Identification of single-nucleotide variants associated with susceptibility to Salmonella in pigs using a genome-wide association approach

Abstract

Background

Salmonella enterica serovars are a major cause of foodborne illness and have a substantial impact on global human health. In Canada, Salmonella is commonly found on swine farms and the increasing concern about drug use and antimicrobial resistance associated with Salmonella has promoted research into alternative control methods, including selecting for pig genotypes associated with resistance to Salmonella. The objective of this study was to identify single-nucleotide variants in the pig genome associated with Salmonella susceptibility using a genome-wide association approach. Repeated blood and fecal samples were collected from 809 pigs in 14 groups on farms and tonsils and lymph nodes were collected at slaughter. Sera were analyzed for Salmonella IgG antibodies by ELISA and feces and tissues were cultured for Salmonella. Pig DNA was genotyped using a custom 54 K single-nucleotide variant oligo array and logistic mixed-models used to identify SNVs associated with IgG seropositivity, shedding, and tissue colonization.

Results

Variants in/near PTPRJ (p = 0.0000066), ST6GALNAC3 (p = 0.0000099), and DCDC2C (n = 3, p < 0.0000086) were associated with susceptibility to Salmonella, while variants near AKAP12 (n = 3, p < 0.0000358) and in RALGAPA2 (p = 0.0000760) may be associated with susceptibility.

Conclusions

Further study of the variants and genes identified may improve our understanding of neutrophil recruitment, intracellular killing of bacteria, and/or susceptibility to Salmonella and may help future efforts to reduce Salmonella on-farm through genetic approaches.

Background

Salmonella is one of the leading causes of foodborne illness and has a significant impact on human health both globally and in Canada [1–3]. While eggs and poultry are the most frequently identified sources of human salmonellosis, pork is also a notable source of Salmonella [4–7]. Studies assessing Salmonella prevalence through serology and/or culture have frequently identified Salmonella in pigs in North America and Europe [8–12]. In pigs, S. Choleraesuis infection typically manifests as swine typhoid that may result in diarrhea, fever, and septicemia, similar to human-infecting typhoidal Salmonella serovars like S. Typhi [13]. Pigs showing visible signs of illness may be treated or removed from the herd to reduce the spread of Salmonella. However, the most frequently identified serovars on Canadian swine farms included S. Typhimurium, S. Typhimurium var. Copenhagen, and S. Infantis [9, 14, 15] which typically result in an asymptomatic carrier state in pigs but are known to cause illness in humans [16]. Pigs carrying Salmonella asymptomatically play a significant role in on-farm transmission of Salmonella within the herd and may limit the effectiveness of control measures implemented on-farm [12]. On-farm control of Salmonella has consisted of stringent biosecurity and sanitation practices, as well as the use of antibiotics, vaccination, and quarantine or culling of infected pigs [17–20]. However, the limited effectiveness of these measures in practice has prompted research into swine genetics as a potential alternative measure to control Salmonella on swine farms.

Traditionally, selective breeding in swine was established to promote desired production traits including growth performance, feed efficiency, fertility, and meat quality [21–23]. However, with the completion and continued updates to the porcine genome, many studies are now investigating the genetic basis of disease susceptibility in swine. One approach in using genetics to improve resistance is to observe immune traits or phenotypes individually (for example; cytokine production, leukocyte proliferation, and serum levels of IgG or acute phase proteins) [24–27]. Differences in these immune traits and disease severity between pigs and between breeding lines has been well documented which suggests the potential of selective breeding for improved resistance in the near future [19, 24, 28–30]. One such study found that piglets with improved recruitment and function of polymorphonuclear neutrophils, but a lower antibody response, were more resistant to Salmonella [28]. As such, it may be possible to select from these breeding lines with more robust immune response phenotypes or desired response traits to promote broad immunity to Salmonella in offspring.

Beyond the assessment of immune traits, several studies in recent years have identified significant associations between single-nucleotide variants (SNVs) and/or candidate genes and susceptibility to Salmonella in pigs. Candidate gene studies have observed variants in porcine toll-like receptor (TLR) genes that were associated with Salmonella fecal shedding [31], and attenuated responses to Salmonella Choleraesuis [32]. Upregulation of TLR5 and TLR9 has been shown in response to S. Choleraesuis and S. Typhimurium though its direct impact on Salmonella susceptibility is unknown [33]. Additionally, SNVs in mannan-binding lectin (MBL) 1 and 2, have been found at higher frequencies in pig populations infected with S. Typhimurium and other pathogens [34]. Further, a candidate gene study of the pigs included in this study identified an MBL1 variant associated with increased Salmonella shedding and a variant in NOD1 associated with isolation of Salmonella at slaughter [35].

