Bacterial isolates were obtained from clinical material from mink submitted to diagnostic laboratories (The National Veterinary Institute DTU, Lyngby, Denmark; Institute for Experimental Pathology, Reykjavík, University of Iceland; veterinary clinic Pecon BV, Gemert, the Netherlands; INVESAGA Group, Department of Animal Pathology, University of Santiago de Compostela, Lugo, Spain; Finnish Food Authority, Seinäjoki, Finland) in the period 2006–2018. Each submission to the laboratory could consist of more than one animal. A maximum of one isolate of each of the four bacterial species was collected from each submission. A farm could be represented more than once if samples were submitted to the laboratory repeatedly for investigation. There was no limitation as to how many times each farm could be represented over the 12-year sampling period. Also, the antimicrobial treatment history for the farms was not a criterion for the inclusion of bacterial isolates. The mink industry follows the same seasonal pattern all over the world, and the animals have been submitted from the beginning of whelping (April) until pelting (November). The following species were included in the study: E. coli (n = 162), S. delphini (n = 63), S. canis (n = 42), and P. aeruginosa (n = 55). Isolates originated from Denmark, Finland, Iceland, Lithuania, the Netherlands, and Spain (Table 1). All isolates included in this study were identified by MALDI-TOF as described in Nikolaisen et al. (8).

All isolates were investigated using the broth microdilution semiautomated technique Sensititre (ThermoFisher Scientific, UK) according to methods described by the Clinical and Laboratory Standards Institute (9). For E. coli, S. delphini, and P. aeruginosa, cation-adjusted Mueller-Hinton broth (CAMHB) was used, and panels were incubated at 35 ±2°C for 16–20 h, whereas for S. canis CAMHB with lyzed horse blood was used and panels were incubated at 35 ± 2°C for 20–24 h (9, 10). Based on data from the national veterinary prescription database VetStat (11, 12), some of the most frequently used antimicrobials in mink production in Denmark were chosen for designing a custom-made panel. This panel contained 2-fold dilutions of amoxicillin (range 0.25–512 μg/mL), colistin (0.06–128 μg/mL), spectinomycin (0.25–512 μg/mL), sulfamethoxazole-trimethoprim 19:1 (0.03–64 μg/mL), doxycycline (0.06–128 μg/mL), lincomycin (0.06–128 μg/mL), sulfamethoxazole (0.5–512 μg/mL), and tylosin (0.12–128 μg/mL). Antimicrobial concentration ranges were based on MIC distributions in the EUCAST MIC database (13) and earlier reports on prevalence of antimicrobial resistance in bacterial pathogens from mink (1, 8). A subset of isolates was further tested for susceptibility to trimethoprim (E. coli: n = 53, S. delphini: n = 38, S. canis: n = 26) and penicillin (S. delphini: n = 18) (Supplementary Tables 1, 2) by broth microdilution (9). Trimethoprim test was performed to determine the added effect of the combinational drug sulfamethoxazole and trimethoprim. Susceptibility to penicillin was tested in isolates harboring the blaZ gene. Escherichia coli ATCC 25922, Staphylococcus aureus ATCC 29213, and Pseudomonas aeruginosa ATCC 27853 were used as quality control strains. Every 10th Sensititre panel was inspected and evaluated by a second investigator.

The protocol for data collection was performed according to the EUCAST SOP 10.1 for ECOFF setting (2). The MICs were determined in three different laboratories by different investigators at (1) the National Food Institute at the Technical University of Denmark, (2) the Department of Veterinary and Animal Sciences at the University of Copenhagen, and (3) the Institute for Experimental Pathology at the University of Iceland.

Firstly, the MIC distributions were visually inspected (e.g., the Gaussian wildtype MIC distributions were identified) to ascertain that the “ECOFFinder” version 2.0 software could be applied (nonlinear regression, at 99%) (4). Additionally, the MIC distributions for each antimicrobial agent and species were compared to the modes and ECOFFs already established and available in the EUCAST database (13).

Prior to analyzing results for SXT, MIC distributions for sulfamethoxazole and trimethoprim were created separately. A “true” SXT wildtype MIC distribution was solely defined on organisms, which were independently wildtype to both agents. Isolates in the SXT wildtype population with MICs > ECOFF for sulfamethoxazole alone were omitted, as the effect of the combinational drug, SXT, must be attributed by the addition of trimethoprim.

