Salmonella pathogenesis and host-adaptation in farmed animals.

: Salmonella is an animal and zoonotic pathogen of global importance. Depending on pathogen and host factors, infections can be asymptomatic or involve acute gastroenteritis or invasive disease. Genomic signatures associated with host-range, tissue tropism or differential virulence of Salmonella enterica serovars, and their variants, have emerged. In turn, it is becoming feasible to predict invasive potential, host-adaptation and zoonotic risk of Salmonella from sequence data to improve outbreak investigation, risk assessment and control strategies. Functional annotation of Salmonella genomes has accelerated with the screening of high-density mutant libraries, revealing host, niche-and serovar-specific virulence factors. As natural hosts and reservoirs, farmed animals provide powerful insights into host-adaptation and pathogenesis of Salmonella not always evident from surrogate rodent or cell-based models.


Introduction
Human non-typhoidal salmonellosis (NTS) is a key foodborne zoonosis, with an estimated 78 million illnesses, resulting in 59,000 deaths and loss of over 4 million disability-adjusted life years per annum [1].Chickens, pigs and cattle are key reservoirs and animal-sourced foods play a significant role in human NTS [2].Use of vaccines in broiler breeders and laying hens is partly credited with reducing Salmonella contamination of chicken meat and eggs in many countries, however effective control strategies for other farmed species are lacking.Among the c. 2600 antigenically-distinct serovars of S. enterica, some are host-specific and cause typhoid-like disease in farmed animals (e.g. S. Gallinarum in chickens), some cause severe illness in a narrow range of hosts (e.g. S. Dublin in cattle and humans), whereas others are more promiscuous and cause gastroenteritis in a wide range of hosts (e.g. S. Typhimurium).In many cases, infections can be asymptomatic or followed by a sub-acute carrier state.
It is now evident that classification of Salmonella into serovars that differ in host range or clinical presentation is crude [reviewed in 3,4].For example, within serovar Typhimurium, pathovariants exist that are host-adapted, including definitive phage type (DT) 2 and DT99 in pigeons [5], DT40 and DT56 in passerine birds [6], U288 in pigs [7], and sequence type (ST) 313 associated with invasive NTS (iNTS) in humans in sub-Saharan Africa [8,9].The molecular mechanisms underlying the differential virulence and tropism of serovars, and pathovariants within serovars, remain ill-defined.Here, we review recent insights into the basis of host-adaptation of Salmonella from comparative and functional genomics, note the importance of studying pathogenesis in natural hosts, and identify priorities for research.

Insights from comparative genomics
Whole-genome sequencing of prototype strains of common Salmonella serovars revealed the synteny and conservation of core genomes, but also identified significant variation in coding capacity due to variable prophage occupancy, serovar-specific genomic islands and loss-of function mutations, as first exemplified by comparing the genomes of S. Typhi and S. Typhimurium [10].The advent of inexpensive whole-genome sequencing is yielding a wealth of data on genetic variation within serovars, with 296,344 draft or complete Salmonella genomes at the time of writing [https://enterobase.warwick.ac.uk/].This revolution has been driven by integration of Salmonella sequencing into the diagnostic workflows of public health laboratories and studies on the genomic epidemiology of Salmonella in settings not well-served by surveillance, as with identification of pathovariants of S. Typhimurium and S. Enteritidis associated with human iNTS in sub-Saharan Africa.
Pathovariants associated with disseminated disease in humans have acquired loss-of-function mutations, particularly in genes associated with intestinal proliferation, such as anaerobic respiration and utilisation of inflammation-and microbiota-derived nutrients that are intact among gastroenteritis-associated variants [8,9,[11][12][13].Some of these loci, such as those associated with butyrate and 1,2-propanediol metabolism, have been proven to play a role during infection by gastroenteritis-associated S. Typhimurium [14,15].Loss-of-function mutations affecting Type III secreted proteins are also associated with invasive disease, as with inactivation of sseI in S.
Gene decay in human typhoidal serovars is balanced by the acquisition or retention of factors that aid immune evasion, such as the via locus encoding Vi capsular polysaccharide in S. Typhi and fepE regulator of very long lipopolysaccharide O chains in S. Paratyphi which confer resistance to oxidative killing [20].Gene decay is also a feature of host-specific or -restricted serovars that cause systemic disease in farmed animals.For example, loss of metabolic capacity was predicted from the genome sequences of avian-specific S. Gallinarum and S. Pullorum, relative to S. Enteritidis which is typically carried asymptomatically in poultry [21].Genome-scale metabolic models for 410 Salmonella strains spanning 64 serovars have since identified metabolic capabilities predicted to be associated with specific hosts and colonisation sites [22].Integrated metabolic, regulatory and protein-protein interaction networks also exist that have revealed common and distinct pathways of host-adaptation [23].Few of these pathways or networks have yet been confirmed experimentally to influence persistence or pathogenesis in farmed animals.Within serovar Enteritidis, a non-classical clade with intermediate levels of gene decay relative to serovar Gallinarum and Pullorum strains was identified that exhibited an intermediate level of systemic virulence [22], highlighting a continuum of ongoing genome evolution linked to host-adaptation.
Host-adaptation and gene decay of human-adapted serovars may have evolved as a consequence of the Neolithic transition towards an agricultural and pastoralist economy [24].Such evolution is ongoing in pathovariants of serovars in wild and domesticated animal populations on timescales measured in decades, as shown for S. Typhimurium [18] (Figure 1).Anthropogenic selection may also drive the emergence of new variants.For example, the widespread use of therapeutic levels of copper in the diet of domestic pigs as a growth-promoter in lieu of antibiotics may be associated with the emergence of a multi-drug resistant monophasic S. Typhimurium (4, [5],12:i-) ST34 clone in pigs that contains a novel genomic island implicated in resistance to copper, loss of which impairs fitness [19,25].
Single nucleotide polymorphisms (SNPs) can profoundly affect phenotypes.For example, SNPs affecting the FimH adhesin alter the ability of Salmonella to bind cell lines from different host species [26].Further, a SNP in the promoter region for pgtE in ST313 iNTS isolates markedly alters virulenceassociated phenotypes [27].Understanding variation in genome regulation is also important, as variation in the repertoire or activity of regulators can alter expression of virulence-associated loci.
For example, a mutation affecting the Salmonella pathogenicity island-1 master regulator HilD has been associated with impaired invasion of human, but not porcine, epithelial cells by S. Derby strains common in pigs, and may partly explain their low zoonotic risk [28].Moreover, quantitative proteomic analysis of phylogenetically similar S. Enteritidis and S. Dublin strains under gut-mimicking conditions indicated greater degradation of metabolic functions in S. Dublin than could be inferred from genome sequence data [29].
Machine learning has begun to identify genomic signatures associated with the animal host-oforigin of S. Typhimurium isolates.For example, Support Vector Machine predicted host-adapted S.
Typhimurium from analysis of protein variants [30], and Random Forest analysis of predicted functional variants reliably discriminated between iNTS and gastroenteritis isolates [13,18].Once trained to predict host-of-origin from sequence data, machine learning algorithms show promise for source attribution during outbreak investigations and predicting zoonotic risk [31,32].Studies have indicated rapid within-host evolution of S. Enteritidis and S. Dublin in chronically-infected human patients and identified genes and pathways that appear dispensable [33,34].The extent to which this applies in farmed animals during carrier states or with co-morbidities is unknown.

