A total of 30 non-duplicate veterinary strains of P. aeruginosa showing non-susceptibility to imipenem and/or meropenem according to the current Clinical and Laboratory Standards Institute breakpoints (CLSI, MIC >2 mg/L), were included in this study. These strains were collected between 2009 and 2014 by the French bacterial resistance surveillance network Resapath (http://www.resapath.anses.fr), from dogs (n = 24), cats (n = 5), and a bovine (n = 1). Wild-type reference strain PAO1 and its derived mutants PT629, PAO7H, and CMZ091 which overproduce efflux pumps MexAB-OprM, MexEF-OprN, and MexXY/OprM, respectively were used as controls in RT-qPCR experiments (Köhler et al., 1997a,b; Muller et al., 2011).

The minimal inhibitory concentrations (MICs) of selected antibiotics were determined by the conventional microdilution method in Mueller-Hinton broth (MHB, Biorad) containing adjusted concentrations of Ca²⁺ (from 20 to 25 mg/L) and Mg²⁺ (from 10 to 12.5 mg/L) (CLSI, 2017). Categorization of strains as S/I/R referred to the current CLSI breakpoints. Screening of carbapenemase-producing strains was performed by using the I-C4000 test, previously shown to have excellent sensitivity and specificity (Fournier et al., 2013a). Briefly, a disk of imipenem (10 μg load) and a disk of imipenem plus 4,000 μg of cloxacillin were deposited on to the surface of a Mueller-Hinton agar plate previously inoculated with a bacterial suspension at 0.5 McFarland. After 18 h of incubation at 37°C, a difference >5 mm between the inhibition diameters around the two disks—due to inhibition of natural ß-lactamase AmpC by cloxacillin, was indicative of absence of carbapenemase production.

The clonal relatedness of strains was investigated by both MLST (MultiLocus Sequence Typing) and MLVA (Multiple-Locus Variable-number tandem-repeat Analysis) targeting 10 VNTR (Variable Number Tandem Repeats), as previously reported (Vu-Thien et al., 2007; van Mansfeld et al., 2009). The allelic profile of an isolate was expressed as the number of repeats for each VNTR in the following order: ms142, ms211, ms212, ms213, ms214, ms215, ms216, ms217, ms222, and ms223. A difference of one repeat between two strains for any of the VNTR was considered as discriminant.

The transcripts of genes copR, czcC, czcR, mexB, mexE, mexY, oprD, and PA1797 were quantified in a Rotor gene RG6000 apparatus (Qiagen) with fluorescent intercalating dye SybrGreen (Fast SybrGreen kit, Qiagen). The primers used for RT-qPCR are listed in Table S1. Briefly, RNA was extracted en masse from mid-log phase cultures (A_{600 nm} equal to 0.8) in MH broth with the RNeasy Plus kit (Qiagen). DNA was removed by RNase-free DNAse treatment, and 2 μg of purified RNA were incubated with ImpromII reverse transcriptase according to the manufacturer's recommendations (Promega). The transcript levels were relativized to the PAO1 values after internal gene normalization (uvrD), as previously described (Jo et al., 2003). Data presented are means of four determinations from two independent experiments.

Potential mutations in gene oprD and its promoter region were searched in the tested isolates. The PCR amplicons were sequenced on both strands by using the BigDye Terminator V3.1 cycle sequencing kit (Applied Biosystems) and specific primers (Table S1), in an RUO3500 Genetic Analyzer (Applied Biosystems). Genes mexR, nalC, nalD, mexZ, parS, parR, mexS, copS, copR, and/or czcR were sequenced in strains showing significant overexpression of mexB, mexY, mexE, copR, and/or czcC genes. The resultant nucleotide and amino acid sequences were aligned with that of reference strains PAO1 (BenBank, accession no. NP249649), PA14 (accession no. CP000438), LESB58 (accession no. WP003107743), and PA7 (accession no. WP012076895), deposited in GenBank (https://blast.ncbi.nlm.nih.gov).

