Significant variability exists in the cytotoxicity of global methicillin-resistant Staphylococcus aureus lineages

Staphylococcus aureus is a major human pathogen where the emergence of antibiotic resistant lineages, such as methicillin-resistant S. aureus (MRSA), is a major health concern. While some MRSA lineages are restricted to the healthcare setting, the epidemiology of MRSA is changing globally, with the rise of specific lineages causing disease in healthy people in the community. In the past two decades, community-associated MRSA (CA-MRSA) has emerged as a clinically important and virulent pathogen associated with serious skin and soft-tissue infections (SSTI). These infections are primarily cytotoxin driven, leading to the suggestion that hypervirulent lineages/multi-locus sequence types (STs) exist. To examine this, we compared the cytotoxicity of 475 MRSA isolates representing five major MRSA STs (ST22, ST93, ST8, ST239 and ST36) by employing a monocyte-macrophage THP-1 cell line as a surrogate for measuring gross cytotoxicity. We demonstrate that while certain MRSA STs contain highly toxic isolates, there is such variability within lineages to suggest that this aspect of virulence should not be inferred from the genotype of any given isolate. Furthermore, by interrogating the accessory gene regulator (Agr) sequences in this collection we identified several Agr mutations that were associated with reduced cytotoxicity. Interestingly, the majority of isolates that were attenuated in cytotoxin production contained no mutations in the agr locus, indicating a role of other undefined genes in S. aureus toxin regulation.


INTRODUCTION
Staphylococcus aureus is responsible for a wide array of diseases ranging from superficial skin infections to severe, life-threatening cases of pneumonia and bacteraemia [1]. The emergence of antibiotic resistant lineages including methicillin-resistant S. aureus (MRSA) has complicated treatment. For decades the circulating MRSA lineages appeared to be limited to causing infections within healthcare settings in patients with predisposing conditions who were susceptible to infection [2]. However, in the 1990s distinct MRSA lineages, unrelated to earlier circulating MRSA lineages, started to emerge outside of healthcare settings to cause infections in otherwise healthy individuals [2,3]. Community-associated MRSA (CA-MRSA) isolates result in similar clinical manifestations, such as severe skin and soft-tissue infections (SSTIs), despite broad genetic diversity among CA-MRSA lineages [4]. Understanding the differences between the healthcare restricted MRSA lineages and the more recently emerged CA-MRSA lineages has been the focus of much attention [3].
Molecular and epidemiological studies of CA-MRSA isolates have identified multiple putative virulence factors, namely toxins, associated with the hypervirulent phenotype characteristic of CA-MRSA isolates. The majority of CA-MRSA isolates harbour lukSF-PV which encodes the bi-component pore-forming leucocidin [4]. Human neutrophils have been shown to be highly susceptible to PVL-mediated lysis due to the expression of complement receptors, C5aR and C5L2, OPEN ACCESS which are required for PVL binding [5]. Importantly, the role of PVL in pathogenesis is likely dependent on the site of infection; a clear role is observed in a rabbit necrotizing pneumonia model [6], but the role of this toxin in dermonecrosis is less evident [7,8]. Aside from PVL, over-expression of core genome virulence determinants, notably α-haemolysin (Hla) and α-type phenol-soluble modulins (α-PSMs), has been hypothesised to significantly contribute to the enhanced virulence of CA-MRSA isolates [9][10][11][12]. Hla is the prototypical β-barrel pore-forming cytotoxin [13]. Multiple studies utilising serological and animal models of infection have indicated a prominent role for Hla in the pathogenesis of disease [14]. The α-PSMs are characterised as small amphipathic α-helical peptides that efficiently lyse numerous cell types, independent of cell specific receptor [11,15]. As with Hla, α-PSMs significantly contribute to virulence in a murine infection models of bacteraemia and skin lesions [11].
Toxin regulation in S. aureus is governed primarily by the accessory gene regulator (agr), which employs a cell density dependent, quorum sensing system to upregulate a suite of secreted virulence factors such as toxins and downregulate surface binding proteins [16]. Accordingly, the agr system plays a central role in the development of a range of S. aureus infections, most notably in CA-MRSA skin infections [17]. Furthermore, transcriptomic data indicate enhanced agr regulation of important toxins (PVL, Hla and α-PSM), which are frequently associated with CA-MRSA virulence [17]. Together, these studies have contributed to the frequent use of the term 'hypervirulent' when referring to CA-MRSA lineages.
Recently, studies of toxicity at a population level have reported that this phenotype varies widely within individual MRSA multi-locus sequence types (STs) [18], suggesting that the virulence of an individual MRSA isolate should not be inferred from its genotype [18]. Non-cytolytic clinical isolates are commonly referred to as ' Agr dysfunctional' , due to the frequency at which mutations occur within the sensor kinase (AgrC) or response regulator (AgrA) encoding genes, impacting significantly on toxin expression. To examine this in greater detail, and across a globally representative collection of isolates, we focussed on a collection of 475 MRSA isolates representing five major MRSA STs (ST22, ST93, ST8, ST239 and ST36; Table S1, available in the online version of this article for further details) collected from Europe, North and South America, Asia and Australia, and examined their toxicity based on their ability to lyse human cells. As no single cell line exists that is susceptible to all of the toxins secreted by S. aureus, we use the THP-1 monocyte-macrophage cell line, which based on our empirical evidence, is susceptible to the widest range of S. aureus toxins, expressing receptors for PVL [5] and Hla [19] and susceptible to δ-haemolysin and α-PSMs [20]. Our findings indicate that toxicity varies significantly within STs and therefore lineage should not be used as a metric to infer pathogenicity. In addition, we have identified novel Agr mutations associated with attenuated toxicity and confirm that multiple isolates with reduced cytotoxicity have no mutations within the Agr operon, indicating the existence of other undefined toxin regulating genes in S. aureus.

