Using genomics to understand meticillin- and vancomycin-resistant Staphylococcus aureus infections

Resistance to meticillin and vancomycin in Staphylococcus aureus significantly complicates the management of severe infections like bacteraemia, endocarditis or osteomyelitis. Here, we review the molecular mechanisms and genomic epidemiology of resistance to these agents, with a focus on how genomics has provided insights into the emergence and evolution of major meticillin-resistant S. aureus clones. We also provide insights on the use of bacterial whole-genome sequencing to inform management of S. aureus infections and for control of transmission at the hospital and in the community.


INtRoDuctIoN
The facultative pathogen Staphylococcus aureus is associated with asymptomatic carriage in 25 % of adults [1] and with a wide spectrum of clinical conditions ranging from skin and soft tissue infections, through to invasive infections such as pneumonia, bacteraemia, infective endocarditis, septic arthritis and osteomyelitis [2]. Invasive S. aureus infections still carry a high mortality (for example around 20 and 10 % for endocarditis [3,4] and pneumonia [5], respectively) and their management can be very complex, particularly when complicated by antimicrobial resistance [6].
The clinical introduction of penicillin in the 1940s dramatically improved the outcome of S. aureus infections (the mortality of S. aureus bacteraemia in the pre-antibiotic era was as high as 80 %); however, after the introduction of penicillin, resistance spread rapidly, and by 1948 more than the half of tested isolates in one centre were resistant to penicillin [7]. Interestingly, the rise of penicillin-resistant S. aureus was subsequently found to be linked to the spread of a single clone, phage type 80/81 [8], the first example of the 'epidemic waves' that now characterize the molecular epidemiology of resistant S. aureus [9]. A similar phenomenon was observed after the introduction of penicillinase-resistant penicillins (e.g. meticillin, oxacillin) in 1959. Two years later, a report described three clinical S. aureus isolates that were resistant to this newly introduced anti-staphylococcal antibiotic [10]. Recent work has established that meticillin-resistant S. aureus (MRSA) was already circulating prior to the introduction of meticillin and was likely selected for by penicillin [11]. MRSA subsequently disseminated in the hospital environment, and then separate epidemic waves occurred in the community. By contrast, resistance to the last-line antibiotic vancomycin has developed slowly following its introduction in 1958, with the first report of vancomycin-intermediate S. aureus (VISA) by Hiramatsu et al. in 1997 [12]. The relatively late emergence of vancomycin resistance was probably related to limited use of vancomycin until the 1980s, when the surge in MRSA infections boosted its use [13]. Resistance to the most recently introduced anti-staphylococcal antibiotics (daptomycin and linezolid) has also been readily acquired: for example, secondary resistance under treatment was described in the randomized controlled trial that led to FDA (Food and Drug Administration, USA) approval of daptomycin [14]; and linezolid resistance, albeit rare, has been reported in series of isolates [15].
In this mini-review, we provide an overview of the major genomic-based insights into the two major clinically relevant mechanisms of staphylococcal resistance (resistance to meticillin and vancomycin), and highlight the contribution of genomic epidemiology to the understanding of the establishment and spread of resistant clones (especially MRSA). Finally, we provide an outline for the future use of genomics beyond resistance research and epidemiology, towards improved individual patient management of invasive S. aureus infections, by prediction of antibiotic response, persistence and virulence.

