Defining and Evaluating a Core Genome Multilocus Sequence Typing Scheme for Genome-Wide Typing of Clostridium difficile

ABSTRACT

Clostridium difficile, recently renamed Clostridioides difficile, is the most common cause of antibiotic-associated nosocomial gastrointestinal infections worldwide. To differentiate endogenous infections and transmission events, highly discriminatory subtyping is necessary. Today, methods based on whole-genome sequencing data are increasingly used to subtype bacterial pathogens; however, frequently a standardized methodology and typing nomenclature are missing. Here we report a core genome multilocus sequence typing (cgMLST) approach developed for C. difficile. Initially, we determined the breadth of the C. difficile population based on all available MLST sequence types with Bayesian inference (BAPS). The resulting BAPS partitions were used in combination with C. difficile clade information to select representative isolates that were subsequently used to define cgMLST target genes. Finally, we evaluated the novel cgMLST scheme with genomes from 3,025 isolates. BAPS grouping (n = 6 groups) together with the clade information led to a total of 11 representative isolates that were included for cgMLST definition and resulted in 2,270 cgMLST genes that were present in all isolates. Overall, 2,184 to 2,268 cgMLST targets were detected in the genome sequences of 70 outbreak-associated and reference strains, and on average 99.3% cgMLST targets (1,116 to 2,270 targets) were present in 2,954 genomes downloaded from the NCBI database, underlining the representativeness of the cgMLST scheme. Moreover, reanalyzing different cluster scenarios with cgMLST were concordant to published single nucleotide variant analyses. In conclusion, the novel cgMLST is representative for the whole C. difficile population, is highly discriminatory in outbreak situations, and provides a unique nomenclature facilitating interlaboratory exchange.

INTRODUCTION

Clostridium difficile, recently renamed Clostridioides difficile, is an anaerobic, Gram-positive, endospore-forming rod-shaped bacterium and the most common cause of antibiotic-associated nosocomial gastrointestinal infections in Europe and the United States (1, 2). Over the last decades, severe C. difficile infections (CDI) have been increasingly detected in hospitals, making C. difficile an important nosocomial pathogen. CDI develop either from endogenous colonization under selecting conditions such as an antibiotic treatment or from an exogenous source, i.e., spores from the contaminated environment (3).

Several methods are described for C. difficile typing, of which PCR ribotyping is currently becoming a gold standard worldwide (1, 4). For an initial grouping of strains, multilocus sequence typing (MLST) (5, 6) and toxinotyping (7) are also widely used methods. For highly discriminatory subtyping of strains, which is necessary in the case of a suspected outbreak, these methods are sometimes complemented with pulsed-field gel electrophoresis (PFGE) or multilocus variable-number tandem-repeat (VNTR) analysis (MLVA) (4); both methods are able to differentiate among closely related isolates. Except for MLST, where a central database hosting the typing nomenclature is in place, the interlaboratory exchange of such typing data is hampered by the lack of a publicly available database ensuring a unique nomenclature and—in the case of PCR ribotyping and PFGE—by difficulties to standardize the interpretation of DNA banding patterns (4, 8).

Nowadays, sequence-based typing approaches using whole-genome sequence (WGS) data are overcoming these obstacles. Several studies on various bacterial species have already shown that WGS-based typing, based either on single nucleotide variants (SNVs) (9, 10) or on gene-by-gene allelic profiling of core genome genes, frequently named core genome MLST (cgMLST) (11–13), currently represents the ultimate tool for strain subtyping. Moreover, it was recently shown in an international ring trial that cgMLST is highly reproducible (14).

For C. difficile, initial studies also confirmed the general applicability of WGS-based typing (9, 15, 16). Nevertheless, the broad use of WGS-based typing of C. difficile is still hampered by the lack of standardized nomenclature (17); this has already been established for other pathogens (18–21) and would facilitate interlaboratory exchange of data.

Therefore, to obtain the basis of a standardized nomenclature for WGS-based C. difficile typing, we defined a novel C. difficile cgMLST scheme covering the genetic diversity within the C. difficile population based on well-characterized reference strains and subsequently challenged this scheme using a diverse set of strains from sporadic cases and outbreak investigations.

