Vancomycin resistance evolves rapidly in C. difficile during in vitro experimental evolution

We first generated genetically barcoded ancestral strains in an avirulent background with either a wild-type or an elevated mutation rate. Specifically, R20291, a clinically relevant ribotype 027 C. difficile strain, was rendered avirulent through complete deletion of 18 kb spanning the entire PaLoc that includes the genes encoding both major toxins and associated regulatory proteins, creating strain R20291ΔPaLoc. A subsequent deletion, removing the mutSL genes encoding a DNA-damage repair system, generated a hyper-mutable variant R20291ΔPaLocΔmutSL, with an approximately 20-fold higher mutation rate than the wild type. Five distinct derivatives of each ancestral strain were then generated through introduction of a 9-nucleotide barcode sequence downstream of the pyrE gene, resulting in 10 individually barcoded replicate lines used in the evolution experiment (R20291ΔPaLoc pyrE::barcode 1–5; R20291ΔPaLocΔmutSL pyrE::barcode 7–11). Each of the 10 barcoded strains was used to inoculate a 6-well plate containing media supplemented with vancomycin at 0.25×, 0.5×, 1×, 2×, 4×, and 8× the initial MIC of 1 μg/ml. Populations were passaged every 48 h, whereby cells from the well with the highest antibiotic concentration permitting growth were propagated in a 1:400 dilution to a fresh 6-well plate. This process was repeated for a total of 30 serial transfers per replicate line, with adjustment of the vancomycin gradient over time as the growth-permitting vancomycin concentration in each evolving line increased. Ten corresponding control populations were propagated under equivalent conditions in the absence of vancomycin. Populations underwent approximately 8.64 generations per transfer, yielding approximately 259 generations throughout the course of the experiment.

Resistance evolved rapidly in all 10 replicate populations propagated with vancomycin selection (Fig 1A). Nine of the 10 replicate lines evolved to grow in 2 μg/ml vancomycin (an apparent MIC of 4 μg/ml, the EUCAST breakpoint) by the end of the second passage (P2) and all 10 grew in the presence of 8 μg/ml (Bc2, 3, 4) or 16 (Bc1, 5, 7, 8, 9, 10, 11) μg/ml vancomycin by P30 (Fig 1A). The emergence of high-level resistance was significantly accelerated in the hyper-mutable replicate lines (Fig 1B). At least 6 individual clones were isolated from each evolved population at the endpoint and their vancomycin MIC was determined. Of the 82 clones tested, 38 (46%) had an MIC consistent with the vancomycin concentration permitting growth of the population from which they were isolated, while the remainder had an MIC slightly lower than expected [for 42 clones their MIC was 2-fold lower, whereas for 2 clones their MIC was 4-fold lower than the vancomycin concentration permitting growth of the population from which they were isolated (S1 Data)], demonstrating that significant variation still existed within some evolved populations by P30.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Evolution of vancomycin resistance.

(A) Changes in apparent vancomycin MIC over the course of a 30 transfer experimental evolution. MIC was determined as the well with the lowest vancomycin concentration showing no clear growth. (B) Shown are the means and standard deviations of the apparent MIC for 5 wild-type (open squares) and 5 hyper-mutating (closed circles) populations. Linear regressions fitted to each data set, blue and pink, respectively, are significantly different by ANCOVA, P = 0.0008. The data underlying panels A and B can be found in S5 Data. (C) Shown are the chromosomal locations of non-synonymous variant alleles in isolated wild-type (top) and hyper-mutating (bottom) C. difficile endpoint clones, excluding any mutations that were also identified in any of the control strains. Each circle represents a single C. difficile genome, colour coded according to population as indicated in the key on the left. The data underlying this panel, including a full list of all variants shown here along with synonymous and intergenic mutations, can be found in S2 Data. (D) PCA of P30 isolates from populations Bc1-5 (coloured points) vs. the ancestral strain (black triangle), with PC1 vs. PC2 plotted, accounting for 93% variance. The loadings (sporulation efficiency, growth, MIC, cell length) are shown in respective locations. Arrows show the evolutionary trajectories of wild-type replicate lines from their ancestor in multivariate phenotype space. The data underlying this figure can be found in S5 Data. MIC, minimum inhibitory concentration; PCA, principal component analysis.

https://doi.org/10.1371/journal.pbio.3002741.g001

Genetic bases of evolved vancomycin resistance

To understand the genetic bases of increased vancomycin resistance in individual endpoint clones, one clone per replicate population that had an MIC representative of their population was selected for whole genome sequencing. The parental strains for each barcoded lineage, and one random endpoint clone from each replicate control population, were also sequenced. For analyses, we focused on mutations that were observed in the vancomycin-treated lines but never in ancestral or in control evolved clones because these are the most likely to have evolved in response to vancomycin selection. Sequence variants unique to the vancomycin-evolved clones were identified using Varscan [21], validated with Breseq and IGV [22] after mapping to the reference R20291 genome (Fig 1C and S2 Data). We observed between 1 and 3 unique mutations per genome in wild-type evolved clones and between 14 and 26 in hyper-mutator evolved clones from the vancomycin treatment. Genic SNPs accounted for 43% of the variants, of which 67.4% were nonsynonymous and 32.6% were synonymous, with frameshifts accounting for a further 31% of these variants.

