Introduction

Since 1990, extensive spread in the United States of the community-associated methicillin-resistant Staphylococcus aureus (CA-MRSA) USA300 clone has been responsible for severe infections, including bone and joint infections (BJIs), especially in children [1]. BJIs represent up to 38% of pediatric CA-MRSA infections in the United States [2]. The global severity of CA-MRSA BJIs has been linked to local musculoskeletal involvement, including extraosseous abscesses, and the frequent need for surgical management [3,4], and systemic complications including severe sepsis and dissemination to the lungs [5,6].

Numerous CA-MRSA virulence factors have been identified [7], some of which appeared to play a specific role in the course of osteomyelitis. Panton—Valentine leukocidin (PVL), a phage-borne pore-forming toxin highly prevalent in CA-MRSA, was able to induce local complications, e.g. bone deformation and muscular abscesses in a rabbit osteomyelitis model [8]. PVL involvement in this context was further supported by the epidemiological association of PVL expression and clinical methicillin-susceptible S. aureus BJI severity [1]. CA-MRSA overexpression of another pore-forming toxin, α-hemolysin (Hla), was also shown to contribute to BJI pathogenesis [9]. We previously showed that Hla was associated with systemic complications, like severe sepsis-related mortality in CA-MRSA rabbit osteomyelitis [10].

In addition to PVL and Hla, secreted peptides called phenol-soluble modulins (PSMs) have been identified as key virulence factors that are also strongly expressed in staphylococci and CA-MRSA [11]. PSM-encoding genes appear in three distinct loci on the S. aureus chromosome. PSMs encoded by the first two loci have been designated PSMα and β. The third PSM locus encodes the δ-toxin; its open-reading frame is part of the RNAIII effector of the staphylococcal accessory-gene regulator (agr), a major two-component system coupled to a density-sensing cassette controlling the expression of most S. aureus virulence factors. All PSM-encoding genes are under agr control, either through the AgrA-mediated regulation pathway for psmα and β, or as a consequence of the co-transcription with RNAIII, as for hld [12]. Thus, PSM expression is tightly coupled with staphylococcal quorum sensing through agr.

Several PSM biological functions impact pathogenesis and, possibly, the course of S. aureus osteomyelitis. PSMs are small peptides with amphipathic properties, allowing them to destabilize lipid bilayers at high concentrations. This activity has been linked to receptor-independent cytotoxicity to host cells, including neutrophils and osteoblasts, the bone-forming cells [13,14]. Moreover, receptor-dependent proinflammatory activation of neutrophils by PSMs was found, resulting from PSM detection by the neutrophil formyl-peptide receptor 2 (FRP2) [15]. Finally, PSMs can assemble into amyloid-like fibrillae, which contribute to stabilizing staphylococcal biofilms [16,17] and induce a tolerogenic phenotype in dendritic cells, contributing to bacterial interference [12] and cell-cycle disruption [18]. These cytotoxic, proinflammatory and biofilm-enhancing properties could suggest PSM involvement in the in vivo outcome of osteomyelitis.

In previous in vivo studies on PSMs using a model of skin-and-soft-tissue infection, the LAC psmα-deleted strain, but not the other psm-deleted strains, was significantly less able to cause skin lesions in mice [11]. Also, Kobayashi et al demonstrated that PSMα and Hla contributed to the pathogenesis of USA300 skin infections in rabbits [19], whereas PSMα peptides had no impact in a PVL-negative ST72 CA-MRSA strain in a mouse model of skin infection [20]. Moreover, PSMs have been shown to facilitate dissemination from an infected catheter in a mouse model of biofilm-associated infection [21]. Furthermore, in a rabbit model of experimental endocarditis, Spaulding et al showed that deactivating PSMs delayed lethal sepsis but did not prevent mortality or valve lesions [22] and, thus, that they do not play a major role in infective endocarditis.

