Salmonella enterica Diversity in Central Californian Coastal Waterways

Abstract

Salmonella enterica is one of the most important bacterial enteric pathogens worldwide. However, little is known about its distribution and diversity in the environment. The present study explored the diversity of 104 strains of Salmonella enterica isolated over 2 years from 12 coastal waterways in central California. Pulsed-field gel electrophoresis (PFGE) and multilocus sequence typing were used to probe species diversity. Seventy-four PFGE patterns and 38 sequence types (STs) were found, including 18 newly described STs. Nineteen of 25 PFGE patterns were indistinguishable from those of clinical isolates in PulseNet. The most common ST was consistent with S. enterica serovar Typhimurium, and other frequently detected STs were associated with the serovars Heidelberg and Enteritidis; all of these serovars are important etiologies of salmonellosis. An investigation into S. enterica biogeography was conducted at the level of ST and subspecies. At the ST and subspecies level, we found a taxon-time relationship but no taxon-area or taxon-environmental distance relationships. STs collected during wet versus dry conditions tended to be more similar; however, STs collected from waterways adjacent to watersheds with similar land covers did not tend to be similar. The results suggest that the lack of dispersal limitation may be an important factor affecting the diversity of S. enterica in the region.

INTRODUCTION

Salmonella enterica is the most common etiologic agent of bacterial gastrointestinal illness worldwide (1). It is acquired via direct or indirect contact with feces from an infected animal, usually through contaminated food or water. In countries with improved sanitation, Salmonella enterica is most commonly spread via contaminated food products. In the United States it is estimated that approximately a million cases of salmonellosis occur annually (2), costing over a billion dollars (3).

Salmonella is frequently isolated from fresh and marine waters, including those used for recreation and shellfish harvesting (4–6). Upstream land activities can contribute Salmonella to environmental waters via human sewage, urban and agricultural runoff, and feces of wildlife and domestic pets. Salmonella has been isolated from marine organisms—including mammals and invertebrates (7, 8), reported to persist in the environment for extended periods of time (9)—and from natural waters and wildlife in the same geographic region (10). Thus, natural waters provide a vehicle for dissemination of Salmonella in the environment and a route of transmission among hosts.

There are more than 2,500 Salmonella serovars (11), and little is known about the occurrence and distribution of these in the environment. Most serovars are zoonotic pathogens encompassing a broad host range, including rodents, birds, domestic livestock, pets, wildlife, and humans (10). The few serovars that are considered host specific include Typhi and Paratyphi A (human), Gallinarum and Pullorum (poultry), and Dublin (primarily cattle) (12, 13). A few studies have examined the distribution of Salmonella serovars in aquatic environments (Table 1). A study of coastal waters in northern Africa serotyped 10 Salmonella isolates from seawater and found two serovars, with S. enterica serovar Senftenberg being the most common (14). In Spain, S. Senftenberg was the most common serovar of 127 isolates recovered from shellfish and seawater comprising 20 different serovars (15). Polo et al. (16) identified 55 different serovars among 823 isolates from Spanish environmental waters; S. Enteritidis was the most prevalent. In Portugal, 45 Salmonella isolates were recovered from coastal waters, and S. Virchow was the most prevalent of the 17 identified serovars (17). Baudart et al. (18) found 35 different serovars among 413 strains isolated from French aquatic environments; S. Typhimurium was the most common. In Salinas Valley, CA, 6 different serovars were identified among 18 isolates from water, the most common having the antigenic formula 6,8:d:− (10). Haley et al. (6) identified 13 serovars among 197 isolates from surface waters in Georgia, USA; the majority of the isolates were S. enterica subsp. arizonae. Finally, a San Francisco Bay study identified 8 different serovars in wetlands and sloughs managed as bird habitat (19). Each of these studies used similar methods to cultivate and serotype Salmonella: cultivation in nonselective broth, in selective broth, and then on selective solid media, followed by biochemical confirmation tests using serotyping with antisera (Table 1). Although these studies provided some data on the temporal or spatial occurrence of different Salmonella serovars, full investigations of their biogeography were not undertaken.

Table 1.