GENoMIc INSIGhtS INto MRSa
Genetic basis of meticillin resistance S. aureus acquires resistance to anti-staphylococcal penicillins through expression of an additional penicillin-binding protein (PBP) (PBP2a) [16]. Unlike other PBPs, PBP2a is resistant to the inhibitory effects of all β-lactams (with the exception of ceftaroline and ceftobiprole) and is almost always encoded by the accessory gene mecA [17]. The expression of mecA is inducible and controlled by a signal-inducer protein and a repressor located within the mecA operon [17]. Accordingly, most MRSA strains express PBP2a at low level, but harbour highly resistant subpopulations (heteroresistance) [18]. High-level resistance can be expressed in special circumstances. An example is the stringent response, i.e. the intracellular accumulation of the second messenger (p)ppGpp secondary to nutritional stress [19,20]. In vitro studies identified genes involved in the stringent response (such as relA) as 'auxiliary genes' that alter the expression of oxacillin resistance, along with several other determinants

Impact Statement
Meticillin-and vancomycin-resistant Staphylococcus aureus have been included by the World Health Organization in the global priority list of antibiotic-resistant bacteria, given the high mortality and morbidity associated with invasive S. aureus infections such as endocarditis and osteomyelitis, and the suboptimal outcome of treatment when anti-staphylococcal β-lactams are not available. Whole-genome sequencing (WGS) studies have not only highlighted how meticillin-resistant S. aureus spreads in the community and at the hospital, but also shown how use of antibiotics and biocides in the community initiates and amplifies the establishment of drug-resistant S. aureus. Moreover, emerging resistance to last-line antibiotics like vancomycin, daptomycin and linezolid can now be dissected at the molecular level by genomic studies. Through increased understanding of the genomic basis of resistance and emerging work on S. aureus virulence and persistence, there is likely to be a growing role of WGS in the direct clinical management of S. aureus infections.
Recently, alternative mec alleles have been described. For example, mecC shares 70 % nucleotide identity with mecA and is typically found in livestock-associated MRSA (LA-MRSA) [23]. Based on two reviews of epidemiological studies [24,25], this mec variant appears to be infrequent and restricted to Europe (with one single case report from Australia [26]). Interestingly, mecC-MRSA appears to have a low oxacillin minimum inhibitory concentration (MIC) due to the distinctive characteristics of its PBP2a-homologue (PBP2c), including higher binding affinity for oxacillin than cefoxitin [27], and susceptibility to penicillin-β-lactam inhibitor combinations [28]. Accordingly, mecC-MRSA was successfully treated with β-lactams in an experimental endocarditis model [29]. Rarer mec homologues are also reported in other staphylococci or related species, such as mecA1 (Staphylococcus sciuri), mecA2 (Staphylococcus vitulinus) or mecB and mecD (Macrococcus spp.) [30][31][32]. Based on genomic studies, it is hypothesized that mecA was acquired several years prior to the first clinical detection of MRSA in 1961. Harkins et al. applied a Bayesian phylogenetic inference to a collection of early MRSA isolates  and concluded that meticillin resistance emerged in the mid-1940s, suggesting that the introduction of penicillin may have contributed to the selective pressure that lead to the advent of MRSA [19].
Horizontal transfer of mecA is made possible by carriage on a specific mobile genetic element (MGE) ranging between 23 and 68 kb in size, the staphylococcal cassette chromosome (SCC) [33]. The association of mecA with the SCC (termed the SCCmec) not only is important for mecA acquisition or transfer, but also is a key factor mediating antimicrobial co-resistance, since genes conferring resistance to non-βlactam antibiotics can be co-located in the same locus [34][35][36]. To date, 13 variants of SCCmec have been described [37], but with the increasing number of sequenced strains, new variants are likely to be discovered in the future. Beyond SCCmec, other MGEs are critical for acquisition and dissemination of resistance to antibiotics and biocides (particularly plasmids and transposons, see the review by Firth and colleagues [38]) and virulence determinants (particularly bacteriophages, see the review by Xia and Wolz [39]).
The conserved structure of SCCmec and mecA has facilitated the molecular detection of meticillin resistance; molecular point-of-care tests have streamlined the rapid detection of MRSA from clinical samples. However, correlation between the presence of mecA or mecC and phenotypic resistance to oxacillin is not absolute -approximately 3 % of S. aureus strains harbouring mecA are phenotypically susceptible to oxacillin [40,41]. As mentioned above, this phenomenon has been previously explained by heterogeneous synthesis of PBP2a [42], but a genomic study provided an interesting alternative mechanism. The authors investigated two clinical isolates of mecA-positive meticillin-susceptible S. aureus (MSSA) and demonstrated that mecA expression was suppressed by disruption of the gene through insertion of IS1181 in one case and a mecA frameshift mutation in the other [41]. β-Lactam sensitivity in MRSA has been investigated in a recently published study by Harrison et al., where they identified a subset of MRSA strains that were susceptible to penicillin/clavulanic acid combinations. The genomic basis of this phenomenon was found to be the association of mutations both in the promoter and coding sequence of mecA [43].
