A total of 49 tandem repeat regions that met the selection criteria (repeat size, 15–40 bp; repeat copy number, > 5; consensus sequence match, < 90%) were detected in the genome sequence from strain 630 by using the program Tandem Repeats Finder version 4.00 4142. For 36 of these repeat regions, it was possible to design PCR primers targeting flanking sequences, and from 28, PCR amplification products could reliably be generated from a panel of reference isolates. However, at 25 of these loci, sequence variation was insufficient to discriminate widely distributed strains, including ribotypes 027, 017, and 001 (not shown). The remaining three repeat regions could discriminate most of the ribotypes examined. The two most variable loci were designated TR6 and TR10 (Table 1). They are located at positions 0.7 Mb and 3.7 Mb of the C. difficile 630 chromosome, respectively, and exhibited both, sequence and length polymorphisms. Locus TR6 is composed of 21-basepair repeat units and resides within an open reading frame encoding a hypothetical protein (orf CD0603 in the 630 genome sequence). A homology search in public databases did not identify any significant similarities with known proteins. In contrast, TR10 is located within a predicted non-coding region. It consists of 22-basepair repeats.

We developed a DNA based typing scheme for C. difficile based on the sequence variation of TR6 and TR10. To facilitate the application of the tandem repeat sequence typing (TRST) scheme, a duplex PCR was designed which allowed simultaneous amplification of both loci (Figure 1). Sequence data were generated from duplex PCR products using the same primers as for amplification. Nucleotide sequences from TR6 and TR10 were concatenated and unique repeat successions were assigned distinct TRST types (tagged with consecutive numbers, prefixed with "tr"; Figure 2, Additional files 1, 2). A detailed comparison of TRST with PCR ribotyping is described in the following.

Genomic regions with short tandem repeat regions may evolve fast due to intra-molecular recombination and frequent polymerase slippage during DNA replication 434445. Accordingly, loci TR6 and TR10 displayed both, sequence polymorphisms, generated through exchange of individual nucleobases (Additional files 3, 4), and length polymorphisms, as a consequence of repeat copy number variation (Additional file 2). Sequences of individual repeats were highly variable, with a nucleotide diversity π of 0.28 ± 0.01 for TR6 and 0.23 ± 0.01 for TR10. The majority of nucleotide substitutions at locus TR6 were synonymous, i. e., they left the encoded amino acid sequence unaffected, and hence may be considered selectively neutral. This was reflected by a Ka/Ks value of 0.39, suggesting TR6 sequences evolve under purifying selection. Locus TR10 does not encode any protein and, hence, sequence variation likely is neutral, too.

Furthermore, there is evidence of rare recombination between chromosomes from different strains, affecting tandem repeat sequences. One homologous recombination event apparently generated TRST type tr-021. While tr-021 shares an identical TR6 sequence with tr-011 (Additional file 2), its TR10 allele differs profoundly from that of tr-011 in both, length and sequence (Additional files 4 and 2), even though isolates displaying tr-011 (isolate N551) and tr-021 (SMI037) are affiliated to the same MLST type (ST-39) and ribotype (011; Figure 3). Interestingly, the TR10 allele of tr-021 is identical to the one of tr-005 (Additional file 2). Hence, the drastic difference between central parts of TR10 in tr-011 and tr-021 may be explained through a single event of horizontal gene transfer from an unrelated strain. Very similarly, tr-066 and tr-045 share identical alleles with closely related TRST types at either TR6 or TR10, respectively, yet differ drastically along a contiguous stretch of central repeats at the other tandem repeat locus. Again, identical alleles may be found elsewhere in the database (Additional file 2), suggesting they were horizontally transferred. In our dataset, these three TRST types displayed the only such discrepancies. We conclude that genetic recombination between unrelated chromosomes was involved in the evolution of maximally three TRST types out of 72 that were included in our set of isolates. Hence, the evolution of tandem repeats TR6 and TR10 is driven largely through clonal diversification, whereas the impact of recombination is extremely small. These results fully corroborate a previous estimate of a very low recombination rate in C. difficile, which had been based on MLST data 31.