Summary of studies reporting on serovars present in natural waters

Location	Matrix	Vol/mass	Detection method	Typing method	No. of isolates	No. of serovars	Most common serovar in water	Reference
Spain	Seawater and shellfish	25 g	ISO 6579 (51): buffered peptone water, selenite cysteine broth, RV10 broth, and Hektoen enteric, phenol red-brilliant, and bismuth sulfite agar, followed by biochemical screening	Antisera	127	20	S. Senftenberg	15
France	River water	1 liter	Buffered peptone water, RV10 broth, and Salmonella-Shigella agar, followed by biochemical screening	Antisera	413	35	S. Typhimurium	18
Morocco	Seawater	100 ml	ISO 6579 (51): buffered peptone water, selenite cysteine broth, RV10 broth, and Hektoen enteric, phenol red-brilliant, and bismuth sulfite agar, followed by biochemical screening	Antisera	10	2	S. Senftenberg	14
Georgia, USA	River water	100 ml	MPN method: buffered peptone water, Rappaport-Vassiliadis broth, and xylose lysine deoxycholate agar, followed by biochemical screening	Antisera	197	13	S. enterica subsp. arizonae	6
California, USA	Freshwater	100 ml or Moore swabs	Various methods used, including those modified from Andrews and Hammack (52) and USEPA method 1682 (30): nonselective enrichment, selective enrichment, selective agar, and biochemical screening	Antisera	18	6	6,8:d:−	10
Portugal	Estuarine and seawater	1 liter	MPN method (53): buffered peptone water, NRV(10, 54) broth, Hektoen enteric and xylose lysine deoxycholate agar followed by biochemical screening	Antisera	45	17	S. Virchow	17
Spain	Seawater and freshwater	1 liter	Buffered peptone water, semisolid Rappaport medium, selenite enrichment broth, and xylose lysine deoxycholate and Salmonella-Shigella agar, followed by biochemical screening	Antisera	823	55	S. Enteritidis	16
California, USA	Wetland water	500 ml	MPN method modified from EPA method 1682: tryptic soy broth, modified semisolid Rappaport-Vassiliadis broth, and xylose lysine deoxycholate, followed by biochemical screening	Antisera	Not reported	8	Not reported	19

There are three different standard techniques for investigating Salmonella diversity: serotyping by immunologic characterization of two surface structures (O polysaccharide [O antigen] and flagellin protein [H antigen]), pulsed-field gel electrophoresis (PFGE), and multilocus sequence typing (MLST) (20). Repetitive extragenic palindromic sequence-based PCR (REP-PCR), has also been used to investigate Salmonella diversity (21, 22). In the present study, we utilize PFGE and MLST for Salmonella isolate characterization. These methods are useful in ecological and epidemiological studies for describing distribution patterns, revealing environmentally persistent strains, and providing clues about the evolution and epidemiology of disease. In the United States, the Centers for Disease Control and Prevention (CDC) in cooperation with state and local public health entities monitors and investigates Salmonella food-borne outbreaks and epidemiology. Public health agencies including the CDC, the U.S. Department of Agriculture (USDA), and the U.S. Food and Drug Administration (FDA) use PFGE in conjunction with other epidemiologic or investigational data to link Salmonella outbreaks to their source(s). Grouping indistinguishable PFGE patterns in a given time frame into PulseNet clusters is an integral part of this process (23). PFGE analysis is the gold standard for molecular subtyping because of its reproducibility, resolution, standardized protocols, and long-time utilization in national databases (PulseNet/VetNet) (http://www.cdc.gov/pulsenet/). MLST is based on sequence analysis of chosen housekeeping genes and is becoming the method of choice for determining the global epidemiology of bacterial pathogens (e.g., Vibrio parahaemolyticus and V. cholerae) (24–27). Due to the inherent sequence-based analysis, MLST provides consistent characterization of bacterial isolates from one laboratory to the next. The sequences typically are stored in a public database that can be readily accessed (Salmonella enterica MLST database; http://mlst.ucc.ie/mlst/dbs/Senterica). MLST studies provide a better understanding of the genetic relatedness of strains within a species and have identified lateral-gene-transfer events (28, 29). MLST has recently been argued as a superior method for classifying S. enterica than serotyping using antisera (20).

Previously, we isolated Salmonella in 13 of 14 coastal water bodies along the central California coast during a 2-year period (4). Salmonella was more likely to be isolated from bodies of water adjacent to subwatersheds with urban land cover than from those abutting forested or agricultural land covers (4). We sought to extend the previous study by performing genetic subtyping of Salmonella isolates. We used PFGE and MLST to investigate Salmonella diversity and biogeography among Salmonella isolates recovered from 12 of the sites.

MATERIALS AND METHODS

Sampling locations and times.

Fourteen central Californian bodies of water were sampled during the study (Table 2). Samples were collected every 6 weeks from January 2008 to November 2009 (n = 18 for each site) from all sites but the Salinas and Pajaro Rivers. The Salinas and Pajaro Rivers were sampled every 6 weeks from February 2008 and February 2009, respectively (n = 17 and n = 8, respectively). Details of water sample collection and transport are reported elsewhere (4). In the field, water quality sensors (Hach Hydrolab Quanta, Loveland, CO, or YSI85, YSI, Inc., Yellow Springs, OH) were used to record temperature and salinity.

Table 2.

Locations where sampling was conducted^a

Site	Latitude	Longitude
San Francisquito Creek	37.465	–122.118
Petaluma River	38.1124	–122.5
Napa River	38.0945	–122.258
Lagunitas Creek	38.0633	–122.82
Bolinas Lagoon	37.9075	–122.682
San Pedro Creek	37.5962	–122.506
Pescadero Creek	37.266	–122.412
Wadell Creek	37.0964	–122.278
Soquel Creek	36.9716	–121.952
San Lorenzo River	36.964	–122.013
Salinas River	36.7436	–121.799
Kirby Park (Elkhorn Slough)	36.8398	–121.744
Pajaro River	36.8496	–121.807
Moss Landing	36.8039	–121.613

^aA map can be found in reference 4.