The candidate gene studies may offer insight into pig susceptibility to Salmonella on-farm and at slaughter and benefit efforts in breeding for resistance to common pathogens on-farm. However, a major drawback of candidate gene studies is that they require a priori knowledge of these genes and their functions, and there is still much that is unknown about the pig immune response and the complex interplay between pathogen and host [36]. With recent technological advancements improving the feasibility of genome-wide association studies (GWAS), we can potentially identify novel variants associated with resistance to Salmonella shedding and colonization across the entire genome [37]. This study aimed to identify SNVs associated with Salmonella IgG antibody response from the end of nursery to market, Salmonella shedding from weaning to market, and presence of Salmonella in tonsil and lymph node tissues at slaughter in commercial pigs using a GWAS approach.

Results

Of the pigs included in the GWAS seropositivity model, 32.3% (254/786) of pigs were seropositive at least once from the end of the nursery stage to the end of the finisher stage, for the shedding model 34.2% (269/786) of pigs shed at least once from weaning to end of finisher, and for colonization model 21.6% (111/515) of pigs tested positive for Salmonella at slaughter. Salmonella positivity for each trait is shown in Table 1.

Table 1.

Demographics in pigs positive and negative for the trait of interest after quality control of genotypic data

Phenotype	Number of negative pigs (controls*)	Number of positive pigs (cases*)
Salmonella seropositivity	254 (32.3%)	532 (67.7%)
Salmonella shedding	269 (34.2%)	517 (65.8%)
Salmonella isolation at slaughter	111 (21.6%)	404 (78.4%)

*Control: never tested seropositive, never shed Salmonella, or tested negative at slaughter; Case: tested seropositive for Salmonella at least once, shed Salmonella at least once on-farm, or was positive for Salmonella at slaughter

Genome-wide association study

After quality control filtering of the data, 51,969 SNVs remained for GWAS analysis from 786 pigs in the seropositivity and shedding models and 515 pigs in the slaughter model. Manhattan plots and quantile-quantile plots for the seropositivity, shedding, and colonization phenotypes are shown in Figs. 1, and 2, respectively. The allele frequencies between case and control pigs for the five significant SNVs and SNVs approaching significance in the GWAS analysis can be seen in the Supplementary Table 1.

Fig. 1.

Manhattan plots of the GWAS analysis for Salmonella seropositivity from end of nursery to end of finisher (a), Salmonella shedding from weaning to end of finisher (b), and Salmonella isolation from tissues at slaughter (c). The horizontal solid and dashed red lines indicate the genome-wide threshold for significant (p = 5.0 × 10^− 7) and suggestive (p = 1.0 × 10^− 5) associations, respectively [38]

Fig. 2.

Quantile-quantile plots for Salmonella seropositivity from end of nursery to end of finisher (a), Salmonella shedding from weaning to end of finisher (b), and Salmonella isolation from tissues at slaughter (c). λ = the average genomic inflation factor

Analysis of the shedding trait identified an intron variant in the gene PTPRJ on chromosome 2 (p = 6.6 × 10^− 6) as well as an intergenic variant on chromosome 6, downstream of the ST6GALNAC3 gene (p = 9.9 × 10^− 6), that had a suggestive association (Table 2) and were observed at a higher frequency in the pigs that did not shed Salmonella. GWAS analysis of Salmonella isolation at slaughter indicated three intergenic SNVs downstream of the gene DCDC2C on chromosome 3 (p = 1.6 × 10^− 6, 3.7 × 10^− 6, 8.6 × 10^− 6, respectively) under the suggested association threshold of 5.0 × 10^− 5 (Table 3), two of which were seen at higher frequency in pigs that tested negative at slaughter, while one was more common in pigs positive at slaughter. Three intergenic SNVs on chromosome 1, upstream of the gene AKAP12 (p = 2.0 × 10^− 5, 3.1 × 10^− 5, 3.6 × 10^− 5, respectively), also approached the suggested association threshold and were more common in the pigs that were negative at slaughter. An intron SNV located in AKAP12 was also observed in the top 15 SNVs for the slaughter model (Table 3). No SNVs were under the suggestive association threshold of 5.0 × 10^− 5 for the analysis of the seropositivity trait. An intron variant located within the RALGAPA2 gene on chromosome 17 was the most significant SNV associated with a pig being seropositive at least once from nursery to finisher stage (p = 7.6 × 10^− 5). Additionally, three intron SNVs in the RALGAPA2 gene were observed within the top 15 significant SNVs for the seropositivity model (Table 4).