While TR6 and TR10 displayed remarkable sequence variation, both loci seemed sufficiently stable to identify genetically related isolates collected over time. For one, the stability of TR6 and TR10 was demonstrated by two VPI 10463 and three 630 strains (including the published genome sequence), that prior to our analysis each had been handled in different laboratories (Additional file 1) and, hence, had independently been subcultured multiple times, but yet shared the same respective TRST sequence types (Additional file 1). Furthermore, stability of both tandem repeat regions was circumstantially suggested through identical sequences found in multiple isolates sharing the same ribotype but originating from different geographical regions (Additional file 1).

Results were compared to PCR ribotyping on the basis of 154 isolates including international reference strains and clinical isolates collected at various German laboratories (Additional file 1). These isolates had been preselected from the material available to represent maximal diversity as judged on the basis of PCR ribotyping and geographic origin. They represented 75 different ribotypes (Additional file 1). Figure 2 shows a neighbor joining dendrogram based on the repeat successions in concatenated TR6 and TR10 sequences.

All 154 isolates were typeable by TRST. Considering both, differences in length and nucleotide sequence, 43 distinct alleles were identified at locus TR6, and 53 alleles at locus TR10 (Table 2, Additional file 2). Sequencing either one of the two loci had less discriminatory power than PCR ribotyping, as reflected by slightly lower discriminatory indices (0.93 and 0.95, respectively, versus 0.97 for ribotyping; Table 2). When considered in combination, however, sequence analysis of TR6 and TR10 resulted in the identification of 72 different TRST sequence types among the 154 isolates investigated (Additional file 2, Figure 2). This way, TRST and PCR ribotyping had equal discriminatory power, reflected by identical discriminatory indices (Table 2) based on the set of isolates included. It has to be considered, however, that this estimate will be skewed to some extent in favour of ribotyping, since ribotype diversity was the basis of initial isolate selection. Many ribotypes were represented by single isolates, and the potential ability of TRST to further discriminate within these ribotypes was thus not tested.

TRST demonstrated high overall concordance with PCR ribotyping for the set of strains typed in this study, resulting in a calculated Adjusted Rand's index of 89.8% (Table 2). The probability that a pair of isolates with the same ribotype also shared identical TRST sequence types was 89.6% (Wallace index 0.896). Accordingly, ribotypes usually corresponded to specific TRST sequence types (Figure 2). For example, 18 isolates with ribotype 027, originating from six different European countries, displayed identical sequences at TR6 and TR10 that discriminated them from all other isolates, and jointly were assigned TRST sequence type tr-027 (Additional file 1, Figure 2). Similarly, four isolates with ribotype 017 from three different countries, including the reference strain for toxinotype VIII, were assigned sequence type tr-017 (Additional file 1, Figure 2). Future work on larger numbers of isolates may reveal that sequencing a single locus (TR6 or TR10) will suffice to identify epidemiologically relevant strains. For the sake of concordance with PCR ribotyping, however, we presently suggest to sequence both loci. As outlined above, this strategy will also detect the impact of recombination.

Discrepancies between TRST and ribotyping were apparent where either method split a particular group of isolates into two or three classes, whereas the other lumped them into one (Figure 2). In virtually all of these cases, however, the respective isolates were affiliated to identical MLST sequence types or to single locus variants with respect to MLST (i. e., identical sequences at six out of seven MLST loci), indicating their close phylogenetic relatedness. Phylogenetic coherence of these additional (sub-)classes will remain unclear as long as there are no phylogenetic markers available to investigate the detailed evolutionary history of C. difficile within MLST sequence types.

MLVA typically resolves dozens of distinct genotypes within individual ribotypes 2021. However, MLVA provided little insight to the genetic relatedness within our collection, since almost all isolates differed from each other at four or more loci 20, even when they were affiliated to identical TRST sequence types or ribotypes (Figure 3). The sole useful exception was represented by isolates JW611148 and CL39, which shared identical alleles at five MLVA loci (Figure 3). The summed tandem-repeat difference between these two isolates was four repeats, which is below the threshold (= 10) previously suggested to indicate close genetic relationship based on MLVA 21. MLST identity confirmed the relatedness of these isolates (Figure 3), and their close phylogenetic relationship also was correctly reflected by identical sequences at TR6 and TR10 (tr-070, Figure 3). However, these isolates displayed a distinct one-band difference between their ribotyping patterns, corresponding to ribotypes 078 and RKI35, respectively (Figure 4). This result illustrates the fact that ribotypes may differ widely with respect to the phylogenetic divergence they encompass. It may be noted that two other pairs of isolates shared highly similar MLVA patterns (AB403/CL45, NCTC11204/P5732; Figure 3). The summed tandem-repeat difference for the former pair is seven repeats, and hence these two isolates would be suggested to be extremely closely related based on MLVA alone 21. These similarities, however, clearly reflect homoplasies, since MLST indicated these isolates were entirely unrelated (Figure 3). Thus, the application of MLVA as currently used is inappropriate when attempting to resolve distant phylogenetic relationships of C. difficile isolates. Again, in these cases, phylogeny was correctly indicated by TRST. We therefore conclude that it may be useful to combine TRST and MLVA in a nested hierarchical fashion, where TRST may resolve phylogenetic diversity to a level equivalent to PCR ribotypes, and MLVA may add additional resolution where desired.

