Identifying emerging trends in antimicrobial resistance using Salmonella surveillance data in poultry in Spain

Abstract Despite of controls and preventive measures implemented along the food chain, infection with non‐typhoidal Salmonella (NTS) remains one of the major causes of foodborne disease worldwide. Poultry is considered one of the major sources of NTS. This has led to the implementation of monitoring and control programmes in many countries (including Spain) to ensure that in poultry flocks infection is kept to a minimum and to allow the identification and monitoring of circulating NTS strains and their antimicrobial resistance (AMR) phenotypes. Here, we investigated the information from the monitoring programme for AMR in Salmonella from poultry in Spain in 2011–2017 to assess the diversity in phenotypic resistance and to evaluate the programme's ability to detect multi‐resistance patterns and emerging strains in the animal reservoir. Data on serotype and AMR to nine antimicrobials obtained from 3,047 NTS isolates from laying hens (n = 1,060), broiler (n = 765) and turkey (n = 1,222) recovered during controls performed by the official veterinary services and food business operators were analysed using univariate and multivariate methods in order to describe host and serotype‐specific profiles. Diversity and prevalence of phenotypic resistance to all but one of the antimicrobials (colistin) were higher in NTS from broiler and turkey compared with laying hen isolates. Certain combinations of serotype and AMR pattern (resistotype) were particularly linked with certain hosts (e.g. susceptible Enteritidis with laying hens, multi‐drug resistant (MDR) Derby in turkey, MDR Kentucky in turkey and broiler). The widespread presence of certain serotype‐resistotype combinations in certain hosts/years suggested the possible expansion of MDR strains in the animal reservoir. This study demonstrates the usefulness of the analysis of data from monitoring programmes at the isolate level to detect emerging threats and suggests aspects that should be subjected to further research to identify the forces driving the expansion/dominance of certain strains in the food chain.


| INTRODUC TI ON
Despite the control measures implemented in food production systems, foodborne infections remain one of the top causes of disease worldwide . Salmonellosis due to non-Typhoidal Salmonella (NTS) is one of the leading causes of foodborne illness, with a global estimate of 129 million cases every year, of which 100,000 to 1 million result in death (mostly in immunocompromised patients) . Even though incidence may be severely underestimated due to underreporting, levels of salmonellosis have lately remained stable in the United States (since the late 90's) (Henao et al., 2015) and the European Union (since 2013) (EFSA & ECDC, 2018), thus highlighting the challenge posed by NTS for public health. The issue is further amplified by the emerging threat of multi-drug resistant (MDR) Salmonella, as though the disease is usually self-limiting, antimicrobial therapy is commonly necessary for invasive infections and recommended for patients at increased risk (Shane et al., 2017). Moreover, MDR NTS strains can act as reservoirs of resistance genes that can be horizontally transferred to other bacteria of the same or different species via mobile genetic elements, therefore contributing to continued spread of antimicrobial resistance (FAO, 2016).
According to recent source attribution studies, poultry, especially laying hens but also broilers and turkeys, is considered one of the major sources of foodborne salmonellosis (De Knegt, Pires, & Hald, 2015;Glass et al., 2016;IFSAC, 2015;Pires, Vieira, Hald, & Cole, 2014). Although Salmonella contamination is frequent in poultry production, predominant serotypes may change depending on host species, time period, region or production system. In response, European Union (EU) regulations have set targets for the reduction of specific serotypes, selected due to their public health significance in flocks of breeding hens, laying hens, broilers, breeding turkeys and fattening turkeys (Messens et al., 2013) (i.e. S. Enteritidis and S. Typhimurium, including its monophasic variant, for all species, plus S. Hadar, S. Virchow and S. Infantis in breeding hens). However, Salmonella infection in poultry (and other animal reservoirs) is known to show considerable variation, with frequent shifts in predominant serotypes (Foley, Lynne, & Nayak, 2008;Shah, Paul, Sischo, Crespo, & Guard, 2017) often driven by the expansion of highly dominant clones, such as S. Typhimurium phage type DT104 (Glynn et al., 1998;Threlfall, 2000). Still, the mechanisms by which such clones become dominant or why their frequency eventually decreases are not well understood (Branchu, Bawn, & Kingsley, 2018).
In Spain, Salmonella monitoring and control programmes in poultry are regulated by EU and Spanish laws and are implemented to ensure that all poultry populations reach the specific prevalence targets set for serotypes of public health relevance. Monitoring involves periodical testing of breeding flocks of Gallus gallus (every two weeks), flocks of laying hens (every 15 weeks) and flocks of broilers and turkeys (birds leaving for slaughter) as a responsibility of the food business operators (auto-control checks) (Anon., 2003). In addition, the official veterinary services perform annual controls (official controls) on commercial farms depending on farm size (Anon., , 2011(Anon., , 2012a(Anon., , 2012b. All the Salmonella isolates coming from official and a proportion of those originating from auto-control checks (which are used when isolates from official checks do not reach the minimum required number) are then subjected to antimicrobial susceptibility testing, again following EU rules, to monitor the occurrence of antimicrobial resistance (AMR) in zoonotic agents (Anon., 2013). This allows the identification of trends of AMR occurrence and the comparison with trends in isolates recovered from food and human samples at a European and national level, typically expressed as changes of the aggregated proportion of resistant isolates to a given antimicrobial over time (EFSA & ECDC, 2018). However, analysis of this information at an isolate level can also help to identify multi-resistance patterns, detect the spread of possible emerging strains and identify possible factors associated with their expansion (Aerts & Jaspers, 2012). The increasing availability of information and analytic approaches facilitated by the revolution in the field of information technology opens new possibilities for surveillance of infectious diseases and detection of emerging threats, including specific Salmonella clones (Besser, 2018).
In the present study, isolate-level data from the Spanish national surveillance programme on AMR of Salmonella from poultry (laying hen, broiler and turkey) during 2011-2017 were analysed to assess the diversity in phenotypic resistance and to detect multi-resistance patterns and emerging strains using univariate and multivariate analytic methods, including multiple correspondence analysis and hierarchical clustering.

