The S-layer contributes to severe disease in the murine model of C. difficile

In a murine model of infection, loss of body weight offers a strong correlative measure of C. difficile disease severity [6, 33, 35]. Infection of antibiotic pre-treated mice with strain R20291 resulted in significant weight loss of up to 15%, peaking between 24 and 48 h post-infection (hpi). In contrast, mice infected with the S-layer deficient derivative strain FM2.5 showed consistently and significantly less weight loss than R20291, peaking at around 6% at 48 hpi, which is not statistically significant different to that determined in the PBS control group (Fig 1a). It is worth noting that, despite this modest weight loss, animals infected with FM2.5 failed to return to their pre-infection weight, even after the R20291-infected mice had fully recovered (96 hpi, Fig 1a). Although measurement of total C. difficile in faecal material showed comparable levels of total numbers of bacteria shedding at 24 and 72 hpi (Fig 1b), the recovered number of spores was significantly lower in animals infected with FM2.5 at 24 h. This suggests that while FM2.5 appears to be growing in the lumen, it is not sporulating as efficiently as R20291; as previously reported [20]. In addition, a significant reduction in both total and spore counts were also observed at 24 hpi in both caecal (Fig 1c) and colonic (S1 Fig) luminal contents from mice infected with FM2.5 compared to R20291, indicating that this strain was less prevalent. In contrast, by 48 hpi, comparable numbers of both FM2.5 and R20291 (total and spore counts) were recovered from both luminal contents. Equivalent recoveries were also observed in faecal material assessed at 72 hpi and in luminal contents at 96 hpi. It should be noted that acute disease, typified by diarrhoea at 48 hpi, made comparative analysis of faecal material unfeasible at this time point.

Download:

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

Fig 1. SlpA deficient C. difficile FM2.5 causes less severe disease in a murine model of infection.

Female C57/Bl6 mice were challenged with spores of R20291 (green) or FM2.5 (purple), or mock infected with sterile PBS (grey). (a) Weight loss was monitored every 24 h for four consecutive days following infection. Each point is the average of several replicate experiments (n>3), with at least 10 animals per time point. Infection with R20291 strain results in statistically significant weight loss at 24 and 48 hpi, compared to both FM2.5 and PBS, while no significant difference was observed between FM2.5 and PBS at these time points. Mice infected with FM2.5 showed a slightly, but statistically significant, lower weight at 96 hpi, compared to the PBS control group. (b) CFU ml^-1 of total (no pattern) and spore recovery (pattern) from faecal material collected at 24 hpi (n = 8, R20291; n = 15, FM2.5) and 72 hpi (n = 7, R20291; n = 14, FM2.5). Variation in sample numbers reflects either difficulties associated with collection from infected mice due to diarrheal symptoms, or loss of individual animals due to disease severity. (c) CFU ml^-1 of total (no pattern) and spore recovery (pattern) in caecal contents at 24, 48 and 96 hpi (n = 5 at each time point except R20291 at 24 hpi; n = 4). (d) Toxin activity of caecal content at 24, 48 and 96 hpi; through challenge of Vero cells in vitro (n = 5 at each time point). Results displayed indicate the reciprocal of lowest dilution at which toxin activity could be measured. (e) Histological scoring of sections of the caecum from mice challenged with PBS, R20291 or FM2.5. Results displayed are the mean ± SEM of sections from at four mice per group (indicated by different symbols) with three sections scored from each tissue. (f) Histopathological sections representing caecal (i, ii and iii) sections following challenge with PBS (i); R20291 (ii); or FM2.5 (iii). Scale bars represent 100 μm. Statistical tests were conducted using GraphPad Prism software v.12. Statistical significance is indicated: ns—not significant; *p < 0.05; **p < 0.01; and ***p < 0.001.

https://doi.org/10.1371/journal.ppat.1011015.g001

Interestingly, mice infected with R20291 that survived infection showed full recovery by 96 hpi, returning to pre-infection weights, equivalent to those of non-infected mice. In contrast, FM2.5-infected mice failed to return to their pre-infection weight even when animals were monitored for a further five days (9 days pi) despite animals remaining asymptomatic, with no evidence of loose faeces.

Assessment of in vivo toxin production showed that less toxin was detected in mice infected with FM2.5 in both the caecum (Fig 1d) and in the colon (S1b Fig) at 24 hpi compared to R20291. However, at 48 and 96 hpi, comparable levels of toxin were recovered from all infected animals. Histological examination at 48 hpi of caecal tissue from R20291 infected mice showed greater tissue damage and inflammation than those infected with FM2.5 or control mice challenged with PBS. Cumulative scoring of caecal and colonic tissue histology are shown in Fig 1e and S1c Fig, indicating changes are significant in caecal and lower colon tissue, and show a non-significant trend in the upper colon. Typical images of tissue damage are also displayed in Fig 1f (caecum) and S1d Fig (lower colon).

