The 160 reference strains included in this study represent 41 different serovars of S. enterica subsp. enterica (subsp. I). The six Salmonella subspecies belong to the Salmonella Reference Collections SAR A (72), B (72), and C (16) (Beltran et al., 1991; Boyd et al., 1993, 1996). Recently corrections have been made to certain serovars in the SAR A and B collections (Achtman et al., 2013). Figures 1A,B show the distribution of Salmonella species, subspecies and serovars. The Salmonella fliC, gnd and mutS genes were selected as candidate targets for the development of the PCR-RFLP. The fliC gene encodes for the phase 1 flagellar antigen and it is present in all Salmonellae (Mcquiston et al., 2004). For the phase 1 antigen, 52 antigenic factors and 61 serotypes (single factors or combinations of factors) have been distinguished (Li et al., 1994). The gnd gene codes for 6-phosphogluconate, an enzyme of the pentose-phosphate pathway, and is located between the rfb locus and the highly variable cld gene (Nelson and Selander, 1994; Thampapillai et al., 1994). The mutS gene, a key component of the methyl-directed mismatch repair system, acts as barrier to horizontal gene transfer by blocking recombination of diverged DNA (Brown et al., 2002, 2003).

Isolates were grown in Tryptic Soy agar (TSA) plates (Difco, BD, Sparks, MD). A single colony was grown in a shaking incubator overnight at 37°C in Brain Heart Infusion broth (BHIB) (Difco). One ml of the broth culture was transferred to a 1.5 ml microcentrifuge tube and centrifuged at 12,000 rpm for 3 min. Total genomic DNA was isolated using the Promega Wizard Genomic DNA Purification kit (Promega Corporation, Madison, WI). The primers used for the amplification of the selected genes are described in Table S1. DNA amplification by PCR was performed in a reaction volume of 50 μl consisting of 25 μl of Qiagen Hot StarTaq Plus master mix (Qiagen, Valencia, CA), 20 μM primer mix, 10 ng of total genomic DNA and volume was completed with molecular biology grade water. Initial denaturation was carried out for 5 min at 95°C. Thirty cycles of amplification were performed in a DNA Engine Tetrad2 Peltier Thermal Cycler (Bio-Rad, Hercules, CA). Each cycle consisted of three steps: denaturation for 30 sec at 94°C, annealing for 30 s at 60°C, and extension for 1 min at 72°C. An additional step of extension for 7 min at 72°C was performed at the end of the amplification to complete extension of the primers. Amplification products were detected by resolving 1 μl of the PCR product using the Agilent DNA 7500 kit and the 2100 Agilent Bioanalyzer (Agilent Technologies, Inc., Santa Clara, CA).

We used the In silico (http://insilico.ehu.es) database to virtually test PCR primers and select possible restriction enzymes to test experimentally during RFLP (Bikandi et al., 2004; Roberts et al., 2007). The Salmonella database consisted of 27 genomes representing 15 species (S. bongori), subspecies (S. enterica subsp. arizonae) and S. enterica serovars: Agona, Choleraesuis, Dublin, Enteritidis, Gallinarum-Pullorum (2), Heidelberg (2), Newport, Paratyphi A (2), Paratyphi B, Paratyphi C, Schwarzengrund, Typhi (3) and Typhimurium (8). Enzymes showing the most number of different restriction patterns among the 27 Salmonella genomes were selected for pilot experiments.

The PCR products were cut using the following restriction enzymes: fliC was cut with HhaI and Sau3AI; gnd with AciI and AluI; and mutS with AciI and HaeII. Restriction enzymes from New England Biolabs (Ipswich, MA) and Fermentas (Glen Burnie, MD) were used during the development of the molecular typing method. Single digestions were done by mixing 5 μl of the selected PCR product and 2.5U of NEB endonucleases or 1 Fast digest unit of the Fermentas endonucleases in final volume of 10 μl. NEB endonuclease mixtures were incubated for 1 h at 37°C. Fast digest mixtures were incubated for 10 min at 37°C. After incubation, DNA digestion was terminated by heat inactivation at 65°C for 20 or 10 min depending on the enzyme used or by the addition of 20 mM EDTA. In selected experiments, restriction digestions were cleaned using the MinElute Reaction Cleanup kit (Qiagen). Restriction digestions were repeated two to three times to test reproducibility of the restriction patterns. Restriction patterns were analyzed using the Agilent DNA 1000 kit and the 2100 Agilent Bioanalyzer (Agilent Technologies, Inc.).

