In this study, a convenience sample collection of 195 S. aureus isolates, including MSSA (n = 37) and MRSA (n = 158), was investigated (Table 1). Isolates were collected between 1993 and 2011 from twelve different host species in ten different countries (Table S1). Molecular typing of the 195 isolates revealed fourteen different spa types (t011, t034, t108, t571, t779, t899, t1197, t1344, t1451, t2576, t2876, t2974, t5972 and t6867) (Table S1). Spa types t011 and t034 were the most common, representing 45% and 40% of the isolates, respectively. Furthermore, approximately 50% of the isolates (n = 99) harboured SCCmec type V, while 27% of the isolates (n = 52) carried SCCmec type IV (Table S1).

We used denaturing high-pressure liquid chromatography (dHPLC) for mutation discovery at 97 genetic housekeeping loci (≈400 bp per locus) distributed along the S. aureus chromosome; in total, they constituted 1.4% (40,230 bp) of the CC398 genome.

Our analysis revealed 96 bi-allelic polymorphisms (i. e., polymorphic sites at which two alleles were observed) associated with 63 haplotypes. Among these polymorphisms were 34 synonymous point mutations in the protein coding genes, 58 non-synonymous point mutations, and 4 insertions or deletions ranging in size from 1 to 14 bp (Table S2). Of these, 41 polymorphisms were informative for maximum parsimony analyses. The nucleotide diversity, π (the average number of nucleotide dissimilarities per site among two isolate sequences), was 0.00008±0.00001 for the coding regions. The mean nucleotide substitution rate was estimated at 5.4×10⁻⁶ substitutions/nucleotide site/year (95% confidence interval, 3.5×10⁻⁶ to 7.5×10⁻⁶). This estimated mutation rate for the isolate collection is relatively faster than a previously reported evolutionary rate for other S. aureus strains [41], [42]. To investigate the time of the most recent common ancestor (TMRCA) of the 195 CC398 isolates, we applied a Bayesian coalescent method of phylogenetic inference as previously described [41]. According to the calculated mutation rate, the sequences variations and the isolation date (1993–2011) of our isolates dataset, we estimated that the TMRCA was ≈1974 (95% confidence interval, 1955 to 1991).

Based on these 96 polymorphisms, a minimum spanning tree (MST) was constructed (Figure 1). The MST demonstrated very limited diversity among the 195 investigated isolates. The ancestral node was determined by comparing concatenated sequences from the investigated loci of all investigated CC398 isolates with the concatenated sequences of N315 as an out-group. Rooting the phylogenetic tree of CC398 using N315 as an out-group revealed that isolates with spa type t899 (n = 2) were the most divergent group in comparison with the remaining CC398 isolates (Figure 1C). The t899 isolates had 10 mutations compared to the root (au200-2, au200-3, au201-1, au201-2, au201-3, au202-1, au202-2, au202-3, au202-4 and au202-5), which were located on the isolates chromosomes within a region of ≥111,139 bp (between 23,209 -134,348) (Table S2). This finding is in agreement with a study based on whole genome sequencing, which suggested that CC398 with spa type t899 had acquired a fragment of 123,000 bp from ST9 through horizontal gene transfer. This fragment included the spa gene and the SCCmec insertion site [28].

The MST revealed six main clades (A to F) within CC398 (Figure 1) (mutations defining each clade are listed in table S2). Mapping the 14 spa types and the SCCmec types onto the MST revealed that clade (B) consisted of isolates (n = 5) from different countries that shared the same spa type (t108), and four of them carried SCCmec type V (Figure 1C & D). Similarly, clades (A, E, and F) were composed entirely of isolates characterised by spa type t034, with the exception of one isolate within clade (F), which was represented by spa type t011 (Figure 1C). Furthermore, the Bayesian tip-association significance test (BaTS) [43] revealed that certain spa types (t034, t011, t571, t108, t1457, and t899), and SCCmec types (IV, and V) were significantly associated with phylogeny (p<0.01; Table S3). Nevertheless, our findings confirm that the usage of spa sequencing as the single typing tool for S. aureus might occasionally lead to misinterpretation, which is in agreement with previous observations in similar studies of other clonal complexes of S. aureus [28], [41], [44], [45].

Our analysis included CC398 isolates from 11 host species. A Bayesian statistical test (BaTS analysis; [43]) indicated that among these isolates, those from turkey meat and those from horses each displayed a significant association with phylogenetic structure within CC398 (p<0.01; Table S3). In contrast, other host species including humans were more dispersed on the phylogenetic tree (Figure 1A), not significantly different from a random distribution (p>0.1; Table S3).

The number of turkey isolates in our collection was very small (n = 4), and their geographic origins are not known with precision. Interestingly, however, we detected the φAvβ prophage in these four isolates by using targeted PCR (Table S1). This prophage was previously reported from S. aureus from several other bird species, suggesting CC398 in turkey may have adapted to the host through acquisition of an avian-specific prophage, similar to S. aureus CC5 in industrially fattened broiler chickens [46].

