the screening tool was assessed with Cohen’s Kappa > 0.4 (Cohen, 1960). Discrepancies during both the title/abstract and full text screening processes were resolved by a third, independent reviewer (SM). Reasons for exclusion were recorded at the full text screening level.

References of included studies were screened by reviewers after completing data extraction. Titles and abstracts of cited references of included studies were screened by one reviewer (CC). The full texts of references or citations that passed the title/abstract screening were assessed by two reviewers (CC, GL).

Data extraction and risk of bias assessment

All three reviewers discussed the data extraction form and procedures in Covidence to clarify any areas of confusion and to modify data extraction to improve reliability. Outcome data tables were customized for each study since reported outcomes varied. Detailed instructions were created to guide reviewers in filling out the data extraction forms (Appendix B). Each manuscript was randomly assigned two reviewers (CC, GL, or SM) via the Microsoft Excel (version 1811) random number generator, and each reviewer independently extracted data from text, tables, graphs and figures. The GetData Graph Digitizer (S. Fedorov, 2013, version 2.26.0.20) was used to extract data from graphs and figures. Disagreements were resolved by consensus. Authors for conference proceedings without corresponding published manuscripts were contacted for data collection when author contact information was available.

The same procedure was followed for study quality assessment. All three reviewers discussed the bias assessment tool (Appendix B) to clarify any uncertainties. Reviewers assessed confounding bias, selection bias, information bias and external validity and assigned each study low risk, high risk or unclear risk of having bias or violating external validity. A consensus was reached via discussion if the two reviewers gave a study different quality assessments.

Data synthesis

Outcome data were separated by outcome reporting method (proportion of resistant isolates, proportion of cattle with resistant isolates, proportion of pens with resistant isolates), antimicrobial, and bacteria. Outcomes reported for a non-random subset of isolates within a study (e.g. ‘select isolates’ in Table 1) were excluded from data synthesis. Within each study, the outcome was compared between tylosin and control or non-tylosin groups at each sampling point; we used the published 95% confidence intervals or calculated 95% confidence intervals from graphical error bars or the sample proportion standard error $\left(S{E}_p=\sqrt{\frac{p(1-p)}{n}}\right)$. Each sampling point was then classified as showing an increase, decrease or no change in resistance for a given bacteria and antimicrobial. The number of sampling points that showed an increase, decrease, or no change in resistance was summarized.

Table 1:

Characteristics and quality of studies included in the systematic review.

