Association between herd management practices and antimicrobial resistance in Salmonella spp. from cull dairy cattle in Central California

Farms and sampling

The study was approved by the University of California, Davis’s Institutional Animal Care and Use Committee (protocol number 18019). Six dairy farms located in the San Joaquin Valley of California were identified and enrolled in the study as a convenience sample as described previously (Abu Aboud et al., 2016) (Table 1). Cull cows were identified for fecal sampling once during each season between 2015 and 2016, specifically during summer (July 1–September 30, 2015), fall (October 1–December 31, 2015), winter (January 1–March 31, 2016) and spring (April 1–June 30, 2016). The choice of week to sample cull cows during any of the four seasons was also by convenience. From the list of cows selected by the dairy farms for culling, 10 cows were randomly selected for fecal sampling on the day of their removal from the herd using a random number generator (Excel; Microsoft Corp., Redmond, WA, USA). Random numbers were prepared specific to the total possible number of cows being presented for sampling with a specific list for each of the sampling frames consisting of multiples of 10 ranging from 11–20 to 91–100 cows. If a producer had less than 11 cows available for sale on a given sampling day, then all cows were sampled. Individual disposable polyethylene sleeves were used to manually collect fecal samples from the rectum of randomly selected cows, and the samples were transported to the Dairy Epidemiology Lab (Aly Lab) on wet ice for processing within 2–6 h of sampling.

Questionnaire

On the day of sample collection, study personnel completed a survey based on responses of the herd manager to questions related to the previous 4 month’s herd management including herd size, breed distribution, milk production, culling rate, number of times cows were culled per month, percent of cull cows sold for beef (compared to dairy purpose), percent of cull cows condemned and reason for condemnation. Herd managers were also asked questions about the percent of cull cows in the previous 4 months that received injectable medical treatments, percent of culled cows that received injectable treatments 3 weeks prior to culling, personnel allowed to administer drugs, drug residue avoidance (use of specific drugs, observing withdrawal time, testing milk and/or urine prior to culling, or other actions), tracking of drug withdrawal periods, use of a drug inventory system and extralabel drug use (familiarity and frequency). Herd level information collected was based on farm manager recollection. In addition, a backup of the herd’s Dairy Herd Improvement software file was obtained within a week of the visit to extract cull cows’ milk production and health events data. Data from all sources were housed and linked in a relational database using dairy and cow identification, and date of sampling (Access; Microsoft Corp., Redmond, WA, USA).

Bacteriological culture

Fecal samples were submitted to the California Animal Health and Food Safety (CAHFS) lab for Salmonella isolation. For each sample, one g of feces was homogenized in nine ml of Tetrathionate Broth (Hardy Diagnostics, Santa Maria, CA, USA) and incubated for 18–20 h at 37 °C. The Tetrathionate broth cultures of individual samples were inoculated on Xylose lysine tergitol 4 (XLT-4) agar plates using a cotton swab and streaked with a sterile loop. The plates were incubated for 18–24 h at 37 °C and suspect Salmonella colonies were identified if they had a red to pink periphery with black centers. Up to three distinct and spatially isolated suspect Salmonella colonies on the XLT-4 plate were selected per fecal sample and each colony was streaked onto Sheep Blood Agar for further biochemical testing. If at least one isolate was confirmed as an Salmonella isolate, the sample was labeled as Salmonella positive. For biochemical testing, the selected colonies were inoculated into urea agar slants, Motility Indole Ornithine (MIO), Citrate, O-Nitrophenyl-β-D-galactopyranoside (ONPG), Lysine Iron Agar (LIA) and Triple Sugar Iron (TSI) agar slants. Colonies were designated as suspect Salmonella if they were urease negative, motility positive indole negative ornithine positive (MIO), citrate positive, ONPG negative, lysine positive iron positive (LIA), dextrose fermenting, and produced H₂S (TSI). Suspect Salmonella colonies were confirmed using commercial polyvalent A1 and Vi antisera (DIFCO; Becton Dickinson Co., Sparks, MD, USA) following the manufacturer’s instructions.