It is remarkable that the majority of isolates from horses under veterinary care clustered in clade (C) (41 out of 53 equine isolates total; Figure 1A). Of note, human isolates in the same clade (n = 6; 07-00334, 07-00471, 07-01238, 07-01239, 07-01335, 07-01730) were from veterinary personnel of an equine clinic in a large Austrian veterinary hospital (Stationary Care 1) who had close contact to infected horses (Table S1) [31]. Isolates in clade (C) (n = 53) had been collected from four different countries (Austria, Belgium, Germany, and the Netherlands) between 2006 and 2011 (Figure 1E), and clade (C) isolates from Germany (n = 29) had been collected from 13 equine clinics and veterinary practitioners distributed over seven different federal states (Table S1). These findings imply that clade (C) within CC398 is disseminated among hospitalised horses and veterinary personnel all over Germany and in several neighbouring European countries. At the same time, we found that clade (C) was extremely rare among S. aureus isolates from human infections in Germany. Among >6,700 isolates that had been submitted to the German National Reference Centre for Staphylococci and Enterococci in Robert Koch Institute between 2010 and 2011, there were 48 MRSA from human infections that displayed spa type t011 (Table S4). Among these, only four isolates carried the synonymous base substitution that defines clade (C) (i. e., they carried a thymidine residue at genomic position 2,533,404; SNP au309-2; Tables S2, S4), as revealed by targeted PCR and sequencing. Hence, the association of clade (C) with infections in horses is highly significant (p<0.0001; chi² test).

We assume that the emergence of MRSA CC398 clade (C) in horses from different European equine clinics may be due to epidemic spread, possibly comparable to several epidemic MRSA strains that rapidly spread within and between medical care hospitals and cause a large number of health-care-associated infections in humans [47]–[49]. A previous study based on multilocus sequence typing (MLST), spa-typing, and SCCmec-typing demonstrated that MRSA-CC398-t011-IV caused nosocomial infections in horses in an equine clinic in Switzerland [36]. The authors reported that MRSA-CC398-t011-IV was first detected in one of the personnel members who formerly worked in an equine clinic in Belgium. Later, this CC398-t011-IV was detected in infected horses and subsequently replaced ST1-t2863, which was prevalent in wound infections in this equine clinic [36]. While samples from Switzerland were not available to us, it is well possible that the strain in this clinic was affiliated to clade (C), since the majority of clade (C) isolates in our collection also displayed spa type t011 (95%) and SCCmecIV (91%), respectively (Table S1, Figure S1).

One possible explanation for the spread of this CC398 sub-clone may be insufficient hygiene practices in veterinary settings; however, this requires further research. Several studies have reported that the nasal carriage rate of MRSA among veterinary practitioners is much greater than in medical staff in human hospitals [50]–[54]. In addition, the nosocomial spread of MRSA in equine clinics and between veterinary personnel was previously demonstrated [55]–[59]. Hence, personnel in veterinary settings may play an important role in the introduction and spread of MRSA into equine clinics. In addition, humans with frequent contact with horses can represent a reservoir for MRSA and subsequently transmit it to their household. A metapopulation model demonstrated that the occurrence of a relatively large proportion of MRSA-CC398 carriers among a susceptible human population might result in an outbreak [60].

Of note, the association of equine origin with a phylogenetic clade within CC398 observed here does not immediately indicate any specific genetic adaptation. Such adaptation is difficult to detect in general. Even for healthcare-associated MRSA, which have been studied for decades and for which abundant genome sequence data is available, it has proven extremely difficult to identify the specific adaptive traits that render these strains successful [47], [61], [62]. Notably, clade (C) contained isolates from other hosts (e.g. human (Stationary Care 1); calf, dog, and pig (Farm 1)) who had been in contact with horses, suggesting that genetic specialisation to the equine host may be limited or lacking.

Although our collection of S. aureus CC398 isolates represents the broadest host species coverage studied to date, its composition is fragmentary with respect to both, the spatial and temporal coverage of the global population of CC398. While we have taken considerable efforts to assemble a broadly representative strain collection, it includes convenience isolates that a limited number of collaborating microbiologists had considered worth to be preserved over years for various reasons. For example, even though our equine isolates had been collected in several European countries between 2006 and 2011, they by no means represent the demographics of the underlying horse population, let alone that of the more widely distributed population of S. aureus CC398. Several other categories (country of origin, host species) contained very few samples (e. g. only one isolate from Canada, one from a cat, etc.). In addition, very limited clinical and other meta-data was available for many of the isolates, because they initially had not been stored with the goal of any global epidemiological inferences in mind. Hence, for an in-depth investigation of the distribution of CC398 among different host species, it would be highly desirable to extend this study by including additional isolates from each of the various hosts, with an even distribution over several years and over a large geographic area, and by systematically recording epidemiological data.

The dHPLC-based mutation discovery method applied here covered 1.4% of the CC398 genome. This approach delivered improved discriminatory power compared to spa typing and standard MLST [44], [63], and provided some novel insights into S. aureus population structure. However, the resolution of analyses and the strength of any inferences would be much improved by whole-genome sequencing [42], [47].