Table 2.

The top 15 SNVs ranked by significance (p value) for the GWAS analysis of Salmonella shedding from weaning to end of finisher

SNV ID	SSCa	Location (bp)b	Variant type	Genec	Gene locationb	p-valued
rs324041697	2	14,596,159	intron	PTPRJ	2: 14549537–14,726,715	6.6 × 10−6
rs81476180	6	136,607,113	intergenic	ST6GALNAC3	6: 136633437–137,201,213	9.9 × 10−6
rs81259485	16	3,248,143	intron	DNAH5	16: 3116189–3,364,956	6.6 × 10− 5
rs81458700	16	36,370,940	intergenic	ACTBL2	16: 36470503–36,471,633	9.9 × 10− 5
rs80889714	4	104,920,198	intron	CASQ2	4: 104918501–104,982,852	1.0 × 10− 4
rs81346930	17	44,032,145	intergenic	LPIN3	17: 43993710–44,006,188	1.2 × 10−4
rs330624049	15	90,476,754	intergenic	ENSSSCG00000034554	15: 90309385–90,309,702	1.2 × 10−4
rs81403742	8	117,467,555	intergenic	TACR3	8: 117661792–117,721,591	1.2 × 10−4
rs80947769	7	19,456,263	upstream	KIAA0319	7: 19386111–19,451,666	1.2 × 10−4
rs339478050	1	188,917,080	intron	RTN1	1: 188704139–188,926,054	1.9 × 10−4
rs81237965	3	3,693,136	intron	RADIL	3: 3618093–313,033	1.9 × 10−4
rs81344023	6	153,022,279	intron	FGGY	6: 152754730–153,213,125	2.1 × 10−4
rs81349902	1	188,903,236	intron	RTN1	1: 188704139–188,926,054	2.3 × 10−4
rs318950111	3	43,561,730	downstream	ZNF169	3: 43513397–43,558,613	2.5 × 10−4
rs81452423	15	32,935,177	intergenic	DLGAP2	15: 33173120–33,264,155	3.0 × 10−4

^aSSC = Sus scrofa chromosome

^bLocation in Ensembl Sscrofa11.1

^cIf the variant was intergenic, the closest gene within a 1 Mbp window was indicated

^dSNVs under the suggestive significance threshold of 1.0 × 10⁻⁵ in bold [39]

Table 3.

Top 15 SNVs ranked by significance for the GWAS analysis of isolation of Salmonella from tissues at slaughter

SNV ID	SSCa	Location (bp)b	Variant type	Genec	Gene locationb	p-valued
rs322440805	3	130,637,256	intergenic	DCDC2C	3: 131097730–131,194,726	1.6 × 10−6
rs326411709	3	130,676,894	intergenic	DCDC2C	3: 131097730–131,194,726	3.7 × 10− 6
rs319944764	3	130,689,632	intergenic	DCDC2C	3: 131097730–131,194,726	8.6 × 10−6
rs81348815	1	15,081,855	intergenic	AKAP12	1: 14906823–15,016,841	2.0 × 10−5
rs80951933	1	15,109,785	intergenic	AKAP12	1: 14906823–15,016,841	3.1 × 10−5
rs80903645	1	15,064,589	intergenic	AKAP12	1: 14906823–15,016,841	3.6 × 10−5
rs326617356	18	26,691,633	intergenic	KCND2	18: 26554190–26,623,034	1.3 × 10− 4
rs323563819	19	109,560,455	intron	GPC3	X: 109536447–110,060,245	2.0 × 10−4
rs80820138	14	122,383,711	intergenic	GPAM	14: 122542205–122,577,239	2.0 × 10−4
rs80961723	7	120,400,325	intergenic	SETD3	7: 120412549–120,486,627	2.2 × 10−4
rs320610499	6	141,349,179	intron	NEGR1	6: 141115729–141,647,820	2.5 × 10−4
rs81296290	2	91,504,717	intron	VCAN	2: 91287931–91,811,674	2.8 × 10−4
rs324243793	18	38,747,572	intron	DPY19L2	18: 38693003–38,764,024	3.3 × 10−4
rs80840697	1	14,988,130	intron	AKAP12	1: 14906823–15,016,841	4.1 × 10−4
rs81361262	2	90,665,766	downstream	ATG10	2: 90437020–90,665,380	4.5 × 10−4

^aSSC = Sus scrofa chromosome

^bLocation in Ensembl Sscrofa11.1

^cIf the variant was intergenic, the closest gene within a 1 Mbp window was indicated

^dSNVs under the suggestive significance threshold of 1.0 × 10⁻⁵ in bold [39]

Table 4.

