Clonal Confinement of a Highly Mobile Resistance Element Driven by Combination Therapy in Rhodococcus equi

MDR clades arise upon acquisition of resistance traits, but the determinants of their clonal expansion remain largely undefined. Taking advantage of the unique features of Rhodococcus equi infection control in equine farms, involving the same dual antibiotic treatment since the 1980s (a macrolide and rifampin), this study sheds light into the determinants of multiresistance clonality and the importance of combination therapy in limiting the dissemination of mobile resistance elements. Clinically effective therapeutic alternatives against R. equi foal pneumonia are currently lacking, and the identified macrolide-rifampin MDR clone 2287 has serious implications. Still at early stages of evolution and local spread, R. equi 2287 may disseminate globally, posing a significant threat to the equine industry and, also, public health due to the risk of zoonotic transmission. The characterization of the 2287 clone and its resistance determinants will enable targeted surveillance and control interventions to tackle the emergence of MDR R. equi.

R hodococcus equi is a soil-dwelling facultative intracellular actinobacterium that causes pyogranulomatous infections in animals and immunocompromised people (1)(2)(3). While affecting a variety of animal species, R. equi is most commonly isolated from foals, in which it causes a life-threatening multifocal pneumonic disease with frequent extrapulmonary involvement (1,3). Attack rates in farms where the disease is endemic are typically 10 to 20% or higher, and a larger proportion of foals can be subclinically affected (4,5). Equine isolates harbor a host-adapted virulence plasmid of Ϸ80 kb designated pVAPA (6)(7)(8)(9), required for intramacrophage survival and pathogenesis (10). pVAPA is easily lost in the absence of host selection but can be readily regained, and transmitted between isolates, via conjugal transfer (7,9,(11)(12)(13). Maintenance of environmental pools of pVAPA-bearing (i.e., "virulent") R. equi through fecal-oral cycling is thought to contribute to farm-level endemicity. Together with two other host-associated virulence plasmids (porcine pVAPB and ruminant pVAPN) (3,8,9,13), pVAPA plasmids can be found in human isolates (7), consistent with equine settings being a potential source of human R. equi infection (2).
Control of foal rhodococcosis is challenging due to the lack of an effective vaccine and relies on antibiotic therapy (4,14). Treatments are not only applied to clinically affected animals but also preventatively to presumptive cases identified by thoracic ultrasonographic screenings (4,5,(15)(16)(17). Although R. equi is susceptible to a variety of antimicrobials in vitro (18,19), many drugs are ineffective in vivo (2,20). Inconsistent in vitro susceptibility to some antimicrobials is also observed, with intrinsic resistance noted for ␤-lactams and quinolones (18,(21)(22)(23)(24). Thus, the mainstay of R. equi therapeutics on equine farms has been and remains the combination of a macrolide (erythromycin, clarithromycin, or azithromycin) and rifampin (4).
Since its introduction in the late 1980s (20,25), the macrolide-rifampin combination has dramatically reduced foal mortality, until recently with little evidence of resistance (4,26). A changing trend was, however, observed in the 2000s, with rates of up to Ϸ4% of high-level resistance to both macrolides (MICs, 12 to Ͼ256 g/ml) and rifampin (MIC, Ն32 g/ml) reported in the United States (28). Alarmingly, a recent study found up to 40% of foals yielding isolates highly resistant to macrolides and rifampin as a result of mass antimicrobial treatment (16). While rifampin-only resistance due to rpoB mutations has been reported for R. equi (19,23,(27)(28)(29)(30), including at least two instances documented during monotherapy in foals (30,31), macrolide resistance has so far always been associated with rifampin resistance (16,28,32).
We recently identified a novel erm gene, erm (46), as the cause of the emerging macrolide resistance among equine R. equi isolates in the United States (32). erm (46) confers resistance against all macrolides (MIC 90 64 to Ͼ256 g/ml for azithromycin, clarithromycin, and erythromycin), lincosamides, and streptogramins B (MLS). In contrast to the closely related mycobacterial erm rRNA methylase genes erm (38) and erm (39), which are part of the intrinsic core resistome, erm (46) was transferable by mating at frequencies of up to 10 Ϫ2 to 10 Ϫ3 (32). This was highly concerning not only due to the risk of rapid spread among equine-associated R. equi organisms but also because macrolides are critically important antibiotics in human medicine (33) and are used in the treatment of R. equi infections in people (2).
Here we report the detailed genetic characterization of the erm (46) determinant. erm (46) is part of a highly mobile transposable element carried on a conjugative plasmid, pRErm46, from which it colonizes the R. equi genome, including the virulence plasmid. Despite its high mobility and transfer potential, the erm(46) element is confined to a specific R. equi clone, likely as a result of strong coselection due to concomitant chromosomal rifampin resistance.

