Antimicrobial Susceptibility of Mycoplasma bovis Isolates from Veal, Dairy and Beef Herds

Mycoplasma bovis is an important pathogen causing mostly pneumonia in calves and mastitis in dairy cattle. In the absence of an effective vaccine, antimicrobial therapy remains the main control measure. Antimicrobial use in veal calves is substantially higher than in conventional herds, but whether veal calves also harbor more resistant M. bovis strains is currently unknown. Therefore, we compared antimicrobial susceptibility test results of M. bovis isolates from different cattle sectors and genomic clusters. The minimum inhibitory concentration of nine antimicrobials was determined for 141 Belgian M. bovis isolates (29 dairy, 69 beef, 12 mixed, 31 veal farms), and was used to estimate the epidemiological cut-off. Acquired resistance was frequently observed for the macrolides, while no acquired resistance to oxytetracycline and doxycycline, minimal acquired resistance to florfenicol and tiamulin, and a limited acquired resistance to enrofloxacin was seen. M. bovis isolates from beef cattle or genomic cluster III had higher odds of being gamithromycin-resistant than those from dairy cattle or genomic clusters IV and V. In this study, no cattle industry could be identified as source of resistant M. bovis strains. A single guideline for antimicrobial use for M. bovis infections, with a small remark for gamithromycin, is likely sufficient.


Introduction
In the last decade, Mycoplasma bovis (M. bovis) has come to the forefront as an economically important bacterium with a large impact on health, welfare and antimicrobial use (AMU) in cattle operations worldwide [1]. The bacterium is mainly feared as the cause of pneumonia, arthritis and otitis in calves, and pneumonia and mastitis in adult cattle [2,3]. In the absence of an effective vaccine, antimicrobial therapy remains a crucial factor to control an outbreak.

Interpretation of MIC Values
The epidemiological cut-off values (ECOFF), as determined using the different methods, and the percentages of isolates belonging to the wild type and non-wild type population, are shown in Table 2. For all three ECOFF methods, acquired resistance was frequently observed for macrolides (gamithromycin, tylosin), while only a few isolates showed acquired resistance against florfenicol, enrofloxacin and tiamulin. Following the visual estimation method, no acquired resistance against oxytetracycline and doxycycline was observed, although the statistical methods categorized part of the population (3.6-13.0%) as non-wild type for doxycycline.  We were able to determine the ECOFF for seven out of the eight antimicrobials with the visual estimation method, whereas normalized resistance interpretation (NRI) and ECOFFinder determined a reliable ECOFF in eight out of eight and three out of eight of the antimicrobials, respectively. The NRI method was able to determine an ECOFF for every antimicrobial tested, even when this needed extrapolation from the tested MIC range, such as for tilmicosin. However, when the standard deviation of the normal distribution of the wild type MIC value is >1.2 log 2 , the program provides only the "putative wild type population", as was the case for the macrolides. The ECOFFinder method was only able to determine reliable results (good fit plots for residuals) for florfenicol and tetracyclines. Truncated distributions influence the reliability or possibility of interpreting some of the results, such as those for the macrolides. All three methods determined the ECOFF values for the corresponding antimicrobials within one dilution, except for doxycycline and tiamulin. This results in substantial differences in the non-wild type population. When ECOFFinder 95% was used for doxycycline, 13% was categorized as non-wild type in comparison to 0% when using the visual estimation method. For tiamulin, the visual method indicated 0.7% non-wild type, whereas NRI indicated 15.1%. This might be due to the combination of "tailing" and the lack of a normal distribution, which complicates the interpretation of the MIC distributions, both visually and statistically.

