Concordance of disk diffusion, broth microdilution, and whole‐genome sequencing for determination of in vitro antimicrobial susceptibility of Mannheimia haemolytica

Abstract Background Extensive drug resistance (XDR) is an emerging concern with Mannheimia haemolytica, and a variety of testing methods are available for characterizing in vitro antimicrobial susceptibility. Objectives To compare the concordance among disk diffusion, broth microdilution, and whole genome sequencing (WGS) for susceptibility testing of M. haemolytica before and after mass treatment using tulathromycin. Animals Forty‐eight M. haemolytica isolates collected from high‐risk beef stocker calves before and after mass treatment (metaphylaxis) using tulathromycin (Draxxin, Zoetis, Parsippany, NJ) given at the label dosage of 2.5 mg/kg body weight SC in the neck. Methods In vitro antimicrobial susceptibility was determined for all 48 isolates using disk diffusion, broth microdilution, and WGS. Concordance was calculated between pairs of susceptibility testing methods as follows: number of isolates classified identically by the 2 testing methods for each timepoint, divided by the number of isolates tested at that timepoint. Discordance was calculated as follows: number of isolates classified differently by the 2 testing methods for each timepoint, divided by the number of isolates tested at that timepoint. Results Concordance between testing methods ranged from 42.3% to 100%, depending on antimicrobial evaluated, timing of sample collection, and testing method used. Very major errors were identified in up to 7.7% of classifications whereas minor errors were seen in up to 50% of classifications depending on antimicrobial evaluated, timing of sample collection, and testing method used. Conclusions and Clinical Importance Our results show that discrepancies in the results of different susceptibility testing methods occur and suggest a need for greater harmonization of susceptibility testing methods.


| INTRODUCTION
Choosing the right antimicrobial can mean the difference between success and failure, health and disease, and even life and death in cattle with bovine respiratory disease (BRD). However, in the recent past, the bacterial pathogen most often associated with BRD in beef cattle, Mannheimia haemolytica (Mh), has begun to develop resistance to an increasing number of antimicrobials. 1,2 Currently, it not uncommon to see strains of this microbe resistant to ≥3 classes of antimicrobials, which has hindered implementation of effective treatment protocols. [2][3][4] Veterinarians have long relied on the results of antimicrobial susceptibility testing (AST) to facilitate decision making for the design of effective treatment regimens. Disk diffusion, concentration gradient agar diffusion (E-test), and broth macro or microdilution are methods that have commonly been used for assessing phenotypic susceptibility, with surveys showing that disk diffusion is the methodology employed by most laboratories. [5][6][7] Whole genome sequencing (WGS) has been used to evaluate mechanisms and epidemiology of antimicrobial resistance at the molecular level and has potential for use in evaluating susceptibility at the genotypic level to complement the currently available phenotypic tests. [8][9][10] In addition, WGS can provide information about the epidemiology of pathogen spread and provide a better understanding of genetic diversity and distribution of specific traits within a bacterial population. 11 Although WGS has encouraging potential, it is currently not known how genotypic resistance agrees with phenotypic resistance in Mh. Also, it is not fully understood how the more commonly used AST methods compare for the testing of field isolates with different susceptibilities and genetic backgrounds.
In studies from the human medical literature, WGS technologies have performed very well relative to standard testing methods. One study analyzed 1379 isolates of Staphylococcus aureus obtained from British hospitals and found that the 3 methods of genotypic susceptibility prediction assessed agreed with culture-based phenotypic assessment in 98.3% of cases. 12 Another study found 99% agreement between genotypic and phenotypic susceptibility in 640 Salmonella spp. isolates collected from retail meat samples and human clinical cases. 13 Numerous other examples of the high concordance among different methodologies exist and suggest that WGS technologies have utility in the assessment of antimicrobial susceptibility. [14][15][16][17] In the veterinary medical literature, work evaluating the utility of WGS for antimicrobial susceptibility testing of Mh is limited. 18 Therefore, our goals were 2-fold. First, we sought to assess concordance among disk diffusion, WGS, and broth microdilution for determining the in vitro antimicrobial susceptibility of Mh obtained from high-risk beef stocker calves before and after mass medication using tulathromycin. Second, because studies have shown that the susceptibility of Mh can change markedly after mass medication, we sought to evaluate how different susceptibility testing methodologies performed in these genetically distinct populations. 3,9 2 | MATERIALS AND METHODS