Carbapenems are not licensed for the treatment of animals. However, 30 imipenem and/or meropenem non-susceptible P. aeruginosa were retrospectively identified in a collection of 527 veterinary strains gathered between 2008 and 2014 by the French network Resapath (Table 1). These resistant strains had been the cause of otitis in dogs (n = 13), and pulmonary infections in cats, dogs, and a bovine (n = 6). Their clonal relatedness was explored by MLVA, which allowed the identification of 25 different genotypes. Most were singletons, but 3 clones included 2 (clones A, G) and 4 (clone B) isolates, respectively. Suggesting that some strains may be more transmissible than others, analysis of the Resapath database indicated that clones A and G had infected animals from distant geographical regions in France (Moselle, Val-de-Marne, and two distinct towns in the district of Alpes Maritimes, respectively). Interestingly, clone B strains were isolated from unrelated animals admitted to the same veterinary clinic in 2010 (three strains between July and October) and in 2014 (one strain in December), probably as part of an outbreak. MLST genotyping experiments showed that clone B belongs to international high-risk complex ST395, known for its propensity to spread in hospitals and to develop multidrug resistance (Libisch et al., 2009; Fernandez-Olmos et al., 2013; Martin et al., 2013; Valot et al., 2014). ST395 can also be encountered in the natural environment as well as in waste waters (Slekovec et al., 2012; Teixeira et al., 2016). Furthermore, we found that 21 strains of the collection (70%) belonged to various sequence types initially identified among clinical isolates (Table 1) (Samuelsen et al., 2010; Garcia-Castillo et al., 2011; Cho et al., 2013; Gomila et al., 2013; Perez et al., 2014). In apparent contradiction with previous studies, our data thus indicated that clones ST395 and ST233 are able to infect humans and animals (Wiehlmann et al., 2007; Kidd et al., 2012; Haenni et al., 2015).

As mentioned in Table 1, only 10 strains (33.3%) were intermediate (I) or resistant (R) to both imipenem and meropenem with respect to the CLSI breakpoints. Some strains were also I or R to major antipseudomonal antibiotics such as cefepime (6.6%), piperacillin/tazobactam (6.6%), ticarcillin (83.3%), ciprofloxacin (40%), and aztreonam (36.6%), but all were susceptible (S) to ceftazidime, tobramycin, amikacin, and colistin (Table 2). According to an international consensus, 12 (40%) strains fitted with the definition of multi-drug resistant bacteria (MDR) (Magiorakos et al., 2012).

Screening of carbapenemase production with IC-4000 test was negative for all the veterinary strains. As the loss of specific porin OprD is the main cause of carbapenem resistance in human isolates of P. aeruginosa (Lister et al., 2009; Fournier et al., 2013b), we analyzed the sequence of gene oprD and its promoter region among the selected strains. In the four clone B strains belonging to ST395, oprD was disrupted by a same deletion of 11 nucleotides, while in other two strains the gene was interrupted by an insertion sequence, ISPa1328 or ISPst3 (Table 1). Of note, ISPa1328 has previously been identified in carbapenem-resistant hospital strains in France, United-States, Lebanon, and Japan (Wolter et al., 2004; Sanbongi et al., 2009; Fournier et al., 2013b; Al Bayssari et al., 2015).

The primary sequence of porin OprD is known to be highly variable in clinical and environmental P. aeruginosa, including those susceptible to carbapenems (Ocampo-Sosa et al., 2012). Most of the veterinary strains exhibited OprD sequences identical to that of reference strain LESB58 (n = 12), PA14 (n = 5), or PAO1 (n = 4). A so far uncharacterized G₄₀₄C substitution was identified in one strain, 25334, producing a LESB58-like porin. Finally, the last two strains, 25752 and 36150, were found to harbor the same S₅₇E and S₅₉R changes in a PP2-like OprD. According to Ocampo-Sosa A. et al., these variations, which are located in porin loop L1 would not impact the diffusion of carbapenems through the outer membrane, suggesting that 25752 and 36150 resist to carbapenems via a mechanism other than membrane impermeability (Ocampo-Sosa et al., 2012).

