Proteomic analysis of in vivo biofilm matrix
Through our previous study, the exteriors of mouse catheter-associated biofilms observed under scanning electron microscope (SEM) were distinct upon receiving different treatments, enabling us to speculate that biofilm matrix components were different in each of the samples9. Protein expression profiles from mouse subcutaneous catheter and in vitro biofilm samples were then initially analyzed using SDS-PAGE (Fig. S1). An abundant protein band at 66 kDa was noted in all in vivo samples and identified as albumin. Although strain ATCC 12598 used in this study is PIA-producing, the protein components were fundamentally different between in vivo and in vitro biofilms, as well as their appearance under SEM (Fig. S2). In samples with equal wet weights, the amount of proteins was approximately seven-fold higher in the in vivo biofilm matrix than in the in vitro biofilm matrix using the Bradford assay. A proteomic analysis was then employed to identify proteins within the in vivo biofilm matrix from healthy mice, and the results are listed in Table 1. Protein abundance was normalized based on total spectral count/protein molecular weight, and the relative quantification was compared against the most abundant histone H4. Although more than one thousand staphylococcal proteins were identified from the in vivo biofilm matrix, their abundance was much lower than that of host proteins and did not show a strong association with biofilm formation among the top five most abundant staphylococcal proteins (Table 1).
Albumin is specifically used as a biofilm matrix component by S. aureus
As shown in Fig. S1, albumin was the most abundant protein in the in vivo biofilm samples. A static biofilm assay was performed to investigate whether albumin was indeed incorporated into the biofilm matrix in vitro. Sequence alignment indicates that mouse serum albumin (MSA) shares 72% and 69% identity with human and bovine serum albumin (HSA and BSA), respectively, whereas it shares 78% identity between HSA and BSA. Therefore, BSA was used in most of our experiments unless indicated otherwise. The normal values of HSA are 3.5 to 5.0% (w/v), whereas they are 2.0 to 2.7% for MSA15,16. Then, the concentration of albumin used in this study was 2.5% or 5.0% as indicated.
Upon (2.5%) BSA treatment, biofilm formation was significantly increased in two S. aureus strains (12598 and SA113), even in their PIA nonproducing counterparts (Δica) and Escherichia coli, but not in Staphylococcus epidermidis and Streptococcus pneumoniae (Fig. 1a). Similar results were observed when BSA was replaced by HSA (2.5%) or human serum (Fig. 1b). Compared with the control groups, biofilm formation was only slightly increased in the presence of albumin-depleted human serum, suggesting that albumin played a major role in biofilm formation in the serum (Fig. 1b). Gelatin (~ 95 kDa) and tryptone were then employed to investigate whether albumin was specifically incorporated into the biofilm matrix, and there was no significant effect (Fig. 1c). Hemoglobin is also one of the abundant proteins within the biofilm matrix (Table 1), whereas γ-globulin is the second most abundant serum protein. Their impact on biofilm formation was investigated. Biofilm formation was significantly increased upon hemoglobin treatment, whereas there was no effect on γ-globulin (Fig. 1c). A highly structured and thick albumin-based biofilm matrix was also observed in both PIA-producing and nonproducing strains under SEM (Fig. 2).
eDNA and FnBP play important roles in the formation of albumin-based biofilms
Albumin had no effect on biofilm formation either when strain SA113ΔcidA (eDNA release deficiency) was employed or SA113 was treated with DNase I. Biofilm-forming capacity was recovered and even stronger in strain SA113ΔcidA when exogenous DNA was added, indicating that eDNA was essential in albumin-based biofilm formation (Fig. 3a).
We then investigated whether SrtA-mediated cell wall-anchored proteins interacted with albumin. Biofilm formation was significantly decreased among icaA, srtA, and icaA/srtA knockout strains and was recovered only in strain SA113ΔicaA when albumin was added (Fig. 3b). The above strains were then employed to compare their adhering capacity between albumin-coated and noncoated well surfaces. Adherence of both SA113 and SA113ΔicaA to albumin-coated wells was increased, whereas it was dramatically decreased for strains SA113ΔsrtA and SA113ΔicaA/srtA regardless of the coating with albumin (Fig. 3c).