MATERIALS AND METHODS

C. difficile strains and genomes.

All strains and genome sequences used for the development of the novel C. difficile cgMLST scheme are listed in Table 1. The isolates were selected by covering the whole diversity of C. difficile organisms, i.e., representative isolates for each clade (downloaded from https://pubmlst.org/cdifficile/) and—based on a Bayesian analysis of the genetic population structure (BAPS; see below) using all available MLST sequence types (STs) as input data—randomly selected representative isolates for each BAPS partition were included (17, 22). The well-defined C. difficile strain 630 (23) was used as the reference sequence during cgMLST target definition. Moreover, the NCBI RefSeq sequences of C. difficile strains CD196 and M120 were used.

TABLE 1.

List of Clostridium difficile isolates and genomes used for cgMLST target definition

Isolate	Clade	BAPS partition	MLST ST	PCR ribotype	Toxinotypea	NCBI RefSeq/ENA SRA accession no. (reference)
630 (reference)	1	Cd06	54	012	0	NC_009089
2402	1	Cd06	199	SLO 086	XXXIII	ERS2050168 (this study)
CD196 (R12087)	2	Cd06	1	027	IIIb	NC_013315
8785	2	Cd06	196	109	IXc	ERS2050173 (this study)
C00007686	3	None	5			SAMEA2240504 (39)
1470	4	Cd05	37	017	VIII	ERS2050166 (this study)
M120	5	Cd03	11	078	V	NC_017174 (40)
SUC36	5	Cd03	195	078	XVI	ERS2050188 (this study)
173070	C-IIb	Cd01	200	151	XXXII	ERS2050167 (this study)
ZZV13-5576	C-I	Cd02	297	SLO 229	Paloc negative	ERR2216002 (this study)
ZZV14-6045	C-IIIb	Cd04	343	SLO 205	PaLoc negative	ERR2216003 (this study)

^aToxinotypes in accordance with recent update on C. difficile toxinotyping (7).

^bPutative new lineage (41).

For subsequent evaluation of the scheme, we used two different sets of isolates/genome sequences: first, a total of 70 well-defined C. difficile isolates (Table 2) were used comprising (i) the reference strains of all published toxinotypes (n = 38) to cover the diversity of toxigenic strains (7), (ii) isolates from two published clusters (n = 8) as examples to rule in or out nosocomial transmissions (9), and (iii) isolates detected during a surveillance study for infection control (n = 24) (15). As a second set for evaluation of the cgMLST scheme, we downloaded 268 assembled genome sequences from the NCBI database (ftp://ftp.ncbi.nih.gov/genomes/) and sequence reads (only data generated with any Illumina sequencing platform) from 3,482 C. difficile isolates from the NCBI Sequence Read Archive (SRA) that were available until 21 October 2015 and were assembled prior to use.

TABLE 2.

List of 70 C. difficile isolates and genomes (toxinotypes and cluster/outbreak isolates) for evaluation of the novel cgMLST scheme

