Evaluation of antimicrobial resistance and risk factors for recovery of intrauterine Escherichia coli from cows with metritis on California commercial dairy farms

The goals of this study were to evaluate factors affecting recovery and antimicrobial resistance (AMR) in intrauterine E. coli in post-partum dairy cows with and without metritis from commercial California dairy farms. Using a cross-sectional study design, a total of 307 cows were sampled from 25 farms throughout California, from which a total of 162 intrauterine E. coli isolates were recovered. During farm visits, cows within 21 days post-partum were categorized in one of three clinical presentation groups before enrollment: metritis (MET, n = 86), defined as a cow with watery, red or brown colored, and fetid vaginal discharge; cows with purulent discharge (PUS, n = 106), defined as a non-fetid purulent or mucopurulent vaginal discharge; and control cows, (CTL, n = 115) defined as cows with either no vaginal discharge or a clear, non-purulent mucus vaginal discharge. Cows diagnosed as MET had significantly higher odds for recovery of E. coli compared to cows diagnosed as CTL (OR = 2.16, 95% CI: 1.17–3.96), with no significant difference observed between PUS and CTL, and PUS and MET. An increase in days in milk (DIM) at the time of sampling was significantly associated with a decrease in the odds ratio for E. coli recovery from intrauterine swabs (OR = 0.94, 95% CI: 0.89–0.98). All intrauterine E. coli were resistant to ampicillin (AMP), with an AMR prevalence of 30.2% and 33.9% observed for chlortetracycline and oxytetracycline, respectively. Only 8.6% of isolates were resistant to ceftiofur (CEFT), one of the most common drugs used to treat cows on farms sampled. No significant difference in the prevalence of AMR was observed among clinical groups at the individual cow level. At the farm level, a significantly higher odds for isolating intrauterine E. coli resistant to chlortetracycline (OR: 2.6; 95% CI: 3.7–58.0) or oxytetracycline (OR: 1.9; 95% CI: 1.4–33.8) was observed at farms that used an intrauterine infusion of oxytetracycline as a treatment for metritis when compared to those farms that did not use this practice. Findings from this study indicate the need for further research supporting a broader understanding of farm practices driving AMR in cows with metritis, as well as data to increase the accuracy of breakpoints for AMR classification of intrauterine E. coli from cattle.


Results
Descriptive data. The number of E. coli samples recovered and information on animal samples by farm is presented in Table 1. A total of 307 cows were sampled from the 25 enrolled farms. All enrolled farms had at least one cow assigned to each of the three clinical classifications of vaginal discharge, except for two farms where no MET cows were identified during our visit. DIM at time of diagnosis for cows with culture positive results for E. coli for MET, PUS, and CTL were 8.1 (95% CI 7.1-9.1), 10.6 (95% CI 9.1-12.0), and 10.6 (95% CI 8.9-12.3), respectively.
Risk factors for E. coli isolation. Risk factors for E. coli isolation are presented in Table 2 and supplemental table 1. The odds ratio for isolating intrauterine E. coli isolate from MET cows when compared to CTL cows was 2.0 (95% CI: 1.1-3.7, P value = 0.03). No significant difference between cows diagnosed as PUS and CTL, or a MET and PUS was observed for isolation of E.coli from intrauterine swabs. Days in milk of the cow sampled were significantly associated with lower odds of isolating intrauterine E. coli for each day increase in DIM for MET and PUS cows ( Table 2).

E. coli antimicrobial resistance. The distribution of minimum inhibitory concentration (MIC) and
resistance for intrauterine E. coli (n = 162) by individual drug for the BOPO6F panel are shown in table 3. The most common resistance profiles and the resistance profile for each isolate and are presented in supplemental tables 2 and 3, respectively. No significant association (P > 0.05) was observed at the animal level between E. coli AMR for the nine drugs with clinical breakpoints and the clinical presentation group.
The percent of all E. coli isolates classified as susceptible for the four antimicrobial drugs commonly used to treat cows with metritis in the U.S. 8 , are presented in Fig. 3. Of the nine antimicrobials tested with available MIC breakpoints, AMP had the highest prevalence of AMR, with all isolates being classified as resistant (Table 3). Although all isolates were resistant to ampicillin, nearly 60% of isolates (n = 97) were resistant to AMP alone (supplemental table 10. A total of 8.6% of isolates (n = 12) included ceftiofur resistance within their total AMR resistance profile (supplemental table 2).
