Array comparative hybridisation reveals a high degree of similarity between UK and European clinical isolates of hypervirulent Clostridium difficile

Array verification and coverage

Forty thousand and ninety seven 60-mer probes were designed to cover the sequenced and annotated genome of C. difficile CD630. This essentially corresponded to a probe every approximately 200 bp. A further 687 probes were designed to extra genes in the preliminary 454 sequence produced for R20291 by the Sanger in 2007 and the available unannotated QCD-32g58 sequence. Additionally, 17 extra genes including the toxin genes, cwp66 and slpA were represented at high density by 346 oligonucleotides. Initial experiments were performed with a set of control strains that included CD630, R20291, R23052 and CD196 (R12087). A CD630 self-self hybridisation was also performed. Analysis of the data obtained showed that the genome of strain CD630 hybridised to 57 of the '027-specific' probes; and a BLASTN in silico analysis of the array oligonucleotides against the CD630, R20291 and QCD-32g58 sequences used to design the array showed that these oligonucleotides had highly significant matches in these strains. Accordingly, these oligonucleotides were excluded from further analysis.

Analysis of the remaining PCR-ribotype 027 oligonucleotides with genomic DNA of the control strains showed that these probes produced a positive signal with the DNA of PCR-ribotype 027 strains. Figure 1 shows a condition tree clustering for all the strains against all of the probes, with those representing CD0001 listed at the top, pCD630, the extra genes and finally the extra PCR-ribotype 027 genes at the bottom. Strains are grouped by PCR-ribotype and, on initial inspection, demonstrate that each PCR-ribotype exhibits a visually similar pattern of hybridisation. Additional files 1, 2, 3, 4, 5, 6, 7 and 8 presents a full list of the probes that are present or absent in each strain.

Core

Mobile elements

C. difficile is known to have a highly mosaic genome with many mobile genetic elements such as conjugative transposons and prophages [28]. Of the 1392 probes representing mobile or extrachromosomal elements in strain CD630, only 92 probes were present in the core of all the strains hybridised. Additional file 10 summarises the presence of the known CD630 mobile elements in each PCR-ribotype. In the majority of PCR-ribotype 027 and 001 strains, CTn1 is absent or highly divergent. It is absent or highly divergent in all PCR-ribotype 078 and 015 strains. CTn2 is absent from all of the PCR-ribotypes except from the PCR-ribotype 12 strains, CD630 and ECDC 012. CTn3 is absent or highly divergent from all PCR-ribotypes except 078 and 012. CTn3 or Tn5397 is the only known mobile C. difficile element containing erythromycin and tetracycline resistance [31]. Therefore resistance to these antibiotic classes in any of the strains tested, including R20291 and PCR-ribotypes 078 and 106 strains (which are resistant to erythromycin) must be provided by an as yet undefined genetic element or mutation [32, 33].

CTn4 is detectable in the PCR-ribotype 027 Quebec strain 23M63 but is absent or highly divergent in 27/28 of the PCR-ribotype 027 strains tested on the array utilised in this study. It is also partially present in one of the PCR-ribotype 001 strains tested but absent from all the other PCR-ribotypes. CTn5 is absent or highly divergent in 6 PCR-ribotypes; 001, 002, 014, 015 003 and 020. In all PCR-ribotype 017 strains only genes CD1864-9 are absent or highly divergent. These genes are also absent or highly divergent in 1-2 strains of the remaining 3 PCR-ribotypes; 027, 078 and 106. One PCR-ribotype 014 strain exhibits hybridisation between CD3330-44, but CTn6 is absent or highly divergent in all the other strains tested. Conversely Ctn7 is present in some form in all PCR-ribotypes except PCR-ribotypes 002 and 015. Prophage 1 is absent from all the strains tested except the PCR-ribotype 012 strains. Prophage 2 hybridises between CD2927-59 in all but PCR-ribotype 001, 002, 014, 015 and 078 strains.

Toxins

The C. difficile genome contains the PaLoc (pathogenicity locus) which harbours five genes (tcdABCDE) responsible for the synthesis and regulation of the two major virulence factors, toxins, TcdA and TcdB. Variation in this region is extensive and as a consequence toxinotyping is a frequently used molecular method used to discriminate between strains [34, 35]. Variable sequences include both the structural genes encoding the toxins, and the associated regulatory genes. Thus, the ability of some PCR-ribotype 027 strains to produce more of both toxins is attributed to a deletion at position 117 in the negative regulator of toxin production, tcdC[15, 16], leading to a truncated TcdC protein. The occurrence of similar deletions in other strains not generally associated with epidemics suggests, however, that such changes are not indicative of hypervirulence [20]. PCR-ribotype 027 strains are usually toxinotype III strains, whereas CD630 is toxinotype 0.