Conversely, oxacillin resistance can be mediated by other mechanisms than PBP2a production. Such mecA-(and mecC)negative, oxacillin-resistant strains [borderline oxacillinresistant S. aureus (BORSA)] are increasingly recognized and may be associated with failure of oxacillin clinical therapy, typically in complex, deep-seated infections [44,45]. While previous work has investigated β-lactamase hyperproduction or PBP mutations [46], recent genomic studies of BORSA isolates have linked this phenotype to mutations in the regulatory gene gdpP [47,48], which degrades the second messenger c-di-AMP. gdpP mutations have been linked to changes in cell-wall metabolism and increased resistance to antibiotics targeting the cell wall like β-lactams and vancomycin [49].

Genomic epidemiology of MRSa
The genomic epidemiology of MRSA is multifaceted. MRSA is typically clonal, with epidemic waves of infections characterized by the temporal rise and decline of clones [50,51]. Parallel to these chronological changes, geographical segregation can be observed, with some adaptation to specific environments, including health-care facilities, community settings and animal populations. These broad groupings form the basis for the often used classifications of health-careassociated MRSA (HA-MRSA), community-associated MRSA (CA-MRSA) and livestock-associated MRSA (LA-MRSA). Previously, molecular epidemiology using lower-resolution approaches, such as pulsed-field gel electrophoresis (PFGE), multilocus sequence typing (MLST) or spa typing, helped delineate dominant MRSA clones and track their emergence, expansion and spread [52]; however, over the past 10 years, whole-genome sequencing (WGS)-based studies have defined the complex epidemiology of MRSA, from tracking global dissemination of successful clones, to dissecting chains of transmission in hospitals and the community, and between livestock and humans.
Harris et al. performed the first study that applied largescale bacterial WGS to explore global dissemination of the HA-MRSA clone sequence type (ST)239 [53]. While the coregenome phylogeny was consistent with PFGE and spa typing, genomic data provided detailed insights into the phylogeographical structure of the ST239 lineage, the emergence of antibiotic-resistance mutations and transmission within a health-care facility. The impact of genomics on the investigation and control of hospital-associated MRSA outbreaks has also been demonstrated in subsequent papers [54][55][56], while others have analysed MRSA transmission networks in the community setting [57] or described the emergence of LA-MRSA and the complex interplay associated with transmission from humans to farm animals and vice versa [58,59]. Table 1 provides a selection of the major genomic epidemiological studies associated with MRSA and their key findings. Given the large number of studies that have applied WGS to address MRSA epidemiology, we have selected studies based on (i) their novelty at the time they were published, and (ii) the range of MRSA genomic epidemiology (e.g. global or local transmission, adaptive evolution, source attribution in LA-MRSA).
More recent S. aureus genomic epidemiological studies have evolved in two complementary directions: (i) expanding the breadth of the analysis by providing local, national or international surveillance frameworks for MRSA (and S. aureus in general); or (ii) performing in-depth investigations of single clones or lineages, and exploring the interplay between adaptive evolution and antibiotic pressure. For example, Aanansen et al. performed a population genomic study of 308 S. aureus isolates across 21 European countries. Their data provided a 'snapshot' view of the genetic diversity of S. aureus across a continent, and allowed the investigation of evolution within single clones and intercontinental transmission, as well as identification of both MSSA and MRSA 'high-risk clones' (e.g. CC22, CC30) based on speed of expansion as assessed from phylogenomics, phylogeographical structure and the presence of key resistance or virulence genes. Their study also showed that the population structures of MRSA and MSSA are fundamentally different, with the former being more clonal and geographically clustered [60]. Building on a similar approach, Reuter et al. described the population structure of over 1000 invasive MRSA isolates from the UK and Ireland [61]. An important finding of their study was strong phylogeographical clustering around hospital referral networks, highlighting the potential for the use of WGS in epidemiological surveillance and early identification of new hospital outbreaks [61,62]. Two recent studies expanded the framework of surveillance of MRSA transmission: Price et al. showed that genomics can be applied to interrogate the complex transmission interplay between patients, health-care workers and the environment in the health-care setting [63], while a study by Coll et al. integrated genomic data with epidemiological data retrieved from various sources (hospital admissions, postcodes, general practice attendance) to reconstruct MRSA transmission networks both in the hospital and the community [64].