Parallel evolution, where mutations affecting the same locus arise in multiple independently evolving replicate populations, is strong evidence for the action of selection and suggests a potential role for these mutations in adaptation. In the 5 wild-type vancomycin selected lines, we observed parallel evolution occurring at 3 genomic loci: vanT in 3 clones, CDR20291_3437 (simplified to CD3437 hereafter) in 2 clones, and comR in 2 clones. Whereas mutations in vanT and CD3437 were mutually exclusive in wild-type evolved clones, mutations in comR always co-occurred with mutations in vanT. VanT is a putative Serine racemase (mutations in Bc3-5) encoded within a VanG-type cluster [17] that was previously implicated in decreased vancomycin susceptibility in C. difficile [19] (S2 Data). CD3437 encodes a predicted two-component system histidine kinase (mutations in Bc1 and 2), with its cognate response regulator encoded by CD3438. These genes had not previously been implicated in vancomycin resistance; however, the nearby CD3439 encodes a putative D,D-carboxypeptidase that likely plays a role in modification of peptidoglycan through removal of the stem peptide terminal D-Ala. Based on these predicted functions, we propose renaming these genes dacS (CD3437, histidine kinase), dacR (CD3438, response regulator), and dacJ (CD3439, D,D-carboxypeptidase). Consistent with mutations at these loci playing key roles in vancomycin resistance, all 5 hyper-mutator replicate lines had mutations in the vanG operon (1 had a nonsynonymous mutation in vanT and 1 in vanS) or dacS, dacR (encoding the cognate response regulator) and dacJ. Interestingly, 1 strain (Bc8) had mutations in both dacS and CD1523 (vanS), encoding a two-component system sensor histidine kinase that is thought to regulate the vanG operon in response to vancomycin, suggesting that the 2 pathways to resistance are not entirely mutually exclusive. comR encodes a homologue of the RNA degradosome component PNPase, suggesting that RNA stability may play a role in the vanT-associated mechanism of vancomycin resistance. Consistent with this possibility, in the third clone carrying a vanT mutation we observed coexisting mutations affecting maa, encoding a putative maltose O-acetyltransferase, and a 75-bp deletion that completely removed an intergenic region downstream of rpmH and before rnpA, encoding another predicted component of the RNA degradosome. By contrast, in the 2 clones carrying mutations in dacS we did not observe any mutations likely to affect RNA stability: one clone had no additional unique mutations, while the other had nonsynonymous mutations in bclA3 and CD3124. bclA3 encodes a spore surface protein with no known function in vegetative cells, while CD3124 encodes an orphan histidine kinase of unknown function. Interestingly, three of the hyper-mutating lineages also had mutations in CD3124, an identical frameshift mutation in all three.

Vancomycin-resistant clones display reduced fitness

To assess the wider consequences of evolved vancomycin resistance for bacterial phenotype, endpoint clones were phenotypically characterised for growth in vitro, sporulation efficiency, and cell morphology. All 10 strains displayed significantly impaired growth in rich liquid media (S1 Fig), with particularly severe defects apparent for Bc10 and 11. Several strains were also impaired in sporulation (S2 Fig), with a wide range of phenotypes apparent, from a mild defect for Bc2 and 4 and delayed sporulation for Bc8, to more severe defects for Bc7, Bc9, and 10 and complete loss of sporulation for Bc11. These growth and sporulation defects were also accompanied by changes in cell length relative to the parental wild-type strain (S3 Fig). Principal component analysis (PCA) of the 5 wild-type-derived endpoint evolved clones (Fig 1D) revealed all 5 resistant strains followed similar evolutionary trajectories away from the parental wild type, associated with lower sporulation efficiency and growth defects, albeit with divergence among replicate lines in the extent of defects and cell size. However, there was no apparent subclustering by resistance mechanism.

Population dynamics in evolving populations

Taken together, the phenotypic and genome sequence data for endpoint clones suggests that there are 2 alternative mechanisms of vancomycin resistance, one involving dacS and the other involving vanT. To better understand how selection acted upon these mechanisms, we next performed pooled population sequencing at passage 10, 20, and 30 to track mutation frequencies over time (Figs 2, S4, and S5). In total, discounting variants found in Bc1 P20, we identified 535 unique variants across the 10 parallel populations and 3 time points. We have removed Bc1 P20 from this analysis as the additional 520 variants identified in that sample alone likely reflect random mutation due to the emergence of a spontaneous hyper-mutator phenotype. Impacted genes clustered within 17 distinct functional classes by KEGG analysis (S6 Fig), with two-component systems and ABC transporters being particularly well represented. Focusing on the 2 main routes to resistance identified in endpoint clones, these data revealed highly contrasting selection dynamics, particularly in our wild-type replicate populations: mutations in dacS rapidly rose to high frequency, reaching fixation by P10. By contrast, mutations in vanT arose later and only reached fixation by P20 or 30, and were preceded by mutations at other sites which reached high frequency by P10 but that ultimately did not survive, being replaced by vanT mutants presumably conferring higher levels of vancomycin resistance. The 2 preceding high-frequency mutations in Bc3 (both T>TA) were very close together, separated by only 7 bp in an intergenic region downstream of CD0482, encoding a uridine kinase, and approximately 250 bp upstream of glsA, encoding a glutaminase. These mutations are outside of the likely promoter region [23] but it is possible that they affect regulation of glsA through alteration of the binding site for a regulatory protein or RNA. Interestingly, changes in glutamine metabolism have previously been linked to vancomycin resistance in Staphylococcus aureus [24]. The single high-frequency mutation in Bc5 at P10 is a nonsynonymous substitution in CD3034, introducing a Gly255Asp mutation in the encoded D-hydantoinase that may play a role in the synthesis of D-amino acids. Taken together, these data suggest that vanT is not required for first-step resistance; vanT mutations either provide higher-level vancomycin resistance, allowing them to supplant earlier mutations, or they require potentiating mutations to arise and be selected first. However, no consistent secondary mutations common to all populations with vanT mutations were identified.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Genomic location of gene variants over time.