Variability of Antimicrobial Susceptiblity Per Production System
The distribution of M. bovis' MIC values for the different antimicrobials and per production system are available in Supplementary File 2. Logistic regression did not show significant differences in antimicrobial resistance between production systems, except for gamithromycin (Table 3). Beef M. bovis isolates (58.21% acquired resistance) had three-times higher odds (CI 95%: 1.23-7.69) of gamithromycin resistance than dairy isolates (31.03%; p = 0.02).
The MIC 50 and MIC 90 values are shown per sector in Table 4. No differences in MIC 50 were observed between sectors for tilmicosin, doxycycline, and tiamulin. A single two-fold dilution difference in MIC 50 between herd types was observed for florfenicol (highest in veal), oxytetracycline (lowest in dairy) and enrofloxacin (lowest in beef). A difference between herds was seen for gamithromycin and to a lesser extend for tylosin. The MIC 50 of gamithromycin was higher in beef cattle (>128 µg/mL) than in dairy (16 µg/mL) or veal calf isolates (32 µg/mL). No difference was observed in MIC 90 for florfenicol, tilmicosin, and gamithromycin, while a single two-fold dilution was observed in MIC 90 for oxytetracycline (lowest in dairy), doxycycline (highest in beef), tylosin (lowest in dairy), tiamulin (lowest in veal) and enrofloxacin (highest in dairy).

Association between AMR and Genomic Cluster
In Figure 1, the distributions of the wild type and non-wild type M. bovis isolates for gamithromycin, tylosin, and enrofloxacin are shown over the five clusters obtained by whole genome sequencing. Most of the macrolide-resistant isolates were located in clusters II and III, whereas cluster I contained isolates susceptible to all antimicrobials. Statistical analysis showed that M. bovis isolates from cluster III (85% acquired resistance) had 22.7-(CI95%: 4.0-125.0, p < 0.01) and 7.9 (CI95%: 1.5-40.0, p = 0.01) -fold higher odds for gamithromycin resistance compared to clusters IV (19%) and V (41%), respectively. No significant association was found for tylosin or enrofloxacin.

Discussion
In this study, the susceptibility of 141 M. bovis isolates obtained from beef, dairy and veal calves was tested against nine antimicrobial agents covering the 6 antimicrobial classes most frequently used to control M. bovis in Belgium, though the results of gentamicin (covering the aminoglycosides) were excluded. Since neither clinical breakpoints (CBPs) nor guidelines to interpret ECOFFs for M. bovis are available, three methods (visual, NRI and ECOFFinder) to determine ECOFF and interpret the antimicrobial susceptibility of M. bovis were explored.
Although one should be cautious in comparing the results of different studies, because of the lack of internationally recognized standard protocols, the observed MIC values and distributions in this study were similar to previous publications from Western Europe [6,10,23,24]. A large number of isolates had high MIC values for macrolides, resulting in more than half of the isolates being nonwild type for gamithromycin or tylosin. Cross-resistance between tylosin and tilmicosin has been described for M. bovis [25,26], but determination of ECOFF based on the MIC values for tilmicosin was not possible, as no normal distribution was obtained due to the very high MIC values.
Surprisingly, there were no indications of acquired resistance against tetracyclines. The current study showed the decreased MIC50-90 values (1-4 µg/mL) of M. bovis for oxytetracycline compared with a 20-year-old Belgian study reporting MIC50-90 values of 2 and 32 µg/mL, respectively [23]. Additionally, in other European countries, a similar trend has been observed in recent years [6,10]. For doxycycline, the percentages of acquired resistance depended on the ECOFF method used (0% to 13%). When using the 95% rule with ECOFFinder, 13% acquired resistance was obtained, while when using the 99% rule only 3.6% resistant isolates were observed. One should be aware that the decision of using 95% or 99% can influence the outcome by increasing either the sensitivity for non-wild type (95%) or the specificity for the wild type population (99%). The distribution and MIC50-90 for