The above results suggested that albumin may interact with SrtA-mediated cell wall-anchored proteins. Cell wall proteins extracted from strains SA113 and SA113ΔsrtA were employed in the albumin pull-down assay, and protein bands that appeared only in the wild-type sample were then identified using LC-MS/MS. Among them, fibronectin-binding protein A (FnBPA, encoded by fnbA) and protein A (encoded by spa) were further studied (Fig. S3). Two mutant strains, SA113ΔfnbAB and SA113Δspa, were generated and employed in the experiments described above. Since FnBPA and FnBPB share sequence and functional similarity, genes encoding these two proteins were knocked out simultaneously17. Either biofilm formation or bacterial adherence to albumin noncoated wells was significantly decreased in both mutants. However, biofilm-forming capacity and bacterial adherence were recovered in strain SA113Δspa but not in SA113ΔfnbAB in the presence of albumin, suggesting that FnBP may interact with albumin (Figs. 3d and 3e). This interaction was further characterized through blockage with anti-FnBPA antibody. Adherence was gradually blocked in the dilution fold at 64 (from 1 μg/ml of antibody) and below on the albumin noncoated wells. The binding of FnBPA to albumin was significantly blocked by the anti-FnBPA antibody at the same dilution and below in the albumin-coated wells (Fig. 3f).
Whether production of PIA was essential in the presence of albumin was investigated, and it was obviously reduced in the presence of albumin (Fig. 4a). Significantly decreased icaA (0.3-fold) and increased icaR (encoding a repressor of the ica operon; 2.8-fold) expression were observed 2 h post-treatment with albumin. A significant increase in cidA expression (10.0-fold) was also detected 1 h post-treatment suggesting that albumin can promote the release of eDNA (Fig. 4b).
Albumin-based biofilms show superior antibiotic resistance
We then compared the antibiotic resistance capacity of embedded bacteria between PIA- and albumin-based biofilms by determining the minimal bactericidal concentrations (MBCs) of different antibiotics (Table 2). The parental strain ATCC12598 and its derived mutants were susceptible to all the antibiotics used in this study under planktonic growth according to CLSI breakpoints regardless of the presence of albumin (Table S1)18. Generally, bacteria grown within albumin-based biofilms showed a superior antibiotic resistance capacity (Table 2). Biofilms produced by various mutants with either impaired PIA production or eDNA release or both showed markedly reduced antibiotic resistance (2 to 16-fold). Antibiotic resistance capacity was recovered in biofilms produced by 12598ΔicaA in the presence of albumin, while albumin had no or mild effect on the resistance in the two remaining mutants (Table 2).
In vitro biofilm nonproducers can form biofilms in vivo
Generally, biofilm-forming capacity among clinical isolates was evaluated through an in vitro assay system. Isolates with impaired PIA production or eDNA release somehow seemed to be weak biofilm producers or nonproducers in vitro19,20. Whether such strains can form biofilms in vivo if they can acquire host albumin or DNA from damaged cells was investigated.
As shown in Fig. 3, biofilm-forming capacity was significantly reduced in vitro when ica, cidA, and fnbAB knockout mutants were tested. The capacity was only recovered in SA113Δica in the presence of BSA. There was no difference in planktonic growth among those strains (data not shown). The above in vitro biofilm weak or nonproducers were then employed in an in vivo mouse model. Compared with SA113, the biofilm-forming capacity was decreased in SA113Δica three days postinoculation, but it was similar to that of the wild type seven days postinoculation (Fig. 5a). The capacity was significantly decreased when either ΔcidA or Δica/ΔcidA mutants were employed in the assay. However, the capacity was recovered to a level similar to that of their wild-type counterpart when xenogeneic DNA was coinoculated with bacteria simultaneously. When ΔfnbAB mutants were employed, the biofilm-forming capacity significantly decreased regardless of the presence of exogenous DNA (Fig. 5b).