R. equi erm(46)
is carried on a conjugative plasmid. Whole-genome sequences (WGS) of 18 MLS-resistant (MLS r ) R. equi equine isolates previously used to identify the erm(46) gene (32) were analyzed to characterize the nature of the transferable element involved. In most WGS assemblies, erm (46) was found together with an adjacent integrase/transposase gene (32) in a unique 6.9-kb contig (Fig. 1A). In one instance (strain PAM 2275 [see Table S2 in the supplemental material]), the same erm(46)containing 6.9-kb fragment was present in an Ϸ50to 62-kb contig (depending on the assembly) with no homology to any known R. equi genomic DNA (21,34) and which contained genes encoding putative conjugation-related type IV secretion system (T4SS) proteins plus an insertion sequence (IS6100 [see below]) at one of its ends. We hypothesized this larger contig might be part of a conjugative plasmid harboring the erm(46) gene within a 6.9-kb transposable element, which then had moved to different genome locations, explaining its assembly as a separate 6.9-kb contig in most isolates.
All erm(46)-positive strains showed the presence of a distinct plasmid band, strikingly, at the very same position as the plasmid DNA from MLS s 103S even if lacking FIG 1 Integrative elements of the R. equi macrolide resistance plasmid pRErm46. (A) Genetic structure of the 6.9-kb transposon TnRErm46 carrying the macrolide resistance gene erm (46). The ISRe46 transposase is a novel member of the IS481 family. Its closest homolog is ISRae1 from Rhodococcus aetherivorans (amino acid identity, 82% [274/333]). TMP is a 25-kDa Gap-like (TauE/SafE superfamily) putative transmembrane protein with a possible role in small-molecule transport/export. See text and Table S1 for other TnRErm46 components. DR, direct repeats (shaded) at the junction with genomic DNA and adjacent inverted repeats (IR), which comprise the CTAG sequence targeted by TnRErm46. This sequence provides the TAG stop codons for the ISRe46 transposase on the right end and for some target genes-whose function is thus not interrupted (e.g., parB on the PAM 2287 pRErm46 plasmid)-on the left end. The left IR provides the TGA stop codon of the transposon's distal hypothetical gene (HP). Stop codons are boxed. (B) Genetic structure of pRErm46's class 1 integron (C1I). The black rectangles flanking IS6100 represent the 14-bp terminal inverted repeats with the sequence GGCTCTGTTGCAAA. The TnRErm46 insertion within the C1I cassette gene aadA9, unique to PAM 2287's pRErm46, is indicated. The aadA9 gene is uninterrupted in most MLS r isolates where pRErm46 was detected, typically within an Ϸ4.4-kb contig covering the intI1-aadA9-qacE-sul1-orf5-tniΔ C1I sequence. PAM 2287's pRErm46 carries two additional TnRErm46 copies (see Fig. 3 and Table S2). nt, nucleotide.
Emerging MDR R. equi 2287 Clone ® pVAPA (PAM 2350 transconjugant) ( Fig. 2A). The identity of the DNA bands was determined by Southern blotting with specific probes for pVAPA (internal vapA fragment) and the larger Ϸ50-/62-kb contig [internal erm (46) and IS6100 fragments]. The vapA probe identified the expected pVAPA band in the equine isolates PAM 2285 and 2287, transconjugant PAM 2351, and control 103S but did not hybridize with the similarly positioned plasmid DNA band from the erm(46)-positive PAM 2350 transconjugant (Fig. 2B). On the other hand, the erm (46) and IS6100 probes detected a specific band at the same position as the vapA signal in isolate PAM 2287 as well as the PAM 2350 and 2351 transconjugants, but not the control 103S plasmid preparation (Fig. 2B). These data were consistent with erm (46) being present in an extrachromosomal element of a size similar to that of pVAPA also harboring IS6100, perhaps in several copies. Indeed, BLASTN searches identified the insertion sequence in 14 additional MLS r genome assemblies (but not MLS s genomes) as a discrete 881-bp contig exactly corresponding to the coding sequence of the 254-residue IS6100 tranposase plus cognate 14-bp terminal inverted repeats (IR).  (46) or the virulence plasmid is indicated by plus and minus signs, respectively. The macrolide-susceptible, virulence plasmid-harboring R. equi 103S strain was used as a control. (A) Agarose gel of plasmid preparations of the indicated representative strains. M, molecular size marker (Promega 1-kb DNA ladder; top band is 10 kb). The arrowhead indicates the position of the virulence plasmid and the (pRErm46) resistance plasmid band; both migrate at a theoretical height of Ϸ15 kb due to supercoiling. The additional faster-migrating band in the plasmid preparations of PAM 2287 and cognate PAM 2350 transconjugant corresponds to a cryptic plasmid that appears to be inconsistently transferred. (B) Corresponding Southern blots probed with erm (46), IS6100, and vapA (pVAPA virulence plasmid) probes (see text for details).