Discussion
In this study, the susceptibility of 141 M. bovis isolates obtained from beef, dairy and veal calves was tested against nine antimicrobial agents covering the 6 antimicrobial classes most frequently used to control M. bovis in Belgium, though the results of gentamicin (covering the aminoglycosides) were excluded. Since neither clinical breakpoints (CBPs) nor guidelines to interpret ECOFFs for M. bovis are available, three methods (visual, NRI and ECOFFinder) to determine ECOFF and interpret the antimicrobial susceptibility of M. bovis were explored.
Although one should be cautious in comparing the results of different studies, because of the lack of internationally recognized standard protocols, the observed MIC values and distributions in this study were similar to previous publications from Western Europe [6,10,23,24]. A large number of isolates had high MIC values for macrolides, resulting in more than half of the isolates being non-wild type for gamithromycin or tylosin. Cross-resistance between tylosin and tilmicosin has been described for M. bovis [25,26], but determination of ECOFF based on the MIC values for tilmicosin was not possible, as no normal distribution was obtained due to the very high MIC values.
Surprisingly, there were no indications of acquired resistance against tetracyclines. The current study showed the decreased MIC 50-90 values (1-4 µg/mL) of M. bovis for oxytetracycline compared with a 20-year-old Belgian study reporting MIC 50-90 values of 2 and 32 µg/mL, respectively [23]. Additionally, in other European countries, a similar trend has been observed in recent years [6,10]. For doxycycline, the percentages of acquired resistance depended on the ECOFF method used (0% to 13%). When using the 95% rule with ECOFFinder, 13% acquired resistance was obtained, while when using the 99% rule only 3.6% resistant isolates were observed. One should be aware that the decision of using 95% or 99% can influence the outcome by increasing either the sensitivity for non-wild type (95%) or the specificity for the wild type population (99%). The distribution and MIC 50-90 for florfenicol were similar (2-8 µg/mL) to previously published data [4,6,10], and only four isolates showed acquired resistance (MIC 32 µg/mL). In general, a small non-wild type population was observed for tiamulin. In Europe, pleuromutilins are not registered for use in cattle, while this class of antimicrobials is registered for the treatment of Mycoplasma infections in pigs and poultry. In addition, valnemulin appears to be very effective against M. bovis in vivo [27], and low numbers of acquired resistance have been reported in M. bovis isolates in France, Spain, and Hungary as well [4,7,26]. One possible non-wild type isolate for tiamulin was identified in this study. In a previous study, all tiamulin-resistant mutant strains showed cross-resistance against florfenicol [26], which was not observed in the current study.
Gentamicin did not pass the QC with S. aureus (ATCC ® 29213 TM ) and E. coli (ATCC ® 25922 TM ), and also the MIC for M. bovis PG45 deviated from previous studies [12,22]; based on these observations the results were excluded. Previous studies did not include these quality control strains, and therefore we are not aware of whether this is a reoccurring problem [12,22,23]. It is likely that specific medium components, resulting in an adjusted pH, have altered the results of the quality control strains [28], but as there is no standard protocol or quality control standard for M. bovis PG45, more research is necessary. Notwithstanding this, we included the results in the supplementary data to contribute to this research (Supplementary File 1).
We did not observe significant sector-specific antimicrobial resistance, except for with gamithromycin. M. bovis isolates from beef cattle had higher odds of being non-wild type than those from dairy cattle. Dairy cattle isolates also had the lowest MIC 50 values for tylosin. This could possibly be explained by the non-registration of macrolides for use in lactating animals, and the high use of macrolides to combat bovine respiratory disease in beef cattle and veal calves. Additionally, other factors, such as age, housing conditions and milk diet, could play a part in the evolution of antimicrobial resistance in different production systems [29]. Considering the small difference in AMR over sectors, together with previously obtained knowledge of the lack of sector-specific M. bovis strains in Belgium [18], a single guideline for the antimicrobial use for M. bovis infections covering all different cattle sectors in Belgium, with a small remark for gamithromycin, is likely sufficient.
All used methods to determine the ECOFF are in some way based on a normal distribution. As a consequence, problems occurred with truncated MIC distributions (e.g., tilmicosin, tiamulin). Although the NRI method was able to determine more ECOFFs than the ECOFFinder (only florfenicol and tetracyclines) in an objective manner, the visual estimation method was mostly in agreement with these methods. Even though it was more subjective, the visual estimation method has the advantage that expert opinion and additional information from MIC data obtained from other class representatives or previous reports can be taken into account. For instance, even though a bimodal distribution was observed for tylosin, the population showing the lower MIC values might not represent the true wild type population. This is supported by a previous MIC study showing a much lower ECOFF (2 µg/mL), while similar QC values were obtained [30]. In addition, it has been previously shown that the specific mutations associated with macrolide resistance were absent in isolates with MIC values between <0.5 and 4 µg/mL [25]. Therefore, an overall shift from M. bovis wild type to non-wild type for tylosin is suspected. The same line of reasoning is applicable to tilmicosin. Lerner et al. (2014) did not find any mutations associated with macrolide resistance in isolates with tilmicosin MIC values between <0.5 and 32 µg/mL. Therefore, it seems that all the isolates in this study acquired resistance to tilmicosin to some extent, except for one isolate with an MIC of 1 µg/mL for tylosin and 8 µg/mL for tilmicosin, probably representing the wild type population for 16-ring macrolides. Indeed, a recent study showed very high MIC values (≥256 µg/mL) for >80% of the M. bovis population against tilmicosin [10], whereas an older study showed a large population with lower MIC values between 0.5 and 32 µg/mL [21].
Finally, we observed an association between gamithromycin susceptibility patterns and previously published genomic clusters. M. bovis isolates in clusters II and III were more frequently belonging to the non-wild type population than those in clusters IV and V. This might be due to the higher heterogeneity in clusters IV and V, caused by genetic drift [18]. Yet, we should be careful in our conclusions, as we are not aware of the influence of genetic drift within any of the clusters on antimicrobial susceptibility data. We feel on this point that even when strain typing can be done very fast and easily, this should always be supplemented with phenotypic antimicrobial susceptibility testing (AST) to detect acquired resistance in M. bovis outbreaks. Nevertheless, strain typing could support the surveillance of AMR by pointing out whether isolates are clonally spread or (closely) related to each other.

