Extended-Spectrum β-lactamase-Producing Enterobacteriaceae Shedding in Farm Horses Versus Hospitalized Horses: Prevalence and Risk Factors

Simple summary This prospective study investigated the prevalence, molecular characteristics and risk factors of extended-spectrum β-lactamase (ESBL)-producing Enterobacteriaceae (ESBL-E) shedding in three equine cohorts: (i) farm horses (13 farms, n = 192); (ii) on admission to a hospital (n = 168) and; (iii) horses hospitalized for ≥72 h re-sampled from cohort (ii) (n = 86). Bacteria were isolated from rectal swabs, identified, antibiotic susceptibility patterns were determined, and medical records and owners’ questionnaires were analyzed for risk factor analysis. ESBL shedding rates significantly increased during hospitalization (77.9%, n = 67/86), compared to farms (20.8%, n = 40/192), and horses on admission (19.6%, n = 33/168). High bacterial species diversity was identified, mainly in cohorts (ii) and (iii), with high resistance rates to commonly used antimicrobials. Risk factors for shedding in farms included horses’ breed (Arabian), sex (stallion), and antibiotic treatment. Older age was identified as a protective factor. We demonstrated a reservoir for antibiotic-resistant bacteria in an equine hospital and farms, with a significant ESBL-E acquisition. In light of our findings, in order to control ESBL spread, we recommend conducting active ESBL surveillance programs alongside antibiotic stewardship programs in equine facilities. Abstract We aimed to investigate the prevalence, molecular characteristics and risk factors of extended-spectrum β-lactamase (ESBL)-producing Enterobacteriaceae (ESBL-E) shedding in horses. A prospective study included three cohorts: (i) farm horses (13 farms, n = 192); (ii) on hospital admission (n = 168) and; (iii) horses hospitalized for ≥72 h re-sampled from cohort (ii) (n = 86). Enriched rectal swabs were plated, ESBL-production was confirmed (Clinical and Laboratory Standards Institute (CLSI)) and genes were identified (polymerase chain reaction (PCR)). Identification and antibiotic susceptibility were determined (Vitek-2). Medical records and owners’ questionnaires were analyzed. Shedding rates increased from 19.6% (n = 33/168) on admission to 77.9% (n = 67/86) during hospitalization (p < 0.0001, odds ratio (OR) = 12.12). Shedding rate in farms was 20.8% (n = 40/192), significantly lower compared to hospitalized horses (p < 0.0001). The main ESBL-E species (n = 192 isolates) were E. coli (59.9%, 115/192), Enterobacter sp. (17.7%, 34/192) and Klebsiella pneumoniae (13.0%, 25/192). The main gene group was CTX-M-1 (56.8%). A significant increase in resistance rates to chloramphenicol, enrofloxacin, gentamicin, nitrofurantoin, and trimethoprim-sulpha was identified during hospitalization. Risk factors for shedding in farms included breed (Arabian, OR = 3.9), sex (stallion, OR = 3.4), and antibiotic treatment (OR = 9.8). Older age was identified as a protective factor (OR = 0.88). We demonstrated an ESBL-E reservoir in equine cohorts, with a significant ESBL-E acquisition, which increases the necessity to implement active surveillance and antibiotic stewardship programs.


Introduction
Extended-spectrum beta-lactamase (ESBL)-producing Enterobacteriaceae (ESBL-E) poses a clinical challenge to both human and veterinary clinicians. ESBLs confer resistance to penicillins, cephalosporins, and aztreonam and are often accompanied by fluoroquinolone resistance, which even further narrows antibiotic treatment options [1]. Moreover, many ESBL genes are encoded on large plasmids, which enables lateral transfer between different bacterial species, within the same host and between different hosts [2]. In human medicine, ESBL production is associated with increased morbidity, higher overall and infection-related mortality, increased hospital length of stay, delay of targeted appropriate treatment, and higher costs [3,4]. Risk factors for colonization and infection in humans include severe illness with prolonged hospital stays, the presence of invasive medical devices for a prolonged duration and antibiotic use [2].
Within the last decade, a growing burden of ESBL-E in companion animals is being observed, both as gut colonizing bacteria and as infecting pathogens, causing wounds, respiratory, urogenital, gastro-intestinal, umbilical infections, and bacteremia [5][6][7][8]. Horses were described as carriers, as well as infected by ESBL-E, in equine clinics and in farm settings [9,10]. Prevalence of ESBL-producing E. coli carriage in horses varies between 4-44% in different European countries [11][12][13], with a lower carriage prevalence in equine riding centers in comparison with equine clinics [10]. In equine community settings, being stabled in the same yard with a recently hospitalized horse was identified as a risk factor for ESBL-producing E. coli carriage [14]. Risk factor analysis in the level of the farm revealed that the odds of being an ESBL/AmpC-producing E. coli premises were higher among riding schools than breeding premises, if premises housed a horse that had been medically treated with antibiotics within the last three months, and also in premises where the staff consisted of more than five persons [13]. However, risk factors for shedding of different ESBL-E species within horses were not yet reported.
We aimed to investigate and compare ESBL-E shedding in different equine cohorts, including farm horses, horses on admission to an equine hospital and during hospitalization, as well as to determine risk factors for shedding. We hypothesized that shedding rates increase during hospitalization, that previous antibiotic treatment is a risk factor for shedding and that shedding on admission and during hospitalization is associated with clinical signs, prolonged hospitalization, and severe outcome.