| Isolate selection
The isolates used in this study were collected as part of a previous investigation into the prevalence of multidrug resistant Mh in high risk stocker cattle. 3 Briefly, the isolates were cultured from deep nasopharyngeal swabs (NPS) collected from calves (n = 20) that were Mh culture-positive at both arrival and 10 to 14 days later at revaccination, and a maximum of 3 isolates could be selected from a given time point if they all had unique antimicrobial susceptibility profiles. This approach yielded 48 isolates, 26 isolates from the calves at arrival, and 22 at revaccination. Any calves displaying signs of BRD at the time of arrival processing were excluded from the study. No calves with isolates included here were diagnosed with BRD during the course of the study.

| Disk diffusion
The disk diffusion susceptibility data utilized in this study was obtained from susceptibility data performed as part of a previous work. 3  (Illumina, San Diego, California). The genomic data were submitted to a consultant for cleaning and assembly. The program FastQC version 0.11.5 was used to assess initial quality, followed by Trimmomatic version 0.36 to trim the reads, with all reads having < 50 base pairs removed; FastQC was run a second time to assess trimmed read quality. 20,21 Coverage for isolates ranged from 144 to 369X, with an average coverage of 253X.
De novo assembly, using paired end reads, was performed using SPAdes version 3.11.1, and quality of the assemblies assessed using QUAST version 4.5. 22,23 Assemblies were individually queried using BLASTn against the National Center for Biotechnology Information (NCBI) database to identify Mh isolates with which they shared the greatest homology, and then aligned to these reference sequences using MAUVE 2.4.0 to place the scaffolds into a more biologically relevant orientation before annotation. 24 Broth dilution interpretive criteria (μg/mL)

| RESULTS
A summary of the disk diffusion susceptibilities of the 48 selected isolates is presented in Table 2. Isolates collected from calves at arrival processing were routinely susceptible to florfenicol, ceftiofur, enrofloxacin, and gamithromycin. In contrast, isolates collected from the same calves at revaccination 10 to 14 days later were consistently resistant to enrofloxacin, tilmicosin, gamithromycin, and tulathromycin, whereas they retained susceptibility to ceftiofur and susceptibility to florfenicol varied.
Minimum inhibitory concentration distributions of the selected isolates as determined by broth microdilution are presented in Table 3. As with disk diffusion, isolates collected from calves at the time of arrival processing were routinely susceptible to most of the antimicrobials evaluated. In addition, isolates collected from calves at revaccination were routinely resistant to most antimicrobials evaluated. The major exceptions were consistent susceptibility to ceftiofur and variable susceptibility to florfenicol. The genes and mutations identified in the isolates by whole WGS BLAST queries, the antimicrobial class for which they encode resistance, and the proportion of isolates positive for each gene at the 2 sampling time points are presented in Table 4. Genes and mutations identified include the phenicol resistance gene floR, β-lactamases bla-ROB1 and bla-OXA2, tetracycline resistance genes tetH/R, macrolide resistance genes erm42, msrE and mphE, as well as fluoroquinolone resistance mutations gyrA S83F, gyrA D87N, and parC E84K.

| Disk diffusion vs broth microdilution
Concordance between disc diffusion and broth microdilution is presented in

| Disk diffusion vs WGS
Concordance between disc diffusion and WGS is presented in

