Streptococcal Protein FOG, a Novel Matrix Adhesin Interacting with Collagen I in Vivo*

Group G streptococcus (GGS) is a human pathogen of emerging clinical significance. It causes skin and soft tissue infections, occasionally resulting in life-threatening conditions such as sepsis and necrotizing fasciitis. We recently identified FOG, a novel surface protein of GGS with fibrinogen binding and immune evasion properties. Here we investigated the role of FOG in streptococcal primary adhesion to host tissue. A FOG-expressing clinical isolate adhered more efficiently to human skin biopsies ex vivo and to the murine dermis in vivo than a FOG-deficient strain. Scanning and transmission electron microscopy of skin specimens exhibited that this property was assigned to the ability of FOG to interact with collagen I, a major interstitial component of the dermis. Overlay experiments with human skin extracts and radiolabeled FOG followed by matrix-assisted laser desorption/ionization time of flight mass spectrometry analysis identified both the α1- and α2-chains of collagen I as targets for FOG. Transmission electron microscopy of the molecular complexes revealed thread-like FOG molecules binding via their NH2 termini to distinct sites on collagen I monomers and fibrils. The results demonstrate that FOG is important for GGS adhesion in vivo, implying a pathogenic role for this surface protein.

Streptococcus dysgalactiae equisimilis, a group G streptococcus (GGS), 5 can cause mild infections like pharyngitis and more severe skin infections such as erysipelas and abscess formation. Deep soft tissue infections may result in life-threatening conditions such as sepsis and necrotizing fasciitis (for references see Ref. 1). Although certain cases of rheumatic fever have been linked to pharyngeal carriage of GGS (2), primary infections of the skin with these pathogens were often found to precede serious invasive conditions and bacteremia (3). Thus, the skin is to be considered an important site of entry. The clinical relevance of skin infections caused by GGS is underlined by epidemiological reports indicating that GGS are at least as frequent as group A streptococci (GAS) in such diseases (4,5).
Primary adhesion to the host tissue is a crucial initial step in bacterial infection. The firm interaction enables the bacteria to persist at the site of infection and subsequently colonize specific locations in the tissue. Certain streptococcal surface components with the potential to adhere to host tissues have been identified previously (for references see Refs. 6 and 7). For GAS a mechanism has been postulated in which lipoteichoic acid acts in concert with bacterial surface proteins to facilitate adhesion to host cells (8 -10). Frequently fibronectin has a key function by bridging bacterial adhesins and eukaryotic integrins (11)(12)(13)(14)(15). Interestingly, some of the M proteins, which are important multifunctional virulence factors of GAS, meet the criteria of adhesins (for references see Refs. 6 and 7). As an example, M1 protein involves both fibronectin and laminin in adhesion to human lung epithelial cells (16).
A recent study indicates that the majority of GAS strains use the adhesin SfbI to recruit collagen types I and IV via surface-bound fibronectin (17). Alternative mechanisms for collagen binding, such as a direct interaction with an M protein of the M3 serotype or binding via a capsule of hyaluronic acid, as seen for an M18 strain, have been published recently (18). Staphylococcus aureus has recently been reported to express the adhesin CNA to directly bind collagen in a mouse model of septic arthritis (19).
Despite the clinical and epidemiological importance and the occasionally dramatic pathogenesis of skin infections with GGS, molecular mechanisms for colonization of human skin by these pathogens are still elusive. A fibronectin-binding protein of GGS, GfbA, has been described earlier as a mediator of adhesion to human fibroblasts. The encoding gene was detected in 36% of tested strains, suggesting that there might be other tissue-specific adhesins expressed by these pathogenic streptococci (20). The recently discovered GGS surface protein FOG has been shown to share sequence homology and structural similarity with M-like proteins of other streptococci. Using primers from the signal sequence and the cell wall anchoring domain of FOG, a PCR product corresponding to the 1.8-kilobase pair FOG gene was generated in 97% of examined clinical isolates of GGS (21).
In the present study, we provide evidence that GGS adhere to collagen I fibrils of the dermis in vivo employing protein FOG as an adhesin. Binding sites essential for this interaction were identified in the NH 2terminal part of FOG and on two distinct positions on the collagen I triple helix. Taken together, the data demonstrate that this surface protein is an important virulence determinant with the potential to trigger streptococcal pathogenesis in vivo.