In the present study, a collection of 195 S. aureus CC398 (MSSA; n = 37 and MRSA; n = 158) isolates was investigated (Table S1) and some of these isolates were included in previous studies [26], [28], [31], [51], [66]–[69]. CC398 convenience isolates were collected from nine different countries (mainly from Europe), and various hosts (humans: n = 80; animals: n = 115). The isolates investigated in this study were selected by animal species, geographical origin, and approximate period of time. Veterinary care facilities in this study were divided into stationary care (where the animals must be hospitalized for at least one night in order to receive medical treatment) or ambulatory care (medical care is provided to animals without being admitted to a hospital for treatment). MRSA isolates were chosen as follows:

[1] 10 isolates from horses were collected in Austria/Vienna. Eight of these isolates were from infected horses treated in Vienna veterinary hospital (Stationary Care 1; Table S1) from 2006 until 2007. We had previously collected and investigated isolates from nasal colonization of the veterinary personnel of this hospital, due to the emergence of CC398 over a long period in this facility, [31]. These human isolates (n = 6) were also included.

[2] 37 clinical isolates from horse were collected in Germany, from 17 different veterinary facilities (3 stationary care, 14 ambulatory care), which were distributed over 6 different German federal states (Baden-Württemberg (1), Hesse (1), Lower Saxony (3), North-Rhine-Westphalia (9), Schleswig-Holstein (2), Saxony (2), and Saarland (1)). Most of these horse isolates were sent for typing to the German Reference Centre for Staphylococci and Enterococci in Robert Koch Institute - branch Wernigerode by the Labor Dr. Boese which is providing diagnostic service for veterinarians treating horses in all the German federal states. Isolates from other animal species from Germany were also included, which originated from nasal colonization in pigs, pig farmers and their family members; colonization of posterior nares of goose, broiler chicken carcasses (thawing liquid). Isolates causing mastitis in cattle as well as various kinds of infections in humans emerging at different geographical locations in Germany were included as well (Table S1).

[3] Finally, isolates from other European countries (e.g. Belgium, Denmark, Italy, the Netherlands, UK) as well as from overseas (Canada, Thailand, USA) were included to maximize geographic distribution and range of host species. Additionally, to monitor the dissemination of one particular S. aureus strain among different animal species, six isolates from a Dutch farm derived from horses, dog, and cattle were included (farm 1 in Table S1).

SCCmec- and spa-typing were performed for all isolates as previously described [70]. Briefly, spa-typing was performed by following the Ridom Staph Type standard protocol and the spa-types were assigned to the Ridom database (www.ridom.org) (Ridom GmbH, Würzburg, Germany). In addition, antimicrobial susceptibility was tested using the broth dilution method according to the DIN58940 guidelines.

In this study, we investigated mainly metabolic housekeeping genes because polymorphisms in these genes provide the most reliable phylogenetic markers [71]. In total, we investigated 97 genetic housekeeping loci, which made up 1.4% (40,230 bp) of the CC398 genome and were scattered over the core genome of CC398. These loci had been analysed previously to investigate the population structure of other clonal complexes of S. aureus [41], [44], [72]. PCR primers were used to amplify 97 genetic housekeeping loci distributed along the 195 S. aureus isolate chromosomes (Table S2). Mutation discovery for the amplified gene fragments was performed using dHPLC (WaveR Nucleic Acid Fragment Analysis System, Transgenomic, Inc., Omaha, NE, USA) as described previously [44], [46], [72], [73]. Briefly, PCR amplicon from each isolate was compared to a reference strain for detecting the heteroduplexes. Heteroduplexes result in double-stranded DNA that contains a point mutation site in comparison to the reference strain. Identified SNPs were confirmed through capillary Sanger sequencing of the PCR products from both ends using the PCR primers which are listed in Table S2.

For identification of phages possessing integrase group φSa3, we performed PCR using the primers int3, f2: 5′GTCAGCTTTAGATGACGC and int3, r2: 5′AGCGCTAATGATGAACGA according to NC_00227452. For PCR demonstration of sak, chp and scn, we followed the protocol as described previously [74]. The presence of φAvβ prophage was determined by PCR as previously described [46].

Based on the discovered SNPs within the 97 genetic loci, a minimum spanning tree was constructed using Bionumerics software version 6.5 (Applied Maths, Ghent, Belgium). Additionally, sequences from the 97 housekeeping genes were concatenated for each isolate, constituting a 40,230 bp sequence alignment. A maximum likelihood tree based on this alignment was assembled using PhyML 3.1 [75]. The ancestral node was distinct by including distantly associated S. aureus genomic sequences. DnaSP was used to estimate the nucleotide diversity (π) and nucleotide variation (θw) and for calculating the mean pair-wise distance between alleles at synonymous (Ks) and non-synonymous (Ka) sites [76]–[78]. The rate of evolution and the divergence times were estimated as described previously using BEAST software (Version 1.7.5, http://beast.bio.ed.ac.uk/) [79]. The Bayesian tip-association significance test (BaTS, version 1.0) was applied to estimates of the association of the phylogeny traits with hosts, spa types, geographical origin, and SCCmec types [43].

Statistical significance of the association between SNP 309-2 and the host species was assessed using a chi-square test (http://www.r-project.org/).