Accumulation of variants in the wild-type C. difficile lineages Bc1 (A), Bc2 (B), Bc3 (C), Bc4 (D), and Bc5 (E). Each circle plot represents the 4.2 Mb genome of a single evolving population after 10 (inner ring), 20 (middle ring), and 30 passages (outer ring), with the locations of non-synonymous within gene variants indicated with black circles and the penetrance of each mutation in the population indicated by the size of the circle. The line graphs show the frequency of all variants (intergenic, synonymous, non-synonymous, frameshifts and nonsense) in each population. The vancomycin MIC for each population is also indicated by the shaded region. Mutations also identified in the respective end point clone (Fig 1C) are highlighted by the coloured lines. Note population Bc1 evolved an apparent hyper-mutator phenotype prior to P20, with 520 variants identified at that time point. For simplicity, only variants present in P10 and P30 are labelled. The data underlying this figure, including a full list of all variants shown here and those in Bc1 P20, can be found in S3 Data. MIC, minimum inhibitory concentration.

https://doi.org/10.1371/journal.pbio.3002741.g002

Recapitulation of dacS mutations confirms role in resistance

As the dacJRS cluster had not previously been implicated in vancomycin resistance, we validated the role of DacS in vancomycin resistance by recapitulating individual mutations in a clean genetic background. We chose the variant dacS allele identified in Bc1 (714G>T) and the variant that evolved in parallel in Bc8 and Bc9 (548T>C). Recapitulated strains carrying only the dacS mutation of interest were constructed in the parental R20291ΔPaLoc by allelic exchange. Introduction of either mutation alone resulted in a 4-fold increase in the vancomycin MIC compared to the parental strain, confirming that DacS is playing a significant role in the evolved resistance we observed (Fig 3A). Interestingly, both mutations resulted in statistically significant growth defects (S7 and S9 Figs), suggesting they were responsible, at least in part, for the growth defects observed in the Bc1, Bc8, and Bc9 endpoint evolved clones. Neither mutation affected sporulation efficiency (S8 and S9 Figs). AlphaFold prediction of the DacS structure yields a plausible dimer model (Fig 3B) that is highly similar to previously characterised histidine kinases [25]. The Bc1 and Bc8/9 mutations described here both result in amino acid changes in the DacS cytoplasmic domain: Bc1 Glu238Asp within the predicted catalytic ATPase (CA) domain and Bc8/9 Val183Ala within the dimerization and histidine phosphorylation (DHp) domain. The impact of these mutations on the function of DacS was not clear but we hypothesised that DacS, along with its cognate response regulator DacR, could be regulating the expression of the D,D-carboxypeptidase encoded by dacJ. To examine this possibility, we extracted RNA from R20291ΔPaLoc, R20291ΔPaLoc dacSc.714G>T, and R20291ΔPaLoc dacSc.548T>C, both in the absence and presence of 0.5 μg/ml vancomycin, and assessed the expression of dacS, dacR, and dacJ by qRT-PCR (Fig 3C). Either point mutation resulted in a substantial up-regulation of expression of all 3 genes, varying from 4.8-fold increased transcription of dacS in R20291ΔPaLoc dacSc.714G>T to 94.6-fold increased transcription of dacJ in R20291ΔPaLoc dacSc.548T>C. Interestingly, these effects were independent of vancomycin suggesting that the two-component system responds to an as yet unknown signal. The genomic organisation of the dacJRS region (Fig 3D) and previous global transcription site mapping [23] suggests that dacS and dacR are transcribed from a single promoter upstream of dacR and that dacJ is transcribed from its own separate promoter. Our data demonstrated that both of these promoters are subject to regulation by the DacS/DacR two-component system, but could not differentiate between activation of a TCS that positively regulates these promoters or deactivation of a repressor. However, as complete deletion of dacRS had no effect on dacJ expression (Fig 3C), we conclude that the DacRS two-component system is a positive regulator of dacJ, and likely also dacRS, and that the mutations identified in Bc1, Bc8, and Bc9 result in increased TCS activity.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. dacS mutations lead to dysregulation of dacJRS.

(A) Vancomycin MICs of R20291ΔPaLoc, R20291ΔPaLoc dacSc.714G>T, and R20291ΔPaLoc dacSc.548T>C as determined by agar dilution. Assays were performed in biological triplicate and technical duplicate, 4 representative spots are shown for each strain. (B) AlphaFold model of DacS as a dimer [26]. The transmembrane domains were identified using DeepTMHMM [27] and the Catalytic ATPase and Dimerization and Histidine Phosphorylation domains were predicted using InterProScan [28]. The locations of Val183 and Glu238 are highlighted in purple on one chain. (C) qRT-PCR analysis of dacJRS expression in R20291ΔPaLoc (black bars), R20291ΔPaLoc dacSc.714G>T (yellow bars), R20291ΔPaLoc dacSc.548T>C (green bars), and R20291ΔPaLocΔdacRS (blue bars). Expression was quantified against a standard curve and normalised relative to the house-keeping gene rpoA. Assays were performed in biological and technical triplicate. Statistical significance was calculated using a two-way ANOVA with the Tukey–Kramer test, ** = P < 0.001. The data underlying this panel can be found in S5 Data. (D) Lack of dacJ up-regulation in R20291ΔPaLocΔdacRS suggests that phosphorylated-DacR acts as an activator of the 2 promoters in the dacJRS cluster, with DacS Glu238Asp or Val183Ala substitutions constitutively activating the function of the TCS, respectively. The consequence is over-expression of DacJ which is then translocated to the cell surface where it can cleave the terminal D-Ala residue in nascent peptidoglycan, thereby preventing vancomycin binding. MIC, minimum inhibitory concentration.