Isolate	Cladea	BAPS partition	MLST STa	PCR ribotypeb	Toxinotypec	% cgMLST targets	NCBI or ENA SRA accession no. (reference)
EX623	1	Cd06	24	102	I	99.5	ERS2039514 (this study)
IS 25	1	Cd06	58	258	XII	99.6	ERS2050180 (this study)
IS 58	5	Cd03	11	033	XIa	97.7	ERS2050181 (this study)
J9965	2	Cd06	194	SLO 032	Xb	98.7	ERS2050182 (this study)
K095	1	Cd06	2	014	XVIII	99.9	ERS2039515 (this study)
KK2443/2006	1	Cd06	19	SLO 037	XXVII	98.9	ERS2039516 (this study)
OCD 5/2	5	Cd03	11	033	XIc	97.7	ERS2039517 (this study)
R 10870	2	None	114	111	XIVa	99.1	ERS2050183 (this study)
R 11402	5	Cd03	11	288 (CE)	XIb	97.6	ERS2050184 (this study)
R 9385	2	Cd06	116	122	XIVb	99.6	ERS2050185 (this study)
R 9367	1	Cd06	55	070	XIII	99.8	ERS2039518 (this study)
SE 881	5	Cd03	11	045	V	98.8	ERS2050186 (this study)
SE 844	2	None	192	080	IIIa	99.6	ERS2050187 (this study)
TFA/V14-10	2	Cd06	231	153 (CE)	XId	99.4	ERS2039519 (this study)
TR13	1	Cd06	17	018	XIX	99.4	ERS2039541 (this study)
TR14	1	Cd06	182	SLO 005	XX	99.9	ERS2039542 (this study)
1732874	2	Cd06	226	SLO 228	IXd	99.6	ERS2039496 (this study)
3073	2	Cd06	41	SLO 042	IIId	99.6	ERS2039506 (this study)
51377	5	Cd03	11	127	VI	98.8	ERS2050169 (this study)
51680	2	Cd06	67	019	IXa	99.6	ERS2050170 (this study)
57267	5	None	193	063	VII	98.4	ERS2050171 (this study)
597B		None	122	131	0/v	99.3	Reference 22
55767	3	None	5	023	IV	98.7	ERS2039507 (this study)
7325	2	Cd06	1	027	XXV	99.5	ERS2050172 (this study)
7459	1	Cd06	16	050 (CE)	XXVI	99.3	ERS2039509 (this study)
8864	2	Cd06	62	591 (CE)	Xa	99.7	ERS2050174 (this study)
AC008	1	Cd06	53	103	II	99.9	ERS2039510 (this study)
AI 541	2	Cd06	231	251	IIIe	99.6	ERS2039511 (this study)
CD07-468	2	Cd06	197	027	XXII	99.6	ERS2050175 (this study)
CD07-140	1	Cd06	3	001	XXIX	98.8	ERS2039512 (this study)
CD08-070	5	Cd03	11	126	XXVIII	99.1	ERS2050176 (this study)
CD10-055		Cd04	369	SLO 201	XXXIV	96.2	ERS2039513 (this study)
CH6223	4	None	198	SLO 035	XXI	98.5	ERS2050177 (this study)
CH6230	2	Cd06	123	251	IIIc	99.3	ERS2050178 (this study)
ES 130	5	Cd03	166	SLO 101	XXX	98.7	Reference 42
WA 151	5	Cd03	167	SLO 098	XXI	98.4	Reference 42
VPI 10463	1	Cd06	46	087	0	99.4	ERS2039543 (this study)
TFA/V20-1	2	Cd06	41	244	IXb	99.5	ERS2039540 (this study)
C00006623	1	Cd06	2			99.8	ERX103559 (9)
C00006624	1	Cd06	10			99.8	ERX103560 (9)
C00006625	4	Cd05	37			98.9	ERX103561 (9)
C00006626	4	Cd05	37			99.5	ERX103562 (9)
C00006627	3	None	5			98.5	ERX103563 (9)
C00006628	1	Cd06	10			99.8	ERX103564 (9)
C00006629	1	Cd06	54			99.3	ERX103565 (9)
C00006630	3	None	5			98.7	ERX103566 (9)
M68	4	Cd05	37	017	VIII	98.8	NC_017175 (15, 40, 43)
R20291	2	Cd06	1	027	III	99.6	NC_013316 (15, 40, 43)
P1	2	Cd06	1			99.6	SRX821661 (15)
P2	2	Cd06	1			99.7	SRX821763 (15)
P3	2	Cd06	1			99.7	SRX821764 (15)
P4	2	Cd06	1			99.7	SRX821765 (15)
P5	2	Cd06	1			99.7	SRX821766 (15)
P6	2	Cd06	1			99.7	SRX821767 (15)
P7	2	Cd06	1			99.7	SRX821768 (15)
P8	1	Cd06	2			99.8	SRX821769 (15)
P9	4	Cd05	37			99.6	SRX821770 (15)
P10	4	Cd05	37			99.6	SRX821771 (15)
P11	1	Cd06	2			99.9	SRX821772 (15)
P12	4	None	81			99.5	SRX821773 (15)
P13A	2	Cd06	1			99.7	SRX821774 (15)
P13B	2	Cd06	1			99.7	SRX821775 (15)
P13C	2	Cd06	1			99.7	SRX821777 (15)
P14	1	Cd06	8			99.7	SRX821778 (15)
P15	1	Cd06	8			99.7	SRX821779 (15)
P16	2	Cd06	1			99.7	SRX821780 (15)
P17	2	Cd06	1			99.7	SRX821781 (15)
P18	2	Cd06	1			99.7	SRX821782 (15)
P19	4	None	81			99.5	SRX821783 (15)
P20	4	None	81			99.5	SRX821784 (15)