Isolates were stored at −80 °C until completion of sampling in 2016, at which time all the Salmonella isolates were thawed and cultured for minimum inhibitory concentration (MIC) determination.

Antimicrobial susceptibility testing

Salmonella antimicrobial resistance was evaluated using broth microdilution method using a gram-negative Sensititre plate (CMV2AGNF) (Trek Diagnostic Systems Inc., Westlake, OH, USA) according the manufacturer’s instructions. Escherichia coli strain ATCC25922 was used as a quality control strain. The MIC values were the lowest concentrations of antibiotics that inhibited visible growth of bacteria. Interpretations of antibiotic resistance were set by the criteria of the MIC breakpoints recommended by the Clinical and Laboratory Standards Institute (Clinical & Laboratory Standards Institute (CLSI), 2014, 2018). A S. enterica serovar was defined as MDR if resistance to at least one antibiotic in each of three or more drug classes was observed (Magiorakos et al., 2012). Three drugs belonging to the gram-negative assay plate (Sensititre, Trek Diagnostic Systems, Cleveland, OH, USA), namely cefoxitin, streptomycin and gentamicin, were not classified using CLSI breakpoints. As outlined in the CLSI guideline, for Salmonella aminoglycosides, first- and second-generation cephalosporins, and cephamycins may appear active in vitro, but are not effective clinically and should not be reported as susceptible (Clinical & Laboratory Standards Institute (CLSI), 2018). Although descriptive data for MIC results were reported, susceptibility classification for these three drugs were not part of the analysis for factors affecting antimicrobial resistance in Salmonella isolates.

Statistical analysis

Descriptive analysis of the antimicrobial susceptibility classification of Salmonella isolates by antimicrobial drug was performed using the FREQ procedure in SAS (SAS Institute Inc., Cary, NC, USA). Descriptive analysis of Salmonella positive sample distribution, resistance phenotypes, and the proportion of Salmonella resistant to antimicrobial drugs were also performed using the FREQ procedure. To adjust for disproportionate sampling at different sampling times and between different farms, we utilized sampling weights in our statistical analysis (Pfeffermann, 1993). Weights were calculated in SAS using DATA, and were reciprocals of the probabilities of selection for the samples collected during a sampling visit; this is in relation to the total number of animals in the sampling population, which were cows to be culled that day on that farm (SAS, 2018). The survey-weighted prevalence of Salmonella shedding in the population of cull dairy cattle, accounting for number of animals sampled by farm, was estimated using the SURVEYMEANS function in SAS. Results for prevalence of samples were rounded to the closest integer.

Descriptive statistics

A summary of descriptive data for farm sampled is present in Table 1. Additional details of herd management and culling practices can be found in a report on Salmonella shedding in cull cows from the same herds (Abu Aboud et al., 2016).

Antimicrobial susceptibility results for Salmonella

A total of 239 fecal samples were collected from six different farms during the four seasons. A total of 60 fecal samples were collected for each season (10 from each farm). From these samples, a total of 62 Salmonella isolates were isolated from the 60 fecal samples. The only exception was for herd number 5 for a sample collected from a cow in the summer, where the cow’s identification number was erroneously entered and therefore was removed from the data set due to inability to match her with information from the farms record keeping system. The samples from this cow were culture negative for Salmonella. The survey-weighted prevalence of Salmonella positive fecal samples was 31% (95% confidence interval (CI) [26–35]). The most common drug class for which Salmonella isolates were resistant was tetracycline, followed by the penicillin and the cephalosporin classes (Table 2). Overall 12% (95% CI [3–20]) of Salmonella isolates were MDR (Table 2). Survey adjusted prevalence of Salmonella resistant to all 14 drugs tested as well as the MIC distribution is displayed in Table 3. The three top drugs for which isolates were resistant were tetracycline (39%; 95% CI [27–51]), ampicillin (18%; 95% CI [9–27]) and ceftriaxone (10%; 95% CI [2–17]). All isolates were susceptible to azithromycin, nalidixic acid and sulfisoxazole. AMR patterns for Salmonella are displayed in Table 4.