In P. aeruginosa, the intracellular penetration of carbapenems may be impaired consecutively to mutational loss of porin OprD (as reported above), mutation-driven changes in the OprD structure, or the presence of reduced amounts of OprD in the outer membrane (Fournier et al., 2013b). Indeed, some mutations in the promoter region of oprD or in regulatory genes that control efflux pumps such as CzcCBA, MexEF-OprN, and MexXY negatively influence oprD transcription, leading to a 2- to 8-fold increase in imipenem MIC (Lister et al., 2009; Muller et al., 2011; Fournier et al., 2013b). RT-qPCR experiments demonstrated that 14 veterinary strains including isolates 25334, 25752, and 36150 that exhibit novel OprD variants, expressed the oprD gene 2- to 16.8-fold less than wild-type reference strain PAO1 (Table 1). None of these isolates harbored mutations in the promoter region of oprD. On the other hand, three of them (25752, 36150, and 37249) significantly overexpressed gene czcC (3.25-, 4.90-, and 17.27-fold more than PAO1, respectively). The efflux system CzcCBA contributes to the intrinsic resistance of P. aeruginosa to zinc and cadmium ions, and is under the positive control of two-component regulatory systems CzcRS, and CopRS (Hassan et al., 1999; Okamoto et al., 2002; Caille et al., 2007). Phosphorylation of response regulators CzcR and CoprR by histidine kinases CzcS and CopS, respectively results in activation of operon czcCBA and concomitant down-regulation of gene oprD (Caille et al., 2007). In agreement with this, sequence analysis of loci czcRS and copRS demonstrated the occurrence of amino acid substitutions in sensor protein CzcS from strains 25752 (A₁₁₀Y, S₄₆₈A), 36150, (S₄₆₈A), and 37249 (P₂₂₆D), and in CopS from strains 25752 (A₃₀₄Y) and 37249 (L₂₅₈R) compared with PAO1, pointing to a possible cause of activation of CzcS, CopS, and in fine CzcCBA. Supporting this interpretation, expression of genes czcR and copR in the three strains was found to be 3.11- to 7.42-fold higher than in PAO1. Both genes are known to be positively controlled by their respective sensors at the transcriptional level (Okamoto et al., 2002). Mutational alteration of two component system(s) CzcRS and/or CopRS is thus clearly associated with carbapenem resistance in both clinical and animal strains of P. aeruginosa (Fournier et al., 2013b).

Four strains showing a downregulated oprD gene (namely, 25334, 26103, 26860, and 37249) appeared to have mexE transcripts 89.5- to 3,057-fold more abundant than in PAO1. As demonstrated previously, overproduction of efflux system MexEF-OprN in clinical nfxC strains mainly results from mutations inactivating or impairing the activity of putative oxidoreductase MexS (Richardot et al., 2016). Analysis of the four veterinary mexE-overexpressing strains indeed revealed disruption of mexS in three of them (del_20nt in isolates 25334 and 37249, lack of amplification of mexS gene probably due to an IS in 26103 and a M₂₂₇V substitution in the enzyme from 26860). To the best of our knowledge, disruption of mexS by an insertion sequence has never been reported so far in clinical or in vitro-selected nfxC mutants. The same also applies to the M₂₂₇V change observed in strain 26860, though its contribution to the MexS-dependent activation of global regulator MexT, with subsequent upregulation of efflux MexEF-OprN and downregulation of porin OprD, remains to be confirmed. All the veterinary nfxC mutants exhibited a significant resistance to imipenem (MIC = 8 mg/L). Of interest, the concomitant dysregulation of systems CzcCBA and MexEF-OprN in strain 37249 had cumulative effects on imipenem MIC (16 mg/L) (Table 1).

