Genome-encoded ABCF factors implicated in intrinsic antibiotic resistance in Gram-positive bacteria: VmlR2, Ard1 and CplR

Abstract Genome-encoded antibiotic resistance (ARE) ATP-binding cassette (ABC) proteins of the F subfamily (ARE-ABCFs) mediate intrinsic resistance in diverse Gram-positive bacteria. The diversity of chromosomally-encoded ARE-ABCFs is far from being fully experimentally explored. Here we characterise phylogenetically diverse genome-encoded ABCFs from Actinomycetia (Ard1 from Streptomyces capreolus, producer of the nucleoside antibiotic A201A), Bacilli (VmlR2 from soil bacterium Neobacillus vireti) and Clostridia (CplR from Clostridium perfringens, Clostridium sporogenes and Clostridioides difficile). We demonstrate that Ard1 is a narrow spectrum ARE-ABCF that specifically mediates self-resistance against nucleoside antibiotics. The single-particle cryo-EM structure of a VmlR2-ribosome complex allows us to rationalise the resistance spectrum of this ARE-ABCF that is equipped with an unusually long antibiotic resistance determinant (ARD) subdomain. We show that CplR contributes to intrinsic pleuromutilin, lincosamide and streptogramin A resistance in Clostridioides, and demonstrate that C. difficile CplR (CDIF630_02847) synergises with the transposon-encoded 23S ribosomal RNA methyltransferase Erm to grant high levels of antibiotic resistance to the C. difficile 630 clinical isolate. Finally, assisted by uORF4u, our novel tool for detection of upstream open reading frames, we dissect the translational attenuation mechanism that controls the induction of cplR expression upon an antibiotic challenge.

ABCF ATPases share a common molecular architecture with two ABC nucleotide-binding domains NBD1 and NBD2 connected by an interdomain linker element which is variable in both amino acid sequence and length (24). This linker is referred to as the antibiotic resistance determinant (ARD) in the case of ARE-ABCFs (34,41,42) and as the P-site tRNA-interaction motif (PtIM) in the case of housekeeping ABCFs (21,22). The ARD is crucial for displacement of the antibiotic by ARE-ABCFs, and its deletion renders the resistance determinant inactive (24). Macrolide-resisting factors such as ARE1 MsrE have the longest ARDs which extend into the ribosomal NPET (41). ARDs of PLS A -resisting ARE-ABCF are shorter, as they make contact with the PTC-bound antibiotic and/or induce conformational changes in the PTC that promote antibiotic dissociation (26,34,43). PhO-resisting PoxtA and Op-trA have even shorter ARDs, which act by perturbing the positioning of the CCA-end of the P-site tRNA, rather than directly distorting the PTC (33). Finally, the P-site tRNAcontacting PtIM elements of the housekeeping ABCFs are in general shorter than the ARD elements of ARE-ABCFs (21)(22)(23).
Antibiotic resistance mediated by ARE-ABCF can synergise with other mechanisms, importantly with resistance mediated by 23S rRNA methylation. This rRNA modification is installed by clinically relevant and widely-spread Erm (Erythromycin resistance methylase) and Cfr (Chloramphenicol and florfenicol resistance) rRNA methyltransferases and confers resistance to macrolide, PhO and PLS A antibiotic classes (44)(45)(46)(47). Recently we have found that B. subtilis ARE-ABCF VmlR can synergise with Cfr to grant significant resistance to lincosamides, including the highly potent fully-synthetic antibiotic iboxamycin (48). The cfr gene is often encoded next to lsa ABCF in bacterial pathogens such as Clostridioides difficile (formerly Clostridium difficile) (49) and staphylococci (50), which is supportive of the two resistance determinants cooperating to grant high levels of protection. However, synergy of ARE-ABCFs with Cfr/Erm has not as yet been systematically tested experimentally. Importantly, Erm-mediated resistance is associated with an increased risk of C. difficile infection (CDI), and the risk further increases upon clindamycin use (51). Neither the role of ARE-ABCFs in C. difficile resistance, nor potential ARE-ABCF-Erm synergy, has been studied.
To further understand the molecular mechanisms underlying intrinsic antibiotic resistance in Gram-positive bacteria, we have characterised four genome-encoded ARE-ABCF factors. First, we show that well-studied B. subtilis ARE2 VmlR mediates resistance to A201A and hygromycin A, which expands the spectrum of ABCF-mediated resistance to nucleoside antibiotics. Second, we characterise the resistance spectrum of a VmlR representative with an unusually long ARD encoded in the genome of soil bacterium Neobacillus vireti (formerly Bacillus vireti)--VmlR2. We rationalise the antibiotic sensitivity data by solving the cryo-EM structure of the VmlR2-70S complex and probe its mechanism through mutagenesis. Third, we demonstrate that actinobacterial AAF1 Ard1 encoded by the A201A producer Streptomyces capreolus (52,53) is a narrow spectrum ARE-ABCF that specifically mediates selfresistance against nucleoside antibiotics. Fourth, we establish Clostridial pleuromutilin and lincosamide resistance protein (CplR) as an ARE1 ABCF resistance factor. We characterise CplR of the important human pathogen C. difficile (originally annotated as a multidrug resistance (MDR)-type ABC transporter CDIF630 02847 (54)), of Clostridium perfringens, a common causative agent of food poisoning, as well as of the mutualistic bacterium Clostridium sporogenes. We demonstrate that CplR contributes to intrinsic resistance of Clostridioides to lincosamides lincomycin, clindamycin and iboxamycin as well as pleuromutilin retapamulin. While lincosamide resistance mediated by CplR can be overcome by the fully synthetic lincosamide iboxamycin (12), ARE1 CplR provides high levels of iboxamycin resistance when acting together with transposon-encoded ErmB present in clinically isolated C. difficile strain 630 (55). Given that clindamycin treatment is associated with increased risk of CDI (51,(56)(57)(58)(59), our discovery and characterisation of CplR-mediated lincosamide resistance mechanism in C. difficile is of direct clinical importance.

Bacterial culture
C. difficile 630 erm and its derivatives were grown anaerobically in Brain-Heart Infusion (Difco) supplemented with 0.5% yeast extract and 0.1% cysteine (BHIS) or BHIS agar under 80% N 2 , 20% CO 2 , 4% H 2 atmosphere in the Coy anaerobic chamber (COY). C. sporogenes and its derivatives were anaerobically grown in Gifu anaerobic medium (GAM) (Nissui Co., Japan) in the Coy anaerobic chamber. C. perfringens and its derivatives were grown at 37 • C in GAM under anaerobic conditions using Anaeropack system (Mitsubishi Gas Chemical Co. Inc., Tokyo, Japan). B. subtilis were routinely aerobically cultured in LB broth. When necessary, antibiotics were supplemented in the media: 20 g/ml erythromycin and 10 g/ml thiamphenicol in the case of C. difficile and 20 g/ml chloramphenicol in the case of C. perfringens.

