In vivo gene expression in a Staphylococcus aureus prosthetic joint infection characterized by RNA sequencing and metabolomics: a pilot study

Staphylococcus aureus gene expression has been sparsely studied in deep-sited infections in humans. Here, we characterized the staphylococcal transcriptome in vivo and the joint fluid metabolome in a prosthetic joint infection with an acute presentation using deep RNA sequencing and nuclear magnetic resonance spectroscopy, respectively. We compared our findings with the genome, transcriptome and metabolome of the S. aureus joint fluid isolate grown in vitro. From the transcriptome analysis we found increased expression of siderophore synthesis genes and multiple known virulence genes. The regulatory pattern of catabolic pathway genes indicated that the bacterial infection was sustained on amino acids, glycans and nucleosides. Upregulation of fermentation genes and the presence of ethanol in joint fluid indicated severe oxygen limitation in vivo. This single case study highlights the capacity of combined transcriptome and metabolome analyses for elucidating the pathogenesis of prosthetic infections of major clinical importance.


Background
Staphylococcus aureus is one of the leading causes of community-and hospital-acquired infections worldwide. The clinical spectrum ranges from superficial skin lesions to deep-sited or generalized infections. Besides acute infections, S. aureus can adapt to a biofilm mode of growth in response to certain environmental cues and thereby infections become persistent and recurrent, particularly in association with prosthetic implants [1]. Moreover, the emergence and spread of resistance to many classes of antibiotics pose an increasing threat to public health. Consequently, staphylococci have been studied extensively both in vitro and in vivo with special focus on resistance and virulence. An arsenal of virulence factors has been identified including toxins, cell surfaces proteins that facilitate attachment and colonization, and factors that contribute to immune evasion and tissue damage [2]. However, few studies have investigated nutrient acquisition and metabolism of S. aureus in vivo during infection, which is an important aspect of S. aureus pathophysiology.
Recently, the increasing number of genome sequences of S. aureus have provided deeper insights into its virulence, antibiotic resistance and physiology in general [3]. It is recognized that the success of S. aureus depends not only on its virulence genes and development of antibiotic resistance, but also on a coordinated and timely expression of genes upon infection of its host. To elucidate this complicated orchestration of gene expression, the transcriptome has been studied in vitro and in vivo using rabbit [4] and mouse [5,6] infection models. However, pathogens are likely to make host-specific adaptions by altering gene expression, which necessitates studies in humans. To our knowledge, Date et al. [6] is the only published investigation of the transcriptome of S. aureus in humans with cutaneous infections caused by the methicillin-resistant USA300 strain.
The aim of this study was to compare the in vivo expression of virulence and metabolic genes of S. aureus in a prosthetic joint infection in a human subject with growth in vitro as reference using RNA sequencing (RNA-seq). Moreover, using nuclear magnetic resonance (NMR) spectroscopy we analyzed the metabolites in the joint fluid and in culture supernatants in order to determine the biochemical composition of the environments.

