A detailed investigation of the porcine skin and nose microbiome using universal and Staphylococcus specific primers

MRSA is an increasing problem in humans as well as livestock. The bacterial co-colonization of the skin in MRSA carriers has been poorly investigated and moreover, there have been no methods for high resolution investigations of the Staphylococcus genus apart from tediously culturing or doing multiple PCRs. On 120 samples from pig ear, skin and nose, we generated amplicons from the V1-V2 region of the 16S rRNA gene to gather an overview of the genus-level microbiome, along with using MRSA specific plates to count MRSA. In parallel with this, amplicons of the tuf gene were generated, targeting only a region of the tuf gene found only in the Staphylococcus genus. Using these methods, we determined a core microbiota across the healthy pig and determined the Staphylococcus genus to be dominated by S. equorum. Moreover, we found Streptococcus to be inversely associated with Staphylococcus and MRSA, suggesting a role for this genus in combating MRSA. In this work, we have thoroughly investigated the skin and nose microbiome of the pig and developed a high throughput method for profiling the Staphylococcus genus which we believe will be useful for further investigations.

depth resolution of the different staphylococcal species which, as pointed out by the authors, could be necessary to gain information regarding the potential protective effect against MRSA.
The use of the 16S rRNA gene is routinely used for profiling microbiotas due to its ubiquity and discriminatory power, but other genes, such as the Elongation Factor Thermo unstable (EF-Tu) gene (tuf), may be useful for more targeted investigations. This gene universally conserved in bacteria, and codes for a protein that binds tRNA in the cytoplasm and mediates entry into the ribosome 16 . It has previously been used to distinguish Lactobacillus and Bifidobacterium 17 as well as Staphylococcus 18,19 , but not yet in a high-throughput context.
In this study, we aimed to investigate the staphylococcal community in high resolution to identify which species dominate the pig nasal and skin microbiota and if any are associated with MRSA. As part of the instigation we have developed a high-throughput method based on selective primers targeting the tuf gene, e.g. an extension of the tuf-based classification used previously 18,19 . This enables one to achieve a large number of reads within the Staphyloccocus-genus, i.e. a large quantum of information, even at very low abundance. Swabs from clinical mink samples were included in the study as verification of the versatility of our approach. For cross-host verification, we included samples from both mink and pigs, as the major staphylococcal pathogens in mink are separate from the ones in pigs.
There were significant differences between sites according to multivariate analysis (P < 0.001 for ANOSIM, Adonis and PERMANOVA), as indicated in Figs 1 and 2. Pairwise comparisons using PERMANOVA showed that the ventral and dorsal surfaces were not significantly different, whereas the nose and the ear was significantly different from all other sites, the nose being the most dissimilar site (Figs 1 and 2). Canonical analysis of principal coordinates revealed that the genera separating the nose from the other sites were higher levels of Streptococcus, Rothia and Globicatella and lower levels of Aerococcus and Facklamia, whereas the ear was separated by higher levels of Staphylococcus and lower levels of Facklamia and Streptococcus.
In univariate analysis, the nose uniquely harbored Rothia and had higher levels of Streptococcus and Moraxella, while levels of Aerococcus, Facklamia and Jeotgalibaca were lower (Fig. 3). The ears had the highest levels of Staphylococcus, but correspondingly lower levels of Streptococcus and Prevotella. The dorsal and ventral skin surface had higher levels of bacteria normally associated with the gut, such as Prevotella, Bacteroides and Enterococcus, although the dorsal surface was higher in Staphylococcus. The species within the Staphylococcus genus mapped almost exclusively to S. equorum or were unclassifiable beyond the genus level.
The Shannon index, a measure of diversity, was not different across groups (P = 0.95). The Streptococcus genus was inversely correlated with Staphylococcus (pearsons ρ = −0.38, P < 0.001).  Supplementary Table S1). Using 16 S rRNA gene amplicons on the positive control did not find S. delphini, S. xylosus or S. sciuri and incorrectly found S. saprophyticus. 16S rRNA gene amplicons did however estimate more realistic proportions of the species herein, but when the Staphylococcus genus is a minority constituent, the species differentiation is low or non-existent.
In the animal samples, 31,508 reads were available for classification per sample on average, and 95-99% of these reads were classifiable to an unambiguous single species using blastn. Several Staphylococcus species were found in all samples (Figs 4 and 5), but all were heavily dominated by S. equorum followed by varying proportions of most species included in the positive control. S. aureus was detected in all samples. Overall, the ear was the most dissimilar site (Fig. 6), as S. schleiferi, S. microti, S. simulans, S. hominis, S. lentus, S. sciuri and S. succinus were relatively lowered here. Moreover, the ventral skin surface was lower in S. equorum, and correspondingly higher in S. agnetis, S. simulans and S. sciuri. Using a qPCR targeting the S. equorum SodA gene, the presence of S. equorum was confirmed in all samples. There was a linear correlation between qPCR estimates of S. equorum (pearsons ρ = 0.35, P < 0.001), although several ear samples were poorly described by this relation. Omission of ear samples showed a substantially higher correlation (pearsons ρ = 0.75, P < 0.001).