Data files containing RFLP patterns from the 2100 Agilent Bioanalyzer were exported as data set tables in CSV format. These were then imported into BioNumerics version 6.6 (Applied Maths, Inc., Austin, TX). The relationships between restriction patterns were calculated by cluster analysis for each and/or combination of restriction patterns using Ward and DICE coefficient with optimization of 1% and tolerance of 0.25%. Ward and Dice were used as recommended by the Guidelines for the validation and application of typing methods for use in bacterial epidemiology (Van Belkum et al., 2007). All the nucleotide sequencing was performed in both directions through MCLAB (San Francisco, CA) and assembled into single complete sequences using the CLC Main Workbench software version 6.8.2 (Aarhus, DK). The fliC, gnd, and mutS genes were sequenced using the primers in Table S1 in all Salmonella collections. The mutS in the SAR B; mutS and gnd in the SAR C collections were obtained from GenBank (NCBI) in FASTA format (Brown et al., 2002, 2003). Sequences for the seven MLST housekeeping genes, aroC, dnaN, hemD, hisD, purE, sucA, and thrA for SARA and B collections were obtained from the NCBI database (Bell et al., 2011) and the MLST Databases at the ERI, University College Cork, respectively. MLST was performed on the 16 Salmonella strains composing the SAR C collection (Boyd et al., 1996; Maiden et al., 1998; Enright and Spratt, 1999). All primers sequences of the seven MLST genes, for amplification and sequencing are described in Table S1. These primers contain M13/pUCR forward (5′-CCCAGTCACGACGTTGTAAAACG-3′) and reverse (5′-AGCGGATAACAATTTCACACAGGAA-3′) universal sequencing priming sites.

PCR cycling conditions were as follows. Initial denaturation was carried out for 5 min at 95°C. Thirty-five cycles of amplification were performed in a DNA Engine Tetrad2 Peltier Thermal Cycler (Bio-Rad). Each cycle consisted of three steps: denaturation for 1 min at 94°C, annealing for 1 min at 55°C, and extension for 1 min at 72°C. An additional step of extension for 5 min at 72°C was performed at the end of the amplification to allow complete extension of the primers. Amplification products were detected by resolving 1 μl of the PCR product using the Agilent DNA 1000 kit and the 2100 Agilent Bioanalyzer (Agilent Technologies, Inc.).

The fliC, gnd, and mutS genes were aligned using BioEdit version 7.1.11, and trimmed with GeneStudio version 2.2.0.0. The seven housekeeping genes sequences were aligned with allele templates from the MLST Database, and then aligned and trimmed as described above. Then, the sequences were queried to the MLST Database website for allele number assignment. Concatenated analyses of fliC, gnd, and mutS; and the seven housekeeping genes were conducted using MEGA software version 5.0.5, using neighbor-joining method with Tamura-Nei distance and 1000 bootstrapping replicates (Felsenstein, 1985; Saitou and Nei, 1987; Tamura et al., 2011).

Salmonella enterica serovars Newport, Saintpaul, and Typhimurium were selected for artificial contamination of alfalfa sprouts, jalapeno peppers and tomatoes, respectively (CDC, 2006, 2008, 2010). These serovars have been previously implicated in outbreaks related to these food commodities. Salmonella inocula for artificial contamination of produce were prepared as described before (Zhang et al., 2011). Alfalfa sprouts, jalapeno peppers and tomatoes were obtained from local supermarkets. These food commodities were processed as described before (Zhang et al., 2011). Briefly, for each Salmonella serovar and their corresponding produce, two 25 g portions of food were placed aseptically into a sterile Seward stomacher bag (Seward, United Kingdom). The two portions were designated as A, for no inoculation and B, for high-level inoculation (10⁵ CFU/ml). The jalapeno peppers and tomatoes were chopped aseptically in a blender into sizes similar to what is present in regular, chunky salsa and then weighed before being placed into preenrichment bags. One ml of the selected Salmonella serovar at the indicated concentration was added to the 25 g produce portion. For the no inoculated control one ml of the MDR buffer was used. Bags were massaged gently by hand for 1 min and kept at 4°C for 2 h. For enrichment, 225 ml of universal enrichment broth (Difco) were added to the bags. Bags were then shaken vigorously by hand for 30 s, and incubated (without shaking) at 35 ± 1°C for 24 ± 1 h.

