|Home | About | Journals | Submit | Contact Us | Français|
Persistent methicillin-resistant Staphylococcus aureus (MRSA) bacteremia (PB) (positive blood cultures after ≥7 days of therapy) represents a clinically challenging subset of invasive MRSA infections. In this investigation, we examined the potential correlation of specific virulence signatures with PB versus resolving MRSA bacteremia (RB) (negative blood cultures within 2 to 4 days of therapy) strains. Thirty-six MRSA isolates from patients enrolled in a recent multinational clinical trial were studied for (i) susceptibility to host defense cationic peptides (HDPs) (i.e., thrombin-induced platelet microbicidal proteins [tPMPs] and human neutrophil peptide 1 [hNP-1]); (ii) adherence to host endovascular ligands (fibronectin) and cells (endothelial cells); and (iii) biofilm formation. We found that PB isolates exhibited significantly reduced susceptibilities to tPMPs and hNP-1 (P < 0.001 and P = 0.023, respectively). There was no significant association between the PB outcome and fibronectin binding, endothelial cell binding, or biofilm formation (P = 0.25, 0.97, and 0.064 versus RB strains, respectively). However, multiple logistic regression analysis revealed that the PB outcome was significantly associated with the combination of reduced susceptibilities to HDPs and extent of biofilm formation (P < 0.0001). Similar results were obtained in a second analysis using days of bacteremia as a continuous outcome, showing that reduced HDP susceptibilities and increased biofilm formation cocontributed to predict the duration of bacteremia. Our data indicate that PB isolates have specific pathogenic signatures independent of conventional antimicrobial susceptibility. These combinatorial mosaics can be defined and used to prospectively distinguish PB from RB strains in advance and potentially to predict ultimate clinical outcomes.
Staphylococcus aureus is a leading cause of bacteremia and infective endocarditis (IE) throughout the industrialized world (18, 22, 42, 54). A growing proportion of these bloodstream infections are due to methicillin-resistant Staphylococcus aureus (MRSA), which is associated with worse clinical outcome, longer hospitalization, and higher net cost than similar infections caused by methicillin-susceptible S. aureus (MSSA) (3, 6, 14, 15, 26, 49, 52, 60). Persistent-bacteremia (PB) outcomes comprise 20 to 30% of all episodes of MRSA bacteremia and are especially relevant to endovascular infections (13, 22, 23, 31). Why some MRSA bacteremia strains persist while others resolve (RB) despite similar baseline clinical and microbiologic characteristics and identical medical and surgical therapeutic strategies is poorly understood.
In two recent studies, we investigated S. aureus virulence factors that could potentially differentiate PB from RB strains in the context of endovascular infections (23, 56). We initially studied PB or RB isolates from the Duke University Medical Center and delineated several in vitro parameters that appeared to distinguish these two strain cohorts, including enhanced matrix ligand and endothelial cell binding, and reduced susceptibility to polymorphonuclear leukocytes (PMN) and platelet-specific host defense peptides (HDPs) (23, 56). Several other recent investigations have also reported associations between putative pathogenic markers of MRSA and the PB outcome in patients with both IE and non-IE syndromes (24, 33, 43, 45). Taken together, these experimental and clinical data suggest that PB isolates may have specific pathogenic characteristics that distinguish them from RB isolates and that might potentially predict a worse clinical outcome. However, a decade of literature studying MRSA virulence factors has identified a recurring theme: no single virulence factor alone appears to be sufficient to describe MRSA virulence in endovascular infection. Rather, we hypothesize that combinations of virulence factors are necessarily or uniquely involved in PB.
Building on our prior investigations, the current study aimed to identify further potential mosaic (“combinatorial”) signatures associated with the PB outcome. To circumvent possible geographic, single-center, and/or clonality biases from previous studies, we utilized a PB and an RB strain cohort from 2002 to 2005, involving a well-characterized, multinational S. aureus bacteremia antibiotic trial (20). We profiled relevant genotypic and in vitro phenotypic characteristics, including susceptibility profiles to prototypic innate HDPs, endovascular host cell and ligand binding, and biofilm formation.