Resistance genes were deducted from whole-genome sequencing of randomly selected 64 E. coli (Danish) and 34 S. delphini isolates originating from Denmark, Spain, Iceland, the Netherlands, and Finland. Briefly, DNA was isolated from culture material using a Maxwell®16 equipment and the 16 LEV Blood DNA Kit according to the manufacturer's instructions (Promega Corporation, USA). The S. delphini isolates were treated with lysostaphin before extraction as described in Strube et al. (14). DNA purity and concentration were assessed using NanoDrop ND-1000 (NanoDrop Technologies, USA) and Qubit® (Life Technologies, USA). Library preparation (NextEra XT DNA sample preparation kit, Illumina, USA) and sequencing (Illumina NextSEQ-based technologies in a 150 base pair paired-end configuration with an expected coverage of 50) was outsourced to Statens Serum Institut, Denmark.

Sequences were quality-checked by fastx_quality_stats from the FASTX-Toolkit (FASTX-Toolkit, RRID:SCR_005534) (http://hannonlab.cshl.edu/fastx_toolkit/). Using Centrifuge (Centrifuge Classifier, RRID:SCR_016665), the reads were classified and checked for contamination (15). Contigs were assembled using SPAdes (SPAdes, RRID:SCR_000131) with the setting: settings “-k 21,33,55,77–careful” (16). The assemblies were checked using Quast (QUAST, RRID:SCR_001228) and annotated using Prokka (Prokka, RRID:SCR_014732) (17, 18). Subsequently, antimicrobial resistance genes were identified by running sequences through the ResFinder pipeline (19).

For E. coli, MIC distributions for six antimicrobial agents are presented (Figures 1–6). Data and derived TECOFFs were in accordance with the EUCAST ECOFFs (Table 2). Antimicrobial resistance genes were not detected in 18 of the 64 sequenced isolates. With only few exceptions, these isolates were found in the wildtype populations (Figures 1–6). With the exception of colistin and spectinomycin, a high number of isolates were part of the non-wildtype populations (Table 6). For three of the agents, the ECOFFinder suggested a value one dilution lower than the EUCAST ECOFF. However, there were no indications other than that the range and mode of colistin, spectinomycin and SXT were in accordance with the EUCAST ECOFFs (Figures 2, 4, 6). Hence, these TECOFFs were visually determined (Table 2).

For amoxicillin, a bimodal distribution was identified. The MIC range, TECOFFand mode for amoxicillin are presented in Figure 1 and Table 2, respectively. Beta-lactam resistance genes were not detected in 23 of the 64 sequenced isolates. All of these were in the wildtype population. Forty-one of the sequenced isolates harbored a β-lactam resistance gene. None of these isolates were in the wildtype population (Figure 1). Genes belonging to the bla_TEM−1 family were most prevalent, while one isolate carried the bla_CTX−M−1 gene encoding an extended-spectrum beta-lactamase (ESBL).

For colistin, the distribution was mono-modal exhibiting a Gaussian distribution in the range 0.25–2 μg/mL (Figure 2). The mode and TECOFF of the colistin MIC values are presented in Table 2. No colistin resistance genes were detected in any of the sequenced E. coli isolates (Figure 2).

Two apparently overlapping populations were detected for doxycycline. The range, TECOFF and mode of doxycycline MIC distribution are presented in Figure 3 and Table 2, respectively. The finding of two overlapping populations was supported by the results and distribution of the sequencing data (Figure 3). Three isolates had an MIC > 128 μg/mL and might represent a third population. Thirty-six of the sequenced isolates had no tetracycline resistance genes and were part of the wildtype population. One isolate had no known tetracycline resistance genes despite having an MIC >128 μg/mL. Twenty-seven of the isolates harbored a tetracycline resistance gene [tet(A) or tet(B)]. None of these isolates were in the wildtype population (Figure 3).

For spectinomycin, the MIC distribution, TECOFF and mode is presented in Figure 4 and Table 2, respectively. Forty-two of the sequenced isolates had no spectinomycin resistance genes and were part of the wildtype population. Twenty-two of the sequenced isolates harbored a spectinomycin resistance gene (aadA5 or aadA1). Seven of these had an MIC < ECOFF, and 15 of these had an MIC > ECOFF (Figure 4).

For sulfamethoxazole, a bimodal distribution was identified. The range, TECOFF and mode of sulfamethoxazole MIC values are presented in Figure 5 and Table 2, respectively. Resistance genes were not detected in 32 of the sequenced isolates. All of these were part of the wildtype population. Thirty-two of the sequenced isolates harbored a sulfonamide resistance gene with sul2 being the most prevalent. None of these isolates were in the wildtype population (Figure 5).