| Study population and laboratory methods
All isolates included in this study were retrieved through official and auto-control checks performed in flocks of laying hens (adults), broilers (within three weeks before slaughter) and fattening turkeys (within three weeks before slaughter) during 2011in Spain according to Commission Regulations No 517/2011, 200/2012and 1190/2012respectively (Anon, 2011, 2012a, 2012b

| Descriptive analysis
The proportion of isolates belonging to different serotypes and exhibiting specific resistance phenotypes was compared within/between host species using chi-square tests corrected for multiple comparisons using Holm's method (Holm, 1979). The number of antimicrobials to which an isolate was resistant for each host species was compared using the Kruskal-Wallis test followed by the Dunn's post hoc test correcting for multiple comparisons through the Holm's method. In addition, information on the phenotypic resistance and yearly frequency of isolation of the most abundant serotypes (n > 200) were evaluated individually. Analyses were carried out in R (R core Team 2014) using the package FSA (Ogle, Wheeler, & Dinno, 2018).

| Analysis of diversity
The resistotype of each isolate was defined as the concatenation of its phenotype (resistant/non-resistant) for each of the nine antimicrobials included in the study. Similarity between resistotypes present in isolates from each host species was evaluated using Venn diagrams, and within host species resistotype diversity distribution was evaluated using Simpson's Index of diversity D as follows (Hunter & Gaston, 1988): where s is the total number of resistotypes, n j is the number of isolates belonging to resistotype j, and N is the total number of isolates for each host. This index expresses the probability that two isolates that are randomly selected among the collection from a given host species will have different serotype/resistotype. Confidence intervals for the Simpson's indexes were estimated through 1,000 bootstrap replicates. Additionally, a rarefaction analysis to estimate the resistotype richness controlling for the different sample size in each host species was carried out as previously described (Ahlstrom et al., 2017). For serotypes in which at least 15 isolates from each host were recovered, the proportional similarity index (PSI), as described by Muellner et al. (2010), was used to compare resistotypes in each host. The PSI is an objective and simple measure of the area of intersection between two frequency distributions. It estimates the similarity between the frequency distributions of, for instance, bacterial subtypes (here resistotypes in a given serotype) from different sources.
The value for the PSI ranges from 1 for identical frequency distributions to 0 for distributions with no common types, and bootstrap confidence intervals were estimated. Analyses were carried out in R (R core Team, 2014) using the Packages VEGAN and Venn.Diagram.