MRSA levels on pig and mink. The presence of MRSA was assayed using MRSA selective plates. MRSA was present in all samples with levels around 10 3 cfu/sample in ear and skin swab samples, whereas slightly lower levels, generally between 10 2 and 10 3 cfu/sample were recorded in nasal swab samples (Fig. 7A). MRSA plate counts was negatively correlated with Streptococcus (pearsons ρ = −0.42, P < 0.001) (Fig. 7B), but poorly related to total Staphylococcus, possibly reflecting that most of the staphylococci are S. equorom. We did not find any large correlations (pearsons ρ > 0.2) between the bacteria found using the tuf-primers and MRSA counts.
For mink, included for cross-validation of the used methods, the V1-V2 primers showed a diverse group of bacterial families, including Staphylococcus. The staphylococci in these V1-V2 samples were not classifiable at a species level or classified mainly as S. pseudintermedius. The tuf primers generated amplicons in most of the samples, mainly classified as S. delphini which is in agreement with classification by MALDI-TOF (data not shown). S. delphini was not detected using V1-V2 primers.

Discussion
In this set of data, the microbiota of the nose, the outer ear and two sites of the skin were investigated using three different methods in order to investigate associations between the microbiota, MRSA and individual species of the Staphylococcus genus.
The overall skin and nasal microbiota was rich in Aerococcus, Staphylococcus, Lactobacillus, Streptococcus, Facklamia and Rothia (nose only), which is in some contrast to previous studies where bacteria belonging to    the phylum Proteobacteria have dominated. One paper found a high abundance of the phylum Proteobacteria (including Moraxella, Psychrobacter and Pseudomonas) in the nose, whereas the present results are much higher in Firmicutes. The authors also found negative associations with MRSA for Lactobacillus and Staphylococcus, and proposed a more thorough investigation using specific primers 15 . A study looking at the tonsil microbiota of pigs also found high levels of Proteobacteria, especially Pasteurella, and an overall composition substantially different than in the present data, possibly explained by the different body site investigated 20 . A more recent paper reported high nasal levels of Gammaproteobacteria, mainly mapped to the genus Moraxella, which we also found abundantly in the nose. Compared to the present data, the paper also found similar levels of Lactobacillus and Staphylococcus 21 . Moraxella is well known as an inhabitant of mucosal surfaces 22 , and consequently, its dominance in nasal samples is not unexpected. Swe et al. 23 investigated the microbiota of the ear thoroughly and found high levels of Streptococcus, Lactobacillus and Corynebacterium, but less Aerococcus and in contrast to our results, Rothia was found in the ear. Rothia nasimurium is a gram positive coccoid bacterium that was first described from the nose of healthy mice 24 , but it is found frequently in nasal swab samples from healthy pigs in our lab. It grows on MRSA2 agar with colonies similar to MRSA and thus constitutes a differential diagnostic issue. Regarding the dorsal and ventral skin of the animal, this has been investigated recently in a study by McIntyre et al. 25 , where high levels of Firmicutes, including Ruminocococeae, Lachnospiraceae, Streptococcus and Prevotellaceae were found and similar compositions were seen in the ventral and dorsal sites. Espinosa-Gongora et al. 14 specifically investigated the nasal microbiota of MRSA carrier pigs vs. non-carriers, and apart from a core microbiota dominated by Proteobacteria and Firmicutes, species within Lachnospiraceae and Leuconostoc were shown to be differentially abundant in these animals. A differential analysis of carriers vs. non-carriers could not be carried out in the present data, as all animals were positive for MRSA in agreement with estimates of MRSA prevalence in Danish pigs 10 . The inverse relation of Streptococcus to MRSA and Staphylococcus observed in our data has previously been reported in Bessesen et al. 26 where a significant negative relation between Streptococcus mitis and MRSA was found in human samples through 16S rRNA gene sequencing and confirmed with in vitro inhibition assays. In future work, individual strains of Streptococcus from the samples of interest should be isolated and assayed for possible antagonistic effects towards MRSA.
MRSA specific plates were used to enumerate MRSA as there currently is no reliable molecular technique to quantify MRSA in mixed samples. The use of plates unfortunately is sensitive to the load of bacteria in the samples, which may explain why MRSA loads were lower in the nose.