(This work was presented in part at the 110th General Meeting of the American Society for Microbiology in San Diego, CA, 23 to 27 May 2010, abstract D-631.)
All MRSA strains used in this study originated from a multinational S. aureus bacteremia clinical trial collection conducted between 2002 and 2005 (20). Detailed descriptions of the study patients have been previously reported (20). In brief, the patients in this study were well matched for epidemiologic, demographic, and clinical features (e.g., underlying diseases). All 36 patients in our analysis had complicated MRSA bacteremia and/or infective endocarditis, and all received either daptomycin monotherapy or dual-drug standard therapy consisting of vancomycin plus low-dose, short-course aminoglycosides. As demonstrated by the data, patients in daptomycin and vancomycin treatment groups had similar clinical outcomes (“noninferior”) (20). Clinical blood culture data were used to group our study isolates into PB and RB cohorts. PB (n = 18) was defined as ≥7 days of positive MRSA blood cultures despite receiving antibiotics to which the isolate was susceptible in vitro (37). RB (n = 18) was defined as an initial positive MRSA blood culture, with subsequent blood cultures being negative 2 to 4 days after therapy initiation. These definitions of PB and RB are the same as those used in our previous strain set investigations of this clinical outcome (23, 56). When not otherwise specified, bacteria were grown overnight in tryptic soy broth (TSB) at 37°C.
The staphylococcal cassette chromosome mec (SCCmec) type was determined by previously published PCR methods (25, 29, 51). The spa typing was defined following analysis of the gene's tandem repeat polymorphisms as determined by nucleotide sequencing (32, 51). spa types were used to provisionally assign isolates to multilocus sequence typing-defined clonal complexes (CCs), as described previously (39). The presence or absence of 30 additional standard staphylococcal virulence genes and agr typing were determined by multiplex PCR (11).
Thrombin-induced platelet microbicidal proteins (tPMPs) were prepared from thrombin-stimulated platelets isolated from fresh rabbit blood as previously described (55). The bioactivity of the tPMPs was assessed by a Bacillus subtilis ATCC 6633 assay as detailed previously (57, 59). Human neutrophil peptide 1 (hNP-1) was purchased from Peptides International (Louisville, KY). The HDP susceptibilities were assayed by exposing 103 CFU/ml bacteria to tPMPs (1 μg/ml equivalent) or 105 CFU/ml bacteria to hNP-1 (10 μg/ml) for 2 h at 37°C as previously described (23, 56). These HDP concentrations were selected based on extensive pilot studies identifying peptide levels that did not rapidly kill control S. aureus strains over the 2-h period of exposure to HDPs (data not shown). The results are expressed as the percentage (± standard deviation [SD]) of the initial inoculum that survived exposure to HDPs. A minimum of two independent assays were performed for each strain.
To measure the ability of S. aureus to adhere to fibronectin, 96-well plates were coated with purified human fibronectin (10 μg/ml; Sigma Chemicals) for 18 h at 4°C, treated with 3% bovine serum albumin (Sigma Chemicals) for 2 h to prevent nonspecific adhesion, and washed with phosphate-buffered saline (PBS) three times before organism seeding. Mid-logarithmic-phase bacterial cells (100 μl) grown in TSB were added at a final inoculum of 109 CFU/ml (optical density at 600 nm [OD600] = 1.0) to the plates and incubated for 2 h. Then, the wells were washed three times with PBS and fixed at 55°C for about 30 min. Bound bacteria were detected by staining them with crystal violet and measuring the OD570 with an enzyme-linked immunosorbent assay (ELISA) plate reader (61). Before all isolates were tested, the optimal fibronectin concentration was determined by establishing a standard curve with different fibronectin concentrations (2.5 to 50 μg/ml) using S. aureus strain NCTC 8325-4. The results (±SD) shown are the averages of at least two independent experiments for each isolate.