MATERIALS AND METHODS
Bacterial Strains-Human clinical isolates of GGS were collected at the Department of Clinical Microbiology of Lund University Hospital (Lund, Sweden). Strains G10, G11, G14, and G19 were joint isolates, G30 and G35 were isolated from wounds, and G41 and G148 were throat specimens. The bacteria were cultured in Todd-Hewitt broth (Difco) at 37°C and 5% CO 2 . Strain G148 lacks the fog gene (21) and served as a negative control.
Recombinant Proteins-Purification of the FOG variants was performed as described (21). FOG 1-C (amino acid residues 1-493) lacks the COOH-terminal D-domain including the bacterial cell wall anchoring LPXTG motif, and FOG 1-B covers the NH 2 -terminal 278 residues (see Fig. 3E). The recombinant proteins carry an NH 2 -terminal GST tag used for protein isolation.
Radiolabeling and Bacterial Binding Assay-Collagen I from calf skin (Vitrogen Inc.) and FOG 1-C were radiolabeled with 125 I using IODO-BEADS (Pierce). The labeling reactions were carried out on ice for collagen I and at room temperature for protein FOG using 20 g of protein in 30 mM phosphate (KH 2 PO 4 /Na 2 HPO 4 ), 120 mM NaCl, pH 7.4 (PBS), and 5 min of initial incubation. Unbound iodine was removed on a PD10 column (Amersham Biosciences). 125 I-Labeled proteins were bound to streptococcal cells as described earlier (22). All of the reactions were performed on ice. For competition assays, the concentration of bacteria was first optimized to 3 ϫ 10 8 bacteria/ml, and subsequently, using a fixed amount of 125 I-Labeled collagen I, increasing amounts of binding competitors were added. CDC (14 kDa), the IgG-binding part of streptococcal protein G, was purchased from Amersham Biosciences, and BSA (66 kDa) was from Sigma. Full-length FOG without GST tag was used. The procedure was as described (22) and performed on ice.
Human Skin Samples-Punch biopsies (4-mm diameter) of human skin were obtained in connection with skin transplant surgery. Informed consent was obtained from the patients. The Ethics Committee at Lund University approved the use of this material (permit numbers LU 509-01 and LU 708-01).
Ligand Blot and Slot Blot-Human skin components were extracted by boiling in 20% SDS for 20 min. Solubilized proteins were separated by reducing SDS-PAGE on a 3-12% gradient gel in the presence of 5% ␤-mercaptoethanol and stained with Coomassie Brilliant Blue or electroblotted onto a PVDF membrane (Immobilon; Millipore). The membrane was washed twice for 30 min in blocking buffer (PBS containing 0.25% Tween 20 and 2% bovine serum albumin), incubated with 125 I-FOG 1-C overnight at 4°C, and washed three times for 20 min in blocking buffer. Signals of bound ligand were detected using the Fuji Imaging System. The bands corresponding to the signals on the blot were excised from the Coomassie-stained gel and prepared for mass spectrometry. Collagen I denatured with guanidine hydrochloride and nondenatured collagen I were applied directly onto a PVDF membrane using the Milliblot-D system (Millipore). The membranes were blocked, washed, and incubated with radiolabeled probe as described above.
Mass Spectrometry-Excised gel bands were washed in water and destained with 40% acetonitrile in 25 mM NH 4 HCO 3 , pH 7.8 (Buffer A). The gel pieces were vacuum-dried and rehydrated in 10 -20 l of 10 mM dithiothreitol (55°C, 30 min). Excess liquid was removed and the reduced proteins were alkylated using 10 -20 l of 55 mM iodoacetamide (20°C, 30 min). The samples were then washed with buffer A and vacuum-dried. Rehydration and overnight incubation at 37°C with 10 -20 l of buffer A containing sequencing grade trypsin (25 ng/ml; Promega) ensued digestion into peptides, which were extracted in 10 l of 2% trifluoroacetic acid and purified using C18 Ziptips (Millipore). Precrystallized 2,5-dihydroxybenzoic acid was provided on an Anchorchip target (Bruker Daltonik GmbH). The purified peptides were eluted onto this matrix using 50% acetonitrile, 0.1% trifluoroacetic acid and allowed to co-crystallize. Spectrometry was carried out on a Bruker Reflex III matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometer operated at an acceleration voltage of 26 kV in the positive ion/delayed extraction mode and detection in the reflector mode. Autolysis fragments of trypsin were used for external calibration of the spectra that were summed of 75-100 single-shot measurements each. The proteins were identified using the Web tool ProFound (23). The settings accounted for a complete alkylation of the cysteine residues, partial oxidation of methionine, and hydroxylation of lysine and proline. Missed cuts were excluded. Other parameters were used in their default settings.