https://doi.org/10.1371/journal.pbio.3002741.g003

We hypothesised that the overexpression of DacJ, and the resulting removal of terminal D-Ala residues from peptidoglycan precursors, depleted vancomycin binding sites within the cell wall resulting in the observed reduction in vancomycin susceptibility in both strains with mutated dacS. To confirm the loss of accessible D-Ala-D-Ala in the cell walls of resistant strains, we incubated cells with vancomycin-BODIPY and visualised binding by fluorescence microscopy (Fig 4). The parental R20291ΔPaLoc strain displayed clear mid-cell vancomycin labelling in 34% of cells, consistent with the growth phase-dependent presence of pentapeptide precursors at the site of maximal peptidoglycan synthesis [19,29,30]. By contrast, dacJ overexpression in R20291ΔPaLoc dacSc.714G>T or dacSc.548T>C was associated with complete loss of vancomycin labelling. To confirm that this difference was due to DacJ, we constructed a dacJ deletion in resistant endpoint isolate Bc1, which partially restored vancomycin labelling and reduced the vancomycin MIC 8-fold (Fig 4 and Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. DacJ activity depletes vancomycin binding sites in the cell wall.

(A) Representative phase contrast (left) and fluorescence (right) images of R20291ΔPaLoc, R20291ΔPaLoc dacSc.714G>T, R20291ΔPaLoc dacSc.548T>C, Bc1 and Bc1ΔdacJ labelled with vancomycin-BODIPY. A control of unlabelled R20291ΔPaLoc (top right) is included to show the background fluorescence that C. difficile exhibits in the presence of oxygen [31]. Vancomycin labelling of accessible D-Ala-D-Ala, present during active synthesis of peptidoglycan, is indicated with white arrows. (B) The normalised maximum fluorescent intensities at the centre of individual cells showed that 34% of R20291ΔPaLoc cells had bound vancomycin, compared to no cells in the unlabelled control of the same strain. R20291ΔPaLoc dacSc.714G>T and R20291ΔPaLoc dacSc.548T>C and the evolved endpoint clone Bc1, which all overexpress dacJ, were not labelled when incubated with vancomycin-BODIPY. However, when dacJ was deleted in Bc1 partial labelling was restored (4% of cells were labelled at their mid-point). This was consistent with the incomplete re-sensitisation to vancomycin also observed with this mutant. The black dashed line indicates the normalised max intensity threshold used to delineate cells that have detectable levels of vancomycin-BODIPY at their midplane from those without (obtained by direct visual inspection, normalised max intensity = 1.18). Asterisks indicate the P-values from pairwise comparisons using a two-tailed Fisher’s exact test; ** = P < 0.0001, * = P < 0.05. The data underlying panel B can be found in S5 Data.

https://doi.org/10.1371/journal.pbio.3002741.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Vancomycin MICs of strains recapitulating Bc1 mutations.

https://doi.org/10.1371/journal.pbio.3002741.t001

Vancomycin resistance in Bc1 is multifactorial

Recapitulation of the dacSc.714G>T mutation, originally identified in the genome sequenced resistant endpoint clone from replicate Bc1, resulted in only a 4-fold increase in vancomycin MIC in the recapitulated mutant, compared with the 16–32-fold increase seen for the evolved clone. Because dacSc.714G>T was the only mutation we observed in the evolved clone, this discrepancy in MICs suggested that the evolved Bc1 clone had additional mutations that were not detected by our short-read Illumina-based pipeline. To address this, we re-sequenced endpoint isolates Bc1-5 using a long-read technology (Oxford Nanopore) and combined this with the existing short-read data to generate high-quality closed genomes. No additional mutations were identified in isolates Bc3-5 (S4 Data). However, previously unobserved additional mutations were identified in the evolved clones from replicates Bc1 and Bc2. Two additional single-nucleotide insertions were identified in evolved clone Bc2, causing frameshifts in CD0794 and CD1871, encoding a conserved hypothetical protein and an ABC transporter permease, respectively. Two additional large indels were identified in Bc1, a 44-bp deletion downstream of CD0978 and upstream of CD0980 (1,197,357_1,197,400del) and a 30-bp duplication within the ORF of vanS (vanSc.367_396dup), encoding the vanG cluster histidine kinase. In the annotated R20291 genome, there is a very small 168-bp ORF upstream of CD0980, annotated CD0979. However, as other C. difficile genome sequences, including that of the well-characterised strain 630 [32] do not show an equivalent ORF in this region and comprehensive RNAseq analysis [23] did not detect any transcription associated with CD0979, we conclude that this is likely a misannotation and that the observed deletion affects an intergenic region. To assess the contribution of these large indels to vancomycin resistance in Bc1, we constructed a series of R20291ΔPaLoc mutant strains with dacSc.714G>T, 1,197,357_1,197,400del and vanSc.367_396dup in all possible combinations of single, double, and triple mutations. The vanS/dacS double mutant fully recapitulated the increased vancomycin resistance seen in the evolved Bc1 clone (16 μg/ml, Table 1), whereas the vanS and dacS single mutants only displayed a 2- and 4-fold increase in MIC, respectively, suggesting that these mutations interacted synergistically with respect to the resulting vancomycin resistance level. The 1,197,357_1,197,400del mutation made no apparent contribution to vancomycin resistance in any of the constructed mutants.