Cow-level factors associated with Salmonella antimicrobial resistance

Summer was significantly associated with detection of a fecal sample positive for Salmonella that was resistant to ceftriaxone (odds ratio (OR) 3.2; 95% CI [1.3–7.6]), or MDR (OR 2.3; 95% CI [1.03–5.2]) (Table 5). Spring was significantly associated with detection of a fecal sample positive for Salmonella that was resistant to ciprofloxacin (OR 7.7; 95% CI [2.0–28.5]), or ampicillin (OR 2.7; 95% CI [1.4–5.4]).

Cow-level factors tested for association with AMR Salmonella from a fecal sample are shown in Table 6. Using the two by two table analysis, the only cow-level factor associated with a lower OR for isolation of resistant Salmonella was culling a cow due to low milk production when compared to culling a cow due to any other reason; this was observed for both ciprofloxacin (OR 0.1; 95% CI [0.03–0.6]), and tetracycline (OR 0.4; 95% CI [0.2–0.6]) resistant isolates. All other cow-level factors for which a significant difference was observed was associated with an increase in the OR for isolating antibiotic resistant Salmonella. These include culling due to lameness when compared to culling for any other reason being associated with isolates resistance to ciprofloxacin (OR 14.9; 95% CI [4.0–54.8]), tetracycline (OR 2.4; 95% CI [1.3–4.4]) and ampicillin (OR 3.1; 95% CI [1.4–6.8]). Treatment of cows with ampicillin prior to culling was associated with isolates resistant to ciprofloxacin (OR 7.3; 95% CI [1.1–45.9]), and treatment with ceftiofur prior to culling being associated with isolates resistant to tetracycline (OR 2.0; 95% CI [1.05–3.8]).

Herd-level factors associated with Salmonella antimicrobial resistance

Spring was significantly associated with the probability of a herd being classified as having at least one fecal sample positive for ciprofloxacin resistant Salmonella when compared to other seasons combined (OR 13.6; 95% CI [2.9–62.4]) (Table 7). Winter was significantly associated with the probability of a herd being classified as having at least one fecal sample culture positive for Salmonella resistant to ceftriaxone (OR 4.3; 95% CI [1.3–14.4]).

Analysis of herd-level factors tested for association with isolating resistant Salmonella are shown in Table 8. Farms with number of milking cow greater than 3,000 was associated with higher odds of isolating at least one Salmonella resistant to tetracycline (OR 3.2; 95% CI [1.1–8.9]) when compared to farms with number of milking cows less than or equal to 3,000. Monthly percent of cows culls greater than 5% was associated with higher odds of isolating at least one Salmonella resistant to tetracycline (OR 9.5; 95% CI [3.1–29.0]) or ampicillin (OR 5.4; 95% CI [1.8–15.9]) when compared to farms reporting a monthly cull rate of less or equal to 5%. All herd-level factors for which a significant difference was observed was associated with an increase in the OR for isolation of resistant Salmonella. The selection of the cut-off point for creating the binomial variables number of milk cows in the herds was based on the mean number of milking cows in herds sampled, which was 3,083 cows (standard error (SE) 512%), with the cut-off point rounded to 3,000 (Table 1). The selection of the cut-off point for creating the binomial variable monthly cull rate was based on the mean monthly cull rate reported by farms in the study, which was 4% (SE 1%), and was rounded to 5% (Table 1).