Heat map of MICs for antimicrobials for the 162 E. coli isolates grouped by clinical presentation group (CTL, MET, PUS) and farm, are shown in Figs. 1 and 2, respectively. When visually comparing CTL, MET, and PUS groups in Fig. 1, no clinical presentation group had a noticeable visual trend for percentile distribution toward higher MIC quantiles. When comparing farms in Fig. 2, there was a visual clustering of isolates in a higher MIC quantile for specific farms, particularly for chlortetracycline and oxytetracycline for farms 18 and Table 1. Distribution of E. coli (n = 162) isolated from intrauterine swabs collected at 25 commercial dairy farms by clinical presentation group (CTL, MET, and PUS). A, swabs positive for E. coli; B, total number of swabs collected from cows in clinical group; C, total number of swabs collected at each farm. a Clinical presentation of cows when intrauterine samples were collected. b Percentage, (Swabs positive for E. coli / total number of swabs collected from cows in clinical presentation group). c Percentage, (Swabs positive for E. coli / total number of swabs collected at each farm). d Percentage, (Swabs positive for E. coli / total number of swabs from all farms for cows in clinical group).   Antimicrobial treatment on E. coli AMR. A total of 11% (18/162) of animals sampled for which E. coli was isolated received antimicrobial treatment within the fourteen days prior to sampling. For cows diagnosed with metritis, 26% (12/46) had a prior treatment with antimicrobials, and 11% (5/45) and 2% (1/53) for PUS and CTL cows, respectively. Based on survey data collected for farms, we would expect a higher number of cows with metritis being treated; however, our research team selected cows for enrollment independent of prior diagnosis of metritis, resulting in a metritis diagnosis occurring prior to that of the farm. This is not surprising, given that most farms used systemic signs of disease to diagnose metritis (e.g., depressed attitude, drop in milk) (Table 4). Indicating a more severe case; the approach used by our study was based on vaginal discharge, which allows an earlier diagnosis of metritis, and therefore explaining the reason for metritis cases identified in not having received a prior antimicrobial treatment by the farm. Results for Fisher's Exact Test analysis evaluating the effect of individual animal treatment with antimicrobials is shown in supplemental table 4, and was not associated with a significant increase of AMR in E. coli. When this analysis was stratified by clinical presentation, again, no significant effect of individual animal treatment on AMR within E. coli was observed. Results for Fisher's Exact Test analysis evaluating the effect of individual animal treatment with ceftiofur alone is shown in supplemental table 5; Individual animal treatment with only ceftiofur 14 days prior to sampling did not result in any significantly increased odds ratios for antimicrobial resistance to the nine antimicrobials analyzed. When this analysis was conducted stratifying by clinical presentation group (CTL, MET, and PUS), there were also no significantly increased odds for antimicrobial resistance to the seven antimicrobials analyzed for (supplemental table 6).
Results for the Fisher's Exact Test evaluating the effect of individual animal treatment with tetracycline alone is shown in supplemental table 7; individual animal treatment with tetracycline 14 days prior to sampling did not result in any significantly increased odds ratios for antimicrobial resistance to the nine antimicrobials analyzed. When this analysis was conducted stratifying by clinical presentation group (CTL, MET, and PUS), there were also no significantly increased odds for antimicrobial resistance to the seven antimicrobials analyzed for (supplemental table 8).
The mixed-effect multinomial logistic regression model used to evaluate the association between intrauterine E. coli AMR and animal-level variables, did not identify treatment group or any other animal-level variable evaluated, including prior antimicrobial treatment, as being significantly associated with AMR in E. coli.
Farm-level antimicrobial treatment and management practices. Descriptive data by farm related to the number of dairy cows and metritis diagnosis and treatment criteria are presented in Table 4. Farms in which cows were treated for metritis using intrauterine treatment with oxytetracycline had a significantly higher www.nature.com/scientificreports/ farm-level prevalence of intrauterine E. coli with AMR to oxytetracycline (LSM ± SEM: 0.82 ± 0.09) and chlortetracycline (LSM ± SEM: 0.70 ± 0.14) when compared to farms in which cows were not treated for metritis using intrauterine treatment with oxytetracycline (LSM ± SEM: 0.03 ± 0.03 and LSM ± SEM: 0.25 ± 0.04, respectively) ( Table 5).