Surface Plasmon Resonance Measurements-Interactions were studied in the BIAcore 2000 system (BIAcore AB) using 10 mM HEPES, 100 mM NaCl, pH 7.4, as running buffer. A CM5 sensor chip was activated by a 4-min injection of 0.05 M N-hydroxysuccinimid, 0.2 M N-ethyl-N-(3-dimethylaminopropyl)-carbodiimide hydrochloride in water. Collagen I from calf skin (Roche Applied Science) was diluted in 15 mM NaAc, pH 5.3, to a final concentration of 15 g/ml. Injection of 5 l at a flow rate of 5 l/min lead to immobilization of 400 response units of collagen I. Residual reactive groups were inactivated by 6-min injection of 1 M ethanolamine, 0.1 M NaHCO 3 , 0.5 M NaCl, 5 mM EDTA, pH 8.0. Interaction measurements with FOG, FOG 1-C, FOG 1-B, and GST, respectively, were performed at a flow rate of 50 l/min. No mass transfer effects were detected. Surface regeneration was achieved by injection of a 30-s pulse of 0.2% SDS in water. The data were analyzed using the BIAevaluation 3.0 software and shown curves represent the difference between the signal of the collagen-coupled surface and a deactivated control surface devoid of protein. They were further corrected by subtraction of the curve obtained after injection of buffer alone.
Bacterial Binding to Human Skin Sections-Skin biopsies were fixed in PBS containing 4% formaldehyde (18 h, 4°C) and processed for paraffin sectioning. Deparaffinized and rehydrated 5 m sections were incubated with bacterial suspensions (2 ϫ 10 9 cfu/ml) in PBST (PBS containing 0.05% Tween 20) for 1 min, 15 min, 30 min, 1 h, or 2 h at room temperature. Prior to experimental colonization conditions, 1 h of primary adhesion was carried out. After removing nonbound bacteria with PBST, a drop of Todd-Hewitt broth medium was placed on the sections, and the slides were incubated at 37°C, 5% CO 2 for 5 h to allow bacterial multiplication. In both types of experiments nonbound bacteria were finally removed by extensive washing with PBST. The samples were fixed in PBS containing 4% formaldehyde and prepared for light microscopy. The samples for scanning electron microscopy were fixed and processed as described below.
Bacterial Colonization of Human Skin Biopsies-Skin biopsies were incubated with bacteria (2 ϫ 10 9 cfu/ml) in PBST for 1 h at room temperature. After removal of nonbound bacteria by washing three times with PBST, the infected skin was cultivated for 24 h in Dulbecco's modified Eagle's medium (Invitrogen) at 37°C, 5% CO 2 . The samples were washed three times in PBS, and specimens were either fixed in PBS containing 4% formaldehyde prior to light microscopy or prepared for scanning electron microscopy as described below.
Primary Adhesion to the Collagen Matrix of Human Fibroblasts-Bacteria were grown in Todd-Hewitt broth to early stationary growth phase, washed three times with PBS, and resuspended in PBS to 2 ϫ 10 9 cfu/ml. Human foreskin fibroblasts were cultured in 24-well tissue culture plates (Nunc) in minimum essential medium (Invitrogen) supplemented with 10% fetal bovine serum, 100 units/ml penicillin, 100 M streptomycin, and 0.1 mM L-glutamine. Specimens for scanning electron microscopy were cultured under identical conditions on coverslips placed at the bottom of the wells. All of the reagents were from Invitrogen. When cells approached confluence, a daily dose of ascorbic acid (25 g/ml) was added for 3 days to stimulate the expression of collagen networks. Cell monolayers were washed with PBS and incubated with streptoccci (6 ϫ 10 8 cfu/well) for 1 h at 37°C, 5% CO 2 , and nonbound bacteria were removed with PBS. Analysis of adherent bacteria was performed either by determining cfu as described (24), or by scanning electron microscopy.