One ml aliquots were taken from each bag (A and B) for DNA extraction, serial dilutions further microbiological testing. Four different DNA extraction methods were tested. First, a 1 ml sample was heated at 100°C for 12 min and then centrifuged for 2 min at 16,000× g (Eppendorf, New York). One ml samples were centrifuged and the pellets resuspended in 100 μl of sterile distilled water and boiled, or DNA was extracted using either the Epicenter Quick DNA extraction (Madison, WI) and the Promega Wizard Genomic DNA Purification following the instructions of their manufacturers. Samples were stored at −20°C. Microbiological analysis of 24 h per-enrichment samples was as previously described (Andrews et al., 2007; Zhang et al., 2011). Identification and confirmation of Salmonella were done using Biolog GEN III plates (Biolog, Inc.; Hayward, CA). Salmonella serotyping was done following the standard protocol for molecular determination of serotype in Salmonella based on the Bioplex technology (Fitzgerald et al., 2007; Mcquiston et al., 2011).

All primers and probes used in this study were purchased from IDT (Coralville, IA) and are given in Table S1. Real-time PCR was done as described before (Deer et al., 2010). Briefly, qPCRs were done using the QuantiFast Multiplex PCR using all the DNA templates following the recommended protocol (Qiagen). Each 25 μl reaction contained 1× Master Mix (HotStarTaq Plus DNA Polymerase, QuantiFast Multiplex PCR Buffer, and dNTP mix), 400 nmol/l IAC primers, 200 nmol/l IAC probe and 1 μl DNA IAC template (1 · 1 pg/μl). For the multiplex reactions, invA primers, invA_176F and invA_291R; and probe invA_Tx_208 were added at 200 and 150 nmol/l, respectively. The qPCR conditions were as follows: 95°C for 5 min (for polymerase activation) and 40 cycles of 95°C for 45 s and 60°C for 45 s with fluorescence acquisition for both, Cy5 and Texas Red, following each 60°C step. All qPCR assays were run in CFX96 Real-Time System (Bio-Rad). The term Cq is equivalent to the original CT (threshold cycle) terminology according to the Minimum Information for Publication of Quantitative Real- Time PCR Experiments (MIQE) guidelines (Bustin et al., 2009, 2010). Conventional PCR for fliC, gnd, and mutS was done as described in the Materials and Methods Section.

Virtual PCR was done using the designed fliC, gnd, and mutS specific gene primers against the Salmonella database (Table S1). The virtual analysis showed all (100%) of the Salmonella strains were PCR positive for the fliC gene (Table 1). However, the virtual PCR simulation predicted negative results for three out of 27 (11%) database strains for the gnd and mutS genes: S. bongori str. NCTC 12419, S. enterica subsp. arizonae 62:z4,z23:- and S. enterica subsp. enterica serovar Newport str. SL254 (Table 1). Experimental PCR confirmed fliC amplification in all the tested strains. Contrary to the PCR database predictions, S. bongori (2) and S. enterica subsp. arizonae (2) were PCR positive for the mutS gene. As previously reported, one out of the three S. enterica subsp. enterica serovar Newport (SAR B37) was PCR negative for mutS (Brown et al., 2003). In agreement with the In silico database, 100% of the S. bongori (2) were PCR negative for the gnd gene experimentally (Table 1 and Figure 2).