Endothelial cells harvested from umbilical cord veins were cultured as described elsewhere (17). Confluent endothelial cell monolayer 6-well plates were washed twice with Hanks balanced salt solution (HBSS) before organism seeding. Bacterial isolates were added to plates at a final inoculum of 5 × 103 CFU/ml and incubated for 1 h under static conditions in a 5% CO2 incubator at 37°C. Unbound bacteria were removed by washing the plates with HBSS, and TSB agar was added. Adherence was expressed as the average percentage (±SD) of the initial inoculum bound from at least two independent experiments for each isolate (56).
MRSA cells from fresh blood culture plates were resuspended in physiological NaCl solution to a density of 0.5 McFarland standard and diluted 1:100 into brain heart infusion (BHI) supplemented with 0.5% glucose; 200 μl of this suspension was transferred to wells of 96-well Nunc Delta tissue culture plates (Roskilde, Denmark) and incubated for 18 h at 37°C. After incubation, the wells were washed with PBS three times, air dried, stained with Safranin (0.1% in distilled water; Acros Organics) for 30 s, and washed three times with distilled water. After visual observation, the adhering dye was dissolved in 30% acetic acid, and absorption was measured at 490 nm (53). An A490 of >0.5 was considered biofilm. The results (±SD) represent a minimum of three independent assays for each strain.
In order to determine if the net composition of PB-associated biofilms differed from that of RB-associated biofilms, we carried out biofilm stability assays in the presence of carbohydrate, protein, and DNA dispersal agents. For these investigations, we selected the five RB isolates with the strongest biofilm formation (A490 > 1.0) and, in parallel, selected five PB strains, also with strong biofilm formation (A490 > 1.0). The supernatants of 18-h-old biofilms were replaced by fresh medium supplemented with either 10 mM sodium metaperiodate (Alfa Aesar, Ward Hill, MA), 100 μg/ml protease K (Acros Organics), or 140 U/ml RNase-free DNase I (Takara Bio Inc., Shiga, Japan) and incubated for 2 h at 37°C as previously described (30, 50, 53). Medium without supplement served as a negative control. After treatment, the biofilms were quantified as described above. Each experiment was performed independently at least twice on separate days, using triplicates for each run.
A two-sample test for binomial proportions was used to compare the genetic backgrounds of PB and RB isolates; P values of ≤0.05 were considered statistically significant. Means and standard deviations were calculated for all predictor variables (phenotypes). For analysis of the relationships of variables to PB versus RB, comparisons of individual variables were performed using simple logistic regression, and multiple logistic regression was used to assess the joint relationship of the predictors with the outcome. All variables with a P value of <0.2 were included in a stepwise regression procedure, and the criterion for remaining in the model was significance at an α of 0.05.
To address the somewhat arbitrary categorical definition used for grouping the strains into PB and RB cohorts, we performed a supplementary analysis using days of bacteremia as a continuous outcome. A log transformation was applied to make the outcome more normally distributed. Simple linear regression was conducted for all predictor variables, and a multivariable model was created using stepwise multiple linear regression procedure. Once the final model was identified, the regression coefficients and their confidence intervals were transformed back. Criteria for entry into and remaining in the model were the same as in the primary analysis.
Vancomycin MIC ranges were 0.5 to 1.0 μg/ml, and the median MICs did not differ between the PB and RB groups. Population analyses of the PB isolates revealed no evidence of vancomycin heteroresistant subpopulations (data not shown).
A total of 6 CCs were represented among the 36 isolates (Table (Table1).1). The most common CC type observed was CC5 in both PB (44.44%) and RB (44.44%) isolates. RB isolates were significantly more likely to be CC1 than PB isolates (16.7% versus 0%; P = 0.012). The most common spa type was 2 in both PB (27.8%) and RB (33.3%) isolates. SCCmec IV was the most commonly observed type, with no differences between PB and RB strains (53% of total isolates). Collectively, 47% of all isolates were agr group I and 41% were agr group II. Taken together, there were no significant associations of either SCCmec type, spa type, or agr group and the PB or RB outcome. Carriage of the V8 protease gene (sspA) was significantly overrepresented among PB isolates versus RB isolates (P = 0.042). There were no other significant differences between the PB and RB strain sets with regard to the presence or absence of the other 29 virulence genes studied (data not shown).