Cytotoxicity assay
The cytotoxicity assay was optimised previously [20]. Briefly, to evaluate S. aureus toxicity, 20 µl of harvested supernatant (either used as 100 % or diluted to 70%, 30 % or 10 % in TSB) was incubated with 20 µl of washed THP-1 cells for 12 min at 37 °C under static conditions. Cell death was quantified using the Guava ViaCount reagent and easyCyte flow cytometry according to manufacturer's instructions. The toxicity of each isolate was measured with two technical repeats and three biological repeats with error bars indicating the standard deviation (SD). LAC (a highly toxic CA-MRSA isolate) supernatant and TS broth were used as positive and negative controls, respectively.

Statistical analysis
A one-way ANOVA with Tukey's multiple comparison test was used to examine differences between experimental results (GraphPad Prism v9.0), where a P value<0.05 was considered to be statistically significant.

S. aureus cytotoxicity is highly variable both between and within sequence types
In this study we compared the cytotoxicity profiles of previously published collections of MRSA isolates (i.e. sequence type (ST) 8 [20], ST22 and ST36 [24]; n=330) with newly derived toxicity profiles (ST93 and ST239; n=145) to provide a comprehensive and globally distributed picture of toxin expression by S. aureus. What was immediately apparent was that the individual STs contained numerous isolates that were either extremely toxic (100 % cell death) or non-toxic (0 % cell death), and could not be reliably assayed under the same conditions. As a result, we used the supernatant of two of the STs (ST239 and ST36) undiluted (100%) and for the other three STs (ST22, ST93 and ST8) we diluted them to 30 % (vol/vol) in TSB (Fig. 1). The cytotoxicity of each isolate was measured using three biological repeats with high reproducibility (Fig. S1). The proportion of ST239 and ST36 isolates that killed more than 50 % of the THP-1 cells was 44 and 39 %, respectively. By contrast, the proportion of ST8, ST22 and ST93 isolates that killed more than 50 % of cells was 87, 84 and 85 %, respectively. Given both the difference in the proportion of isolates killing more than 50 % of the cells and the differences in dilutions required (100 vs 30 %), this demonstrates that ST8, ST22 and ST93 contain a higher proportion of highly toxic isolates than ST239 and ST36. When comparing the median difference in cytotoxicity between STs, we observe that ST22 are significantly more cytotoxic than ST8 isolates (P=0.025). However, no difference in cytotoxicity is observed between ST22 and ST93, ST93 and ST8 or ST239 and ST36. However, what was equally striking from this initial analysis is the scale of the variation in the toxicity of the isolates within each ST (Fig. 1).