To confirm our experimental results, we tested six more strains of S. bongori belonging to the Systems and Assays for Food Examination (SAFE) Reference Collection (Mcquiston et al., 2008): 94-0708 (V 48:i:-), 95-0123 (V 40:z35:-), 96-0233 (V 44:z39:-), CNM-256 (V 60:z41:-), CNM262 (V 66:z41:-), 95-0321(V 48:z35:-), and repeated strains SAR C11 and 12 for PCR amplification of the gnd gene. One hundred % of the of S. bongori were PCR negative for the gnd gene under our experimental PCR conditions (Figure 2). These data validate the predicted results from the In silico website with respect to S. bongori. A possible cause for the lack of amplification of the gnd gene in S. bongori strains can be primer-template mismatches. Mismatches located in the 3′-end region of a primer have significantly larger effects on priming efficiency than mismatches located at the 5′-end (Beard et al., 2004; Johnson and Beese, 2004; Stadhouders et al., 2010). We identified mistmatches in the 3′-end of the gnd F-1 primer (Table S1) by aligning 16 gnd Salmonella gene sequences from the SAR C Reference Collection (Boyd et al., 1996) (Figure S1). This mismatch resulted in the differentiation of S. bongori strains from the other Salmonella specie and subspecies. To confirm the specificity of these results, with conducted an exclusivity PCR test (defined here as the lack of the signal or negative reaction on closely related non-Salmonella strains) against a panel of 20 different gram-positive and gram-negative bacterial strains (Table S2). In this test, 100% of the exclusivity strains tested were PCR negative for the fliC and gnd genes. Five out 20 (25%) showed lower levels of mutS PCR product as compared to a Salmonella positive control. A PCR profile with positive fliC and mutS genes in combination with a negative PCR amplification of the gnd gene suggests the presence of S. bongori, while S. enterica will exhibit a positive PCR profile for the three genes. These data suggest that the two species of Salmonella can be differentiated by PCR using the described three PCR amplification profile.

To select the restriction enzymes to be used for the RFLP experimentally, we conducted virtual RFLP of fliC, gnd, and mutS genes for 27 Salmonella sequenced strains in the In silico database. Analysis of the banding patterns showed several enzymes that produced four or more different restriction patterns per specific gene tested. Pairs of such restriction enzymes were chosen specifically for each of the three genes to generate experimental restriction patterns: fliC gene, HhaI and Sau3AI; gnd, AciI and AluI; and mutS, AciI and HaeII.

Given the fact that we had four out the 27 Salmonella strains with complete genomes in the database, we decided to validate the predicted RFLP patterns. We used the data obtained from the following sequenced available genomes: S. enterica subsp. arizonae 62:z4,z23:- (SAR C5), serovar Paratyphi A str. ATCC 9150 (SAR B42), Paratyphi C str. RKS4595 (SAR B49) and Typhimurium str. LT2 (SAR A2). The predicted restriction patterns were compared to the experimental data generated using the 2100 Agilent Bioanalyzer (Table S3). The degree of agreement between the predicted number of fragments and the total size, and our experimental RFLP for the four Salmonella strains was evaluated. All (100%) of the simulated restriction patterns were different as compared to the experimental ones. The differences in total size of the predicted and the experimental fragments varied from 0.6 to 10.4%.

Restriction patterns were resolved using the Agilent DNA 1000 kit (Agilent Technologies). This kit reports a sizing accuracy of ±10%, depending upon the fragment size range, and a sizing resolution that varies from ±5 bp, ±5% and ±10% in the fragments ranging from 25–100, 100–500, and 500–1000 bp, respectively (Agilent Technologies). When resolving restricted DNA using the Bioanalyzer, adding EDTA and/or using heat inactivation of the restriction enzymes is recommended to avoid possible degradation of the internal DNA marker (Agilent Technologies). We tested the effect of adding 20 mM EDTA, heat inactivation, and the use of a commercially-available method for cleaning restriction digestion reactions on the resolution of restriction fragments and reproducibility of restriction patterns in the 2100 Agilent Bioanalyzer. No significant differences in the number of restriction fragments or the sizes obtained among treatments were observed (Figure 3). Although minor differences were detected among fragment sizes between 2 and 5 bp, no degradation of the 1500 and 15 bp internal markers were observed (Figure 3). Given these results, we chose the use of 20 mM EDTA for the inactivation of the restriction enzymes. Although heat inactivation is a cheaper alternative, this step adds 10–20 min to the procedure depending on the restriction enzyme in use.