Primary Adhesion to the Murine Dermis in Vivo-Female Balb/c mice of 6 -8 weeks were anesthetized with isoflurane. Groups of three mice each received a subcutaneous dorsal injection of 6 ϫ 10 8 cfu G41 or G148. The area of injection was marked with a pencil for later identification. After 30 min the animals were sacrificed, and the marked area was excised. For determination of adherent bacteria skin pieces were washed with PBS and homogenized in a volume of 400 l of PBS using a Multi-Gen 7 homogenizer (Pro Scientific Inc. Monroe, CT). The samples were vortexed, diluted, and plated in duplicate on blood agar plates. The plates were incubated overnight at 37°C, and the number of cfu was counted. For determination of contaminating normal flora one mouse was injected with PBS alone and treated as above. In parallel, skin samples were prepared for examination by scanning electron microscopy. The animal experiments were approved by the regional ethical committee for animal experimentation (permit M294-03).
Light Microscopy-For light microscopy of colonized skin biopsies, 5-m paraffin sections were stained with Mayers Hematoxylin (Histolab AB) and Eosin Y (Surgipath Inc.). Alternatively, using the affinity of protein G to goat IgG, the bacteria were detected with the Vectastain ABC Elite kit (Vector Laboratories Inc.) and subsequent staining with a DAB kit (Vector Laboratories Inc.), following the manufacturer's instructions. The samples were examined in a Carl Zeiss Axioplan 2 imaging e microscope. Digital images were acquired using a Carl Zeiss AxioCam HR camera and the software Axiovison 3.1 (Zeiss).
Scanning Electron Microscopy-Deparaffinized skin sections on glass slides were incubated with bacteria as described above. The whole slide was fixed in 4% formaldehyde, 2.5% glutaraldehyde in PBS, for 2 h at 4°C. Fixed specimens were dehydrated for 10 min at each step of an ascending ethanol series and critical point dried in a Balzers critical point dryer in liquid carbon dioxide using absolute ethanol as intermediate solvent. The glass slides were cut into appropriate pieces with a diamond knife, mounted on aluminum stubs, and coated with a 30-nmthick layer of gold. They were examined in a Jeol J-330 scanning electron microscope at an acceleration voltage of 5 kV and a working distance of 10 mm.
Transmission Electron Microscopy-Skin biopsies incubated with bacteria (see above) were fixed, embedded, and sectioned with a diamond knife as previously described (21). The structure of FOG constructs and complexes thereof with collagen I was analyzed by mica sandwich squeezing/rotary shadowing (25). Complexes between FOG, labeled with colloidal gold (26), and reconstituted collagen I fibrils were visualized by negative staining as described (27). To prepare collagen I fibrils an acidic solution of collagen I from fetal calf skin (3 mg/ml; Roche Applied Science) was neutralized with PBS to a final concentration of 25 g/ml. Reconstitution of fibrils was carried out for 24 h at 37°C. Sample concentrations were usually in the range of 10 -20 g/ml in 50 mM Tris-HCl, 0.15 M NaCl, pH 7.4 (TBS). Collagen I fibrils and gold-labeled FOG were incubated in TBS buffer for 30 min at 4°C. Specimens were observed in a Jeol JEM 1230 electron microscope operated at 60 kV accelerating voltage. The images were recorded with a Gatan Multiscan 791 CCD camera.