Strain and plasmid construction
The plasmid and oligonucleotides used in this study are listed in Supplementary Table 1, respectively. To construct C. difficile cplR (CDIF630 02847) deletion mutant we constructed the plasmid for the allele exchange mutagenesis as previously reported by Peltier and colleagues (67). The CD2517.1 type I toxin gene under the control of the tetracycline-inducible promoter as well as ≈1 kb-long 5 and 3 flanking regions of the target gene were cloned into pMTL83151 plasmid (CHAIN biotech, Nottingham, UK), yielding the pMSRNO plasmid. Using E. coli HB101/pRK24, the pMSRNO plasmid was then introduced into C. difficile by conjugation. pRK24 (74) was a gift from Farren Isaacs (Addgene plasmid # 51950). Briefly, 500 l of overnight cultured C. difficile were heated at 52 • C for 2 min. Meanwhile, 1 ml culture of E. coli HB101/pRK24 harbouring pNO201 were centrifuged and pelleted at 1500 × g for 3 min. The E. coli pellet was suspended to 200 l of the heated C. difficile culture in the anaerobic chamber. The mixture was spotted onto a BHI agar plate and incubated for 8 h. The mixed colonies were resuspended in 1 ml BHI broth, and then spread onto BHIS plates supplemented with 10 g/ml thiamphenicol, 250 g/ml D-cycloserine and 50 g/ml kanamycin (BHIS-TCK). After incubation for two to three days, individual colonies were restreaked on BHIS-TCK for isolation. The protocol for allele exchange was based on that of Peltier and colleagues (67). The deletion of the target gene was confirmed by PCR amplification.
C. sporogenes mutants were constructed using the Targetron system (Merck). A DNA fragment containing group II intron and ltrA gene from pJIR750ai (Merck) as well as a Clostridium-E. coli shuttle vector pMTL83153 (CHAIN biotech, Nottingham, UK) containing a constitutive promoter P fdx were PCR-amplified, and the intron-containing plasmid was produced by ligation of these DNA fragments pre-digested with HindIII and XhoI. Then, a PCRamplified C. sporogenes cplR-targeted intron was introduced into HindIII-BsrGI site in the plasmid. The resulting pNO201 plasmid was used for cplR mutant construction. The pNO201 plasmid was introduced into C. sporogenes by conjugation as described above, with one modification: the heat treatment prior to conjugation was omitted. While this step is crucial for obtaining transconjugants with C. difficile, in the case of C. sporogenes transconjugants are easily obtained without heat shock. To screen the mutant in which the target gene was disrupted, the colonies were restreaked onto BHIS containing 20 g/ml erythromycin. The insertion of the intron in the appropriate locus was verified by PCR amplification.
The C. perfringens mutant was constructed as described previously (75,76). Briefly, ≈1 kb of the 5 and 3 flanking regions of the target gene were PCR-amplified, and the resultant fragments were cloned into pCM-GALK using the In-Fusion cloning system (Takara Bio, Japan). Resulting plasmid was introduced into C. perfringens HN13 via electroporation, and the desired mutant was isolated on agar plates supplemented with 20 g/ml chloramphenicol or 3% galactose. The deletion of the target gene was confirmed by PCR and DNA sequencing.
For complementation of cplR genes in C. difficile and C. perfringens, cplR-expressing plasmids were constructed. The C. difficile cplR-expressing plasmid was based on pRPF185 plasmid which contains a tetracycline-inducible promoter (77). The pRPF185 was a gift from Robert Fagan and Neil Fairweather (Addgene plasmid # 106367; http://n2t.net/addgene:106367;RRID: Addgene 106367). A PCR-amplified cplR CDS was cloned into pRFP185 using the In-Fusion cloning system. For complementation of C. perfringens cplR, the DNA fragment containing a native promoter and CDS of C. perfringens cplR gene was amplified by PCR and cloned into pJIR750 (78).
An mCherry2-L variant codon-optimized for low-GC bacteria was used to construct the reporter constructs used to assay the regulation of cplR expression (79). The DNA fragment encoding a codon-optimized mCherry2-L CDS was synthesized at Eurofin Genomics (Tokyo, Japan). DNA fragments containing cplR promoter and 5 leader region were amplified and ligated with the mCherry2-L by PCR. The resulting DNA was introduced into pMTL84151 (CHAIN biotech) using the In-Fusion cloning system.
The B. subtilis strains were derivatives of wt168 (trpC2). The strains were constructed through homologous recombination by transformation with plasmids listed in Supplementary Table 1. The plasmids were constructed by standard cloning methods including PCR and Gibson assembly using oligonucleotides listed in Supplementary Table 1. Successful integration of a gene into the chromosome was accomplished by double crossing-over at the target loci. The resulting recombinant clones were checked for their antibiotic resistance markers, including the absence of those originally present on the plasmid backbone, as well as for inactivation of the thrC target locus.

Antibiotic susceptibility testing
The Minimum Inhibitory Concentrations (MIC) were calculated according to guidelines from the European Committee on Antimicrobial Susceptibility Testing (EUCAST) (http://www.eucast.org/ast of bacteria/ mic determination). Bacterial cells were cultured in 96-well plates in medium supplemented with 2-fold serial dilutions of the antibiotics. BHIS, GAM, or LB media was used to grow C. difficile, C. perfringens/C. sporogenes and B. subtilis, respectively. The 200 l cultures were inoculated at the concentration of 5 × 10 5 CFU/ml and grown in the 96-well plates. The MIC was determined as the lowest concentration of antibiotics in which no bacterial growth was observed after incubation for 24 h at 37 • C. The reproducibility was confirmed by at least three individual experiments.

Preparation of cryo-EM grids
Elutions from pull-downs were kept on ice until being applied within two h to glow discharged cryo-grids (Quantifoil 2/2 Cu 300 coated with 2 nm continuous carbon). 3.5 l of sample was loaded on grids in Vitrobot (FEI) under conditions of 100% humidity at 4 • C, blotted for 5 s and vitrified by plunge-freezing in liquid ethane. Samples were imaged on a Titan Krios (FEI) operated at 300 kV at a nominal magnification of 165 000× (0.82Å/pixel) with a Gatan K2 Summit camera and a BioQuantum energy filter with a slit width of 20 eV. The exposure rate was 5.277 electrons/pixel/s with a 4 second exposure and 20 frames using the EPU software.