Reference First Author, Year	Bacteria investigated	Laboratory Procedure P: Phenotypic G: Genotypic	Study Design	Intervention Groups	Number of pens, animals and samples	Outcome Reporting*	Bias Assessment
CANADA
Alexander, 2008	Escherichia coli	P: Selective plating, agar dilution	Cluster RCT	Tylosin (61 days + 42 days) and control; pen-level	5 pens of 10 steers in each group. 2 rectal samples per steer per timepoint.	Proportion of isolates resistant, pooled across time. Proportion of steers with resistant E. coli (AMP 50 ug/mL, GEN 2 ug/mL, TET 4 ug/mL).	Confounding: Low Information: Low Selection: Low Violating External Validity: Low
Benedict, 2015	Escherichia coli	P: Broth microdilution, disk diffusion	Longitudinal cross-sectional	Tylosin and not-tylosin; individual and pen level	807 cattle from 205 pens at arrival; 923 cattle from 215 pens at later timepoint. 1 rectal fecal sample per animal.	Logistic models of antimicrobial resistance on antimicrobial use.	Confounding: Unclear Information: Low Selection: Low Violating External Validity: Low
Beukers, 2015	Enterococcus	P: Selective plating, disk diffusion of select isolates G: PCR of select isolates	Cluster RCT	Tylosin (197 days) and control; pen-level	5 pens of 10 steers in each group. 1 rectal sample per steer per timepoint.	Proportion and 95% confidence interval of isolates resistant (ERY 8 ug/mL, TYL 32 ug/mL). Proportion of steers with resistant (ERY, TYL) isolates.†	Confounding: Low Information: Low Selection: Low Violating External Validity: Low
Inglis, 2005	Campylobacter jejuni, Campylobacter hyointestinalis	P: Agar dilution	Cluster RCT	Tylosin (56 days + 42 days) and control; pen-level	5 pens of 10 steers in each group. 1 fecal sample per steer per timepoint.	Proportion of steers with resistant (TET, 8 ug/mL) isolates at each timepoint and cumulative carriage rates.†	Confounding: Low Information: Low Selection: Low Violating External Validity: Low
Rao, 2010	Escherichia coli, Campylobacter jejuni, Salmonella	P: Agar dilution (E. coli, C. jejuni); broth microdilution (Salmonella)	Longitudinal cross-sectional	Tylosin and not-tylosin; pen-level	Total of 84 pens from 21 feedlots. 25 fresh pen-floor fecal samples per pen.	Coefficients and resistance prevalence estimates from regression models.	Confounding: Low Information: Low Selection: Low Violating External Validity: Low
Sharma, 2009	Escherichia coli	P: Selective plating G: Real-time PCR on fecal samples	Cluster RCT	Tylosin (197 days) and control; pen-level	1 pen of 50 steers in each group. 1 sample from each pen at start of composting.	Number (log CFU/g) of E. coli on selective plates (AMP 32 ug/mL, TET 16 ug/mL).	Confounding: Unclear Information: Low Selection: Unclear Violating External Validity: Unclear
Zaheer, 2013	Enterococcus	P: Selective plating G: PCR of select isolates	Experimental trial‡	Tylosin (28 days) and control; animal-level	10 steers in individual pens in each group. 1 rectal fecal sample per steer per timepoint.	Proportion of isolates resistant (ERY 8 ug/mL).	Confounding: Low Information: Low Selection: Unclear Violating External Validity: High
UNITED STATES
Amachawadi, 2014	Enterococcus	P: Microbroth dilution on select isolates G: PCR	Prospective cohort of tylosin (RCT for copper, supplement)	Tylosin (117 days or 131 days), 2x2 factorial with copper and supplement; pen-level	6 pens of 10-11 heifers in each group. 8 fresh pen-floor fecal samples per pen per timepoint.	Gene prevalence (ermB, tet(M)) pooled across time.†	Confounding: Low Information: Low Selection: Low Violating External Validity: High
Amachawadi, 2015	Enterococcus	P: Microbroth dilution of select isolates G: PCR	RCT	Tylosin (28 days) and control, 2x2 factorial with copper; animal-level	20 steers in individual pens in each group. 1 fresh pen-floor fecal sample per pen per timepoint.	Gene prevalence (ermB, tet(M)) pooled across time.†	Confounding: Low Information: Low Selection: Low Violating External Validity: Low
Jacob, 2008	Escherichia coli, Salmonella, Enterococcus	P: Microbroth dilution G: Real-time PCR on fecal samples	Cluster RCT	2x3 factorial of distillers grain and tylosin plus monensin (150 days) vs monensin vs control; pen-level	9 pens of 6-7 heifers in each group. 1 fecal sample from each heifer pooled by pen.	Proportion and 95% confidence interval of enterococci resistant (ERY 8 ug/mL, TYL 32 ug/mL). Hypothesis tests of proportion of resistant E. coli by intervention group. Salmonella not reported by intervention group.†	Confounding: Low Information: Low Selection: Low Violating External Validity: High
Molitoris, 1986	Enterococcus (Streptococci)	P: Not reported	Not reported	Tylosin and not-tylosin; unclear intervention level	48 feedlots total.	Proportion of isolates resistant (ERY, concentration not reported).	Confounding: Unclear Information: Unclear Selection: Unclear Violating External Validity: Unclear
Mollenkopf, 2017	Salmonella	P: Broth dilution G: PCR of select isolates	Cross-sectional	Tylosin and not-tylosin; pen-level	62 tylosin pens, 140 not-tylosin pens from a total of 68 feedlots. 25 fresh pen-floor fecal samples per pen.	Logistic model of blaCMY-2 positive Salmonella on tylosin treatment.†	Confounding: Low Information: Unclear Selection: Unclear Violating External Validity: Unclear
Muller, 2018	Enterococcus	P: Selective plating	RCT	Tylosin (119 days) and control; animal-level	8 pens of 13 steers in each group. 1 fecal pat from 8 random steers per pen per timepoint.	Proportion and 95% confidence interval of isolates resistant (ERY 8 ug/mL, TET 16 ug/mL).	Confounding: Low Information: Low Selection: Unclear Violating External Validity: Low

