Defining the Core Genome of Salmonella enterica Serovar Typhimurium for Genomic Surveillance and Epidemiological Typing

Abstract

Salmonella enterica serovar Typhimurium is the most common Salmonella serovar causing foodborne infections in Australia and many other countries. Twenty-one S. Typhimurium strains from Salmonella reference collection A (SARA) were analyzed using Illumina high-throughput genome sequencing. Single nucleotide polymorphisms (SNPs) in 21 SARA strains ranged from 46 to 11,916 SNPs, with an average of 1,577 SNPs per strain. Together with 47 strains selected from publicly available S. Typhimurium genomes, the S. Typhimurium core genes (STCG) were determined. The STCG consist of 3,846 genes, a set that is much larger than that of the 2,882 Salmonella core genes (SCG) found previously. The STCG together with 1,576 core intergenic regions (IGRs) were defined as the S. Typhimurium core genome. Using 93 S. Typhimurium genomes from 13 epidemiologically confirmed community outbreaks, we demonstrated that typing based on the S. Typhimurium core genome (STCG plus core IGRs) provides superior resolution and higher discriminatory power than that based on SCG for outbreak investigation and molecular epidemiology of S. Typhimurium. STCG and STCG plus core IGR typing achieved 100% separation of all outbreaks compared to that of SCG typing, which failed to separate isolates from two outbreaks from background isolates. Defining the S. Typhimurium core genome allows standardization of genes/regions to be used for high-resolution epidemiological typing and genomic surveillance of S. Typhimurium.

INTRODUCTION

Salmonella enterica serovar Typhimurium is one of the leading causes of Salmonella-related gastroenteritis in humans (1). Rapid and accurate identification and characterization of S. Typhimurium isolates is one of the most important tasks of public health laboratory surveillance for outbreak investigation and long-term epidemiology, especially when S. Typhimurium outbreaks pose threats of national/international spread (2). Currently, a variety of techniques have been employed to subtype S. Typhimurium for epidemiological purposes (3–6). Traditionally, phage typing was used in Australia and Europe (7, 8). However, our studies showed that phage types may arise independently through mutation, and thus isolates of the same phage type may not share the same evolutionary history (9). Multilocus sequence typing (MLST) has a low resolution to type S. Typhimurium, as the majority of isolates belong to sequence type 19 (ST19) (10). Multilocus variable-number tandem-repeat analysis (MLVA) has been adopted as a standardized method in Australia (9, 11) and Europe (12). However, the numerical dominance of endemic MLVA types reduces the power of MLVA for outbreak investigations. In addition, MLVA has limited value for long-term epidemiology (13).

Next-generation sequencing (NGS) has been increasingly employed to prospectively identify and track outbreaks (14). NGS has major advantages over other pathogen-typing methods, as it promises a standardized universal solution for high-resolution typing. We recently demonstrated the utility and resolution of NGS for outbreak investigation of S. Typhimurium by sequencing 57 isolates from five distinct and epidemiologically confirmed point-source S. Typhimurium DT170 outbreaks in Australia (15). Our study utilized all single nucleotide polymorphisms (SNPs) identified from the genomes of the outbreak isolates. In contrast, Leekitcharoenphon et al. (16) utilized SNPs identified from 2,882 Salmonella core genes for outbreak investigation of S. Typhimurium. The Salmonella core genes were defined by analyzing 73 representative and publicly available genomes from 23 serovars and proposed a core genome typing scheme for S. enterica (16). The use of the core genome for typing allows standardization of genome-based typing. However, since Salmonella is a diverse species with 7 subspecies and more than 2,400 serovars, the species core genome may be substantially smaller than the core genome of a serovar, and thus the use of the species core genome for typing may lead to substantial loss of resolution at the serovar level.

In this study, we defined the core genome of S. Typhimurium using 21 S. Typhimurium reference strains from Salmonella reference collection A (SARA) and 47 publicly available genomes. To obtain the core genome of the serovar, it is critical to select strains to represent the spectrum of diversity. SARA is a collection of 21 diverse S. Typhimurium strains representing 17 electrophoretic types and four sequence types (STs)and 51 strains from four closely related serovars, which are referred to as the S. Typhimurium complex (4). The strains were collected internationally in the 1980s or earlier from various host sources and have been characterized using a range of molecular methods, which showed considerable diversity (5, 6). We further evaluated the performance of epidemiological typing based on Salmonella core genes (SCG), S. Typhimurium core genes (STCG), and STCG plus core intergenic regions (IGRs) using published genomic data from epidemiologically confirmed outbreaks. To avoid confusion over the terminologies used, the S. Typhimurium core genome includes STCG and core IGRs while STCG denotes S. Typhimurium core genes only. S. Typhimurium core genome typing refers to typing using either STCG or STCG plus core IGRs as a generic term if not specified.