Results and discussion
S. aureus infection: culture, genome and transcriptome Standard culture of joint fluid, tissue biopsies, and prosthesis components revealed a pure growth of S. aureus with a pansusceptible antibiogram (see case history in Methods). Amplicon sequencing was used for detection of bacteria in fluid obtained by sonication of prosthesis components (all joint fluid was used for RNA-seq). Approximately 44000 reads were obtained, all of which were clustered into operational taxonomic units (OTUs) identified as S. aureus (data not shown).
The joint infection had an acute presentation although a previous indolent period cannot be precluded (see case history). Assuming an acute infection [7], we chose to compare gene expression of the in vivo sample with the isolate in an exponential growth phase. Additionally, we sequenced the genome of the isolate (SAU060112) to gain insight into the virulence and antibiotic resistance capacity and to facilitate high fidelity RNA-seq read mapping.
To reconstruct the genome 17.8 million reads were generated. The assembly resulted in 17 contigs with an average coverage of 729 and N50 of 601492 bp. The total length of contigs was 2.68 Mb which is close to the average (2.86 Mb) gapless chromosome length of S. aureus (currently 66 strains in total available at NCBI, May 2015). No plasmids were found. The genome assembly is predicted to contain 2562 protein-coding genes. Details of the assembly and analysis of the COG classification distribution of the protein-coding sequences can be found in Additional file 1: Tables S1, Additional file 2: Table S2 and Additional file 3: Figure S1. The isolate was spa type t908 and belonging to Clonal Complex 45. Interestingly, according to Driebe et al. [8] CC45 show less homoplasy density than other S. aureus clades indicating little recombination with other clonal complexes. Furthermore, in contrast to other CC45 strains included in the study, but similar to USA600-BAA1754 (spa type t671), the entererotoxin genes entC, sel and sen is present in SAU060112. We thus believe that within the CC45 complex, SAU060112 is relative closely related to USA600-BAA1754 despite the different spa type. Approximately 25 and 350 million RNA-seq reads were obtained for in vitro cultures and the in vivo sample, respectively (Table 1). Between 26.8 and 37.8 % of total reads from the in vitro cultures were mapped to the protein-coding sequences of the genome with the mapping criteria employed (95 % similarity, 80 % length fragment). Relaxation of the mapping criteria led to increased mapping efficiency (data not shown), however, this also increased the risk of erroneous mapping of human host transcripts to the bacterial genome. Thus, this conservative approach was chosen for all samples. As expected, the majority of the sequences from the in vivo sample originated from the human host, and only 1.2 % (4.1 million) reads were mapped to the S. aureus genome and 0.086 % (0.3 million) to the protein-coding sequences. While 0.3 million reads might be considered a relative low number of reads compared to modern RNA-Seq studies that frequently have many millions of reads per sample, it is still expected to be enough to detect reads from about 85 % of bacterial genes according to [9]. It is possible that other methods of purification of bacterial RNA from background host RNA than the one we employed can yield a higher proportion of bacterial RNA. A total of 430 genes (17 % of total) were found to be differentially expressed, of which 317 were upregulated and 113 downregulated in vivo. The complete list of differentially expressed genes is available in Additional file 4: Complete list of differentially expressed genes.

Antibiotic resistance genes
SAU060112 was susceptible to β-lactams (including penicillin and methicillin) and 5 additional antibiotic classes. Analysis of the genome by the Resistance Gene Identifier (RGI) at the Comprehensive Antibiotic Research Database [10] predicted absence of resistance genes to β-lactams, macrolides and aminoglycosides in accordance with the antibiogram, but identified several efflux pumps related to other antibiotics (Additional file 5: Figure S2). Some of the efflux pumps (tet38 40-fold, p-val = 8.2*10 −40 ; mepA 7-fold, p-val = 2.5*10 −5 ) and cell wall biosynthesis genes (mgt 12-fold, p-val = 4.0*10 −10 ; pbp2 5-fold, p-val = 0.0027; murZ 8-fold, p-val = 5.8*10 −6 ) had increased expression in vivo, possibly induced by antibiotic treatment received by the patient for two days. The peptidoglycan biosynthesis pathway has been shown to be upregulated in S. aureus treated with subinhibitory doses of cell wall active antibiotics [11,12]. However, several studies [12][13][14] have shown responses in bacteria is a global process not only involving proteins directly affected by antibiotics, but also proteins with no apparent relationship to the antibiotics. Therefore, it is unknown to which extent the differentially regulated genes found in this study was induced by antibiotic treatment or an in vivo response.
Notably, the expression of 32 virulence genes present in the genome was negligible (≤5 reads/100,000 mapped mRNA reads) in vivo. Also, 9 virulence genes were found downregulated including transcription regulator sarS (13-fold, p-val = 6.1*10 −6 ) [25], immunoglobulin G-binding protein A (spa) (6-fold, p-val = 0.005), and six of eight genes in the putative ESAT-6-secretion system, while expression of these genes was reported unchanged in [6] ( Table 2). SarS belongs to the SarA protein family, global regulators of virulence gene expression in S. aureus [26]. SarS, which is controlled by many regulators, activates spa expression and represses α-hemolysin [18,27]. This correlates with the finding in this study of expression of spa being reduced while expression of α-hemolysin is increased (91-fold) in vivo. ESAT-6 proteins have been reported to be important for staphylococcal infection in mice, but their functions during human infection remain unclear [28].