In vivo pressure drives selection for S-layer variants

When grown on selective chromogenic agar (ChromID; BioMerieux), the morphology of R20291 presents with the typical ‘fried egg’ C. difficile colony, visible after approximately 16 h of incubation (Fig 2a). In contrast, FM2.5 produces smaller and smoother colonies, which take ~24 h to emerge (Fig 2b). Following infection of mice with R20291 and FM2.5, faecal material was recovered and plated daily. During examination of resultant colonies, it was noted that, while colonies from R20291 infected mice showed the expected morphology, material retrieved from FM2.5-infected mice showed a mixture of both large (FM2.5_large) and small colony types (FM2.5_small) (Fig 2b). Additionally, FM2.5_large, were countable after 16 h incubation, while the expected FM2.5-like colonies, FM2.5_small, were only observable from 24 h. Several colonies from both types, FM2.5_small and FM2.5_large, were streaked from the original plates and sub-cultured twice to ensure clonality. Colony morphology of individual clones remained consistent on subsequent subcultures, suggesting this was not a phase variable trait. Individual clones were then stored at -80°C.

Download:

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

Fig 2. Recovery of S-layer variants following in vivo challenge.

Following challenge with spores of R20291 and FM2.5, faecal material was recovered and plated on C. difficile selective chromogenic agar (Biomerieux). (a) Colonies of R20291. A dashed red box indicates the area enlarged to show colony morphology in the panel below. (b) The two colony types of FM2.5. An equivalently sized dashed red boxed area has been enlarged and colonies from each plate are shown below. FM2.5 is indicative of the typical FM2.5-like, smaller colony morphology, while the FM2.5_variant is representative of the larger colony type. (c) Sequencing of a region of slpA shows the nucleotide and protein sequence of SlpA_R20291 (green); the nucleotide insertion (red) in SlpA_FM2.5 (light purple), introducing a stop codon that results in premature translational termination, leading to truncation of the SlpA protein. A single nucleotide deletion (light blue) in the SlpA_FM2.5 sequence, results in the modification of 3 amino acids (shown in light blue) in SlpA_varA, (dark blue); a five-nucleotide insertion (light crimson) in SlpA_FM2.5 sequence results in modification of 13 amino acids (shown in light crimson) in SlpA_varB (crimson). These modifications in the sequence of SlpA_varA and SlpA_varB, restore the reading frame of slpA, allowing transcription and translation of the entire protein. (d) SDS-PAGE analysis of surface layer proteins extracted by low pH preparation. Lane 1: MW Marker; Lane 2: R20291; Lane 3: FM2.5; Lane 4: FM2.5_varA; Lane 5: FM2.5_varB. (e) Western immunoblot analysis using an anti-SLP_H antibody. (f) Western immunoblot analysis using an anti-SLP_L antibody. Bands corresponding to SLP_L and SLP_H indicated with blue and gold arrowheads, respectively. (g) Relative proportion of SlpA_FM2.5 and SlpA_varC sequences in samples analysed in both the spore preparations used for mouse inoculations, and in faecal samples recovered up to 10 days post infection. Raw images are provided in S3 Fig.

https://doi.org/10.1371/journal.ppat.1011015.g002

Colonies of different sizes were observed in several independent animal experiments, from which several clones were recovered. Amplification of the slpA sequence from these clones revealed that the FM2.5_large colonies contained modifications in the genomic sequence upstream of the FM2.5 mutation site (single nucleotide insertion, Fig 2c). The first variant identified FM2.5slpA246delT (subsequently referred to as FM2.5_varA) showed a single nucleotide deletion (246delT), which restored the original reading frame and rescued translation of the full SlpA; modifying three amino acid residues in the translated protein. An insertion of five-nucleotides was observed in a second variant, FM2.5slpA249insCTTAG (subsequently referred to as FM2.5_varB; 249-253insCTTAG), resulting in the modification of 13 amino acids within the mature protein (Fig 2c). Genomic sequencing of individuals clones of these variants confirmed that they were closely related to FM2.5 and were not unrelated strains carried as contaminants by the mice (S2 Fig).

To confirm restoration of SlpA expression in these strains, low pH cell surface extracts of FM2.5_varA and FM2.5_varB were analysed by SDS-PAGE. This confirmed that both SLP_H and SLP_L proteins were present in R20291, absent in FM2.5 but restored in FM2.5_varA and FM2.5_varB (Fig 2d). The proteins were confirmed as SLP_H and SLP_L by western immunoblot analysis using anti-SLP_H and anti-SLP_L antibodies (Fig 2e and 2f).

Interestingly, these and other variants were also identified in subsequent in vivo experiments, even when animals were infected with batches of independently prepared spores. This reproducible recovery of S-layer variants in vivo raised the possibility that low numbers of genetic variants, present within the FM2.5 population, were being amplified within the in vivo environment. To test this hypothesis, we undertook amplicon sequencing of slpA in a preparation of FM2.5 spores used to infect mice. and from faecal material recovered from these animals at 24, 48, 72 and 96 hpi (Fig 2g,). Analysis of this unbiased amplification of slpA amplicons from faeces from these mice, revealed a third variant, FM2.5slpA252delA (subsequently named FM2.5_varC), in which the original frameshift mutation was corrected by deletion of the extra nucleotide (252delA). This variant was also identified, albeit at a low proportion of the population (< 5%, Fig 2g), in the spores used for inoculation of these animals. Neither of the previous variants FM2.5_varA or FM2.5_varB were detected in either the spore or faecal samples tested, although this does not exclude the possibility that these or others were present in this preparation but that detection was beyond the discretionary power of the sequencing data. Based on our overall observations, each spore preparation is likely to contain low numbers of different variants, clones of which outgrow and outcompete the mutant within the population in vivo. However, isolation of variants as early as 24 hpi (> 94% of the population) suggests that expression of an intact S-layer provides a competitive advantage in vivo for these clones which express an intact S-layer.