The array results confirm that tcdB is conserved among all PCR-ribotype 027 isolates examined and diverged in the 3' region of tcdC (the negative regulator of toxin production) as indicated by a lack of hybridisation to EXP_CD630_800001_805000_s_PSO-60-77, the last tcdC probe on the array. Naturally occurring toxin A-B+ strains cause diarrhoea and colitis in humans [36] and are generally PCR-ribotype 017 (toxinotype VIII). From the observed hybridisation obtained with our array, all of the PCR-ribotype 017 strains examined here lacked tcdA and exhibited divergence in tcdB when compared to the corresponding CD630 and SM probes (data not shown).

Some C. difficile strains also produce a third toxin in addition to TcdA and TcdB, a binary ADP-ribosyltransferase toxin encoded by cdtA and cdtB. The role of binary toxin in pathogenesis is unclear, although it has been linked to increased disease severity [2]. The genes cdtA and cdtB are conserved in PCR-ribotypes 027 and 078. Our hybridisation results agree with those previously reported for CD630, showing divergence in both of these genes which cause these genes not to be active in this strain [37]. PCR-ribotype 017 also displays similar results to previous publications and, shows limited hybridisation to some CD630 cdtA and cdtB reporters as concluded by Rupnik [35]. The results from this study for the other PCR-ribotypes examined show that this region is divergent.

Flagella and motility genes

Flagella are important in pathogenesis for many enteric pathogens including Campylobacter jejuni and Salmonella enterica serovar Enteridis [38, 39]. Chemotaxis and motility are inextricably linked and both are important for bacterial survival allowing the bacteria to move towards nutrients and away from substances that may prove detrimental.

Genes that allow for flagella modification by glycosylation have recently been described in C. difficile QCD-32g58 and R20291 upstream of the flagellar biosynthesis locus [32, 40]. Reporters representing 2 of the 4 loci (CDR0223 and 5) are present on the array and are conserved in all strains but two PCR-ribotype 017 strains (L22 and 23). Stabler et al[32] described the flagella related genes in 2 loci of the CD630 genome, F1; CD0226-40 and F3; CD0245-71 [29]. Loss of, or significant divergence in the F1 and inter-flagella region (F2; CD0241-4) was observed in PCR-ribotype 027 strains; this was shown to be due to 84-90% sequence identity in this region [32].

Our data shows that only 7/93 strains are divergent in these genes and this includes the two PCR-ribotype 017 strains discussed above, two non-toxigenic strains and two PCR-ribotype 078 strains. PCR-ribotype 078 strains have previously been reported to be non-motile [32] and although the CD630 flagella loci appears to be highly divergent or absent in these stains, the corresponding R20291 flagella and flagella glycosylation genes are present, indicating that another mechanism of variation is responsible for their non-motility.

Antibiotic resistance

Another contributing factor to the spread of C. difficile infection is the acquisition of antibiotic resistance. The genome sequence of CD630 allowed the identification of many genes associated with antibiotic resistance, including those already known such as ermB and tetM, and those with no prior experimental data, such as the putative lantibiotic antibiotic resistance genes (CD0478-CD0482, CD0820-CD0824 and CD1349-CD1352). In contrast to strain CD630, the epidemic 027 strains have been shown to be highly resistant to fluoroquinolones due to point mutations in the DNA gyrase genes which cannot be detected by this microarray [4, 29].

In agreement with previous array data, the lantibiotic resistance loci, CD0643-6 and CD01349-52 are absent or highly divergent in all the PCR-ribotype 078 strains tested and appear to be divergent in some of the tested PCR-ribotype 027 strains. The putative ABC transporter that confers daunorubicin resistance (CD0456) was absent from PCR-ribotype 078, 106 and 020 strains, but present in all others. The R20291 sequence showed that chloramphenicol resistance was conferred by CDR3461, part of the CTn027. The array shows that this gene or its homologue is present in all of our PCR-ribotype 027 and 001 strains, present in the majority of PCR-ribotype 078 strains and divergent in the remaining PCR-ribotype strains.

Regulatory systems

Regulatory genes form a large part of the C. difficile genome comprising 11% of the CD630 genome [28]. In Staphylococcus aureus, the agr quorum sensing locus (agrCABD) has been implicated as a key regulator of many virulence factors [41, 42]. In strain CD630 only homologues of agrD and agrB were present, respectively encoding a prepeptide of a secreted small autoinducer peptide and a transmembrane protein involved in AgrD processing. The homologous system in S. aureus also contains two further genes; agrC and agrA encoding a two-coponent system. Preliminary 454 sequencing of the PCR-ribotype 027 had shown that R20291 contained a second complete copy of an agr locus (agrCABD) in addition to the agrBD genes of strain CD630. Accordingly, oligonucleotides corresponding to this extra agrCABD locus was incorporated on our array at high density with an additional 25 probes.