With over 40 000 S. aureus genomes now publicly available, large-scale genomic surveillance is now possible [65,66]. Fig. 1 demonstrates the global distribution of MRSA clones based on publicly available S. aureus genomes processed through the Staphopia platform [65]. Despite the great interest of this large and ever-growing public dataset of S. aureus sequences, it should be noted that these data are not necessarily representative of the actual S. aureus epidemiology (sources of bias include the larger availability of sequencing in a small group of developed countries, increased sequencing of MRSA for public-health reasons and lack of metadata including country of collection for a large proportion of the isolates). Other available genomic platforms for S. aureus (and other bacteria) offer access to publicly available genomes and allow comparison of sequences uploaded by the user through analysis pipelines, e.g. patric (https://www. patricbrc. org), National Center for Biotechnology Information (NCBI) Pathogen detection (https://www. ncbi. nlm. nih. gov/ pathogens/; however, S. aureus is not included yet) and Pathogenwatch (https:// pathogen. watch). These repositories and new computational approaches allow high-throughput analysis of stored sequence data for both rapid and efficient genomic surveillance [67] and discovery of genetic determinants of resistance or pathogenesis [68].

In-depth genomic studies of specific MRSa lineages
An early example of an in-depth, clone-specific approach is reported in a study by Holden et al., who applied Bayesian phylogenetics methods to dissect the evolutionary history of the hospital-associated ST22 clone, the dominant MRSA clone in the UK. The analysis reconstructed the acquisition of important antibiotic-resistance determinants (mecA and resistance-associated mutations in the gyrA and grlA genes) and showed that rapid expansion and dissemination of sublineage ST22-A2 was likely promoted by acquisition of fluoroquinolone resistance [69]. Other authors have combined population genomics and phenotypic studies to dissect adaptive micro-evolution of MRSA both in the hospital [70] and community environment [71][72][73]. For example, two genomics studies of ST93, a community-acquired MRSA clone in Australia, have revealed how this lineage likely arose in a remote area in North-West Australia and subsequently disseminated across the continent and overseas [71,73]. One study also showed that this high-virulence clone changed its phenotype towards reduced virulence (e.g. expression of alpha-toxin) and increased susceptibility to oxacillin [71], possibly indicating adaptive evolution to the health-care environment, as previously shown in other clones such as CC30 [74]. Another Australian study explored adaptive evolution of the hospital-associated ST239 clone. Phylogenetic analysis showed that the epidemics had been enhanced by the introduction of a previously unrecognized sublineage from Asia. Both the Australian and the Asian sublineage of ST239 exhibited patterns of convergent evolution, namely decreased virulence and increased resistance to antibiotics (including vancomycin) -both characteristics of hospital adaptation. The transmission potential of ST239 was also highlighted in a study performed in an hospital in Thailand [75]. Further, in a worldwide study of the CA-MRSA clone ST59 (dominant in East Asia), the authors applied a Bayesian phylogenetic method ('Markov jumps') to identify 'source' countries (USA, Taiwan) and 'sink' countries (Australia, the UK, the Netherlands) of the ST59 epidemic [76]. Collectively, these studies have provided clear examples of the utility of population genomics (complemented with relevant phenotypic testing) in understanding the evolutionary mechanisms that underpin the success of MRSA clones.

co-resistance in MRSa
An area of ongoing interest is the role of co-resistance to non-β-lactam antibiotics in the spread and expansion of  [77]. Co-resistance is of both epidemiological and clinical relevance, since it has been shown that use of selected antibiotics (e.g. fluoroquinolones) may drive the MRSA epidemic [78]; thus, offering potential targets of preventive interventions through antibiotic stewardship in human health or in agriculture [79]. This is not only true for systemic antibiotics, but also for topical antibiotics and biocides [80].