RESULTS
GGS Expressing FOG Adhere to Human Dermis ex Vivo-The FOGexpressing GGS strain G41 was compared with the FOG-deficient strain G148 (21) for the ability to adhere to and subsequently colonize human dermis ex vivo (Fig. 1). Bacteria were allowed to adhere to thin   4) were analyzed by reducing SDS-PAGE. One gel was stained with Coomassie (STAIN), and the other gel was transferred to a PVDF membrane (BLOT) and incubated with radiolabeled FOG 1-C. M r standards, origin, and front (arrows) of the separating gel are indicated to the left. Indicated are also ␣1and ␣2-chain and ␤and ␥-fragments of collagen I. Lines a, b, and c indicate the mobility of dermis components that interact with FOG and were then analyzed by MALDI-TOF mass spectrometry. B, collagen I denatured in a solution of 4 M guanidine hydrochloride (ϩ) and nondenatured collagen I (Ϫ) were applied in slots to a PVDF membrane and incubated with radiolabeled FOG 1-C. The amount of applied collagen I is indicated to the right.

Protein FOG Interacts with Collagen I in Vivo
sections of human skin biopsies on glass slides. For subsequent colonization the adhered bacteria were multiplied by incubation on the slides with culture medium. As shown by light microscopy, G41 colonies adhered to the fibrous matrix of the dermis, whereas binding of G148 was considerably less pronounced (Fig. 1A). This was confirmed by scanning electron microscopy of infected skin sections at higher resolution (Fig. 1B). Large numbers of colonies of G41 bound to fibrillar structures in the dermis (arrowheads). In contrast, only a few colonies of G148 could be detected. Fresh human skin biopsies were also infected with G41 or G148, followed by cultivation to simulate colonization of the tissue. With light microscopy the formation of large colonies in the dermis was obvious (arrowheads), where colonization by strain G41 was much more pronounced as compared with G148 (Fig. 1C). Scanning electron microscopy (Fig. 1D) demonstrated a frequent intimate association of G41 colonies with fibrillar networks (inset), which was observed to a much lesser extent for G148.
FOG Binds to Collagen I ␣1and ␣2-Chains of Skin Extracts-To identify skin components serving as substrates for FOG, we set up ligand blot experiments (Fig. 2A). The FOG 1-C fragment lacking the COOHterminal membrane anchor domain (Fig. 3E) was used to minimize potential self-aggregation by hydrophobic interactions. Human skin biopsies were extracted with SDS, separated by SDS-PAGE using bovine collagen I as a standard, and transferred to a PVDF membrane. Interestingly, radiolabeled FOG 1-C bound only to a few discrete bands of the skin extract with apparent molecular masses of 129, 141, and 207 kDa, respectively ( Fig. 2A, BLOT). In the control lane FOG 1-C interacted with the ␣1and ␣2-chains as well as ␤and ␥-fragments and slower migrating fragments of bovine collagen I. Comparison with the Coomassie-stained gel showed that the bands exhibited a similar migration pattern as the ␣1and ␣2-chains and the ␤-fragment of bovine collagen I ( Fig. 2A, STAIN).
Corresponding bands on the Coomassie-stained gel were excised, digested with trypsin, and analyzed by MALDI-TOF mass spectrometry. The obtained peptide fingerprints identified all bands as collagen I ( Table 1). The data evaluation took into account the hydroxylation of proline and lysine residues characteristic for collagens. This led to uncertainties in the exact peptide assignment of the two homologue collagen I ␣-chains. However, the combination of the MALDI-TOF fingerprint data and the migration pattern in SDS-PAGE ( Fig. 2; for identification of the bands see Ref. 22) led to the assignment presented in Table 1, concluding that both collagen I ␣-chains are targets for FOG.
To test the influence of collagen denaturation occurring in SDS-PAGE on the FOG-collagen interaction, the binding of FOG 1-C to nondenatured and to guanidine hydrochloride-denatured bovine collagen I was examined in a slot blot experiment. Radiolabeled FOG 1-C bound to both collagen preparations in a similar concentration-dependent manner (Fig. 2B).

FOG Binds to the Collagen I Triple Helix in Surface Plasmon
Resonance Experiments-GST-tagged recombinant full-length FOG, as well as the shorter variants FOG 1-C and FOG 1-B, were tested for binding to immobilized collagen I. All of the fragments exhibited a dose-dependent binding to collagen I (Fig. 3, A-C), whereas the GST tag alone did not bind (Fig. 3D). Binding curve fits based on the 1:1 model of Langmuir revealed apparent dissociation constants of 8 ϫ 10 Ϫ11 M for FOG, 4 ϫ 10 Ϫ10 M for FOG 1-C, and 3 ϫ 10 Ϫ10 M for FOG 1-B, respectively. Binding of the FOG 1-B fragment localized the site of collagen binding to the NH 2 -terminal half of FOG (Fig. 3E).