Discussion
Recovery of E. coli from intrauterine swab samples. E. coli was recovered from 53% of intrauterine swab samples collected from all post-partum dairy cows in our study, which is within the range previously observed by other researchers. Bicalho et al. 16  MET cows had significantly higher odds ratio for isolation of intrauterine E. coli when compared to CTL cows (Table 2), and is in agreement with previous studies 17,18 . A study by Pohl et al. (2018) that isolated intrauterine E. coli from cows using two clinical signs to define metritis (reddish-brown fetid discharge and rectal temperature > 39.5 °C) had a 90% recovery of intrauterine E. coli, and 70% recovery when using solely one clinical sign (reddish-brown fetid discharge or rectal temperature > 39.5 °C). In contrast, cows not displaying clinical signs of metritis had an E. coli recovery of 54% 13 . The Pohl et al. 13 study also observed that cows with two clinical signs of metritis had 7.16 times the odds of having intrauterine E. coli compared to cows without metritis, in agreement with our findings. The difference in the magnitude of E. coli recovery in cows with metritis between Pohl et al. 13 and our study may be explained by differences in herd management practices and geographical factors from German dairy farms that resulted in a different magnitude of recovery of E. coli in cows with metritis.
A higher days in milk (DIM) at the time of MET or PUS diagnosis was found to be significantly associated with a lower odds of isolation of intrauterine E. coli ( Table 2). The relationship between DIM and odds to isolate E. coli agrees with previous findings 19 , where the progression of the uterine microbiota from calving was evaluated, with an observed rapid decrease in the relative abundance of Proteobacteria, a major phylum of Gram-negative bacteria that includes E. coli, from 0 to 6 ± 2 DIM. Jeon et al. 19 also observed a subsequent increase in the relative abundance of bacteria in the phylum Bacteroidetes from 0 to 6 ± 2 DIM. While the dynamics of the uterine microbiome are complicated, particularly at the time of parturition, increases in relative abundance of other microbes likely drive the decrease in abundance of Proteobacteria; therefore decreasing the odds of isolating intrauterine E. coli 20 . A definitive explanation behind this phenomenon remains elusive and continues to be a topic of research. E. coli AMR. When comparing CTL, PUS, and MET cows, no significant difference in the odds ratio for isolating AMR E. coli isolates to the AMDs tested was observed. This is in discordance with our hypothesis, that had an assumption that cows with MET may have been colonized by E. coli carrying both virulence and antimicrobial resistance genes. Previous studies have shown that specific virulence genes are associated with intrauterine E coli isolated from cows with metritis 16 . Furthermore, a study from cows from a single farm, using whole genome sequencing to characterize intrauterine E. coli, observed a correlation between intrauterine pathogenic E. coli (characterized based on presence of virulence genes) and extended spectrum β-lactamase (ESBL) genes, which confer resistance to expanded-spectrum cephalosporins 21 . Discordance between our results using phenotype methods for AMR diagnosis and those using genomic approaches could be due to disagreements that have been reported between these two methods.
Resistance to ceftiofur, the most common systemic antimicrobial treatment for metritis in California, was low in our study with 8.6% of isolates (n = 12) phenotypically resistant 8 . Similar studies conducted in New York State, New Zealand, and Germany also observed low AMR (as specified by CLSI breakpoints available at the time of publication) to CEFT within uterine E. coli; with 1.2%, 0%, and 5.9% of isolates resistant, respectively [11][12][13] .