Analysis of growth profiles of all recapitulated single, double, and triple mutants revealed severe growth defects similar to that seen in the evolved Bc1 clone for all recapitulated strains that included the dacS point mutation, suggesting that this mutation alone was largely responsible for the growth defect (S7 Fig). Sporulation of the triple mutant was very similar to both the evolved Bc1 clone and the parental R20291ΔPaLoc (S8 Fig).

Mutations in vanS have previously been implicated in reduced vancomycin susceptibility through increased expression of the enzyme-encoding genes in the vanG cluster. To assess if a similar mechanism was at play in Bc1, we performed qPCR on all of the genes in the vanG cluster in R20291ΔPaLoc vanSc.367_396dup (Fig 5A). The vanS mutation had no effect on the expression of vanR in either the presence or absence of 0.5 μg/ml vancomycin but, surprisingly, a significant decrease was observed in the vanS transcript copy number in both conditions (4.5- to 4.8-fold decrease). As these 2 genes are thought to be transcribed in a single bicistronic mRNA [23], this observation suggests that the duplication within the vanS ORF either impacts the stability of the 3′ end of the transcript or transcriptional efficiency past vanR. Despite the apparent decrease in the vanS transcript copy number, the expression of vanG, vanXY, and vanT were all significantly increased in the vanS mutant (from a 10-fold increase in transcription of vanG in the presence of vancomycin to a 136-fold increase in vanG without vancomycin).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. vanS mutations lead to dysregulation of the vanG cluster.

(A) qRT-PCR analysis of expression of the genes in the vanG cluster in R20291ΔPaLoc (black bars) and R20291ΔPaLoc vanSc.367_396dup (orange bars). Expression was quantified against a standard curve and normalised relative to the house-keeping gene rpoA. Assays were performed in biological and technical triplicate. Statistical significance was calculated using a two-way ANOVA with the Tukey–Kramer test, ** = P < 0.001. (B) AlphaFold model of VanS as a dimer with (right) and without (left) the vanSc.367_396dup encoded 10 amino acid duplication [26]. The transmembrane domains were identified using DeepTMHMM [27] and the Catalytic ATPase and Dimerization and Histidine Phosphorylation domains were predicted using InterProScan [28]. The location of the 10 amino acid duplication is highlighted on both chains in dark blue. The insertion results in an approx. 90° rotation of the transmembrane domains relative to the catalytic ATPases. (C) Genomic organisation of the vanG cluster. The data underlying this panel A can be found in S5 Data.

https://doi.org/10.1371/journal.pbio.3002741.g005

Strains and growth conditions

All strains used or generated in the course of this study are described in S1 Table. C. difficile was routinely cultured on brain heart infusion (BHI) agar and in tryptone yeast (TY) broth. C. difficile was grown in an anaerobic cabinet (Don Whitley Scientific) at 37°C, with an atmosphere composed of 80% N₂, 10% CO₂, and 10% H₂. Media was supplemented with thiamphenicol (15 μg/ml) (Sigma) and colistin (50 μg/ml) (Sigma) as appropriate. For counter selection against plasmids bearing the codA gene, C. difficile defined media with 5-fluorocytosine (CDDM 5-FC) was used as described previously [42]. E. coli was routinely cultured in Luria–Bertani (LB) broth or agar at 37°C and supplemented with chloramphenicol (15 μg/ml) (Acros Organics) or kanamycin (50 μg/ml) (Sigma) as appropriate.

Molecular biology

All oligonucleotides and plasmids are described in S2 and S3 Tables, respectively. Plasmid miniprep, PCR purification, and gel extractions were performed using GeneJET kits (Thermo Fisher). High-fidelity PCR amplification was performed using Phusion polymerase (NEB), and standard PCR amplification was performed using Taq mix red (PCRBIO), according to manufacturers’ instructions. Gibson assembly primers were designed using NEBBuilder (NEB) and restriction digestion and DNA ligation was performed using enzymes supplied by NEB. Plasmids were transformed into NEB5a (NEB) or CA434 competent E. coli cells according to the NEB High Efficiency Transformation Protocol. The sequences of cloned fragments were confirmed using Sanger sequencing (Genewiz, Azenta Life Sciences, Germany).

C. difficile mutagenesis

C. difficile strain R20291 was modified for use in evolution experiments and to recapitulate individual mutations. Homology arms for introducing mutations onto the C. difficile chromosome by allelic exchange were generated either by Gibson assembly of PCR products or synthesised by Genewiz (Azenta Life Sciences) and then subsequently cloned between BamHI and SacI sites in pJAK112 [23], a derivative of pMTL-SC7215 [42]. Following confirmation by Sanger sequencing, plasmids were transformed into E. coli CA434 and transferred to C. difficile by conjugation [43]. Homologous recombination was performed as previously described [42] and mutations were confirmed by PCR and Sanger sequencing of mutated regions.