Structural Characterization of FOG-Visualization of GST-tagged recombinant full-length FOG and its truncated variants FOG 1-C and   amino acids (Fig. 3E), resulting in averages of 1.0 Ϯ 0.1, 0.8 Ϯ 0.1, and 0.8 Ϯ 0.1 Å for the axial distances between vicinal amino acids, respectively. In summary the different recombinant proteins consist of an NH 2 -terminal globular GST tag and a protruding thread-like structure formed by the FOG moiety that is more tightly packed than expected for an ideal ␣-helix (1.5 Å axial distance). Like the closely related M-proteins, the thread-like FOG moieties might represent coiled-coil dimers.

FOG Binds to Distinct Sites on the Collagen I Triple Helix-
The interaction between FOG and collagen I was visualized by transmission electron microscopy after mica sandwich squeezing/rotary shadowing (Fig.  4A). Large fields of FOG molecules in association with the 300-nm-long collagen I molecules became apparent. The contact between FOG and collagen I occurred at the very NH 2 -terminal portion of FOG, which is marked by the globular GST tag. Consistent with the surface plasmon resonance measurements, an interaction between GST alone and collagen I was not observed (data not shown). Therefore we concluded that the site of interaction resided in the NH 2 -terminal of the FOG moiety, without any contribution by the GST tag. On the collagen I triple helix two distinct binding sites for FOG were found (Fig. 4B). Because pepsinextracted collagen I was used we could not distinguish between NH 2and COOH-terminal. However, collagen molecules were found with both binding sites occupied by FOG (Fig. 4B, bottom panels). They allowed defining the relative position of the two binding positions, which are located on one tip and at 235-nm distance on the 300-nmlong triple helix (Fig. 4C). Additional binding sites were not observed.
The truncated recombinant variants FOG 1-C and FOG 1-B exhibited the same interactions with collagen I (data not shown).
FOG Binds to Distinct Sites on Collagen I Fibrils-For a more detailed understanding of GGS adhesion to dermis collagen networks in situ, complexes between FOG and reconstituted collagen I fibrils were visualized by electron microscopy after negative staining with uranyl formate (Fig. 5). To facilitate identification of individual FOG molecules on the collagen fibrils, FOG was conjugated with 4-nm colloidal gold. Both negative staining (Fig. 5A) and positive staining (Fig. 5B) of collagen fibrils were observed on different locations on the grid, resembling the banding pattern after negative staining with sodium phosphotungstic acid, indicating the polarity of the fibrils (22). On negatively stained fibrils gold-labeled FOG was found at the borders of the more intensely stained gap regions (Fig. 5A). Consistently, binding close to the c 2 -band and to the quartet of a-bands of positively stained fibrils was observed (Fig. 5B) (for assignment of the bands see Ref. 28). Evaluation of 300 complexes confirmed the preferential localization of FOG to the border between gap and overlap regions on the collagen fibril (Fig. 5C). A schematic representation (Fig. 5D) corroborates this localization with the  binding sites determined on collagen I monomers, despite the above mentioned uncertainty to distinguish between NH 2 -and COOH-terminal (Fig. 4). Binding of FOG either to the NH 2 terminus of the triple helix or to the COOH terminus, together with the corresponding site at 235-nm distance, in each case results in the observed FOG binding pattern on the striated collagen I fibrils.