Toxicity cannot be inferred solely from the MRSA sequence type
To further compare the toxicity between STs we took the five most and five least toxic isolates from each MRSA ST and quantified their cytotoxicity over a range of dilutions of their supernatant (Fig. 2). The cytotoxicity of each supernatant dilution was measured using three biological repeats with high reproducibility (Fig. S2). The mean cytotoxicity of the five least toxic ST8 isolates was slightly higher than those from the other STs, but this was not statistically significant (P>0.05, Fig. 2a). This demonstrates that each of the five MRSA STs contain isolates expressing comparable low levels of Fig. 2. Cytotoxicity of the least and highest toxic isolates from each of the five MRSA sequence type. The supernatant of the five least (a) and highest (b) toxic isolates from each sequence type was diluted to 10, 30, or 70 % supernatant in sterile TS broth or used undiluted (100%) and the percentage cell death of THP-1 cells examined. The cytotoxicity of each isolate was quantified using three biological repeats and the data is presented as the mean and standard deviation across the five isolates. cytotoxicity. Of the most toxic isolates, there were significant differences in cytotoxicity across the five STs (Figs 2 and S2). At a supernatant dilution of 10 % the ST22 and ST8 isolates were on average more toxic than the others (P<0.05 for each comparison, Fig. 2b). At supernatant dilution of 30%, the ST22, ST93 and ST8 were on average more toxic than the other STs (P<0.05 for each comparison values), while at dilution 70%, the ST36 isolates were statistically significantly less toxic than the other STs (P<0.05 for each comparison values). ST22, ST93 and ST8 all contain the community-associated Type IV SCCmec element, which prior to this study would have led to them being referred to as hypervirulent [9,25]. By comparison, ST239 and ST36 isolates contain the type III and II healthcare-associated SCCmec elements, respectively. While the ST8 and ST22 collections did contain the most toxic isolates, the most toxic ST93 isolates were no more toxic than those from the ST239 collection but was statistically more cytotoxic than the most toxic isolates of ST36 when compared at 70 % dilution (ST93 vs ST36 P=0.043) and at 30 % dilution (P=0.017). (Figs 2b and S2b). This may be a result of differential expression of cytotoxins to which our cell line is not susceptible. However, the ST93 collection did contain a higher proportion of highly toxic isolates than the ST239 and ST36 collections, which is what we observed with the other type IV SCCmec carrying MRSA STs studied here (Fig. 1). This observation aligns with earlier work demonstrating that the type IV element has less of a down-regulating effect on toxicity compared to the larger hospital associated SCCmec types [26].

Agr mutations alone do not explain all low toxic isolates
To understand the molecular mechanisms behind the observed variation in cytotoxin production, we sought to examine the impact of sequence variability within the Agr regulatory locus on this phenotype. AgrC and AgrA represent the sensor kinase and response regulator of the Agr system, respectively (Fig. 3a), and mutations within the genes encoding these key proteins are frequently associated with reduced toxin production by clinical isolates (Fig. 3a) [16]. As the genome sequences were available for all the isolates studied here, we interrogated these and found that of the 475 isolates, 14 isolates had non-synonymous mutations in agrA (Table 1) and 35 isolates had non-synonymous mutations in agrC ( Table 2). The location of each of the amino acid changes inferred by these mutations is indicated in Tables 1 and 2 and have been mapped onto a representation of each protein (AgrA, Fig. 3b; AgrC, Fig. 3c) where the critically active regions are indicated.
Given the major role the Agr system plays in regulating cytotoxin production in S. aureus we examined what proportion of low toxic isolates could be explained by these Agr mutations. We set an arbitrary threshold where we considered any isolate that killed less than 20 % of the cells as 'low toxicity' , and any that killed more than 80 % as 'high toxicity' . Of the 475 isolates, 17.7 % (n=84) of the isolates were categorised as low toxicity and of these 34 (40.5 %) had non-synonymous mutations in the agr locus. While of the high toxicity isolates (n=215 (45.2 %)) eight had non-synonymous mutations in the agr locus, suggesting that these amino acids are not critical to the activity of the Agr system (Tables 1 and 2). Of the 34  Continued low toxicity isolates with mutations in the Agr locus, while we cannot assume these mutations are causative without making a series of isogenic mutants, having mapped them to the functional regions of the proteins (Fig. 3b, c) we can hypothesise those mutations that are likely to be causative of the low toxicity phenotype.