For sulfamethoxazole in combination with trimethoprim (SXT), three populations were apparent, and the wildtype population displayed a Gaussian distribution in the range 0.03–0.25 μg/mL (Figure 6). A “true” SXT wildtype MIC distribution was solely defined on organisms, which were independently wildtype to both agents. Therefore, 26 sulfamethoxazole non-wildtype and concomitantly SXT wildtype were omitted. One isolate in the SXT wildtype population was omitted due to high trimethoprim MIC (64 μg/mL) (Supplementary Table 1A). All included isolates with SXT MICs of 0.12 μg/mL and 0.25 μg/mL were sensitive to trimethoprim alone (MIC ≤1 μg/mL) (Supplementary Table 1A). Thirty-nine of the isolates with an SXT MIC of 0.06 μg/mL were tested and proved sensitive to trimethoprim alone (MIC ≤ 1 μg/mL) (Supplementary Table 1A). The mode and range of the SXT MIC values are presented in Table 2. There was no indication other than that the MIC distribution from the current study was in accordance with the EUCAST database. All isolates without detected sulfonamide nor trimethoprim resistance genes were in the wildtype population. Three non-wildtype isolates (MIC of 0.5 and 2 μg/mL) harbored only a sulfonamide resistance gene (Figure 6B). Eighteen of the sequenced isolates harbored both sulfonamide (sul1, sul2, sul3) and trimethoprim resistance genes (dfrA1, dfrA5, dfrA8, dfrA14). Of these 18 isolates, 14 had an MIC > 64 μg/mL, two had an MIC = 64 μg/mL, and two had an MIC = 4 μg/mL (Figure 6).

For S. delphini, the results of the seven tested antimicrobials are presented in Figures 7–13. Tentative ECOFFs were suggested for six of the antimicrobials (Table 3). In seven of the 34 sequenced isolates, no resistance genes were detected, and these isolates were mostly located in the wildtype population (Figures 7–11, 13). For doxycycline and lincomycin, high fractions of isolates were in the non-wildtype populations (Table 6).

All isolates had an MIC ≤ 0.25 μg/mL to amoxicillin (Figure 7), truncating the dataset to the left. Since the test range did not cover the MIC distribution, it was not possible to suggest a TECOFF. Beta-lactam resistance genes were not detected in 16 of the sequenced isolates. Eighteen of the sequenced isolates harbored the β-lactam resistance gene blaZ; these isolates were tested against penicillin. Five of those were non-wildtype against penicillin (MIC ≥ 0.25 μg/mL, Supplementary Table 2) when using the EUCAST ECOFF for S. aureus (13).

A bimodal distribution was apparent for doxycycline; the wildtype population was truncated to the left in the range ≤ 0.06–0.25 μg/mL. A TECOFF of 0.12 μg/mL was suggested (Table 3, Figure 8). Nineteen of the sequenced isolates had no tetracycline resistance genes and were part of the wildtype population. Fifteen of the sequenced isolates harbored the tetracycline resistance gene tet(M), none of these isolates were in the wildtype population (Figure 8).

For spectinomycin, the apparent wildtype population was in the range 16–64 μg/mL, but due to the lack of a Gaussian distribution it was not possible to apply the ECOFFinder 2.0 software [(4), Figure 9]. A TECOFF of 128 μg/mL was suggested by visual inspection (Table 3). Thirty-two of the sequenced isolates had no spectinomycin resistance genes and were part of the wildtype population. Two of the sequenced isolates harbored the spectinomycin resistance gene spc, and both had MICs above the test range > 512 μg/mL (Figure 9).

For tylosin, three populations could be identified. The wildtype population displayed a Gaussian distribution in the range 0.25–2 μg/mL. A TECOFF of 2 μg/mL was suggested (Table 3, Figure 10). Twenty-nine of the sequenced isolates had no macrolide resistance genes and were part of the wildtype population. Five of the sequenced isolates harbored macrolide resistance genes, none of these isolates were in the wildtype population (Figure 10). Four different macrolide resistance genes were identified, all belonging to the erm gene family encoding macrolide, lincosamide and streptogramin B resistance (MLS_B).