Cryo-EM data analysis
MotionCor2 was used to correct for beam-induced motion in 6384 starting micrographs (81). Contrast transfer functions were estimated with Gctf (82). Subsequent processing was performed in RELION 3.1 unless specified otherwise (83,84). Micrographs with MaxRes >4Å and/or a CtfFigureOfMerit <0.05 were discarded, resulting in 5014 micrographs used for further processing. From the resulting set of 5 014 micrographs, 279 082 particles were picked with Gautomatch [https://www2.mrc-lmb.cam.ac. uk/download/gautomatch-056] and extracted with a pixel size of 2.46Å. 2D classification was used to remove nonribosomal particles, resulting in 193 343 particles which were subjected to 3D auto refinement. The initial reference, which was low-pass filtered to 60Å, was a B. subtilis 70S ribosome that contained a P-tRNA and no factor in the E site (EMD-0176, (34)). The output volume was used as a reference for subsequent 3D refinements. 3D classification with four classes and without angular sampling was then performed, resulting in two classes of interest that contained density in the E site. The best-resolved of the two classes, class 2, was used for most analyses, on the grounds that this class had the most continuous and feature-rich density for the protein in the E site. Particles from class 2 were re-extracted at the original pixel size and used for 3D autorefinement, and then anisotropic magnification, per-particle defocus, per-micrograph astigmatism, beam tilt, and higher-order aberrations were refined before final 3D autorefinement and post-processing (84).
A combined volume, containing particles from both classes with VmlR2 density in the E site, was also reextracted as above and partial signal subtraction followed by 3D classification around the A and E sites was performed. For the A-site classification, three classes with 29.1%, 37.6% and 33.3% occupancy were observed. The first and last of these classes contained a tRNA, each in a slightly different conformation, while the second class had no density in the A site, consistent with the poorly resolved density for the A tRNA in the parent map. Each of these classes was reverted to full density and used for 3D autorefinement. Partial signal subtraction and 3D classification around the E site and L1 stalk yielded three maps that differed mostly in the conformation of the L1 stalk, consistent with the poor resolution of this region in the parent map. However, density for VmlR2 was not improved in any of these classes, so class 2 from the initial 3D classification was used for further analyses.
For post-processing, soft solvent masks and a B factor automatically estimated by Guinier analysis were used. See also Supplementary Figure S2 for an overview of processing and Supplementary Table 2 for collection and processing statistics.

Model building and refinement
Starting models for B. subtilis ribosomal proteins and VmlR2 were obtained from the AlphaFold2 database (85). Ribosomal RNA was taken from PDB 6HA1 (34). Models were first adjusted using Coot (86) into a high-resolution B. subtilis 70S map (to be described elsewhere). The resulting model was then fitted in the VmlR2-70S map and adjusted manually with Coot. The starting model of VmlR2 was fitted domain-by-domain and adjusted by hand. A prediction created by ModelAngelo was also used to guide modelling of VmlR2 in places (87). For the P-site tRNA and 23S rRNA of the L1 stalk, PDBs 7NHK and 7NHL were used as starting points, respectively (26). Density for the uL1 protein and the CCA-3 end nucleotides of the Psite tRNA was very unclear and these regions were therefore omitted from the model. PDB 7K00 was used as a reference to guide modelling, especially around 23S rRNA helix 69 (88). The resulting model was refined first with REFMAC5/Servalcat (89,90), then with Phenix using the starting model as a reference for restraints to decrease bond length and angle root-mean-square deviations (91). MolProbity through Phenix was used for model validation (92

RNA extraction and RT-PCR
Overnight culture 630 erm C. difficile strain in BHIS was diluted 1:20 in fresh BHIS medium and incubated anaerobically for 2 h to reach the log phase. Then we added various concentrations of antibiotics to the culture. After incubation for 0.5-6 h, the cells were collected from 1 ml of the culture by centrifugation at 10 000 × g for 2 min and flash frozen in liquid nitrogen. Total RNAs were extracted as described previously (94) with minor modifications. Briefly, the cell pellets were flash-frozen in liquid nitrogen until use. The cells were resuspended in 500 l of LETS buffer (100 mM LiCl, 20 mM EDTA, 20 mM Tris-HCl pH 8.0, 1% SDS) and disrupted by addition of 500 l of glass beads and 500 l of phenol/chloroform/isoamylalchol followed by shaking for 5 min using a Shake Master NEO (Bio Medical Science, Tokyo, Japan). After centrifugation at 20 000 × g for 5 min, the aqueous layer was transferred to a fresh tube and mixed with equal volume of phenol/chloroform/isoamylalchol. To recover the total RNAs, the aqueous layer was mixed with 2.5 volumes of ethanol and 0.1 volumes of 3 M NaOAc, pH 5.2. The RNA was harvested by centrifugation at 20 000 × g for 10 min, washed with 70% ethanol, and contaminated DNA was removed by DNase I treatment (Promega). Purified RNA samples were resuspended in 50 l of RNase-free water and stored at −80 • C.
Reverse transcription was preformed with PrimeScript™ RT Master Mix kit (TaKaRa, Japan) using 10 or 0.5 ng (genomic or plasmid encoding genes, respectively) of total RNA. The cDNAs for the target genes were amplified using Tks Gflex™ DNA Polymerase (TaKaRa, Japan) and the primers used are listed in Supplementary  Table 1.

mCherry fluorescent reporter assays
C. difficile harbouring reporter plasmids were cultured in BHIS supplemented with 10 g/ml thiamphenicol. Cells were harvested from 1 ml cultures by centrifugation at 10 000 × g for 2 min and washed with phosphate buffer saline (PBS). The resultant cell pellets were fixed with 4% formaldehyde for 8 min, washed with PBS and resuspended in 1 ml of PBS. After exposure to oxygen for 90 min for chromophore maturation, fluorescence (ex/em = 579/616 nm) and absorbance (600 nm) of each sample were measured using a Synergy H1 microplate reader (BioTek, Vermont, USA). To calculate the relative promoter activity, relative fluorescence intensity normalised by O.D. 600 was further normalised by the signal from the constitutive P fdx -mCherry reporter.

The ABCF ARE1 subfamily member CplR is conserved in Clostridia
While horizontally mobile ARE-ABCFs that are encoded in mobile genetic elements or plasmids mediate acquired antibiotic resistance, chromosomally-encoded ARE-ABCFs provide intrinsic antibiotic resistance and are likely to be predominantly inherited vertically. The three Gram-positive ARE-ABCF representatives characterised here--CplR (standing for Clostridial pleuromutilin and lincosamide resistance; we propose this new name based on the resistance spectrum determined in the current study, see below), VmlR2 and Ard1--are of the latter, intrinsic type. Despite this commonality, they belong to very different subfamilies of the ABCF family tree ( Figure 1A). C. difficile CplR encoded by the gene CDIF630 02847 (RefSeq protein accession WP 011861613.1) is a member of the ARE1 subfamily, although it is phylogenetically distinct from the well-characterised mobile Vga-and Msr-like ARE1 proteins. CplR is found in a range of different Clostridia, including C. difficile, C. perfringens and C. sporogenes ( Figure  1A). The core CplR group is found in various clostridia genera: Terrisporobacter, Clostridium, Clostridoides and Paeniclostridia (Supplementary Figure S3). However, there is a large diversity of CplR close relatives in bacillota, for which it is hard to establish orthology versus paralogy. The ARE2 subfamily of ABCFs can be subdivided into two phylogenetically distinct types: one subtype has a short 'arm' subdomain, as is seen in VmlR from B. subtilis (34), while the second form has insertions in the arm and -sometimes -the ARD/linker element, as is the case with Neobacillus vireti VmlR2 ( Figure 1B, Supplementary Figure S4). The Ard1 ABCF from Streptomyces capreolus is classified as being in the actinobacterial subfamily AAF1, closely related to ARE4 and ARE5 subfamilies (24). Just as with C. difficile CplR ARE-ABCF (54), Ard1 was originally mis-annotated as a transporter (52). While ABCF proteins do carry the ATP-binding cassette domain that is also found in many transmembrane transporters, they lack the necessary transmembrane domains for such a role. Ard1 is encoded as part of the A201A antibiotic biosynthesis gene cluster, acting as a self-resistance determinant (53).

