Data sets. (i) Assembled genomes.

In August 2010, 1,212 completely sequenced and assembled bacterial genomes were collected from the NCBI Genome database (http://www.ncbi.nlm.nih.gov/sites/genome). For 336 of these genomes, MLST schemes have been developed and are available through the MLST databases (Table 1).

(ii) Sequence reads.

Table 2 shows an overview of the species for which we had short sequence reads, along with the sequencing platforms used. Campylobacter jejuni, E. coli, Klebsiella pneumoniae, Staphylococcus aureus, and Salmonella enterica isolates were sequenced on the Illumina platform generating paired-end reads by TGen (United States). Streptococcus thermophilus, Bifidobacterium animalis, and Lactococcus lactis isolates were sequenced on the Illumina platform generating paired-end reads by Source BioScience (United Kingdom), BaseClear (The Netherlands), and BGI (Hong Kong). Other S. thermophilus, B. animalis, B. longum, and L. lactis isolates were sequenced on the Illumina platform generating single reads by Source BioScience (United Kingdom). P. aeruginosa isolates were sequenced on the Illumina platform generating single reads by Partners HealthCare (Boston, MA) or on the Roche 454 GS platform by the Allegheny-Singer Research Institute (Pittsburgh, PA). WGS data for the 2011 German O104:H4 E. coli outbreak were obtained from publicly available sources. Data from 7 × 314 chips sequenced on the Ion Torrent platform (Life Technologies) for a single isolate were obtained from BGI (23). Data from 8 × 314 chips sequenced on the Ion Torrent platform (Life Technologies) for the outbreak strain LB226692 were obtained from Life Technologies and the University of Münster (19). Illumina MiSeq single-read data for five different isolates of the outbreak strain were obtained from the British Health Protection Agency (HPA). From the Göttingen Genomics Laboratory, we obtained data for two isolates of the outbreak strain sequenced on a Roche 454 GS sequencer. The ABI SOLiD data were the 50 × 50 mate pair data set with 600× coverage of E. coli DH10B, available from the SOLiD software development community.

Table 2

MLST of completely sequenced bacterial isolates using short sequence reads

Sequencing platform	Species	No. of isolates	MLST scheme	Proportion of loci with:		Avg N50	Log avg N50
				Minor mismatches	Major mismatches
Illumina
Paired-end reads	B.animalis	5	Bifidobacterium	0.000	0.029	33,113	4.52
	C. jejuni	53	Campylobacter	0.005	0.000	131,571	5.12
	E. coli	15	E. coli scheme 1	0.010	0.000	195,822	5.29
	E. coli	15	E. coli scheme 2	0.075	0.000	196,463	5.29
	K. pneumoniae	4	K. pneumoniae	0.000	0.000	207,167	5.32
	L. lactis	34	L. lactis	0.564	0.039	44,525	4.65
	S. aureus	83	S. aureus	0.017	0.009	196,736	5.29
	S. enterica	50	S. enterica	0.000	0.009	249,501	5.40
	S. thermophilus	13	S. thermophilus	0.090	0.000	47,686	4.68
Single reads	E. coli	6	E. coli scheme 1	0.000	0.000	46,479	4.67
	B. animalis	2	Bifidobacterium	0.000	0.071	24,979	4.40
	B.longum	2	Bifidobacterium	0.000	0.000	22,548	4.35
	L. lactis	7	L. lactis	0.238	0.119	14,114	4.15
	P. aeruginosa	81	P. aeruginosa	0.019	0.125	9,380	3.97
	S. thermophilus	9	S. thermophilus	0.000	0.000	28,823	4.46
Roche 454	E. coli	3	E. coli scheme 1	0.000	0.000	92,131	4.96
	P. aeruginosa	2	P. aeruginosa	0.000	0.000	60,477	4.78
Ion Torrent	E. coli	2	E. coli scheme 1	0.000	0.500	13,779	4.14
SOLiD	E. coli	1	E. coli scheme 1	0.286	0.000	165,835	5.22

Open in a separate window

Draft assembly of short sequence reads.

If sequence reads are given as input to the MLST server, the reads are assembled de novo prior to ST prediction. Short-read data produced from all major next- and third-generation sequencing platforms, such as the Illumina, Roche 454 GS, and Applied Biosystems SOLiD platforms and the Life Science Ion Torrent personal genome machine (PGM), are supported (8, 18a, 24, 26, 30). The de novo assembly creates contiguous sequences without gaps from the DNA sequence reads, termed contigs, and when paired-end or mate-paired reads are available, these are used to combine the contigs into scaffolds. As a measure of the quality of the draft assembly, the N₅₀ value is calculated for the assembled genomes. The N₅₀ value for contigs or scaffolds is defined as the length of the shortest contig or scaffold in the set of the largest contigs or scaffolds that represents at least 50% of the assembly (20). The assembly is available for download from the MLST server.

