Abstract
Every year salmonellosis is responsible for $2.3 billion in costs to the U.S. food industry, with nearly 6% of the reported cases associated with pork and/or pork products. Several studies have demonstrated the role of pigs as Salmonella reservoirs. Furthermore, this pathogen has been identified as a potential biological hazard in many livestock feeds. The overall objective of this research was to characterize Salmonella enterica isolates in selected U.S. swine feed mills by whole-genome sequencing (WGS) and evaluate isolates in association with the season and feed production stages. Salmonella isolates were collected from 11 facilities during a previous study. Samples were analyzed for Salmonella prevalence following the U.S. Department of Agriculture guidelines and confirmed by PCR. WGS was carried out on either the MiSeq or NextSeq sequencer. De novo genome assemblies were obtained with the Shovill pipeline, version 0.9. ResFinder and SPIFinder were used to identify antibiotic resistance genes and pathogenicity islands. Finally, their phylogenetic relationship and diversity were determined by core genome multilocus sequence typing. Overall, our analysis showed the presence of S. enterica in the feed mill environment. Isolates belonged to 16 different serotypes. Salmonella Agona, Salmonella Mbandaka, Salmonella Senfenberg, and Salmonella Scharzengrund were the most frequently found, and 18 single-nucleotide polymorphism clusters were identified. In silico analysis showed that 40% of the strains carried at least one antimicrobial resistance gene. All isolates in this study could be considered of public health concern and pathogenic potential. Our findings underscore the potential role of the feed mill environment as the pathogen entry route into the human food value chain.
Introduction
Annual estimates list Salmonella as responsible for ∼11% of total foodborne illnesses, 35% of total hospitalizations, and 28% of total deaths (Scallan et al., 2011). In addition to the public health burden, this zoonotic pathogen can also negatively affect swine production, resulting in huge economic losses (Argüello et al., 2018).
Salmonella outbreaks have been associated with animal feed (Österberg et al., 2006; Molla et al., 2010) and pork products (Moreno Switt et al., 2009; Gossner et al., 2012; Andrés-Barranco et al., 2014), raising a flag on an underestimated entry route of the pathogen into the feed-to-fork model. Contaminated feed can lead to swine being infected, and studies show that nearly 6% of reported human salmonellosis cases are associated with the consumption of contaminated pork and/or pork products (Dickson and Hurd, 2013). A recent study conducted in Europe highlighted the risk of salmonellosis linked to consumption of pork products derived from animals fed with contaminated feed (Rönnqvist et al., 2018). The Food and Drug Administration (FDA), under the Food, Drug, and Cosmetic Act section 402(a)(1), deems animal feed adulterated when contaminated with a Salmonella serotype that is considered pathogenic to the animal intended to consume the animal feed and the animal feed will not subsequently undergo a commercial heat step or other commercial process that will kill the Salmonella. The Salmonella serotype considered as an adulterant in swine feed is Salmonella Choleraesuis (Food and Drug Administration, 2013). In 2015, the same agency released a rule regarding the current good manufacturing practices, hazard analysis, and risk-based preventive controls for food for animals as a result of the changes brought into action after Food Safety Modernization Act implementation. This new regulation makes facilities responsible for manufacturing, processing, packing, and holding of food for animals and promotes hazard analysis and risk-based preventive controls (Federal Register, 2015). Recent, multistate foodborne outbreaks have highlighted the need of implementing controls and analysis in accordance with the new regulatory framework, and the importance of using rapid methods to trace contamination in the food supply chain, compared with traditionally and time-consuming routine tests (Zhang et al., 2015). Genomic techniques allow for rapid identification of pathogens and have led to the development of databases (e.g., GenomeTrakr) with epidemiological information able to assist microbial monitoring and surveillance across the food industry.
In our previous studies, we investigated the presence of Salmonella in selected U.S. swine feed mills (Magossi et al., 2019a, b). The collected data suggested the role of these environments as potential Salmonella entry routes into the food chain. Therefore, the objectives of this study were to (1) characterize previously collected Salmonella enterica isolates by whole-genome sequencing (WGS) and (2) use WGS data to correlate isolates with season, geographical location, and feed processing stages and evaluate the prevalence of antimicrobial resistance (AMR) genes.
Materials and Methods
Isolates and sample collection
The 57 Salmonella strains used in this research were isolated in a previous study (Magossi et al., 2019a). Each isolate was assigned a U.S. FDA Center for Food Safety and Applied Nutrition (CFSAN) number ID as part of the GenomeTrakr network. The strains were isolated in 2016 and 2017 from 11 feed mills distributed among eight states representative of the main swine production areas within the United States (Magossi et al., 2019b). The strains were stored on tryptone soy broth containing glycerol at −80°C until further use.