Extra-label use of ampicillin has previously been reported as the second most common treatment option for metritis in California dairy cows 8 . In the literature, intrauterine susceptibility of E. coli to AMP has varied, possibly due in part to the use of breakpoint values that have been periodically updated 12 . As an example, a study by 11 observed that approximately 34% of early post-partum cows harbored ampicillin-resistant E. coli. Due to Table 5. Association of farm-level management of using oxytetracycline as an intrauterine infusion as the most common drug for the treatment of metritis and farm-level prevalence of AMR to tetracycline drugs in intrauterine E. coli. a Parameter estimate for the multivariate model evaluating resistance to the referred drug. b Least-square means (LSM) of farm-level prevalence of intrauterine E. coli resistant to the referred drug. c Standard error of the means (SEM) for the LSM. 4 P value from analysis comparing farm-level prevalence of AMR between farms for the referred drug, adjusted using Bonferroni. This large discrepancy is a reflection of how MIC breakpoints are defined for Enterobacterales in humans and animals. Traditionally, MIC breakpoints are set using a range of data, including in vitro microbiological data, animal and human pharmacokinetic/pharmacodynamic (PK/PD) data, and clinical and bacteriological outcome data from prospective clinical studies 24 . Furthermore, the accuracy of the data used to determine MIC breakpoints for specific animal species and tissues will directly affect the validity of the results related to SIR classification. Oxytetracycline is approved in the US for the systemic treatment of metritis caused by species of staphylococci and streptococci. A study by 12 reported that of 209 intrauterine E. coli, 83.2% of were susceptible and 4.8% were resistant to oxytetracycline. In Germany, a study reported that of 85 intrauterine E. coli isolates, 81.1% were susceptible and 9.5% were resistant to tetracycline 13 . In our study, 30.3% of E. coli tested against chlortetracycline and 33.9% of E. coli tested against oxytetracycline were classified as resistant. The observed higher prevalence of resistance to tetracyclines in our study when compared to previous studies may reflect specific practices for managing metritis in California. In Fig. 2, the visual clustering of isolates in a higher MIC quantile collected from the two farms that indicated using tetracycline drugs as their first choice for treatment of metritis in dairy cows suggests that treatment of cows with metritis using tetracyclines may increase the selection of AMR to drugs in that class. Future studies should be designed to allow for the determination of causation.

Farm-level risk factors
Our study revealed that using oxytetracycline intrauterine infusion as a treatment tended to increase the odds of recovering intrauterine E. coli carrying resistance to either chlortetracycline or oxytetracycline. In California, approximately 27% of farms used intrauterine infusion with oxytetracycline for treatment of metritis 8 . However, researchers have recommended against intrauterine infusions due to lack of evidence to support any added benefit in reproduction or cure 25,26 . Given the extra-label nature of the use of intrauterine oxytetracycline in cattle, this practice can only occur though the prescription of a veterinarian, and appropriate milk withhold periods should be followed 27 .
The fluoroquinolone enrofloxacin is approved for use in treating respiratory disease in non-lactating dairy cattle under 20 months of age, however, its extra-label use to treat bovine metritis is illegal 28 . Due to drug dilutions present on the plate, we could only classify isolates as susceptible and intermediate. As such, we can only report that enrofloxacin susceptibility was high with 96.9% of isolates classified as susceptible (Fig. 3). None of the farms in our study reported using enrofloxacin as a treatment option for metritis. Our findings suggest very low resistance of intrauterine E. coli to enrofloxacin, supporting the expected lack of use of this drug in lactating dairy cattle. Fluoroquinolone resistance within our isolates could potentially be originating from horizontal gene transfer or from co-selection due in part to the use of approved AMDs to treat metritis; however, further study to investigate this would be necessary 29,30 .
While resistance to each of the antimicrobials discussed above is concerning, multidrug resistance (MDR) is an especially pressing issue if an isolate is resistant to multiple common treatment options. The most common resistance profile was to AMP alone (59.9% of isolates), while resistance to ampicillin-chlortetracycline-oxytetracycline (18.5% of isolates) was the 2nd most common profile. Our study showed the presence of resistance to the common drugs used to treat metritis in California with 2.5% of isolates (n = 4) resistant to ampicillin-ceftiofur-chlortetracycline-oxytetracycline. The relatively low prevalence of MDR within intrauterine E. coli isolates   Santos et al. (2010) in which a total of 35% (n = 80) of E. coli isolates from cows with metritis were MDR, with the major MDR profile being ampicillin-chloramphenicol-florfenicol resistance, observed in 96.4% of MDR isolates 11 . However, our results do more closely resemble those of Abdelfattah et al. 31 in which a total of 14.14% (n = 307) of 2171 E. coli isolates, recovered from fecal samples from healthy, adult dairy cows from 10 farms in California, were MDR. Their major MDR profile being florfenicol-sulphadimethoxine (16.2%), and tetracycline-florfenicol-sulphadimethoxine (6.82%). For farm level analyses, an important consideration when extrapolating results, is that the results represent findings for a specific population within the herd. Specifically, fresh cows (within the first 21 days in milk) that were sampled and also had an E. coli culture-positive result (cows without a culture positive result were not included in this analysis). Within these constraints, our results aimed to have internal validity for this specific population.