Identification of Salmonella isolates.

Salmonella cells were enumerated by adapting U.S. Environmental Protection Agency (EPA) method 1682 for detection of Salmonella in biosolids (30) to a most-probable-number (MPN) method for water samples (4). In all, nine tubes were included in the MPN assay; three different volumes of water (10 ml, 100 ml, and 1 liter) were run in triplicate. Presumptive colonies from each of the nine tubes yielding positive culture and biochemical test results were confirmed as Salmonella by using PCR to assay for the presence of the invA gene (4). Each putative Salmonella colony was placed in 20 μl of DNase-free water (Invitrogen) and boiled to lyse the cells. The lysate was centrifuged to sediment cell debris, and 2 μl of the lysate was used as a template in invA PCR. Each PCR mixture consisted of 22 μl of PCR SuperMix (Invitrogen) and a 0.2 μM concentration of each primer (139F and 141R) (31). Cycling conditions included an initial denaturation step of 94°C for 2 min, followed by 30 cycles of 94°C for 30 s, 64°C for 30 s, and 72°C for 30 s and a final extension step at 72°C for 7 min. PCR products were separated in 1.5% agarose gels and examined by UV transillumination using a UV gel imager (Bio-Rad, Hercules, CA). All confirmed Salmonella isolates were further characterized by PFGE or MLST. If two or more isolates from the same water sample, collected at the same place and time, were determined to be genetically identical based on both PFGE and MLST, then only one was retained for our data set.

PFGE.

Salmonella isolates were further typed using PFGE according to the 24-h standardized protocol described by PulseNet (32). Plugs were digested using either 50 U of XbaI or 30 U of BlnI (New England BioLabs, Ipswich, MA). Gels were photographed using a Gel Doc 1000 (Bio-Rad). PFGE patterns were analyzed using the BioNumerics software (version 6.0; Applied Maths, Kortrijk, Belgium). PFGE pattern analysis and genetic similarity coefficients were calculated using Dice's correlation at 1.5 and 1.2% band position tolerance, respectively. Dendrograms were constructed by the complete linkage method using the combined XbaI and BlnI PFGE patterns. Pattern numbers were assigned to isolates using both PFGE patterns together.

MLST.

Salmonella isolates were further characterized using the seven housekeeping gene scheme MLST for S. enterica (http://mlst.ucc.ie/mlst/dbs/Senterica). The targeted genes included thrA (aspartokinase plus homoserine dehydrogenase), purE (phosphoribosylaminoimidazole carboxylase), sucA (α-ketoglutarate dehydrogenase), hisD (histidinol dehydrogenase), aroC (chorismate synthase), hemD (uroporphyrinogen III cosynthase), and dnaN (DNA polymerase III beta subunit). The primers used for PCR are those described for sequencing of each gene, except that the M13F and M13R sequencing primers (IDT, Coralville, IA) were added to the sequence of each PCR primer at the 5′ end to facilitate subsequent sequencing (see Table S1 in the supplemental material). Once the sequences were obtained, the allele numbers and sequence types (STs) were assigned according to the S. enterica MLST database (http://mlst.ucc.ie/mlst/). Further, serovar and subspecies assignments were inferred from the ST by querying the S. enterica MLST database. The program eBURST v3.0 was used to identify the different clonal complexes (http://eburst.mlst.net). The most restrictive group definition was used to define the clonal complexes; i.e., at least six of the seven alleles had to be identical for isolates to be included in the same group or clonal complex (33). The statistical confidence levels of the ancestral types were assessed using 1,000 bootstrap resamplings. Two different STs are considered single-locus variants (SLVs) when they differ from each other at a single locus. Double-locus variants (DLVs) are any two different STs differing in two loci.

Concatenated sequences of the seven housekeeping genes (3,336 bp) were used for phylogenetic analysis. Minimum-evolution (ME) trees for the concatenated sequences were constructed with Mega 3.1 software (34) using the Kimura two-parameter model to estimate the genetic distances. The statistical support of the nodes in the ME tree was assessed by 1,000 bootstrap resampling.

Comparison with clinical strains in PulseNet.

Isolates predicted, based on their ST, to be one of the top 10 serotypes causing salmonellosis infection by the CDC (http://www.cdc.gov/foodnet/data/trends/tables/table5.html) were chosen for comparison to clinical isolates in PulseNet. A total of 25 isolates generating unique PFGE patterns by restriction enzyme digestion with XbaI were queried against PulseNet PFGE patterns uploaded between 1 June 2010 and 28 February 2013. The data uploaded to PulseNet during this period of time (2.5 years prior to the query) are readily accessible by authorized users.

Ancillary data.