MATERIALS AND METHODS

Bacterial strains and genomic DNA isolation.

Twenty-one S. Typhimurium strains from SARA were used (4) (Table 1). These strains were isolated from various sources, four of which came from human sources and eight of which came from animal sources while the remaining nine came from unknown sources. The phenol-chloroform method was used to extract genomic DNA from each strain as described previously (17).

TABLE 1.

Strains sequenced in this study

Strain	ETa	Original name	Source	Locality	STb	Year
SARA1	Tm 1	INSP 24	Human	Mexico	ST19
SARA2	Tm 1	LT2		Lab strain	ST19
SARA3	Tm 1	NVSL 7095	Horse	Rhode Island	ST19	1987
SARA4	Tm 1	NVSL 5820	Rabbit	Indiana	ST19	1986
SARA5	Tm 1	IVB 232		Mongolia	ST19
SARA6	Tm 2	CDC B1213	Human	Ohio	ST19
SARA7	Tm 3	IVB 665/81		Norway	ST36
SARA8	Tm 5	IVB 5560		Finland	ST36
SARA9	Tm 7	NVSL 2816	Parrot	California	ST98	1987
SARA10	Tm 9	NVSL 6814	Opposum	California	ST19	1987
SARA11	Tm 10	IVB 276/25		Thailand	ST19
SARA12	Tm 11	NVSL 6993	Horse	Louisiana	ST19	1987
SARA13	Tm 12	IVB 1430		France	ST19
SARA14	Tm 13	IVB 75/67		Panama	ST19
SARA15	Tm 14	NVSL 6968	Dog	Texas	ST19	1987
SARA16	Tm 15	CDC B1236	Human	North Carolina	ST19
SARA17	Tm 16	IVB 48/81		Yugoslavia	ST19
SARA18	Tm 17	NVSL 6938	Horse	Iowa	ST19	1987
SARA19	Tm 21	INSP 85	Human	Mexico	ST19
SARA20	Tm 22	IVB 1544		France	ST19
SARA21	Tm 23	USFW 318	Heron	Oregon	ST99

^aET, electrophoretic type (4).

^bST, sequence type (5,6).

Genome sequencing, de novo assembly, and core genome analysis.

Genomic DNA was sequenced using the MiSeq genome analyzer (Illumina). We used 250-bp paired-end sequencing. Contigs were assembled de novo using the software package Velvet version 1.0.8 and VelvetOptimiser (18). Large scaffolds and short contigs generated by Velvet were aligned to the S. Typhimurium LT2 genome (GenBank accession number NC_003197) using progressiveMauve version 2.3.1 (19). RAST was used to annotate the sequences from each NGS genome (20). These annotated genes were also grouped into functional categories.

Twenty-one strains from SARA and 47 S. Typhimurium genomes from GenBank were used for the identification of the core genome content (Table 1; see also Table S1 in the supplemental material). The assembled genomes were aligned to the reference S. Typhimurium strain LT2 in progressiveMauve (19). Genes and intergenic regions (IGRs) (>100 bp) that appeared in all isolates were considered core genes and core IGRs, respectively. Mobile genes and repetitive elements were excluded from the core genome. In order to find the core genome of the Typhimurium complex, 9 Salmonella enterica serovar Heidelberg genomes, 2 Salmonella enterica serovar Saintpaul genomes, 3 Salmonella enterica serovar Muenchen genomes, and 2 Salmonella enterica serovar Paratyphi B genomes were used in addition to the 68 S. Typhimurium genomes (see Table S1 in the supplemental material).