Conclusion
An increase in DIM at the time of sampling was significantly associated with a decreased odds for E. coli recovery, while classification within the MET clinical presentation group at the time of sampling was significantly associated with increased odds of recovery of E. coli from intrauterine swabs. A low prevalence of AMR to CEFT, the most common metritis treatment in our study, within E. coli was observed. The extra-label use of intrauterine infusion with oxytetracycline on two farms was observed as a significant factor for increased farm-level prevalence of intrauterine E. coli with AMR to oxytetracycline and chlortetracycline, highlighting the potential impacts of this practice on antimicrobial stewardship. Our findings support the need for further research to support a better understanding of farm practices driving AMR in cows with metritis, as well as data to support breakpoints that will result in more accurate AMR identification within intrauterine E. coli from cattle.

Materials and methods
The University of California Institutional Animal Care and Use Committee (IACUC; #20620) approved all experimental procedures conducted with animals for this study. The UC Davis Institutional Review Board (IRB) Administration granted an exemption (IRB ID 1307716-1) for all experimental procedures for this study.

Study design. A convenience sample of 25 commercial dairy farms from the Sacramento and San Joaquin
Valleys in California was recruited with the help of local veterinarians and UC Davis faculty and extension advisors. The study was conducted between September 2018 and November 2019.
Using a cross-sectional study design, intrauterine swabs were collected from post-partum cows between 3 and 21 DIM; cows that were unable to stand were not eligible for enrollment in the study. Three clinical presentation groups were defined based on vaginal discharge (VD) characteristics 32 as metritis discharge (MET): watery, reddish or brownish, and fetid), purulent discharge (PUS): non-fetid purulent or mucopurulent vaginal discharge), and normal discharge (CTL): clear lochia, clear mucus, or no vaginal discharge. Due to sampling time limitations, five cows per clinical group were targeted as the maximum number per dairy.
Researchers visited each of the 25 farms once during the morning lockups of fresh cow pens, while farm employees were performing their own health checks. Researchers (R.V.P. and A.G.) collected vaginal discharge from cows using a Metricheck™ device (Simcrotech, Hamilton, New Zealand) cleaned with 2% chlorhexidine gluconate solution between cows. The VD was evaluated by sight and smell, and cows were assigned to the corresponding clinical presentation group. Evaluation of animals was conducted independent of data from the farm on prior diagnosis of metritis, and were conducted independently of farm employee findings.
For animals selected to be enrolled in the study, rectal temperature was measured using the GLA M900 thermometer (GLA Agricultural Electronics, San Luis Obispo, CA, USA). Prior to intrauterine sample collection, researchers cleaned the vulva using dry paper towels and 70% isopropyl alcohol. A 30-inch double-guarded sterile culture swab (McCullough; Jorgensen Labs Inc., Loveland, CO, USA) was gently passed through the vulva and cervix until reaching the uterine body. The swab was exposed and rolled against the uterine wall, retracted within the double sheath, removed from the cow, and immediately placed in Amies transport media with charcoal (BBL™ CultureSwab™ Plus; Becton, Dickinson, and Company, Sparks, MD, USA). The swabs were kept on ice until inoculation on to solid medium in the laboratory within 24 h of sample collection. Individual animal antimicrobial treatment in the last 14 days for animals sampled was recorded.
A sample size calculation was conducted based on our first hypothesis that dairy cows diagnosed with metritis will have a higher risk for isolation of intrauterine E. coli when compared to CTL cows. For this purpose, an a priori sample size estimation for the proportion of E. coli culture positive cows in MET and CTL was made based at a 90% power (σ: 0.1; α: 0.05; μ MET. : 0.7, μ CTL : 0.4) was calculated in JMP Pro 16.0 (SAS Institute Inc., Cary, NC), resulting in a minimum of 53 animals per clinical group.