^aMLST STs were in accordance to the C. difficile MLST database (https://pubmlst.org/cdifficile/), and clades were determined in this study (see Table S1).

^bPCR ribotypes were in accordance to recent publications (7, 40).

^cToxinotypes were given in accordance with the recent update on C. difficile toxinotyping (7).

BAPS.

To determine the overall C. difficile species variation, we used Bayesian Analysis of Population Structure (BAPS) version 6.0 (17, 24, 25). Sequences of all MLST STs available as of 31 March 2016 (n = 347 STs) were downloaded from the MLST website (https://pubmlst.org/cdifficile/) (6), and all allelic gene sequences per locus were multiply aligned using MUSCLE (26) and finally concatenated for each ST. BAPS was carried out using the clustering of linked molecular data functionality. Ten runs were performed, setting an upper limit of 30 partitions. Admixture analysis was performed using the following parameters: minimum population size considered, 1; iterations, 50; number of reference individuals simulated from each population, 50; and number of iterations for each reference individual, 10.

DNA extraction, whole-genome sequencing, and assembly.

Prior to sequencing, the isolates were cultured anaerobically for 48 h at 37°C on Columbia blood agar plates (Oxoid, Wesel, Germany) and DNA was extracted using a fast glass bead method (27). Sequencing libraries were prepared using Nextera XT chemistry (Illumina Inc., San Diego, CA) for a 250-bp paired-end sequencing run on an Illumina MiSeq sequencer. Samples were sequenced to aim for minimum coverage of 120-fold using Illumina's recommended standard protocols. The resulting FASTQ files were de novo assembled using the SPAdes assembler version 3.11 (28) integrated in Ridom SeqSphere⁺ software (29) (version 5.0 beta; Ridom GmbH, Münster, Germany) using the following SPAdes parameters: k, automatic selection based on read length and mismatch careful mode turned on.

cgMLST target gene definition.

To determine the cgMLST gene set, a genome-wide gene-by-gene comparison was performed using the cgMLST target definer (version 1.4) function of SeqSphere⁺ (Ridom GmbH) with relaxed parameters (≥80% gene sequence identity and 100% gene sequence overlap) reflecting the high diversity within C. difficile. These cgMLST target definer parameters comprised the following filters to exclude certain genes of the C. difficile strain 630 reference genome (GenBank accession number NC_009089.1) from the cgMLST scheme: a “minimum length filter that discards all genes that are shorter than 50 bases,” a “start codon filter that discards all genes that contain no start codon at the beginning of the gene,” a “stop codon filter that discards all genes that contain no stop codon, more than 1 stop codon or if the stop codon is not at the end of the gene,” a “homologous gene filter that discards all genes that have fragments that occur in multiple copies in reference genome (with identity of ≥90% and more than 100 bases overlap),” and a “gene overlap filter that discards the shorter gene from the cgMLST scheme if the two genes affected overlap more than 4 bases.” The remaining genes were then used in a pairwise comparison with BLAST version 2.2.12 (parameters used were word size 11, mismatch penalty −1, match reward 1, gap open costs 5, and gap extension costs 2) with the query C. difficile chromosomes. All genes of the reference genome that were common in all query genomes with a sequence identity of ≥80% and 100% overlap (with the default parameter stop codon percentage filter turned on; i.e., more than 80% of the query genomes do not contain internal stop codons) formed the final cgMLST scheme.

