With the cooperation of the field veterinarians from The Japanese Association of Swine Veterinarians,¹ consent for participation in this observational study was obtained from ten farrow-to-finish pig farms on a convenient basis. Between November 2017 and July 2020, each farm was visited twice in the warm (spring and summer) and cold (autumn and winter) seasons, respectively, except for farm E (visited only once in the cold season). Brief descriptions of these farms are presented in Table 1 with the varied farm size of 70–1,790, based on sow number. At each visit, aerosol samples were collected from five pig houses of different life stages, including sow stall and farrowing houses as the sow piggery group and the weaners, growers, and finishers houses as the fattening piggery group. Using a commercial air sampler (CORIOLIS MICRO, Bertin Technologies SAS, France) placed at the center of each piggery, 3,000 L of air was passed into 10 mL of sterilized phosphate-buffered saline (PBS, Dulbecco’s PBS (−) “Nissui” Nissui Pharmaceutical Co., Ltd., Tokyo, Japan) for 10 min. These PBS samples were brought to the National Institute of Animal Health for laboratory investigations.

In total, 19 sampling visits were made in ten cold and nine warm seasons. Samples were obtained from both piggery groups during all nine warm season visits, meaning nine sow and nine fattening piggeries. However, during the ten cold season visits, due to the technical condition, samples were collected from only the fattening piggery group of Farm D, which meant nine sow and ten fattening piggeries were targeted. This sampling frame is summarized in Figure 1.

Within 20 h after the on-farm sampling, 100 μL of the PBS sample obtained from each piggery was inoculated on 5% sheep blood in trypticase soy agar (TSA) (BD Trypticase Soy Agar with 5% Sheep Blood, Nippon Becton, Dickinson, and Company, Japan) and mannitol salt agar (MSA) (Mannitol Salt Agar “Nissui,” Nissui Pharmaceutical Co., Ltd., Tokyo, Japan) and aerobically cultured for 16–20 h at 37°C. MSA was used as gram-positive bacteria selective agar, especially, salt-tolerant bacteria, which included some members of the Staphylococcus genus. Then, 10 isolates were randomly selected from each medium and stored in 10% glycerol-added Muller Hinton broth (Difco; BD, New Jersey, United States) at −80°C until identification.

All the isolates were identified using species-specific PCR for staphylococci, assumed as the dominant genus by the authors, following previously established procedures (14, 15). For those not identified using this PCR, partial gap gene sequencing (16) or 16S rRNA partial sequencing using a commercial kit (Bacterial 16S rDNA PCR Kit, Takara Bio Inc., Shiga, Japan) was applied following the manufacturer’s instructions. In addition, sequence data were analyzed to determine the most likely species, referring to the EzBioCloud Database.²

The identified isolates’ distribution by genus was summarized. In particular, the Chi-square test statistically evaluated the proportion of staphylococci among all the bacterial isolates in each farm by the seasons and the piggery groups.

For all the isolated staphylococci, the minimum inhibitory concentration (MIC) values of the 11 antimicrobials below were determined using a commercial kit (Dry Plate “Eiken,” Eiken Chemical Co., Ltd., Japan). Antimicrobial phenotypes were interpreted based on the breakpoints provided by the CLSI guidelines: 0.5 μg/mL for oxacillin (OXA), 0.5 μg/mL for ampicillin (AMP), 8.0 μg/mL for cefazoline (CFZ), 16.0 μg/mL for kanamycin (KAN), 16.0 μg/mL for gentamycin (GEN), 8.0 μg/mL for erythromycin (ERY), 16.0 μg/mL for tetracycline (TET), 32.0 μg/mL for chloramphenicol (CHL), 32.0 μg/mL for vancomycin (VAN), 8.0 μg/mL for lincomycin (LCM), and 4.0 μg/mL for ciprofloxacin (CIP), respectively (17). In addition, the following quality control strains were also assessed: Staphylococcus aureus ATCC 29213, Enterococcus faecalis ATCC 29212, Escherichia coli ATCC 25923, and Pseudomonas aeruginosa ATCC 27853.