Illumina sequence data are assembled using Velvet, version 1.1.04 (34). Prior to assembly, paired-end data are filtered and trimmed using the following steps. (i) All reads containing the character N are removed. (ii) If a read matches at least 15 nucleotides (nt) of a sequencing primer/adaptor, the read is trimmed at the 5′ coordinate of the match. (iii) The 3′ tail is trimmed up to a quality score of 15 (phred scale). (iv) The minimum average quality of the read after trimming is 20. (v) The length of the read after trimming is at least 15 nt. We do not trim Illumina single-end data, since benchmarking showed that this reduced the overall quality of the assemblies and of MLST prediction for the data sets used in the study. Then, in parallel, several assemblies using k-mer sizes from 33% to 80% of the average read length are run, and the assembly with the best cumulative rank for N₅₀, number of contigs, and length of the largest contig is selected as the best assembly.

Both Roche 454 GS and Ion Torrent PGM sequence data are assembled using the Roche proprietary GS De Novo Assembler software, version 2.6 (Newbler 2.6). If given standard flowgram files (.sff), the assembler clips and trims the data prior to assembly.

For Applied Biosystems SOLiD sequence data, assembly is performed using the SOLiD System de novo Accessory Tools, version 2.2. The assembly pipeline uses colorspace Velvet 0.7.55 (34) for the assembly and is run without read error correction (with SAET) or postassembly analysis in order to decrease run time. For all sequencing technologies, single-end, paired-end, or mate-paired reads can be used for assembly.

After uploading of short read data, the assembly is available for download from the MLST server.

MLST of 336 assembled bacterial genomes.

To evaluate our method, we used it for identification of the STs of 336 completely sequenced and preassembled bacterial genomes. These bacteria cover 56 MLST schemes. Table 1 shows the results with regard to the proportion of the MLST alleles in the tested genomes that were previously unseen and hence were not registered in the MLST databases. For 34 MLST schemes, all alleles in the MLST loci in the tested genomes matched perfectly to an allele already registered in the MLST databases (the proportion of new alleles equaled zero), while for the remaining 22 MLST schemes, 0.4% to 73.3% of the MLST alleles in the genomes were not in the MLST databases.

Two MLST schemes exist for E. coli: E. coli scheme 1, which employs seven genes (adk, fumC, gyrB, icd, mdh, purA, recA) (32), and E. coli scheme 2, which employs eight genes (dinB, icdA, pabB, polB, putP, trpA, trpB, uidA) (13). When the 36 completely sequenced E. coli isolates were typed using E. coli scheme 1, only one allele (0.4%) was not in the database. When E. coli scheme 2 was used, 10 alleles (3.1%) were not in the database. This difference in the proportion of previously unseen alleles may reflect either the coverage of the MLST databases (that is, how large a fraction of the total number of alleles they contain) or the rates of evolution of the genes used by the two schemes. The database for E. coli scheme 1 contains on average 228 alleles per locus, while that for E. coli scheme 2 contains on average 143 alleles per locus. Accordingly, the higher number of previously unseen alleles found by using E. coli scheme 2 seems mostly to reflect the fact that this database is less complete than the database for scheme 1.

The three MLST schemes that resulted in the highest number of previously unseen MLST alleles were the Brachyspira (57.1% new alleles), Leptospira (66.7% new alleles), and Streptomyces (73.3% new alleles) schemes. These schemes are meant to cover a whole genus rather than a specific species. As a consequence, these databases are expected to contain far more alleles than databases that aim at covering only a single species. However, this was not the case, as can be seen from Table 1. The Neisseria scheme also aims at covering a whole genus, but here no new alleles were found in the eight Neisseria genomes tested (two Neisseria gonorrhoeae and six Neisseria meningitidis genomes). Indeed, the Neisseria database is the second largest database, containing 561 alleles per locus, in accord with the early establishment of the database in 1998 (18). Of interest, the Helicobacter pylori database contains an average of 2,088 alleles per locus and as such is by far the largest database. Apparently, this does not mean that the database is in any way complete, since more than half of the alleles in the 10 H. pylori genomes tested are not in the database. This observation indicates that the genes selected for the H. pylori MLST scheme (33) are evolving faster than the genes that are generally used in the MLST schemes. This idea is in line with studies showing that in general, H. pylori has high rates of recombination and mutation (5, 7, 29).

Eight bacterial MLST schemes were not tested in this analysis, since we did not have access to complete genomes from these species. However, it is possible to use the MLST Web server with these species as well (Brachyspira intermedia, Burkholderia cepacia complex, Campylobacter helveticus, Campylobacter insulaenigrae, Streptococcus oralis, Streptococcus equi subsp. zooepidemicus, Clostridium septicum, and Chlamydiales spp.).

This work was supported by the Center for Genomic Epidemiology at the Technical University of Denmark and was funded by grant 09-067103/DSF from the Danish Council for Strategic Research.

We are grateful to Mads Bennedsen, Birgitte Stuer-Lauridsen, and colleagues at Chr Hansen A/S (Hørsholm, Denmark) for sharing unpublished genome sequence data. We are grateful to Hans-Henrik Stærfeldt and John Damm Sørensen for excellent technical assistance.