The primary land cover within each subwatershed from which samples were collected was deemed to be primarily urban, agricultural, or forested according to a protocol described elsewhere (4). Each sampling date was characterized as wet or dry based on the amount of rainfall during the 7 days proceeding the sampling date, as described by Walters et al. (4). Dissolved inorganic nitrogen concentrations in the water samples were reported previously (4) and were measured after filtration, using a nutrient autoanalyzer.

Biogeographic analysis.

Analyses were carried out using PRIMER version 6 (PRIMER-E, Ltd., Plymouth, United Kingdom) unless otherwise noted. Species accumulation curves (35, 36) and Chao1 diversity estimators (37) were generated using observed PFGE and ST data. The species accumulation curve allows one to assess the degree to which taxon diversity has been sampled in the system; a steep curve indicates that diversity is undersampled, whereas a curve that asymptotes suggests that increased sampling will not uncover more taxa. The Chao1 estimator uses a nonparametric approach to provide the number of taxa predicted to be present as the number of isolates typed tends toward infinity; it is calculated using information on the number of taxa that appear 1 or 2 times in the pooled samples (37). The similarity of Salmonella PFGE patterns and STs collected among sampling events (defined by a location and time) was determined using Bray-Curtis similarity coefficients (38). STs were also classified into their respective clusters (formed during creation of the ME tree, described above; clusters are associated with subspecies designations), and the similarity analysis was repeated. Taxon-distance and taxon-time relationships were investigated with Matlab R2009b (Natick, MA), using nonparametric Mantel tests with dissimilarity matrices and geographic and temporal distance matrices (39). The geographic distance between the sampling events was determined by using the spherical law of cosine distance from latitude and longitude coordinates of each site (39). Taxon-environment relationships were investigated by determining the correlation between sampling event Salmonella isolate dissimilarities and sampling event environmental distances (39). Log-transformed dissolved inorganic nitrogen concentration, log-transformed salinity, log-transformed rainfall in the last 7 days, and temperature were standardized, and the Euclidean distance, between sampling events was calculated. Environmental variables were used individually and in combination with other environmental variables to identify the best predictors for the taxon distribution (BEST function in PRIMER v.6). Finally, an analysis of similarity (ANOSIM) (40) was used to explore the extent to which primary land cover (urban, agricultural, or forested) and wet versus dry conditions could explain the genetic similarity between the Salmonella strains found during each sampling event. The results were deemed statistically significant if P was less than 0.05.

Nucleotide sequence accession numbers.

New alleles found for thrA, purE, sucA, hisD, aroC, hemD, and dnaN genes were deposited in GenBank (KC262729 to KC262745) and are available at http://mlst.ucc.ie/mlst/dbs/Senterica.

RESULTS

Occurrence and concentration of Salmonella spp.

The occurrence and concentration of Salmonella spp. during the sampling campaign were published by Walters et al. (4) and are summarized briefly for context. Salmonella was detected at least once at each site except for Kirby Park. Salmonella was detected in 0 to 72% of the samples collected per site, at concentrations ranging from <0.75 to 7.25 MPN per liter. The highest concentrations were consistently found at San Pedro Creek; this site also had the greatest percentage of positive detections. Note that the single isolate obtained from Bolinas Lagoon was lost, so isolates from 12 of the 14 original sites were used in the present analysis. In total, 306 isolates from 60 sampling events were characterized by using PFGE and MLST methods. Of these 306, 104 isolates were retained for our analysis. The remaining 202 isolates had identical PFGE and MLST patterns as an isolate cultivated from the exact same water sample and thus were considered replicates for the purpose of the present study. Between one and seven unique Salmonella genotypes were collected during each of the 60 sampling events (Fig. 1).

Fig 1.

Diversity of S. enterica isolates from bodies of water along the coast of California. PFGE patterns and the corresponding dendrogram for 102 of 104 isolates obtained in the present study are depicted. The different clusters observed are designated on the left side of the figure. The ST and PFGE pattern numbers are provided. Isolate identification numbers are also indicated. These include a number composed of the date in a “MonthDayYear” format that the strain was collected (e.g., strain “61208” was isolated 12 June 2008), followed by the site-dilution-replicate information. The latter gives the site name, followed by the volume of the water sample (4, 40, or 400 ml), followed by the replicate (A, B, or C) from which the isolate was obtained (see Materials and Methods). Site name definitions are as follows: Lagunitas (L), Moss Landing (ML), Napa River (N), Salinas River (NSA), Pescadero Creek (Pes), Petaluma River (Pet), Pajaro River (PR), San Francisquito Creek (SF), San Lorenzo River (SL), Soquel Creek (SO), San Pedro Creek (SP), and Wadell Creek (W).

PFGE analysis.