Sequence modifications can affect SlpA structure and assembly

To understand the effects of the detected variations in amino acid sequence on SlpA structure and S-layer assembly, crystallisation of SlpA_R20291, SlpA_varA and SlpA_varB was carried out. Although crystals were obtained for all strains, only SlpA_varB crystals were of sufficient quality for X-ray diffraction data collection and structural determination by molecular replacement, using previous SlpA structures as models. As we have so far been unable to obtain an experimental structure of the complete SlpA_R20291, our previously described variant—SlpA_RΔD2—(PDB ID: 7ACZ), which lacks the most exposed region of SLP_L [24], was used in our comparisons as a model of the S-layer protein in R20291. SLP_H and the interacting domains were easily traceable in the electron density but D1 was only partially built, whilst density for domain D2 was very poor and this region could not be traced in the final SlpA_varB model (Fig 3a, PDB ID: 8BBY and S1 Table). This implies that D2 is flexible and/or unstructured, while the structure of the core domains required for S-layer assembly—SLP_H, and, to a lesser extent, D1 and LID/HID [24]—seems to be generally maintained. However, the relative orientation of these domains in the SlpA_varB molecule is altered (Fig 3a), with D1 and the interacting domains rotated towards the SLP_H plane by ~30 ° (Fig 3a). The 13 altered residues in α2_L in SlpA_varB result in disruption of the α-helix secondary structure and introduce disorder in the upstream loop that links the preceding β-strand (β3_L) and α2_L. It is worth noting that SLP_H in R20291, which belongs to SLCT 4, has several insertions within the cell wall binding 2 (CWB2) sequence motifs that define CWPs in C. difficile, when compared to other SlpA types. These insertions could not be traced in our previous SlpA_RΔD2 model [24] but were traceable in SlpA_varB and result in several loops protruding above the SLP_H plane, towards the environment, partially occluding the CWB2 motifs (Fig 3a, right). Together with the movement of the interacting domains and D1 towards the SLP_H tiles, this creates a more compressed arrangement (~66 Å thickness compared to ~76 Å in SlpA_CD630, PDB ID: 7ACY, Fig 3b, bottom).

Download:

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

Fig 3. Structure of SlpA_varB shows a different assembly arrangement.

(a) Structural model of SlpA_varB (SLP_L—pale red, SLP_H—slate blue, PDB ID: 8BBY), superimposed on the R20291-derived SlpA_RΔD2 model (SLP_L—gold, SLP_H—slate blue, semi-transparent), with the rotation angle of the D1 and LID/HID domains shown by an arrow. Three distinct structural features are observed: SLP_H, LID/HID and D1. Cartoon representation of the SLP_H/SLP_L (H/L) complex, as seen from the environmental side (left) and side view (right). Sequence of α2_L, with paler colours indicating differences, is shown schematically. (b) Cartoon representation of the H/L planar array (PDB ID 8BBY, interacting molecules coloured and viewed as in a). (c) 2D schematic of H/L complex crystal packing in SlpA_varB (top), SlpA_RΔD2 (centre) and SlpA_CD630 (bottom), indicating the interaction network linking a single H/L (slate blue/crimson or slate blue/gold) complex with neighbouring molecules in a planar arrangement generated by SLP_H tiling. The missing D2 in the SlpA_varB model is represented as dashed lines. Notably, D1-D1 interactions seen in other models are missing in SlpA_varB and the SLP_H tiles are shifted, with new HID-CWB2₃ interactions stabilising the lattice. Array is depicted as seen from the extracellular environment, with symbols representing key interaction types in the crystal lattice, detailed in S2 Table.

https://doi.org/10.1371/journal.ppat.1011015.g003

In the previously described crystallographic models of SlpA_CD630 (PDB ID: 7ACY), SlpA_R7404 (PDB ID: 7ACX) and SlpA_RΔD2 (PDB ID: 7ACZ), α2_L was responsible for closing a gap between neighbouring SlpA molecules via D1-D1 interactions [24]. In the crystallographic model of the R20291-derived SlpA_RΔD2 variant, D1-D1 interactions are mediated by hydrogen bonds between S50_L-S50_L and Q70_L-A49_L from neighbouring molecules.

In the SlpA_varB structure, disruption of α2_L and reorientation of D1 and LID/HID relative to SLP_H leads to changes in the interactions between neighbouring molecules and, consequently, a rearrangement of the S-layer array. Strikingly, the D1-D1 interactions seen in previous models were not observed here, possibly due to the flexibility of α2_L and preceding loop caused by the changes in the variant sequence leading to a different orientation of D1 domains. Unlike in the previously determined structures, neighbouring D1 domains in the SlpA_varB structure are too far apart to mediate contacts (> 12 Å). In the previous models, SLP_H tiling creates two wide channels, which are stabilised by interactions mediated by the interacting domains and D1 [24]. A different mode of stabilising the SLP_H tiling is observed in SlpA_varB, with the interacting domains now partially inserted in those cavities (Fig 3b and S4a Fig). A new interacting interface between HID from one molecule and CWB2₃ from a neighbouring molecule occludes these gaps and stabilises the S-layer lattice (Fig 3c and S4a Fig). This new arrangement of the crystal lattice is in line with our proposed assembly model, where the S-layer 2D array is maintained mostly by hydrogen bonds and salt bridges across surfaces with complementary charges [24], largely dependent on SLP_H-SLP_H interactions and stabilised by varying degree of interactions involving SLP_L (Fig 3c and S2 Table). The structural model of SlpA_varB confirms that changes in SLP_L can be accommodated with minor structural changes to the (sub)domains, by exploring flexible loops and hinges to provide a stable S-layer.