Equine Study Cohorts, Study Design, and Sampling Methods
This prospective study was performed on 13 farms throughout Israel and in the Koret School of Veterinary Medicine-Veterinary Teaching Hospital (KSVM-VTH). The study was approved by the Internal Research Review Committee of the KSVM-VTH (Reference numbers: KSVM-VTH/15_2015, KSVM-VTH/23_2015). Rectal swabs were collected from the horses with owner consent. On admission, sampling was performed prior to any medical treatment in the hospital. When horses survived and were not discharged, a second sample was taken 72 h post-admission. Farm horses were located in different regions of Israel to roughly represent the population.

Demographic and Medical Data
For farm horses (cohort (i)), owners' questionnaires were reviewed for data regarding individual horses, including the originating farm, signalment (age, sex, and breed), duration of the horse's accommodation in the farm, hospitalization and antibiotic treatments within the previous year.
For hospitalized horses (cohort (ii)), medical records were reviewed for the following information: signalment (age, sex, and breed), geographic origin, previous admission to the hospital within the previous year (yes/no), clinical signs, duration of illness before admission, antibiotic therapy before and during hospitalization, surgical procedures, other medications, hospitalization length, short-term outcome, and admission charge.

ESBL-E Isolation and Species Identification
Rectal specimens [14] were collected using bacteriological swabs (Meus s.r.l., Piove di Sacco, Italy) and were inoculated directly into a Luria Bertani infusion enrichment broth (Hy-Labs, Rehovot, Israel) to increase the sensitivity of ESBL-E detection [15]. After incubation at 37 • C (18-24 h), enriched samples were plated onto Chromagar ESBL plates (Hy-Labs, Rehovot, Israel), at 37 • C for 24 h. Colonies that appeared after overnight incubation at 37 • C were recorded, and one colony of each distinct color was re-streaked onto a fresh Chromagar ESBL plate to obtain a pure culture. Pure isolates were stored at −80 • C for further analysis.
Isolates were subjected to Vitek-MS (BioMérieux, Inc., Marcy-l'Etoile, France) for species identification or to Vitek-2 (BioMérieux, Inc., Marcy-l'Etoile, France) for species identification and/or antibiotic susceptibility testing (AST-N270 Vitek 2 card). Chloramphenicol, enrofloxacin, and imipenem were analyzed using disc diffusion assay (Oxoid, Basingstoke, UK). ESBL-production was confirmed by combination disk diffusion using cefotaxime and ceftazidime discs (Oxoid, Basingstoke, UK), as well as cefotaxime and ceftazidime with clavulanic acid (Sensi-Discs BD, Breda, The Netherlands). Results were interpreted according to the Clinical and Laboratory Standards Institute (CLSI) guidelines [16]. Multidrug-resistant (MDR) bacteria were defined as such due to their in vitro resistance to three or more classes of antimicrobial agents [17].

Molecular Characterization of ESBL-E
Isolates were examined for the presence of the blaCTX-M group using a multiplex polymerase chain reaction (PCR) from ESBL-E DNA lysates, as previously described [18]. Isolates that were found to be blaCTX-M PCR negative were further examined for the presence of blaOXA-1, blaOXA2, blaOXA10 [19], blaTEM, and blaSHV groups [20]. ESBL-producing E. coli isolates were subjected to PCR for the detection of mdh and gyrB genes in order to determine the presence of the worldwide pandemic E. coli ST131 lineage [21].