Analysis of 104 Salmonella isolates by PFGE using enzymes XbaI and BlnI resulted in 74 distinguishable patterns, demonstrating a high level of genetic diversity among the isolates (Fig. 1; also see Table S2 in the supplemental material). Five Salmonella isolates were untypeable by PFGE. After the addition of thiourea to the running buffer, only two isolates remained untypeable. Most PFGE patterns were represented by a single isolate. The six most frequently occurring PFGE patterns accounted for 25% (n = 26) of all isolates (n = 102). The frequencies of the most common PFGE patterns were as follow: 8 (4% of the isolates), 44 (3%), 48 (7%), 56 (3%), 57 (4%), and 68 (5%). The PFGE patterns had an overall similarity of 43.4% and grouped into four main clusters (Fig. 1). The majority of the isolates (65%, n = 68) clustered into group III, the PFGE patterns of which showed an overall similarity of 45.4%. Three other groups were observed (representing 35%, n = 34) clustering with diverse overall similarities (I, within cluster similarity = 44.5%, n = 17; II, 46.7%, n = 11; and IV, 51.2%, n = 6) (Fig. 1). The PFGE pattern accumulation curve (Fig. 2) indicates that more sampling effort would uncover more genotypes. The Chao1 diversity index, which represents an estimate of the number of PFGE patterns actually present in the sampled waters had a complete census been taken, is 149 ± 31.

Fig 2.

Observed species-accumulation curves using MLST and PFGE data. Error bars represent the standard deviations.

Salmonella MLST profiles.

Novel allele types were identified for all housekeeping genes (Table 3). Thirty-eight discrete STs were identified among the 104 Salmonella isolates, indicating a high degree of genotypic diversity. Of these 38 STs, 23 were represented by single isolates, 15 were represented by more than one isolate (n = 2 to 27), and 18 were new STs (Table 3). The observed ST accumulation curve, created with data from all sites, is provided in Fig. 2. A comparison of the ST and PFGE curves indicates that the ST curve saturates more quickly. The Chao1 diversity index for STs is 82 ± 27.

Table 3.

Sequence types and allelic profiles, predicted serotype, and subspecies of the Salmonella isolates obtained in this study and analyzed according to the S. enterica database^a

ST	Allelic profile							No. of isolates	Serovar	Subspeciesb
	aroC	dnaN	hemD	hisD	purE	sucA	thrA
4	43	41	16	13	34	13	4	1	Montevideo	I
11	5	2	3	7	6	6	11	10	Enteritidis	I
13	3	3	7	4	3	3	7	3	Agona	I
15	2	7	9	9	5	9	12	12	Heidelberg	I
18	9	9	6	12	9	12	2	2	Manhattan	I
19	10	7	12	9	5	9	2	27	Typhimurium	I
24	13	12	17	16	13	16	4	1	Javiana	I
32	17	18	22	17	5	21	19	1	Infantis	I
33	2	5	6	7	5	7	12	3	Hadar	I
52	23	9	15	12	17	20	12	1	Blockley	I
63	33	26	30	26	21	32	28	3	NK	IIIb
81	43	41	16	42	35	13	4	1	Montevideo	I
126	11	10	25	13	10	35	4	3	NK	I
307	2	14	24	14	2	19	107	1	Paratyphi B var. Java	I
329	82	38	26	12	115	78	70	1	Ohio	I
404	46	122	3	18	6	138	133	1	Paratyphi B var. Java	I
413	15	70	93	78	113	6	68	5	Mbandaka	I
592	189	70	68	132	175	9	172	2	Worthington	I
645	145	26	30	26	21	87	28	1	NK	IIIb
654	111	47	49	42	12	58	3	2	NK	I
816	234	198	96	267	231	208	142	1	NK	I
817	92	107	93	156	64	151	87	1	NK	I
818	19	20	3	20	5	87	22	1	NK	I
819	33	26	30	266	230	27	28	1	NK	IIIb
868	11	10	25	13	10	35	211	1	NK	I
869	235	25	28	24	237	26	78	2	NK	IIIa
870	26	24	28	24	20	25	27	1	NK	IIIa
871	145	26	30	268	141	87	28	2	NK	IIIb
872	189	70	152	132	34	9	172	1	NK	I
873	11	10	13	264	10	13	4	1	NK	I
874	56	143	28	263	48	50	78	1	NK	IIIa
875	33	26	30	180	141	87	134	1	NK	IIIb
876	33	26	159	144	21	87	28	2	NK	IIIb
898	56	24	29	83	20	50	27	1	NK	IIIa
899	157	142	33	16	40	35	3	1	NK	I
909	81	283	101	12	124	130	17	1	NK	I
910	145	26	30	26	21	288	28	3	NK	IIIb
1480	92	187	8	262	95	6	210	1	NK	I

^aNew STs and alleles found in this study are indicated in boldface. NK, unknown. The database is located at http://mlst.ucc.ie/mlst/dbs/Senterica.

^bI, IIIa, and IIIb indicate S. enterica subspecies I (enterica), S. enterica subspecies IIIa (arizonae), and S. enterica subspecies IIIb (diarizonae), respectively.