As no crystal data was obtainable, we initially used AlphaFold 2 [36] to predict models for SlpA_R2021 and SlpA_varA. However, in the absence of a specific template, the predicted models show considerable variations of the relative orientation of SLP_L and SLP_H or some of the proteins sub-domains. The observed orientations are not compatible with our knowledge that SLP_H faces the cell wall and SLP_L the environment [24, 37]. This misorientation was also observed in AlphaFold 2 predictions of our experimental structures of SlpA_CD630 and the SlpA_varB and therefore we did not proceed with AlphaFold 2 predicted models. Instead, we used the SWISS-MODEL server, either without a template input, or using SlpA_RΔD2 [24] or the SlpA_varB structure determined here. Depending on which template was used (automatically selected SlpA_R7404, PDB ID: 7ACX; SlpA_RΔD2 or SlpA_varB), different predicted structures of SlpA_varA were obtained, varying mostly in the orientation of D1 and interacting domains relative to SLP_H (S4b Fig). Interestingly, one common feature was that the changes resulting from the altered sequence seem to be accommodated not by altering the α-helix but by varying the length of the upstream loop that links the preceding β-strand (β3_L) and α2_L (S4b Fig). It is therefore unclear if SlpA_varA is more likely to adopt a R20291-like as observed in the SlpA_RΔD2 model or SlpA_varB-like S-layer assembly, as both can accommodate the modified sequence.

In vivo S-layer selection is independent of toxin expression

FM2.5 has previously been reported to show a delay in toxin production [20], consequently we chose to investigate whether recovery of an intact S-layer was a toxin-dependent, or toxin-independent event. To answer this question, mice were infected with FM2.5ΔPaLoc, in which the Pathogenicity Locus (PaLoc), encoding toxins A and B, had been deleted. These animals, in contrast to those infected with FM2.5, showed no weight loss over the 96 hours of infection (Fig 4a) and as expected, no toxin was observed in samples from the caecum or colon (S5a Fig). Total numbers of bacteria (S5b Fig) and spores (S5c Fig) recovered from the caecum and colon of these mice were comparable at 96 hpi to that observed in animals infected with FM2.5.

Download:

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

Fig 4. In vivo challenge of mice with FM2.5ΔPaLoc.

Female C57/Bl6 mice were challenged with spores of R20291 (green), FM2.5 (purple) and FM2.5ΔPaLoc (blue) or PBS (grey). (a) Weight loss was monitored every 24 h for four consecutive days following infection. Each point is the average change of weight, calculated from a minimum of 5 mice per group. (b) Comparison of the sequences of slpA shows the sequence for R20291 (green) and the reported insertion in FM2.5 (red) responsible for the truncation of the SlpA protein; deletion of four nucleotides (orange) leads to changes in four amino acids (light orange) in the sequence in FM2.5ΔPaloc_varD (orange). For comparison, the intact sequence of slpA from R20291 is shown in green. Statistical tests were conducted using GraphPad Prism software v.12. Statistical significance is indicated: ns—not significant; *p < 0.05; **p < 0.01; and ***p < 0.001.

https://doi.org/10.1371/journal.ppat.1011015.g004

Interestingly, S-layer variants were also recovered from these mice, with sequence modifications in the same region of slpA as previously identified. This variant FM2.5ΔPalocslpA237delTTAT (subsequently referred to as FM2.5ΔPaloc_varD) had four nucleotide deletions (Fig 4b; A237delTTAT). These changes also restore an intact SlpA, indicating that any potential selection advantage is independent of toxin production.

The variant strains display differing levels of virulence in mice

To assess whether recovery of SlpA by these variants correlated with rescued virulence, mice were infected with spore preparations of FM2.5_varA and FM2.5_varB, alongside R20291 and FM2.5 (Fig 5a). Interestingly, infection with FM2.5_varA resulted in significant weight loss within the first 48 h of infection, similar to that observed in mice infected with R20291. Mice infected with FM2.5_varB, in contrast, showed a limited pattern of weight loss, like those animals infected with FM2.5, which stabilized from 48 hpi.

Download:

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

Fig 5. Functional analysis of FM2.5_varA and FM2.5_varB in vivo and in vitro.

(a) Female C57/Bl6 mice were challenged with spores of R20291, FM2.5, FM2.5_varA, FM2.5_varB or mock infected with sterile PBS. Weight loss was monitored at the same timepoint each day for four consecutive days following infection. Each point shows the average weight change for a minimum of 5 animals per treatment. (b) In vitro toxin activity as measured through challenge of Vero cells. Samples were prepared by filtering supernatant following growth of C. difficile in vitro for 36 or 72 h; activity was measured by the challenge of Vero cells. Supernatants were harvested at the same phase of growth for each strain. Cells treated with PBS or with purified Toxin B were included as negative and positive controls respectively. OD₆₀₀ represents the optical density of Giemsa stain incorporated and released from intact Vero cells, hence high OD represents limited toxicity. Results displayed are the mean ± SEM of at least three independent replicates. Statistical tests were conducted using GraphPad Prism software v.12. Statistical significance is indicated: ns—not significant; *p < 0.05; **p < 0.01; and ***p < 0.001.

https://doi.org/10.1371/journal.ppat.1011015.g005

To determine whether these differences were associated with changes in toxin production, the variants were cultured in vitro for up to 72 h. Filtered spent growth medium removed from the cultures at 36 and 72 hpi were used to determine the production of toxin B. Both FM2.5_varA and FM2.5_varB produced levels of toxin comparable to R20291 at these time points (Fig 5b), with toxin-mediated damage to Vero cells resulting in cell rounding, cellular loss and reduced levels of staining with Giemsa. In agreement with previous reports [20], FM2.5 produced significantly less toxin than R20291 at 36 h, although toxin production levels were comparable in all strains by 72 h.