^*Concentration following antimicrobial indicates the concentration used (selective plates broth microdilution) or resistance breakpoint (agar dilution)

^†MIC or disk diffusion results available but pooled across intervention groups

^‡Would qualify as a RCT except randomization was not described or stated in methods

ERY: erythromycin. TYL: tylosin. AMP: ampicillin. TET: tetracycline. GEN: gentamicin. RCT: randomized controlled trial.

The proportion of resistant isolates or proportion of cattle with resistant isolates after tylosin administration was synthesized with a meta-analysis; studies without a non-tylosin group were excluded. The following meta-analysis protocol was applied. If the appropriate standard error was not published, it was calculated from available 95% confidence intervals, graphical error bars, or from the sample proportion $\left(S{E}_p=\sqrt{\frac{p(1-p)}{n}}\right)$. For each bacteria-antimicrobial combination with two or more studies, a mixed-effects, inverse-variance weighted, linear model (R metafor) was created to account for potential confounding variables, study design and study quality. Outcome was the difference in percent resistant (proportion x 100%) in the tylosin and non-tylosin groups. For outcomes reported at specific time-points, a linear mixed-effects, inverse-variance model (R metafor: rma.mv) was created with a random slope and intercept for day within each study, with first day of tylosin administration as day 0. Different variance structures were tested for day and an auto-regressive structure was chosen based on the model AIC and BIC. Polynomial effects of day (squared and cubed) were tested and compared to a model with only a linear effect; the best model with significant coefficients for the day effects was selected based on AIC and BIC. Heterogeneity was assessed with τ² (estimated amount of residual heterogeneity) and the Cochran’s Q_E test for residual heterogeneity that is not accounted for by variables included in the model (Viechtbauer, 2010). However, Cochran’s Q test has limited statistical power when the number of studies is small (Hardy and Thompson, 1998). Meta-analyses with significant residual heterogeneity were excluded from our conclusions and further analysis. For bacterial-antimicrobial combinations with three or more studies, a sensitivity analysis and publication bias assessment were performed. The effect of each individual study on the random effects model was assessed by ‘leave-one-out’ sensitivity analysis. Each study was removed from the model and the impact on model coefficients was assessed. Missing studies due to publication bias were identified with funnel plots and the rank correlation test for funnel plot asymmetry. Meta-analysis results were presented only if there was no evidence of residual heterogeneity, no evidence of publication bias, and no sensitivity to single studies.

Results

The initial search on February 9, 2018 identified 493 records from the 11 databases, of which 212 were duplicates (Figure 1). The search was repeated on August 9, 2018 and identified 32 records published since February 9, 2018, 5 of which were duplicates. After screening titles and abstracts, 280 records were excluded; agreement between reviewers (CC, GL) was moderate (kappa = 0.5). The remaining 27 studies were assessed for eligibility with full-text review and 11 were included. There was disagreement between reviewers during full-text review (kappa = 0.36) and seven studies were reviewed by a third reviewer (SM) to resolve the disagreement. Studies that met all inclusion criteria were cited by 565 manuscripts, of which 171 had already been screened (duplicates) and two met the inclusion criteria (passed full-text review). Screening the reference titles from the 13 included studies did not yield any new relevant records (536 references screened).

Figure 1: PRISMA flow diagram of the search and selection process.