Evaluation of the cgMLST target gene set.

To evaluate the representativeness and the discriminatory power of the novel C. difficile cgMLST target gene set, we used the above-mentioned genomes (Table 2; see also Tables S3 and S4 in the supplemental material). To ensure the sequence quality of the downloaded genomes/reads prior to further analyses, only isolates with a coverage of ≥50 and a consensus base count that deviated at most ±10% from the median consensus base count were included. A well-defined cgMLST scheme should result, on average, in 97.5% extracted cgMLST target genes (30). To extract the cgMLST genes, the default parameters were used in the SeqSphere⁺ software: (i) for processing options, “Ignore contigs shorter than 200 bases”; (ii) for scanning options, “Matching scanning thresholds for creating targets from assembled genomes” with “required identity to reference sequence of 90%” and “required alignment to reference sequence with 99%”; and (iii) for BLAST options, word size 11, mismatch penalty −1, match reward 1, gap open costs 5, and gap extension costs 2. In addition, the target genes were assessed for quality, i.e., the absence of frameshifts and ambiguous nucleotides. A core genome gene was considered a “good target” only if all of the above-listed criteria were met, in which case the complete sequence was analyzed in comparison to the reference sequence. Alleles for each gene were assigned automatically by the SeqSphere⁺ software to ensure a unique nomenclature. The combination of all alleles in each strain formed an allelic profile that was used to generate minimum spanning trees (MST) using the parameter “pairwise ignore missing values” during distance calculation.

In order to maintain backwards compatibility with classical C. difficile MLST, the sequences of the seven genes comprising the allelic profile of the MLST scheme were extracted separately from the genome sequences and queried against the C. difficile MLST database in order to assign STs in silico using the SeqSphere⁺ software that queries the respective gene sequences, compares them with the allele library of each of the seven MLST target genes, and assigns alleles and STs.

Accession number(s).

All raw reads generated and/or contig sequences were submitted to the European Nucleotide Archive (http://www.ebi.ac.uk/ena/) under accession number PRJEB23450. The NCBI accession numbers for the sequences determined for this study are ERR2216002, ERR2216003, ERS2039496, ERS2039506, ERS2039507, ERS2039509 to ERS2039519, ERS2039540 to ERS2039543, ERS2050167 to ERS2050178, and ERS2050180 to ERS2050188 (see Tables 1 and 2).

RESULTS

To develop a cgMLST scheme that sufficiently covers the diversity of the species C. difficile, we initially determined—besides the known partitioning into clades—the diversity using BAPS. This approach based on 347 STs resulted in six partitions comprising 306 STs; 41 STs were not assigned to any BAPS group (Table S1). Based on this grouping, 11 genome sequences, including that of C. difficile strain 630 (Table 1), were used to define the cgMLST scheme. Their comparison resulted in selection of 2,270 genes out of 3,756 genes present in strain 630 (50.4% of the 630 strain chromosome nucleotides) (Table S2). Figure 1 illustrates the diversity of the 11 isolates used for cgMLST target definition.

FIG 1.

Neighbor-joining tree of the 11 C. difficile isolates used for cgMLST target definition based on cgMLST target genes with pairwise ignore missing values. In addition to the sample name, the clade is given and the BAPS partitions are colored. The distance is given as the number of cgMLST genes.

This novel cgMLST scheme was then challenged with different sets of strains (Table 2; see also Table S3). Out of the genomes of the 38 reference strains of all published toxinotypes, 2,184 to 2,268 cgMLST targets (mean, 99.1%; median, 99.4%) could be extracted. Similarly, for the two published outbreaks, all isolates contained 2,237 to 2,267 cgMLST targets (mean, 99.5%; median, 99.7%), underlining the representativeness of the cgMLST scheme.