Modification of the S-layer results in large changes in gene expression

To gain a greater understanding of the differences in gene expression between R20291, FM2.5 and variant strains, comparative RNA-Seq analysis was conducted following in vitro growth. SlpA was the most highly expressed transcript in all strains (R20291, FM2.5, FM2.5_varA and FM2.5_varB) which is unsurprising given that the S-layer is comprised of approximately 600,000 copies of SlpA. In FM2.5, slpA is transcribed normally but is not translated to a full-length protein due to the frame-shift early in the open reading frame. Consequently, as transcripts are generated to an equivalent level, this gene does not appear in our list of differentially expressed genes. In contrast, comparative analysis of transcripts generated by R20291 and FM2.5 revealed over 287 differentially expressed genes (DEGs) (Fig 6a), linked to alterations in metabolism, transport, membrane integrity and sporulation (Fig 6b). In contrast, less differences were observed when R20291 was compared to FM2.5_varA (44 DEGs), than FM2.5_varB (185 DEGs), which showed similar numbers of DEGs to FM2.5. This correlates well with the observed phenotype of these strains within animals—FM2.5_varA associated with wild type like disease and FM2.5_varB with FM2.5-like attenuation. Analysis of these data suggest that differences in observed disease severity could be linked to changes in transcription of several virulence-associated traits, including toxin A and B, and genes associated with sporulation (Fig 6c). While in the case of FM2.5_varA, the recovery of an intact S-layer would appear to be sufficient to restore wild type gene transcription, only partial transcription profile restoration, including toxin expression, was observed in FM2.5_varB. However, as toxin activity was observed at 36 h in culture, the alterations in transcription control seen in FM2.5 and FM2.5_varB would appear to be limited to timing rather than absolute prevention of toxin production.

Download:

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

Fig 6. Global transcriptional differences between isolates of C. difficile following in vitro growth.

Analysis of mRNA recovered from in vitro grown cultures of C. difficile isolates R20291, FM2.5, FM2.5_varA and FM2.5_varB. (a) Total number of differentially expressed genes (DEGs) between experimental groups are highlighted in turquoise (upregulated) and orange (downregulated). (b) DEGs from experimental comparison of R20291 and FM2.5 were categorised based on function. (c) Transcriptional differences in select genes of FM2.5, FM2.5_varA and FM2.5_varB relative to R20291. Lists of all identified differentially expressed genes are provided (S1 Spreadsheet).

https://doi.org/10.1371/journal.ppat.1011015.g006

Taken together, these data suggest that expression of the S-layer plays a key role in C. difficile gene expression which contributes to disease severity within the host.

Bacterial strains and growth conditions

The bacterial strains used in this study include C. difficile strain R20291, its derivative FM2.5 [20], FM2.5_varA, FM2.5_varB, FM2.5_varC, FM2.5ΔPaLoc and FM2.5ΔPaLoc_varD (this study). Strains were routinely grown under anaerobic conditions on Braziers cycloserine, cefoxitin egg yolk (CCEY) agar (Oxoid, UK); CHROMID C. difficile Chromogenic medium (BioMérieux); or in Tryptone yeast (TY) broth (Oxoid, UK).

Generation of FM2.5ΔPaLoc

Homologous recombination was used to generate a derivative of strain FM2.5 that lacked the entire pathogenicity locus (PaLoc). Briefly, 1.2 kb up and downstream of the PaLoc was amplified by PCR using oligonucleotides RF920 and RF921, and RF922 and RF923 (S3 Table), respectively, and cloned by Gibson assembly into plasmid pJAK112 [46] that had been linearised by PCR using RF311 and RF312. The resulting plasmid, pJAK143, was then conjugated into C. difficile [47] and mutagenesis to knock out the PaLoc was carried out using standard allele exchange [4].

Murine model of infection

All procedures were performed in strict accordance with the Animals (Scientific Procedures) Act 1986 with specific approval granted by the Home Office, UK (PPL 60/8797 and PPL PI440270). These applications were also considered and approved by the University of Glasgow Animal Welfare Ethical Review Body (AWERB). Food and water were provided ad libitum and animals kept at a constant room temperature of 20–22 °C with a 12 h light/dark cycle. Groups of up to six C57/bl6 female mice aged 6–8 weeks supplied by Charles River (Edinburgh) were used in each treatment group. All animals were screened for colonisation with C. difficile in advance of the start of each experiment by plating faeces on ChromID selective media (Biomeuriex). An antibiotic cocktail (kanamycin, 0.40 mg ml^-1; metronidazole, 0.215 mg ml^-1; colistin, 850 U ml^-1; gentamicin, 0.035 mg ml^-1 and vancomycin, 0.045 mg ml^-1 [all Sigma Aldrich, UK]) was administered ad libitum in the drinking water as previously described [35] with clindamycin sulphate (150 mg Kg^-1), administered by oral gavage following cessation of the antibiotic cocktail. Animals were each challenged with approximately 10⁶ spores of C. difficile 72 h after clindamycin treatment. Mice were monitored closely post-infection and weighed daily to determine the severity of the disease. Animals with a weight loss greater than 10% of pre-challenge weight were given soft food and were culled if weight loss reached 20%.