Survey and treatment records. A survey questionnaire was developed to collect information on farm characteristics, management practices, and antimicrobial treatment regarding metritis. The survey was administered by R.V.P or A.G. and targeted dairy managers and animal handlers. On-farm antimicrobial treatment history for each sampled cow for the last fourteen days was collected via either electronic records or after interviewing workers during farm visits. Data collected from interviews were entered into spreadsheets for analysis (Microsoft Office Excel 2010, Microsoft Corp., Redmond, WA). Within  www.nature.com/scientificreports/ cultured in 10 mL of brain heart infusion broth (Difco; Becton, Dickinson, and Company, Sparks, MD, USA) at 37 °C for 24 h. The broth culture (500 µL) was mixed with 50% sterile glycerol/50% sterile water solution (500 µL) prior to storage at − 80 °C. For all isolates, antimicrobial susceptibility testing was conducted in batches after completion of sample collection using a broth microdilution method following the Clinical Laboratory Standards Institute (CLSI) guidelines 14 . The Sensititre Vet Bovine/Porcine plate (BOPO6F, Trek Diagnostic Systems, Oakwood Village, OH, USA) was used for testing susceptibility to the following antimicrobial drugs: penicillins (penicillin and ampicillin), cephalosporins (ceftiofur), fluoroquinolones (danofloxacin and enrofloxacin), phenicols (florphenicol), sulfas (sulphadimethoxine and sulfamethoxazole/trimethoprim), tetracyclines (chlortetracycline and oxytetracycline), macrolides (tylosin tartrate, tulathromycin, and tilmicosin), aminoglycosides (gentamicin and neomycin), lincosamides (clindamycin), pleuromutilins (tiamulin), and aminocyclitols (spectinomycin). Sensititre plates were read manually, and minimum inhibitory concentrations were interpreted using current CLSI breakpoints when available 14 (supplemental table 9).

Bacterial isolation and antimicrobial susceptibility testing.
Prior to the veterinary CLSI guidelines, the only option to define the susceptible, intermediate or resistant (SIR) classification of E. coli isolates from cows with metritis was to utilize human-based breakpoints, which have been used for currently available studies in literature. However, as defined by CLSI VET09 guidelines for extrapolating breakpoints for veterinary pathogens, the use of human-based breakpoints result in SIR interpretations that have very low confidence, and are not recommended 15 . Instead, as per Chapter 8 of the CLSI VET09 document, which focuses on bovine-specific breakpoints, the recommendation for defining breakpoints when they are not available for specific bacteria or anatomical locations, is to apply a different bacterial species or infection site from a bovine-specific source. Based on these guidelines by CLSI, we utilized this updated approach to maximize the accuracy of SIR classification of isolates in the study (supplemental table 9).
Data management and statistical analysis. CLSI MIC breakpoints were available for only nine of the 18 AMDs tested (ampicillin, ceftiofur, chlortetracycline, oxytetracycline, florfenicol, gentamicin, danofloxacin, spectinomycin, and enrofloxacin) (supplemental table 9). The nine drugs tested and later used in the analysis had CLSI breakpoints for either Enterobacterales or Pasturella multocida in cattle and horses, as recommended by CLSI. Isolates that grew in all dilutions of an antimicrobial assessed were classified as "Growth in all dilutions" (GAD) because their MIC was higher than the highest dilution tested in our study (supplemental Fig. 1). By using GAD, we stratified the data between isolates for which the highest concentration in the plate was the actual MIC (e.g., CEFT for isolates where the MIC = 8) from those that grew at the highest concentration available on the MIC plate (e.g., CEFT for isolates where the MIC > 8), for which the actual MIC value is unknown. Antimicrobial drug resistance profiles for intrauterine E. coli isolates are presented in supplemental table 2.
Risk factors for E. coli isolation. A logistic regression model in SAS (SAS Institute Inc., Cary, NC; version 7.15) using PROC GLIMMIX logit function was used to evaluate animal-level risk factors collected at the time of sampling on the odds of isolating E. coli from an intrauterine swab sample. The dependent variable was the binomial variable for culture-positive or negative for E. coli, and the independent variables were clinical presentation group (MET, PUS, or CTL), lactation number (1, 2, and 3 or greater), days in milk (DIM), and fever (categorical variable, with 39.5 ºC as a fever benchmark) at time of sampling 1 . All interactions were considered in the model. Univariate analysis for each explanatory variable was conducted; all variables with a P < 0.3 were selected to be offered to the model using a backward stepwise elimination process. Farm was controlled as a random effect in the model. The quadratic association between DIM and outcomes of interest were evaluated and retained in the model if significant. Pairwise comparisons between the clinical presentation groups were conducted, adjusting for multiple comparisons using the Tukey-Kramer approach. The Akaike information criterion (AIC) was used for model selection and to ensure a more parsimonious model was selected. Clinical presentation group was forced into all models regardless of the p-value. A variable was considered a confounder if the coefficient of a significant variable in the model changed ≥ 20% after removal from the model. All models included farm as a random effect.