The phenotypic characteristics of PB and RB strains are summarized in Table Table2;2; individual data are shown in Fig. Fig.1.1. The PB isolates exhibited significantly reduced susceptibilities to in vitro killing by hNP-1 and tPMPs (P = 0.02 and P < 0.001, respectively, using simple logistic regression analyses). The association between the PB outcome and fibronectin binding, endothelial cell binding, or biofilm formation was not significant (P = 0.25, 0.97, and 0.064 versus RB strains, respectively). Stepwise multiple logistic regression identified the combination of relative hNP-1 and tPMP resistance, together with biofilm formation, as the best composite model to predict the PB outcome (P < 0.0001; pseudo R2 = 0.486) (Table (Table2).2). Model quality was assessed using the Hosmer-Lemeshow goodness-of-fit statistic, and it was determined that all models fit the data well.
Additionally, we performed a supplementary analysis using days of bacteremia as a continuous outcome. As in the primary analysis, higher relative resistances to in vitro killing by hNP-1 or tPMPs were significantly associated with longer duration of bacteremia in univariate analysis (P = 0.014 and P < 0.001, respectively). The associations were nonsignificant for fibronectin binding, endothelial cell binding, and biofilm formation (P > 0.1 for all comparisons). Multiple linear regression revealed that the combination of relative hNP-1 (P = 0.007) and tPMP (P < 0.001) resistance, together with biofilm formation (P = 0.021), was the best composite model to predict the duration of bacteremia (P < 0.0001; adjusted R2 = 0.492) (Table (Table3).3). The appropriateness of the model was assessed using residual plots, and it was determined that all models fit the data well.
To investigate the net chemical composition of the extracellular matrix of PB and RB biofilms, we performed biofilm stability assays in the presence or absence of various dispersal agents. Sodium metaperiodate oxidation of polysaccharide intercellular adhesions (PIA) or polymeric N-acetyl-glucosamine (PNAG) led to an average reduction of 44% in the A490 across all 10 isolates. Reductions were significantly higher in RB versus PB strains, suggesting that RB isolates had a substantial carbohydrate content (58.4% versus 30.6%; P = 0.04) (Fig. (Fig.2).2). In addition, the biofilms of all 10 isolates were significantly reduced by treatment with proteinase K. Moreover, 8 of the 10 isolates (4 PB and 4 RB) were significantly reduced by treatment with DNase I. The average reductions in the A490 were 89% and 70%, due to proteinase K and DNase I, respectively, and the extents of dispersal were significantly higher than dispersal by sodium metaperiodate (P < 0.001 and P < 0.05 for proteinase K and DNase I versus sodium metaperiodate, respectively). The extents of dispersal due to proteinase K and DNase I were similar in the PB and RB groups (Fig. (Fig.22).
Conventional approaches to MRSA virulence assessments have predominantly focused on putative virulence determinants individually, using “molecular Koch's postulates” (16). Such studies have provided valuable insights into the contributions of specific virulence factors to MRSA pathogenesis, yet an important theme that has emerged from these efforts is that no single virulence determinant alone is sufficient to enable full virulence of MRSA. In this study, we are, to our knowledge, the first to employ a combinatorial strategy to assess the impact of phenotypic mosaics that potentially enable an MRSA PB outcome in the context of endovascular infections. The use of clinical MRSA isolates from a recent multinational clinical trial allowed us to exclude possible geographic, single-center, and/or clonality biases that may have substantially limited previous investigations.