DNA preparation
Genomic DNA from each strain was obtained using the DNeasy blood and tissue kit (Qiagen, Hilden, Germany), following the manufacturer's instructions. DNA concentration was determined using a Qubit 4.0 fluorometer (Invitrogen). The resultant DNA extract was stored at −20°C until WGS analysis.
WGS, contig assembly, and annotation
Sequencing was carried out as previously described (Lomonaco et al., 2018). Paired-end DNA libraries were prepared with the Nextera XT DNA library preparation kit, and WGS was carried out on either the MiSeq or NextSeq sequencer, using a 500-cycle MiSeq reagent V2 kit or a 300-cycle NextSeq 500/550 high-output V2 kit, respectively (Illumina, San Diego, CA). De novo assemblies were obtained with Shovill, version 0.9 (
In silico molecular analysis (serotyping, multilocus sequence typing, core genome multilocus sequence typing, AMR genes, and pathogenicity islands)
The serotype of each isolate was determined in silico by prediction from draft genomes using SeqSero 1.0 (
NCBI pathogen detection, single-nucleotide polymorphism cluster, and AMR
To increase the size of our analysis, we included comparisons of the 57 isolates sequenced herein with more than 190,000 other Salmonella isolates obtained worldwide, using isolates available in the NCBI Pathogen Detection database (
Nucleotide sequence accession numbers
The draft genome sequences of all 57 S. enterica strains used in this research are available in GenBank under accession numbers QUTN00000000 to QUVQ00000000 and QUZX00000000 (Lomonaco et al., 2018).
Results and Discussion
In our previous research (Magossi et al., 2019a, b), a total of 383 environmental and feed samples were collected from 11 U.S. swine feed mills over a year. From the total isolates, 14.8% (n = 57) were confirmed as Salmonella and characterized in the present study (Table 1). The 57 isolates originated mostly from Kansas (n = 22) and North Carolina (n = 22) from 2 different facilities, followed by Iowa (n = 5), Indiana (n = 5), and Oklahoma (n = 3) (Magossi et al., 2019a). Sixteen Salmonella serotypes were identified among the 57 isolates, 8 of which were detected in multiple isolates: Salmonella Agona (n = 14, 24.5%), Salmonella Mbandaka (n = 13, 22.8%), Salmonella Senftenberg (n = 7, 13.2%), Salmonella Schwarzengrund (n = 5, 13.2%), Salmonella Rissen (n = 3, 5%), Salmonella Hartford (n = 3, 3.5%), Salmonella Kiambu (n = 2, 3.5%), and Salmonella Typhimurium (n = 2, 3.5%). Influence of sampling site, geographical location, and season was as follows. The prevalence of specific serotypes seems to be influenced by the geographical location. Studies conducted in Spanish and Brazilian feed mills reported the detection of specific serotypes mainly in Spain (Salmonella Mbandaka and Salmonella Typhimurium) and Brazil (Salmonella Senftenberg and Salmonella Worthington) (Torres et al., 2011; Pellegrini et al., 2015). Diversity in serotypes was also observed in feedstuff from Costa Rica (Molina et al., 2016). In our study, we also observed that the distribution of Salmonella serotypes varied among seasons and locations. As previously reported (Magossi et al., 2019a), Salmonella isolates had a significantly higher prevalence during fall months (46%) compared with early spring (14%) or summer (38%) (Fig. 1). This is in line with the observed seasonality variation of Salmonella, with higher number of cases during warmer months compared with colder months (Ravel et al., 2010; Jahne et al., 2015). Herein, of the top three most frequently detected serotypes, two (i.e., Salmonella Agona and Salmonella Mbandaka) were identified during all three sampling seasons exclusively in Kansas and in North Carolina, Kansas, and Oklahoma (Fig. 1). Salmonella Senftenberg was instead observed in all states except Oklahoma (Fig. 1). Overall, fall isolates showed the greatest diversity as they belonged to 11 serotypes, followed by summer with 8 and spring with 3. Only two serotypes (Salmonella Agona and Salmonella Mbandaka) were noticed across all three seasons.
Distribution of the 16 serotypes detected overall across the 57 Salmonella enterica isolates according to season (fall, spring, and summer) and geographical location where the mills are located (Indiana, Iowa, Kansas, North Carolina, and Oklahoma). The bars indicate the different serotype detected (by color) and the number of isolates for that serotype. Color images are available online.
Summary of the Metadata, MLST, and NCBI Pathogen Detection results for the Fifty-Seven Salmonella enterica Strains Isolated from Different Feed Mills in the United States of America Between 2016 and 2017
Isolates from the environment include soil, sediment, water, and animals.
For updated information, refer to
SNP, single-nucleotide polymorphism; NCBI, National Center for Biotechnology Information; MLST, multilocus sequence typing; AMR, antimicrobial resistance.