Siderophores
In response to iron limitation, S. aureus has two known iron acquisition mechanisms: one is the iron-regulated surface determinant (isd) gene set that mediates heme acquisition from mammalian heme-containing proteins, and the other is a Fe(III)-siderophore acquisition system, which is capable of removing iron from human transferrin and lactoferrin. S. aureus produces two distinct siderophores: staphyloferrin A and staphyloferrin B [29]. The Ferric Uptake Regulator (Fur) controls expression of genes encoding all these systems [30], but mechanisms for fine-tuning of expression of these systems are unknown. We found 3-fold upregulation of fur (p-val = 0.004) during  in vivo infection, but no difference in expression of isd and staphyloferrin A genes. However, the sbn operon (locus SAU060112_20052 -20060) encoding staphyloferrin B was upregulated in vivo (3-to 27-fold, p-val = 6.9*10 −5 -2.3*10 −17 ) in this study. The ninth protein SbnI encoded by the sbn operon is recently found to play an important role in transcription control of the sbn operon [31]. Staphyloferrin B production has been found to be important for S. aureus growth in iron-limited medium and for its pathogenicity in a murine kidney abscess model [32]. In human cutaneous abscesses expression of both isd and sbn operons was elevated as well as two genes of the staphyloferrin A operon [6].

Metabolism
We observed upregulation of several genes related to anaerobic/hypoxic conditions, which include the genes involved in pyruvate to ethanol fermentation (pflB 23fold, p-val = 8.  (Fig. 1) and pyruvate formate-lyase-activating enzyme (pflA) (63-fold, p-val = 7.3*10 −10 ). The anaerobic/hypoxic condition was further supported by the high concentration of lactate (~40 mM) and presence of ethanol in the infected joint fluid (Fig. 2). The ADI operon was the most upregulated amino acid catabolic pathway in the current study as well as in human cutaneous abscesses [6] and chronic human and murine osteomyelitis [16]. This operon also includes arginine/ ornithine antiporter arcD, which is the only transporter for free arginine [33]. Arginine is utilized by S. aureus as a source of energy under anaerobic conditions [34]. We think that this pathway is essential for the direct production of ATP without generating organic acids under anaerobic conditions. This hypothesis is indirectly supported by the overexpression of the ethanol fermentation pathway. Under microaerophilic or anaerobic conditions, S. aureus ferments the majority of pyruvate to lactic acid in vitro [33]. However, lactic acid concentration was nearly 40 mM in the joint fluid (Fig. 2), which was higher than average lactate level in septic arthritides and probably was produced mainly by human host cells under hypoxic condition [35]. To avoid the unfavorable production of additional lactic acid while still oxidizing NADH to NAD + for continuation of glycolysis and ATP generation, genes promoting pyruvate fermentation to ethanol were upregulated instead.
Besides ADI, high expression of catabolic threonine dehydratase tdcB (52-fold, p-val = 2.5*10 −19 ), alanine dehydrogenase ald (24-fold, p-val = 1.3*10 −14 ) and several additional amino acids catabolic enzymes were observed (Fig. 1), while several genes involved in amino acid synthesis including tryptophan, arginine, cysteine and histidine were among the 113 downregulated genes in vivo. Moreover, NMR data showed high concentration of free amino acids in the infected joint fluid compared to LB culture (Fig. 2). Taken together, our data suggest that free amino acids were a major source of carbon and energy for S. aureus in vivo.
Besides amino acids, several genes involved in carbohydrate catabolism had increased expression in vivo, including N-acetylneuraminate lyase nanA (9-fold, p-val = 8.9*10 −7 ) and the lac operon (125-to 464-fold, p-val = 2.4*10 −43 -2.7*10 −23 ). The enzyme NanA catalyzes the cleavage of N-acetylneuraminic acid (Neu5Ac), which is the predominant sialic acid in humans and is present as a terminal sugar on a wide range of glycoproteins and glycolipids. Host glycoproteins can be used as nutrient for bacteria [36], for example, Streptococcus pneumoniae can utilize human glycoconjugates as the sole source of Fig. 1 Overexpressed metabolic pathways in the infection. Pathway names are according to the MetaCyc database. Each pathway is assigned with a specific color and the upregulated enzymes in each pathway are indicated. On the bottom, under each pathway fold change of the upregulated enzymes in the current study are listed in the second column while fold change of these