Concerning BJIs, PSMα expression was associated with extensive bone damage inducing the death of infected human osteoblasts in an ex vivo model of intracellular infection [14], whereas in a murine osteomyelitis model, PSMs enhanced cortical bone destruction [23]. Addressing in vivo PSM BJIs requires using a model sufficiently close to the human situation in terms of PVL susceptibility. In particular, murine models used in previous studies of PSM involvement in CA-MRSA osteomyelitis could not account for the PVL effect, because murine immune cells respond poorly to PVL [24], unlike those of rabbits that are highly PVL-sensitive [25]. Hence, disease outcomes as a function of PVL may differ between these species.

In this context, we sought to determine the contribution of PSMs to local and systemic osteomyelitis severity in a PVL-sensitive rabbit model of acute BJI, using the highly virulent USA300 CA-MRSA strain LAC, which expresses PSMs, Hla and PVL, and its isogenic derivatives lacking PSM expression.

Materials and Methods

We used the clinical S. aureus Los Angeles County wild-type strain (LAC-WT), and its isogenic Δpsm derivatives (LAC Δpsmα and LAC Δpsmαβhld, respectively), all kindly provided by Frank Deleo, as for our previous studies. The strains were created and originated in Dr. Michael Otto’s laboratory and were previously described [11].

The Δpsm derivatives were constructed via allelic replacement with a spectinomycin-resistance cassette of the psmα and psmβ operons, as previously described [26], and by disrupting codon usage within the hld gene inside RNAIII to disable δ-toxin. Technical difficulties and cost prevented us from verifying modifications/confirming plasmid transfer with a gene-recomplementation assay.

Microorganisms were stored at –80°C until use. Prior to the experiments, bacteria were cultured in casein hydrolysate and yeast extract medium (CCY) at 37°C with shaking for 18 h. After centrifugation, the supernatants were passed through 0.22-μm filters, and the pellets were washed and resuspended in phosphate-buffered saline (PBS). All inocula were quantified by optical density (OD), then serial dilutions were plated on tryptic soy agar (bioMérieux, Paris, France).

Norden’s method [27] was used to induce osteomyelitis in female New Zealand white rabbits, weighing between 2 and 3 kg, housed in individual cages with ad libitum access to food and water, in compliance with French legislation on animal experimentation and with the approval of the Animal Use Committee of Maisons-Alfort Veterinary School. Rabbits were anesthetized by intramuscular injection of 25 mg/kg each of ketamine (Virbac, Carros, France) and 2% Xylazine (Bayer Santé, Division Santé Animal, Loos, France). An 18-gauge needle was inserted percutaneously through the right tibial metaphysis into the medullary cavity to aspirate 0.4 mL of bone marrow. Infection was induced by direct injection of 0.1 mL of a sclerosing agent (3% sodium-tetradecyl sulfate), followed by 0.2 mL of inoculum (3 × 10⁸ colony-forming units (CFUs)) and 0.1 mL of saline to rinse the syringe. Fentanyl-patch analgesia was given for 7 days following surgery.

Animals were assigned to receive either LAC Δpsmα (n = 24) or LAC Δpsmαβhld (n = 24) to evaluate the in vivo impact of PSMs. Nineteen LAC-WT—infected rabbits served as controls.

Statistical Analyses

Percentages of hematogenous spread (positive blood, kidney and/or spleen cultures), bone deformations and abscesses were compared using Student’s t-test. Non-parametric Mann—Whitney U-tests were used to compare tibial bacterial counts and 2-way ANOVA with Tukey’s multiple comparisons tests for histological scores.

The Kaplan—Meier method was used to estimate survival mortality, with inoculation as time 0 and censoring at sacrifice on D14. Percentages of survivors were compared with the log-rank (Mantel—Cox) test (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Kaplan—Meier survival curves compared with the log-rank (Mantel—Cox) test.

Severe sepsis-related mortality tended to be lower for Δpsmα (··) and Δpsmαβhld (- -) (P = .06) vs. LAC WT (–).

https://doi.org/10.1371/journal.pone.0157133.g001

Survivors’ antibody-titer changes between D0 and D14 were analyzed with paired Welch’s t-test after log₁₀ transformation.