Interestingly, no erm (46) or IS6100 probe signal was detected in PAM 2285 plasmid preparations (Fig. 2B) despite the 6.9-kb erm (46) contig being present in the WGS assemblies. Since erm (46) transfer was not observed with PAM 2285, we surmised that in this isolate the 6.9-kb erm(46)-containing putative transposable element might have jumped onto the chromosome with subsequent loss of the conjugative plasmid carrying it. Characterization of pRErm46, a self-transmissible plasmid responsible for emerging macrolide resistance in R. equi. Single-molecule real-time (SMRT) sequencing was performed on a plasmid-enriched DNA preparation from the pVAPA-negative transconjugant PAM 2350. This yielded the complete 5.0-Mb 103S chromosome and a second scaffold of 87,665 bp comprising the previously identified Ϸ50-/62-kb contig. This was assumed to correspond to the erm(46)-containing putative conjugative plasmid that comigrated with pVAPA (80.6 kb) and was named pRErm46 (Fig. 3).
pRErm46 is a covalently closed DNA molecule as determined with Circlator (35) and verified by PCR (Fig. 3). It shares no overall homology with database entries, except for discrete segments strongly similar to sequences from corynebacterial and (many) ␤-proteobacterial plasmids (see below). pRErm46 contains 104 coding DNA sequences (CDSs), of which 92% had homologs in other Actinobacteria, particularly Rhodococcus spp., suggesting a rhodococcal origin. The GϩC content (65.09%) is similar to the average for R. equi genomic DNA (68.7%) (21,34).
A function could be predicted for 49% of the CDSs. This included products involved in plasmid maintenance (Fic/Doc-like protein), self-replication (DnaK and DnaB homologs, putative DNA gyrase/topoisomerase) and partitioning (ParA and duplicated ParB homologs) (Table S1). Self-transmissibility is mediated by a MOB-P relaxase (36), MobC-like relaxase-accessory protein (RAP) (37), and T4SS apparatus with a TraG-like FIG 3 Genetic structure of pRErm46 from R. equi PAM 2287 (reference sequence, GenBank accession no. KY494640). Highlighted in color are the class 1 integron (yellow) and the TnRErm46 transposon (three copies, in red; underlined, TnRErm46 insertion shared by all pRErm46 plasmids, presumably the original site from which secondary transpositions took place). The macrolide resistance erm (46) gene within TnRErm46 is in black. Excluding the integrative elements, the pRErm46 backbone (in gray) is a conjugative replicon of 56.7 kb. pRErm46 (PAM 2287) sequence annotation is in Table S1.
Emerging MDR R. equi 2287 Clone protein and VirB4-like ATPase (38). A LysM-like murein endopeptidase and a secreted cutinase (Table S1) may help the DNA translocation complex through the bacterial cell envelope. Actinobacterial cutinases have lipolytic/esterase activity (39) and in mycobacteriophages are thought to act as auxiliary LysB lysins aiding the penetration of the thick mycolic acid layer (40,41). A cutinase has been shown to promote conjugation of the R. fascians virulence plasmid pFiD188 (42).
pRErm46's class 1 integron carries three directly repeated copies of the IS6 family insertion sequence IS6100 ( Fig. 1B and Fig. 3). First identified in Mycobacterium fortuitum (46), IS6100 is widespread among a diversity of Gram-negative organisms, consistent with an active horizontal exchange of this element between different bacterial groups (47). In Salmonella, IS6100 is found to be associated with the complex class 1 integron that constitutes the mobilizable multidrug resistance SG1 genomic island (48). Three IS6100 copies are also found with a similar arrangement in the homologous regions of the pTET3 (Fig. 1B) and pLEW279a plasmids, suggesting a recent common origin for the Corynebacteriales mobile class 1 integrons.
Two of pRErm46's IS6100 copies flank the core integron sequence. At the left, IS6100 is inserted into the intI1 coding sequence (Fig. 1B), which was interpreted in pTET3 as causing the truncation of the intI1 gene (44). However, analysis of the (identical) region in pRErm46 reveals that the intI1 open reading frame (ORF) extends into the 3= end of the convergent IS6100 sequence in the opposite strand and may code for a full-length integrase. The right IS6100 copy is next to a putative NTP-binding protein/transposon resolvase pseudogene at the end of orf5 of the 3= CS (Fig. 1B). This is likely the remnant of a Tn402-like transposon typically carried by "preclinical" class 1 integrons, aka the tniΔ module (49,50), which IS6100 presumably functionally replaces.
The third IS6100 copy in pRErm46 is at the end of an additional module that extends to the right of the integron and harbors a putative tetR-tetA tetracycline efflux determinant. An identical IS6100-flanked module (BLASTN E value, 0.0; Ͼ99.9% identity) is found in pTET3 at the opposite (intI1) side of the integron core (Fig. 1B), in pLEW279a, and in Trueperella (Corynebacterium) pyogenes chromosomal DNA, suggesting that it is independently mobilizable. IS6100 transposition has been experimentally demonstrated in Actinobacteria, including Mycobacterium (46,51), Streptomyces (52), and Corynebacterium (44). Due to the replicative transposition mechanism of IS6-like elements, involving cointegrate resolution by homologous recombination (47), directly repeated flanking IS copies are formed favoring local rearrangements (48,49). Together with transposition itself, such recombination-driven rearrangements may explain the different location of the tetR-tetA module in the otherwise identical class 1 integrons from pRErm46 and pTET3.
The pRErm46 plasmid sequence of the PAM 2350 transconjugant (originally from the MLS r equine isolate PAM 2287, which we establish as the reference sequence for pRErm46; GenBank accession no. KY494640) carries in addition three copies of the 6.9-kb erm(46) element (Fig. 3).