| Association between phenotypic resistances
For the most abundant serotypes (n > 50 isolates) the existence of associations between the simultaneous phenotypic resistance to any pair of antimicrobials was evaluated through Fisher's exact test using Holm's correction to adjust for multiple comparisons as previously described (Boerlin et al., 2005).
Then, a multiple correspondence analysis (MCA) was performed to analyse the multivariate pattern of relationship between the phenotypic resistances while accounting also for other nominal variables (here host species and serotype) and reduce the dimensions of the data as previously performed (Jaspers, Ganyani, Ensoy, Faes, & Aerts, 2016;Pages-Monteiro et al., 2017). Clusters in the reduced dimensions obtained from the MCA, determined based on Euclidean distances, were investigated using hierarchical clustering with Ward's minimum variance method (Ward, 1963). The number of clusters was set empirically as described in Husson, Josse, & Pages (2010). Analyses were carried out using the R package FactoMineR (Le, Josse, & Husson, 2008).  Table S1). Most of the isolates from laying hens were retrieved from official controls (82.7%) while for broiler and turkey the main source of isolates were auto-controls performed by the food business operators (60.6 and 77.9% of the isolates, respectively), with large differences depending on the year (Table S1).

| Descriptive results
Overall, a total of 127 serotypes were identified, with the distribution of the main serotypes varying largely with the host (Table   S2). In laying hens, no serotype accounted for more than 25% of the isolates, and serotypes other than S. Enteritidis (20.3%), S. Infantis (11.7%), S. Corvallis (8.8%) and S. Ohio (7.9%) represented less than 5% of the isolates. In total, those four most frequent serotypes accounted for 48.7% of all isolates from laying hens. In broilers, only six serotypes accounted for over 5% of the Salmonella isolates in this host, and jointly made up 73% of all isolates (S. Mikawasima, 16.1%; S. Kentucky, 14.4%; S. Virchow, 8.2%; S. Infantis, 7.2%; S. Havana, 5.9%; and S. Typhimurium, 5.1%). In the case of turkey, the top four serotypes (S. Derby, 51.4%; S. Hadar, 13.1%; S. London, 11.1%; and S. Kentucky, 6.4%) accounted for over 80% of the total with over half of the turkey isolates belonging to the most common serotype,

S. Derby.
Frequency of phenotypic resistance was also highly dependent on the host species, with the proportion of isolates from laying hens resistant to any antimicrobial remaining below 20% (and below 11% except for the quinolones) compared with resistance rates above 20% for six and seven of the nine antimicrobials evaluated in the case of broiler and turkey isolates, respectively (Table 1). Significant host-specific differences in the proportion of resistant isolates were identified for all antimicrobials, with the most common pattern being turkey > broiler>laying hen (for Amp, Chl, Cip, Smx, Tet and Tmp) followed by broiler > turkey>laying hen for Gen and Nal. The only antimicrobial in which the highest proportion of resistance was found for laying hens was Cst (Table 1). This was reflected by statistically significant (p < .0001) differences in the mean number of antimicrobials to which an isolate was resistant depending on the species, which was significantly (p < .0001) higher in the case of turkey (median = 5) compared with broiler (median = 2) and laying hen (median = 0), with the latter difference also being significant (p < .0001).

| Diversity of AMR phenotypes
Overall, 94 out of the total 2^9 = 512 possible resistotypes were found in the isolate collection, of which 67 were found among broiler isolates, 70 among turkey isolates and 40 among laying hen isolates; approximately one-third of the resistotypes (33/94) were present in isolates from all three hosts, while 22, 19 and 3 were exclusive to broiler, turkey and laying hen isolates, respectively (Figure 1).
Resistotype diversity was similar for isolates from broiler (Simpson's Index = 0.89, 95% confidence interval 0.83-0.94) and turkey (0.86, 95% CI 0.78-0.93), and higher than that for isolates from laying hen (0.49, 95% CI 0.05-0.65) ( Figure 2). These results were in agreement with the rarefaction analysis: the resistotype accumulation curves for broiler and turkey showed considerably steeper slopes so that an increasing sample size was translated into a more drastic increase of resistotypes compared with laying hen, thus pointing out a larger richness of resistotypes in turkey and broiler isolates compared with those from laying hen ( Figure 3).  Table S3). Interestingly, S. Typhimurium resistotypes from broilers were significantly less similar to those from laying hens (0.14; 95% CI 0-0.34) than to turkey isolates. A similar pattern, but not significant, could be observed for S. Kentucky (Table S3).