The relationship among restriction patterns was analyzed by cluster analysis. A total of possible 63 individual and multiple combinations were analyzed. Dendrograms were drawn using BioNumerics (Applied Maths). S. enterica is comprised of six subspecies: enterica (I), salamae (II), arizonae (IIIa), diarizonae (IIIb), houtenae (IV), and indica (VI). For simplicity, S. bongori is still commonly referred to as subsp. V (Tindall et al., 2005). Some derivatives of S. enterica subsp. houtenae (IV) have been reported and identified as subgroup VII (Boyd et al., 1996). Based on biotype these are very similar to subsp. IV but can be distinguished by multilocus enzyme electrophoresis (Boyd et al., 1996). To establish whether PCR-RFLP has the potential to differentiate among Salmonella subspecies we conducted cluster analysis as described in the Material and Methods Section. Based on the distribution of subspecies and the number of members in each subspecies group (Figure 1A), we expected that the best clustering would consist of seven to eight clusters depending on whether the derivatives of subsp. IV could be separated in two distinct clusters. These theoretical subspecies clustering show a discriminatory power (DP) equal to 0.167 (Hunter and Gaston, 1988; Hunter, 1990). The current RFLP cluster analysis showed that restriction patterns obtained cutting the mutS gene with the restriction enzyme AciI was indeed sufficient to differentiate the different subspecies of Salmonella (Figure 4). S. enterica subsp. II, IIIa, IIIb, IV, V, and VI were grouped in single homogeneous clusters (Figure 4). S. enterica subsp. I was grouped into six homogeneous clusters consisting of 109, 2, 15, 3, 5, and 12 members, respectively. This clustering corresponds to a DP of 0.5219. This suggests that Salmonella subspecies can be differentiated by mutS-AciI RFLP cluster analysis.

There are a total of 2,579 serovars in the genus Salmonella distributed between the two species and six subspecies, the bulk of which (1531 serovars) are in S. enterica subsp. enterica (Grimont and Weill, 2007). The current study represents 41 serovars of S. enterica subsp. enterica (Figure 1B). To establish whether PCR-RFLP has the potential to differentiate among Salmonella species, subspecies and serovars, we conducted cluster analysis as described in the Material and Methods Section. Based on the distribution of S. enterica subsp. enterica serovars and the five subspecies we expect that the best clustering for our subset should consist of the following: 28 homogeneous clusters containing at least two representatives of a selected serovar or subspecies, and 19 individual serovars (due to their representation with 1 member), for a total of 47 different types among the 160 strains. This clustering distribution corresponds to a DP of 0.9331 (Hunter and Gaston, 1988; Hunter, 1990).

The six restrictions patterns obtained by the digestion of the fliC, gnd, and mutS genes were analyzed in BioNumerics. We obtained best differential clustering using the combination of the fliC gene cut with HhaI and Sau3AI; gnd gene cut with AciI and AluI; the mutS gene cut with HaeII. Forty-three different clusters and eleven single serovars were identified for a total of 54 different types (Figure 5). This cluster distribution corresponds to a DP of 0.9725. The 43 clusters and the relationship among strains on each cluster are described in Table 2. Twenty-six out of 43 (60.5%) clusters consisted of different homogeneous serovar groups. Nineteen out of the 28 (68%) serovars and/or subspecies represented by more than one strain were grouped into homogeneous clusters (Table 2). In twelve out of these 19 (63.2%) homogeneous clusters representing serovars and subspecies containing more than one strain, 100% of the representing strains were grouped together. Seventeen out of 43 (39.5%) clusters were defined as Type I clusters consisting of S. enterica subsp. enterica strains. Five (29.4%), five (29.4%) and eight (47%) of the 17 Type I clusters did not share, shared one or two elements in their antigenic formula, respectively. Four out of five (80%) clusters sharing one element shared either H1 or H2 antigen. Five out of eight (62.5%) clusters sharing two elements shared the O and the H2 antigen. Thirty-seven % (3/8) shared both H1 and H2 antigens.