TnRErm46 transposon. The TnRErm46 transposon is 6,919 bp in length and comprises seven ORFs encoding, in this order, a hypothetical protein, a TetR family transcriptional regulator, a 25-kDa Gap-like (TauE/SafE superfamily) membrane transporter, an ABC ATPase with a duplicated NTP-binding domain, a putative nucleotide kinase with an AAA ATPase motif, the erm(46) 23S rRNA methyltransferase, and an IS481 family transposase ( Fig. 1A and Table S1). Except for the hypothetical protein, homology searches suggest an actinobacterial origin, specifically rhodococcal. Maximum identity values of 82 to 89% with its closest homologs (from Nocardia/Rhodococcus spp.) indicate that the transposase is a novel IS481 member, which we named ISRe46.
IS481 family members are widely distributed in bacteria and may be associated, as in this case, with cargo genes such as antibiotic resistance determinants and regulators, constituting self-mobilizable transposons (53). IS481 transposases belong to the Asp-Asp-Glu (DDE) catalytic motif type and are thought to operate through a copy-paste mechanism (53). The erm(46) element was confirmed as an ISRe46-driven transposable unit by the presence in all its multiple insertion sites (see below) of the directly repeated (DR) hexanucleotide NCTAGN, invariably followed inwards by the inverted repeat (IR) sequence 5=-TGTTGCGGGTCATGACGTTG-3= (Fig. 1A). Analysis of the insertion sites in the R. equi genome (Fig. 4) indicates that CTAG is the target DNA sequence for transposition, of which the TAG triplet in the right flanking DR provides the UAG stop codon for the transposase (Fig. 1A). We named this novel macrolide resistance transposable element TnRErm46.
One of the three TnRErm46 copies in pRErm46 from PAM 2350/2287 is inserted into the integron's passenger gene (aadA9) (Fig. 1). The two other copies are found in a 16.7-kb region between positions 32562 at the 3= end of the parB gene and 49273 before parA in pRErm46's backbone ( Fig. 3 and Table S1). In the left copy the ISRe46 element is a pseudogene due to two frameshift mutations. The second transposition appears to have involved a duplication of the three left flanking genes from the plasmid backbone (pRErm46 CDS_0440, pRErm46 CDS_04450, and parB), which as a result are directly repeated between the two transposon copies (Fig. 3 and Table S1). In both TnRErm46 copies the TAG of the left NCTAGN DR provides the stop codon for the . In parentheses is the number of instances in which a particular insertion was detected. In pRErm46, 16/16 corresponds to 100% of isolates in which pRErm46 sequences were detected in the genome assemblies. One of the 20 chromosomal insertion sites was in a resistant clone-specific region. The accessory genome accounts for Ϸ20% of an R. equi isolate's gene content (34).
Emerging MDR R. equi 2287 Clone ® repeated parB1 and parB2 genes (Fig. 3). Interestingly, these transposition events took place at the parA/B plasmid partitioning module, previously identified in the R. equi virulence plasmids and related rhodococcal replicons as a hot spot for integration of foreign DNA (8,13).
Pervasive genomic colonization by TnRErm46. The multiple copies of TnRErm46 in pRErm46 (PAM 2350/2287) suggested that it is an actively transposable element. To gain insight into this issue, we searched for the TnRErm46 sequences in the WGS assemblies of the 18 MLS r isolates. Contigs with 100% matching nucleotide sequences to either of the TnRErm46 ends were aligned with BLASTN against the R. equi chromosome and the reference sequences of the equine-type virulence plasmid pVAPA1037 (GenBank accession number AM947677) (8) and resistance plasmid pRErm46. To pinpoint the chromosomal insertions, we used a draft SMRT sequence of PAM 2287 (reference isolate of the MLS r clone; see below) and the reference (complete) R. equi 103S genome sequence (21) for coordinate assignment. Figure 4 shows the TnRErm46 insertions identified for all the MLS r clinical isolates. Transposition into the bacterial chromosome was detected in all but one of the isolates (PAM 2275), up to five times in some cases. A total of 35 insertions at 20 different chromosomal sites were identified. In addition, TnRErm46 was found in the pVAPA virulence plasmid in four isolates (one copy in PAM 2283, 2284, and 2287 and four copies in PAM 2286). Except nucleotide position 55131 shared by pVAPA2286 and pVAPA2287, the element was inserted at different pVAPA locations ( Fig. 4 and Table S2). Again, it is noteworthy that four of the seven pVAPA insertions were in the plamid's replication/partitioning module. This region was previously identified in pVAPA and related circular rhodococcal plasmids as the integration site for the horizontal gene transfer (HGT) DNA which constitutes the plasmid-specific, niche-adaptive variable region (VR) (8,13). In pVAPA, the VR corresponds to the horizontally acquired vap pathogenicity island (PAI), which itself provided one of the insertion sites for TnRErm46. Consistent with the clinical source of the bacteria and the requirement of a functional virulence plasmid/vap PAI for host colonization (54,55), none of the transposon insertions, including the one within the theoretical boundaries of the vap PAI ( Fig. 4) (at nt position 59315 targeting the stop codon of a pseudogene), are likely to affect the expression of critical pVAPA genes, specifically virulence-associated determinants.