C. difficile shedding and organ colonization

Fresh faecal samples collected daily were weighed, serially diluted in phosphate buffered saline (PBS) and cultured on CCEY agar at 37 °C for 48 h. At the experimental endpoint, animals were culled, and the caecum and colon harvested. Enumeration of total counts and spore-specific counts in lumen associated material were performed as previously described [35]. In brief, total viable counts were determined by plating serial dilutions on ChromID selective media. Spores were enumerated following standard heat treatment of the samples at 65 °C for 20 min [33].

Quantification of toxin expression

Quantification of toxin activity was performed using monolayers of Vero cells (kidney epithelial cells) as described previously [48]. Briefly, toxin was recovered from the spent filtered TY medium used to support bacterial growth for 36–72 h. Spent medium was recovered at the same stage of the growth cycle and at the same OD_600nm. In vivo toxin activity was measured by filtering using a 0.2 μm filter, luminal content collected from the caecum and colon of infected mice. Luminal contents collected from the caecum and colon of uninfected mice served as negative controls. Samples for toxin measurement were tested by the addition of serial dilutions to confluent monolayers within 72 h of collection. Cells and toxin were co-cultured for 24 h before cells were washed with phosphate buffered saline (PBS), fixed with 5% formal saline (Fisher), and stained with Giemsa for 30 min, before thorough washing to remove excessive stain. For data presented in Fig 1d and S1b and S5c Figs, toxin activity was determined as the reciprocal of last dilution in which toxin activity was observed, i.e., showing cell destruction. In Fig 5, to quantify the toxin activity more precisely, excess stain was removed by washing with PBS before cells were permeabilised to release internalised stain using 200 μl 1% SDS. 100 μl of the supernatant was transferred to a fresh plate and the OD_620nm values determined. In this context, higher values indicate Vero cells are intact and unaffected by the toxin, lower values indicate that toxin mediated damage prevents uptake and retention of the dye. Cells treated with PBS only or with 0.0005 μg ml^-1 purified Toxin B (Sigma, UK) were included as negative and positive controls respectively.

Histology and immunohistochemistry

Tissue samples were harvested from the caecum and colon of antibiotically susceptible animals infected with either R20291 or FM2.5 at post-mortem, 48 hpi. These tissues were gently washed in sterile PBS and immediately fixed in 10% formalin. Embedded tissue sections were cut and stained with Hematoxylin and Eosin [35]. Blind histological scoring of tissue was performed on three independent sections of caecal and colonic tissue from each mouse. Each section was scored out of a total of 15, with a score of 0 indicating no change, 1 slight change, 2 moderate change and 3 significant change, for the following categories: epithelial damage, neutrophil migration, haemorrhagic congestion, tissue oedema and crypt hyperplasia. Data presented represents the mean scores calculated for four mice for caecal, and three mice for colon assessment for each treatment.

slpA sequencing from isolated variant clones

Individual clones of bacteria, recovered from faecal or tissue associated material that showed different morphology on ChromID plates, where subject to at least two rounds of clonal selection. Genomic DNA was isolated from a 20 ml culture grown anaerobically for 18 h in tryptic soya broth (TSB). Bacterial cells were initially disrupted enzymatically by resuspending the pellet in lysis buffer (20 mM Tris-Cl, pH8.0, 2 mM Na EDTA, 1.2% Triton X-100, lysozyme 200 mg ml^-1), and incubated at 56 °C for 90 min. The DNA was recovered using the DNeasy Blood and Tissue Kit (Qiagen), following manufacturer’s instructions. A 478 bp fragment of slpA, centred on the FM2.5 point mutation [20], was amplified by using oligonucleotides RF110 and RF111 (S3 Table). The resulting fragments were subjected to Sanger sequencing and compared to wild type and FM2.5 sequences.

Isolation and sequencing of slpA from faecal extracts

Faecal samples were also collected for unbiased sequencing of slpA within the population, by directly extracting DNA from faecal samples using the FastDNA SPIN kit for soil (MP Biomedicals). Briefly, approximately 200–600 mg of faeces sample were suspended in 978 μl sodium phosphate buffer with 122 μl MT buffer lysis solution. Samples were then homogenised in a FastPrep instrument using two 30 second pulses, at speed setting 6.5, in lysing matrix E. Samples were centrifuged for 10 min at 14,000 x g to remove debris. 250 μl protein precipitation solution was added to the lysate supernatant, and the precipitant formed removed by centrifugation at 14,000 x g for 5 min. DNA was then bound to a silica matrix, washed using the kit wash buffer, and eluted with water. For analysis of slpA sequences in spores used for challenge experiments in mice, samples were treated with 0.1% taurocholic acid for 60 min to induce germination and then boiled for 5 min in the presence of Chelex resin to release DNA. DNA extracted from either faeces or germinated spores was used as a template for PCR amplification of a 330 bp fragment of slpA using Phusion polymerase (NEB) and oligonucleotides RF2193 and RF2194 (S3 Table). Resulting DNA fragments were purified and sequenced using the amplicon-EZ service offered by GENEWIZ (Azenta Life Sciences).