Top 15 SNVs ranked by significance for the GWAS analysis of Salmonella seropositivity from end of nursery to end of finisher

SNV ID	SSCa	Location (bp)b	Variant type	Genec	Gene locationb	p-valued
rs81241392	17	28,616,449	intron	RALGAPA2	17: 28394827–28,679,538	7.6 × 10− 5
rs323410857	1	201,886,813	intron	HACD4	1: 201886332–201,910,664	8.2 × 10− 5
rs81459294	16	1,133,302	intron	CTNND2	16: 508245–1,521,550	1.0 × 10− 4
rs80868434	7	30,048,501	intergenic	MLN	7: 30003291–30,013,128	1.2 × 10− 4
rs80849858	1	77,472,299	intergenic	TRAF3IP2	1: 77484191–77,523,887	1.2 × 10− 4
rs336677749	17	28,600,733	intron	RALGAPA2	17: 28394827–28,679,538	1.3 × 10− 4
rs330020208	6	89,629,789	intron	PHC2	6: 89574035–89,685,278	1.3 × 10− 4
rs81370878	3	56,211,114	intergenic	VWA3B	3: 55950591–56,171,826	1.4 × 10− 4
rs323186575	6	91,486,462	intron	ZMYM4	6: 91455778–91,623,818	1.4 × 10−4
rs80970182	17	28,210,436	intron	CFAP61	17: 28111660–28,372,543	1.4 × 10−4
rs80930168	1	77,338,889	intron	REV3L	1: 77195253–77,401,051	1.9 × 10−4
rs80801203	17	28,465,351	intron	RALGAPA2	17: 28394827–28,679,538	2.1 × 10−4
rs338087144	14	77,361,157	intergenic	KAT6B	14: 77404443–77,597,673	2.4 × 10−4
rs81241392	17	28,616,449	intron	RALGAPA2	17: 28394827–28,679,538	2.6 × 10−4
rs81290595	15	6,253,250	intron	ENSSSCG00000036561	15: 6222423–6,266,771	2.8 × 10−4

^aSSC = Sus scrofa chromosome

^bLocation in Ensembl Sscrofa11.1

^cIf the variant was intergenic, the closest gene within a 1 Mbp window was indicated

^dSNVs under the suggestive significance threshold of 1.0 × 10⁻⁵ in bold [39]

Discussion

This study aimed to assess susceptibility to Salmonella in commercial pigs by using a GWAS approach to identify potential variants associated with Salmonella IgG antibody response from end of nursery to end of finisher, Salmonella shedding from weaning to end of finisher, and isolation of Salmonella from tissues at slaughter. GWAS is a useful tool in preliminary identification of novel SNVs or genes and associations between genetic variants and disease resistance or susceptibility. This has promise for future efforts in breeding for resistance to Salmonella on-farm and at slaughter, which may lead to a reduction of Salmonella benefiting both public health and animal welfare.

GWAS analysis of shedding from nursery to finisher stage

GWAS analysis identified two associated SNVs. The first was an intron variant located in the PTPRJ gene, which encodes protein tyrosine phosphatase (PTP) receptor type J (PTPRJ) and was observed at a higher frequency in pigs that never shed Salmonella. These PTPRJ receptors are found on many cell types and are highly expressed in the macrophage rich tissues of the intestines [40]. PTPRJ receptors are responsible for negative regulation of protein tyrosine kinases, which are involved in pathogen recognition and clearance, as well as resolution of inflammation [41–44]. PTPs may also be involved in B cell activation, recruitment of neutrophils, and the negative regulation of T-cell signaling [39, 45–47]. The timely recruitment of neutrophils is a crucial part of the immune response and often plays a key role in the clearance of pathogens in the gut [48]. A challenge study on PTPRJ-deficient mice showed impairment of B cell development but an unexpected significant increase in early neutrophil recruitment and rapid clearance of Staphylococcus aureus in a subcutaneous air pouch and suggested that PTPRJ plays both a positive and negative role in attraction of neutrophils [49].