Mycoplasma Bovis Collection
One hundred forty-one epidemiologically independent M. bovis isolates, originating from 29 dairy, 69 beef, 12 mixed (both dairy and beef) and 31 veal farms, were included in this study. Isolates were obtained from the respiratory tract (128), middle ear (4), milk (5), joint (2), abscess (1) and seroma (1), collected in Belgium between 2016 and 2019, with the exception of one isolate which was obtained in 2014. One hundred of these isolates have been strain-typed previously [18]. All isolates were obtained from diagnostic samples collected by field veterinarians from clinical cases, in compliance with the EU legislation on ethics in animal experimentation [2010/63/EU]. All samples were cultured on a modified pleuropneumonia-like organism (PPLO) agar plate and incubated for 7-10 days (37 • C, 5% CO 2 ). Presumptive M. bovis identification was based on the typical fried-egg colony appearance on modified PPLO agar and the presence of lipase activity as tested on medium containing Tween-80 [31]. Final identification was performed with MALDI-TOF MS as described before [32]. All samples were stored at −80 • C until further analysis.
Plates were sealed with adhesive foil and incubated at 37 • C. The interpretation of color change from red to orange/yellow was done after 48 and 72 h of incubation. The lowest antimicrobial concentration without color change was recorded as the MIC at the earliest time point at which the growth control well had the same color as the pH control well. When for a certain isolate, more than one skipped well was observed, the results for these isolates were excluded. When only one well was skipped, the highest MIC value was listed [34]. Quality control was performed in every run (six in total) by determining the MIC values of the M. bovis strain PG45 (ATCC 25523) and comparing these to previously published values [7,12,20,22]. Reference strains Staphylococcus aureus ATCC ® 29213 TM and Escherichia coli ATCC ® 25922 TM were included as extra QC strains in the same broth as M. bovis, but were observed after 24 h of incubation.