Sample Size and Statistical Analysis
The minimal sample size (number of animals sampled) for farm horses was calculated using WinPepi, based on an estimated shedding rate of 25% for ESBL-E in equine community livery premises [22] and on the fact that Israel is endemic for ESBL-E [23], with a confidence level of 95% and an acceptable difference of 7%, resulting in n = 147.
The minimal sample size for horses on admission to hospital was based on the expected difference between ESBL-E shedding and non-shedding horses and the percentage of admitted horses that were treated with antibiotics before admission since antibiotic treatment was assumed to be a risk factor for shedding [12]. Since there is no previous study revealing percentages of antibiotic-treated horses and ESBL shedding, data for this calculation was based on a human study [24]. Estimating that 25% of horses on admission are ESBL shedders (representing the equine community) and that 72% and 44% of horses were treated with antimicrobials within shedders and non-shedders, respectively, with a 5% significance level and power of 80%, the total required sample size is 145 horses, including 116 non-shedders and 29 shedders.
Risk assessment was performed using Chi-square or Fisher's exact tests for association between individual variables, shedding and ESBL-E acquisition. Descriptive statistics were used to describe shedding rates. Continuous variables were analyzed using t-tests or Mann-Whitney U-tests. p ≤ 0.05 was considered statistically significant. For risk factor analysis of farm horses, a logistic regression model (multivariable analysis) was conducted using all the significant variables in the univariable analysis at a significance level of p < 0.2 using the ENTER method (IBM SPSS Statistics 25). Categorical data were summarized by the number of cases (percentage) and confidence intervals (95%) were calculated by Fisher's (WinPEPI 11.15 Describe A).
In order to compare between shedding rates and antibiotic resistance rates within horses on admission and during hospitalization (cohorts (ii) and (iii), respectively), a mixed effect logistic regression model was conducted (STATA version 13). Resistance was defined as complete resistance (not including "intermediate resistance"). Odds ratio (OR) for a significant change in antibiotic resistance rates is defined as OR for a change in one resistance category (e.g., a change from "susceptible" to "intermediate" or from "intermediate" to "resistant"). A comparison between shedding rates and antibiotic resistance rates between farm horses (cohort (i)) and horses on admission (cohort (ii)) was performed using Chi-square. (Table 1) Overall, 192 horses were sampled, originating from 13 farms across Israel (June 2016-September 2018). The average number of sampled horses per farm was 15 (range: 3-26 horses).
Horses hospitalized for ≥72 h were re-sampled (n = 86).  1 Breed distribution was not significantly different for Arabians, Quarter horses, and Warmbloods in comparison to farm horses, and was significantly different for the pacers horses (significantly higher in farms, p = 0.012) and Friesians (significantly higher on admission, p < 0.001); 2 Median age of horses on admission was significantly lower than the median age of farm horses (p < 0.0001); 3 Sex distribution was not significantly different between farm horses and horses on admission; 4 Data was not available for seven horses.

Antibiotic Therapy, Surgical Procedures, Length of Stay, and Outcome
A proportion of 8.3% (n = 16/192) of farm horses was hospitalized within the previous year, ranging from 0-30% between farms. A proportion of 19.8% (n = 38/192) of horses were treated with antibiotics within the previous year, ranging from 0-61% between farms. On admission, 9.5% (n = 16/168) of horses were reported to be previously hospitalized (within a year period), and 16.1% (n = 27/168) of horses were treated with antibiotics within the previous year. Previous hospitalization and antibiotic treatment prevalence rates were not significantly different in comparison with farm horses.
During hospitalization, 50.6% of horses (n = 85/168) were treated with antibiotics, a proportion which is significantly higher than antibiotic treatment in farms and prior to admission (p < 0.0001). Surgical procedures were performed in 36.9% of horses (n = 62/168). The median length of stay was three days (range: several hours-21 d). Out of all horses admitted to hospitalization, 84.4% survived to discharge (n = 142/168).