Finally, all pRErm46 plasmids harbored at least two TnRErm46 copies in a combination of common and unique locations. The only exception is PAM 2275, with a single transposon insertion at position 32567. This specific insertion is shared by all pRErm46 plasmids analyzed (Table S2), suggesting that it corresponds to the original transpositional event through which the pre-pRErm46 replicon acquired the erm(46) determinant (Fig. 5). Of note, the TnRErm46 insertion within the aadA9 gene found in pRErm46 from PAM 2350/2287 ( Fig. 1 and 3) is unique to this isolate. However, TnRErm46 was found within the integron's 3= CS in three other isolates, targeting sul1 (once) and orf5 (three different sites) (Table S2). Of a total of 11 identified insertion sites in the pRErm46 plasmids, five were within the class 1 integron (Fig. 4), consistent with bacterial integrons serving as flexible recruitment platforms for resistance-encoding DNA and transposable elements (43).
For three isolates, PAM 2280, 2285, and 2289, class 1 integron sequences were not found in the WGS assemblies despite the fact that the isolates carried the TnRErm46 element. However, pRErm46 (including the class 1 integron) was detected by PCR and BLASTN mapping of the TnRErm46 insertion sites in PAM 2289 and by PCR in PAM 2280 (Table S2), suggesting that perhaps the resistance plasmid was present in only a small fraction of these isolates' populations. PAM 2285, in contrast, appears to have actually lost the pRErm46 plasmid and to carry TnRErm46 only on the chromosome (Table S2). This would explain the inability of this isolate to transfer the erm(46) element/MLS r phenotype in mating experiments.
Overall, the data reveal a pattern of active, widespread dissemination of TnRErm46 across the host R. equi genome. This includes the extrachromosomal replicons, i.e., the pRErm46 resistance plasmid and the pVAPA virulence plasmid, both of which are conjugally transferable. Conjointly, all these mechanisms ensure the stable maintenance of the newly acquired resistance element in R. equi (Fig. 5).
Cotransfer of macrolide resistance and virulence determinants. In mating experiments with PAM 2287, 74% of the erm(46)-positive (erythromycin resistant [Erm r ]) 103Stransconjugants had also acquired the virulence plasmid. To further document that pVAPA can transfer erm (46), we examined one such erm(46)/pVAPA-positive transconjugant, PAM 2351 (see above), by SMRT sequencing. The sequence analyses confirmed that pVAPA from PAM 2351 carried TnRErm46 inserted at position 55132 (3= end of parB) as in the donor PAM 2287, consistent with the transposon having been conjugally transferred in situ within pVAPA. Notably, this insertion was accompanied by three additional copies of TnRErm46 in the virulence plasmid (Fig. S1). These data  Table S2).
Emerging MDR R. equi 2287 Clone ® confirm that the TnRErm46 element can actively transpose within a relatively short time frame. They also confirm that the virulence plasmid, essential for host colonization, may serve itself as a spread vector for the macrolide resistance determinant erm (46).
Clonal spread of pRErm46/TnRErm46. R. equi isolates from equine settings or clinical specimens are genotypically diverse not only at a global population level (34) but also within the same farm (56) or even the same animal (56,57). The various conjugal exchange routes for TnRErm46, via either pRErm46, pVAPA, or both, hence would predict a situation of rapid spread across a diversity of R. equi strains. To investigate this, we performed a phylogenomic analysis of the 18 erm(46)-positive isolates available for this study and 27 control MLS s strains. The resistant isolates were all obtained from different US states over a period of a decade (34) while the control susceptible set comprised six coetaneous random equine isolates from the same geographic areas (Table S2) plus 22 strains representative of the global genomic diversity of R. equi (34) (Table S3). The latter included the type strain DSM 20307 (ϭATCC 6939) and 103S (ϭ DSM 104936 ϭ NCTC 13926) (21) used as the reference genome. Figure 6 shows a core genome phylogeny based on a Parsnp alignment (58). While the control equine isolates were randomly distributed across the R. equi population structure, all 18 TnRErm46-harboring isolates clustered together within a distinct clade at the top of the tree. In a context where all members of the R. equi species radiate at a short genetic distance from each other (34), the macrolide-resistant group differed from the nearest control isolates by Ϸ25,000 to 27,000 single nucleotide polymorphisms (SNPs) (over Ϸ3,800 core genes/4.2 Mbp). In contrast, the resistant group differed only by an average of 60 (43 to 102) SNPs in pairwise comparisons, indicating that it corresponds to a recently emerged R. equi clone (Fig. S2).
The erm(46)-positive R. equi bacteria also exhibited rifampin resistance (Rif r ) (MIC 90 Ͼ32 g/ml, versus 0.25 g/ml for MLS s equine isolates) (32,59). Analysis of the rpoB sequences revealed in all of them the same missense mutation at codon 443 resulting in a Ser¡Phe substitution. None of the susceptible isolates carried any rpoB mutation. The location corresponds to Escherichia coli RpoB position 531 within rifampin resistance-determining region 1 (RRDR-1; residues 507 to 533), where the same substitution is known to cause the Rif r phenotype (60). Different RRDR-1 Rif r mutations have been previously reported for R. equi, including Ser531Leu and Ser531Trp (E. coli numbering), associated with MICs of 8 g/ml and Ͼ64 g/ml, respectively (19,(28)(29)(30). Like Ser531Trp, the Ser531Phe rpoB mutation, which had never been described before for R. equi, introduces a large aromatic side chain at position 531 and is also associated with a relatively high level of rifampin resistance.