Interpretation of MIC Values
Due to the lack of clinical breakpoints (CBPs), the interpretation of the MIC values of M. bovis is not straightforward [3,21,35]. The best option for interpreting the M. bovis MIC data is probably to determine ECOFF. With this method, wild type bacterial populations are distinguished from those with acquired resistance (non-wild type) by observing the MIC distribution. Three methods to estimate the ECOFF were compared in this study-the visual estimation ("eyeball") method based on the uni-, bior multimodal MIC distribution and/or tailing, as described previously [35,36], as well as two statistical methods, "Normalized Resistance Interpretation (NRI)" (http://www.bioscand.se/nri/, Bioscand AB, Täby, Sweden; [37]) and the "Iterative Statistical Method" processed in ECOFFinder (version 2.1; https://www.eucast.org/mic_distributions_and_ecoffs/, EUCAST) [38,39]. Instructions provided by the founders were followed. When using the NRI method, standard deviations of the normal distribution of wild type MIC values exceeding 1.2 log 2 result in a tentative estimate of the ECOFF, and one can only speak of the "putative wild type group". With ECOFFinder, plots for residuals were checked and categorized, corresponding to whether the residuals scattered on either side of the horizontal axis in the center (well fit, selected subset values are considered reliable), only partly (poor fit) or not at all (no fit). As users can choose the cut-off value (95% to 99.9%) with ECOFFinder, depending on the intended use and influencing the sensitivity and specificity of the (non-)wild type population, both the 95% and 99% cut-offs were determined. In addition, the MIC 50 and MIC 90 were calculated as the lowest MIC at which at least 50% and 90% of the isolates in a test population are inhibited in their growth, respectively. Since no ECOFF within the testing range of our study could be obtained by any of the three methods for tilmicosin, previously published data and cross-resistance with tylosin were taken into account to make the decision that the isolates with MIC ≥32 µg/mL belonged to the non-wild type population (see discussion section). The latter tilmicosin ECOFF, together with the ECOFFs obtained with the visual estimation method, were used in further analysis to compare the AMR in M. bovis isolates obtained from different cattle sectors or belonging to specific genomic clusters.

Statistical Analysis
To determine whether there are significant differences between conventional herds (dairy, beef) and veal calves, a logistic regression was performed on binary variables, representing acquired resistance (1) and wild type (0) isolates. A p-value smaller than 0.05 was considered statistically significant. The Hosmer-Lemeshow test was included to determine the goodness of fit of the model (SPSS Statistics 26). To allow a meaningful statistical analysis, only for the antimicrobials for which 5 to 95% acquired resistance was observed was statistical analysis done.
In a former study, a subset of 100 isolates of the currently used M. bovis database had been strain-typed, as described earlier [18]. These isolates have been categorized into 5 phylogenetic clusters, based on single nucleotide polymorphism (SNP) analysis [18]. In the present study, the association between the presence of acquired resistance according to the visual estimation method and the phylogenetic clusters of this subset of strains was investigated for antimicrobials for which 5 to 95% acquired resistance was observed, and further visualized in MEGA-X [40]. Logistic regression on binary variables (1: acquired resistance; 0: wild type) was only performed for clusters III to V, as clusters I and II did not contain enough M. bovis isolates for the model to run.

Conclusions
The high acquired resistance percentages of M. bovis in Belgium were observed for macrolides, with almost all isolates having acquired a resistance to 16-ring macrolides and a large proportion to 15-ring macrolides. In addition, a minimal acquired resistance to florfenicol and tiamulin was observed, a limited acquired resistance to enrofloxacin, and almost no resistance to the tetracyclines. A higher AMR for gamithromycin was observed in beef cattle compared to dairy, but the veal industry could not be identified as a reservoir of resistant M. bovis strains. A single guideline for the antimicrobial use of M. bovis infections covering all different cattle sectors in Belgium, with a small remark for gamithromycin, is likely sufficient. In addition, only M. bovis strains belonging to clusters II and III had more isolates with acquired resistance for gamithromycin compared to IV and V. Therefore, this study shows that strain-typing cannot replace the phenotypic AST of M. bovis in surveillance programs.