To determine whether a stable core genome was obtained, a regression analysis of the number of isolates against their shared genes was performed by fitting a double exponential decay function (21), N_c = θ + k₁ × exp(m₁ × N_g) + k₂ × exp(m₂ × N_g), where θ is a constant value representing the predicted minimum number of core genes, N_c is the number of core genes, N_g is the number of genomes, and k₁, m₁, k₂, and m₂ are parameters where the model was determined by a weighted least-squares regression analysis.

Identification of single nucleotide polymorphisms.

For SNP detection in the draft genomes, SNP calling was conducted as described previously (17). The Burrows-Wheeler alignment (BWA) tool (version 0.7.5) was used to map the reads against the S. Typhimurium LT2 genome (22). SAMtools version 0.1.19 was used to further filter the SNPs identified from BWA mapping by SNP quality (23). SNP calling followed the previously described criteria (17). SNPs with quality score of less than 30 were removed. A custom script was used to determine whether an SNP in the genic region was a synonymous SNP (sSNP) or a nonsynonymous SNP (nsSNP). For the complete genomes, SNPs were determined by using the NUCmer program “show-snps” (with options “-CIlrT”) in the MUMmer package version 3.0 (24).

Phylogenetic analysis.

To evaluate the performance of the S. Typhimurium core genome (STCG with and without core IGRs) typing, 21 strains from SARA sequenced in this study and 94 isolates from 13 different outbreaks were used (see Table S2 in the supplemental material) (14, 15, 25, 26). The selection of outbreaks was either based on representative phage type or ST, including DT135, DT135a, U292, DT3, DT104, DT12, DT120, and ST313. Phylogenetic trees based on SNPs were constructed using the minimum evolution algorithm in MEGA 6.0 (27). Bootstrap analysis was performed with 1,000 replicates. The percentage of concordance was calculated by Ridom EpiCompare (version 1.0) (28). The discriminatory power was assessed by calculating Simpson's index of diversity (D value) (29).

Sequence divergence in coding genes and IGR.

Sequence divergence was estimated using the Kimura 2-parameter (K2P) model implemented in MEGA 6.0 (27). The Student t test was performed with SPSS for Windows version 11.5 (SPSS Inc., Chicago, IL, USA).

Nucleotide sequence accession numbers.

The annotated S. Typhimurium whole-genome sequences were deposited in the Sequence Read Archive under accession numbers SAMN03470046 to SAMN03470066.

RESULTS

General statistics of the SARA genome sequencing.

The 21 SARA S. Typhimurium strains were sequenced using MiSeq 250-bp paired-end sequencing in a multiplexed format. The average number of reads generated per genome was ∼2,651,000. The average coverage depth for all genomes was ∼42×, with a lowest coverage depth of 24× (Table 2). Percentage match of reads to reference strain LT2 chromosome ranged from 96.5% to 99.0%. The reads were assembled de novo. The number of contigs ranged from 228 to 619, with an average of 330. SARA10 had the largest genome size of 5.0 Mb, while SARA16 had the smallest (4.7 Mb).

TABLE 2.