Our data thus indicate that the emerging macrolide resistance detected among R. equi equine isolates in the United States is due to a specific clone that acquired the TnRErm46 element and is characterized by a distinct Rif r mutation.
Fitness neutrality of pRErm46. We finally considered to what extent the clonal spread of erm(46)-mediated macrolide resistance could have been influenced by a specific adaptation of the pRErm46 plasmid to the host strain. To explore this, we tested whether pRErm46 acquisition entailed any fitness cost upon transfer to the 103S strain, which belongs to an early clonal bifurcation of R. equi (lineage I) different from that of the resistant clone (lineage II) (34). The analyses were performed with low-passagenumber PAM 2350 cultures stored after conjugation to minimize the potential impact of adaptive plasmid cost amelioration (61,62).
No significant differences in exponential growth rate () or maximum growth (A) were observed between pRErm46-bearing PAM 2350 and the plasmid-devoid 103S recipient in either complex media (brain heart infusion [BHI] and Luria-Bertani [LB] broth) or chemically defined medium (Fig. 7A and B). The two strains also showed no significant differences in competitive ability in vitro in the absence of antibiotic pressure (Fig. 7C). Thus, pRErm46 does not seem to impose significant fitness costs upon transfer to a different R. equi genomic background.

DISCUSSION
The rise of antimicrobial resistance (AMR) among leading human pathogens is often associated with specific clonal complexes that become globally disseminated (63). While phylogenomic studies have clearly linked the emergence of such successful clades to founder genetic events involved in the acquisition of resistance (63)(64)(65)(66)(67)(68), the drivers of the clonal expansion remain largely speculative. Our study captured one such founder event at the early stages of its evolution in a major animal pathogen, R. equi, providing valuable insight into the determinants of AMR clonality.

FIG 6
Clonal spread of TnRErm46-mediated macrolide resistance. Phylogenetic tree of 46 R. equi isolates based on core genome SNP analysis using Parsnp (58). The genomes analyzed are from 18 MLS r and 6 control MLS s equine isolates from different U.S. states (highlighted in red and blue, respectively) plus 22 isolates representative of the genomic diversity of the species, including the reference genome from strain 103S (DSM 104936 ϭ NCTC 13926; GenBank accession no. FN563149 [21]) and the type strain DSM 20307 (ϭ ATCC 6939; GenBank accession no. LWTX00000000 [34]). Numbers in the nodes indicate bootstrap values for 1,000 replicates. The tree graph was constructed with FigTree (http://tree.bio.ed.ac .uk/software/figtree/). See also Fig. S2.

Emerging MDR R. equi 2287 Clone
We show that the emerging erm(46)-mediated MLS r detected among equine R. equi isolates in the United States (32) is due to a specific clone that carries a mobile element and a chromosomal mutation conferring dual resistance to, respectively, macrolides and rifampin, the mainstay of antibiotic therapy against foal rhodococcosis since the 1980s (20,26). Diverse Rif r rpoB mutations have been previously reported for R. equi (19,(28)(29)(30), indicating that such chromosomal mutations occur regularly and alone are insufficient to support the clonal expansion of a specific strain. It seems likely that the conjugal acquisition of the hitherto undescribed pRErm46 plasmid and cognate erm(46)-carrying TnRErm46 transposon has been the decisive trigger in the emergence of the R. equi multidrug-resistant (MDR) clone.
Mobile resistance elements have been traced to the origin of major MDR clonal complexes (63). The elements often become stabilized in the host genome and evolve with the clone, with little evidence of active, short-term mobilization or lateral transferability. Examples include SCCmec in methicillin-resistant Staphylococcus aureus (MRSA) (69), the AbaR genomic island/composite transposon in Acinetobacter baumannii clonal complex GC1 (68,70), and the SG1 MDR island in Salmonella enterica serovar Typhimurium DT104 (71). On other occasions, exemplified by the plasmid-borne genes bla KPC in Klebsiella pneumoniae clonal group 258 and bla CTX-M-15 in Escherichia coli ST131 (66,67,72), the mobile resistance element promotes the clonal amplification while at the same time it exploits the dominance of their host clone to disseminate and drive the expansion of other diverse bacterial lineages/species (63).
In our case, the erm(46) macrolide resistance determinant is restricted to a specific R. equi clone yet its vehicle is a highly mobile, potentially actively transferable element. Moreover, transferability is maximized through co-option of the conjugative machinery of the pVAPA virulence plasmid (Fig. 5), physically linking two essential traits required for the colonization of a macrolide-rifampin-treated equine host. Furthermore, evidence indicates that pVAPA plasmids are actively exchanged between R. equi lineages (9,34), facilitated by alternation of spontaneous plasmid loss in the absence of host pressure (7,11) and high-frequency conjugative reacquisition (12). While these features theoretically anticipate the rapid spread of TnRErm46 across the array R. equi populations typically colonizing the equine farm environment (56,57), the MDR clone (dubbed 2287 after the isolate from which the pRErm46 reference sequence derives) has been circulating since at least 2002 (Table S2), until now with no apparent TnRErm46 spillover to other R. equi lineages.