GGS Expressing FOG Bind Directly to Collagen I in Vitro-Eight GGS clinical isolates originating from various infection sites were tested for their ability to bind soluble radiolabeled collagen I. Seven FOG strains bound between 30 and 50% of the added collagen (Fig. 6A). The FOGdeficient isolate G148 bound considerably less collagen I (11%). Preincubation with fibronectin did not enhance collagen binding to G41 (data not shown). Taken together, these findings demonstrate the specificity of the direct interaction of FOG-expressing GGS with collagen I without requiring fibronectin as a bridging molecule. This specificity was further confirmed by inhibition experiments in vitro with soluble components. Radiolabeled collagen I was added to G41 bacteria in the presence of increasing amounts of full-length FOG, the IgG-binding domain of streptococcal protein G (CDC), or BSA, respectively. In the presence of CDC or BSA, the bacteria were able to recruit collagen I, whereas the addition of FOG resulted in a significant inhibition (Fig. 6B).
GGS Expressing FOG Adhere to Collagen I Fibers in the Human Dermis-The kinetics of primary adhesion of the strains G41 and G148 to human dermis was visualized by scanning electron microscopy. Numbers of bound colonies/mm 2 on dermis thin sections were quantified in the microscope and determined at different time points during 2 h. Both strains exhibited a time-dependent adherence, where G41 bound considerably faster than G148. After 2 h, adhesion of G41 resulted in a significantly higher density of 1800 colonies/mm 2 , as compared with only 200 colonies/mm 2 for strain G148 (Fig. 7A). Preincubating bacteria with soluble collagen I decreased the adhesion of G41, whereas no such effect was observed for G148, emphasizing the specific interaction between G41 and the dermal collagen I network (Fig. 7B).
Next, streptococcal primary adhesion to collagen fibers in situ was examined at higher resolution by embedding infected human skin biopsies, followed by ultrathin sectioning and transmission electron microscopy. G41 bacteria intimately associated with cross-striated collagen I fibers in the dermis (Fig. 8A), whereas G148 bound to a much lesser extent to noncollagenous dermis components (Fig. 8B). The contact between G41 and collagen was mediated by FOG protruding from the surface of G41 as elongated hair-like structures (Ref. 21 and Fig. 1B). In areas with optimal contrast and resolution, these FOG structures were observed bound to the gap regions of the collagen fibrils (Fig. 8A, inset). This is consistent with the observation that gold-labeled FOG binds to the borders of the gap regions of reconstituted collagen I fibrils in vitro (Fig. 5).
GGS Expressing FOG Adhere to Fibrillar Collagen I Networks in Vivo-To investigate the in vivo relevance of our findings, we allowed streptococci to adhere to collagen I networks of human fibroblasts in monolayer culture. Scanning electron microscopy showed extensive binding of G41 bacteria to the pericellular collagen matrix (Fig. 9A) as opposed to the considerably weaker binding of G148 (Fig. 9B). These findings were confirmed by determining numbers of cfu of adherent bacteria (Fig. 9E, left panel). In further experiments we examined streptococcal primary adhesion in an in vivo skin infection model. Balb/c mice received subcutaneous dorsal injections of G41 or G148 streptococci, followed by a 30-min incubation time, which was considered short enough to minimize leukocyte infiltration. Bacterial adherence to the infected areas was examined by scanning electron microscopy and quantified by determination of cfu. Colonies of G41 streptococci were frequently associated with the fibrillar collagen I-containing networks in the dermis (Fig. 9C). In contrast, G148 bacteria interacted with the dermis to a much lesser extent (Fig. 9D). Similar results were obtained by quantification of GGS adherence by counting cfu of homogenized infected skin areas (Fig. 9E, right panel).
Taken together, our findings demonstrate the specific interaction of the GGS surface protein FOG with the collagen I network of the dermis in vivo. This property facilitates primary adhesion of GGS to the host tissue and implies an advantage in pathogenesis of FOG-expressing GGS.

DISCUSSION
The human skin is an important site of entry for GGS infections. In this study we show that the bacterial surface protein FOG is a novel matrix adhesin possessing an important role for GGS primary adhesion to the human dermis. FOG binds specifically to both collagen I ␣-chains on two distinct locations of the collagen monomer. Consequently, collagen fibrils provide two binding sites for FOG in the gap region and serve as substrates for GGS adhesion in vivo. This implies that upon skin injury, when the complex collagen matrix in the dermis is exposed, these pathogens can establish primary adhesion to collagen I fibrils in the wound. The immune evasion properties of FOG (21) may facilitate subsequent colonization of the site of infection. Moreover, because collagen I is a ubiquitous constituent of other connective tissues, this collagen is an attractive target for GGS primary adhesion to a variety of locations in the human body.