Hybridisation against our array demonstrated that the extra agrCABD locus found in R20291 is entirely present in the genomes of 82 of the 94 (86%) strains tested, including two of four non-toxigenic strains (Figure 2). Additional file 11 details the presence, absence and divergence (signal around 1) for each probe in the remaining 12 strains. The hybridisation to a few probes by DNA isolated from each of these 12 strain implies that this region is divergent rather than absent. PCR primer walking was performed on the strains detailed in Additional file 12 and primers were designed to the region CDR3184-3190. These primers generated amplicons of the expected size when DNA was derived from the positive control, R20291. No such amplicons were generated when DNA was derived from the 12 test strains. The positive control primers designed to amplify CDR3190 produced an amplicon with DNA isolated from all strains (data not shown). Overall, these results indicate that the absence of this additional agrCABD locus is the exception, rather than the rule.

Other virulence factors

The ability to sporulate is an important mechanism for the dissemination of all clostridia. A recent study has suggested that epidemic PCR-ribotype 027 isolates are more prolific in terms of spore formation than non-epidemic strains [43]. The sporulation related genes represented on the array are conserved throughout all the strains tested.

Another set of genes that have been implicated in virulence are those encoding cell surface proteins, including Cwp84 [44]. The majority of the genes coding for cell surface proteins are conserved in all of the strains tested. The genes which appear to show divergence are cwp66, CD2791 and CD3392.

Strains and growth conditions

Ninety-four clinical strains were investigated in this study and these included 29 PCR-ribotype 027 strains, 17 PCR-ribotype 106 strains, 10 PCR-ribotype 001 strains, 9 PCR-ribotype 078 strains, 8 PCR-ribotype 002 strains, 8 PCR-ribotype 017 strains and 7 PCR-ribotype 014 strains (Additional file 16). Four non-toxigenic strains were also hybridised to the array for further investigation. The majority of the strains examined this study were isolated in the UK or the Netherlands.

A 10 μl loop was used to inoculate pre-reduced BHIS agar from frozen bacterial stock. The plates were then incubated anaerobically at 37°C under an atmosphere of N₂:H₂:CO₂ (80:10:10, vol:vol:vol) in an anaerobic workstation (Don Whitley, Yorkshire, UK). A single colony was then used to inoculate a 10 ml BHIS broth and incubated overnight prior to DNA extraction.

DNA Extraction

A traditional DNA extraction method utilising phenol chloroform extraction was used [48]. Briefly, overnight cultures were pelleted and the cells were resuspended in 260 μl buffer EB (Qiagen). After the addition of 20 mg/ml lysozyme (Sigma-Aldrich, Gillingham, Dorset, UK) and 10% SDS (Sigma-Aldrich, U.K.), the solution was incubated at 37°C for 1 hour. The solution was then incubated for a further hour with 100 mg/ml DNase free RNase (Roche, Burgess Hill, U.K.) and Proteinase K (20 mg/ml; Qiagen, Crawley, West Sussex, U.K.). DNA was extracted by phenol:chloroform:IAA (Sigma-Aldrich) washes and phase-lock gel (5 Prime, Gaithersburg, MD, USA). The genomic DNA was then precipitated using ice-cold 100% ethanol and sodium acetate and purified with two washes of 70% ethanol. Purity and quantity were assessed using a NanoDrop1000 spectrophotometer (Thermofisher Scientific, Waltham, MA, USA) and visualisation by gel electrophoresis. Genomic DNA used for hybridisation to the microarray was fragmented by sonication and the fragment size was examined by gel electrophoresis.

Array design

The array was designed to cover the previously sequenced strain C. difficile 630, the preliminary 454 sequence data of the 027 strain R20291 and the unannotated sequence of the Canadian 027 isolate QCD32g58 in a strategy similar to that used by Witney et al[49]. The R20291 genome sequence was generated by 454/Roche GS20 as discussed in Stabler et al[32]. Genome annotation of strain R20291 and QCD32g58 was based on previously published annotations of C. difficile strain 630 [17]. The genomic sequences were compared against the database of strain 630 proteins by blastx, and a CDS feature in the query genome was created when a hit of over 90% identity was found. Glimmer3 was used to predict CDSs in genomic regions where no significant hits were found [50]. Any unique genomic regions left were examined and annotated manually in Artemis [51]. The genome comparisons were visualized in Artemis and ACT (Artemis Comparison Tool; [52]. In silico c omparison against the Canadian strain QCD32g58 was also performed.