A number of STs were observed more than once. ST19 (predicted to be S. Typhimurium according to the MLST database; hereafter, all serovar names should be interpreted as “predicted” since they were inferred from the S. enterica MLST database) was most frequently identified (27 of 104 isolates, 28%). Other STs were also represented by more than one isolate: ST11 (Enteritidis, n = 10, 10%), ST15 (Heidelberg, n = 12, 13%), ST413 (Mbandaka, n = 5, 5%), ST13 (Agona), ST33 (Hadar), ST63 (unknown), ST126 (unknown), and ST910 (unknown) (n = 3, 3%), ST18 (Manhattan), ST592 (Worthington), ST654 (unknown), ST869 (unknown), ST871 (unknown), and ST876 (unknown) (n = 2, 2%). Singleton STs associated with a known serovar included Paratyphi B var. Java, Ohio, Blockley, Infantis, Javiana, and Montevideo.

eBURST analysis did not reveal any clonal complexes among the isolates analyzed. However, eBURST analysis did demonstrate distinguishable genetic relationships among some STs, resulting in the formation of four groups. Group 1 was formed by ST645, ST910, ST63, and ST871 (all unknown serovars; ST645 and ST910 were SLVs). Group 2 was formed by ST592 and ST872 (DLVs; both were unknown serovars). Group 3 was formed by ST4 and ST81 (DLVs; both predicted to be Montevideo). The last group was formed by ST126 and ST868 (SLVs; both unknown serovars). The remaining STs were determined to be singletons in the eBURST analysis (Fig. 3).

Fig 3.

S. enterica “population snapshot” obtained using eBURST v3. No clonal complexes were identified. STs that are SLVs of each other are shown connected by lines (black). STs that are DLVs of each other are shown connected by dotted lines. The circle diameters are relative to the number of strains having the same ST.

Isolates characterized as the same ST did not necessarily have the same PFGE pattern. For example, the 27 isolates characterized as ST19 had 13 distinct PFGE patterns (Fig. 1). All of the isolates that shared a PFGE pattern had the same ST.

Clustering and phylogenetic analysis.

An ME tree was generated from the concatenated sequences of the seven loci of the 38 STs (Fig. 4). The ME tree uncovered phylogenetic relationships that were not resolved by eBURST. Three main clusters could be easily visualized. Each cluster was composed of isolates predicted (from the MLST database) to belong to one of three subspecies: S. enterica subspecies I (enterica) cluster I, S. enterica subspecies IIIa (arizonae) cluster II, or S. enterica subspecies IIIb (diarizonae) cluster III. Interestingly, 86 isolates (83%) belonged to subspecies I. Overall, the relationship observed by eBURST was retained in the ME tree from the MLST data. Cluster I is the most diverse. STs 126 and 868 (SLVs) were grouped together, as seen by eBURST, and also showed a phylogenetic relationship to ST873 (not observed by eBURST). STs 4 and 81 (DLVs) and STs 872 and 592 (DLVs) also clustered together, respectively. Cluster II was composed of STs 645 and 910 (SLVs) grouped together, as well as with the two other STs that showed relatedness (ST63 and ST871, both DLVs of ST645) as determined by eBURST analysis. However, additional STs also grouped into this cluster (cluster II) did not show relatedness by eBURST (e.g., STs 875, 876, and 819). Cluster III was composed of only 4 STs, none of them linked by eBURST.

Fig 4.

ME tree constructed from the concatenated sequences of the seven loci for each of the 40 STs observed in this study. The scale represents the evolutionary distance and bootstrap values >50% are shown in the branches. Evolutionary distances were computed using the Kimura two-parameter method and are expressed in numbers of base substitutions per site. Each branch of the tree represents a different cluster or subspecies (clusters I [subsp. I], II [subsp. IIIa], and III [subsp. IIIb]), which are subsequently used in the biogeographic analysis. The four groups observed by eBURST analysis are indicated by boldface numbers 1 to 4.

Comparison to clinical isolates.

Of the 25 isolate PFGE patterns queried against PulseNet, 19 (76%) were found to be indistinguishable from clinical entries (Table 4). Of the 19 patterns found in PulseNet, 8 were associated with one or more PulseNet clusters. The number of clinical entries in PulseNet for each matching pattern varied from 1 to >500; most matching clinical isolates were obtained from California. The STs listed in PulseNet matched those predicted from ST (compare Tables 3 and 4).

Table 4.