enzymes in the human cutaneous abscesses study [6] are in the third column carbon for growth [37]. The increased expression of nanA is consistent with the higher concentration of Neu5Ac in the joint fluid than the in vitro supernatant where it was undetectable (Fig. 2). The S. aureus lac operon is inducible by galactose and suppressed by glucose [38]. The concentration of galactose in vivo was at the baseline level in the NMR spectra, hence, it is unknown to which extent galactose is used as a nutrient. The increased expression of purine and pyrimidine deoxyribonucleoside degradation pathways (deoA 5-fold, p-val = 0.0001; deoB 4-fold, p-val = 0.001; deoC 9-fold, p-val = 5.8*10 −8 and deoD 81-fold, p-val = 8.3*10 −10 ) indicated that the pathogen probably also acquired nucleosides as nutrients. The end products of these pathways are acetyl-CoA, a central metabolic intermediate, and D-glyceraldehyde-3-phosphate, an intermediate of glycolysis (Fig. 1). The metabolite measurement shows increased levels of nucleosides, particularly uracil, in vivo (Fig. 2). Uracil has been found elevated in joint fluid from rheumatoid arthritis patients [39]; however, the mechanism behind this is unknown.
Although the concentration of free amino acids, some glycans and nucleosides were higher in the joint fluid, the expression level of all hydrolytic exoenzymes but lipases remained low in vivo (Additional file 6: Table S3). This is in contrast to findings reported by Szafranska et al., who observed upregulation of many genes encoding secreted proteolytic enzymes in S. aureus during acute and chronic murine osteomyelitis [16]. A possible explanation for the low expression of hydrolytic exoenzymes in the current study is that hydrolysis of proteins and glycans might have been done by host enzymes as part of the inflammatory response. Neutrophils both release proteases themselves and activate proteases expressed by cells Fig. 2 Concentration of metabolites determined by NMR analysis. In vitro (OD 600 = 0) (blue) and joint fluid (green) were analyzed in technical triplicates while in vitro (OD 600 = 0.5) (red) was done in biological replicates. The detection limit of NMR is~2 μM. a: amino acids. b: nucleobases. c: glycans. d: metabolites resident in tissues. Thus, the host response could provide S. aureus with the free amino acids, the glycans and other nutrients needed for growth in vivo.
Among the transport systems, oligopeptide permease (opp) transporters encoded by the opp-1 operon (locus SAU060112_40296 -40300) were the most overexpressed transporter system (up to 101-fold, p-val = 2.2*10 −21 ) along with the genes surrounding the operon (locus SAU060112_40295 -40303). This operon was also highly overexpressed in cutaneous abscesses in humans [6]. The exact role of opp-1 remains unknown, although it was found to impact in vivo growth of S. aureus in mouse and rabbit infection models [40].
A major limitation of our study is the lack of biological replicates, as we did not obtain other samples of S. aureus infected joint fluid during the study period. In an attempt to find similarities of S. aureus gene expression in infections in human subjects, thus corroborating the findings in independent experiments, we compared our RNA-seq data extensively with microarray data from S. aureus cutaneous abscesses in humans [6]. Although the two studies differed in type of infections, genetic background of S. aureus isolates, experimental setups and analytic methods, they had 113 upregulated and 13 downregulated genes in common, which correspond to 36 % upregulated and 12 % downregulated genes found in this study. The upregulated virulence genes included saeRS, a few toxins (particularly γ-hemolysin and two uncharacterized leukocidin-like proteins), and chp (Table 2). With regard to nutrient acquisition and metabolism, the elevated transcripts were those of the sbn operon, ADI operon, tdcB, ald, and several enzymes involved in nucleoside catabolism as well as ethanol fermentation (Fig. 1). Additionally, the opp-1 operon was overexpressed in both studies. The 13 downregulated genes in both studies included the virulence regulator sarS (13-fold, p-val = 6.1*10 −6 ), cystathionine γ-lyase (mccB 6-fold, p-val = 0.001, glyoxal reductase (yvgN 5-fold, p-val = 0.004), glycosyl-4,4′-diaponeurosporenoate acyltransferase (crtO 668-fold, p-val = 0.004), phosphoribosylformylglycinamidine synthase 1 (purQ 5 fold, p-val = 0.002), and a few conserved proteins of unknown function. All in all, the biological function and regulation of these up-and downregulated genes need to be investigated by future in vivo studies.