How can the paradoxical clonal confinement of the otherwise highly mobile TnRErm46 element/pRErm46 plasmid be explained? Our data appear to exclude fitness constraints, because acquisition of the pRErm46 plasmid does not seem to entail significant costs in a different R. equi genomic background. In addition, any fitness cost would be largely compensated in antibiotic-treated animals in which resistance confers a selective advantage (73). We suggest that clonality is primarily determined by the inability of pRErm46/TnRErm46 to ensure alone the evolutionary viability of the host R. equi bacteria in the absence of a chromosomal rpoB mutation under combined macrolide-rifampin pressure. Under these conditions, the probability of concurrent acquisition of the pRErm46 plasmid and a Rif r mutation reoccurring de novo would be vanishingly small (as low as Ϸ10 Ϫ10/-14 , considering individual in vitro frequencies of, respectively, Ϸ10 Ϫ2/-6 [most often 10 Ϫ5 ] and Ϸ10 Ϫ8 [30; our unpublished observations]) compared to the likelihood of the existing R. equi 2287 clone spreading through horse trading or cross-exposure at racecourses.
Human-pathogenic MDR clones typically begin as localized expansions and rapidly-within a few years-disseminate pandemically and diversify in clonal complexes (63)(64)(65)(66)72). Despite the high prevalence of R. equi in horse breeding countries, after more than 15 years R. equi 2287 still appears to be at the initial stages of local dissemination. This apparently slow expansion arguably stems from the fewer opportunities for spread afforded by the significantly smaller size and restricted geographical mobility of the equine versus human populations. Bar the diversity of TnRErm46 insertion sites, R. equi 2287 isolates also show very limited genetic variability, presumably a reflection of the ecological stability of the equine niche to which the clone is restricted. We predict that, over time, R. equi 2287 will probably become globally disseminated, accumulate additional resistance mechanisms (63,64,68,72,74), and eventually enter a path of fast genome evolution fueled by the continued transposition and amplification of the ISRe46 element, as observed in Bordetella pertussis with the homologous insertion sequence IS481 (75). pRErm46 (or TnRErm46 carried by other mobile elements, e.g., bacteriophages [41]) is also likely to be horizontally transferred to other R. equi strains or equine-associated microbiota.
In addition to being a veterinary proxy of the nosocomial setting in terms of intensive antimicrobial use and close-contact transmission, the particularities of R. equi antibiotic control in equine farms, based on the same drug combination for decades, offer a unique "controlled" scenario that helps illuminating the factors underlying AMR clonal spread. Our findings illustrate that under sustained combination therapy, AMR requires the co-occurrence of independent founder events, thereby effectively reducing the likelihood of resistance emergence and expansion to a specific bacterial clone. They also highlight the impact of ecological bottlenecks imposed by antimicrobial use in shaping the population dynamics of pathogens (74) and the potential effectiveness of combination therapies in limiting not only the emergence (76,77) but also, critically, the spread of resistance-even if conferred by highly mobile elements.
While consubstantial to and underpinning AMR, clonal spread provides at the same time a predictable framework to develop control strategies. The pRErm46 plasmid and TnRErm46 element here reported, together with the identification of the Rif r 2287 lineage harboring these, lay the foundations for the effective tracking and control of emerging macrolide-rifampin multiresistance among equine R. equi isolates and its potential zoonotic transmission to humans.

MATERIALS AND METHODS
Bacteria, culture conditions, and reagents. The 18 MLS r and 6 control MLS s isolates analyzed in this study were selected from a collection of R. equi cultures obtained from tracheobronchial aspirates or postmortem tissue of infected foals from different U.S. states between 2002 and 2013 (32) (Table S2). WGS shotgun assemblies thereof were previously used in the identification of the erm (46) gene (32). WGS assemblies from 22 additional control isolates representative of the global diversity of R. equi have been previously described (34). The presence of the virulence plasmid and pRErm46 plasmid/TnRErm46 element was routinely checked by PCR using specific oligonucleotide primers (Table S4). R. equi was grown in brain heart infusion medium (BHI; Difco-BD) at 30°C unless otherwise stated, with orbital shaking (200 rpm) for fluid cultures. Bacteriological agar (Oxoid) was used at 1.6% (wt/vol) for plate cultures. Antibiotic supplements were added as required after autoclaving. Chemicals and oligonucleotide primers were purchased from Sigma-Aldrich unless otherwise stated.
Conjugal transfer of pRErm46. Mating experiments were carried out as described by Anastasi et al. (32) using a 1:1 donor to recipient ratio. A virulence plasmid (pVAPA)-cured derivative of the genome strain 103S (21) with a chromosomal hygromycin resistance marker was used as a recipient for counterselection of the wild-type donor isolate (designated 103S -Hyg r ; internal collection no. PAM 2266). After incubation at 30°C for 72 h, transconjugants were selected on BHI agar supplemented with 200 g/ml of hygromycin B and 10 g/ml of erythromycin. pRErm46 transfer was confirmed by PCR analysis of at least 30 Hyg r Erm r colonies per experiment using ad hoc primers (Table S4). Transfer frequencies were calculated by dividing the number of Hyg r Erm r bacteria by the total number of recipient bacteria.