A number of interesting observations emerged from the present investigation. First, an important factor in the pathogenesis of S. aureus infection is the ability of the microbe to circumvent clearance by platelet-mediated mechanisms, principally via resistance to killing by platelet HDPs (e.g., tPMPs) secreted in response to agonists generated at endovascular infection sites (58). Reduced killing by tPMPs in vitro has been correlated with enhanced in vivo virulence and less efficacy of antibiotic therapy (including vancomycin) in a rabbit endocarditis (IE) model in several previous studies from our laboratory (9, 10, 40). In addition, S. aureus bloodstream strains isolated from endovascular infections were significantly less susceptible to killing by tPMPs in vitro than bacteremia strains from other clinical syndromes (e.g., soft tissue abscesses), demonstrating the potential importance of relative tPMP resistance in clinical settings (1, 55). Of note, in both these prior experimental models and clinical investigations, a “biologic breakpoint” utilizing the same 2-h killing assay appeared to emerge. Thus, a survival profile in the range of >40% in this assay correlated well with (i) an endovascular source for bacteremia, (ii) more virulent infection in IE models, and(iii) reduced responsiveness to antimicrobial prophylaxis and therapy in these same models. Interestingly, in the present study, we observed an impressive statistical relationship between reduced tPMP-induced killing in vitro and the PB outcome, with the mean in vitro survival profile also greater than 40% for such strains. Finally, two recent clinical studies demonstrated a similar correlation between relative tPMP resistance and PB-RB outcomes in strains from both IE and non-IE infections (23, 44). Similarly, reduced in vitro susceptibilities to hNP-1 correlated with the PB outcome in our study, in accordance with our previous findings (56). We have postulated that relative resistances to such PMN HDPs may foster PB by favoring MRSA survival within PMN-rich abscesses (e.g., within the kidneys or spleen). Taken together, these data indicate a potential role for relative resistance to tPMPs and hNP-1 as a surrogate biological marker for the PB outcome.
Second, the ability of bacteria to bind to matrix ligands and certain host cells is felt to be an important virulence factor in S. aureus endovascular infections (56). In the present study, we were surprised to find no significant differences in fibronectin-binding profiles between the PB and RB strain sets. This is in contrast to our previous study, in which the PB outcome was associated with better fibronectin binding (56). Interestingly, a recent study showed that S. aureus isolates at either extreme of fibronectin-binding capacity were equally capable of establishing IE in an experimental model (61). Even though the ability to bind fibronectin is crucial for inducing endocarditis (48), the extent of the binding seems to play a minor role in IE induction and in persistence. In parallel to fibronectin binding, no significant difference in endothelial cell binding profiles was noted when the PB and RB cohorts were compared. The fibronectin-binding phenotype and/or expression profiles of genes involved in fibronectin binding may differ substantially between the in vitro conditions we tested and in vivo infection conditions. In addition, the ability to invade and survive inside endothelial cells might be of more importance for persistence in the host than adhesion. Also, the extent of the proinflammatory response induced by the bacteria might play a role. Studies are ongoing in our laboratory to address these questions.
Last, the extensive literature correlating indwelling endovascular devices, biofilm formation on such devices, and S. aureus bacteremia led us to examine the ability of the PB versus RB cohorts to form biofilms in vitro (12, 19, 21, 28, 34). Importantly, multiple linear and logistic regression data analysis showed that combining the extent of biofilm formation with reduced susceptibility to HDPs provided a phenotypic mosaic that was significantly correlated with both the PB outcome and increased days of bacteremia. Since biofilm formation restricts the access of many conventional antimicrobial agents (e.g., vancomycin) to sessile bacterial communities (12), it is reasonable to postulate that biofilm also mitigates the exposure of such colonies to HDPs generated at endovascular infection sites. Another interesting finding of our study is the overrepresentation of the sspA gene, encoding the V8 protease, in the PB strain cohort (100%) compared with the RB strain cohort (89%). V8 protease has been reported to be required for biofilm dispersal mechanisms, which in turn can play a role in colonizing new sites and favor recalcitrant infections (2).