Finally, repression of gene oprD may be caused by mutational activation of ParRS, a two-component system, which positively controls operon mexXY and an adjacent gene of unknown function, PA1797 (Muller et al., 2011). As indicated in Table 3, among the 14 strains exhibiting low oprD mRNA levels, eight significantly upregulated genes mexY (from 9.44- to 72.57-fold) and PA1797 (from 7.50- to 95.35-fold), compared with PAO1. Amino acid substitutions were found in response regulator ParR (strain 35941) and sensor ParS (n = 7 strains) (Table 3). None of these except A₁₃₈T in ParS has been reported so far in agrW2 mutants isolated in the clinical setting (Muller et al., 2011; Guénard et al., 2014). They are located in the periplasmic (L₅₀P), the HAMP (Histidine kinase, Adenylyl cyclase, Methyl-accepting chemotaxis proteins and phosphatases) (R₁₈₅G), the HisKA (A₂₁₅T), and the ATPase (A₃₂₄V) domains of ParS. The finding that ParRS-controlled genes mexXY and PA1797 are overexpressed in those strains strongly supports the notion of a constitutive activation of ParRS by such mutations, as observed in clinical P. aeruginosa isolates (Muller et al., 2011; Guénard et al., 2014). Providing further evidence of ParRS activation, colistin MIC was 2-fold higher for the eight agrW2 mutants than for the rest of the collection. Indeed, in addition to downregulating the oprD gene, ParRS is able to upregulate expression of LPS modification operon arnBCADTEF-ugd in this type of mutants, with consecutive increase in resistance to polymyxins (Muller et al., 2011). Figure 1 depicts the regulatory pathways involved in repression of oprD gene in the veterinary strains studied.

When overproduced, the efflux system MexAB-OprM increases the resistance of P. aeruginosa to ß-lactams including meropenem and doripenem, but not to imipenem (Masuda et al., 2000; Okamoto et al., 2002). To assess the role of this pump in the meropenem resistance of veterinary strains, gene mexB transcripts were quantified in all the isolates by RT-qPCR. As indicated in Table 4, 12 out 30 strains (40%) appeared to overexpress mexB from 3.4- to 15.75-fold, and to be 4- to 16-fold more resistant than wild-type strain PAO1 to meropenem (MIC of meropenen equal to 0.5 mg/L in PAO1) (Table 1). Transcription of mexAB-oprM is subject to a complex regulation involving two repressors, NalD and MexR, which respectively bind to a proximal and a distal promoter upstream of the operon (Poole et al., 1996; Morita et al., 2006). Recently, it was reported that a response regulator, named CpxR, is also able to activate mexAB-oprM via its interaction with the distal promoter mentioned above, in mexR knock-out mutants (Tian et al., 2016) Finally, another regulator, NalC, indirectly influences mexAB-oprM transcription by modulating the expression of a two-gene operon in which armR encodes an anti-MexR protein (Cao et al., 2004; Daigle et al., 2007). Therefore, many genetic events leading to the inactivation or alteration of protein MexR (nalB mutants), NalC (nalC mutants), or NalD (nalD mutants) are expected to upregulate the mexAB-oprM operon, and so to increase meropenem MIC in clinical isolates (Ziha-Zarifi et al., 1999; Higgins et al., 2003; Cao et al., 2004; Llanes et al., 2004; Sobel et al., 2005a; Chalhoub et al., 2016). Sequencing of the corresponding genes in the 12 MexAB-OprM-overproducers identified two isolates containing disrupted mexR genes (deletion of 1 and 10 nucleotides in 26292 and 36140, respectively), and a third one (36171) encoding a L₅₄P substitution in the DNA binding domain of MexR (Table 4) (Lim et al., 2002). Four other strains were found to carry mutations generating a frameshift in gene nalC (deletion of 12 nucleotides in 26860 and 36145), a truncated NalC (R₁₄₃stop in 25827), or amino acid substitutions in this repressor (E₁₅₃D, A₁₈₆D in 25333). Five remaining strains harbored disruptive mutations in nalD (strains 25401, 25572, 25333, 25380, 36163) while in the strain 37248 the gene coding sequence was interrupted by an ISAs2 element (Table 4). Reminiscent of this, mexR was reported to be inactivated by another insertion sequence, IS21, in a MexAB-OprM upregulating clinical strain (Boutoille et al., 2004). As already observed in clinical strains (Ziha-Zarifi et al., 1999; Srikumar et al., 2000; Llanes et al., 2004; Sobel et al., 2005a), mutations in both nalC and nalD were present in one strain, 25333, though with no apparent influence on mexB expression and/or meropenem MIC, as compared with single mutants. To our knowledge, all the nalB (n = 3), nalC (n = 3), nalD (n = 5), and nalC/D (n = 1) mutations reported here are novel.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.