When considering the states, we observed high diversity in Salmonella serotypes in North Carolina (10 serotypes), which might be related to the diverse feed production since these facilities did not exclusively produce swine feed. A higher prevalence was observed in sites corresponding to the receiving of raw ingredients area floor (n = 16, 28.1%), worker shoes (n = 9, 15.7%), control room floor (n = 8, 14.0%), receiving ingredients pit gratin (n = 7, 12.3%), and manufacturing area floor (n = 7, 12.3%) (Fig. 2A). In particular, worker shoes can carry and spread the biological hazard and were recognized as a potential pathogen reservoir in several studies (Amass et al., 2000; Otake et al., 2002; Magossi et al., 2019a, b). When serotypes were stratified by sampling locations (Fig. 2B), Salmonella Agona was most frequently isolated from worker shoes (28.5%) and the manufacturing and control room floors (21.4%), while Salmonella Mbandaka (30.7%) and Salmonella Senftenberg (57.1%) were mostly found in the receiving area floor. A higher frequency of Salmonella Schwarzengrund (33.3%) was observed instead in the manufacturing areas. Dust and floors of feed mill facilities have been identified as pathogen reservoirs also in other studies (Torres et al., 2011; Pellegrini et al., 2015).
Distribution of the detected serotypes stratified for sampling site
In silico molecular analysis
Overall, detected serotypes corresponded with clusters defined by MLST, as also previously observed (Achtman et al., 2012). All isolates belonging to the same serotype shared the same sequence type (ST), as registered in the MLST database (ST) (Table 1), and presented at least one pathogenicity island (SPI 1–4). cgMLST also divided all isolates according to serotypes, which were clearly divided into 16 clonal complexes. Based on the NCBI Pathogen Detection database results, 54 strains were grouped into 18 SNP clusters (comprising more than 10,000 strains overall), while 3 were not assigned to any. Overall, the SNP cluster largely coincided with the serotype/cgMLST/MLST results. The 18 SNP clusters grouped our strains with a wide range of other isolates from the database, ranging from 2 to 6995. Strains were matched with other environmental isolates from the database and while no matches with clinical samples were found, some SNP clusters included clinical isolates related to the strains analyzed herein (Table 1). cgMLST and SNP clustering analysis agreed for all isolates. All 14 Salmonella Agona strains were grouped in the SNP cluster grouping the largest number of strains analyzed herein (i.e., PDS000036374.1). The only other two strains present in this cluster were isolated in Mexico in 2007 (herbal tea ajenjo) and 2003 (Alfalfa Herbal Tea) (SAMN02845921 and SAMN01813429, respectively). These strains (CFSAN071946, CFSAN071947, and CFSAN071951) originated in the fall from three different sampling locations in the same mill: finished feed, control room, and warehouse floor. The three Salmonella Rissen grouped in PDS000031137.14 were separated into two clades (CFSAN071971/CFSAN071970 and CFSAN071961), as also shown with cgMLST. In particular, CFSAN071961 is close to two Salmonella Rissen strains from North Carolina, derived in 2017, collected from the animal swine market and raw product (Pornsukarom et al., 2018). All three correlate with four U.S. clinical cases (SAMN08823502, SAMN08832556, SAMN09501102, and SAMN09534158). The closest neighbor of the two Salmonella Typhimurium isolates in PDS000026658.14 (overall grouping 46 isolates) is a clinical case from the United States of America (SAMN09703939), while the closest neighbor of the monophasic isolates in PDS000042752.37 (overall grouping 6995 isolates) is a monophasic S. enterica from California collected in 2016 from raw ground pork products (SAMN05968677). As previously observed with cgMLST, the two isolates each belonging to Salmonella Hartford (PDS000029644.4) and Salmonella Kiambu (PDS000029314.27) were highly correlated. The eight serotypes detected in only one isolate were either not assigned to any SNP cluster (Salmonella Bareilly and Salmonella Cubana) or were the only strain from this study in the SNP cluster (e.g., Salmonella I 4,[5], 12:i-, Salmonella Barranquilla, Salmonella Braenderup, Salmonella Javiana, Salmonella Soerenga, and Salmonella Worthington) (Table 1). Among strains belonging to the same serotype, Salmonella Mbandaka and Salmonella Senftenberg isolates were divided into three different SNP clusters and three SNP clusters plus an unassigned isolate (Salmonella Mbandaka). In particular, the three SNP clusters for Salmonella Senftenberg were represented by PDS000031803.29 (n = 5, CFSAN071937, 071938, 071969, 071993, and 071994); PDS000003890.32 (n = 1, CFSAN071972); and PDS000031814.12 (n = 1, CFSAN071954). The first cluster grouped isolates collected from the receiving of raw ingredients area floor or the receiving ingredients pit gratin, while the second and third clusters grouped isolates from the finished product discharge bin boot (Table 1). Regarding Salmonella Mbandaka, three SNP clusters were observed: (1) PDS000026971.3 (n = 6, CFSAN071945, 071953, 071956, 071957, 071958, and 071959); (2) PDS000026950.68 (n = 5, CFSAN071932, 071933, 071934, 071936, and 071960); and (3) PDS000028188.3 (n = 1, CFSAN071948) and one unassigned (CFSAN071943), in addition to the two strains mentioned previously. In PDS000026950.68, CFSAN071932-071933-071934-071936 and CFSAN071960 are separated into two different clades, as also observed in cgMLST. It must be noted that the cgMLST was carried out only on the 57 strains analyzed herein and thus the gap observed might be explained by the absence of connecting strains, underlining the importance of large public databases such as GenomeTrakr. Finally, while Salmonella Senftenberg isolates from the same mill clustered together, this was not observed for Salmonella Mbandaka, where isolates from the same facility were divided across two different SNP clusters.