The studies that met the inclusion criteria are detailed in Table 1. Briefly, seven studies isolated Enterococcus species, five isolated Escherichia coli, three isolated Salmonella species, and two isolated Campylobacter species. One study also examined Mannheimia haemolytica resistance but M. haemolytica outcomes were not included in this review (Zaheer et al., 2013). All studies examined phenotypic resistance via: selective plating (five studies), broth dilution (six studies), agar dilution (three studies) or disk diffusion (two studies); the proportion of isolates resistant and/or proportion of cattle with resistant isolates were commonly reported. Although six studies reported minimum inhibitory concentrations (MIC) for isolates, these results were pooled across intervention groups. Selective plating does not require the isolation and characterization of individual bacterial colonies; samples are plated on agar with or without antimicrobial supplementation and the number of colonies on each type of plate are compared to determine the proportion of resistant bacteria in the sample. The other phenotypic methods (broth dilution, agar dilution, and disk diffusion) characterize individual isolates. As expected, there was variation in laboratory methodology across studies. For example, bile-esculin-azide agar (Zaheer et al., 2013; Beukers et al., 2015) or M-Enterococcus agar (Jacob et al., 2008; Amachawadi, 2014; Amachawadi et al., 2015; Müller et al., 2018) was used for isolation and selective plating of enterococci with varying incubation times and temperatures. Five studies also assessed genotypic resistance by PCR and two studies performed real-time PCR on fecal samples prior to bacterial isolation.

Eight of the studies were randomized controlled trials or experiments, three were cross-sectional studies, one (conference abstract) did not provide sufficient information to determine study design (Molitoris et al., 1986), and one study was considered a prospective cohort study because it was a randomized controlled trial for elevated dietary copper and a nutritional supplement but fed tylosin to all animals (Amachawadi, 2014). The cross-sectional studies did not report the prevalence of resistance at an isolate or animal level for the tylosin exposed and unexposed groups; rather, they used regression models to examine the association between tylosin exposure and antimicrobial resistance (Rao et al., 2010; Benedict et al., 2015; Mollenkopf et al., 2017).

Only studies performed in the United States or Canada were identified in this systematic review, although in-feed tylosin is also labeled in Brazil, Australia, and Mexico for the reduction of liver abscesses. We also did not identify any relevant studies performed in other countries with high beef production (e.g. India, China, Argentina, New Zealand). No studies were excluded due to country at the full-text review stage and a search of the excluded titles and abstracts returns no relevant results for these beef-producing countries. However, we were limited to examining only studies published in English, which may have resulted in failing to identify studies from non-English speaking countries. Only 31 studies excluded at the title/abstract review stage had non-English titles and six of these had abstracts in English plus at least five were translations of English titles. Tylosin dosage, duration or indication was not included in the search terms or eligibility criteria. Yet, all the prospective studies and controlled trials fed tylosin at the labeled dosage for prevention of liver abscesses and most studies fed it continuously for the entire feeding period. The two studies that used multiple feedlots (Molitoris et al., 1986; Mollenkopf et al., 2017) did not report the tylosin dosages or durations used on each feedlot.

Study quality was assessed in four domains: confounding bias, information bias, selection bias, and risk of violating external validity. Randomized controlled trials were generally rated high quality, with low risk of bias and low likelihood of violating external validity (Table 1). Occasionally, key details on the animal selection process, source population, and confounders (e.g. prior antimicrobial use) were not reported and we were therefore unable to make a quality determination in one or more bias domains. In this scenario, we have reported the quality as ‘unclear’. Two studies eligible for the meta-analyses with low risk of internal bias were rated as higher risk of violating external validity because they did not replicate a feedlot environment by either housing animals in individual pens (Zaheer et al., 2013) or by using only heifers (Jacob et al., 2008). We included them in the meta-analyses because the studies were otherwise of high quality and were conducted at research feedlots.