Interestingly, the PTPRJ variant was observed at a significantly higher frequency in pigs with no Salmonella shedding observed from weaning to end of finisher. It is possible that, this variant may be causing a reduction in the expression of PTPRJ resulting in a subsequent rapid increase in neutrophil recruitment and clearance of Salmonella infection in the control population [49], while RPTP, another member of the PTP family, is rescuing some B cell functionality and signalling [48]. Currently, more research is needed into the putative function of PTPRJ in pigs and how this variant and its interactions with other PTPs, like RPTP, may affect B cell signalling and neutrophil recruitment in response to pathogens like Salmonella before a definitive conclusion can be made.

Analysis of the shedding trait also identified an intergenic variant nearest to the ST6GALNAC3 gene, which encodes ST6 (α-N-acetylneuraminyl-2,-3-β-galactosyl-1,3)-N-acetylgalactosamine-α-2,6-sialyltransferase 3 (ST6GALNAC3) and was also more frequent in pigs that did not shed Salmonella. ST6GALNAC3 is a member of a family of sialylatransferases involved in the transfer of sialic acids CMP-sialic acid to O-glycans with α-2,6 linkage [50–52]. Sialylated glycans, like those produced by ST6GALNAC3, are involved in many key cellular processes including cellular adhesion, self-recognition, and signaling [51, 53–55]. Currently, little is known about the in vivo function and differential expression of ST6GALNAC3 in pigs [52, 56]. However, in vitro studies have demonstrated that sialic acid moieties on the mucosal surface may be used by pathogens, including Salmonella, as a nutrient source or as receptors for adhesion and invasion [51, 55, 57, 58]. One such study found that removal of N-acetylneuraminic acid, the most abundant cell surface sialic acid, greatly impaired Salmonella attachment to Caco-2 cells [57]. Additionally GD1a, synthesized by ST6GALNAC3, has also been implicated as a co-receptor for Salmonella flagellin (FliC) and is thought to induce β-defensin-2 in Caco-2 cells [59].

To date, the function and expression of ST6GALNAC3 largely remains an unknown in pigs. In the current study, the variant was seen at a higher frequency in pigs that were not observed to be shedding from weaning to finisher stage. Due to the role of sialic acids in Salmonella adhesion and nutrient acquisition, it is possible that this variant near ST6GALNAC3 is resulting in a change of the sialylation of the mucosal surface. This may be impacting either the availability and localization of host-derived sialylated glycans and impairing the rate of Salmonella adherence or nutrient acquisition and thus may affect Salmonella susceptibility in pigs.

GWAS analysis of colonization at slaughter

The model for isolation of Salmonella from tissues at slaughter identified three intergenic variants with a suggestive association nearest to the gene encoding doublecortin domain containing 2C (DCDC2C). Two of these variants were more frequent in pigs that tested negative at slaughter, while the third was seen at a higher frequency in pigs positive at slaughter. The DCDC2C gene is currently uncharacterized in pigs and little is known about its function in vivo. Studies of DCDC2C in humans have suggested that variants in this gene may be associated with structural defects in cilia in sperm and in cilia length in sensory cells in the ear [60, 61]. Thus, it is possible that the variants observed in the current study also result in structural defects, either in the cilia in the lungs, or perhaps the villi of the intestines of pigs and may be changing the pigs’ susceptibility to Salmonella. How tissue colonization at slaughter may be being altered is yet unclear, as SNVs associated with both negative and positive isolation of Salmonella at slaughter were identified. Additionally, these variants are approximately 400 kbp from the DCDC2C gene, and thus the observed effect on the slaughter trait may be related, in part, to linkage with a functional SNV in a different gene. Regardless, DCDC2C may be a potential target for further investigation Salmonella susceptibility in pigs.

Analysis of the slaughter model identified an additional three intergenic variants near the gene encoding A-kinase anchor protein 12 (AKAP12) that were approaching the suggestive association threshold. An intron variant in AKAP12 was also identified in the top 15 significant SNVs for the slaughter trait. AKAP12 is a member of the AKAP family of scaffolding proteins involved in the recruitment and anchoring of protein kinases (PK) A and B. AKAP12 has be shown to regulate the subcellular localization of PKA and play a role in PKB regulation [62, 63]. PKA and PKB signaling play a crucial role in intramacrophage killing. Salmonella is known to subvert the host cell machinery and manipulate these pathways to persist intracellularly through the reduction of reactive oxygen intermediates or the manipulation of motor proteins [64, 65]. Studies have shown that inhibiting PKA and PKB activation greatly reduces virulence and impairs intracellular growth and survival of Salmonella [66, 67].