Conclusions
This single case study highlights the capacity of combined transcriptome and metabolome analyses for elucidating the pathogenesis of deep-sited infections with and without a foreign body. Future research should explore the in vivo physiology and virulence of S. aureus, which may ultimately lead to new strategies to combat S. aureus infections.

Case history
The patient was an adult male with a sero-negative polyarthritis since his youth. Debut of psoriasis led to a diagnosis of psoriatic arthritis after approximately two decades. He had undergone numerous surgical procedures and had joint implants in one hip, both knees, one elbow and one shoulder. Immunomodulatory therapy with adalimumab (Humira, Abbott US), a tumor necrosis factor (TNF)-α antibody , was started 26 months before the admission. The patient was admitted after a fall with subsequent swelling of the right knee. He was febrile (38.8°C) and had marginal leukocytosis (12.0*10 9 /L) and highly elevated C-reactive protein (304 μg/mL, reference interval <10 μg/ mL). A joint puncture revealed serous joint fluid (60 % mononuclear leukocytes) and 10 4 -10 5 colony forming units of S. aureus, susceptible to penicillin, methicillin and 5 antibiotic classes other than β-lactam [41]. Intra-venous dicloxacillin was commenced on the 2nd day of admission, but changed to cefuroxim in combination with gentamicin due to spiking fever. S. aureus with the same antibiogram was obtained from blood culture and biopsies obtained during revision surgery with removal of the implant on the 4th day of admission. On the same day intravenous therapy was switched to penicillin G. The blood culture isolate was referred to Statens Serum Institut (Copenhagen, Denmark) for spa-typing as part of national surveillance (t908, annotated to Clonal Complex 45). Several months later the patient underwent surgical revision and removal of implants from the left elbow and the left hip. S. aureus infection with the same antibiogram was confirmed.

Culture and antibiotic resistance test
Joint fluid, biopsies and prosthetic components were cultured according to [42] with an incubation period of 14 days (see Additional file 7). Species identification was done with a MALDI Biotyper CA System (Bruker Daltonics, Germany). Antimicrobial susceptibility testing was carried out as above [41]. The S. aureus isolate from prosthetic components was designated SAU060112.
16S rRNA gene amplicon sequencing and data analysis DNA extraction was done using MolYsis complete5 (Molzym, Germany) according to the manufacturer's instructions. For 16S rRNA amplicon sequencing, the V1-3 region was PCR amplified with bacterial primers 27 F and 534R in accordance with the protocol used by the Human Microbiome Project [43] and sequenced on a MiSeq DNA sequencer (Illumina, CA) [44]. The 16S rRNA amplicon data were analyzed using QIIME toolkit [45]. Raw sequences were demultiplexed and qualityfiltered using the default parameters. Sequences were then clustered into OTUs based on 99 % sequence similarity and taxonomy assignment was done using the Greengenes database [46].  [47]. Automatic annotations provided by MaGe were curated manually to validate the presence or absence of genes of interest. Based on the annotations, the protein coding genes were classified into the Cluster of Orthologous Groups (COG) [48] functional categories using COG automatic classification tool at MaGe. Details of genome sequencing and annotation can be found in Additional file 7.

RNA sample collection, extraction and sequencing
Immediately following aspiration the joint fluid was centrifuged at 12100 g for 2 min at room temperature and the pellet and supernatant were snap-frozen separately in liquid nitrogen. RNA from in vitro cultures (3 biological replicates) were isolated from cultures grown to exponential phase (OD600~0.5) in LB medium. The cell suspension was centrifuged and supernatant and pellet were snap-frozen separately. All samples were stored at −80°C until RNA extraction or NMR analysis.
RNA was extracted using RiboPure™ Bacteria Kit (Ambion®, Life Technologies) except that the in vivo sample was homogenized in a mortar (precooled in liquid nitrogen) before RNA extraction. The RNA solutions were purified and concentrated using the MinElute PCR Purification Kit (Qiagen).
Twenty micrograms of in vivo-derived RNA was sequentially treated with the MICROBEnrich™ and MICROBExpress™ kits (Ambion®) to deplete mammalian RNA and enrich bacterial mRNA, respectively. Four to six micrograms of in vitro-derived RNA was used. Sequencing libraries were prepared with the enriched microbial RNA using Illumina® TruSeq® RNA Sample Preparation Kit v2 according to the manufacturer's instructions. Libraries were PE sequenced (2 × 150 bp) using Truseq SBS Kit v.3-HS Sequencing Kit on an Illumina HiSeq 2000.

Differential gene expression analysis
Using the RNA-Seq analysis function in CLC Genomics Workbench, reads were aligned to the annotated SAU060112 genome allowing a minimum length fraction of 0.8 and minimum similarity fraction of 0.95. A table of read counts was used as input for differential gene expression analysis using edgeR using default settings [49]. Only genes with false discovery rate <0.05 using Benjamini and Hochberg's algorithm [50] were classified as differentially expressed.

Availability of supporting data
The annotated genome sequence data was submitted to the European Nucleotide Archive (accession nos. CCXN01000001-CCXN01000017). The RNA-seq data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus [56] and are accessible through GEO Series accession number GSE62091 (http:// www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE62091).

Ethics statement
This study was conducted within the framework of the 'Prosthesis-Related Infection and Pain' (PRIS) -