From a genomic perspective, co-resistance can arise via four different mechanisms: (i) the same genetic determinant (gene or mutation) can confer resistance to multiple antibiotics (e.g. the pleiomorphic effects of rpoB mutations, which include decreased susceptibility to vancomycin and daptomycin [66]); (ii) compensatory mutations that counterbalance the fitness cost associated with resistance to one antibiotic can alter susceptibility to another drug (e.g. increased β-lactam susceptibility in VISA, see below [81]); (iii) the genetic resistance determinants are co-located on the same mobile element (e.g. within the SCCmec) [36]; or (iv) the resistance determinants are co-located within the same strain (e.g. fluoroquinolone resistance in MRSA).
Co-location on the same MGE is important, because it is associated with a risk of horizontal transmission of both genetic determinants. In MRSA, it is enabled by the plasticity of the SCCmec element that can host several genes associated with resistance to antibiotics or heavy metals [82]. For example, the erythromycin-resistance gene erm(A) is found on transposon Tn554, which is in turn nested within type II, III and VIII SCCmec elements [33], while the tetracycline-resistance plasmid pT181 is integrated in type III SCCmec. Similarly, the aminoglycoside-resistance gene aacA-aphD is carried on transposon Tn4001, which can in turn be found not only on several multi-resistance plasmids, but also on some SCCmec elements [83]. An illustrative example of the effect of resistance co-location on MRSA epidemiology is provided by the rapid emergence of fusidic acid-resistant MRSA and MSSA in New Zealand that was likely fuelled by the unrestricted use of topical fusidic acid. Genomic studies showed that fusidic acid resistance was restricted to two dominant MRSA clones (ST5 and ST1) and one MSSA clone (ST1) that had acquired the fusidic acid-resistance determinant, fusC. Crucially, the fusC operon was exclusively located in SCC elements, in both MSSA and MRSA [34,84]. Important MRSA co-resistance determinants located outside SCCmec are the quinoloneresistance genes gyrAB and grlAB, encoding the DNA gyrase and DNA topoisomerase, respectively [85]. Population genomics studies show that a single acquisition of quinolone resistance in the 1990s drove clonal expansion in both ST8 (USA300) [57] and the ST22 lineage (EMRSA-15) [69].
Tolerance to biocides and resistance to topical antibiotics can also be mediated by genes located on plasmids. The qacA gene encodes an efflux pump that is associated with tolerance to monovalent and divalent cations such as chlorhexidine, a widely used disinfectant in the hospital setting. It is generally carried on pSK1 family plasmids, often in combination with other resistance genes such as the β-lactamase blaZ.
A recent adaptive evolution study has shown a progressive decrease of chlorhexidine sensitivity among ST239 isolates. This phenotypic change was associated with a complex rearrangement of the pSK1 plasmids [86]. Mupirocin resistance is linked to mutations in the chromosomal gene ileRS (low-level resistance) or the plasmid-located gene mupA [87]. A recent genomic study showed that mutations in the essential gene ileS appear to have pleiotropic effects [88], highlighting the complexity of antibiotic resistance in S. aureus.

using genomics to explore virulence in MRSa
The complexity of virulence has been recently highlighted [89]. It has been very difficult to identify single genomic determinants of clinical outcome in S. aureus infections [90], with the exception of some toxin-mediated syndromes [91,92]. Nevertheless, it is possible to classify genetic determinants of virulence based on broad phenotypic characterization in experimental models (e.g. adhesion, toxin production, immune evasion and gene regulation) and their genetic context (i.e. core genome or accessory genome). Although a detailed description of virulence determinants is beyond the scope of this review, we will highlight some recent insights into S. aureus virulence that have been specifically provided by genomic studies.