In light of the role that AKAP12 plays in the localization and regulation of PKA and PKB, which are key players in the intracellular survival of Salmonella, it is possible that variants near AKAP12 are altering the expression of AKAP12 and reducing or inhibiting the normal localization of PKA and PKB. This, in turn, may be impairing intracellular survival of Salmonella in host macrophages and preventing or limiting the systemic spread of Salmonella to lymphatic tissues and may account for the significantly higher frequency of these variants in the control pigs that were negative for Salmonella in tissues collected at slaughter.

GWAS analysis of seropositivity from nursery to finisher stage

GWAS analysis did not identify any significant associations with the tested variants. The closest SNV approaching the suggestive association threshold was an intergenic variant nearest to the gene RALGAPA2, which encodes the gene Ral GTPase-activating protein (RalGAP) catalytic alpha subunit 2 (α2) of RalGAP and was observed at a higher frequency in pigs that tested seropositive for Salmonella at least once. An additional two intergenic variants near and one intron variant in RALGAPA2 were identified in the top 15 significant SNVs for the seropositivity model. RalGAPs are involved in the negative regulation of RalGTPases, RalA and RalB, in the Ras/Ral signaling pathways and regulate crucial cellular processes including response to infection and mediation of inflammation [68–72]. Notably, constitutive expression of RalGTPases has been shown to promote tumorigenesis, increased expression of inflammatory cytokines, and increased epithelial permeability [69, 73–76].

It is possible that these variants are altering the expression of RALGAPA2 and, in turn, causing constitutive expression of RalGTPases. Considering the role of Ral GTPases in cell survival, inflammation, and permeability, this constitutive expression may be promoting susceptibility to Salmonella. Firstly, by creating an inflammatory environment in which Salmonella is known to thrive [77–79], or secondly, by impairing infected cell death and promoting the intracellular survival of Salmonella leading to chronic or persistent infection and higher antibody production [80–83]. However, it is important to note that the ELISA chosen for the current study tested for Salmonella IgG antibodies, but it is possible that some pigs had only produced a preliminary IgM response. Further studies are needed to assess the presence of IgM antibodies and whether they may result in stronger associations between the identified variants and Salmonella seropositivity. Additionally, a targeted study into RALGAPA2 and its pathways may identify an association with Salmonella susceptibility that was missed at the genome-wide level.

Conclusion

To the best of the authors knowledge, this is the first GWAS analysis assessing susceptibility to Salmonella in pigs at different stage of production on commercial farms and at slaughter. The variants identified herein were generally in or near genes involved in broad innate immunity and may be of great interest in improving genetic resistance to control enteric pathogens on swine farms. Overall, much of the porcine innate and humoral immune responses to these pathogens remains unknown. Further investigation of these genes on-farm, and the function, expression, and pathways of these genes both in vitro may greatly improve our understanding of the genetic basis of susceptibility to Salmonella and other swine enteric pathogens.

Methods

Animal use in this project was approved by the University of Guelph Animal Care Committee (AUP# 3124) and follows Canadian Council of Animal Care guidelines (CCAC, 2009).

Animals and sample collection

Farm and pig selection for this study have been detailed previously [84]. Briefly, 14 groups of 54–60 pigs (three-way Yorkshire x Landrace x terminal boar line cross) were selected from eight commercial farrowing sources in Southern Ontario; all pigs were housed on commercial farms and kept under the management of the farm personnel. Two groups of pigs were selected from each of six farrowing sources (designated Cohorts One and Two), while the remaining two farrowing sources had only one group each (Cohort One) for a total of 14 groups. During the nursery stage, pigs received either a conventional diet (high complexity) or a lower cost reduced animal protein diet (low complexity) in which animal protein was replaced by plant protein as part of a larger study [85]. After which, all pigs received conventional diets during grower-finisher stage.

Blood and fecal samples were collected at weaning, and at the end of the nursery, grower and finisher stages. Rectal swabs (Starplex®, VWR International, Mississauga, Ontario, Canada) were taken in the event that a fecal sample could not be obtained. Blood was drawn from the jugular or suborbital vein, transported to the lab in a cooler, and then centrifuged at 1500 x g for 20 min. Sera were stored at − 20 °C. At the end of production pigs were shipped to the abattoir for slaughter where palatine tonsils and submandibular lymph nodes were collected from a subset of 580 pigs.