| Association between phenotypic resistances
Bivariate association between phenotypic resistance against any two antimicrobials was assessed in 15 serotypes with n > 50 (

| Salmonella Derby
The most common serotype in the collection in this study was S.
Derby (637/3,047; 21% of all isolates), with almost all isolates recovered from turkey (628/637) and only four and five isolates cultured from broiler and laying hen samples, respectively. As discussed above, more than 50% of the turkey isolates were identified as S.
Derby, although the proportion of turkey isolates belonging to this serotype in our collection decreased sharply during the study period, dropping from > 60% in 2011-2014 to less than 30% in 2017 ( Figure 5).
Resistance to Amp, Cip, Smx, Tet and Tmp was extremely high for Smx (n = 2) and Tet (n = 5)) allocated to Cluster 1 and the remaining two isolates classified in Clusters 2 and 3.

| Salmonella Kentucky
Overall 228 isolates identified as S. Kentucky were retrieved over the study period, of which 110 were isolated from broiler (48%), 40 from laying hen (18%) and 78 from turkey (34%). The proportion of isolates from turkey belonging to this serotype in the collection increased dramatically over the study period, going from less than 3% before 2015 to 18.5% in 2016-2017, while no similar trend during the whole study period was observed in the case of isolates from laying hen or broilers ( Figure 5).
The resistance profile of the S. Kentucky isolates appeared to depend highly on the host, with a median number of 2.5, 5 and 6 resistances to different antibiotics in isolates from laying hen, broiler and turkey, respectively. Thirty per cent of the laying hen isolates were fully susceptible, whereas the most common resistotype in the isolates from the other two hosts included co-resistance to Amp, Cip, Gen, Nal, Smx and Tet, with up to 30% and