A striking feature of CA-MRSA clones (such as ST8 and ST93) has been increased virulence in animal models and clinical examples of severe diseases (such as necrotizing skin or lung infections) [93,94]. This has also been demonstrated in vitro as increased cytotoxicity against human lymphocytes or macrophages [95]. While the genetic basis of increased virulence remains elusive, these clones were characterized by the presence of the Panton-Valentine leukocidin (PVL) toxin encoded by two genes lukS and lukF, located on a bacteriophage [96]. There remains controversy around the true clinical relevance of PVL [97]; however, a recent genomewide association study (GWAS) showed that it was strongly associated with S. aureus pyomyositis among children in Cambodia [98]. Further, it has been shown that cell toxicity resulting from exotoxin production in MRSA might be related to regulatory mechanisms rather that a single gene or locus. For example, there is an inverse relationship between PBP2a expression and toxicity; generally, classic CA-MRSA clones such as ST8 and ST93 have a lower oxacillin MIC and higher toxicity [99].
Other virulence determinants identified in genomic studies are the arginine catabolic mobile element (ACME), a large genetic segment possibly enhancing colonization in ST8 MRSA [100] and sasX, carried on a prophage in ST239 MRSA [101]. It is expected that genomic studies will continue to identify previously unrecognized virulence determinants. For example, a recent analysis of 92 USA300 isolates from an outbreak in a New York community identified mutations in the pyrimidine nucleotide biosynthetic operon regulator pyrR that were associated with enhanced fitness in vitro and enhanced colonization and transmission in a mouse model [72]. Furthermore, genomic analysis revealed that the acquisition of a bacteriophage was associated with larger skin abscesses in an animal model, emphasizing the impact of structural variants and MGEs on clone success and staphylococcal pathogenesis [72].

GENoMIc INSIGhtS INto VaNcoMycIN-RESIStaNt S. aureuS
The first report of vancomycin resistance was published in 1997 [12], 39 years after vancomycin was first introduced [102]. The authors isolated an MRSA strain with a vancomycin MIC of 8 mg l −1 from a patient with a persistent sternal wound infection, who had been exposed to vancomycin for several weeks [12]. From a molecular perspective, vancomycin resistance in S. aureus can arise through acquisition of the vancomycin-resistance determinant vanA, or more commonly via an array of vanA-independent mechanisms, mostly mutations in genes involved in cell-wall biosynthesis [103]. vanA-mediated resistance is associated with high-level resistance (VRSA, with a vancomycin MIC of 16 mg l −1 and higher) and is due to acquisition of the vanA operon, located on transposon Tn1546 [104], more commonly associated with vancomycin resistance in enterococci [105]. It was first described in 2002 in a patient with end-stage renal failure and diabetic foot infection [106]. Subsequent molecular studies demonstrated that the VRSA isolate carried a conjugative plasmid that had acquired Tn1546 from a co-infecting vancomycin-resistant Enterococcus faecalis [104]. While this report and previous experimental work [107] raised concerns of dissemination of high-level vancomycin resistance, VRSA remains very rare, with a total of only 14 cases reported in the USA [108], and a few reports from Iran [109] and India [110]. Although most VRSA strains to date belong to clonal complex 5, genomic analyses of 12 VRSA strains from the USA showed that they were genetically distant, with the most recent common ancestor around 1960 and likely independent acquisition of the plasmid-born vanA operon in each isolate [111].
Since its first description in 1997, several studies have investigated the complex background of vanA-independent vancomycin resistance. Phenotypically, these strains have low-level vancomycin resistance (VISA, vancomycin MIC 4-8 mg l −1 ) or may not be resistant when tested with conventional methods, yet harbour vancomycin heteroresistance (hVISA) [103]. They are also characterized by a thickened cell wall [112], slower growth and increased autolysis [113]. The molecular basis of these changes is complex and polygenic (extensive reviews have been published by Howden et al. [114] and McGuinness et al. [108]). Most mutations involve regulators of cell-wall biosynthesis, such as the two-component regulators vraRS [115], graRS [116] and walKR [117]. However, mutations in the rpoB gene (with or without co-resistance to rifampicin) [118] and in the PP2C phosphatase prpC [119] can also be associated with the VISA phenotype. Interestingly, in one case, decreased vancomycin susceptibility was linked to insertion of the transposon IS256 upstream of walKR [120], possibly altering its expression [121]. To date, two GWASs have assessed putative mutations associated with the VISA phenotype, both on ST239 isolates. The first study of 123 isolates found an association with a SNP in walkR [70], while the second (75 isolates) pointed to the H481Y/L/N rpoB mutation [122]. Further, a study using a machine-learning approach found that the VISA phenotype could be predicted with 84 % accuracy [123]. Although reduced vancomycin susceptibility can be found in any genetic background [124], ST239 strains tend to have a higher vancomycin MIC [125]. ST5 seems also to be more often associated with VISA [126].