A recent report describes the ability of GAS to interact with collagen via surface-bound fibronectin, resulting in bacterial aggregation and immune evasion (17). The majority of the GAS strains exhibited fibronectin-mediated collagen recruitment, whereas only a minority showed direct interaction with collagen IV or collagen I. In contrast, our observations demonstrate a direct interaction of GGS with collagen I fibrils in vivo. This feature enables GGS to infect human skin without the necessity to recruit bridging molecules, which might not be available at certain infection sites. Examination of eight clinical strains indicates that direct binding to collagen I is a common feature among GGS isolates. Because all collagen-binding strains express the fog gene, but not strain G148 that does not bind collagen, we assume a general role for FOG in mediating this interaction. This is supported further by the observation that in an additional 30 tested clinical isolates PCR products, corresponding to the size of the fog gene, were obtained using fog-specific primers (21).
Ligand blot experiments combined with MALDI-TOF mass spectrometry identification revealed FOG binding to the genetically distinct ␣-chains of collagen I, despite denaturation during SDS-PAGE. This shows that the minimal structural motif essential for the interaction is present in both collagen I ␣-chains and is independent of the threedimensional triple helical structure. Binding of FOG to triple helical collagen I, as shown by ligand blot, electron microscopy, and surface plasmon resonance, indicates that the assembly of the ␣-chains into a triple helix does not lead to sterical impairment of the interaction. A binding site for collagen I resides in the NH 2 -terminal half of FOG as shown by surface plasmon resonance measurements. Transmission electron microscopy after rotary shadowing suggests that the site of interaction is located at the very NH 2 -terminal end of FOG. On collagen I two distinct binding sites were identified. Electron microscopy further indicated an interaction of FOG with collagen I fibrils. Thus, FOG is capable to bind to all different collagen I structures that can occur in severed skin. Whereas binding to the fibrillar collagen network implies a function in dermal adhesion, recruitment of soluble forms of collagen I might contribute to biofilm formation and might have additional functions in immune evasion.
The finding that FOG-expressing G41 adhere to the collagen matrix of the dermis in vivo further emphasizes the importance of this collagen as a target for primary adhesion. It can be expected that further characterization of the FOG-collagen interaction will contribute to understand how sequels of infections with GGS develop. An interesting question is whether FOG behaves promiscuously in its binding to different collagen types. Preliminary data do in fact indicate that this might be the case. This feature could broaden the spectrum of potential GGS infection sites and make the bacteria less dependent on local variations in tissue composition and availability of specific tissue components. In this context it is noteworthy that GAS with rheumatogenic potential has recently been shown to interact with collagen IV. A direct association of the M3 protein and the hyaluronic acid capsule with collagen IV was suggested as a basis for poststreptococcal rheumatic disease (18). An emerging general importance of collagen adhesins as virulence factors is supported by another recent report describing the direct interaction of S. aureus with collagen II (19). This feature was shown to be mediated by the adhesin CNA and promoted early colonization of joints of mice in vivo.
Here we show that once FOG-carrying GGS have passed the epithelium, they possess an efficient mechanism to adhere to and colonize the connective tissue of the host. Comparing the results obtained with GGS strains isolated from different sites of infection did not show a connection between the ability to bind collagen I and to the site of infection. The observation is consistent with the wide tissue distribution of the matrix protein. Factors that determine the tissue specificity of different GGS strains remain to be discovered. A possible mechanism might be multiple adhesins participating in GGS colonization in vivo and potentially contributing to tissue specificity.
The processes leading from colonization to development of pathological GGS infections are not fully understood yet. Age and underlying diseases, like diabetes, or injury (burn, chronic ulcer) are common risk factors for the host. Apparently, a particular susceptibility of those patients depends on impaired immunity, possibly in combination with the circumvention or loss of the protective epithelium. The pathogenesis of GGS infections is also determined by the evolved bacterial modes of immune evasion, acquisition of nutrients, or dissemination. The interaction described in this work probably contributes to those additional mechanisms of pathogenesis, which motivates further investigations.