Results

We inoculated 48 rabbits with 3 × 10⁸ Δpsmα (n = 24) or Δpsmαβhld (n = 24) CFUs; the two that died immediately post-anesthesia (1 each in Δpsmα or Δpsmαβhld group) were excluded. Nineteen LAC-WT—inoculated rabbits served as controls.

D1 blood cultures were positive for 50% of the LAC-WT—infected rabbits tested vs. 65% of Δpsmα-infected (P = .47) and 61% of Δpsmαβhld-infected animals (P = .72).

Ten (52%) LAC-WT—infected rabbits died of severe sepsis with disseminated infection between D0 and D7 (median: 4 days) compared to five (22%) of the Δpsmα and seven (30%) of Δpsmαβhld groups (P = .03 and .11, respectively), with respective median survival of 5 and 4 days (P = .53, non-significant (NS)). The lack of significantly different mortality rates between Δpsmαβhld and WT groups, as opposed to the significantly different Δpsmα group, was likely attributable to random fluctuation and small sample size rather than to a real difference between Δpsmα and Δpsmαβhld groups. Indeed, Kaplan—Meier survival curves (Fig 1) showed a trend towards lower mortality of Δpsmα- and Δpsmαβhld-infected rabbits (P = .06).

Most non-survivors had positive spleen (80% of Δpsmα vs. 100% of Δpsmαβhld and LAC WT, NS) and kidney (100% of all groups) cultures. Macroscopic examination of all non-survivors’ lungs found red and congestive lesions, as in the LAC-WT group. Their lung histological scores confirmed the comparability of Δpsmα and Δpsmαβhld vs. LAC-WT infections. However, LAC-WT bacterial lung densities were higher (8.19 (interquartile range (IQR), 7.55–8.23) log₁₀ CFUs/g than Δpsmαβhld (6.39 (IQR, 6.27–6.48) log₁₀ CFUs/g P = .04), but comparable to Δpsmα (6.79 (IQR 5.62–7.10) log₁₀ CFUs/g, P = .25).

All non-survivors had infected bones, with median bacterial densities of 6.9 (IQR 6.85–6.96) log₁₀ Δpsmα CFUs/g and 6.86 (IQR, 6.81–7.38) Δpsmαβhld log₁₀ CFUs/g vs. 6.43 (IQR, 6.29–6.58) log₁₀ LAC WT CFUs/g of bone (P = .04 and .014, respectively) (Fig 2A). As for LAC-WT—infected rabbits, rare microabscesses were found at the inoculation site (0% Δpsmα, 20% Δpsmαβhld vs. 20% LAC WT, NS) but no cortical deformation (NS).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Comparisons of osteomyelitis parameters observed in Δpsmα (●○)-, Δpsmαβhld ▲△)- and with LAC-WT (◆◇)–infected non-survivors (black) and D14 survivors (white).

(A) Median bacterial bone densities, expressed in log₁₀ CFUs/g of bone. The Δpsmα- or Δpsmαβhld-infected non-survivors differed significantly from those infected with LAC WT. (B) Mean histological scores (0–4: none, minimal, mild, moderate, severe) for lung involvement. The bacterial densities in Δpsmαβhld-infected non-survivors differed significantly from those infected with LAC WT (*P = .017), whereas inflammation was more severe in Δpsmαβhld- and Δpsmα-infected D14 survivors than those infected with LAC WT (**P = .04 and ***.01, respectively).

https://doi.org/10.1371/journal.pone.0157133.g002

Survivors were sacrificed on D14 to evaluate bones and organs (Fig 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Macroscopic findings after challenge with a high inoculum of CA-MRSA USA300.