Assessment of agrA mutations
The AgrA protein is part of the LytR family of DNA-binding response regulators that modulate virulence determinants in several pathogenic bacteria [27]. This transcription factor contains the LytTR domain which binds with high affinity to both P2 and P3 promoter regions of the agr locus upregulating toxin production (Fig. 3a) [16,28]. The structure of the C-terminal LytTR DNA binding domain of AgrA (residues 137-238) in complex with a DNA duplex has been solved [29], highlighting the formation of an intricate ten-stranded elongated β-β-β sandwich (Fig. 3b). Key residues within three loop regions extending from the β-sheets makes contact with the major and intervening minor groove of its DNA target resulting in increased transcription from the P2 and P3 promoters and activation of the Agr system [29]. The mutations in eight of the low toxicity isolates resulted in premature truncation of the AgrA protein, which caused a loss of a functional region, and as such can be considered causative of the low toxicity phenotype (Table 1, Fig. 3b). Two of the other low toxicity isolates (ASARM93 and EOE120) had substitution mutations (E163G and D157Y, respectively) within the critical β-β-β sandwich. Mutation E163G occurs within β-strand three which forms the centre of the LytTR domain and plays a role in salt-bridge formation with H174 [29]. Mutation D157Y is located within an α-helix between strands β2 and β3 and forms a salt bridge interaction with both H208 located on helix between β5 and β6 and E141 positioned on the beginning of strand β1. These salt bridge interactions stabilise the LytTR β-β-β fold, and our data indicates that mutations within these residues impairs Agr activity and results in reduced toxicity.
Lastly, isolate ASARM63 had a mutation conferring a H169Y change, which results in the loss of a histidine residue critical to the DNA binding activity of AgrA [32]. Only one low toxic agrA mutant contained a substitution in a region with no ascribed function, and that was AGT9 which had an A47D change (not shown in Fig. 3b).

Assessment of agrC mutations
The AgrC protein is composed of a highly variable polytopic transmembrane sensor which relays auto-inducing peptide (AIP) mediated signals to the highly conserved cytoplasmic histidine kinase (HK) domain (Fig. 3a, c) [30]. The AgrC HK domain is composed of two subdomains; the catalytic ATP-binding (CA) domain which promotes the autophosphorylation of the central histidine (His239) residue within the H-box of the dimerized histidine phosphotransfer (DHp) domain (Fig. 3c) [30][31][32]. The CA subdomain N-box asparagine (N339) and glycine residues situated within the G-box (G394 and G396) are considered essential for ATP binding and Agr activity [33,34]. Importantly, alanine 340 and threonine 388 are conserved residues within this subdomain [32]. Two low toxic isolates (ASARM97 (A340V) and EOE096 (T388I)) contain substitutions in the CA kinase subdomain region of AgrC, and as such may be causative of the low toxic phenotype. Six low toxic isolates (Sa_TPS3148, Sa_TPS3151, Sa_TPS3161, GRE4, ICP5062 and EOE173) occur in conserved residues within the H-box of the DHp subdomain and as such, are likely to be causative of the low toxicity phenotype. Nine isolates, all ST239 belonging to the Turkish lineage [35], contained a double substitution of I311T and A343T, and in previous work we have functionally verified that these mutations result in a significant delay to RNAIII activation [18]. Molecular modelling has previously indicated that these mutations prevent AgrC dimerization and access to the ATP-binding pocket required for Agr activity which explains the low toxic phenotype [36].
Of the low toxic isolates with agrC mutations, four (ASARM204, ASARM84, ASARM154 and Sa_TPS3165) contain substitutions within either the extracellular or cytoplasmic regions of the protein with no ascribed function, and so we cannot claim with a level of confidence that these are causative of the low toxic phenotype.
This study adds to the growing literature indicating that cytotoxin regulation is highly complex in S. aureus and confirms that there are still undiscovered mechanisms at play that modulate this major virulence phenotype. This works reinforces the importance of Agr mutations in S. aureus toxicity. The data generated here provides a clearer understanding of the relationship between Agr mutations and toxicity, which may be exploited for future anti-virulence drug design.  However, considering that close to 60 % of low toxicity isolates had no Agr mutations demonstrates that numerous regulatory mechanisms await discovery, and work dedicated to unravelling the regulatory circuits controlling the toxicity of S. aureus is ongoing. This also suggests that use of the term ' Agr dysfunction' should be used with consideration of the fact that for many of low toxicity clinical isolates the Agr system is likely to be functional.

ST/Strain ID
What this work primarily highlights is the care and consideration needed when inferring terms like 'hypervirulence' to an isolate based on its sequence type. There can be no doubt that CA-MRSA lineages have recently emerged and are highly successful. Whether this is a result of 'hypervirulence' as opposed to an enhanced ability to transmit amongst otherwise healthy individuals needs further investigation. From the perspective of cytolytic toxin production by MRSA, the level of variability is significant and likely to play a major role in the outcome of S. aureus disease.

Conflicts of interest
The authors declare that there are no conflicts of interest.