Despite different genetic resistance mechanisms and phenotypes, VRSA and VISA share common features that distinguish them from MRSA. Unlike MRSA, VRSA and VISA are generally polyclonal, and no significant dissemination has been documented. To date, vancomycin resistance has occurred secondarily, during treatment for complicated S. aureus infections. As such, prevention of this resistance is likely best achieved through optimising the management of complex MRSA infections (including appropriate source control) and implementing antibiotic stewardship, rather than through infection control. However, there remains an omnipresent risk that vancomycin resistance could disseminate more effectively, especially with widespread transfer and expansion of a vanA-harbouring clone [127].
Unfortunately, resistance to newer anti-staphylococcal antibiotics is also emerging. Daptomycin has been proposed as a possible alternative to vancomycin for the treatment of invasive MRSA infections (with the exception of pneumonia) [128]; however, VISA/hVISA are often co-resistant to daptomycin [129] and secondary resistance can develop in vivo, especially in the case of deep-seated infections and poor source control [130]. Genetically, the main mechanism of daptomycin resistance is considered to be gain-of-function mutations in mprF, which encodes for a lysyltransferase producing lysylphosphatidylglycerol, a positively charged cellmembrane lipid that mediates resistance to host antimicrobial peptides [131]. It is hypothesized that mutations associated with daptomycin resistance increase cell-membrane positivity and, hence, impair binding of daptomycin, which is positively charged [132]. Similar to the VISA phenotype, comparative genomics studies of closely related isolates (either from cases of daptomycin treatment failure or from in vitro exposure experiments) have been instrumental in identifying further genetic determinants of daptomycin resistance. Strikingly, some of these genes are the same as those implicated in the VISA phenotype, such as walKR [133], rpoB [118] or prpC [119]. Furthermore, both daptomycin and vancomycin resistance are associated with the 'see-saw' effect, where increased resistance to glycopeptides and lipopeptides leads to reduced resistance to β-lactams [81]. The molecular basis of this phenomenon is complex and only partially investigated; for example, some studies have shown compensatory changes in the mecA gene [134] and reduced mecA expression [135].
Linezolid is a potential alternative anti-MRSA antibiotic that is not known to be affected by co-resistance to vancomycin. Resistance to linezolid can arise through to point mutations in 23S ribosomal RNA [136] and the ribosomal proteins L3/ L4 [137]; however, it can also be acquired through transfer of the accessory gene cfr, which produces a 23S rRNA methylase [138]. This gene is often carried by a plasmid [139] and a small multi-clonal outbreak of cfr-positive MRSA has been described in a Spanish hospital [140]. Ceftaroline, a nextgeneration cephalosporin with specific activity against PBP2a, might be used either as salvage therapy or as part of combination treatment for invasive MRSA infections [141]. However, ceftaroline resistance has been described in several MRSA lineages, both at baseline and on treatment [142], mainly through point mutations in mecA or in pbp4 [143,144]. Interestingly, mecA polymorphisms associated with ceftaroline resistance were found in a Korean study, despite the fact that ceftaroline had not yet been used in the country [145]. In a study of 421 strains, 17 % were nonsusceptible to ceftaroline (>1 mg l −1 ), with a higher proportion in ST239 MRSA [146].