Different factors, such as the production of PIA/PNAG by icaADBC operon-encoded enzymes (7); extracellular DNA (eDNA) (30, 38, 50); and several surface proteins, including Bap, SasG, FnBPs, and Spa (5, 8, 41, 46), have all been shown to be involved in S. aureus biofilm formation. In accordance with previous findings (35, 36, 47), we found that biofilm formation in our tested MRSA isolates depended more on the protein content than on the PIA/PNAG content. Biofilms of most of the isolates tested also contained eDNA, a recently described and important S. aureus biofilm component (30, 38, 50). Interestingly, our RB isolate biofilm matrices seemed to contain more PIA/PNAG than those of the PB isolates tested. Thus, the biofilm “component type” might be another useful marker for distinguishing PB from RB isolates. Further investigations are ongoing in our laboratory to investigate this notion.
We recognize that there are some important limitations in the present study. First, we assessed a relatively small number of PB-RB strains and phenotypes, affording a substantial chance for statistical bias. We also realize that in vitro testing of HDPs in sublethal concentrations much lower than would be encountered in vivo may also present limitations. We are currently examining larger PB-RB strain cohorts and additional phenotypic biomarkers to circumvent some of these issues. To complement our in vitro findings, in vivo studies are ongoing in our laboratories using an experimental IE model and selected isolates from our current MRSA strain cohort. These investigations are designed to address the issues of whether “PB” isolates are intrinsically more virulent than “RB” MRSA isolates or whether PB outcomes are relevant only during antimicrobial therapy. The outcomes from our previous experimental IE studies in other PB-RB strain collections would predict that the latter concept is more valid than the former concept (56).
We recognize that there is no clearly defined days of bacteremia “breakpoint” to distinguish PB and RB outcomes. As in our prior studies of PB and RB outcomes in other strain cohorts, we utilized the ≥7 days definition of PB for several reasons. First, our decision to define “persistent bacteremia” as ≥7 days is based on well-accepted definitions in the peer-reviewed literature (23, 27, 31, 37). Second, the definition is consistent with our own clinical experiences. The clinical signs and symptoms of patients with bacteremia lasting ≥7 days are typically quite different from those of patients with shorter bacteremic durations, even those with complicated infection (e.g., febrile morbidity). Thus, while far from perfect, our definition of PB is well accepted and allows investigators a reasonable yardstick for studying this fascinating clinical entity.
Collectively, our data strongly support our overall hypothesis that PB isolates have specific and combinatorial pathogenic “signatures” that are independent of conventional antimicrobial susceptibility. Importantly, the signatures that are predominant in one PB strain cohort (e.g., fibronectin binding ) may not be contributory in others, such as in our current strain set. Overall, defining common and consensus PB signatures among large well-defined strain collections might eventually be useful to identify patients at higher risk for PB to optimize anti-MRSA therapeutic strategies.
This work was supported by the Swiss National Science Foundation grant PBZHP3-123284 to K.S., American Heart Association grants SDG 0630219N and AID 09GRNT2180065 to Y.Q.X., and U.S. National Institutes of Health grants (AI-39108 to A.S.B., AI-39001 to M.R.Y., and AI-068804 to V.G.F.). Human endothelial cells were provided by the General Clinical Research Center in the Los Angeles Biomedical Research Institute of Harbor-UCLA Medical Center (NIH NCRR MO1-RR00425). We recognize the statistical support obtained for this project through the MO1-RR00425 grant to the General Clinical Research Center at Harbor-UCLA Medical Center.
We thank the laboratories of Scott G. Filler and Ashraf Ibrahim at Los Angeles Biomedical Research Institute at Harbor-UCLA Medical Center for providing endothelial cells. We also thank Catherine Sugar (UCLA) and Peter Christenson (Los Angeles Biomedical Research Institute at Harbor-UCLA Medical Center) for their guidance with statistical analysis of our data.
Published ahead of print on 22 November 2010.