Antibiotic resistance spectra of S. capreolus AAF1 Ard1, N. vireti ARE2 VmlR2, C. dfficile ARE1 CplR expressed in B. subtilis vmlR surrogate host
To determine the spectra of antibiotic resistance conferred by N. vireti VmlR, S. capreolus Ard1, and C. difficile CplR, we first ectopically expressed these ARE-ABCFs in a B. subtilis surrogate host under the control of an IPTG-inducible P hy-spank promoter (95) and determined the Minimum Inhibitory Concentrations (MICs) for a comprehensive panel of translation-targeting antibiotics (Table 1). We used a B. subtilis strain lacking the VmlR ARE-ABCF factor since the vmlR strain is highly sensitive to PLS A antibiotics (28,31,34,48).
While the wild-type B. subtilis is virtually immune to A201A and hygromycin A (MICs in excess of 256 g/ml, beyond the concentration levels that are experimentally fea-sible given the paucity of these antibiotics), the vmlR strain is sensitive (MIC for A201A of 4 g/ml, MIC for hygromycin A of 16 g/ml) ( Table 1). This observation expands the spectrum of known VmlR-mediated resistance to nucleoside compounds. When B. subtilis VmlR is overexpressed in vmlR B. subtilis, the strain becomes PLS Aand nucleoside-resistant; no protective effect is detectable when an ATPase-deficient EQ 2 variant of VmlR is expressed ( Table 1). Expression of N. vireti VmlR2 confers the same PLS A and nucleoside resistance spectrum, despite the relatively long ARD, without any detectable resistance to macrolides ( Table 1). Expression of S. capreolus Ard1 in Streptomyces lividuns is known to confer resistance to A201 (52), however no systematic characterisation of the Ard1 resistance spectrum was performed. The P hy-spankdriven expression of wild-type S. capreolus Ard1--but not that of EQ 2 Ard1--confers vmlR B. subtilis moderate resistance to nucleoside--but not PLS A --antibiotics. The narrow spectrum of Ard1-mediated resistance suggests that Ard1 specifically confers self-resistance to A201A (Table  1). Finally, ectopic expression of C. difficile CplR confers vmlR B. subtilis resistance to PLS A antibiotics and has a moderate (2-to 4-fold increase in MIC) protective effect against nucleosides.
Collectively, these results (i) expand the resistance spectrum of ABCF-mediated resistance to nucleoside antibiotics, (ii) demonstrate that despite its relatively long ARD, N. vireti ARE2 VmlR has a PLS A and nucleoside resistance spectrum identical to that of the short-ARD B. subtilis ARE2 VmlR, (iii) establish S. capreolus Ard1 ARE-AAF1 as a dedicated narrow-spectrum nucleoside resistance factor and (iv) establish genome-encoded C. difficile ARE1 CplR as a PLS A resistance determinant.