We sought to further elucidate the composition of the staphylococcal genera by the use of primers specific for the tuf gene, an approach already considered for Staphylococcus by previous papers 18,19 . The primers used by Martineu et al. 18 were chosen for a metataxonomic approach as their product length was better suited for the 2 × 250bp MiSeq platform. The primers were entirely specific for staphylococci and were successful in amplifying all species tested, albeit with differing specificity. In contrast to non-specific primers, such as for the 16 S rRNA gene, the tuf primers will selectively amplify and provide resolution in samples very low in staphylococci, even to the degree where these are undetectable by 16S rRNA gene primers. The use of tuf primers also allows detection S. delphini as in mink. The primers cannot, though, distinguish between MRSA or methicillin sensitive S. aureus, as these differentiated by the presence of a mecA cassette and not a variation in the tuf gene. Inspection of the tuf gene of the negatively biased species revealed various levels of mismatch in the primers, e.g. 2 out of 5 sequences  classified as S. haemolyticus in the database generated from NCBI had a single mismatch in both the forward and reverse primer. On the other hand, all 5 sequences classified as S. saphrolyticus were perfectly matched by both primers, although these were biased against in the sequencing. The reasons for the positive bias of S. equorum are less clear, as the region of the primer binding site of other negatively biased species such as S. cohnii or unbiased species such as S. aureus were identical to those in S. equorum. Ideally, the primers would amplify all staphylococcal species equally, which will require further development. Alternative tuf primers or use of the SodA gene as in Blaiotta et al. 27 should be investigated. Further degeneration of the primers could be beneficial to encompass all species, although it is doubtful if all bias can be completely removed. The issue of different sequences classified as the same species (e.g. as for S. haemolyticus) further complicates intelligent primers design.  In the pigs, we observed very high levels of S. equorum, a species first found on the skin of healthy horses 28 and described in the nose of pigs 29 , followed by S. schleiferi, S. cohnii and S. microti, as well as S. aureus in all samples, whereas the mink samples were dominated by S. delphini. The high prevalence of S. equorum, a species hitherto not having received much attention, was further investigated with a specific qPCR, showing that all samples were positive for S. equorum and that the relative abundance estimated from sequencing was in agreement with the relative abundance estimated by qPCR.
In conclusion, we have conducted a thorough investigation of the skin, nose and ear microbiota of the pig and implemented a high resolution method for elucidation of the Staphylococcus genus. The use of tuf-specific primers, despite not being entirely quantitative, allows for detection of individual Staphylococcus-species even in very low abundance, making future investigation of staphycoccal populations more straightforward.
We found that the microbiota of the nose differed significantly from the microbiota at the skin sites. The microbiome in all sites was dominated by Aerococcus, Streptococcus, Lactobacillus, Facklamia, Rothia and Staphylococcus, whereas the nose was enriched with Streptococcus and Moraxella and uniquely harboring Rothia. The staphylococcal population was heavily dominated by S. equorum followed by S. schleiferi, S. cohnii and S. microti, as well as S. aureus. In the nose, the level of MRSA was negatively correlated with the level of Streptococcus which makes Streptococcus a possible target of interest regarding manipulation of the natural microbiota towards an anti-MRSA environment.

Materials and Methods
Sample collection. Pig samples were obtained at a commercial slaughterhouse (Danish Crown, Ringsted, Denmark), immediately following controlled atmosphere stunning by CO 2 but prior to euthanasia. Each animal (n = 30) was sampled behind one ear, in the nose and on the ventral and dorsal skin surface by use of individual ESwabs (Copan Diagnostics Inc., Murrieta, CA, USA) for a total of 120 samples. The ear was sampled by three gliding motions behind the pinna, the nose by a rotary motion ~1 cm inside each of the nares and on the two skin surfaces by three consecutive ~3 cm strokes in the same location. All handling of animals was in accordance with regulations from the Danish Ministry of Justice.
Mink samples, some positive and some negative for MRSA were acquired from routine diagnostics of clinical mink sent to DTU-VET. Swab samples from throat were obtained as described by Hansen et al. 2 .
A Staphylococcus control was made by a mixing cultures of S. aureus (both resistant and sensitive to methicillin), S. cohnii, S. delphini, S. equorum, S. epidermidis, S. haemolyticus, S. hyicus, S. schleiferi, S. sciuri, S. succinus, and S. xylosus equally, which subsequently was subjected to the same purification, PCR-amplification and sequencing procedure as the pig and mink samples.
MRSA enumeration by culturing. Samples were processed within 2 hours of sampling and were not centrifuged or filtered due to concerns regarding flocculation of staphylococci and/or adhesion to skin cells and dust particles. For direct quantification of MRSA, 100 µL sample material, undiluted and in 10 −1 dilution, was plated on MRSA-selective plates (Brilliance MRSA2 agar, Oxoid, Basingstoke, UK). Plates were incubated at 37 °C overnight and suspected MRSA colonies were counted after 18-24 h. One colony from each positive sample was confirmed as MRSA by MALDI-TOF and PCR detection of the mecA and nuc genes 30 .