All the reported bacterial resistance outcomes were tabulated by cumulative days of tylosin administration as a semi-quantitative assessment to determine whether tylosin increased, decreased, or had no effect, compared to a control group, on proportions of resistance at the isolate, animal or pen level (Figure 2). Sub-group analyses of multi-drug resistant isolates were not included and only three observations of resistance genes (erm(B): erythromycin resistant enterococci; tet(M): tetracycline resistant enterococci, bla_CMY-2: beta-lactam resistant Salmonella) were eligible for inclusion (Amachawadi et al., 2015). There was an increase in the proportion of macrolide-resistant enterococci after about 100 days of tylosin administration and also an increased proportion of steers that carried erythromycin-resistant enterococci after the same duration of tylosin administration. There was no change in tetracycline-resistant enterococci at the isolate level, although there were no estimates of tetracycline resistance after 118 days or more of reported tylosin administration. There was no clear trend in the proportion of resistant E. coli at the animal or isolate level, although the majority of estimates reported no change. Only one study (Mollenkopf et al., 2017) reported resistance outcomes for Salmonella; the other studies (Jacob et al., 2008; Rao et al., 2010) that isolated Salmonella did not report comparisons of resistance between tylosin and non-tylosin groups. There could be duration-dependent effects of tylosin on C. hyointestinalis; the last two timepoints in Inglis et al. (2005) showed an increase in the proportion of cattle with tetracycline-resistant isolates, although both timepoints occurred after tylosin administration had ceased.

Figure 2: Impact of tylosin on resistant enteric bacteria at the isolate, animal and pen level across 12 studies identified in the systematic review.

One study (Amachawadi, 2014) was excluded because it did not report a control group or pre-tylosin observation for comparison to the tylosin group. Categorization as increased resistance, no change in resistance or decreased resistance was based upon 95% confidence intervals for reported resistance in tylosin and control groups when possible; otherwise, point estimates were compared. Comparisons were classified according to cumulative days of tylosin administration and were classified as “NA” if the resistance estimates were pooled across several timepoints or if the days of tylosin administration were not available. For E. coli, all observations with “NA” days of tylosin are pooled isolate-level observations. Each study could contribute more than one comparison, depending on the outcomes and timepoints reported. ERY: erythromycin. TET: tetracycline. TYL: tylosin. AMP: ampicillin. GEN: gentamicin. Other: any other antimicrobial. bla_CMY2: beta-lactamase resistance gene encoding an AmpC cephamycinase.

There were sufficient data (two or more studies reporting the same outcome type) for only two meta-analyses: the effect of tylosin on the proportion of either: (1) erythromycin-resistant (four studies, not presented due to residual heterogeneity (Jacob et al., 2008; Zaheer et al., 2013; Beukers et al., 2015; Müller et al., 2018)) and (2) tylosin-resistant enterococci (two studies, not presented due to inability to test for publication bias and sensitivity analysis (Jacob et al., 2008; Beukers et al., 2015)). The studies included in the meta-analysis used the same resistance breakpoints (Table 1). The studies using selective plating reported the proportion of resistant enterococci as a percent by dividing the number of enterococci growing on antimicrobial supplemented plates by the number of enterococci on plates without antimicrobials (Zaheer et al., 2013; Beukers et al., 2015; Müller et al., 2018). These studies speciated a subset of enterococci isolated from agar plates; the most common species identified was E. hirae (431/519 isolates (Beukers et al., 2015), 76/182 isolates (Müller et al., 2018), 128/130 isolates (Zaheer et al., 2013)). Jacob et al. (2008) used microbroth dilution to test individual isolates (species not identified) for antimicrobial resistance and reported the percent of all isolates that were resistant. One study (Jacob et al., 2008) pooled results from two timepoints (122 days and 136 days of tylosin administration); it was included in the meta-analysis because the timepoints were close together, so the days of tylosin administration were averaged. The ERY model had significant residual heterogeneity but potential confounders, such as animal weight or age at placement, could not be included in the models because they were not reported for all studies. Polynomial effects of Day did not eliminate the residual heterogeneity in the ERY model and were not significant predictors of ERY resistance. Therefore, the ERY model is not presented. The TYL best-fit model, including Day and Day², did not have significant residual heterogeneity. However, tests for residual heterogeneity have low power when few studies are included in the model (Hardy and Thompson, 1998). Furthermore, we are unable to remark on publication bias or perform a leave-one-out sensitivity analysis because only two studies were included in the TYL metaanalysis. Therefore, the TYL meta-analysis results are also not presented.