Cryo-EM structure of N. vireti VmlR2-70S complex
Next, we aimed to determine cryo-electron microscopy structures of ribosome-bound ATPase-deficient (EQ 2 ) C. difficile ARE1 CplR and N. vireti ARE2 VmlR2. To generate the ribosome-bound complexes we employed our wellestablished affinity purification strategy (26,33). While the C. difficile CplR structure would provide the necessary structural insights into intrinsic PLS A and nucleoside antibiotic resistance of C. difficile pathogen, the N. vireti VmlR structure would be instructive for structural rationalisation of the VmlR2 N.vireti resistance spectrum that is identical to that of short-ARD B. subtilis VmlR. S. capreolus Ard1 has not been prioritised for structural studies for two reasons: (i) the lack of established genetic tools for S. capreolus and (ii) the levels of nucleoside antibiotic resistance conferred in B. subtilis surrogate host are modest, indicative of imperfect functioning of the resistance determinant on the B. subtilis 70S ribosome.
Unfortunately, our attempts to purify the FLAG 3 -tagged C. difficile CplR in complex with the ribosome were unsuccessful. As C-terminal FLAG 3 -tagging of C. difficile CplR abrogated the functionality of CplR in vmlR B. subtilis (Table 1; vmlR thrC::P hyspank -cplR Cd -FLAG 3 ), we did not pursue this direction further. However, we were successful in isolating a complex of EQ 2 N. vireti VmlR2-HTF with B. subtilis 70S ribosomes (Supplementary Figure S1)  Bootstrap Support values, in percentages. Only those 50% or higher are shown. CplR is a member of the ARE1 subfamily, but without any strongly supported specific association with Vga or Msr subtypes. There is strong support for Ard1 falling within the same clade as other actinomycete ARE subfamilies, all of which likely confer self-protection of the antibiotic producer. VmlR2 is a close relative of VmlR, but has distinctive sequence/structure features, including (B) an extended ARD/linker subdomain (see Supplementary Figure S4 for full alignment). munoprecipitated FLAG 3 -tagged EQ 2 VmlR2 was eluted and applied to a cryo-EM grid for imaging in a Titan Krios followed by single-particle analysis using RELION (83). Three-dimensional classification revealed one major class of particles (72.3% of particles after 2D classification) consisting of VmlR2-EQ 2 bound to the 70S ribosome (Supplementary Figure S2). This volume could be refined to an average resolution of 2.9Å (Supplementary Figure S5A, B). The VmlR2 NBDs and C-terminal extension (CTE) could be modelled by domain-wise fitting and adjustment of the AlphaFold2 model (Supplementary Figure S5C). While the density for the NBDs was not sufficient to draw conclusions about the precise geometry adopted by the catalytic region of the enzyme, the VmlR2 interdomain linker, consisting of two ␣ helices and a loop, and which extends into the core of the large ribosomal subunit, was well resolved and could be modelled de novo with high confidence (Supplementary Figure S5C). The VmlR2-70S structure is globally similar to other antibiotic resistance ABCF-70S complexes (26,33,34,41,43) (Figure 2A, B). VmlR2-EQ 2 is bound in the ribosomal E site with closed NBDs and density corresponding to two bound NTP molecules, presumably ATP (Figure 2A-C, Supplementary Figure S6A, B). The VmlR2 ␣-helical interdomain linker extends from the E site into the PTC, overlapping the canonical position of a P-site tRNA ( Figure 2D,E). This results in a distorted tRNA occupying the P site, with the tRNA elbow contacting the VmlR2 NBD2, pulled towards the E site, and the acceptor stem highly distorted. Poor density for the P-tRNA acceptor stem indicates a high degree of flexibility in this region, and in particular the CCA-3 end could not be modelled (Supplementary Figure S6C). As observed previously in ABCF samples prepared by ex vivo immunoprecipitation (26,33,34), the density of the P-site tRNA was most consistent with an initiator fMet-tRNA (Supplementary Figure S6C), and density for the P-site codon was most consistent with AUG (Supplementary Figure S6D). An apparent Shine-Dalgarno-anti-Shine-Dalgarno helix is positioned in the mRNA exit channel, close to, but not visibly contacting, the VmlR2 CTD (Supplementary Figure S6E). A sub-stoichiometric tRNA is present in the A site, as observed previously for the 70S-LsaA-EQ 2 complex from Enterococcus faecalis, and subclassification revealed that the A-site tRNA can adopt multiple conformations ( Figure 2B, Supplementary Figure S7). These observations imply that VmlR2-EQ 2 had bound to and stalled the 70S ribosome after translation initiation, but before the first translocation event, which is the same state stalled by PLS A and nucleoside antibiotics (7,96).
As expected from the high degree of sequence homology, the structure of VmlR2 bound to the 70S ribosome is globally similar to B. subtilis VmlR bound to the 70S, with a root square mean deviation (RMSD) of 1.44Å between the two factors after least-squares fitting. In particular, the ␣- helices of the well-resolved interdomain linker, which travel through the path normally occupied by the P-tRNA to reach into the PTC, are highly structurally conserved (Figure 2D, E). The ends of the ␣-helices and the ARD-loop linking them, diverge from the equivalent region in the B. subtilis VmlR ( Figure 2E), as expected from the sequence alignment ( Figure 1B). Compared to VmlR, the additional eight residues in the VmlR2 ARD are mostly oriented away from the PTC and drug binding site ( Supplementary Figure S8A), similar to that observed for the ARDs of LsaA, VgaA LC and VgaL, which are also extended compared to B. subtilis VmlR despite sharing a similar spectrum of antibiotic resistance (Supplementary Figure S8B-D) (26). Indeed, despite containing eight more amino acids than B. subtilis VmlR, the VmlR2 ARD reaches a similar site in the PTC as many other PLS A -specific ARE-ABCFs ( Figure 2E, Supplementary Figure S8). Residue Ile261 of VmlR2 overlaps with the binding site of PLS A and nucleoside antibiotics, similarly to Phe237 of B. subtilis VmlR except inserted approximately 1.4Å deeper into the PTC ( Figure 2F, Supplementary Figure S9A-H). We next mutated selected residues of the ARD-loop of VmlR2 to alanine and assessed the ability of the resulting VmlR2 variants to mediate antibiotic resistance in a vmlR B. subtilis background ( Table 2). As observed for other antibiotic-resistance ABCFs, mutating the only residue that overlaps with the PLS A binding site, in this case Ile261 to alanine, did not appreciably shift the antibiotic resistance spectrum of VmlR2 (Table 1). However, due to the increased penetration of the VmlR2 ARD into the PTC, an alanine side chain at position 261 is also predicted to still overlap with the PLS A binding site (Supplementary Figure S9I-K). Therefore, we next performed a more radical substitution and replaced Ile261 with glycine. The Ile261Gly substitution, which is predicted to remove overlap between VmlR2 and the drugs ( Supplementary Figure S9L-N), compromised the activity of VmlR2 against retapamulin and virginiamycin M1 but not against lincomycin and iboxamycin (Table 1). This antibiotic-specific loss of protective activity is reminiscent of the effect we have previously observed with the Phe237Val substitution in B. subtilis VmlR (34). We therefore conclude that while the Ile261 sidechain is not required for antibiotic resistance.
Mutation of other selected residues close to the drugbinding site resulted in a complex pattern of changes to antibiotic resistance (Supplementary Figure S10). Residues Arg259 and Gln260, positioned in the tip of the ARD, were insensitive to mutation to alanine despite making defined contacts with the 23S rRNA (Supplementary Figure  S10A, B). By contrast, VmlR2 Y266 was required for resistance to tiamulin and virginiamycin M. This residue is positioned similarly to B. subtilis VmlR Y240 and VgaA LC Y223, with each residue stacking with 23S rRNA nucleotide U2585 (E. coli numbering; Figure 2G), and we previously found that a VgaA LC Y223A variant also displayed reduced resistance against PLS A antibiotics (26). The adjacent residue, E265, also had a drastic effect on antibiotic re- sistance when mutated to alanine. E265 is within hydrogenbonding distance of 23S rRNA residues A2450 and A2451, which form part of a region we have previously termed PTC loop 2 (PL2, Supplementary Figure S10C, (26)). Interactions between VmlR2 and the 23S rRNA in this region may serve to position the VmlR2 ARD in the PTC.
Finally, we have truncated the very tip of the ARD including the above-mentioned Arg259 and Gln260. The 254-261 ( SASERRQI) VmlR2 variant ( Supplementary Figure S8A) displayed (albeit diminished) protective activity against all of the antibiotics tested, i.e. retapamulin, virgini-amycin M1, iboxamycin and lincomycin. This result further reinforces the idea that VmlR2 acts allosterically to protect the ribosome.
In summary, residues that interact with U2585 and A2450/A2451, located at the beginning of interdomain linker ␣2, appear to be important for PLS A resistance mediated by VmlR2, while by contrast, mutating the residue that physically overlaps with the PLS A binding site or other residues in the extended tip of the ARD does not affect antibiotic resistance (Supplementary Figure S10C,D). These observations are consistent with previous mutagene- C. difficile MIC testing was carried out in BHIS media, and growth inhibition was scored after 36-48 h of anaerobic incubation at 37 • C. Plasmid-harboring strains were cultured in BHIS media supplemented with 10 g/ml thiamphenicol (to maintain the plasmids) and 500 ng/ml anhydrotetracycline (to induce the expression of ARE-ABCF proteins).
sis studies of ARE-ABCFs (26,34). Previously, we proposed a model in which ARE-ABCFs confer antibiotic resistance by modulation of the antibiotic binding site (26,33,34). Similar to B. subtilis VmlR, VmlR2 modulates the conformation of several 23S rRNA nucleotides that form the PLS A binding site. In particular, nucleotides around A2451, part of PL2, are displaced in the VmlR2-bound structure compared to the antibiotic-bound structures or a stalled elongating ribosome ( Figure 2H,I, Supplementary Figure S11). We therefore propose that binding of VmlR2 to a PLS Aor nucleoside-bound ribosome triggers a distortion of the antibiotic-binding site, in turn leading to dissociation of the antibiotic and ultimately allowing translation to resume after dissociation of the ARE-ABCF.