Isolates for which XbaI patterns were queried against PulseNet entries uploaded between 1 July 2010 and 28 February 2013 and associated data

Isolate	PFGE pattern XbaI/BlnI	ST	XbaI pattern observed in PulseNet	Serotype predicted by PFGE	PFGE pattern associated with at least one PulseNet cluster	Total no. of PulseNet entries	No. of PulseNet entries in CA
101309.24	1	4	Yes	Montevideo	No	6	0
10808.16	4	11	Yes	Enteritidis	No	>500	193
101309.66	5	11	No
80608.04	7	11	Yes	Enteritidis	Yes	388	1
22008.31	8	11	Yes	Enteritidis	Yes	>500	162
61208.04	10	13	Yes	Agona	Yes	45	1
80608.14	11	13	Yes	Agona	No	2	0
101309.19	60	15	Yes	Heidelberg	Yes	>500	93
50409.05	61	15	Yes	Heidelberg	No	12	7
11309.04	62	15	No
32309.38	63	15	Yes	Heidelberg	Yes	241	46
22209.10	68	15	Yes	Heidelberg	No	71	32
50409.02	45	19	No
32309.30	46	19	Yes	Typhimurium	No	22	3
91608.26	48	19	Yes	Typhimurium	No	24	2
91608.13	49	19	Yes	Typhimurium	No	14	1
91608.08	51	19	Yes	Typhimurium	No	5	1
101309.54	53	19	Yes	Typhimurium	No	49	6
32309.27	55	19	Yes	Typhimurium	Yes	347	22
22008.18	56	19	No
91608.22	57	19	Yes	Typhimurium	No	1	0
101309.67	59	19	Yes	Typhimurium	Yes	165	34
71409.13	17	24	No
101309.52	36	32	Yes	Infantis	Yes	19	3
101309.25	2	81	No

Biogeographic analysis.

The PFGE data were too diverse (most patterns were singletons) and thus dissimilar between sampling events, so they could not be used for biogeographic analysis. When the Bray-Curtis similarity matrices were computed for PFGE patterns, the vast majority were zero (data not shown). Thus, the MLST data were subjected to biogeographic analysis.

The distribution of STs among sampling sites is shown in Fig. 5. Many STs appeared to be endemic. That is, they were detected only once or twice at one or two sites. Other STs appeared more cosmopolitan; for example, ST19 and ST15 were detected at seven and eight different sites, respectively, and multiple times at the same sites. San Pedro Creek appeared to be the most diverse site, harboring 18 different STs. However, the most isolates in our collection were also obtained at San Pedro Creek.

Fig 5.

Number of times each ST was isolated from each site over the course of the study. The sites are indicated as follows: Lagunitas (L), Moss Landing (ML), Napa River (N), Salinas River (NSA), Pescadero Creek (Pes), Petaluma River (Pet), Pajaro River (PR), San Francisquito Creek (SF), San Lorenzo River (SL), Soquel Creek (SO), San Pedro Creek (SP), and Wadell Creek (W).

We determined the extent to which ST diversity during each sampling event (defined as a location and time) could be predicted from watershed data (primary land cover) or environmental conditions (rainfall, salinity, temperature, and dissolved inorganic nitrogen) using multivariate statistics. In addition, we investigated the existence of taxon-time and taxon-area relationships. There were no statistically significant correlations between the similarities of STs or subspecies isolated during each sampling event and geographic distance or environmental distance between the sampling events (P > 0.05). However, there was a significant negative correlation between subspecies and ST similarity and temporal distance (subspecies: r = −0.145, P = 0.007; ST: r = −0.119, P = 0.005]); i.e., the closer the sampling events took place in time, the more similar the Salmonella isolates.

S. enterica similarity between sampling events (at the ST or subspecies level) could not be explained by major land cover classification (urban, agricultural, and forested) for the sites based on ANOSIM. Whether the sampling event occurred during wet or dry weather explained a small amount of the similarity of the ST distributions between sampling events at the alpha = 0.1 level of significance (global R = 0.04, P = 0.07). There was a larger number of Salmonella isolates during wet weather than during dry weather (67 versus 37 isolates, respectively), so richness was higher in wet weather than in dry weather (29 versus 17 STs) (Fig. 6). Twenty-one of the STs were isolated only during wet weather, while 9 STs were isolated only during dry weather. The remaining 8 STs were isolated during both dry and wet weather.

Fig 6.

Number of times each ST was isolated over the course of the study in wet versus dry conditions.

DISCUSSION

A number of genotypically unique S. enterica isolates were detected in this survey of water bodies along the central California coast. Based on ST classifications, S. enterica Typhimurium was the most commonly detected serovar, followed by Heidelberg and then Enteritidis. S. Typhimurium was also the most commonly detected serovar in French environmental waters (18). Most of the STs belonged to subspecies I (based on data in the MLST database), the Salmonella group most commonly associated with human disease. Between 1998 and 2010, there were 273 outbreaks of salmonellosis in California; S. enterica serovars Typhimurium, Enteritidis, and Heidelberg caused the most (24, 75, and 24 outbreaks, respectively) (41). Thus, our most common environmental isolates are also the most common Salmonella pathogens in California, based on reported illnesses. In addition, the PFGE patterns of ca. 75% of the queried isolates were indistinguishable from clinical entries in PulseNet; some were associated with PulseNet clusters. This suggests that S. enterica isolates from the present study have the potential to be important clinically.