Insights from functional genomics
High-throughput screening of libraries of random or defined mutants has transformed understanding of the role of Salmonella genes in vivo.For example, transposon-directed insertion-site sequencing (TraDIS) can simultaneously assign the location of insertions and their effect on fitness by massively-parallel sequencing of transposon-flanking regions in pools of mutants before and after screening in animals (Figure 2).Screening of a library of S. Typhimurium mutants for intestinal colonisation of chickens, pigs and calves assigned phenotypes to 91% of mutants and 2715 different genes, including subsets inferred to play conserved and host-specific roles [35].Moreover, comparison of phenotypes for mutants in the same library between ileal mucosa and draining lymph nodes of calves identified niche-specific virulence factors [36].The existence of niche-specific virulence factors is further supported by screening of tagged S. Dublin mutants by oral versus intravenous inoculation of calves [37].A complementary approach has involved the hierarchical screening of pools of Salmonella mutants with specific deletions (first those lacking gene clusters, then mutants lacking single genes within virulence-associated clusters).This strategy identified S. Typhimurium genes required in chicks [38] and phenotypes for single gene deletions have been defined in bovine ligated ileal loops [39].While such studies substantially reduce the number of animals required to assign roles to bacterial genes in vivo, they are not feasible where output pools of a representative size cannot be obtained, for example from the avian reproductive tract and egg to understand vertical transmission.Moreover, they represent snap-shots during infection and have so far relied on a narrow range of strains, serovars, dosing regimen and host parameters.
As more information accrues on the role of Salmonella genes in colonisation and pathogenesis it should prove feasible to predict genomic signatures underlying host-adaptation with greater precision.Analysis of just 70 virulence-associated genes across 500 Salmonella genomes has already revealed associations with host range and invasive disease in farm animal-associated serovars [40].
Predictions may be aided by other forms of functional annotation, for example the transcriptome of Salmonella in vitro [41], and inside host cells [e.g.42], and from proteomic analysis of Salmonella recovered from animals [e.g.43].Such studies provide clues to the role and regulation of genes of hypothetical or unknown function.A curated and searchable database integrating sequence data, functional data from mutant screens, and data on gene expression would add value to existing datasets and enhance discovery.
Although genetic variation is a primary driver of phenotypic variation, it has also become clear that non-heritable phenotypic variation exists in clonal populations of Salmonella in the host [reviewed in 44] (Box 1).Box 1. Significant heterogeneity can exist in the phenotypes of Salmonella in vivo.Fluorescence dilution has revealed that non-replicating 'persister' cells of S. Typhimurium form readily following infection of macrophages [45].This approach relies on dilution of an inducibly-expressed fluorescent protein upon cell division relative to a constitutively-expressed fluorescent protein with different excitation-emission properties.The formation of slow-growing and dormant cells may have a significant impact on the activation and efficacy of host immune responses.For example, macrophage polarisation by Salmonella has been linked to the rate of intracellular net replication [46], in turn affecting the ability of reprogrammed macrophages to clear Salmonella [47].The extent to which heterogenous populations of Salmonella form across different serovars, pathovariants and hosts is unknown, and may be linked to the repertoire or expression of bacterial toxin-antitoxin modules required for persister cell formation [45].Significant differences among immune-related cells, mediators and genes also exist between cell lines and primary cells, and between mice and farmed animals, such that paradigms established thus far may not always apply.