Extraction and western immunoblot analysis of S-layer and associated proteins

Surface layer proteins were extracted using low pH glycine as previously described [49] and analysed by SDS-PAGE using standard methods [50] (Fig 2 and S3 Fig). Proteins were transferred to nitrocellulose membranes via semi-dry transfer (Bio-Rad Trans Blot Turbo; 25 V, 30 min) for western immunoblot analysis. Transfer efficiency was confirmed by PonceauS staining of membrane post transfer, and Coomassie staining of the polyacrylamide gel following transfer. Membranes were blocked for 1 h in Phosphate-buffered Saline containing 0.1% Tween20 (PBS-T) with 5% milk powder. Blots were subsequently incubated in primary antibody (rabbit anti-SLP_H raised against C. difficile 630 1:100,000 dilution; rabbit anti-SLP_L raised against C. difficile R20291 1:200,000 dilution) in PBS-T containing 3% milk powder, for 1 h at room temperature. Membranes were washed thoroughly in PBS-T before incubation with secondary antibodies (anti-rabbit horseradish peroxidase, Promega WB401B 1:2,500 dilution) for 1 h at room temperature. Blots were washed in PBS-T before detection by chemiluminescence (Bio-Rad). Molecular weight (MW) markers (Thermo Scientific 26616) were imaged (Bio-Rad ChemiDoc XRS+) simultaneously and overlaid onto the blots to aid visualisation. Original full blots are shown in S3 Fig.

Whole genome sequence analysis

Genomic DNA recovered from R20291, FM2.5, FM2.5_varA, FM2.5_varB were sequenced by MicrobesNG (Birmingham UK) using Illumina based technologies. The raw data was then aligned to the reference genome (NC_013316.1) using Bowtie2. Polymorphisms were called using FreeBayes under default settings and VCF file was filtered to include only mutations identified with a read depth of > 50 and in at least 50% of reads. To build the tree (S2 Fig), the filtered mutations across all strains were merged and converted into a bed file. Regions were expanded by 100 bp on each side and these were considered the mutable regions. The phylogenetic tree was constructed using PhyloTree (https://github.com/samandmac/PhyloTree), which uses BLAST and PHYML to generate a tree in Newick format, and subsequently an R script which allows visualisation of the tree using the ggtree package. The raw sequence data has been deposited to the Gene Expression Omnibus; reference ID GSE205747. Note: file name Rv9 contains the genomic data for FM2.5_varA, and RV189 for FM2.5_varB respectively.

Protein purification and X-ray crystallography

C. difficile variant strains were cultured in 400 ml of TYG broth for 16 h. Cultures were then centrifuged at room temperature at 4,696 x g and resulting pellets were washed with 40 ml of 0.01 M HEPES pH 7.4 and 0.15 M sodium chloride (HBS) buffer. S-layer extraction was performed by resuspending the washed pellet in 4 ml of 0.2 M glycine-HCl pH 2.2 and centrifugation for 5 min at 21,100 x g. Collected supernatant was then neutralized with 2 M Tris-base. S-layer extract was filtered and resolved onto a Superdex 200 26/600 column using an ÄKTA Pure FPLC system (Cytiva) in 50 mM Tris-HCl pH 7.5, 150 mM NaCl buffer.

Purified SlpA_varB at 10 mg ml^-1 was subjected to crystallization using a Mosquito liquid handling robot (TTP Labtech), with the sitting drop vapor-diffusion method, at 20 °C. Crystals were obtained in 0.03 M magnesium chloride hexahydrate; 0.03 M calcium chloride dihydrate, 0.12 M ethyleneglycol, 0.05 M Tris (base); 0.05 M bicine pH 8.5, 20% v/v glycerol; 10% w/v PEG 4,000.

Data were collected on the I24 (λ = 0.71 Å) beamline at the Diamond Light Source Synchrotron (Didcot, UK; mx24948-136) at 100 K. Two datasets were collected from one crystal and initially processed by the automatic multi-crystal data-analysis software pipeline xia2.multiplex [51] within the Information System for Protein Crystallography Beamline (ISPyB), re-processed using Automatic Image Processing with Xia-2 (DIALS [52] and Aimless 3d [53]) and scaled with Aimless within ccp4.cloud of CCP4 [54] software suit.

The initial model of the core SLP_H was obtained by molecular replacement in Phaser [55], using an SlpA_varB model of CWB2 domains, derived from the SlpA_RΔD2 model (PDB ID: 7ACZ) and calculated using SWISS-MODEL [56]. The generated solution model was then subjected to automatic model building with Modelcraft [57], followed by manual building with Coot [58] and refinement in Refmac5 [59].

Final models were obtained after iterative cycles of manual model building with Coot and refinement in phenix_refine [60]. Data collection and refinement statistics are summarized in S1 Table.

PDBePISA [61] was used to investigate interdomain and protein-protein interfaces in the crystallographic lattice to identify interacting residues, which were confirmed by manual inspection within COOT.

Structural representations were generated using PyMOL Molecular Graphics System (Schrödinger, LLC).