(A) Pulmonary hemorrhages demarcated by hyperemic regions in both lungs. (B) Abscesses in the left lung indicating disseminated infection. (C) White circled muscle abscess in the right leg. (D) Bone marrow filled with pus indicating osteomyelitis. (E) Splenomegaly with necrosis observed after disseminated sepsis.

https://doi.org/10.1371/journal.pone.0157133.g003

D14 Δpsmα- and Δpsmαβhld-infected survivors did not differ from LAC-WT—inoculated rabbits for hematogenous dissemination: 17%, 19% vs. 11% spleen-positive cultures (NS) and 34%, 44% vs. 11% positive kidney cultures (NS). Histological scores were comparable except for higher lung inflammation in Δpsmα (P = .04) and Δpsmαβhld (P = .01) than LAC-WT infections (Fig 2B), while mean (±SD) lung bacterial densities were comparable for the three groups (2.31 ± 1.04, 2.21 ± 0.39 vs. 2.03 ± 1.18 log₁₀ CFUs/g, respectively; NS).

Unexpectedly, Δpsmα and Δpsmαβhld vs. LAC-WT strains, respectively, induced similar bone deformities (39% and 43% vs. 67%; P = .73 and .41), muscle abscesses (83% and 75% vs. 89%; NS), and bone bacterial density (5.89 (IQR, 4.71–6.63) log₁₀ CFUs/g and 5.91 (IQR, 4.82–6.83) log₁₀ CFUs/g) vs. (6.15 (IQR, 2.47–6.73) log₁₀ CFUs/g) (P = .95 and .82) (Fig 2A).

D14 survivor’s titers were significantly higher than on D0 for anti-PVL (6.4-fold increase, 95% CI 2.9, 14.1; P = 3.5 × 10⁻⁵, paired t-test of log₁₀ values) and anti-Hla antibodies (41.3-fold increase, 95% CI 18.1, 96.8; P = 2.9 × 10⁻¹⁰), thereby confirming a significant immune response towards the two toxins. Survivors’ and non-survivors’ D0 anti-PVL and anti-Hla antibody titers were similar (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Serum anti-PVL—and -Hla—antibody titers in a rabbit model of S. aureus LAC experimental osteomyelitis.

Survivors’ and non-survivors’ Initial anti-PVL (A) and -Hla titers (C) did not differ significantly. Survivors had significantly higher anti-PVL (B) and -Hla (D) antibody levels between D0 (inoculation) and D14 (sacrifice) (P < .001). Welch’s t-test analyses of paired (B, D) or unpaired (A, C) log₁₀ values.

https://doi.org/10.1371/journal.pone.0157133.g004

Discussion

Our experimental model closely reproduces features of severe acute osteomyelitis seen in children with about half of the LAC-WT—infected rabbits dying within 7 days of severe sepsis, bacteremia, and high bone and lung bacterial densities associated with histological lung lesions, while D14 survivors developed severe bone infections with deformation and abscesses.

Our main results demonstrated that, during the early first phase of acute disseminated severe sepsis, deactivating PSMs tended to limit sepsis-attributable mortality, albeit not significantly. Δpsmαβhld-infected non-survivors had more bone CFUs than LAC-WT—infected non-survivors. In contrast, the former non-survivors had lower lung CFUs. This result can appear to be inconsistent with the bacterial lung histological score. However, bacterial enumeration by CFU counting is more reliable because (i) based on a larger sample size (tissue homogenate of a large piece of tissue) and (ii) assessing living but not dead cells.

These findings agree with the previously described role of PSMs in facilitating bacterial dissemination from an infected catheter in a mouse model [21, 28], a biofilm-associated infection like BJI. Conversely, during the second phase of subacute osteomyelitis (D14), deactivating PSMs did not modify bone CFUs, abscesses or deformities, in contrast to what could have been anticipated based on previous in vitro and ex vivo observations of PSMs’ osteoblast cytotoxicities [14].

The impact of PSMs (mainly PSMα) during the early phase of severe sepsis could be due to elevated production of agr-dependent toxins, including Hla and PSMs [29], especially by the LAC-WT strain. Although PSM deletion resulted in lower mortality, the effect was not significant, in contrast to previous observations with Hla [10].