Moreover, we investigated the publicly available genome sequences from the NCBI (n = 268 assembled C. difficile genomes and reads from 3,482 isolates of the SRA). We first determined the median of the consensus base count (4,150,084 bp) and included only genomes of isolates in the analysis that exhibited ±10% of the median consensus base count. Furthermore, genomes of isolates with coverage <50-fold were excluded as well as NCBI assembled genomes, which also existed as SRA isolates. In total, we finally included 2,954 publicly available genomic data in our final analysis. Summarized from Table S4, on average 99.3% cgMLST targets were detected (median, 99.6%; 1,116 to 2,270 targets). Figure S1 illustrates the population structure and relationship to classical MLST STs.

To further ascertain the representativeness of our approach, especially using BAPS, we determined the ST distribution and percentage of isolates from the 2,954 isolates that were not grouped into any of the BAPS groups. After exclusion of 33 isolates with an unknown ST, only 9 of the 41 STs that were not assigned to any BAPS group were present in 123 (4.2%) isolates. Of these 123 isolates, however, all had, on average 98.8% cgMLST targets (Table S4).

After confirmation of the representativeness of the novel cgMLST scheme, we analyzed the capability of the scheme to differentiate among closely related isolates from outbreak investigations. We reanalyzed different scenarios from the literature comprising short- and long-term scenarios (9, 15). In Fig. 2, two short-term spatiotemporal clusters spanning 17 to 22 days illustrate two typical clinical scenarios: while Fig. 2A shows that a clonal spread was detected among two isolates differing in only two cgMLST targets, Fig. 2B indicates that a transmission could be ruled out, as the suspected isolates belonged to the same ST type but differed in 62 cgMLST targets. These findings were in accordance with the published SNV analysis (9). Figure 3 shows the cgMLST typing results from a recent long-term outbreak investigation in a Chinese hospital from 2012 to 2014 (15). Two peaks (March to July 2012 and August 2013 to February 2014) of a clonal spread were recognized; again, our reanalysis using cgMLST corroborated the previous findings that these peaks were linked and belonged to the same outbreak clone. Based on these results, we finally defined the threshold, i.e., the maximum number of differing alleles for isolates that are likely to belong to the same clone, as ≤6 alleles. Isolates sharing genotypes within this threshold are then grouped within the same cluster type (CT) (21).

FIG 2.

Minimum-spanning tree of two spatiotemporal clusters (9). Each node represents a unique cgMLST allele profile. The numbers on connecting lines display the number of differing alleles between the genotypes (line length not to scale). The different nodes are colored by the MLST ST, and closely related genotypes (≤6 different cgMLST alleles) are shaded. (A) Short-term cluster of four cases, where one transmission event was epidemiologically confirmed (2,244 to 2,265 cgMLST target genes [mean, 99.5%] were analyzed). (B) Short-term cluster of four cases, where a clonal transmission was ruled out (2,237 to 2,265 cgMLST target genes [mean, 99.1%] were analyzed).

FIG 3.

Minimum-spanning tree and epidemiological curve illustrating a long-term spatiotemporal C. difficile cluster with two identified peaks (15). The 22 cluster isolates are colored according to their peaks, and the two reference strains are marked in green. (A) Minimum-spanning tree of the reanalyzed sequences based on cgMLST targets. Each node represents a unique allelic profile, and the size of the nodes represents the number of isolates. The numbers on connecting lines are the numbers of differing alleles between the genotypes (not to scale), and closely related genotypes (≤6 different cgMLST alleles) are shaded; 2,243 to 2,267 cgMLST target genes (mean of 99.6% of all cgMLST targets) were analyzed. All isolates of peaks 1 and 2 belonged to ST1. (B) Epidemiological curve. Each box represents one isolate, and boxes are colored according to their peak affiliation.

DISCUSSION

Here we describe the establishment of a novel cgMLST scheme for C. difficile, the most common cause of antibiotic-associated gastrointestinal infections worldwide. Based on a collection of isolates that represent the diversity of C. difficile organisms, we were able to construct a robust cgMLST scheme that contains 2,270 targets. This is in concordance with a previous estimate of the number of C. difficile core genes, where—depending on the number of strains and their characteristics—a range of 600 to 3,000 target genes were predicted (31).