Evolutionary relationships between isolates may be revealed through tandem repeat sequence alignment and phylogenetic analysis. This is also feasible for those isolates that were assigned different TRST types. For example, ribotypes 027, 156, and 019 by MLST are indicated to be closely related, since corresponding isolates are assigned two MLST sequence types that differ at one locus only (Figure 3). Close relationship of ribotypes 027 and 019 previously has also been found on the basis of DNA macrorestriction analysis, when isolates with both ribotypes were assigned to the 'North American Pulsotype NAP1' 23. Concordantly with MLST and macrorestriction, TRST also indicated the relatedness of these types through similar tandem repeat sequences that clustered tightly in the phylogenetic tree (Figure 2), yet it maintained the discriminatory power of PCR ribotyping by assigning three different sequence types (tr-034, tr-027, tr-019) (Figure 2). Similarly, ribotypes 078 and RKI35 were indicated to be closely related to ribotype 066 by both, MLST and TRST (Figures 2 and 3). In contrast, these relationships were not at all apparent on the basis of ribotyping band patterns (Figure 4).

Phylogenetic relatedness was also indicated in cases where TRST was more discriminatory than PCR ribotyping. For example, ribotypes 001, 163, 087, 014, and 117 each were subdivided into several TRST types (Figure 2). Clusters of related tandem repeat sequences in the phylogenetic tree still corresponded to PCR ribotypes (Figure 2), which warrants the comparability of results from both methods. This feature may be highly desirable, since it will facilitate, for example, cross-referencing to ribotyping-based examinations and maintaining the continuity of ongoing surveillance programs.

Ribotyping does not enable phylogenetic analyses based on dissimilar banding patterns, and the relatedness of different ribotypes has not commonly been assessed. In the long run, large-scale mutation discovery and genomic (re-)sequencing will reveal the phylogenetic validity of typing procedures 46.

We anticipate that PCR ribotyping will eventually be replaced by typing procedure(s) based on DNA sequences. The inherent portability of sequence data will obviate the need for the exchange of reference strains and enable decentralised genotyping efforts, which may boost large scale investigations on the molecular diversity of C. difficile. At present, however, our knowledge about the diversity and population biology of this important pathogen is very limited 233132. As a consequence, it is generally not clear if isolate groupings provided by various typing methods, including PCR ribotyping, are concordant with the epidemiology of associated disease 2123. Related to these considerations, one limitation of this present study is the lack of epidemiologically linked isolates in our data set. Investigations in the near future should evaluate the utility of tandem repeat sequencing for infection chain tracking and short-term epidemiological investigations.

A total of 154 C. difficile isolates comprising 75 different ribotypes were used in this study. The strain collection included both, international reference strains and selected clinical isolates from various German hospitals, collected in 2007 and 2008. More detailed information about individual isolates is given in Additional file 1.

Genomic DNA was isolated from cultures grown for 48 h on cycloserine-cefoxitin fructose agar (OXOID, Basingstoke, UK), by using the DNeasy Blood & Tissue Kit (QIAGEN, Hilden, Germany) according to the manufacturer's recommendations.

PCR ribotyping initially was performed at the Reference Laboratory for Clostridium difficile at the Leiden University Medical Center in the Netherlands and later was transferred to the Robert Koch Institute. We followed the protocol of Bidet et al. 26, except that PCR Products were run on 1.5% agarose gels in 1× TBE at 85 volts for 4 hours. Isolates were assigned novel PCR ribotypes if their patterns differed from previously named patterns by at least one band.