The resistance rate (%) for each antimicrobial was defined as the proportion of resistant staphylococci isolates among the staphylococci isolates analyzed using the antimicrobial susceptibility test and the distribution of each antimicrobial based on the sampling seasons [n = 19 for “cold”s, and 18 for “warm”s (Figure 1)] and piggery groups [n = 18 for “sow”s, and 19 for “fattening”s (Figure 1)] was compared using the Mann–Whitney U test.

Annual antimicrobial product purchases were recorded for the piggery groups in each farm. This study used the previous year’s volume as the reference AMU data for a farm visit between January and June. The current year’s data were adopted for visits between July and December. Then, the annual mean treatment days (head*day) were estimated as follows:

The defined daily dose (DDD) is the Japanese pig production-specific indicator established previously (18, 19), based on the original concept and definition by the World Health Organization (20). These annual mean treatment days were the annual DDD-adjusted usage of each commercial product and summed up by the antimicrobial classes, which were tetracyclines (TETs), amphenicols (APCs), penicillins (PENs), cephalosporins (CEPs), sulfonamides (SULs), pyrimidines (PMDs), macrolides (MCLs), lincosamides (LCMs), aminoglycosides (AGDs), quinolones (QUIs), polymixins (PMXs), and pleuromutilins (PLMs), respectively. The class-based annual DDD-adjusted AMU was statistically compared by the seasons [n = 19 of “cold”s, and 18 of “warm”s (Figure 1)] and piggery groups [n = 18 of “sow”s, and 19 of “fattening”s Figure 1)], respectively, using the Mann–Whitney U test.

Association between the resistance rate of staphylococci for each antimicrobial and class-based annual DDD-adjusted AMU was explored using the generalized linear mixed model on each piggery group (n = 18 for the sow group and n = 19 for the fattening group). Considering the difference in the number of staphylococci successfully obtained on each sampling visit, raw data used for resistance rate calculation were incorporated into the model as the dependent variable; both tested and resistant staphylococci isolates were directly employed. Moreover, with the various farms cooperating, as presented in Table 1, sampling season was forced into the model, and the farm was employed as the random effect. Therefore, the model is described as follows:

Where p in logit (p) of the model outcome represents the resistance rate accounting for the tested and resistant staphylococci isolates, α is the model intercept, season is the dichotomous data of cold or warm season, χ is the fixed effect as the dichotomous data classified as “high” or “low” based on the median of the class-based annual DDD-adjusted AMU, β is its coefficient, RF is the farm as the random effect, and e is the binomially distributed residual term.

If the resistance rate to an antimicrobial revealed a positive and significant association with “high” class-based annual DDD-adjusted AMU of its class, a multivariable model for the associations with the AMU of other classes was also explored. The final model met the minimum Akaike’s Information Criterion (AIC), and statistical significance was set at p < 0.05 for the remaining independent variables with positive fixed effects.

The statistical modeling and other tests mentioned above were performed using R version 4.1.0.³ Primarily, the “glmmML” package version 1.1.3⁴ was used for the generalized linear mixed model.

Animal ethics approval was not required for this study as the samples consisted of piggery aerosol and were collected in the presence of the veterinarians during their routine farm visits for veterinary care and consultation.

In total, 915 bacterial isolates were obtained from TSA, and the genus-level description is summarized in Figure 2. The most dominant genus was Staphylococcus (n = 610, 66.7%), followed by Aerococcus (n = 85, 9.3%) and Rothia (n = 50, 5.5%). Finally, 1,113 staphylococci isolated from TSA (n = 610) and MSA (n = 503) underwent the antimicrobial susceptibility test, respectively.

The proportion of staphylococci exceeded 50% in most sampling visits. Farms A and H had over 70%. Apart from Farm E, which was visited once, no intra-farm significant seasonal difference was observed in the staphylococci proportion (Figure 3A, Chi-square test: p > 0.05). Sow and fattening piggeries had over 40% of staphylococci among the isolates. Farm A’s sow piggery group and both piggery groups of Farm H had over 80%. Only Farm C and I had significantly higher proportions of staphylococci in the fattening piggery group than in the sow piggery group (Figure 3B, Chi-square test: p = 0.02 and 0.03 for Farm C and Farm I, respectively).