Antimicrobial susceptibility results for Salmonella

Our study focused on AMR shedding of Salmonella sp. in cull cows, information regarding potential effect of sampling and microbiological methods on differences observed in Salmonella prevalence from samples collected from farms, as used in this study, has been previously discussed in a manuscript that focuses on epidemiology of Salmonella sp. in California cull dairy cattle (Abu Aboud et al., 2016). The most common drug class for which nontyphoidal S. enterica were resistant was tetracycline (Table 2). Similar findings have been observed in other studies, with prevalence of resistance to tetracycline varying from 13% to 44% (Cummings et al., 2013; Ray et al., 2007). Resistance to the penicillin drug class was the second most common, with ampicillin being the most common drug within that class for which Salmonella isolates were resistant (Table 3). Similarly, other studies have reported prevalence of resistance to ampicillin varying from 10% to 42% (Cummings et al., 2013; Ray et al., 2007). An important finding was the identification of the cephalosporin class as the third most common drug class for which Salmonella was resistant, with resistance to ceftriaxone and ceftiofur being the two most common drugs in that class (Table 3). Multiple reports for antimicrobial susceptibility of Salmonella isolated from dairy cattle in the U.S. have observed prevalence of resistance to ceftriaxone and quinolone drugs to be zero, with a higher prevalence of resistance observed for ceftiofur (14% to 20%) (Lundin et al., 2008; Ray et al., 2007). Third generation cephalosporins such as ceftriaxone, and fluoroquinolone drugs, such as ciprofloxacin, are the drugs of choice when treating severe nontyphoidal Salmonella infections in humans (Medalla et al., 2016). However, fluoroquinolones are not routinely prescribed for children due to possible fluoroquinolone-induced joint/cartilage toxicity, and third-generation cephalosporins are particularly important as a therapeutic option in this age group, making resistance to this drug class of greater public health relevance (Leibovitz, 2006). However, it should be noted that neither ceftriaxone nor ciprofloxacin is approved by the US Food and drug administration (FDA) for use in livestock, although other third generation cephalosporin drugs and fluoroquinolones are available for use in cattle. No fluoroquinolone drug is approved by FDA for use in lactating dairy cattle, and all extra-label uses of fluoroquinolone-class antimicrobials in food animals has been prohibited in the US since 1997 (Food and Drug Administration (FDA), 2017).

Two Salmonella isolates resistant to ciprofloxacin were also noted to be resistant to ceftriaxone. The lack of clarity of the role of antimicrobial drug use in cattle compared to other animals species on the spread of antimicrobial-resistant Salmonella to human populations highlights the need for further studies that can use molecular tools to accurately measure and evaluate and generate science based information to direct future efforts. An example is a recent study by Carroll et al. (2017) that used whole-genome sequencing to compare antimicrobial-resistant S. enterica serovars Typhimurium, Newport and Dublin isolated from dairy cattle and humans in Washington State and New York State (Carroll et al., 2017). Although an overlap of AMR genes between S. enterica isolates from dairy cattle and humans was observed in the Carroll et al. study, many genetic components resulting in AMR were confined to human isolates only, indicating that different factors may be playing a role in the emergence and spread of AMR S. enterica in humans and farm animals.

The survey-adjusted prevalence of Salmonella positive fecal samples is in agreement with other studies, including one study that collected samples from cull dairy cows at five non-fed beef slaughter establishments that received cull cows and bulls from cow-calf operations (non-fed beef), representing five regions of the United States, where they observed a non-weight adjusted prevalence of 23% (Troutt et al., 2001). A study by Brichta-Harhay et al. observed the mean prevalence of MDR Salmonella in cull cattle to be 17%, 12% and 0.3% for hides, pre-evisceration carcass samples, and post-intervention carcass samples, respectively. In our study we observed the average of the survey-adjusted prevalence of MDR Salmonella to be 12% (Table 2).