At least two populations were apparent for lincomycin with the wildtype population displaying a Gaussian distribution in the range 0.12–2 μg/mL. A TECOFF of 2 μg/mL was suggested (Table 3, Figure 11). Twenty of the sequenced isolates had no macrolide nor lincomycin resistance genes, all but three were part of the wildtype population. Ten of the sequenced isolates harbored the lincomycin resistance gene, lnu(A); none of these isolates were in the wildtype population. Additionally, five of the sequenced isolates harbored erm genes, all had a lincomycin MIC above the test range (>128 μg/mL, Figure 11).

There was only one apparent population for sulfamethoxazole, and a TECOFF of 128 μg/mL was suggested (Table 3, Figure 12).

For SXT, the wildtype population displayed a Gaussian distribution in the range ≤0.03–0.5 μg/mL. Two isolates were omitted, so that all isolates within the SXT wildtype population (Figure 13) were sensitive to sulfamethoxazole alone (MIC ≤ 128 μg/mL) (Figure 12). Thirty-eight randomly selected isolates in the SXT wildtype population were tested against trimethoprim alone, and all were sensitive (MIC ≤ 8 μg/mL) (Supplementary Table 1B). A TECOFF of 0.25 μg/mL was suggested for SXT (Table 2, Figure 13). Two isolates harbored two different trimethoprim resistance genes, dfrK and dfrG, and these displayed MICs of 2 and 8 μg/mL, respectively (Figure 13).

The MIC distributions of S. canis are presented in Supplementary Figure 1. Tentative ECOFFs were suggested for five of the seven antimicrobials tested (Table 4). With the exception of SXT, a high number of isolates were found in the non-wildtype populations (Table 6).

For amoxicillin, all isolates had an MIC ≤ 0.25 μg/mL (Supplementary Figure 1A), truncating the dataset to the left. Since the test range did not cover the MIC distribution, it was not possible to suggest a TECOFF.

The majority of the isolates displayed a Gaussian distribution for doxycycline in the range 8–32 μg/mL (Supplementary Figure 1B). However, this distribution was most likely not the wildtype distribution, since two isolates had MIC values of 0.25 and 2 μg/mL, respectively, and since the ECOFF for the closely related species S. pyogenes (Table 4) and S. pneumoniae is 0.5 μg/mL (13). Consequently, a TECOFF was not proposed.

Two main distributions were apparent for spectinomycin; the wildtype population displayed a Gaussian distribution in the range 8–32 μg/mL (Supplementary Figure 1C). A TECOFF of 32 μg/mL was suggested (Table 4).

Two distributions were apparent for tylosin, and the wildtype population was truncated in the range ≤0.125–0.25 μg/mL (Supplementary Figure 1D). Visual inspection of the truncated data indicated a tylosin TECOFF of 0.25 μg/mL (Table 4).

Similarly, for lincomycin, two populations were apparent with the wildtype population truncated in the range ≤0.06–0.25 μg/mL (Supplementary Figure 1E). Visual inspection of the truncated data indicated a TECOFF of 0.5 μg/mL (Table 4).

For sulfamethoxazole, probably two overlapping populations were apparent in the range 8–>512 μg/mL. The wildtype distribution was most likely in the range 8–128 μg/mL (Supplementary Figure 1F). A TECOFF of 128 μg/mL was suggested (Table 4).

For SXT, the wildtype population displayed a Gaussian distribution in the range ≤0.03–0.12 μg/mL (Supplementary Figure 1G). Eight isolates were omitted, so that all isolates within the SXT wildtype population were sensitive to sulfamethoxazole alone (MIC ≤ 128 μg/mL, Table 4). Further, 26 randomly selected isolates in the SXT wildtype population were tested against trimethoprim alone, and one isolate with MIC ≥ 4 μg/mL was omitted (Supplementary Table 1C). A TECOFF of 0.12 μg/mL was suggested (Table 4).

For P. aeruginosa, MIC distributions and results of the tested antimicrobials are presented in Supplementary Figure 2 and Table 5.

For colistin, only one population was apparent (Supplementary Figure 2A). The MIC range and mode for colistin were similar to the EUCAST MIC distribution and the ECOFF of 4 μg/mL (Table 5). All isolates were in the wildtype population (Table 6).

For SXT, the wildtype population displayed a Gaussian distribution in the range 2–64 μg/mL (Supplementary Figure 2B). A tentative ECOFF of 32 μg/mL was suggested (Table 5). The TECOFF places 24% of the isolates in the non-wildtype population (Table 6).