Protein structure prediction

Homology models for SlpA_R20291 and SlpA_varA were generated by providing SWISS-MODEL webserver with the SlpA_varB (PDB ID: 8BBY) or SlpA_RΔD2 (PDB ID: 7ACZ) as user templates, as well as without a template. Structural alignments between predicted models and templates were performed using COOT [58]. As SlpA_RΔD2 lacks the D2 domain, predicted models based on this experimental model have a disordered D2 domain. Therefore, overall comparison of the three predicted models was based on models calculated in the default mode, which uses SlpA_R7404 (PDB ID: 7ACX) as a template, while analysis of the modification-containing region in the D1 domain was done using SlpA_RΔD2 or SlpA_varB derived models.

Recovery of mRNA for RNA sequence analysis

RNA was recovered from C. difficile strains R20291, FM2.5, FM2.5_varA and FM2.5_varB which had been cultured in vitro in TY broth. Briefly, bacterial cells reaching an OD_620nm = 0.6 were pelleted (5000 x g, 15 min) and immediately fixed in 1.5 ml RNA-protect (Qiagen) for 10 min before being processed using a PureLink RNA mini kit (Ambion) to extract total RNA. To ensure maximal lysis of bacteria and recovery of RNA, the bacterial pellet was additionally subject to treatment with 100 ml lysozyme solution (10 mg ml^-1 in 10 mM Tris-HCl, pH8.0, 0.1 mM EDTA), 0.5 ml 10% SDS solution and 350 ml of Lysis Buffer (Invitrogen PureLink RNA Mini Kit) containing 2-mercaptoethanol. Cells were then homogenised using MP Biomedical beads (0.1 mm) and bead beater (MP Biomedical FastPrep24) with cells subject to 2 cycles of 60 s beating, followed by incubation on ice for 2 min. Total RNA was then extracted using the standard PureLink RNA mini kit protocol, according to the manufacturer’s specifications. Genomic DNA was removed using a TURBO DNase kit (Ambion) and samples were tested for efficient removal of DNA by conventional PCR. Samples for RNA-Seq were prepared in triplicate on two separate occasions (6 samples for each) for all four bacterial strains.

Illumina library preparation of mRNA samples for RNA-Seq was prepared using a TruSeq Stranded mRNA library prep kit (Illumina) according to the manufacturer’s instructions. Sequencing was performed on the Illumina NextSeq 500 platform (75 bp length; single-end). Library generation, optimisation of amplification and sequencing were performed at the University of Glasgow Polyomics facility. Quality control of sequencing data was performed using FastQC (Babraham Bioinformatics) to assess the minimum Phred threshold of 20 and potential data contamination. The raw data has been deposited to the Gene Expression Omnibus reference ID GSE205747. Note: files with sample numbers 1–6 reflect samples generated and analysed independently for each strain; Rv9 was the original name of FM2.5_varA, and RV189 for FM2.5_varB, respectively.

RNA analysis and identification of differentially expressed genes

Raw RNA-Seq datasets were subject to the following pipeline. Firstly, fastQ files were assessed using FastP [62] and then were aligned to the C. difficile R20291 (accession number NC_013316) reference genome using STAR [63] (v2.6) with–quantMode GeneCounts–outFilterMultimapNmax 1 and–outFilterMatchNmin 35. We used a Star index with a–sjdbOverhang of the maximum read length − 1. Next, read count files were merged and genes with mean of < 1 read per sample were excluded. Finally, the expression and differential expression values were generated using DESeq2 [64] (v1.24). For differential comparisons, we used an A versus B model with no additional covariates. All other parameters were left to default.

The processed data was then visualised using Searchlight [65], specifying one differential expression workflow for each comparison, an absolute log₂-fold cut-off of 1 and adjusted p of 0.05. All other parameters were left to default.

Pathway analysis methods

Functional and metabolic pathways were implied by interrogation of the WP numbers assigned using the C. difficile annotated genome (NC_013316.1) and the Uniprot or NCBI Blastp databases. Gene ontology (GO) was assigned based on Biological Process assignment within Uniprot. Comparative analysis of all differentially expressed genes are provided (S1 Spreadsheet).

Analysis of phase variable switch orientation

To analyse the orientation of the seven C. difficile R20291 phase variable switches, orientation specific qPCR was carried out as previously described [44]. All strains were grown on BHI agar and mixed populations were prepared by pooling 100–200 colonies of each followed by genomic DNA purification using standard phenol-chloroform extraction and isopropanol precipitation. qPCR was performed using SYBR Green JumpStart Taq following the manufacturer’s instructions with 100 ng gDNA and 100 nM of each oligonucleotide (S3 Table). Reactions were run in a Bio-Rad CFX Connect RT-PCR instrument using the following parameters: 95 °C for 3 min, followed by 40 cycles of 95 °C for 10 sec, 60 °C for 15 sec, 72 °C for 15 sec and 64 °C for 10 sec, followed by a melting curve: 95 °C for 10 sec, 55 °C for 5 sec and 95 °C for 5 sec. Products were proportionally quantified using the rpoA gene as a reference and data presented as the percentage of the switch in the orientation of the published R20291 reference genome [66].

Statistical analysis

Statistical analysis was carried out in GraphPad Prism v.12. Tests used included for weight loss a two-way ANOVA with Tukey post test; for bacterial counts and toxin measurements, a two way ANOVA; for tissue histology analysis an ordinary one-way ANOVA. Statistical significance is indicated: ns—not significant; *p < 0.05; **p < 0.01; and ***p < 0.001.