Discussion

Even though tylosin has been used in beef cattle since the 1960s (Shotwell and Carr, 1976), we identified only 13 published studies that examined the effect of tylosin on antimicrobial resistant foodborne pathogens in the cattle gastrointestinal tract. A semi-quantitative analysis (Figure 2) identified trends in antimicrobial resistant Enterococcus species, E. coli, Salmonella enterica, and Campylobacter species. In the United States, tylosin is typically fed throughout the feeding period (four or more months) (USDA, 2013), which could result in an increased prevalence of resistant fecal enterococci. The majority of observations indicated an increase in erythromycin-resistant enterococci at the isolate level after approximately 100 days of feeding tylosin and an increase in tylosin-resistant enterococci after approximately 50 days of tylosin administration (Figure 2). Because the slaughter withholding period after tylosin administration is 0 days (Elanco US Inc), cattle fed tylosin could have increased levels of resistant enterococci in their gastrointestinal flora at slaughter, where these bacteria can contaminate beef products. Therefore, efforts to eliminate or reduce the duration of tylosin administration could decrease the risk of beef contamination with macrolide resistant enterococci. One of the studies included in the systematic review (Müller et al., 2018) tested an intermittent tylosin regimen (cycling between 1 week of administration and 2 weeks off). The authors found no difference in liver abscesses, average daily gain, and weight gain to feed intake ratio between intermittent and continuous tylosin administration (Müller et al., 2018), suggesting that intermittent tylosin administration does not have a negative impact on beef production. Although this study found no significant difference in the proportion of macrolide-resistant enterococci between treatment groups (Müller et al., 2018), further research on the optimal timing and duration of tylosin administration, elimination of tylosin, and alternative methods for controlling liver abscesses, could identify treatment regimens that reduce macrolide resistance in fecal enterococci.

The semi-quantitative analysis of other potential foodborne bacteria identified varying effects of tylosin administration (Figure 2), which may result from variation in study design and laboratory methodology (e.g. the E. coli tetracycline resistance breakpoint varied, Table 1). At the animal-level and across all levels (i.e. summing the panels of Figure 2), the majority of bacteria-antimicrobial combinations tested showed no change in resistance due to tylosin administration. Since macrolides are generally ineffective against gram negative bacteria (Leclercq and Courvalin, 1991), tylosin may exert a smaller selective pressure on E. coli, Salmonella enterica, and Campylobacter species than on Enterococcus species. In addition, the predominance of no change in resistance due to tylosin administration could result from limited co-resistance between macrolide-resistance genes and other resistance genes. However, many of the comparisons within E. coli, Salmonella, and Campylobacter were not associated with a specific duration of tylosin exposure, which could bias the overall conclusion of no change in resistance if the true duration of exposure was short or if animals had not received tylosin immediately prior to sampling. Two of the cross-sectional studies reported animal-daily-doses rather than duration of tylosin exposure and found no relationship between the animal-daily doses and antimicrobial resistance of fecal bacteria but did not evaluate time since tylosin exposure (Rao et al., 2010; Benedict et al., 2015). Others (Molitoris et al., 1986; Mollenkopf et al., 2017) only reported previous exposure and therefore the duration of exposure or time since exposure was unknown. In addition, some randomized controlled trials reported outcomes pooled across several timepoints (Sharma et al., 2009; Amachawadi et al., 2015). Therefore, the observation of no effect of tylosin on many bacteria-antimicrobial resistances should be interpreted as a lack of evidence for an effect rather than evidence of no effect.

The erythromycin-resistance mixed effects models demonstrated significant heterogeneity after accounting for study and days of tylosin administration, likely due to significant variation in study design and methodology across the included studies. We were unable to do subgroup analyses or meta-regression to search for additional sources of heterogeneity because of insufficient sample size (four studies) and some studies did not report adequate details on potential confounders. For example, animal weight or age at placement could alter the pharmacodynamic effect of tylosin because the enteric concentration of oral antimicrobials, dosed per animal, depends on the volume of the gastrointestinal tract, which increases as animals age (Cazer et al., 2014). We recommend that authors follow the reporting guidelines for randomized controlled trials (REFLECT (O'Connor et al., 2010), CONSORT (Schulz et al., 2010)) or observational studies (STROBE (von Elm et al., 2007), STROBE-AMS (Tacconelli et al., 2016), STROBE-VET (Sargeant et al., 2016)) so that consistent details on study populations, microbiological methods, and outcomes are available for future meta-analyses.