Distribution of ESBL-E Species and ESBL Genes
Overall, 192 ESBL-E isolates were analyzed (Table A1). Fourteen bacterial species were identified of which three were identified in all cohorts-E. coli, Klebsiella pneumoniae, and Enterobacter cloacae ( Figure 1). The most prevalent bacterial species in all cohorts was E. coli, consisting of 79.2% of isolates from farms, 66.7% from horses on admission, and 49.0% from hospitalized horses. However, the prevalence of E. coli decreased in horses on admission and in hospitalized horses, as the diversity of other ESBL-E species increased, from four species in farms to five species on admission and twelve species in hospitalized horses. Nosocomial ESBL-E species that were not identified in farms and on admission included Citrobacter freundii (n = 3/105), Salmonella spp (n = 3/105), K. oxytoca, Citrobacter brakii, E. vulneris, Pantoea spp, Proteus mirabilis, and Raoultella ornithinolytica (n = 1/105 each). The pandemic hypervirulent E. coli ST131 [25] was identified in three horses: two horses on admission and one horse during hospitalization. The main ESBL gene was the blaCTX-M-1 group in all cohorts (total 56.8% of all isolates, Table 2). brakii, E. vulneris, Pantoea spp, Proteus mirabilis, and Raoultella ornithinolytica (n = 1/105 each). The pandemic hypervirulent E. coli ST131 [25] was identified in three horses: two horses on admission and one horse during hospitalization. The main ESBL gene was the blaCTX-M-1 group in all cohorts (total 56.8% of all isolates, Table 2).

Antibiotic Susceptibility Profiles
Antibiotic resistance rates varied between cohorts, with a significant increase during hospitalization. All isolates from all cohorts were susceptible to imipenem (Table 3).
Among bacteria that grew on Chromagar ESBL plates, the prevalence of MDR bacteria was 89.6%, 71.8%, and 94.3% in farms, horses on admission, and hospitalized horses, respectively. The prevalence rate was significantly higher in isolates originated from hospitalized horses compared to   1 Horses re-sampled from cohort "horses on admission"; 2 Shedding rate in hospitalized horses is significantly higher than shedding rate on admission and in farms (p < 0.0001, OR=12.12, 95% CI 3.92-37.49); 3 Prevalence of multidrug-resistant (MDR) isolates is significantly higher in isolates originated from hospitalized horses compared to isolates originated from horses on admission (p < 0.001).
Among bacteria that grew on Chromagar ESBL plates, the prevalence of MDR bacteria was 89.6%, 71.8%, and 94.3% in farms, horses on admission, and hospitalized horses, respectively. The prevalence rate was significantly higher in isolates originated from hospitalized horses compared to horses on admission (p = 0.001, Table 2).

Farm Horses
In univariable analysis, horses' breed, sex, hospitalization in the previous year, antibiotic treatment in the previous year, and age were significantly associated with ESBL-E shedding (Table A2). Since the Arabian breed was the most prevalent breed sampled, we clustered all other breeds as one category in the multivariable analysis. In a logistic regression model, the breed (Arabian), sex (stallion versus mare, which was the reference in this category), and antibiotic treatment in the previous year were identified as risk factors for shedding. Age greater than one year was identified as a protective factor (Table 4).

Horses on Admission
Signalment (age, sex, and breed), geographic origin, prior hospitalizations in the last year, clinical signs, length of illness before admission, antibiotic therapy before and during hospitalization, surgical procedures, other medications, hospitalization length, short-term outcome, and admission charge were not associated with ESBL-E shedding on admission (Table A2). Sex, hospitalization length, and admission charge resulted in p < 0.2, therefore, were analyzed via a logistic regression model, which did not yield any significant associations (Table A3).

Horses During Hospitalization
There was no association between ESBL shedding 72 h post-admission and on admission, clinical signs on admission, antibiotic treatment during hospitalization, surgical procedures during hospitalization, length of stay, admission charge and outcome (Table A2).