General features of the 21 S. Typhimurium genomes sequenced in this study

Strain	Total no. of reads	N50	No. of contigs	Total length (bp)	Coverage (×)	Match to LT2 chromosome (%)	No. (%) of nonsynonymous SNPs	No. (%) of synonymous SNPs	No. (%) of intergenic SNPs	No. (%) of indels	Total no. of SNPs
SARA1	1,581,754	194,269	283	4,900,132	46	97.8	88 (49.42)	32 (18.00)	50 (28.09)	8 (4.49)	178
SARA2	1,136,806	324,050	307	4,892,963	35	99.0	22 (47.83)	24 (52.17)	0 (0.00)	0 (0)	46
SARA3	1,003,706	172,050	280	4,789,145	34	96.5	189 (42.76)	150 (33.94)	95 (21.49)	8 (1.36)	442
SARA4	1,500,444	222,200	363	4,967,277	39	97.2	363 (42.52)	319 (34.81)	188 (21.32)	12 (1.07)	882
SARA5	3,212,530	168,092	479	4,965,739	46	97.5	6 (21.88)	34 (40.63)	24 (37.50)	0 (0.00)	64
SARA6	2,683,486	225,480	372	4,937,837	64	98.6	204 (43.68)	143 (30.62)	115 (24.63)	5 (1.07)	467
SARA7	2,700,978	240,373	327	4,918,149	63	97.9	1,913 (16.05)	8,509 (71.41)	1,372 (11.51)	122(1.02)	11,916
SARA8	1,291,676	219,093	289	4,807,626	37	97.1	1,844 (16.32)	8,013 (71.71)	1,258 (11.13)	184(1.63)	11,299
SARA9	1,187,376	214,955	288	4,919,127	34	97.5	280 (36.61)	265 (37.48)	154 (21.78)	8 (1.13)	707
SARA10	1,624,246	194,273	273	5,025,508	45	97.0	296 (41.69)	255 (35.92)	151 (21.27)	8 (1.13)	710
SARA11	1,207,350	277,613	307	4,921,121	37	97.2	302 (39.27)	284 (36.93)	165 (21.46)	21 (2.73)	769
SARA12	1,193,274	257,533	265	4,899,590	40	97.2	454 (41.35)	341 (30.05)	260 (23.68)	43 (3.91)	1,098
SARA13	1,172,002	278,182	274	4,856,144	34	97.0	437 (44.26)	310 (29.70)	240 (23.76)	23 (2.28)	1,010
SARA14	1,789,596	356,416	228	4,822,760	69	97.9	233 (44.13)	153 (28.98)	107(20.27)	35 (6.63)	528
SARA15	2,668,118	270,525	242	4,862,737	49	97.5	20 (7.75)	159 (61.63)	74 (28.68)	5 (1.94)	258
SARA16	2,930,600	144,590	619	4,727,259	37	96.5	77 (31.40)	218 (47.67)	99 (19.77)	0 (0.00)	394
SARA17	545,298	118,536	433	4,905,073	50	97.1	194 (43.69)	156 (35.14)	90 (20.27)	4 (0.90)	444
SARA18	1,284,792	149,251	387	4,769,756	37	97.6	219 (41.17)	192 (36.09)	117 (21.99)	4 (0.75)	532
SARA19	2,430,606	223,024	313	4,940,693	35	97.3	104 (45.61)	61 (26.75)	57 (25.00)	4 (1.75)	228
SARA20	819,486	272,276	332	4,930,836	24	97.3	173 (45.53)	115 (30.27)	86 (22.63)	4 (1.05)	380
SARA21	1,073,366	306,796	250	4,853,595	34	97.0	360 (47.57)	231 (30.52)	162 (21.40)	4 (0.53)	757

Identification of the S. Typhimurium core genes.

Overall, the pan genome that includes all genes present in one or more of the 21 strains contained 4,955 genes, of which 4,177 were present in all strains. We screened the 4,177 genes against 47 genomes selected from available S. Typhimurium genomes in GenBank, from which we excluded genomes identical or closely related to outbreak strains. We removed remnant prophage and mobile genes from the core genes even if they were present in all strains but kept virulence genes and genes on genomic islands (GIs). In the final set, 3,846 genes (see Table S3 in the supplemental material) were defined as the S. Typhimurium core genes (STCG). These core genes were present as 64 conserved segments with an average of 60 genes per segment, the longest segment having 249 genes (see Table S3 in the supplemental material). The 3,846 core genes were unevenly distributed across the functional categories (Fig. 1A). Core genes account for 98.87% and 97.19% in the functional categories of protein metabolism and respiration, respectively.

FIG 1.

Core genome variation of S. Typhimurium. (A) Functional catalogues of core and noncore genes in the LT2 genome. A, Cofactors, vitamins transport, and metabolism; B, cell wall and capsule; C, virulence, disease and defense; D, membrane transport; E, iron acquisition and metabolism; F, nucleotide transport and metabolism; G, carbohydrates metabolism; H, fatty acids, lipids, and isoprenoids metabolism; I, nitrogen, sulfur, and phosphorus metabolism; J, amino acid transport and metabolism; K, metabolism of aromatic compounds; L, cell division and cell cycle; M, motility and chemotaxis; N, regulation and cell signaling; O, secondary metabolism; P, phages, prophages, transposable elements, plasmids; Q, respiration; R, stress response; S, protein metabolism; T, general prediction only or unknown function. The catalogues of gene functions were based on the subsystem of RAST. Error bars indicate variations in the number of accessory genes among the 21 SARA strains. (B) Estimation of core genome size. The graph shows the descending trend of the core genome size of S. Typhimurium with the increasing number of genomes. The number of shared genes is plotted against the number of genomes that were sequentially added. Black squares are the averages of such values. Error bars indicate variations in the number of core genes among different strains.