| D ISCUSS I ON
Given the importance of poultry and poultry products as sources of foodborne salmonellosis, national control programmes are a critical tool to ensure that presence of Salmonella remains at minimum levels in poultry populations, particularly for serotypes of special importance to public health (Messens et al., 2013). In addition, these programmes enable the identification of predominant serotypes and AMR phenotypes that can be used to better understand the epidemiology of Salmonella infection in the food animal reservoir. Here, we have demonstrated that data from the Spanish control programme in poultry can be used to provide valuable insight into Salmonella population dynamics in laying hens, turkey and broiler that could be indicative of expanding clones or trends in serotype or resistotype dominance.
In our collection, serotype distribution was much more heterogeneous in isolates from laying hen compared with broiler and especially TA B L E 3 Number and percentage of antimicrobial-resistant phenotypes for 228 Salmonella Kentucky isolates per host on the basis of the number of antimicrobials to which the isolates were resistant. The most common resistotype in each host and overall is indicated in bold turkey, from which half of the isolates recovered belonged to a single serotype that nevertheless became less frequent at the end of the study period. Care should be taken however when interpreting serotype data from broiler and turkeys since over half of the isolates included in this study for those two hosts were selected through convenience sampling among isolates retrieved during the controls performed by the food business operators ensuring only they originated from different epidemiological units, and may therefore not represent the true distribution of serotypes in these poultry populations.
Therefore the shifts observed in serotypes over time for broiler and turkey isolates may be partially due to sampling bias and not represent a true change in their prevalence. Still, a comparison of the ranks in the proportion of isolates belonging to each serotype found in auto-control checks performed by food business operators with those found in this study revealed no significant differences, thus suggesting that results approximate the field situation (results not shown).
While the rarefaction analysis suggested that a large amount of the resistotype diversity in Salmonella isolates circulating in laying hens was being captured through the achieved sample size, this was not the case for isolates from broiler and turkey even if they were more homogeneous in terms of serotype diversity. This is likely a result of the higher proportion of susceptible isolates to all the antimicrobials except Cst recovered from laying hen compared with those from turkey or broiler, which could reflect a higher use of antimicrobials in the latter two hosts (70.3% of all isolates from laying hens were pansusceptible). Interestingly, among the four serotypes in which at least 15 isolates from all three host types were found, only one (S. Mikawasima) had a highly similar resistotype regardless of the host species, which could be explained by a frequent resistotype (Amp-Cip) found in all three hosts. For the remaining three serotypes (Senftenberg, Kentucky and Typhimurium), isolates from laying hens were always less similar, therefore suggesting that in certain cases the production environment (including antimicrobial use) could play a major role in resistotype occurrence within a serotype versus resistance being mainly determined by the serotype alone.
Colistin was the antimicrobial against which the proportion of resistant isolates was lowest in all three host types. Most (67/91) of the Cst-resistant isolates were identified as S. Enteritidis, which is in agreement with previous reports from other European countries, suggesting that this phenotype may be associated with intrinsic serotype-specific differences in susceptibility (Agerso et al., 2012;EFSA & ECDC, 2018). The presence of several plasmid-mediated Cst-resistance genes (mcr-1, mcr-2, mcr-4, mcr-5 and mcr-9) in Salmonella has previously been reported in isolates with typically high levels of phenotypic/genotypic resistance to a large number of antimicrobials (Borowiak et al., 2017;Carattoli et al., 2017;Carroll et al., 2019;Garcia-Graells et al., 2018;Hu et al., 2019;Skov & Monnet, 2016). The fact that most (>75%) of the Cstresistant isolates found in this study were in contrast susceptible to all other antimicrobials tested except quinolones suggests that the plasmid-mediated mcr genes may not have been a frequent source of Cst-resistance in the Spanish isolates. This is in agreement with previous results obtained in NTS isolates retrieved from poultry in Spain in 2010-2017 with a suspected (MIC = 2) or confirmed (MIC ≥ 4) Cst-resistant phenotype screened for the presence of mcr-1 and mcr-2 (n = 37) or mcr-1 to mcr-5 (n = 115) using PCR. In this analysis, mcr genes were only detected in four (all mcr-1) and six (five mcr-1, one mcr-4) isolates, respectively (Agüero M., personal communication).
Nevertheless, MDR Cst-resistant strains were also found in low frequencies here, including four S. Derby and two S. London isolates resistant to > 5/8 antimicrobials evaluated in addition to Cst, similar to what has been described for mcr-1-positive Salmonella isolates belonging to these serotypes retrieved from food in China (Hu et al., 2019). Therefore, additional genomic characterization studies of S. Enteritidis isolates but also of MDR strains belonging to other serotypes would be required in order to assess the role of plasmid-  (Levings, Djordjevic, & Hall, 2008), which was first identified in the epidemic DT104 S. Typhimurium clone but that based on the phenotypes observed here would be particularly common in other serotypes (with > 45% of all S. Derby and S. London isolates exhibiting a Amp-Chl-Smx-Tet-Tmp resistance profile).
Resistance to Cip and Nal were also significantly associated with most prevalent serotypes ( Wai-Chi Chan, & Chen, 2015) and represents a major public health concern due to their ability to be transferred horizontally.
The existence of detectable clusters in the distribution of the isolates across the first three dimensions identified in the MCA further confirmed the association between groups of isolates based on the information on their AMR phenotype, serotype and host species. While Cluster 1 included the majority of the (susceptible) broiler and laying hen isolates, Cluster 4 included most turkey isolates that were also identified as S. Derby, the most prevalent serotype in the panel, and presented a MDR phenotype. Previous studies have suggested that this serotype, commonly found in pig and poultry, is polyphyletic, with certain lineages being primarily recovered from either pig or poultry (Hayward, Petrovska, Jansen, & Woodward, 2016;Sevellec et al., 2018 (Le Hello, Bekhit, et al., 2013). This strain, initially recovered in Europe from travellers to African countries in the 2000's (Le Hello et al., 2011), has since then been isolated from human and non-human sources in Europe, Africa and Asia, resulting in major public health concerns due to the variant's carriage of multiple resistance genes including extended-spectrum beta-lactamase (ESBL), plasmid-encoded cephalosporinase and carbapenemase (Le Hello, Bekhit, et al., 2013;Le Hello, Harrois, et al., 2013). In the most recent reports on AMR in zoonotic and indicator bacteria from humans, animals and food, S. Kentucky was the seventh/eight most common serotype recovered from humans in 2016/2017, but also displayed one of the highest levels of MDR (76.3% compared with a general level of 26.5% across all serotypes), and around 20% of the human S.

E TH I C S S TATEM ENT
The authors confirm that the ethical policies of the journal, as noted in the journal's author guidelines page, have been adhered to. No ethical approval was required as all isolates analysed here were retrieved through the ongoing Spanish national surveillance program on AMR performed according to national and EU regulations.  -2016-20422).

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interests in relation to this work.