Switching between bacteriostatic and bactericidal antimicrobials for retreatment of bovine respiratory disease (BRD) relapses is associated with an increased frequency of resistant pathogen isolation from veterinary diagnostic laboratory submissions

Although 90% of BRD relapses are reported to receive retreatment with a different class of antimicrobial, studies examining the impact of antimicrobial selection (i.e. bactericidal or bacteriostatic) on retreatment outcomes and the emergence of antimicrobial resistance (AMR) are deficient in the published literature. A survey was conducted to determine the association between antimicrobial class selection for retreatment of BRD relapses on antimicrobial susceptibility of Mannheimia haemolytica, Pasteurella multocida, and Histophilus somni. Pathogens were isolated from samples submitted to the Iowa State University Veterinary Diagnostic Laboratory from January 2013 to December 2015. A total of 781 isolates with corresponding animal case histories, including treatment protocols, were included in the analysis. Original susceptibility testing of these isolates for ceftiofur, danofloxacin, enrofloxacin, florfenicol, oxytetracycline, spectinomycin, tilmicosin, and tulathromycin was performed using Clinical and Laboratory Standards Institute guidelines. Data were analyzed using a Bayesian approach to evaluate whether retreatment with antimicrobials of different mechanistic classes (bactericidal or bacteriostatic) increased the probability of resistant BRD pathogen isolation in calves. The posterior distribution we calculated suggests that an increased number of treatments is associated with a greater probability of isolates resistant to at least one antimicrobial. In addition, the frequency of resistant M. haemolytica isolates was greater with retreatment using antimicrobials of different mechanistic classes than retreatment with the same class. Specifically, treatment protocols using a bacteriostatic drug first followed by retreatment with a bactericidal drug was associated with a higher frequency of resistant BRD pathogen isolation. This effect was more profound with specific treatment combinations; tulathromycin (bacteriostatic) followed by ceftiofur (bactericidal) was associated with the highest probability of resistant isolates among all antimicrobial combinations. These findings suggest that the selection of antimicrobial mechanistic class for retreatment of BRD should be considered as part of an antimicrobial stewardship program.

8 141 antimicrobials used, and non-antimicrobial treatments was manually extracted from these records 142 by one investigator (AS). Isolates from submissions that explicitly stated no usage of 143 antimicrobial drugs were assigned the treatment history classification of none ("0"). Isolates 144 from cases in which information regarding antimicrobial treatments was unclear (e.g., "many" or 145 "everything") or not given were classified as "unknown." Isolates from cases with treatment 146 histories indicating the use of four or more antimicrobials were classified as "4+."

147
Trade names were converted to generic drug names to determine the antimicrobial drug 148 class (bacteriostatic or bactericidal) and the sequence of class administration for first-and 149 second-line treatments. Drug class was assigned based on the established in vitro 150 pharmacodynamics of the antimicrobial agent as summarized in Table 1.
151 Table 1 (Tables S1 and S2). Finalized case report information, such as 156 microscopic evidence of pneumonia, also was noted. Case information was classified as 157 "unknown" if the information was not supplied or unclear. After each eligible record was 158 identified, the submission forms for each case were individually reviewed by a single researcher 159 (AS). Antimicrobial treatments were grouped as -cidal or -static based on antimicrobial activity 160 level.
161 Variable transformations 162 Due to sparse data for cases receiving multiple treatments, we arbitrarily chose to group 163 together animals that received more than three treatments (4+). Animals with unknown treatment 164 histories were excluded from the analysis.

165
For the subset of animals receiving just two treatments, we created two categorical 166 variables. One categorical variable grouped the data into two levels: "same" to designate animals 167 that received first-and second-line treatment from the same drug class (i.e., either bacteriostatic 168 and bacteriostatic or bactericidal and bactericidal) and "different" to designate animals that 169 received first-and second-line treatment from different drug classes (i.e., either bacteriostatic 170 followed by bactericidal or bactericidal followed by bacteriostatic). We also created a four-level 171 categorical variable to capture all possible combinations (4 levels: bacteriostatic followed by 172 bactericidal, bacteriostatic followed by bacteriostatic, bactericidal followed by bacteriostatic, and 173 bactericidal followed by bactericidal). 188 where represents the category (i.e., antimicrobial drug class) and n = 18 represents the 189 number of possible antimicrobials. Thus, the probability density function of is: 190 .