To determine whether the diversity and number of strains used were sufficiently large to derive a stable core genome, we used the statistical model developed by Bottacini et al. (21) by random sampling of the 68 genomes (see Table S1 in the supplemental material) to fit a double exponential decay function. The regression function is expressed as N_c = 3,811 + 679.14 × exp(−1.184 × N_g) + 345.86 × exp(−0.0039 × N_g) (R² = 0.9991) (Fig. 1B), and the predicted minimum number of genes was 3,811. The number of core genes (3,846) for the 68 genomes derived above is close to the predicted value, which suggests that we derived a stable core genome. The fitted curve reached a plateau at 55 genomes (see Fig. S1A in the supplemental material), which is the expected minimum number of genomes that would be required to obtain the core genome content.

The S. Typhimurium complex includes four serovars closely related to S. Typhimurium (6). For the 3,846 S. Typhimurium core genes identified, 3,809 genes were shared by all strains in the S. Typhimurium complex while 37 genes were present only in S. Typhimurium (see Table S4 in the supplemental material). Interestingly, most of the S. Typhimurium unique core genes are located between 3.7 Mb and 4.7 Mb in LT2 (see Fig. S1B in the supplemental material), including a part of the rfb gene cluster, sugR, rhuM, oadA, oadB, ilvA, yjdC, mgtA, and other nonessential genes. These genes were variably present in S. Paratyphi B SPB7, S. Heidelberg SL486, S. Muenchen baa1594, and S. Saintpaul SARA26 (see Table S4 in the supplemental material). Most of these genes encode proteins that are associated with extracellular components or functions, including putative transporters, putative cytoplasmic proteins, and putative membrane proteins. In general, genes encoding proteins located outside the cell are known to be more variable (30). Five genes from rfb gene clusters are absent in S. Muenchen. This may reflect the difference between S. Muenchen (a serovar of the O:8 [C₂-C₃] serogroup) and the other four serovars (O:4 [B] serogroup) in the O-antigen gene cluster.

Characterization of core IGRs.

We identified the S. Typhimurium core IGRs using the procedure similar to that used to identify the core genes. Noncoding IGRs greater than 100 bp were extracted according to the coordinates of their adjacent consecutive coding genes for LT2. There are 1,857 IGRs, of which 281 were variably present in the 68 S. Typhimurium strains, and 1,576 were core IGRs (see Table S5 in the supplemental material). The size of core IGRs ranged from 101 bp to 12,151 bp, with an average length of 285 bp. The largest IGR has a length of 12,151 bp, which is a large repetitive region in S. Typhimurium.

To determine whether the core IGRs evolve at a similar rate to the core genes, we performed a pairwise comparison of core genes and core IGRs between LT2 and D23580, which are the only complete S. Typhimurium genomes. The divergence was 2.12 × 10⁻⁴ substitutions per site and 2.15 × 10⁻⁴ substitutions per site for core genes and core IGRs, respectively, indicating that the core IGRs and core coding regions evolve at a similar rate (t test, P = 0.31; n = 1576) (see Fig. S2 in the supplemental material).

Identification of genome SNPs in the SARA strains.

SNP calling was performed for the 21 SARA strains sequenced in this study. The total SNPs ranged from 46 to 11,916 (Table 2). Note that SARA2 is also a LT2 strain. To examine whether the SNPs observed in SARA2 are due to sequencing errors of the reference sequence, we investigated three other LT2 genomes in the public database (accession numbers ERS007491, ERS007500, and AHUZ00000000). The number of SNPs ranged from 56 to 68. Moreover, all of the 44 SNPs observed in SARA2 also existed in these genomes, indicating that they are probably not due to sequencing errors.