To test the applicability of the PCR-RFLP in the detection of S. enterica subsp. enterica in contaminated produce after 24 h of pre-enrichment, we artificially inoculated food commodities with Salmonella serovars known to have been responsible for past outbreaks associated with the those food commodities: alfalfa sprouts with S. Newport, jalapeno peppers with S. Saintpaul, and tomatoes with S. Typhimurium (CDC, 2006, 2008, 2010). Food commodities were inoculated as described before the Materials and Methods Section. Four 1 ml aliquots were collected after 24 h pre-enrichment and different DNA extraction methods were tested to assess the effect of these different extraction methods on the amplification of the fliC, gnd, and mutS genes. We used a Salmonella-specific qPCR as a comparator (Deer et al., 2010). Conventional PCRs of fliC, gnd and mutS genes were affected by the DNA extraction and the food commodity (data not shown). Salmonella spp. was detected by qPCR in all DNA extraction methods in all food commodities tested. However, a positive amplification of the three RFLP genes by conventional PCR was obtained using the commercially available DNA extraction kit in jalapeño peppers and tomatoes. In tomatoes, pelleting bacteria from pre-enrichment followed by resuspension in water and boiling was also a good source for DNA template for conventional PCR. Although amplification of the three RFLP genes was observed using DNA extracted with the commercially available kit from alfalfa sprouts pre-enrichment, the yield of PCR product was too low for further manipulation (data not shown). Restriction patterns obtained from pre-enrichment samples exhibited identical patterns when compared to pure culture controls.

We next compared our PCR-RFLP method with the BAM standard method (Andrews et al., 2007). After artificial inoculation Salmonella strains were detected by the BAM standard method, confirmed by biochemical fingerprint (Biolog, Inc.) and serotyped using the Salmonella standard molecular serotyping method (Table 3). Using our previously established cluster analysis, PCR-RFLP identified and serotyped S. Saintpaul and S. Typhimurium after 24 h pre-enrichment (Table 3). However, low amplification yield for S. Newport precluded further manipulation of three RFLP genes. Following the BAM Method, all three Salmonella strains used for the artificially inoculation were isolated, identified and serotyped from all food commodities.

The length of the seven concatenated housekeeping genes was 3,138 bp. The number of variable nucleotides was 59.4% among the 160 Salmonella species, subspecies, and serovars. Evolutionary analyses were conducted in MEGA5 (Tamura et al., 2011). The evolutionary history was inferred using the Neighbor-Joining method (Saitou and Nei, 1987). The evolutionary distances were computed using the Maximum Composite Likelihood method (Tamura et al., 2004) and are in the units of the number of base substitutions per site (Figure 6). Forty-one different clusters and 21 single serovars were identified for a total of 62 different types (Figure 6). This cluster distribution corresponds to a DP of 0.9652. The 41 clusters and the relationship among strains on each cluster are described in Table 4. Twenty-six out of 41 (63.4%) clusters consisted of different homogeneous serovar groups. Twenty out of the 28 (71%) serovars and/or subspecies represented by more than one strain were grouped into homogeneous clusters (Table 4). In eleven out of these 20 (52.4%) homogeneous clusters representing serovars and subspecies containing more than one strain, 100% of the representing strains were grouped together. Fourteen out of 41 (34.1%) clusters were defined as Type I clusters consisting of S. enterica subsp. enterica strains. Five (35.7%), one (7.1%) and eight (57.1%) of the 14 Type I clusters did not share, shared one or two elements in their antigenic formula, respectively. Seven out of eight (87.5%) clusters sharing two elements shared O and the H2 antigen. The remaining cluster sharing two elements of the antigenic formula was composed of strains sharing the O and the H1 (1/8; 12.5%). Only one cluster out of 41 (2.4%) was composed of strains from two different subspecies (Table 4).