Alternatively, the absence of PSMs’ effect on bone damage in D14 survivors could be explained by two hypotheses. First, agr might be less expressed during this late phase of osteomyelitis, considered a localized biofilm infection [29]. That possibility underlines the need to study the impact of virulence factors in a subacute osteomyelitis model (D14) with long-term animal survival [14]. Second, a potential PSM effect could have been masked by the impact of PVL, which plays a major role by enhancing rapid local spread of rabbit osteomyelitis with extraosseous infection extension, especially muscle abscesses [8], or by other toxins, e.g. Hla, which has been associated with severe sepsis-related mortality [10]. PSMs are by far the most abundant protein secreted by S. aureus (70–80%), much more than Hla or PVL [30]. Our results do not exclude a possible PSM effect on osteomyelitis caused by PVL-negative S. aureus isolates. A double-mutant strain Δpvl–Δpsmαβhld would have helped test this hypothesis, but we were unsuccessful in constructing one.

Our results do not corroborate those of Cassat et al, who showed that PSMα significantly limited bone remodeling in mice after creation of a cortical defect in the femur and local inoculation of 1 × 10⁶ LAC-WT or Δpsmα1–4 CFUs [23]. However, it should be stressed that their model of post-traumatic localized osteomyelitis is far from what was encountered in children with primary CA-MRSA osteomyelitis [1,2,4,5,31].

Finally, because PSMs were shown to play a proinflammatory role by inducing neutrophil activation and cytokine release through the through human formyl-peptide receptor-2 (FPR2) pathway, the higher pulmonary inflammation scores for Δpsmα and Δpsmαβhld groups were also unexpected [15]. Our results cannot be explained by the comparable lung bacterial densities of Δpsmαβhld and LAC-WT groups. FPR2 is known to be expressed by several epithelial tissues including human, mouse and rat lungs. However, FPR expressed by rabbits is only 68% homologous to FPR2. Also, FPR2 can be triggered by different ligands and induce pro- or anti-inflammatory responses [32]. Therefore, we think the pulmonary inflammation seen in experimental rabbit osteomyelitis might involve other pathways and cytokines, which would support our results.

In conclusion, our results showed that deactivating PSMs prevented bacterial dissemination from bone during the early stage of the infection, but did not impact the local severity of rabbit USA300 osteomyelitis during the later stage. In agreement with our earlier work on PVL and Hla involvements in CA-MRSA osteomyelitis [8,10], these new findings confirmed that PVL expression in a susceptible host, e.g. rabbit, is the major cause of abscesses and bone damage, potentially masking PSM effects. Nevertheless, PSMs and their modulation by the staphylococcal PSM-degrading protease aureolysin might remain important bone-damaging factors in PVL-insensitive hosts [23] and perhaps in PVL-negative S. aureus. This strong dependence of staphylococcal osteomyelitis pathophysiology on host susceptibility factors and the infecting strain’s toxin-gene content are exemplary of the difficulties that must be overcome to improve our understanding of this disease. Beyond the pathogenic role of toxins, other bacterial factors related to quorum sensing and hypoxic response in poorly oxygenated bone tissue are likely equally important to establish bone infection, as recently demonstrated by comparisons of S. aureus genes essential to infection in abscess and osteomyelitis models [33,34]. Among the intricate array of staphylococcal virulence factors known to contribute to osteomyelitis so far, Hla is the only one that: measurably influences local and systemic outcomes, even in the presence of PVL [10]; is common to virtually all S. aureus lineages [35]; and is already targeted by a monoclonal antibody being tested in a clinical trial (study identifier NCT02296320) to treat S. aureus infection [36]. In this context, it seems reasonable to designate Hla as a more promising target than PSMs to treat severe CA-MRSA osteomyelitis, in agreement with recent conclusions drawn from a mice model of skin and soft tissue infection [20].

Acknowledgments

The authors thank the Raymond Poincaré University Hospital and Infectious Diseases Colleagues: Aurore Lagrange and Aurélien Dinh for their support and Janet Jacobson for editorial assistance.