CplR contributes to intrinsic PLS A resistance in Clostridia
Next, we characterised the resistance mediated by C. difficile, C. perfringens and C. sporogenes CplR in the native bacterial hosts. In our C. difficile antibiotic sensitivity experiments we used two strains: the erm-R20291 and the erm + 630. R20291 is an epidemic hypervirulent strain that belongs to PCR-ribotype 027 (97). The MIC values determined with this strain are directly comparable with erm-ATCC 700057 MICs reported by Mitcheltree and colleagues in the original iboxamycin study (12). Despite encoding a genomic copy of CplR, both R20291 and ATCC 700057 are sensitive to iboxamycin (MIC of 0.0625 g/ml for R20291, MIC of 0.25 g/ml for ATCC 700057 (12)) ( Table 2).
The C. difficile 630 wild-type reference strain is a clinical isolate that over the years gained popularity with the research community (98). The prophage region of the mobilizable, non-conjugative transposon Tn5398 present in the genome of C. difficile 630 encodes a tandem expression cassette for two identical copies of the ermB gene for rRNA adenine N-6-methyltransferase (55). This ubiquitous antibiotic resistance determinant confers resistance to macrolides as well as lincomycin and clindamycin (99,100). Compared to wild-type 630, the erm 630 C. difficile is more sensitive to pleuromutilin retapamulin (MIC for of 0.4 g/ml vs 1.6 g/ml) as well as lincosamides (MIC for lincomycin of 16 g/ml vs > 256 g/ml, MIC for clindamycin of 16 g/ml versus 256 g/ml and MIC for iboxamycin of 0.25 g/ml vs 16 g/ml) ( Table 2). Since in B. subtilis VmlR synergises with Cfr to grant high levels of resistance (48), we next tested the effects of CplR loss in erm 630 C. difficile. The strain is further sensitised to lincosamides (MIC for lincomycin, clindamycin and iboxamycin of 0.1, 0.5 and <0.125 g/ml, respectively) as well as virginiamycin M1 (MIC of 16 g/ml) but not florfenicol or hygromycin A. The effect of CplR expression on nucleoside resistance is minor: a mere 2-fold increase in sensitivity to A201A, and no effect on sensitivity to hygromycin A ( Table  2, compare 630 erm cplR to 630 erm as well as 630 erm cplR/pRFP185 to 630 erm cplR/pRFP185-cplR). Our results establish that the synergetic action of Erm and CplR is responsible for the ability of wild-type C. difficile 630 to resist PLS A antibiotics, including iboxamycin (MIC of 16 g/ml).
To test the role of CplR in antibiotic resistance of C. sporogenes, we constructed a strain in which the cplR Cs gene is disrupted by the insertion of the intron-ermB resistance marker (Table 4). To deconvolute the effect of the introduced ermB gene on antibiotic resistance from that of cplR Cs disruption, we also constructed a strain in which the intron-ermB marker is introduced into the sigD locus. Compared to the isogenic wild type, the cplR::intron-ermB strain is more sensitive to retapamulin (MICs of 0.025 g/ml versus 25.6 g/ml, Table 4) suggesting that CplR mediates resistance to pleuromutilins. As expected, the introduction of the ErmB marker the cplR::intron-ermB strain displays higher resistance to lincomycin than the isogenic wild type (MIC for lincomycin of 256 g/ml versus 32 g/ml, Table  4). At the same time, the cplR::intron-ermB strain is more sensitive to iboxamycin (MICs of 0.031 g/ml versus 0.125 g/ml), consistent with the role of CplR in lincosamide resistance. The antibiotic resistance function of ClpR is fur- The MIC test for C. perfringens was carried out in GAM media and the growth inhibition was scored after 24 h of anaerobic incubation at 37 • C. In the case of plasmid-harboring strains GAM media was attentionally supplemented with 10 g/ml chloramphenicol to maintain the plasmids. Expression of the CplR ARE-ABCF was driven by the native promotor. The MIC test for C. sporogenes was carried out in GAM media, and the growth inhibition was scored after 24 h of anaerobic incubation at 37 • C.
ther supported by the clindamycin and iboxamycin MIC of the cplR-disrupted mutant being lower than that of the sigD mutant (compare cplR:intron-ermB to sigD::intron-ermB: clindamycin MICs of 8 g/ml vs 128 g/ml, iboxamycin MICs of 0.031 g/ml versus 16 g/ml) (Table 4). Finally, the effects on nucleoside antibiotic sensitivity are minor (2fold increased sensitivity to hygromycin B, and no effect on sensitivity to A201A). Taken together, our results establish ARE1 CplR as a clostridial genome-encoded PLS A resistance determinant and suggest that high PLS A resistance levels of C. difficile 630 are mediated through a combined action of direct ribosome protection by CplR and 23S rRNA modification by the ErmB methyltransferase.