We further analyzed the nature of the SNPs of the SARA genomes. The SNPs were divided into four categories: nonsynonymous (nsSNP), synonymous (sSNP), intergenic (IG), and single base indels. IG SNPs on average accounted for approximately 21% of the total number of SNPs in each strain with the exception of SARA2, which had no IG SNP (0%). The average proportion of nsSNP for each strain was 37.6% ranging from 7.8% to 49.4% while the sSNPs ranged from 18.0% to 71.7% with an average of 39.1%. Single base indels accounted for only 1.8% of the SNPs. SARA8 had the highest number of indels (184 indels) representing 1.63% of the SNPs. Interestingly, SARA7 and SARA8 had 71.4% and 71.7% SNPs belonging to sSNP, many of which were shared between the two strains. Thirty-two genes among 21 SARA strains terminated earlier due to a stop codon and led to proteins that were >20% shorter compared to strain LT2 (see Table S6 in the supplemental material). Thus, these genes were considered pseudogenes. We then extracted the S. Typhimurium core genome SNPs and SCG SNPs. The average number of S. Typhimurium core genome SNPs and SCG SNPs for 21 SARA strains were 1,267 and 825, respectively, indicating that the S. Typhimurium core genome covers 34.8% more SNP sites than SCG (Table 3).

TABLE 3.

Average number of SNPs found in three typing schemes

Group	Total no. of SNPs	No. of STCG plus core IGR SNPs	No. of STCG SNPs	No. of SCG SNPs
SARA	1,577	1,267	957	825
Outbreak 1	869	719	630	560
Outbreak 2	717	582	454	398
Outbreak 3	670	422	302	261
Outbreak 4	815	625	482	417
Outbreak 5	629	485	381	332
Outbreak 6	649	545	415	345
Outbreak 7	643	510	420	360
Outbreak 8	694	584	457	400
Outbreak 9	788	678	599	249
Outbreak 10	711	620	554	230
Outbreak 11	675	605	539	240
Outbreak 12	672	604	533	233
Outbreak 13	708	623	540	231

Application of S. Typhimurium core genome for epidemiological typing.

With the STCG plus core IGRs defined above, we compared the resolution of the use of STCG, STCG plus core IGRs, and SCG for epidemiological typing of S. Typhimurium. The SCG of 2,882 genes was previously defined by Leekitcharoenphon et al. (16). The publicly available genome sequences of S. Typhimurium strains were used as test samples, including strains from 8 prolonged outbreaks (over 1 year) and 5 short duration outbreaks (ranging from 9 to 46 days) (see Table S2 in the supplemental material). SNPs were called by comparison with LT2 and were used to compare the typing methods.

For the isolates from the 13 outbreaks, the number of SNPs ranged from 629 to 869 (Table 3). The SNPs for STCG plus core IGRs for the 13 outbreaks were between 422 and 719, which were only slightly lower than the total number of SNPs, while the STCG SNPs ranged from 302 to 630. SCG SNPs ranged from 230 to 560, which is 43.5% lower on average than STCG plus core IGRs.

We used several parameters to measure the difference in resolving power between S. Typhimurium core genome typing and that of SCG typing. Phylogenetic trees constructed using the minimum evolution method (31) were used to determine whether STCG and SCG resolve the strain relationships differently. Simpson's index of diversity (D value) was used to measure differences in discriminatory power within an outbreak, while SNP difference was used to measure the separation of outbreak isolates from background isolates or the nearest isolates unrelated to a given outbreak.

As shown in Fig. 2, STCG typing achieved 100% separation of the outbreaks (Fig. 2A). However, the performance of the SCG typing was much poorer (Fig. 2B). Two isolates from outbreak 2 were assigned to different branches. Moreover, it is notable that the SCG tree failed to distinguish some background isolates from outbreak isolates. Strain 0909R12120 (231) was not related to any of the outbreaks but was clustered with isolates from outbreak 6. The epidemiological concordance of SCG typing decreased to 97.9%.

FIG 2.

Phylogenetic relationships of S. Typhimurium isolates based on S. Typhimurium core genes (A) and Salmonella core genes (B). The minimum evolution method was used to infer evolutionary relationships of the isolates. The numbers in brackets represent outbreak numbers. Bootstrap was performed with 1,000 replicates. The bootstrap values (1,000 replicates; >50%) are shown next to the branches. The unit of the scale bar indicates the evolutionary distance in substitutions per nucleotide. The solid circles and squares indicate the wrongly placed isolates from outbreaks 2 and 6, respectively, by SCG typing.

We further examined the level of differentiation to the genotype level and calculated Simpson's index of diversity (D value) within the 13 outbreaks. Overall, compared to those of STCG typing, the number of genotypes and the D values were decreased in SCG typing (Fig. 3A and B).

FIG 3.