Importance of research in natural hosts
Host-adaptation of Salmonella serovars and pathovariants is inferred from their epidemiology, and rarely confirmed by experimental inoculation of a range of hosts.Where strains have been proven to vary in host range and virulence, as with a panel of serovar Choleraesuis, Dublin, Gallinarum and Typhimurium strains tested reciprocally in pigs, calves and chickens, correlates of host-adaptation have emerged.For example, avirulence of S. Gallinarum in calves associates with reduced translocation via the gut lymphatic system compared to S. Dublin [48], and differential virulence of S.
Choleraesuis and S. Typhimurium strains in pigs associates with their net replication in intestinal mucosa and lymph nodes [49].Differential virulence of the same strains in calves was not reflected in phenotypes in primary bovine macrophages [50], however the S. Gallinarum strain survived better in avian primary macrophages compared to the S. Typhimurium and S. Dublin strains [51].Phenotypes in a streptomycin-pretreated mouse model of colitis were also not reliably predictive of virulence of the strains in farm animals [7,52].Taken together with evidence that the role of S. Typhimurium genes can differ across chickens, pigs and calves [35], these studies emphasise a need to study Salmonella in target animal hosts where feasible.Cost, infrastructure and ethical constraints limit this and consequently relatively few strains have thus far been characterised that may not be typical of the wider serovar.A strategy for co-screening wild-type strains based on massively-parallel sequencing of polymorphic alleles can reduce the number of animals needed to assign phenotypes in vivo [7,53].
The gut microbiota plays a key role in Salmonella pathogenesis by conferring resistance to colonisation and supplying metabolites [reviewed in 54].As the microbiota of farmed animals differs considerably, it will be important to study how this may impact on the outcome of infection (Box 2).Box 2. Microbiota differences may impact on the outcome of Salmonella infections in different hosts.
In murine models, Clostridia, Bacteroidaceae and Enterobacteriaceae have been implicated in colonisation resistance to Salmonella [55][56][57][58].This can occur in a manner sensitive to diet [59] and in neonatal chickens is associated with competition with commensal Enterobacteriaceae for oxygen [60] As the microbiota varies between different anatomical compartments and host species, it is plausible that host-adaptation of serovars or pathovariants may partly reflect how they interact with the resident microbiota, and the extent of reshaping of the microbiota by Salmonella-induced inflammation.For example, the microbiome and environment during asymptomatic colonisation of the chicken caeca by S. Typhimurium may be very different to that in the inflamed ileum of calves with S. Typhimurium-induced diarrhoea.

Future perspectives
Salmonella continues to evolve in farmed animals over short timescales as new ecological niches emerge.Its recent history has been punctuated by the rise and decline of epidemic variants (e.g. S. Typhimurium DT9, DT104, DT204 and ST34 in Europe since the 1960s, and ST313 in Africa).Knowledge of Salmonella-host interactions mostly derives from a small number of reference strains that may not reflect natural diversity or epidemiological trends in farmed animals.Moreover, the vast majority of studies employ murine models of infection or cell-based assays to investigate host-pathogen interactions.A pressing need exists to understand Salmonella pathogenesis in its natural hosts, including to interpret the impact of genetic differences between pathovariants, and to translate this knowledge to enhance farm animal health and food safety.[1985 words, not including abstract, highlights or boxes] * Analysis of the pan-genome of Salmonella revealed similar sized core and accessory metabolic reactomes.Carbon source uptake and metabolism formed a large proportion of the accessory reactome and was predictive of growth in defined media with high accuracy.Groups of phylogenetically unlinked serovars with a shared host-adaptation exhibited common patterns of metabolism consistent with convergent evolution to available metabolites.Typhimurium pathovariants.Maximum likelihood phylogenetic tree of S. Typhimurium strains based on variation in the core genome sequence (adapted from Tassinari et al [19]).Association of major clades with host species is indicated (key) with primary (dark colour) and secondary (light colour) host associations established in the literature.

Figure 1 .
Figure 1.Population structure and the primary and secondary host species association of S.

Figure 2 .Figure 1 SFigure 2
Figure 2. Assignment of roles to S. Typhimurium genes in intestinal colonisation of farmed animals