Salmonella enterica

MLST databases for Staphylococcus aureus, Streptococcus pneumoniae, Salmonella enterica, Escherichia coli, Enterococcus faecium, Listeria monocytogenes and Enterobacter cloaceae were downloaded from pubmlst.org using the getmlst.py script included with SRST2 (June 2014).
Antimicrobial resistance gene detection was performed using the ARG-Annot database of acquired resistance genes [18 (link)]. Allele sequences (DNA) were downloaded in fasta format [43 ] (May, 2014). Sequences were clustered into gene groups with ≥80% identity using CD-hit [44 (link)] and the headers formatted for use with SRST2 using the scripts provided (cdhit_to_csv.py, csv_to_gene_db.py). A copy of the formatted sequence database used in this study is included in the SRST2 github repository [35 ].
Representative sequences for 18 plasmid replicons were extracted from GenBank using the accessions and primer sequences specified by Carattoli et al. [45 (link)]. A copy of the formatted sequence database used in this study is included in the SRST2 github repository [35 ].

We implemented memory-efficient indexing schemes for the classification of microbial sequences based on the FM-index, which also permits very fast search operations. We further reduced the size of the index by compressing genomic sequences and building a modified version of the FM-index for those compressed genomes, as follows. First, we observed that for some bacterial species, large numbers of closely related strains and isolates have been sequenced, usually because they represent significant human pathogens. Such genomes include Salmonella enterica with 138 genomes, Escherichia coli with 131 genomes, and Helicobacter pylori with 73 genomes available (these figures represent the contents of RefSeq as of December 2015). As expected, the genomic sequences of strains within the same species are likely to be highly similar to one another. We leveraged this fact to remove such redundant genomic sequences, so that the storage size of our index can remain compact even as the number of sequenced isolates for these species increases.
Figure 1 illustrates how we compress multiple genomes of the same species by storing near-identical sequences only once. First, we choose the two genomes (G₁ and G₂ in the figure) that are most similar among all genomes. We define the two most similar genomes as those that share the greatest number of k-mers (using k = 53 for this study) after k-mers are randomly sampled at a rate of 1% from the genomes of the same species. In order to facilitate this selection process, we used Jellyfish (Marcais and Kingsford 2011 (link)) to build a table indicating which k-mers belong to which genomes. Using the two most similar genomes allows for better compression as they tend to share larger chunks of genomic sequences than two randomly selected genomes. We then compared the two most similar genomes using nucmer (Kurtz et al. 2004 (link)), which outputs a list of the nearly or completely identical regions in both genomes. When combining the two genomes, we discard those sequences of G₂ with ≥99% identity to G₁ and retain the remaining sequences to use in our index. We then find the genome that is most similar to the combined sequences from G₁ and G₂ and combine this in the same manner as just described. This process is repeated for the rest of the genomes.
As a result of this concatenation procedure, we obtained dramatic space reductions for many species; e.g., the total sequence was reduced from 661 to 74 Mbp (11% of the original sequence size) in S. enterica and from 655 to 107 Mbp (16%) in E. coli (see Table 1). Overall, the number of base pairs from ∼4300 bacterial and archaeal genomes was reduced from 15 to 9.1 billion base pairs (Gbp). The FM-index for these compressed sequences occupies 4.2 GB of memory, which is small enough to fit into the main memory (RAM) on a conventional desktop computer. As we demonstrate in the Supplemental Methods and Supplemental Table S1, this compression operation has only a negligible impact on classification sensitivity and accuracy.

A total of 316 strains of Salmonella enterica and 1 strain of S. bongori from the collections of the Institute of Medical Microbiology, Jena, and the Robert Koch Institute, Wernigerode, Germany, were used as samples for the evaluation of the SalmoTyper assay. Twenty-five patient isolates of other Enterobacterales species collected at the Institute of Medical Microbiology, Jena, served as negative control strains. All strains were streaked onto Columbia sheep blood agar and Hektoen enteric agar (Oxoid, Thermo Fisher Scientific, Wesel, Germany) and incubated overnight before LAMP testing.
Species were identified using MALDI-TOF (Vitek MS, bioMeriéux, Nürtingen, Germany) and the identity of Salmonella strains was confirmed by serotyping using group-specific and monospecific antisera (Sifin Diagnostics, Berlin, Germany) according to the White-Kauffmann-Le Minor scheme.

In this study, 49 Salmonella enterica strains from the Bacterial Strain Collection of the Laboratory of Immunology and Molecular Biology were included, and Salmonella enteritidis (ATCC® 13076™) were used as a positive control. The strains were previously serotyped using the Kauffmann−White scheme and correspond to the serotypes, namely, S. enteritidis (n = 4), S. typhimurium (n = 2), S. braenderup (n = 1), S. newport (n = 1), S. grupensis (n = 1), and S. uganda (n = 1) isolated from cases of gastroenteritis in humans [22 (link)] and S. paratyphi B (n = 24) and S. heidelberg (n = 15) isolated from poultry farms located in the region of Tolima [23 (link)] and Santander [24 (link)].