Salmonella antibody detection

Collected sera were tested for IgG antibodies to Salmonella O-antigens 1, 3, 4, 6, 7, 9, 10, and 12 via a commercial enzyme-linked immunosorbent assay (ELISA; pigtype® Salmonella Ab kit, QIAGEN Leipzig GmbH, Leipzig, Germany) as described previously [86]. Sample optical density (OD) was measured at 450 nm using a BioTek Synergy HT Multi-Mode Microplate Reader and BioTek’s Gen5 software version 11.1. The sample-to-positive (S/P) ratios were calculated as follows:

$$\mathrm{S}/\mathrm{P}=\frac{{\mathrm{OD}}_{\text{sample}}-\text{Mean}{\mathrm{OD}}_{\text{negative control}}}{\text{Mean}{\mathrm{OD}}_{\text{positive control}}-\text{Mean}{\mathrm{OD}}_{\text{negative control}}}$$

A sample was considered seropositive if the S/P ratio was greater or equal to 0.3.

Salmonella isolation

Fecal samples and tissues were cultured as detailed previously [84]. Briefly, 10 g of fecal or tissue sample was homogenized in 50 mL of tetrathionate broth (Oxoid, Nepean, Ontario, Canada) using a Seward Stomacher 400 Circulator (Seward Laboratory Systems Inc., Bohemia, New York, USA) and incubated for 24 h at 37 °C. After which, 0.1 mL of tetrathionate broth culture was transferred into 9.9 mL Rappaport Vassiliadis (RV) broth (Oxoid, Nepean, Ontario, Canada) followed by another 24 h incubation at 42 °C. Then, a loopful of RV culture was plated onto xylose-lysine-tergitol 4 agar (Becton Dickinson™, Baltimore, Maryland, USA) and incubated for a final 24 h at 37 °C. Salmonella colonies were confirmed by a Salmonella O Antiserum Poly A-I and Vi (Becton Dickinson™, Grayson, Georgia, USA) slide agglutination test.

Salmonella phenotypes

The following binary traits were defined for each pig: Salmonella seropositivity from end of nursery to end of finisher (never seropositive vs. seropositive once or more), Salmonella shedding from weaning to end of finisher (never shed vs. shed once or more), and positive isolation of Salmonella from tonsils and/or lymph nodes at slaughter (yes/no). Seropositivity at weaning was excluded when categorizing pigs into binary traits due to the concern of maternal antibodies confounding the results in the GWAS. Supplementary Figure 1 depicts a simplified flow chart of the sample collection and case-control groups by phenotype.

Genotyping and quality control

DNA was extracted from tail dockings, ear tissue, or blood using the DNEasy Blood and Tissue kit (Qiagen, Mississauga, Ontario Canada), with either 25 mg of tissue or 100 ul of blood [35]. DNA was genotyped at Eurofins BioDiagnostics, Inc. (River Falls, Wisconsin, USA) using a custom 54 K Affymetrix Axiom® myDesign™ chip designed in consultation with the Canadian Centre for Swine Improvement (Ottawa, Ontario, Canada). Quality control (QC) of SNVs was performed in PLINK v1.9 [87, 88]. As this chip contained select SNVs from multiple research groups, some of which had proprietary labels that precluded their identification in the genome, SNVs were removed if they had no corresponding rsID in Ensembl Sscrofa11.1. Pigs with SNV call rates of less than 90% were excluded and SNVs with a minor allele frequency lower than 5% or a call rate less than 95% were removed from further analysis. The SNV call rates were compared between cases and controls, and Fisher’s exact test was used to exclude any SNV in which the missingness between a case and control was significantly different (p < 1.0 × 10^− 5).

Genome-wide association study

A GWAS model was used to assess each of the phenotypes described above. Pigs that tested seropositive for Salmonella or shed Salmonella at least once on-farm, and pigs that were positive for isolation of Salmonella at slaughter were considered to be the case population. Pigs that never tested seropositive, shed Salmonella, or tested negative at slaughter were assigned as the control population for the GWAS models. The outcome variable for each model was seropositive at least once from end of nursery to end of finisher (seropositivity model), shedding detected at least once from weaning to end of finisher (shedding model), and positive isolation at slaughter (colonization model), respectively.