Cow-level factors Associated with Salmonella antimicrobial resistance

Two seasons were found to be significantly associated with resistance to specific drugs (Table 5). During the summer, higher OR for isolating MDR Salmonella and Salmonella resistant to ceftriaxone was observed, while spring was associated with higher OR for isolation of Salmonella resistant to ciprofloxacin and ampicillin. Further studies are needed to explain these differences, and should focus on identifying specific factors within each season that would results in the differences observed. It should also be noted that only two Salmonella isolates were identified as resistant to ciprofloxacin, and a high level of uncertainty for this analysis can be noted by the wide 95% CI; therefore findings for this drug must be cautiously interpreted. Differences in shedding patterns of resistant Salmonella could potentially be related to differences in weather conditions or management practices on farms in the regions where our study samples were collected.

Drug records indicating treatment at least once with ampicillin versus no exposure to ampicillin resulted in significantly higher OR for isolating Salmonella resistant to ciprofloxacin (Table 6). As already mentioned, no fluoroquinolone drugs are available to treat lactating cows, including in an extra-label matter. Furthermore, selection of resistance to ciprofloxacin under these conditions is most probably due to co-selection, although use of quinolone drugs in non-lactating cows in the herd (e.g., calves) could potentially select for resistant to the quinolone drug class. A study evaluated ciprofloxacin resistance of E. coli (which as Salmonella is an Enterobacteriaceae) in poultry in Australia, a country that has never permitted the use of fluoroquinolone in food-producing animals (Ingram et al., 2013), observed that 30% of samples collected form poultry carried fluoroquinolone non-susceptible E. coli. In their study, they hypothesized that the unexpected high prevalence of resistance to ciprofloxacin was probably related to co-selection, either by a clonal dissemination of chromosomal quinolone resistance determinants (vertical co-selection), or by dissemination of mobile genetic elements conferring resistance to fluoroquinolone drugs, such as aac(6′)-Ib-cr, qnrA, qnrB and qnrS. In their study, specific circumstances that could result in co-selection were not evaluated or suggested. Nevertheless, a study in humans has identified an association between non-fluoroquinolone therapeutic treatments and significant increase in selection of Enterobacteriacea resistant to fluoroquinolone drugs, including the use of beta-lactam drugs (Vien et al., 2012).

Drug records indicating treatment at least once with ceftiofur versus no exposure to ceftiofur resulted in significantly higher OR for Salmonella isolates being resistant to tetracycline (Table 6). Previous studies have identified co-selection to resistance mechanism other than those related with resistance to cephalosporins. A study evaluating the effect of parenteral treatment of cattle with ceftiofur resulted in selection of not only bla CMY-2, which is a gene related with phenotypic resistance to cephalosporins drugs, but also tet(A), a gene related with phenotypic resistance to tetracycline (Kanwar et al., 2014).

Lower OR was observed for Salmonella resistant to ciprofloxacin and tetracycline in cows culled due low milk production when compared to cows culled due to any other reason (Table 6). Factors that could have corroborated for these finding are not clear, and further research is needed to identify additional factors that could corroborate for these findings.

Culling cows due to lameness as compared to culling cows for any other reason resulted in higher odds for Salmonella resistant to ciprofloxacin, tetracycline and ampicillin. The current study did not collect information on the specific condition that resulted in the cow being culled due to lameness. This information could provide better understating about factors that could affect selection of a treatment of a lame cow. For example, a cow diagnosed with foot rot, interdigital dermatitis or a sole ulcer may receive different types of therapeutic treatment that could affect selection and shedding of resistant bacteria (Divers & Peek, 2008). The findings from our study highlight the need to collect further treatment information to better characterize specific factors that could be affecting antimicrobial resistance shedding patterns of cows culled for lameness.

A limitation of this study includes that data related to culling and drug treatment was based on farm records, which could contain error while being entered by farm personnel. Additional limitations include that samples from individual animals were collected in a cross-sectional fashion, and therefore we cannot infer causation for factors found to be associated with presence of resistance Salmonella in cull dairy cows. Other potential challenges that could affect precision of data collected at the farm were potential variability on the definition of disease at each farm.