286
Antimicrobial treatments were grouped based on their anticipated impact on bacterial 287 growth in vitro, i.e., bactericidal ("cidal") or bacteriostatic ("static"). We created a heatmap to 288 illustrate the impact of specific pairs of combinations of first and second antimicrobial treatments 289 on the number of isolates identified as resistant against the listed antimicrobials with CLSI 290 breakpoints (Fig. 2). Red indicates the observed maximum number of resistant isolates and white 291 (i.e., blank) represents no observation of antimicrobial resistance for a specific antimicrobial 292 combination (Fig. 2). A similar examination of the data was not conducted for P. multocida and 293 H. somni because there were an insufficient number of isolates for this to be meaningful.

304
Due to the limited number of isolates available from animals that received more than 2 305 treatments, we did not explore or conduct sensitivity analysis on the impact of other possible 306 antimicrobial combinations on the isolation of resistant organisms. We also did not explore 307 alternatives to the priors chosen for the Bayesian analysis, as we considered the chosen priors to 308 be the most biologically defensible. A variable to account for the non-independence of isolates 309 from the same animal was not included in the model, as the number of these cases was relatively 310 small.

311
The distribution of AMR in bacterial isolates demonstrated an association between the 312 isolation of an AMR bacteria and the number of treatments used (Fig. 3 and Table 3). The data 313 indicate that administration of two or more antimicrobial agents to treat BRD in cattle may 314 increase the likelihood of isolating an antimicrobial resistant pathogen (Fig. 3).   Table 5).    Table 7.  398 As reported in Table 7, the probability of an organism being resistant to at least one antibiotic 399 (ρ) was similar for the different treatment combinations. Specifically, in 62%, 81%, and 35% of 400 cases, the probability of an organism being resistant to at least one antibiotic was higher if 401 animals received a bacteriostatic antimicrobial for first treatment followed by a bactericidal 402 antimicrobial for retreatment of BRD when compared to bacteriostatic-bacteriostatic, 403 bactericidal-bacteriostatic, and bactericidal-bactericidal treatment, respectively.

404
With respect to the treatment, posterior gamma (γ) distributions shifted to the right in 405 animals that received a first line, bacteriostatic antimicrobials followed by retreatment with a 406 bactericidal antimicrobial (Error! Reference source not found.. 9). This suggests that BRD 407 pathogens isolated from these animals would be more likely to test resistant to more than one 408 antimicrobial (Table 5). The probability of obtaining a resistant isolate from an animal receiving 409 first-line bacteriostatic treatment followed by retreatment with a bactericidal antimicrobial being 410 higher than the other sequences was >95% (Error! Reference source not found. 8).
411 Table 8 431 As such, this report provides insights into potential critical control points for antimicrobial 432 stewardship in livestock production systems. For example, the heat map (Fig. 2)  Our study suggests that sequential treatment with different classes of 515 antimicrobials is a risk factor for developing drug resistance. Therefore, a review of 516 antimicrobial pre-exposure should be taken before the initiation of subsequent antimicrobial 517 therapy to prevent the emergence of antimicrobial resistance in cattle infected with BRD.

518
As concern about the impact of AMR microbes on animal and public health increase, 519 additional knowledge from studies such as the current one are needed to investigate interventions 520 that reduce the development of antimicrobial resistance. Furthermore, a microbiological 521 diagnosis should be established before using broad-spectrum antimicrobials to treat BRD of

532
These exploratory data suggest that treatment protocols stipulating first-line treatment 533 with a bacteriostatic followed by second-line treatment with a bactericidal may increase the 534 probability that drug resistance develops. As concern about antimicrobial resistance increases 535 from an animal and public health perspective, this knowledge suggests potential ways to reduce 536 the development of resistance. The hypothesis that the impact of an antimicrobial on bacterial 537 growth may be associated with the risk of increased resistance needs to be tested in a clinical 538 trial. Such a trial would also need to determine whether treatment efficacy is affected by a 539 change in treatment protocol or post-treatment interval. If treatment effectiveness proves to be 540 the same, then we may have an avenue by which to reduce the induction of resistance via the 541 recommendation that veterinarians tailor their treatment regimens to reduce the potential for 542 AMR development.

544
Although this study is hypothesis-generating, it has several strengths. The data set is 545 reasonably large for the questions we asked. Although a great deal of data were missing, we 546 limited our analysis to specific questions to avoid impact due to this missing data. Furthermore, 547 we recognized the limits of the passively collected and hypothesis-generating nature of the data