Our in silico ResFinder analysis showed that 40% of the strains carried at least one AMR gene; most strains were not found to harbor AMR genes (Table 1). A resistance determinant for tetracyclines was detected in all strains from five serotypes: tetB in Salmonella Agona, Salmonella Mbandaka, and Salmonella Schwarzengrund; tetA in Salmonella Rissen; and both tetB and tetM in S. I 4,[5], 12:i-. In Salmonella Mbandaka and Salmonella Agona, not all strains had the same AMR profile, which differed based on the observed clustering (Table 1). CFSAN071935 (S. I 4,[5], 12:i-) had the highest number of resistance determinants (n = 12), while strains belonging to Salmonella Rissen had one resistance determinant. The serotype I,[5], 12:i- is of particular concern because of its known resistance to many common antimicrobials (Moreno Switt et al., 2009). This serotype has been linked to swine feed in Poland and Germany (Hauser et al., 2010; Wasyl and Hoszowski, 2012). Swine is the reservoir most frequently associated, nevertheless it has been rarely found in pig samples collected in slaughterhouses during the U.S. National Antimicrobial Resistance Monitoring System program (Elnekave et al., 2018). The pathogenicity, on-farm distribution, spread, and prevalence in the United States of S. I 4,[5], 12:i- are currently unknown and/or unquantified. Nonetheless, its known association with pork, combined with resistance to antimicrobials, underscores the importance to conduct research to prevent its presence in postharvest products (Moreno Switt et al., 2009).
Conclusion
While the risk of salmonellosis from feed remains difficult to quantify, our data highlight the possible role of the feed mill environment as the Salmonella entry route into the human food chain. Overall, the serotypes detected in our study are not those of concern for swine feed (Salmonella Choleraesuis), nevertheless they have shown pathogenic potential for both humans and animals and might be considered of public health concern (Pedersen et al., 2008; Li et al., 2012; Hoffman et al., 2015). Our data highlight the importance for the feed industry to establish strategies to ensure the safety of animal feed and the need to create a database of the most common Salmonella serotypes associated with these types of products. The presence of biological hazards in contaminated feed, feed environment, animals, and pork products in the same SNP clusters indicates a potential public health risk (Food and Drug Administration, 2016). Our analysis showed that most of our SNP clusters contain environmental, animal (swine, cattle, and chicken), and clinical isolates. The phylogenic tree from the Salmonella Typhimurium SNP cluster (PDS000026658.25) is reported in Figure 3 as an example. Most of the isolates in this cluster belong to clinical samples, but the few obtained from the environment and animals were mainly isolated from pork, hogs, or swine lymph nodes, highlighting the possibility of porcine lineages. Additionally, if Salmonella is found in the feed environment, it may circulate into other vectors that have been shown to be genetically diverse reservoirs (Guard et al., 2019). Additionally, it has been shown that farm-specific Salmonella strains can persist in the farm environment for multiple years and that transmission among different species (i.e., chickens, ducks, and pigs) is plausible (Basler et al., 2016; Alegria-Moran et al., 2017; Toro et al., 2018). Thus, we cannot exclude the potential role of the feed mill environment as the pathogen entry route into the human food value chain. Nevertheless, quantification of pathogen survival from feed to fork will help determine the real risk of contaminated feed for animals and human health.
Example of a phylogenic tree from the Salmonella Typhimurium single-nucleotide polymorphism cluster (PDS000026658.25). Color images are available online.
Footnotes
Funding Information
This work was supported by the Center for Food Safety and Applied Nutrition at the U.S. Food and Drug Administration and by the National Pork Board. This project was also supported, in part, by an appointment to the Research Participation Program at the Center for Food Safety and Applied Nutrition, U.S. Food and Drug Administration, administered by the Oak Ridge Institute for Science and Education through an interagency agreement between the U.S. Department of Energy and the Food and Drug Administration.