Evaluation of antimicrobial treatment on AMR prevalence. Univariate analysis was conducted to evaluate the effect of individual animal antimicrobial treatment on AMR in E. coli (n = 162), independent of clinical presentation group, as well as by stratifying the analysis by treatment group. A binary variable was created for being treated in the preceding 14 days prior to sampling with any antimicrobial drug. Because most animals that had an E. coli isolate and received any antimicrobial treatment were treated with ceftiofur or tetracycline (17/18), antimicrobial specific binary variable were created, where animals were either treated with that specific antimicrobial or did not receive that specific antimicrobial (supplemental Risk factors for E. coli antimicrobial resistance. A logistic regression model using the Logit function in PROC GLIMMIX in SAS (SAS Institute Inc., Cary, NC) was used to evaluate the association between intrauterine E. coli AMR and animal-level variables. Univariate analysis between each explanatory variable and the categorical binomial variables for ampicillin, ceftiofur, chlortetracycline, oxytetracycline, florfenicol, gentamicin, danofloxacin, spectinomycin, and enrofloxacin as resistant or susceptible was used to identify tests with a P < 0.3; these were selected to be offered to the model using a backward stepwise elimination process. A model was generated for each of the nine AMDs with available breakpoints using a categorical binomial variable to clas- www.nature.com/scientificreports/ sify an isolate as resistant or susceptible. Independent individual animal-level variables offered to the model were clinical presentation group (retained in all models), lactation number, rectal temperature, and days in milk at the time of sample collection, and antimicrobial treatment in the last 14 days with either ceftiofur or tetracycline drug. Farm was controlled as a random effect in the model. The Akaike information criterion (AIC) was used for model selection and to ensure a more parsimonious model was selected. Clinical presentation group was forced into all models regardless of the P-value. Confounding effects were evaluated by examining the effect of the removing variables on the coefficients of the remaining variables. A variable was considered a confounder if the coefficient of a significant variable in the model changed ≥ 20% after removal from the model. Mixed-effect multinomial logistic regressions were used for the analysis of binomial data for AMR categorization of an isolate for each AMDs with MIC breakpoints using the logit link function in PROC GLIMMIX; in this model the response variable was a proportion using the events/trials syntax, where the events were the number of intrauterine E. coli isolates at a farm with AMR to the antimicrobial drug being evaluated (events) out of the total intrauterine E. coli isolated from that farm (trials) 33 . More specifically, the dependent variable was the number of isolates with AMR to the antimicrobial drug being evaluated at a farm (events) out of the total intrauterine E. coli isolated from that farm (trials). Using this approach, the models assessed the least square means for the prevalence of AMR at the farm level. The explanatory variables offered to the model were farm-level practices, including antimicrobial drugs commonly used as first choice for treatment of metritis on the farm. Using this approach, the models assessed the association between AMR proportion at the farm level and surveyed farm practices.
A model was generated for each of the nine AMDs with MIC breakpoints to evaluate the association of farmlevel prevalence of E. coli AMR and farm-level management practices as explanatory variables. Individual models were created for ampicillin, ceftiofur, chlortetracycline, oxytetracycline, florfenicol, gentamicin, danofloxacin, spectinomycin, and enrofloxacin. Models were built and evaluated as previously described.
Heat maps for isolate susceptibility to antimicrobials. Heat maps representing each individual isolate and its susceptibility to 12 antimicrobials by clinical presentation group and farm were created using RStudio (Version 1.4.1106) (R Foundation for Statistical Computing, Vienna, Austria) using the heatmap.2 function.
Of the eighteen total drugs tested, five drugs were not included (tiamulin, sulfadimethoxine, trimethoprimsulfamethoxazole, tylosin, and clindamycin) because these drugs had either fewer than two antimicrobial concentrations tested, or more than 98% of isolates had the same MIC value. The percentile scale for susceptibility to antimicrobials was generated after categorizing MIC dilution ranges available for each antimicrobial in ascending order, representing the percent decrease in susceptibility for the evaluated range. As an example, for oxytetracycline, five antimicrobial dilution concentrations were available in the MIC plate (0.5, 1, 2, 4, and 8 μg/ mL), generating a percentile decrease in susceptibility with increments of 25%, assigned a percent category of decreased susceptibility 0%, 25%, 50%, 75%, and 100%.

Data availability
The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.