Until today, SNV analysis was mainly reported for C. difficile WGS comparisons of circumscribed clinical or epidemiological settings (9, 15, 16). In this study, SNV results were calculated in comparison to a reference sequence among all strains included; any addition of strains would result in novel SNV results, which could lead to conflicting results, as SNV typing does not rely on a fixed nomenclature but is (re)calculated as soon as a novel strain is added. In contrast, cgMLST allows an easy curation of allelic data in a central database, which is a prerequisite for ensuring a universal typing nomenclature as already shown nearly 2 decades ago for “classical” MLST (32). The recently established database (http://www.cgmlst.org/) currently hosts (November 2017) the nomenclature for cgMLST schemes of nine different species (18–21, 33, 34), enabling a uniform typing nomenclature for thousands of genes using next-generation sequencing. The allele-based approach comprises another advantage in comparison to SNV-based approaches: it treats both a mutation that creates an SNV and a recombination that is likely to introduce multiple SNVs as a single evolutionary event (12, 19). Thereby, it compensates for recombination (32), which is helpful for a better definition of genetic relationships in bacteria with higher recombination rates, like C. difficile (35).

The subsequent evaluation of the novel cgMLST scheme employing a diverse collection of isolates as well as strains from clearly defined outbreak situations confirmed the representativeness of the novel scheme, i.e., the typeability (36) (99.3% of all cgMLST targets were successfully extracted from all 2,954 publicly available genomic data) and its ability to type all strains with sufficient discriminatory power to differentiate even among closely related isolates within nosocomial clusters. The high discriminatory power combined with a standardized typing nomenclature, which is crucial for outbreak investigations to facilitate comparison with historical data (12, 19, 37), enabled us to differentiate among epidemiologically related isolates detected during short- and long-term scenarios. Clonal transmissions as well as accidental spatiotemporal clusters could be exactly resolved (Fig. 2 and 3). Even isolates detected more than 1 year apart were still grouped together differing in ≤3 alleles (Fig. 3), which is in line with previous observations that expected 0 to 3 SNVs among transmitted samples within 1 year (16). Moreover, Eyre et al. suggested a threshold of ≥10 SNVs for genetically distinct isolates (16); analogously, we would suggest—adding a 2-fold-higher threshold as determined by our data as a precaution—a threshold of ≥7 alleles difference for isolates being unrelated and ≤6 alleles for isolates that are likely to belong to the same clone. Nonetheless, it has to be noted that typing efforts should always be evaluated in the context of the epidemiological situation.

When introducing novel typing approaches, backward comparability with previous typing methods and a high level of typeability, i.e., the representativeness of a method for any sample in a population, are always great demands. Backward comparability is possible for MLST, in which the ST can be easily extracted from the WGS data in silico, and clustering of cgMLST genotypes is concordant to MLST STs (Fig. S1). However, only limited backward comparability is possible with PCR ribotyping, currently the most widely used typing method for C. difficile. Due to the repetitive nature of the ribosomal operon (part of which is the internal transcribed spacer [ITS] region that is the target region for PCR ribotyping), PCR ribotypes cannot be extracted from draft genomes with any current methodology. To some extent, a correlation of PCR ribotypes and STs is known (38). To assign a PCR ribotype to a new cgMLST cluster, a representative strain would need to be PCR ribotyped. With respect to typeability, we have chosen the BAPS partitioning approach (17, 24, 25) to create an unbiased overview of the population diversity and subsequently randomly selected representative isolates for the cgMLST target definition and sequenced them to achieve highest sequence quality. Another way would be to analyze all available data from public databases; however, the quality is frequently unknown.

In summary, here we present the cgMLST typing scheme for C. difficile with a discriminatory power comparable to that of SNV analysis. The new scheme offers an excellent typing platform that enables local and international comparison of C. difficile isolates and could hence contribute to both better detection or clarification of outbreaks and a deeper understanding of the spread of C. difficile lineages.