Concatenated analysis of three RFLP genes showed the following. The length of the concatenated fliC, gnd, and mutS genes was 3,907 bp. The number of variable nucleotides was 73.5% among the 160 Salmonella species, subspecies, and serovars. Evolutionary analysis, history and distances were determined as for the housekeeping genes in MEGA5 (Saitou and Nei, 1987; Tamura et al., 2004, 2011) (Figure 7). Forty-four different clusters and 34 single serovars were identified for a total of 78 different types (Figure 7). This cluster distribution corresponds to a DP of 0.988. The 44 clusters and the relationship among strains on each cluster are described in Table 5. Twenty-two out of 44 (50.0%) clusters consisted of different homogeneous serovar groups. Thirteen out of the 28 (46.4%) serovars and/or subspecies represented by more than one strain were grouped in homogeneous clusters (Table 5). In six out of these third-teen (46.1%) homogeneous clusters representing serovars and subspecies containing more than one strain, 100% of the representing strains were grouped together. Twenty-one out of 44 (47.7 %) clusters were defined as Type I clusters consisting of S. enterica subsp. enterica strains. Six (28.6%), four (19.0%) and ten (47.6%) of the 21 Type I clusters did not share, shared one or two elements in their antigenic formula, respectively. Three out of four (75%) clusters shared only the O antigen. Nine out of 10 (90.0%) clusters sharing 2 elements, shared the O and the H2 antigen. Cluster 20 (4.8%) is a Type I cluster composed of two members that shared the H1 and H2 flagellar antigens and differed in one element of the O antigen. Only one cluster out of 44 (2.7%) was composed of strains from two different subspecies but shared the H1 antigen (Table 5).

MLST assigns an independent allele number based on sequence differences to each of the seven housekeeping genes. The combination of alleles defines an individual strain multilocus sequence type (ST) (Maiden et al., 1998). To test a different approach in the analysis of PCR-RFLP data collected we assigned a numerical identifier to each of the 930 restriction patterns generated by the restriction digestion of fliC, gnd, and mutS genes PCR products and investigated the relatedness of the 160 Salmonella strains by assigning a restriction type (RT). To assign the restriction patterns numbers we took in consideration the following: number of bands, differences in fragments sizes and presence and/or absence of a fragment (Van Belkum et al., 2007). Based on that, we identified 71 and 39 different restriction patterns by cutting the fliC gene with HhaI and Sau3AI, respectively. In the case of the gnd gene, we found 39 and 23 different restriction patterns by cutting with AciI and AluI, respectively. For the mutS gene, 41 and 40 different restriction patterns were assigned after digesting the PCR product with AciI and HaeII, respectively.

To assign the numerical RT we used the same restriction patterns that best clustered the different species, subspecies, and serovars (Figure 5). We assigned 128 different RTs vs. 87 different STs among the 160 strains studied (Tables S4, S5). Among the species, subspecies and serovars with more than one representative (28 possible clusters), we identified in eleven out of 28 (39.3%) the same number of RTs and STs. In four out of the 28 possible homogeneous clusters (14.3%) no STs were identified in the database. These strains belong to S. bongori, and S. enterica subspecies arizonae, houtenae and salamae. In third-teen out of 28 (46.4%) a higher number of RTs than STs were assigned. Higher diversity in the number of RTs increased with the number of representative in a specific Salmonella serovar. In the case of S. Typhimurium, Paratyphi B, Heidelberg, Muenchen and Saintpaul, 15, 12, 8, 7, and 10 different RTs were assigned vs. 4,7,2,5, and 4 STs, respectively (Tables S4, S5). To compare and illustrate the clonal structure of S. Typhimurium, Paratyphi B, Heidelberg, Muenchen and Saintpaul derived from STs and RTs we used eBURST program (Feil et al., 2004). Results are summarized in Table 6. Given the fact that more RTs were assigned as compared to STs, a more complex clonal structure is observed in all five serotypes using the RTs (Table 6). S. Typhimurium clonal structure using the STs consisted of a founder ST 19 containing the majority of the strains (20 out 26, 77%), and connecting two single locus variable STs, 98 and 99, and a singleton ST 36. The S. Typhimurium clonal structure based on RTs is composed of one founder RT 6 connecting two subgroup founders RTs 9 and 38; and 3 singletons RTs 33, 92 and 93. Subgroup founder RT 9 is diversified by a third subgroup founder 69 (Table 6). It is interesting to mention that based on ST only singletons STs were observed in S. Muenchen. However, using RTs three different clonal groups were identified (Table 6). The typeability of the RT approach when compared to cluster and concatenated sequence analysis of fliC, gnd and mutS was superior (Table S6). We tried to determine epidemiology concordance in respect to the source of isolation. Even though some RTs were unique to the source of isolation for some Salmonella strains, the strain specific information was limited to reach a conclusion (Table S7).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.