applyING GENoMIcS to thE MaNaGEMENt of INVaSIVE S. aureuS INfEctIoNS
In addition to population-level studies, genomics has been increasingly used in the clinical microbiology laboratory at the patient level, mainly in the prediction of phenotypic resistance from genotypic data. Several translational studies have shown that genomic prediction of antibiotic resistance is highly accurate in the case of S. aureus [147,148]. The main issue with this approach is related to yet unknown resistance mechanisms [149]; however, it is likely that prediction accuracy will further improve as databases of genetic determinants of resistance are constantly updated, provided that careful genotype-phenotype association studies are also performed.
From a clinical perspective, one area where genomics offers considerable potential is through the use of WGS data to predict clinical outcomes and inform patient management, beyond considerations of antimicrobial resistance [90]. Previous molecular studies using multiple PCR or DNA arrays have suggested an association between certain clonal types and clinical manifestations or outcomes of S. aureus bacteraemia; however, with a few exceptions [150], no consistent link was demonstrated between the presence/absence of specific genes or mutations and clinical outcomes. More recently, Recker et al. used WGS data and applied a machine-learning algorithm to a S. aureus bacteraemia cohort in the UK to map associations between bacterial genetics, phenotypes potentially associated with virulence (cytotoxicity and biofilm formation) and clinical outcome [151]. The main finding of the study was that bacterial phenotype and genotype contributed to infection outcome; however, the effect appeared to be clone-specific, highlighting the complexity of outcome predictions in this setting. Another study from Denmark was not able to determine bacterial genomic predictors of infective endocarditis in S. aureus bacteraemia, despite using multiple genomic approaches [152]. Prediction might be more straightforward for rarer, specific clinical S. aureus syndromes. For example, Young et al. successfully applied GWAS to further highlight the role of PVL in the pathogenesis of pyomyositis [98]. However, to provide findings that can be implemented in clinical management, larger studies of genetic determinants of S. aureus infection outcomes are needed. Crucially, these studies will require additional validation either in independent cohorts or through functional tests [153], as well as integration of clinical covariables and, ideally, host genomics [154].
An alternative approach to uncover bacterial genetic determinants of disease is to investigate bacterial host adaptation through within-host evolution studies [155]. The study of genetically closely related isolates from the same patient offers a unique opportunity to identify new bacterial molecular markers of resistance, virulence or persistence without the need for large patient cohorts and without the analytical problems associated with GWAS approaches. These studies have played an essential part in establishing the genetic factors underlying VISA [114], but may also offer insights into the pathogenesis of invasive S. aureus infections [120,156]. Furthermore, comparative genomics of multiple patient isolates could help manage treatment failure by a reliable differentiation between true relapse and reinfection, or by the identification of de novo resistance mutations, especially if novel techniques are used that allow accurate detection of low-frequency populations [157].

coNcluSIoN aND futuRE DIREctIoNS
S. aureus remains a considerable clinical burden, in both hospital and community settings. This is aggravated by resistance to key anti-staphylococcal antibiotics like flucloxacillin and vancomycin. Molecular and genomic studies have provided invaluable insights into how resistance arises. For MRSA, they have demonstrated how MGEs have facilitated the selection and dissemination of distinct clones in hospital wards, community networks and at a global scale. Further, genomic datasets are now available, allowing the prediction of resistance to common antimicrobials, with ongoing work trying to accurately predict genotypic resistance to last-line antibiotics such as vancomycin, daptomycin, linezolid and ceftaroline. Future studies should also investigate whether bacterial genomics can be used to predict antibiotic synergism and response to combination therapy (e.g. vancomycin/daptomycin combination with β-lactams [158]). However, this large amount of genomic information can only be exploited if high-quality metadata are collected and (where possible) made publicly available. For example, phenotypic antibiotic susceptibility should be submitted along with other metadata (an approach encouraged by the NCBI, as described at: https://www. ncbi. nlm. nih. gov/ biosample/ docs/ antibiogram/). Even more importantly, clinical phenotypes (including relevant outcomes and relevant treatment and confounder factors) should be mapped from carefully designed, prospective cohorts [90]. This integrative approach combining publicly available databases, curated microbiological and clinical phenotypes and powerful computational tools will pave the way for bacterial genomics to move from population studies to patient management. five reasons to publish your next article with a Microbiology Society journal