An antibiotic challenge induces the accumulation of C. dif-ficilecplR full-length mRNA, and expression is re-repressed once bacterial growth is restored
Expression of ARE-ABCFs is typically induced upon a sub-MIC antibiotic challenge through a de-repression mechanism, with the full-length mRNA levels being kept low in the absence of the cognate antibiotic due to premature transcription termination in the absence of ribosomal stalling on a regulatory uORF (27)(28)(29)(30)(31)(32).
Genome-wide mapping of transcriptional start sites (TSSs; identified using dRNA-seq) and termination sites (TTSs; identified using RNAtag-seq) in C. difficile 630 (54) revealed that i) the cplR ORF preceded by a 212 nucleotide-long 5 upstream region (CDIF630nc 084), and that ii) as expected for an ARE-ABCF, under non-inducing conditions, the transcript prematurely terminates at start codon of the cplR ORF (CDIF630 02847) ( Figure 3A). Since antibiotic-induced ribosomal stalling by a regulatory uORF is a common regulatory strategy employed for inducible control of ARE-ABCF expression (27)(28)(29)(30)(31)(32), we have mounted an in silico search for regulatory ORFs of CplR and close relatives using our bioinformatic tool for uORF detection, uORF4u (61). Sequence conservation analysis reveals that the uORF of Clostridial CplR has a consensus of MR[I/M], with C. difficile cplR uORF specifically encoding Met-Arg-Ile ( Figure 3B, C). The consensus is different from the uORF4u-computed uORF for VmlR (MIN, Figure 3C), VmlR2 (MK[L/Q], Figure 3D) and LsaA (MAGN, Figure 3F), despite the four ARE-ABCFs having similar resistance spectra. The identified short uORFs are likely to be stalled by PLS A and nucleoside antibiotics, which inhibit translation after initiation with little contextspecificity (7,96). We could not identify an adequate number of sufficiently closely related homologues for the tool to be able to predict any conserved uORFs for Ard1.
We predicted secondary structures of the 5 cplR leader region in induced (i.e. full-length mRNA with 70S ribosome stalled on the uORF in the presence of antibiotics) and repressed (i.e. prematurely terminated mRNA generated in the absence of antibiotics) conformations using RNAfold (66) and Mfold (101). In the repressed conformation the 5 cplR leader region is predicted to form two stable stem loop structures ( Figure 3G). The 5 hairpin sequesters the putative regulatory uORF(MRI). The 3 terminator hairpin is followed by a single-stranded polyU stretch, thus constituting a canonical signal for efficient intrinsic termination in bacteria (102)(103)(104)(105). Importantly, the stem loop of the terminator hairpin sequesters the Shine-Dalgarno element positioned to drive the expression of the cplR ORF, potentially mediating the suppression of the cplR expression from full-length mRNA transcripts that fail to terminate prematurely. In repressed conformation the terminator hairpin is not formed, with an anti-terminator stem loop formed instead which allows transcription of full-length mRNA (Figure 3H). Furthermore, the Shine-Dalgarno that drives the cplR expression is readily accessible for initiating ribosomes ( Figure 3D).
To test antibiotic-inducible expression of C. difficile cplR, we cultured 630 ermR C. difficile in the presence of increasing sub-MIC concentrations of retapamulin (2-200 ng/ml; MIC of 800-1600 ng/ml) and lincomycin (0.02-2.0 g/ml; MIC of 32 g/ml) and assessed the levels of cplR mRNA by RT-PCR using two primer pairs: one amplifying 5 leader region and the other targeting the ORF ( Figure 4A). As expected, the level of cplR mRNA was increased in a dose-dependent manner for both retapamulin and lincomycin treatment, with the RT-PCR signal for both mRNA regions increasing upon antibiotic challenge ( Figure 4B).
Next, we investigated the time-course of cplR expression upon antibiotic challenge. We hypothesised that when the inhibitory effect of the antibiotic on translation is re-Nucleic Acids Research, 2023, Vol. 51, No. 9  lieved by the ARE-ABCF, the expression of the factor should subside as the antibiotic-induced ribosomal stalling that drives the expression is abolished. To test this hypothesis, we contrasted the RT-PCR data with bacterial growth kinetics. The cplR expression quickly responded to the antibiotic challenge, peaking within 30 min after the addition of antibiotics ( Figure 4C, left). Importantly, while the retapamulin-induced growth inhibition was severe and persisted for the whole duration of the experiment, the lincomycin-induced growth inhibition is much milder and at the later time points the growth kinetics are similar to that one the untreated control ( Figure 4C, right). Consistent with the prediction, while in the presence of 200 ng/ml retapamulin the maximal levels of cplR mRNA are maintained throughout the 6 h of the RT-PCR time course, but the cplR mRNA levels gradually decreased as bacterial growth -and, by inference, protein synthesis -has recovered from the lincomycin challenge ( Figure 4C, left).
Collectively, our results suggest that (i) inhibition of translation by retapamulin or lincomycin induces accumulation of cplR mRNA and (ii) induction of cplR mRNA is abrogated once expression of the factor overcomes the drug-induced translation inhibition.

The 5 leader region of cplR controls the induction of ABCF expression upon an antibiotic challenge
To test the role of the predicted 5 cplR leader region elements in antibiotic-induced expression of cplR in C. difficile we used a series of reporter constructs in which the expression of the mCherry red fluorescent protein was driven by the native cplR promoter (P cplR ) with either wild-type or mutated 5 leader variants ( Figure 5A). We used (i) RT-PCR to assess the effects of 5 leader mutations and/or antibiotic challenge that act on the level of transcription and (ii) mCherry fluorescence measurements to assess the com- pound effects that act both on transcriptional and translational levels simultaneously.
When P cplR is used to drive the expression of mCherry preceded by wild-type cplR 5 leader, while virtually no RT-PCR or fluorescence signal is detectable in the absence of antibiotics, whereas addition of either retapamulin (at 0.2 g/l) or lincomycin (at 2 g/ml) results in strong induction of the reporter ( Figure 5B, C, Supplementary Figure S12). Conversely, when the 5 leader region is replaced with that of constitutively and strongly expressed ferredoxin gene (fdx) (106), both the RT-PCR ( Figure 5B) and fluorescence ( Figure 5C, D) signal drastically increased and became insensitive to the antibiotic challenge. Taken together, these results suggest that our mCherry-based reporter strategy faithfully replicates the regulatory mechanisms that natively control the expression of cplR.
After establishing the validity of the assay, we next introduced point mutations in the cplR 5 leader probing (i) the regulatory uORF (AUG to AUC substitution of the initiation codon; mut uORF ), (ii) the terminator stem loop that sequesters the cplR main ORF Shine-Dalgarno motif (triple C-to-G substitution disrupting the helix; mut termSL ) and, finally, (iii) the terminator polyU tract (triple U-to-A substitution; mut polyU ) ( Figures 3G, H and 5A).
The mut uORF fluorescent reporter is inactive both in the presence and the absence of either retapamulin ( Figure 5C) or lincomycin ( Figure 5D), which is consistent with translation of the uORF being essential for the induction of cplR. The mut termSL reporter in which premature transcription termination is ablated and the Shine-Dalgarno (SD) element that drives expression of cplR is made directly accessible for translation initiation is (i) constitutively active and non-responsive to addition of antibiotics, and (ii) the fluorescent signal is >10-fold higher than that of the fully induced wild-type 5 leader reporter construct ( Figure 5C, D). By contrast, the expression of mCherry from the mut polyU reporter is (i) tightly regulated by antibiotics, with only very weak signal detectable in the absence of retapamulin, (ii) in the presence of retapamulin, the fluorescent signal was about twice (2.3-fold) as strong as that in the case of the wild-type reporter ( Figure 5C). In the case of the mut polyU reporter the full-length mRNA is expected to be constitutively produced (just as in the case of mut termSL ). Our RT-PCR experiments directly support this prediction: in good agreement with consistent production of the mut polyU reporter mRNA, the maximum induced signal is higher for the mut polyU reporter than for the wild-type reporter ( Figure 5B). At the same time, the cplR SD element is expected--just as in the wild-type reporter--to be acces-A B C D Figure 5. The ribosomal stalling on the uORF but not the premature RNAP termination is essential for induction of cplR in response to antibiotic challenge. (A) Experimental setup used for mutational probing of the cplR 5 leader region. Fluorescent reporters relied on mCherry expression driven by the cplR promotor (P CplR ) using wither (i) wild-type cplR 5 leader, (ii) fdxA 5 leader (5 leader fdxA ), (iii-v) cplR 5 leader with substitutions in the uORF start codon (mut uORF ), the intrinsic terminator stem (mut termSL ) or the polyU track downstream of the intrinsic terminator stem loop (mut polyU ).
Locations of the substitutions on the mRNA secondary structure are presented on Figure 3CD. (B) Accumulation mCherry mRNA variants with different cplR 5 leaders in response to a sub-MIC challenge with lincomycin (2 g/ml). The antibiotics were added to the logarithmic phase of C. difficile 630 ermR culture and total RNA was collected 30 mins after the antibiotic challenge. mCherry ORF and fdx ORF were probed by RT-PCR amplification. 23S rRNA was used as loading control. (C and D) Mutational probing of the cplR 5 leader. Expression of mCherry driven by is 5 leader fdxA (P fdx -mCherry) is insensitive to sub-MIC challenge with retapamulin (C) or lincomycin (D). Wild-type P cplR -mCherry reporter is induced by both retapamulin (C) and lincomycin (D), though the signal is much lower than in the case of P fdx -mCherry. The mut termSL reporter is constitutively induced and is insensitive to addition of either retapamulin (C) or lincomycin (D). The mut polyU reporter is repressed in the absence of antibiotics and is induced by the retapamulin challenge (C). To calculate the relative promoter activity, relative fluorescence intensity normalised by O.D. 600 was further normalised by the signal from the constitutive P fdx -mCherry reporter.
sible only upon ribosomal stalling on the uORF, and, evidently, translational attenuation is sufficient to ensure the suppression of cplR expression in the absence of antibiotic.
Collectively, our results establish that translational attenuation and transcriptional attenuation synergise to control the inducible expression of CplR in C. difficile.