Probes were firstly designed to the CD630 genome and then additional genes of interest from other strains (R20291 and Quebec) were included. The CD630 portion of the array had a tiling design. For this, the genome was divided into 5 Kb segments with the aim of producing the best probe for each 100 bp of sequence. All possible 60 mers were considered and ranked on the basis of melting temperature, likelihood of secondary structure and GC content. The highest ranking probe per 100 bp was then selected. Additional genes were compared to other sequences and any regions of homology discounted. Where possible the 3 highest ranking probes were then selected. Sometimes 3 probes per gene were not possible due to gene length or homology. The array design was completed using the Agilent eArray interactive website. http://www.genomics.agilent.com/CollectionSubpage.aspx?PageType=Product&SubPageType=ProductDetail&PageID=1455

Using this method 40, 0097 oligonucleotides were designed to cover the CD630 genome. A further 681 probes covered any extra genes found in R20291 or QCD32g58. Regions such as the PaLoc were also represented by 346 extra probes at higher density.

Array production

Our high density custom microarrays were printed using an in situ inkjet oligonucleotide synthesizer by Agilent Technologies (Stockport, Chesire [53]. The probes were 60 oligonucleotides in length and printed in single copy per array. Four arrays were printed per slide.

Labelling and hybridisations

The genomic DNA was labelled using the Bioprime DNA labelling system (Invitrogen, UK). Hybridizations were performed, using SureHyb technology (Agilent, Stockport, Chesire, U.K.), with 2 μg of test genomic DNA labelled with Cy5-dCTP and 2 μg Cy3-dCTP (GE Healthcare Life Sciences, UK) with labelled C. difficile 630 genomic DNA as a common reference. The labelled DNA was purified using a MiniElute kit (Qiagen, Crawley, W. Sussex, UK) and the extent of Cy dye incorporation was measured using a nanodrop spectrophotometer. The test and control DNA were combined in a final volume of 39 μl and at a concentration of 2 μg each. To this mixture 10× Oligo aCGH/ChIP-on-Chip Blocking agent and 2× Hi-RPM hybridisation buffer (Agilent Technologies, U.K.) were added. The solutions were then denatured at 95°C, and incubated at 37°C for 30 min. The microarray was hybridized overnight using a SureHyb chamber at 65°C for 24 h. Slides were washed once in pre-heated Oligo aCGH/ChIP-on-chip Wash Buffer 1 for 5 min and briefly in Oligo aCGH/ChIP-on-chip Wash Buffer 2. Microarrays were scanned using an Axon 4000b array scanner (Molecular Devices, Sunnyvale, CA, USA) and intensity fluorescence data acquired using GenePix Pro (Molecular Devices).

Technical replicates were performed with our control strains CD630 (self-self hybridisation), R20291 and CD196, and this included dye-swap experiments. No replicates were performed for the clinical strains tested.

Microarray data analysis

The data was normalized and analysed using GeneSpring GX version 7.3 (Agilent Technologies, UK). Initially for each spot, the median pixel intensity for the local background was subtracted from the median pixel intensity of the spot, and any values less than 0.01 were adjusted to 0.01. Background-subtracted pixel intensities for the test strain channel were divided by those for the reference strain channel. The resulting log ratios were normalised by applying Per Spot Per Chip normalization, using 50% of data from that chip as the median.

An arbitrary cut-off of twofold was used to identify those genes that are specific to one of the strains. Therefore, for all strains, the upper cut-off was set at a ratio of 2 and the lower cut-off at a ratio of 0.5. Genes with a ratio greater than the upper cut-off were deemed to be specific to the test strain, genes with a ratio less than the lower cut-off were deemed to be specific to the reference strain, and genes with ratios between 0.5 and 2 were deemed to be present in both strains. Previous studies have shown that using arbitrary twofold cut-offs to determine presence or absence of genes is more conservative than other methods such as GACK or standard deviation from the median [48]. The presence or absence of a sequence was based on the presence or absence of one probe. The presence of absence of a gene was based on the presence or absence of more than one probe.

PCR amplification

PCR amplifications were performed using primers described in Supplementary Table 7 and KOD Hot start DNA polymerase (Novagen, Merck Chemicals, UK). Reactions were performed using a denaturation step at 95°C followed by 30 cycles at 95°C for 30 seconds, 52°C for 1 minute, 72°C for 2 - 7 minutes, followed by a final extension of 72°C for 5- 7 minutes. PCRs used to define the PaLoc used the primers and reaction conditions as described by Braun et al[45]. PCR primer walking used to confirm the results for the second agr locus were performed using the same polymerase as above, with annealing temperatures of 55°C. PCR products were analysed on 1% or 3% agarose gels run at 100-150 mV for 1 hour and stained with ethidium bromide.

Microarray data accession number

Fully annotated microarray data has been deposited in ArrayExpress (E-MTAB-162).