R20291 was rendered avirulent via deletion of the complete pathogenicity locus (encoding toxins A and B) using plasmid pJAK143 [44], yielding strain R20291ΔPaLoc. This strain was then further modified by deletion of DNA repair operon mutSL to create a hyper-mutator strain; 1.2 kb up and downstream of the mutSL operon was amplified by PCR using oligonucleotides RF2066 and RF2067, and RF2068 and RF2069 (S2 Table), respectively, and cloned between BamHI and SacI sites in plasmid pJAK112. The resulting plasmid, pJEB002, was then conjugated into R20291ΔPaLoc and allelic exchange performed as described above, yielding R20291ΔPaLocΔmutSL. Enumeration of colony-forming units (CFUs) on BHI agar containing rifampicin (0.015 μg/ml) suggested this strain has a mutation rate approximately 20-fold higher than R20291ΔPaLoc. Five barcoded variants of both R20291ΔPaLoc and R20291ΔPaLocΔmutSL were then generated by insertion of unique sequencing barcodes. Briefly, pJAK081 (identical to pJAK080 [23] but in the pMTL-SC7215 backbone) was modified by introduction of a synthetic DNA fragment containing a multiple cloning site and a 9-bp sequencing barcode (Barcode 1), flanked by the fdx and slpA terminators, between the existing homology arms for insertion of DNA between CD0188 (pyrE) and CD0189 in the R20291 genome. A second plasmid (pJAK202) containing Barcode 2 was constructed in the same manner and the rest (pJAK203-205 and 207–211) were generated via site-directed mutagenesis using pJAK201 as a template. These 10 plasmids were then used to generate R20291ΔPaLoc Bc1-5 and R20291ΔPaLocΔmutSL Bc7-11. Recapitulated strains were generated by introducing deletion, insertion, or point mutations into R20291ΔPaLoc. For each desired mutation, an approximately 2-kb synthetic DNA fragment, centred on the mutation of interest, was cloned between BamHI and SacI sites in pJAK112. The resulting plasmids (S3 Table) were conjugated into R20291ΔPaLoc, and allelic exchange was performed as described above.

Evolution

Directed evolution of C. difficile was performed using a broth-based gradient approach in which 10 individually barcoded parallel lines were evolved for a period of 30 passages. A 6-well plate for each of the parallel lines was prepared for each passage using 4 ml of TY broth with vancomycin, spanning a gradient of 0.25 to 8× the current MIC, as determined from the most recent passage for each line, allowing the gradient to rise with increasing resistance.

The evolution was initiated using overnight cultures from single colonies, adjusted to OD_600nm 0.1, and 10 μl was added to each well, before incubating for 48 h at 37°C. Plates were visually inspected after 48 h and 10 μl of the well with the highest antibiotic concentration supporting growth was used to inoculate the wells of the subsequent passage. For each parallel line, a control well was passaged without antibiotic; 1 ml of a population, and the corresponding control, was frozen at −80°C in 15% glycerol whenever the MIC increased, and after passages 10, 20, and 30.

gDNA extraction

gDNA was obtained from C. difficile cultures using the phenol–chloroform method as described previously [38]. DNA concentration was quantified using Qubit, and purity was assessed via microvolume spectrometry.

Sequencing

Endpoint (P30) isolates and respective controls were gained through plating P30 populations and culturing single colonies. Vancomycin MICs were determined by agar dilution (described below), and gDNA extraction and quantification were performed as above. Library prep (Nextera XT Library Prep Kit (Illumina, San Diego, United States of America)) and 30× illumina sequencing (NovaSeq 6000, 250-bp paired end protocol) of isolates was performed at MicrobesNG (Birmingham, UK). Reads were trimmed at MicrobesNG using Trimmomatic (v0.30) [45] with a sliding window quality cutoff of Q15.

WT-derived endpoint isolates, the same as those described above, were also sequenced by nanopore. Library prep (Oxford Nanopore Technologies SQK-RBK114.96 kit (ONT, UK)) and nanopore sequencing (GridION, FLO-MIN114 (R.10.4.1) flow cell) was performed at MicrobesNG (Birmingham, UK).

Pooled sequencing of the 10 evolved and 10 control populations was performed at 3 time points—P10, P20, and P30, and 1 ml of the well with highest antibiotic concentration supporting growth was taken after 48 h, harvested via centrifugation, and frozen at −20°C. gDNA extraction and quantification were performed as above. Library prep (Nextera DNA Flex Library Prep Kit (Illumina, San Diego, USA)) and 250× Illumina sequencing (NovaSeq 6000, 150-bp paired end protocol) was performed at SNPsaurus (Oregon, USA). Reads were trimmed using Trimmomatic (v0.39) with the following criteria: leading:3; trailing:3; slidingwindow:4:15; minlen:36.

Sequencing analysis pipeline: Isolates

Trimmed reads were checked using FastQC (v0.11.9) [46] to ensure sufficient quality for analysis. Analysis was primarily performed using a custom pipeline, built based on a resistant mutant analysis pipeline [47]. Reads were aligned to the reference (C. difficile R20291, accession number: FN545816) using BWA-mem (v0.7.17) [48] and sorted using SAMtools (v1.43) [49]. Coverage across the genome was inferred using Bedtools (v2.30.0) [50] genomecov and map functions. PCR duplicates were removed via Picard (v2.25.2) (http://broadinstitute.github.io/picard/). The mpileup utility within SAMtools (v1.43) was used to generate the mpileup file required for Varscan. Variants were then called using Varscan (v2.4.3–1) [21] mpileup2cns (calling SNPs and indels) using the following parameters: min-coverage 4; min-reads2 4; min-var-freq 0.80; p-value 0.05; variants 1; output-vcf 1. Here, a minimum of 4 reads were required to support a variant, and the cutoff for variant calling was 80%. Vcf files were annotated using snpEff (v5.0) [51]. Variants were also called using Breseq (v0.35.5) [22], using default parameters, and putative variants were retained if detected in both analysis pipelines. All variants were manually verified using IGV (v2.8.6) [52]. Variants that were also called in control lines were removed using Varscan (v2.4.3–1) compare, generating a list of variants unique to vancomycin resistant lines.