Further limitations of the statistical approach used include not simultaneously accounting for the association of multiple factors with isolation of AMR Salmonella. The reasons for selecting Chi-square instead of an analysis that would account for more factors in the same model include low number of isolates resistant to specific drug classes in animals that were exposed to an outcome of interest. Furthermore, the lack of consistent disease definitions between the different dairies is another factor limiting further investigation of risk factors for Salmonella shedding. Therefore, our results should be viewed as initial effort to identify cow-level factors associated with shedding of resistant Salmonella, for which further studies could focus.

Herd-level factors associated with Salmonella antimicrobial resistance

At the herd-level, season was significantly associated with isolating AMR Salmonella (Table 7). For spring, the associations observed was similar to that observed at the cow-level analysis, with higher odds for resistance to ciprofloxacin observed during spring. Resistance to ceftriaxone, however was observed to be higher during the winter at the herd-level when compared to summer at the cow-level. As already discussed with results from cow-level association with season, additional studies are needed to elucidate factors resulting in the identified role of season.

Farms with greater than 3,000 lactating cows had significantly greater odds for isolation of at least one Salmonella resistant to tetracycline when compared to farms with 3,000 lactating cows or less (Table 8). Furthermore, monthly culling percent greater than 5% was observed to be associated with higher OR for selection of resistance at the herd level for tetracycline and ampicillin. As observed at the cow-level, culling due to different reasons resulted in higher odds for selection of resistance in some cases (e.g., culling due to lameness), and lower in others (e.g., culling due to low milk) (Table 6). Further studies that focus on specific factors that could better explain the associations between reasons for culling animals with greater odds or not of selection of AMR Salmonella are needed to outline specific factors that are associated with culling and could explain the results observed.

Our findings highlight how herd level management practices can be associated with increased shedding of AMR Salmonella in cull cows. Approaches based on decision-making criteria both at the individual and herd level have been proposed and demonstrated to be a potential alternative for improved economical outcome of culling decisions (Fetrow, Nordlund & Norman, 2006; Haine et al., 2017; Lehenbauer & Oltjen, 1998). Therefore, improving timely culling of cows with early or mild disease problems (or significant risk for disease) before progression to more serious disease which may require antimicrobial therapy or that could result in increased risk of ante or post-mortem condemnation of the carcass could be both economical and reduce AMR selection pressure.

A previous cross-sectional study conducted by Habing et al. (2012) conducted on 265 dairy herds in 17 states observed that some of the farm management practices associated with increased prevalence of Salmonella included using sprinklers or misters for heat abatement (OR 2.8; CI [1.6–4.9]), feeding anionic salts to cows (OR 1.9; CI [1.1–3.5]), and feeding ionophores to cows (OR 2.1; CI [1.2–3.7]) (Habing et al., 2012). In the Habing et al. study, they observed no significant associations between antimicrobial use and detection of AMR Salmonella on dairy farms, although other management practices related to manure management, including application on growing pasture or hay, and use of composted/dried manure for bedding in lactating cows were associated with significantly higher presence of at least one AMR Salmonella on a farm. Ways by which the Habing et al. study differed from ours include collecting samples from multiple classes of cows, as well as collecting samples from multiple states.

One factor to be aware of our study is that we only had six dairy farms for which samples were collected for four seasons for the herd-level analysis (total of 24 herd-level sampling points). Therefore, interpretation of results must take into account the number of dairy farms in the study, which could affect the external validity of the results to other dairy farms in California.

Misclassification, also known as information bias, is a very common source of bias that affects the validity of questionnaire answers (Althubaiti, 2016). Participants in our study were not blinded to the objective of the study and answers to the questionnaire administered and could have been affected by misclassification, due to answers that may have been perceived as more socially desirable. To reduce misclassification in our study, the questionnaire was carefully assembled and an internal validation process conducted. Furthermore, lack of standardized guidelines or benchmarks for many of the questions asked reduces concerns with answers biased toward a more desirable standard.