Univariable analysis of each phenotype was performed using a mixed-effects logistic regression model in Stata (Stata/IC 14.2 for Windows, StataCorp LP, Texas, USA) to assess the significance of associations between covariates and the phenotype (p < 0.05) and subsequent inclusion in the GWAS model [84, 86]. For the seropositivity model, only farm was included as a covariate. Covariates included in the shedding model were diet and cohort, while covariates for the colonization model were age at slaughter, cohort, and season. To account for the cryptic relatedness of the pigs in a study population with no sire information, a genomic relatedness matrix (GRM) was calculated using the genome-wide efficient mixed model association algorithm (GEMMA) v0.96 [37, 38, 89]. Genome-wide analyses of case-control data were performed using a generalized logistic mixed model association test (GMMAT) [90] using the GMMAT v0.9.3 package for R [91, 92]. The covariates analyzed in Stata and the GRM created in GEMMA were included in the GMMAT analysis. The chosen p-value threshold for suggestive and significant associations was 1.0 × 10^− 5 and 5.0 × 10^− 7, respectively based on a study by Burton et al. [39]. The Wald test was used to assess the significance of associations between SNVs and the Salmonella phenotypes. Allele frequencies in case-control populations were calculated and an allelic chi-squared (χ²) test in PLINK v1.9 was used to determine if there was a significant difference between case and control pig populations. In the event that an intergenic variant was identified as significant, the closest gene within a 1 Mbp window, either upstream or downstream, was assumed to be the causative gene. The data generated from this GWAS has been deposited in the Animal Quantitative Trait Loci Database (AnimalQTLdb) and can be accessed as a group via https://www.animalgenome.org/QTLdb/supp/?t=MwAn1O1XiA, or individually via association (QTL) numbers: 194531–194535.

Abbreviations

AKAP

A-kinase anchoring protein

ELISA

Enzyme-linked immunosorbent assay

DCDC2C

Doublecortin domain containing 2C

GWAS

Genome-wide association study

GRM

Genomic relatedness matrix

GMMAT

Generalized logistic mixed model association test

GEMMA

Genome-wide efficient mixed model association

IgG

Immunoglobulin G

MBL

Mannan-binding lectin

NOD

Nucleotide-binding oligomerization domain-containing protein

OD

Optical density

PK

Protein kinase

PTP

Protein tyrosine phosphatase

PTPRJ

Protein tyrosine phosphatase receptor type J

QC

Quality control

RalGAP

Ral GTPase-activating protein

RALGAPA2

Ral GTPase-activating protein catalytic alpha subunit 2

RV

Rappaport Vassiliadis

S.

Salmonella

SNV

Single-nucleotide variant

S/P

Sample-to-positive

ST6GALNAC3

ST6 (α-N-acetylneuraminyl-2,-3-β-galactosyl-1,3)-N-acetylgalactosamine-α-2,6-sialyltransferase 3

TLR

Toll-like receptor

Authors’ contributions

The project was conceived, study designed, and funding obtained by BNL, AF, and RMF. Bacterial culture and host DNA extraction was performed by MHAG. Antibody assays were performed by CHS. Bioinformatics and statistical analysis was performed by CHS, RSF, and AF. CHS, AF, and BNL were major contributors to writing the manuscript. All authors read and approved the final version of the manuscript.

Funding

This work was supported by the Natural Sciences and Engineering Research Council of Canada Collaborative Research and Development Grant [grant number 488429–15]; the Ontario Ministry of Agriculture, Food and Rural Affairs [grant number UofG2013–1472, HQP Fellowship for C. Schut]; Swine Innovation Porc [1239e], the Canadian Center for Swine Improvement, Alliance Genetics Canada, and Ontario Pork [15/015]. Alliance Genetics provided assistance in farm recruitment and sample collection. Genotyping was done on a custom SNP chip designed by the Canadian Centre for Swine Improvement, Inc. Funding sources did not contribute in any other way to the manuscript.

Availability of data and materials

The data generated from this GWAS has been deposited in the Animal Quantitative Trait Loci Database (AnimalQTLdb) and can be accessed as a group via https://www.animalgenome.org/QTLdb/supp/?t=MwAn1O1XiA, or individually via association (QTL) numbers: 194531-194535.

Ethics approval and consent to participate

Animal use in this project was approved by the University of Guelph Animal Care Committee (AUP# 3124) and follows Canadian Council of Animal Care guidelines (CCAC, 2009). Informed verbal consent was obtained from participating producers by phone and on-farm during a questionnaire of farm operating procedures as approved by the ethics committee.

Consent for publication

Not applicable.

Competing interests

None of the authors have any competing interests.

Supplementary information

Supplementary information accompanies this paper at 10.1186/s12917-020-02344-0.