We found that some studies in the systematic review reported non-significant results by omission (i.e. non-significant associations are not presented in the results section), suggesting publication bias. In particular, studies that build many statistical models between antimicrobial use and resistance may not report non-significant models or coefficients, requiring readers to assume that there is no association between unreported antimicrobial use-antimicrobial resistance combinations. Therefore, there may be additional studies that found no significant association between tylosin use in cattle and enteric bacteria antimicrobial resistance that were not identified in our search because tylosin was not mentioned in the title or abstract if it was a non-significant predictor. We encourage authors to publish all available statistical models in the manuscript or supplemental materials and clearly state which comparisons were tested in the results section and, ideally, the abstract. Several included studies also published some data, such as MIC results, pooled across intervention groups. We hope that in the future detailed data tables will be available as supplemental material so that meta-analyses can use all available datasets. We also identified several studies that investigated the fecal resistome via PCR or metagenomic sequencing but did not isolate specific bacteria (Chen et al., 2008; Alexander et al., 2011; Xu et al., 2016). This could be a topic for a future systematic review and meta-analysis to further consolidate knowledge on the effect of tylosin on antimicrobial resistant bacteria in beef cattle.

Uncertainty surrounding the expected and likely range of effects of tylosin use in beef cattle on resistant bacteria is a challenge to policy makers. In the absence of clear, repeatable outcomes from multiple independent studies, policy makers may choose to invoke the Precautionary Principle, restricting tylosin use out of an abundance of caution, in jurisdictions where this has been directly codified in a regulatory framework. However, this could result in negative consequences for animal health and welfare, farm businesses, and farmers. Only one study included in this systematic review reported animal health and economic characteristics associated with tylosin administration; it identified more liver abscesses in cattle that were not fed tylosin compared to cattle treated with tylosin (Müller et al., 2018). However, a recent study found no difference in liver abscesses between tylosin and no-tylosin groups (Schmidt et al., 2020). We encourage future studies to include feedlot performance, animal health, and carcass quality outcomes, as well as antimicrobial resistance outcomes, so that the effect of tylosin or other antimicrobials on resistant bacteria can be considered in the context of judicious antimicrobial use to prevent and treat disease. Furthermore, policy makers must also consider the shedding of resistant bacteria and antimicrobial residues into the environment, which can be an indirect transmission route to humans. One study in this systematic review measured antimicrobial resistant bacteria in aged pen manure (Sharma et al., 2009), which could contaminate the environment through run-off or application to fields. Future systematic reviews could assess the risk of environmental contamination with resistant bacteria or tylosin residues so that this information can then be considered in policy decisions.

Conclusions

Tylosin is commonly used to prevent liver abscesses in beef cattle in the United States and Canada and is also licensed for this use in Brazil, Mexico and Australia. However, we only identified 13 studies (7 randomized controlled trials) that investigated the effects of tylosin on antimicrobial resistance in potential zoonotic enteric bacteria. The results of two meta-analyses on macrolide-resistant enterococci could not be interpreted because of significant residual heterogeneity (ERY model) and an insufficient sample size for assessing publication bias and model sensitivity (TYL model). Based on a semi-quantitative analysis of all 13 studies, there is evidence that longer durations of tylosin exposure (at least 100 days) are associated with an increased percent of macrolide-resistant enterococci. This suggests that the standard practice of administering tylosin throughout a four- to six-month feeding period results in increased proportions of enteric macrolide-resistant enterococci when the cattle are slaughtered—an opportunity for contamination of beef products. However, additional studies, with thorough reporting, are required to confirm this result. There was limited evidence to support an effect of tylosin on other bacteria-antimicrobial resistance combinations at the pen-level, animal-level or isolate-level. Further research to optimize, reduce, or eliminate tylosin administration in feedlot cattle could reduce selection for resistant enteric bacteria.