Comparison of core genome typing. (A) Number of genotypes among 13 outbreaks based on STCG plus core IGRs (blue bars), STCG (red bars), and SCG (green bars). (B) Simpson's index of diversity (D value) among 13 outbreaks based on STCG plus core IGRs (blue bars), STCG (red bars), and SCG (green bars). (C) The average number of SNP difference between outbreak isolates and background strains/closest outbreak-unrelated isolates based on STCG plus core IGRs (blue bars), STCG (red bars), and SCG (green bars).

The SNP differences between outbreak-related strains and background strains in STCG and SCG typing as measured by the average number of SNP difference are shown in Fig. 3C. The SNP differences in STCG typing averaged 39 SNPs with a smallest difference of 7 SNPs. In contrast, SCG typing separated background isolates from outbreak isolates by as few as 3 SNPs (outbreak 13) with an average of 27 SNPs.

We further examined STCG plus core IGRs. It separated all outbreaks from each other and from background isolates (see Fig. S3 in the supplemental material). The discriminatory power of STCG plus core IGRs for typing was slightly higher than STCG typing (Fig. 3). For 3 of the 13 outbreaks, adding core IGRs increased the resolution, while for the remaining 10 outbreaks the resolution was the same as that for STCG typing.

DISCUSSION

In this study, we sequenced 21 SARA strains to determine the core genome of S. Typhimurium, which consisted of 3,846 core genes and 1,576 core IGRs. The S. Typhimurium core genome contained 956 (33.45%) more genes than the Salmonella core genome, which contains only 2,882 genes as defined previously (16). We showed that the use of the STCG or the STCG plus core IGRs increases the resolution of the genome sequencing for epidemiological typing using published and epidemiologically investigated outbreaks (14, 15, 25, 26).

The number of core genes in a serovar depends on the sampled diversity of the given serovar. We used SARA strains that were selected previously based on multilocus enzyme electrophoresis (MLEE) diversity (5, 6) and also cover most of the continents and seven major STs. The 21 SARA strains were obtained from nine different countries with different prevalence profiles of S. Typhimurium infections, including four European countries (Norway, France, Yugoslavia, and Finland), three countries of the American continent (United States, Panama, and Mexico), and two Asian countries (Mongolia and Thailand). Other isolates from public databases include 13 isolates from Denmark, nine isolates from Africa (Malawi), nine isolates from Australia, six isolates each from the United Kingdom and the United States, three isolates from China, and one isolate from Japan. We used a statistical model to estimate the underlying number of core genes by fitting a double exponential decay function and found that the core genome content reached a plateau in the sampling, suggesting that the core genome has reached its stable stage from the strains analyzed and adding more strains would not substantially reduce the core genome size.

We also analyzed the core intergenic regions for consideration as a source of SNPs to increase the power of genome-based typing. Intergenic regions are generally considered to have less or no functional constraints, and thus mutational changes in IGRs are thought to be generally neutral and have little or no selective cost (32). The substitution rates for IGRs are, therefore, expected to be higher than those for housekeeping genes, where there are major functional constraints on mutation as many will be deleterious. Interestingly, the overall substitution rate was similar between core genes and core IGRs. Core IGRs may contain regulatory elements such as small RNAs (sRNAs), which are highly conserved. Hu et al. used IG SNPs for typing different phage types of S. Typhimurium previously (32). Thus, IGRs may increase the resolution of genome-based typing. We found 1,576 core IGRs with an average size of 285 bp and a total size of 448,581 bp, contributing to 11.5% of the S. Typhimurium core genome. For the 68 genomes used for determining core IGRs, we found that 316 had variations, including SNP changes in 296 IGRs. However, no variation was observed in 1,260 core IGRs. The Salmonella core IGRs were not defined previously (16), and a comparison cannot be made. We expect that conservation of Salmonella core IGRs will be similar in proportion to that observed for the core genes at the species level.