Microbiota analysis and preparations. DNA from each sample was purified with a Maxwell ® LEV Blood DNA Purification Kit (Promega Corporation, Madison, WI, USA). First, the swab was vigorously shaken at high speed to loosen the cells from the swab. The whole sample was then transferred to a 2 ml Eppendorf tube and centrifuged for 15 min at 20.000 g. The supernatant was removed and the pellet was incubated for 60 min at 37 °C with 100 µl of lysozyme mixture (20 mM Tris-HCl (pH 8), 2 mM EDTA, 1.2% Triton X, 200 µg/ml lysostaphin and 25 mg/ml lysozyme). Subsequently, the samples were mixed with 350 µl lysis buffer, and one 5 mm stainless steel bead (Qiagen GmbH, Hilden, Germany) was added to the samples followed by shaking on a Qiagen TissueLyser II (Qiagen GmbH, Hilden, Germany) for 2 min at 20 Hz. Samples were then incubated for 1 h at 56 °C with 30 µl Specificity of the primers were tested by PCR on pure culture of Rothia nasimurium, Corynebacterium sp., Clostridium perfringens, Propionibacterium acnes, Streptococcus suis, Escherichia coli and Lactobacillus casei, all of which were negative. The resulting PCR products for both primer sets, included negative controls, were then analyzed on an Agilent 2100 Bioanalyzer using an AgilentDNA1000 kit (Agilent Technologies, Waldbronn, Germany) and further pooled in equimolar ratios (50 ng per barcoded sample). The pooled DNA was then purified of primers and detergents using a Qiagen MinElute PCR purification kit (Qiagen GmbH, Hilden, Germany) according to the manufacturer's instructions. Amplicons were submitted to The National High-Throughput DNA Sequencing Centre at the University of Copenhagen, Denmark, for sequencing on an Illumina MiSeq 250PE platform. The V1-V2 amplicons were then merged, quality filtered, chimera-checked and mapped against the RDP-II Scientific REPoRTS | (2018) 8:12751 | DOI:10.1038/s41598-018-30689-y SSU database using the BION-meta software (Danish Genome Institute, Aarhus, Denmark). The tuf gene amplicons were merged and quality filtered using BION-meta, clustered at 97% similarity with USEARCH 32 , de-novo chimera checked with UCHIME 33 and taxonomically assigned using the blastn algorithm from command line BLAST with a custom made tuf database, using a word length of 22 and a minimum similarity of 90%. An alternative workflow using only USEARCH was also investigated, but was abandoned due to lower sensitivity. The tuf database was built by downloading all Staphylococcus genomes being either classified as "Complete genome" or "Scaffold" from NCBI Genbank (as indexed in ftp://ftp.ncbi.nlm.nih.gov/genomes/genbank/bacteria/assembly_ summary.txt) followed by extraction of all CDS annotated as tuf genes using an in-house Perl-and bash-based script (available upon request). The tuf genes were then checked for correct length (between 1150 bp and 1200 bp) and used as a database for blastn. The tuf gene from S. delphini was added manually.
An S. equorum specific qPCR targeting the S. equorum sodA gene was carried out as described in Blaiotta et al. 27 with the addition of QuantiTect SYBR Green (Qiagen, Hilden, Germany) using 25 ng DNA for each reaction. A standard curve of serial dilutions of S. equorum DNA was used to ensure amplification efficiency. The estimates of S. equorum from the qPCR, expressed in S. equorum DNA per 25 ng total DNA, was compared to the estimates of total Staphylococcus from 16S rRNA sequencing to further investigate the levels of S. equorum. Statistical analysis. All samples were normalized to 100,000 reads before further analysis and data is presented in its entirety when plotted. To avoid issues of normality, individual OTUs were compared using Kruskal-Wallis followed by Conover-Iman test if significant as implemented in the agricolae package 34 . The Shannon-index was used to estimate diversity and was tested with ANOVA. Multivariate patterns were visualized by principal coordinates analysis (PCoA) and tested for groupwise differences with ANOSIM, Adonis and PERMANOVA from the vegan package 35 , all using Bray-Curtis distances. Canonical analysis of principal coordinates was used to evaluate bacteria of interest. Correlations are calculated and plotted as log10 using Pearson correlations. P-values below 0.05 were considered significant, and were adjusted for multiple comparisons using Sidak-corrections 36 when doing multiple univariate analyses. All tests were two-tailed.
Ethics Approval and Consent To Participate. All handling of animals was in accordance with regulations from the Danish Ministry of Justice.