DISCUSSION
We have experimentally characterised five genome-encoded ARE-ABCF proteins that are implicated in the intrinsic antibiotic resistance of S. capreolus (AAF1 Ard1), N. vireti (ARE2 VmlR2), and Clostridia C. difficile, C. perfringens and C. sporogenes (ARE1 CplR). As we show here, the vmlR B. subtilis strain is a convenient surrogate host for uncovering the resistance spectra of different ARE-ABCFs.
Thus, antibiotic sensitivity assays using heterologous expression in vmlR B. subtilis could be used to study other chromosomal ABCFs, guided by bioinformatic surveys.
Our cryo-EM reconstruction of the N. vireti VmlR2-70S complex is broadly similar to structures of other ARE-ABCFs with specificity for PLS A antibiotics. The ARD of these elements is counter-intuitively both essential for resistance and poorly conserved at the sequence level, exemplified by the eight-amino-acid insertion in the VmlR2 ARD compared to the B. subtilis VmlR ARD, despite the similar antibiotic resistance profiles of both these ARE-ABCFs ( Figure 1B, Table 1, Supplementary Figure S8). We observe the extended VmlR2 ARD to be oriented mostly away from the PTC and towards the P-tRNA, reminiscent of the ARDs from more distantly-related PLS A -specific ARE-ABCFs such as LsaA, VgaA LC and VgaL (Supple-mentary Figure S8). Thus, ARDs from PLS A -specific ARE-ABCFs can accommodate addition or deletion of amino acids while only modestly modulating their position with respect to the PTC. To date, all ribosome-bound ARE-ABCF structures have shown either a modest overlap between the factor and the antibiotic binding site, or no overlap at all--a pattern which holds for ARE-ABCFs specific for both PLS A s as well as macrolide, ketolide, and streptogramin Bs (26,34,41,43). This pattern is challenged by our results expanding the specificity of VmlR and VmlR2 to the nucleoside antibiotics A201A and Hygromycin A (Table 1), as each of these antibiotics occupy the PLS A binding site but additionally extend towards the A site. Modelled overlaps between VmlR and these antibiotics are extensive (Supplementary Figure S9). However, consistent with previous ARE-ABCF structures, an overlap between the ARE-ABCF and the antibiotic is not strictly required for resistance, as mutants predicted to abrogate the modelled overlap do not abolish activity (Table 1) (26,34,41). We note that VmlR2 induces distortions in the PTC nucleotides that would disrupt the antibiotic binding site, especially in the region around A2451, which we term PTC loop 2. Such a distortion is common to other PLS A ARE-ABCFs (26).
The most clinically important aspect of this work is the detailed study of the Clostridial ARE1 CplR. The Grampositive anaerobic spore-forming bacterium C. difficile is a causative agent of nosocomial and chronic infections as well as healthcare-associated diarrhoea (107). Disruption of the protective gut microbiota by broad-spectrum antibiotics allows for efficient proliferation of antibiotic-resistant C. difficile in the gut, thus triggering C. difficile infection, CDI (108). C. difficile has high levels of antibiotic resistance, and treatment with lincosamides such as clindamycin has long been recognised as a risk factor for development of CDI, especially in the case of erm + strains (51,(56)(57)(58)(59). As we show here, (i) similarly to L. monocytogenes ARE-ABCF VgaL and 23S ribosomal RNA methyltransferase Cfr (48), C. difficile CplR synergises with the 23S rRNA methyltransferase Erm to grant high levels of PLS A resistance, and, (ii) analogously to other experimentally studied ARE-ABCFs, induction of CplR is controlled by ribosomal stalling on a regulatory uORF. These results warrant a systematic exploration for co-occurrence of ARE-ABCFs and 23S rRNA methyltransferase antibiotic resistance determinants, with a special focus on the cases of their colocalisation in one operon on a mobile element (as is observed for LsaA and Cfr (49,50)). We demonstrate the utility of our bioinformatic tool uORF4u (61) for identification of regulatory uORFs and their consensus sequences. The induction of ARE-ABCFs relies on antibiotic-induced ribosomal stalling on regulatory uORFs (27)(28)(29)(30)(31)(32), and for elongation inhibitors such as macrolides, phenicols and oxazolidinones, antibiotic-induced ribosomal stalling is contextspecific (14,15,17,(109)(110)(111)(112). A dedicated follow-up investigation of the relationship between uORF consensuses and ARE-ABCF resistance and antibiotic inducibility spectrum is warranted. The specific sequence conservation of the uORF could point towards as-yet-uncharacterized patterns of sequence-specific antibiotic-mediated ribosomal stalling that are used by the cell to trigger the resistance factor induction upon an antibiotic challenge. As we have shown earlier, in the case of B. subtilis VmlR, the sequence identity of the regulatory uORF(MIN) is crucial for efficient induction upon a lincomycin challenge (31), despite this antibiotic not being known to act in a context-dependent manner. These insights, in turn, could potentially be leveraged for development of antibiotic variants with unnatural novel stalling profiles that are not detected by uORF-mediated attenuation mechanisms, and, therefore, do not trigger inducible resistance.

DATA AVAILABILITY
Cryo-EM maps have been deposited in the Electron Microscopy Data Bank (EMDB) with the accession code EMD-16246. The molecular model is available from the RCSB PDB with the accession code 8BUU.