The number of STs detected given the total number of isolates is comparable to the number of serovars observed in other environmental studies (6, 10, 14–18). The ST and PFGE pattern accumulation curves suggest that we sampled approximately half the genotypes in the pool, comparable to other studies of microbial diversity (see, for example, references 48, 54, and 55). Future work that extensively samples one of the sites could potentially uncover the full extent of diversity and add to our understanding of the factors that affect the distribution of S. enterica in this region. In addition, next-generation full-genome sequencing may provide more information on the diversity of the isolates already collected, their similarity to outbreak-causing Salmonella isolates already sequenced, and their resistance to antibiotics, which is becoming an increasing important aspect of salmonellosis (42).

The observed STs were predicted to belong to one of three subspecies: S. enterica subspecies I (enterica), S. enterica subspecies IIIa (arizonae), or S. enterica subspecies IIIb (diarizonae). No isolates associated with the other three subspecies (salamae, indica, and houtenae) were detected. This may be attributable to the fact that other subspecies were rare and missed by our sampling efforts. Alternatively, this may be because other subspecies live in niches different from those sampled here. For example, S. enterica subsp. salamae serovar Sofia is infrequently detected and has only been found in commercially produced broiler chickens in eastern Australia (43). It is also possible that other STs were viable but not cultivatable and thus were not captured by our isolation procedures. The majority of the isolates (83%) were predicted to belong to subspecies I, indicating the high prevalence of potentially harmful Salmonella spp. in these water sources. Most food-borne cases belong to S. enterica subspecies I (44). Only a few isolates (17%) predicted to belong to S. enterica subspecies IIIa and IIIb were observed. However, most (82%) of these isolates are new STs that have not been previously reported. In total, 18 new STs were identified here.

PFGE was more discriminatory than MLST. For example, 27 strains belonging to ST19 (S. Typhimurium) were further discriminated into 13 distinct groups by their PFGE patterns. PFGE has been described as more discriminatory than MLST for other bacteria, such as Vibrio vulnificus (45) and S. Newport (46). However, MLST allowed for typing of isolates that were untypeable by PFGE. MLST also allowed for determining the phylogenetic relationships among the isolates. Thus, the two analyses are complementary. Investigators of previous studies have used both PFGE and MLST to explore Salmonella diversity and concluded that although PFGE has high discriminatory capabilities, it has not demonstrated the ability to establish relationships between strain origin and genotype in the manner that MLST has (20, 46, 47). Owing to its high discriminatory power, biogeographical analysis was not meaningful for the PFGE data since PFGE genotypes were mostly singletons. However, biogeographic analysis was feasible with the MLST data.

The biogeographic analysis indicates that sampling events that were geographically close or similar in environmental conditions based on salinity, temperature, dissolved inorganic nitrogen, and/or rainfall did not have similar STs present. This held true even when STs were classified as their associated subspecies. The lack of association with rainfall, salinity, nitrate, and temperature may indicate that these factors are not important in selecting for S. enterica diversity at the spatial and temporal scales of this study. This is consistent with the notion that these waters do not represent niches where Salmonella can be metabolically active; however, Salmonella survival in the aquatic environment has been described as prolonged (9). Lack of dispersal limitation may be the main factor driving diversity at these sites. This would be consistent with the lack of a taxon-distance relationship (48, 49). Although primary land cover, which may serve as a proxy for watershed Salmonella sources responsible for dispersal, was not correlated to Salmonella diversity, wet versus dry weather was, albeit weakly. During wet weather, more watershed sources are active because stormwater runs over impervious surfaces and soils and enters bodies of water. Alternative explanations for the lack of observed biogeographic relationships include the following: (i) environmental factors placing selective pressure on S. enterica were not recorded during the study, (ii) S. enterica diversity was undersampled, and (iii) genetic relatedness needs to be explored using a finer or coarser scale (49).

A relationship between S. enterica species genetic similarity and time was found, with sampling events close in time tending to be more similar. This suggests that ST turnover rates are longer than the time between sampling events. It may be a result of wet and dry sampling events, which were shown to harbor similar S. enterica genotypes, being clustered in time.

Primary watershed land cover did not explain the diversity of STs observed during the sampling events. This indicates that specific STs do not emanate from sources that are strictly associated with specific land covers or that the differential fate of STs as they are transported from the watershed source to the sampling location interferes with the identification of land cover-ST relationships. For example, photoinactivation experiments carried out in freshwater with S. Mbandaka, S. Heidelberg, and S. Typhimurium indicate that S. Mbandaka is more resistant to sunlight than the other two serovars (50). Alternatively, the primary watershed sources of Salmonella may be wildlife that is not restricted to particular land covers.

The factors modulating the presence, distribution, and diversity of waterborne Salmonella are complex and poorly understood. This investigation into the distribution and diversity of waterborne Salmonella adds to our understanding of these factors. In particular, it appears that the lack of dispersal limitation may be important in controlling Salmonella diversity at the temporal and spatial scale of the study. Additional studies with deeper sampling at greater spatial and temporal scales may be warranted to fully investigate Salmonella diversity and the factors driving it.