| DISCUSSION
The availability of reliable and reproducible AST methods is essential strains of Mh has complicated disease control and treatment strategies. 2,3,29 Broth microdilution has been considered the reference method for AST because it provides a numerical assessment of in vitro antimicrobial activity, and summary statistics can provide quantitative data that can be used in development of treatment protocols and also in surveillance or epidemiologic monitoring programs. 10 We chose disk diffusion and WGS for comparison to broth microdilution for several reasons. First, disk diffusion is commonly employed by diagnostic laboratories, and surveys have shown that it is the AST method used most often in these settings. [5][6][7] Second, WGS has the potential to provide additional data beyond antimicrobial susceptibility, and can enhance our understanding of microbial pathogenesis and pathogen transmission. 8 As a result, more data are needed regarding the relative concordance between these methods so that interpretation can be made appropriately in clinical scenarios.
In our study, concordance among the different AST methods varied substantially in some cases. In addition, the variation in concordance among methods was dependent upon the antimicrobial evaluated and sampling time. More specifically, concordance was lower for tilmicosin, tulathromycin, and florfenicol among the different testing methods than was seen with other antimicrobial classes, and much of the discordance seen was specific to a given sampling time point. For example, concordance for tilmicosin and tulathromycin was lower in arrival isolates across all testing methods, whereas concordance for florfenicol was lower in revaccination isolates. In addition, concordance for penicillin between WGS and broth microdilution was only 42% for arrival isolates but 86% for isolates collected at revaccination. Although most of the discordant results were minor errors, this difference in interpretation still could alter a clinician's choice of antimicrobial and potentially impact treatment outcome adversely. Consequently, our results suggest that bacterial strain and carriage of specific resistance determinants, along with history of previous antimicrobial exposure, might impact the utility and interpretation of AST results for certain classes of antimicrobials.
Generally speaking, concordance between disk diffusion and broth microdilution was not as high as expected, and for some testing methods, fell below the minimum level of concordance considered acceptable by the United States Food and Drug Administration (FDA) for the number of isolates tested. 30 Although only 2 isolates were classified as discordant with very major errors, numerous minor errors occurred in which an isolate was classified as intermediate by 1 of the methods evaluated. Similar work comparing agreement between disk diffusion and agar dilution for classifying in vitro susceptibility of Mh found that agreement between methods for florfenicol, ceftiofur, and enrofloxacin susceptibility was 100%, whereas agreement for tetracycline was only 85.3%. 31 Although not the same methodology, agar  oxytetracycline, and tulathromycin concordance was 100%. 18 Interestingly, concordance for ceftiofur was found to be only 31%. 18 An intermediate category of susceptibility, however, was not included and specific types of errors were not determined in that study. 18 Nevertheless, this study similarly illustrates the difficulties in applying the complexities of genomic resistance to clinical scenarios and shows that the presence or absence of a specific gene is not always sufficient to predict phenotypic antimicrobial susceptibility. Moreover, the discordance seen in our study is higher than reported for some pathogens affecting humans and suggests that the concordance between WGS and other AST methods might be species-specific. [12][13][14][15][16][17] Moving forward, several assumptions must be made if WGS is to be relied on to accurately predict a given resistance phenotype. Most importantly, we must assume that no epistasis is present. 11 In other words, whether a single gene determines the resistance phenotype or if a resistance phenotype is controlled by many genes, these genes or separate gene loci must not interact in a complex or antagonistic manner. A good example of fulfillment of this assumption would be the genes controlling macrolide and tetracycline susceptibility. 11 For these drug classes, known resistance genes all work together to confer phenotypic resistance and do not work against each other in an overly complicated manner. Other assumptions that must be considered are that resistance genes are highly expressed, the origin of a given strain has no impact on gene expression, and that all of the genes responsible for resistance in a microbe of interest are known. 11 Regardless of the method evaluated, the number of errors, particularly very major errors, detected in our study is concerning and falls outside of the target accuracy established by the FDA for the number of resistant isolates tested. 30

CONFLICT OF INTEREST DECLARATION
Authors declare no conflict of interest.

OFF-LABEL ANTIMICROBIAL DECLARATION
Authors declare no off-label use of antimicrobials.

INSTITUTIONAL ANIMAL CARE AND USE COMMITTEE (IACUC) OR OTHER APPROVAL DECLARATION
Authors declare no IACUC or other approval was needed.

HUMAN ETHICS APPROVAL DECLARATION
Authors declare human ethics approval was not needed for this study.