Using the isolates from 13 outbreaks, we demonstrated that the use of STCG and S. Typhimurium core genome (STCG ± core IGRs) increased the resolution of typing. SCG typing offered the poorest resolution, as some outbreak isolates cannot be differentiated from background isolates (outbreak 6). By total number of bases, SCG covers 2.9 Mb while STCG covers 3.5 Mb and STCG plus core IGRs covers 3.9 Mb. The resolution of the typing scheme increases with the total number of bases under consideration. This was reflected in the number of SNP sites observed in the 21 SARA strains with 957 and 1,267 SNP sites in total for STCG and STCG plus core IGRs, an increase of 16% and 54%, respectively, compared to the 825 SNP sites covered by SCG. Additionally, typing using STCG and STCG plus core IGRs increased the differentiation of isolates within an outbreak. For 3 outbreaks, the inclusion of core IGR increased the differentiation compared to that of STCG typing. Compared to SCG typing, the use of STCG or STCG plus core IGRs increased the differentiation of isolates within an outbreak for 7 and 8 outbreaks, respectively. Although the discriminatory power of typing using STCG ± core IGRs is no better than that of SCG typing to subtype isolates for some outbreaks, this situation is likely to change when there is an increased number of background isolates sequenced from the same region and/or time period (15). There will be an increased need for higher resolution to differentiate outbreak isolates from closely related and/or endemic isolates.

NGS data may be used to trigger an investigation of an outbreak either locally or globally. However, considering that there are variations between isolates within an outbreak and background isolates may be very closely related to isolates from a potential outbreak, genome sequence alone may not be adequate for the determination of an epidemiological link. Our previous study showed that to rule in or rule out whether an isolate is part of an outbreak requires the determination of a cutoff of the number of SNP differences between isolates being analyzed (15). We found a cutoff of a maximum of 4 SNP differences for an ex vivo/in vivo evolution time of up to 40 days using the fastest mutation rate known (15). However, for prolonged outbreaks such as outbreak 1, which lasted more than 2 years, one may need to set a larger cutoff as the expected number of SNPs is proportional to the ex vivo/in vivo evolution time. Therefore, for prolonged outbreaks, differentiation of outbreak isolates from background isolates becomes more challenging, as larger cutoffs may falsely rule in background isolates as outbreak isolates. Defining the S. Typhimurium core genome will help to further determine appropriate cutoff values for outbreak investigations.

Our study highlights the importance of defining the core genome at an appropriate level for epidemiological typing to retain a sufficient number of core genes/regions to achieve the resolution required for short- and long-term epidemiology. For Salmonella, there is a clear subspecies structure and good separation of serotypes; defining the core genome at serovar level offers the best resolution for epidemiological typing of a serovar. For species that contain high diversity and with less distinctive subspecies structure, the core genome at species level is likely to become progressively smaller as the number of genomes sequenced increases, which clearly would lead to a substantial loss of discriminatory power. The core genome of Escherichia coli is estimated to be only around 1,500 to 2,000 genes (33, 34). This led to the proposal of a soft core genome by Kaas et al. (34), which is defined as core genes found in 95% of all the analyzed genomes in contrast to the usual 100%. The E. coli soft core genome was found to consist of 3,051 genes, nearly twice the core genome defined traditionally (33, 34). We did not determine whether the soft core genome approach would alleviate the need to define the core genome at serovar level in Salmonella. Serovar-level core genome typing offers the best resolution to meet the needs of outbreak investigation, as only a few serovars predominate in human infections (35, 36).

The SARA genomes also contained rich genome content variation, including plasmids, prophages, and antibiotic resistance genes (see Text S1 in the supplemental material). An alternative approach to core genome typing is pan-genome typing, including all accessory genes to distinguish isolates based on the presence or absence of genes (14). The study by Leekitcharoenphon et al. (14) compared pan-genome typing and core genome typing and found that pan-genome typing had only 64% to 65% concordance with core genome typing based on the phylogenetic trees derived, and four outbreaks cannot be separated by pan-genome typing. Clearly accessory genome variation within and between outbreaks is not sufficient to support a high-resolution epidemiological typing.

In conclusion, our analysis of 21 SARA strains and 47 publicly available genomes found that the core genome of S. Typhimurium is significantly larger than the Salmonella core genome. We show that the S. Typhimurium core genome offers higher resolution for epidemiological typing of outbreaks. In this study, we employed a core genome SNP-based approach for epidemiological typing for the comparison of discriminatory power. Genome typing can be based on either SNP comparison or gene by gene comparison (37). The S. Typhimurium core genome defined in this study allows standardization of the genome regions to be used for high-resolution typing for genomic surveillance and public health investigation of outbreaks.