DNA techniques. Plasmid DNA for Southern blotting was extracted from exponential BHI cultures (optical density at 600 nm [OD 600 ] Ϸ1.0) using an alkaline precipitation protocol as previously described (7). Total and plasmid-enriched DNA for sequencing was obtained using the GenElute bacterial genomic DNA kit (Sigma-Aldrich) and QIAfilter midikit (Qiagen), respectively. Total DNA for routine PCR tests was prepared by heating single bacterial colonies at 100°C in 100 l of ultrapure water and centrifugation for 90 s at 16,000 ϫ g. PCR was performed as previously described (13). Oligonucleotide primers used are shown in Table S4. For Southern blotting, plasmid preparations were electrophoresed in 0.5% agarose and transferred to positively charged nylon membranes (Roche) after treatment of the gels with denaturing solution (1.5 M NaCl 0.5 N NaOH) for 30 min and neutralizing solution (1.5 M NaCl, 0.5 M Tris-HCl pH 7.5) twice for 20 min. Probes consisted of digoxigenin-labeled internal fragments of the genes of interest [vapA, erm (46), and IS6100 transposase] generated with the PCR DIG probe synthesis kit (Roche) using suitable oligonucleotide primers (Table S4). Membranes were hybridized at high stringency and developed according to the manufacturer's instructions. Hybridized membranes were stripped with 0.2 N NaOH-0.1% SDS at 45°C for 30 min before reprobing.
DNA sequencing and analysis. Illumina shotgun libraries of 24 R. equi equine clinical isolates (18 MLS r and six MLS s ) obtained by Anastasi et al. (32) were used in this study. Genomes were assembled de novo as previously described (34) or using a modified protocol with Cutadapt software for adapter removal (78) and QUAST for quality assessment (79). SMRT PacBio genome sequencing of PAM 2350 and 2351 transconjugants was performed by BGI Genomics (Hong Kong) and BaseClear sequencing services (The Netherlands), respectively. Minimap and Miniasm (de novo assembler) software was used for mapping and de novo assembling single SMRT data (80), and Racon for consensus calling (81). Circlator (35) was run with error corrected reads from the Canu v1.3 error correction module for assembly circularization to obtain an accurate linear representation of circular sequences (82). Annotation of the assemblies was performed using Prokka V1.11 (83) and InterProScan v5.17. pRErm46 annotations were manually inspected and curated based on Blast analysis. Whole-genome phylogenies were inferred using Parsnp in the Harvest suite, designed for SNP analysis between closely related species/strains (Ն97% average nucleotide identity [ANI]) (58). Parsnp uses FastTree 2 (84) to build approximate maximum likelihood trees from core-genome SNPs.
Fitness assays. Effects on bacterial fitness were investigated by monitoring the growth and competitive ability in two complex media (BHI and LB) and in chemically defined medium (modified R. equi minimal medium [mREMM]) (34) supplemented with two rapidly metabolizable carbon sources (50 mM sodium lactate and 25 mM sodium acetate), as previously described (85). Briefly, exponentialphase BHI cultures were washed in phosphate-buffered saline (PBS), resuspended in fresh medium to give an OD 600 of Ϸ0.05, and triplicate 400-l aliquots of the suspension transferred to 48-well plates. Bacterial growth was monitored at 30°C every 30 min using an automated plate reader (Optima apparatus, BMG). The maximum growth rate () and maximum bacterial cell density (A) were estimated from spline-fits of OD 600 values using the Grofit package in R (86) as previously described (85). Pairwise competition experiments were carried out as described above with a bacterial inoculum consisting of a 1:1 mix of the two test bacteria. At specific time points, cultures were sampled and appropriate dilutions spread on LB agar supplemented with suitable antibiotics for differential counting. The competitive index was calculated using the formula CI ϭ (test/reference log CFU ratio at t ϭ n)/(test/reference log CFU ratio at t ϭ 0) (13,85).
Statistics. Growth parameters were analyzed using one-way analysis of variance (ANOVA) followed by Šidák post hoc multiple-comparison tests unless otherwise stated. One᎑sample Student t᎑tests were used to determine if competition index (CI) values differed significantly from 1 (the theoretical CI value if the ratio of the competing strains remains the same with respect to that at time zero). Statistical analyses were performed using Prism 8.0.2 software (GraphPad Software Inc., San Diego, CA).

ACKNOWLEDGMENTS
This article is dedicated to the memory of our collaborator Steeve Giguère, equine clinician and R. equi researcher, tragically deceased in 2018.
We thank J. Navas for the 103S -Hyg r strain, F. de la Cruz for confirmation of conjugative relaxase family, J. L. Martínez and T. M. Coque for stimulating discussions, and A. Wawsczyk for performing preliminary fitness experiments as part of her undergraduate project in our laboratory.