Among the 1,113 staphylococci isolates, the most dominant specie was S. sciuri (which was renamed Mammaliicoccus sciuri in 2020) (n = 265, 23.8%), and others had <10% each (Table 2). The top five species S. sciuri, S. cohnii, S. saprophyticus, S. haemolyticus, and S. chromogenes dominated over 40% of each farm and some over 80% (data not shown).

For 10 of the 11 tested antimicrobials (all the isolates were susceptible to VAN), datasets of the resistance rate of staphylococci by the seasons (Figure 4A and Supplementary file 1) and piggery groups (Figure 4B and Supplementary file 2) were obtained. Regardless of the seasons and piggery groups, resistance rates for OXA, ERY, and LCM were > 40% of the median and tended to be higher than those of other antimicrobials.

A significant seasonal difference was only identified in the resistance rate for OXA, with the median for the cold and warm seasons being 65.4 and 80.7%, respectively (Figure 4A and Supplementary file 1, p = 0.03 as revealed by the Mann–Whitney U test). In contrast, a significant between-piggery group difference was identified in OXA, AMP, ERY, and CHL. The resistance rates for these four antimicrobials in the fattening piggery group were significantly higher than those in the sow piggery group, with a median of 78.8 and 58.0% for OXA, 57.1 and 31.8% for AMP, 82.4 and 48.4% for ERY, and 45.5 and 21.1% for CHL for the fattening and sow piggery groups, respectively (Figure 4B and Supplementary file 2, all p < 0.05 as revealed by the Mann–Whitney U test).

Figure 5 illustrates the distribution of the class-based annual DDD-adjusted AMU. The AMU varied by farm; however, no intra-class difference was identified by season (Figure 5A and Supplementary file 3, all p > 0.45 as revealed by the Mann–Whitney U test). In contrast, a between-piggery group difference was identified in the AMU of all the classes, except PMXs and PLMs. Therefore, the fattening piggery group had a significantly higher AMU than the sow piggery group, with zero medians for all classes, except TETs and MCLs (Figure 5B and Supplementary file 4, all p < 0.05 as revealed by the Mann–Whitney U test).

Table 3 presents four final models obtained by statistical modeling from the datasets of the fattening piggery group for the association between the class-based annual DDD-adjusted AMU and resistance rate of staphylococci. Out of the 11 evaluated antimicrobials, the resistance rate for OXA, ERY, TET, and CHL was significantly associated with the AMU of the corresponding PENs, MCLs, TETs, and APCs, respectively.

Regarding OXA, the final model included only PENs, and its “high” usage was associated with a higher resistance rate for OXA [odds ratio (OR) and 95% confidence interval (CI)] = 2.36 (1.11, 5.05), p = 0.03). For ERY, the final model included MCLs, APCs, and LCMs. A “high” usage of these three antimicrobial classes was independently associated with a higher resistance rate for ERY (OR (95% CI) = 6.89 (2.53, 18.73), p = 0.0001 for MCLs, OR (95% CI) = 4.81 (1.17, 19.69), p = 0.03 for APCs, and OR (95% CI) = 20.91 (3.60, 121.51), p = 0.001 for LCMs, respectively). For TET, the final model included only TETs, and its “high” usage was associated with a higher resistance rate for TET (OR (95% CI) = 2.48 (1.03, 5.99), p = 0.04). In addition, for CHL, the final model included APCs and TETs. A “high” usage of these antimicrobial classes was independently associated with a higher resistance rate for CHL (OR (95% CI) = 3.22 (1.14, 9.12), p = 0.03 for APCs, and (OR (95% CI) = 3.39 (1.73, 6.62), p = 0.0004 for TETs, respectively). No significant interaction terms were identified in all the final models.

Conversely, analyses of the sow piggery group did not reveal any significantly positive association between the AMU and resistance rate. However, the resistance rates for KAN and TET had a marginally positive association with “high” AGDs and TETs use, respectively (p = 0.09 and 0.11, respectively, data not shown).