Nonsynonymous mutations occurring within genes were plotted using a previously published custom script in RStudio (v4.1.0) utilising the Plotrix package [53] to visualise parallel evolution.

Sequencing analysis pipeline: Isolates (long-read)

Long-read genome assembly was performed at MicrobesNG (Birmingham, UK): Reads were assembled using Flye (v2.9.2-b1786) [54], polished using Medaka (v1.8.0) (https://github.com/nanoporetech/medaka), and annotated using Bakta (v1.8.1) [55]. For a given isolate, Nanopore and Illumina data were amalgamated using Polypolish (v0.6.0) [56]. Short read sequences were aligned to the Nanopore assembly using BWA-mem (v0.7.17) [48], using the -a parameter. Alignments were filtered by insert size using Polypolish with default parameters. Assemblies were polished using Polypolish and annotated with Prokka (v1.14.6) [57]. Variants were called using Snippy (v4.6.0) (https://github.com/tseemann/snippy) with default parameters. Identified variants were manually verified using IGV (v2.8.6) [52].

Sequencing analysis pipeline: Populations

The 10 evolved and 10 control populations, sequenced at P10, P20, and P30, were evaluated using a population analysis methodology that followed the same custom pipeline as for isolate analysis, with modifications to SNP calling parameters and filtering. InSilicoSeq (v1.5.4) [58] was used to rapidly simulate realistic sequencing data at a range of coverage depths (80×, 100×, 300×) based on the R20291 genome, with SNPs seeded at 5% frequency. Simulated sequences were analysed using the custom pipeline, and a range of Varscan parameters (P-values, min-coverage, min-reads2) were trialled to generate a set of parameters to accurately call variants without false positives.

Varscan parameters used to call population variants were dependent on average coverage depth, allowing a strong evidence base while calling low-frequency variants. Samples were separated by average coverage: the following Varscan (v2.4.3–1) mpileup2cns parameters were used for samples with 100× or higher average coverage: min-coverage 80; min-var-freq 0.05; P-value 0.05; variants 1; output-vcf 1. A minimum of 4 reads were required to support a variant, and the cutoff for variant calling was 5%. For samples with average coverage below 100×, the following Varscan (v2.4.3–1) mpileup2cns parameters were used: min-coverage 4; min-reads2 4; min-var-freq 0.05; p-value 0.05; variants 1; output-vcf 1. As with high-coverage samples, a minimum of 4 reads were required to support a variant, meaning the minimum variant frequency which could be called was inversely scaled (4/coverage at position), with a minimum frequency of 5%. This process identified a list of variants within each line at each time point and their respective frequencies.

As in the isolate pipeline, population variants were filtered using Varscan (v2.4.3–1) compare, removing variants that were also present in control lines to generate a list of variants unique to vancomycin-resistant lines. Further filtering of this dataset, again using Varscan (v2.4.3–1) compare, discarded variants that never reached above 10% frequency by the end of the evolution. Variants were also manually inspected to ensure no variants remained the same frequency across all time points.

Mutations occurring within genes (except those in Bc1 P20) were coloured according to their barcoded replicate line and displayed in KEGG Mapper–Color [59] to view KEGG pathways.

Nonsynonymous mutations occurring within genes and rRNAs were plotted using a previously published custom script in RStudio (v4.1.0) utilising the Plotrix package [53] to visualise population dynamics, where point size indicated mutation frequency in the population.

Strain fitness analysis

Growth curves were performed anaerobically in 96-well plates using a Stratus microplate reader (Cerillo). Overnight cultures of C. difficile were diluted to OD_600nm 0.05 and incubated at 37°C for 1 h. Plate lids were treated with 0.05% Triton X-100 + 20% ethanol. Plates were prepared using 200 μl of equilibrated culture. Samples were measured at minimum in biological and technical triplicate. The OD_600nm was measured every 3 min over a 24-h period. Data was plotted in Graph pad Prism (v9.0.2) and was analysed in RStudio (v4.1.0) using the GrowthCurver package (v0.3.0) [60].

To assess sporulation efficiency, triplicate overnight C. difficile cultures were adjusted to OD_600nm 0.01 and grown for 8 h. Cultures were then adjusted again to OD_600nm 0.01, subcultured 1:100 into 10-ml BHIS broth, and grown overnight to obtain early stationary phase spore-free cultures (T = 0). At T = 0, and the following 5 days, total viable counts were enumerated by spotting 10-fold dilutions in technical triplicate onto BHIS agar with 0.1% sodium taurocholate. Colonies were counted after 24-h incubation. Spore counts were enumerated using the above method following heat treatment (65°C for 30 min).

To visualise cell morphology, C. difficile samples were harvested via centrifugation, washed twice in PBS, and fixed in 4% paraformaldehyde, before harvesting and resuspension in dH₂O. Samples were mounted in 80% glycerol and imaged using a 100× Phase Contrast objective on a Nikon Ti eclipse widefield imaging microscope using NIS elements software. Images were analysed in Fiji (v2.9.0) using MicrobeJ (v5.13l) [61].