Discussion
This study investigates ESBL-E shedding in three equine cohorts, including farm horses, representing community equine, as well as horses on admission to the hospital and during hospitalization. Studies regarding antibiotic-resistant pathogens shedding, either in farm horses or in hospitalized horses were reported previously from different European countries [13,22,26,27]. Our study compares different equine cohorts within the same country. Both community and hospital cohorts are of great interest, from a veterinary and a 'one health' perspective, therefore it is highly valuable to compare these cohorts.
We found high ESBL-E shedding rates (Table 2), an increased bacterial species diversity ( Figure 1) as well as in the ESBL-E genes variety (Table 2). An increase in shedding rates may be due to the acquisition of bacteria, plasmids or resistance genes. The main bacterial species in all cohorts was E. coli, with decreased incidence on admission and during hospitalization, due to increased incidence of other nosocomial ESBL-producing bacterial species. The main ESBL gene group was CTX-M-1, as was previously reported in community horses [26]. However, on admission and during hospitalization, CTX-M-1 incidence decreases, alongside an increase in the number of ESBL genes. A study conducted in an equine hospital in the UK demonstrated the emergence of ESBL-producing E. coli during a decade [26], whereas we demonstrated a significant increase in ESBL-E shedding during individual horses' hospitalization. These findings support an urgent necessity in active surveillance and infection control programs in veterinary facilities and hospitals.
In addition, there is a need to set strict antibiotic stewardship programs in veterinary medicine, specifically in companion animals' facilities, with specific guidance and enforcement. According to a recommendation published by the Committee for Medicinal Products for Veterinary Use (CVMP) of the European Union, there is a need to reserve fluoroquinolones, third and fourth generation cephalosporins for treatment when other options are likely to fail, and whenever possible, treatment should be supported by an antimicrobial susceptibility testing [28]. In practice, fluoroquinolones and cephalosporins are in use in equine medicine, sometimes as a first-line choice [29,30]. In our study, ESBL-E shedding as well as resistance rates for chloramphenicol, enrofloxacin, gentamicin, nitrofurantoin, and trimethoprim-sulpha increased significantly during hospitalization, resulting in a significant increase in MDR bacterial species shedding (Tables 2 and 3). In light of our findings, as well as increasing resistance rates in other equine studies, we recommend implementing antibiotic stewardships in equine clinics and hospitals [31,32].
We also aimed to determine risk factors for shedding. We did not find significant associations between shedding on admission and during hospitalization to medical data. During the study period, we sampled all horses on admission, which represented a heterogeneous population, including critically ill horses alongside healthy mares, which were hospitalized together with their sick foals. Therefore, the lack of significant risk factors may be due to high variation in the equine population. Many of the pathologies on admission were attributed to the gastro-intestinal system, which might influence the intestinal microbiome. However, clinical signs on admission and during hospitalization were not associated with shedding. In farm horses, we detected several risk factors for ESBL-E shedding ( Table 4). The Arabian breed was the main breed within farm horses and horses on admission to hospital. These horses in Israel are used mainly for breeding and shows and are held under intensive management, which may explain the risk for ESBL shedding. Interestingly, we detected the 'stallion' sex as a risk factor. In human medicine, it is reported that males are more susceptible to diverse bacterial illnesses than females, including an ESBL-E infection [33], presumably related to hormonal influences [34]. This may explain also our findings in veterinary medicine, however, it requires further investigation. Previous antibiotic treatment was identified as a risk factor as well, in agreement with other human and veterinary studies [2,13]. Age older than one year was identified as a protective factor, which may be due to the maturation of immunity. In a national survey of cattle farms in Israel, the prevalence of ESBL-E was higher in calves versus adult cows, where the use of antimicrobial prophylaxis was more common [35]. In human medicine, elderly age is associated with ESBL-E infections [33]. However, in our study, elderly horses older than 20 years old [36] were not prevalent and consisted of 3% (n = 12/360) of the study population. Therefore, elderly age may not be identified as a risk factor.
Our results should also be addressed from a 'one health' perspective. We detected resistant zoonotic bacteria both in farms and in hospital settings, which underlines the necessity for awareness and improved management. The human-animal interaction has great psychological and physical established benefits, with a great emphasis on equine-assisted therapy [37][38][39]. Therefore, there is pronounced importance in establishing safety policies involving therapists, physicians, and veterinarians, in order to ensure safe human-equine interactions in community settings [40]. This also applies to veterinary hospital staff. In a longitudinal study involving veterinary hospital staff and students, a higher level of ESBL-producing E. coli carriage was observed longitudinally [41], which underlines the necessity to implement gold standards biosecurity programs in veterinary hospitals.

Conclusions
Multi-drug resistant potentially zoonotic bacteria were detected both in farm horses and in hospitalized horses, with a significantly increased shedding during hospitalization. Therefore, we recommend implementing active surveillance programs alongside with infection control and antibiotic stewardship policies, in order to decrease resistance burden and to allow safe human-equine interactions.

Acknowledgments:
We are grateful to the KSVM-VTH equine department staff, farm owners, employees, and veterinarians for their collaboration in conducting this study.

Conflicts of Interest:
The authors declare no conflict of interest.