To facilitate the application of tandem repeat sequence typing, a duplex PCR was designed using the following primers: TR6-F (5'-TTTCAACTTGTCCAGTTTTTAAGTC-3') and TR6-R (5'-ATGACATAGCGTTTGTGGAAT-3'); TR10-F (5'-TGCATCAAATTGGTCAAGACTC-3') and TR10-R (5'-TGAAATCATTGACTATAAAGCAAAA-3'). DNA amplification was performed on 1 μl of purified genomic DNA in a final volume of 50 μl containing 0.1 μM of TR6 and 1 μM of TR10 primers, 200 μM of each deoxynucleoside triphosphate, 1× PeqLab PCR buffer Y (20 mM Tris-HCL, 16 mM (NH₄)₂SO₄, 0.01% Tween 20, 2 mM MgCl₂) and 1.25 units Hot Taq-DNA-Polymerase (PeqLab, Erlangen, Germany). After an initial denaturation of 96°C for 3 min, the protocol consisted of 35 cycles at 96°C for 45 s, 52°C for 45 s, and 72°C for 45 s following a final extension at 72°C for 7 min. PCR products were prepared for sequencing using the QIAquick^® PCR Purification Kit (QIAGEN, Hilden, Germany) and 0.35 μl of the purified products were applied for sequencing using the BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Foster City, USA) with identical primers employed in the PCR. Automated sequence detection was performed on an ABI capillary sequencing system and sequences were analysed using the BioNumerics 5.10 software (Applied Maths, Belgium).

Data processing was performed with BioNumerics 5.10 by using a novel, dedicated "Repeat Typing" plugin that allowed automated batch assembly of trace files. The assignment of TRST sequence types was based on the successive occurrence of user-defined repeats in concatenated sequences from both tandem repeat loci. A repeat distance matrix for matching and clustering were calculated based on the DSI model 47, a mutation model comprising substitutions, indels (insertions or deletions), and duplications. Subsequent cluster analysis was performed based on the neighbor joining algorithm.

Clostridium difficile isolates were typed by MLST as described previously 31. Sequence data were submitted to the C. difficile MLST database http://www.pasteur.fr/recherche/genopole/PF8/mlst/Cdifficile.html to assign allele profiles and the resulting sequence types. Sequence types were analysed by constructing a dendrogram based on the UPGMA (Unweighted Pair Group Method with Arithmetic mean) clustering algorithm using the multistate categorical similarity coefficient (tolerance 0%) available in the BioNumerics software.

Seven-locus MLVA was conducted as described previously 2022, except that the different loci were PCR-amplified individually and PCR products were sequenced for repeat copy number determination. To facilitate sequence analysis of MLVA locus C6 20, two novel oligonucleotide primers were used: C6-F 5'-CCAAGTCCCAGGATTATTGC-3' and C6-R 5'-AACATGGGGATTGGAATTGA-3'. Repeat copy numbers were determined manually using BioNumerics 5.10 software. The summed tandem-repeat difference was calculated where appropriate; it is the sum of repeat differences between two isolates at all seven MLVA loci 21.

An index of discrimination was calculated to compare the discriminating capacity of ribotyping, and TRST. The discriminatory index was defined as the average probability of two consecutively sampled strains being characterized as the same type. This probability depends on the number of strain types and their frequency distribution in the population. Discriminatory indices were calculated based on Simpson's index of diversity 48. Confidence intervals for discriminatory indices were determined as described previously 49. The Concordance of two typing schemes was calculated based on the adjusted Rand's and Wallace's coefficients 50. While the Rand's coefficient allows a quantitative evaluation of the global congruence between two typing systems, the Wallace's coefficient compares the congruence of schemes depending on the directionality of typing by estimating the probability that a pair of isolates sharing the same type in system 1 also share the same type in system 2, and vice versa. Calculation of all parameters was performed with EpiCompare software, version 1.0 (Ridom GmbH, Würzburg, Germany).

The nucleotide diversity (π) and the ratio (Ka/Ks) of the average number of non-synonymous substitutions per non-synonymous site (Ka) to the number to synonymous substitutions per synonymous site (Ks) was calculated by using DnaSP, version 4.5 51.