Vancomycin MICs

MICs were obtained via standard agar dilution methods [62]. Briefly, overnight cultures were adjusted to OD_600nm 0.1, corresponding to between 1.06 × 10⁷ and 1.83 × 10⁷ CFUs for all strains analysed, and 2.5 μl samples were spotted in biological triplicate and technical duplicate onto BHI plates with ranging antibiotic concentrations. MICs were determined after 48 h incubation, and plates were imaged using a Scan 4000 colony counter (Interscience).

Principal component analysis

WT (Bc1-5) P30-resistant isolates were characterised in terms of their sporulation efficiency, growth rate, MIC, and cell length. These were compared, along with the WT ancestor from the start of the evolution, in a PCA. The PCA was computed using the prcomp() function in Base R (http://www.rproject.org/) and visualised using the factoextra package. The first 2 PC were plotted, as these accounted for 93% variance. The loadings were added in their respective locations, and isolates were coloured based on resistance mechanism.

qRT-PCR

Total RNA was extracted using the FastRNA pro blue kit (MP Biomedicals). Cells were grown to an OD_600nm of approx. 0.4, and 2 volumes of RNA protect (Qiagen) were added. Cells were incubated for a further 5 min, before harvesting via centrifugation. Cell pellets were stored at −80°C. Pellets were resuspended in RNA Pro solution and transferred to a tube containing lysing matrix B (MP Biomedicals). Cells were lysed via FastPrep (2 cycles of 20 s, 6 m/s) and centrifuged (16,200 × g, 4°C, 10 min) to remove insoluble cell debris. The supernatant was transferred to a microfuge tube, and 300 μl chloroform (Sigma) was added. Samples were centrifuged again (16,200 × g, 4°C, 15 min), and the supernatant was precipitated at 20°C overnight after addition of 500 μl 100% ethanol. After precipitation, RNA was harvested by centrifugation (16,200 × g, 4°C, 15 min), washed with 70% ethanol, and dried. Precipitated RNA was resuspended in 50 μl nuclease-free water, residual DNA was removed using the Turbo DNA-free kit (Invitrogen), and the RNA was cleaned and concentrated with the RNeasy Minelute cleanup kit (Qiagen), as per manufacturers’ instructions.

cDNA was generated using Superscript III (Invitrogen), and 5 μg RNA was mixed with 2 μl dNTP mix and 1 μl 100 mM random primer (Eurofins), heated at 65°C for 5 min, and cooled on ice; 8 μl 5× buffer, 2 μl 0.1 M DTT (Invitrogen), 1 μl RiboLock RNase inhibitor (Thermo Scientific), and 2 μl Superscript III were added, before incubation at 25°C (5 min), 50°C (30 min), and 70°C (15 min). cDNA was adjusted to 40 ng/μl. RT negative controls were made as above, without presence of Superscript III.

Expression was measured against an exact copy number control, via standardisation with a plasmid containing 1 copy of each target DNA sequence [63] and normalised against expression of the housekeeping gene rpoA. Plasmids pJEB029 (containing approximately 200-bp target gene fragments of dacS, dacR, dacJ, and rpoA) and pJEB032 (containing approx. 200-bp target gene fragments of vanR, vanS, vanG, vanXY, vanT, and rpoA) were synthesised with the pUC-GW-Kan backbone (Genewiz). These were purified using the GeneJET plasmid miniprep kit (Thermo Fisher) and linearised using NdeI (NEB) and diluted to known copy number (2 × 10⁸ per μl). A qPCR mastermix was assembled, containing 25 μl SYBR Green JumpStart Taq ReadyMix (Sigma), 7 μl MgCl₂ in buffer (Sigma), forward and reverse primers (concentration determined by prior optimisation) and nuclease-free water up to 45 μl. Approximately 5 μl of cDNA (40 ng/μl), 5 μl of RT negative control, or 5 μl qPCR template plasmid pJEB029 or pJEB032 (diluted serially in lambda DNA (Promega)) were added, and qPCR was performed (BioRad CFX Connect Real Time System). Copy numbers were calculated using BioRad CFX manager (v3.1), and data were analysed in Microsoft Excel (2016) to generate copies per 1,000 copies of rpoA. Data were graphed and statistically analysed in GraphPad Prism (v9.0.2).

Structural modelling

DacS and VanS were modelled as dimers using AlphaFold [26] and the resulting output files were visualised using ChimeraX [68].

Statistics

Statistical analysis was performed in Graphpad Prism (v9.0.2), and P < 0.05 was considered significant. Data are presented as mean ± SD, unless otherwise stated. Differences in growth were analysed using the R package GrowthCurver outputs. A cross-correlation was performed in RStudio using Hmisc and corrplot packages [69,70], which determined area under curve (AUC-E) to be the most representative measure of growth. The differences in AUC-E, compared to the control strain curves, were calculated using t tests with Welch’s correction. To test differences in sporulation efficiency, AUC was chosen as a representative measure of sporulation across all time points. AUC was calculated in GraphPad Prism (v9.0.2), and the difference between AUC for P30 isolates was compared with the WT using a one-way ANOVA with Brown–Forsythe and Welch’s correction. Statistical analysis of cell length was performed in GraphPad Prism (v9.0.2). Fiji (v2.9.0) MicrobeJ (v5.13l) cell length outputs for P30 isolates were compared to the WT using a one-way ANOVA with Brown–Forsythe and Welch’s correction to test for differences in cell length. To determine whether differences in gene expression between strains were statistically significant, a two-way ANOVA with the Tukey–Kramer test was performed in GraphPad Prism (v9.0.2). In the experiments that quantified vancomycin-BODIPY binding, the function “fishertest” in Matlab (vR2023a) was used to compare the different strains/treatments using a Fisher’s exact test.