Introduction

Streptococcus pyogenes (group A streptococcus, GAS) is one of the most prevalent human pathogens. It causes acute diseases ranging from trivial to life-threatening, such as pharyngotonsillitis (strep throat), scarlet fever, impetigo, meningitis, necrotizing fasciitis and streptococcal toxic shock-like syndrome [1]. The highest disease burden is caused by the post-infection sequelae acute rheumatic fever, rheumatic heart disease and glomerulonephritis [2, 3] but there is also a worrying global rise of severe invasive infections [4]. A key virulence factor involved in many if not all these diseases is the GAS M protein. Encoded by the chromosomal emm gene, the M protein is covalently anchored at its C-terminus to the cell wall, covering the GAS surface at high density and acting as an adhesin and a potent immune evasion factor [5–7]. M proteins are known, or can confidently be predicted, to form linear fibrils extending approximately 50 nm from the bacterial surface [8, 9]. Their hair-like architecture is based on dimerization of α-helical chains into parallel coiled coils, the defining and dominant structural feature of this class of proteins. Whilst sharing similar overall structure, M proteins are sequentially highly diverse, with over 220 distinct known variants of the emm gene defining GAS serotypes [10]. The C-terminal regions of M protein variants are highly conserved and predicted to form well-defined coiled coils. Sequences diverge increasingly towards the N-terminal hypervariable region (HVR) that is projected away from the bacterial surface. Experimental high-resolution structural information for M proteins is limited [11–14].

In addition to their role in adhesion and subverting the host’s immune response, M proteins have also been implicated in biofilm formation [15]. Bacterial biofilms are a major cause of difficult-to-treat infections largely contributed by their recalcitrance to antibiotics and immune responses [16]. They are classified as either surface-attached biofilms, typically associated with implants and medical device-associated infections, or non-surface-associated biofilms. The latter are commonly observed in respiratory infections of patients with impaired mucociliary function, and in persistent soft tissue infections seen in chronic wounds resulting from diabetes or impaired vascularization [17]. Biofilm is also a potentially complicating feature associated with severe invasive GAS diseases. Analyses of biopsies collected during the acute phase identified biofilm in over 30% of patients with necrotizing soft tissue infections caused by GAS [18]. Some GAS types, such as emm1 and emm3, are over-represented among isolates from severe invasive infections, i.e. necrotizing soft tissue infections and streptococcal toxic shock syndrome [1, 19, 20], but no clear association between biofilm formation and emm type is known [21]. Köller et al. highlighted that the propensity of GAS to form biofilm in vitro was directly dictated by the experimental setting where both the medium and coating with specific extracellular matrix proteins influenced the outcome [22].

M proteins have been reported to interact with a wide range of host proteins, such as fibrinogen [23], C4-binding protein [24], plasminogen [25], fibronectin [26], collagens [27] and immunoglobulins [28]. Importantly, binding activities vary between emm types. Phylogenetic clusters of M proteins broadly share binding propensities [10]. However, very few M protein interactions have been confirmed through biophysical and structural analyses. The M1 protein binds to the fibrinogen D domain in the variable “B-repeat” region [11]. Several M protein types have the ability to bind to the C4-binding protein through their HVRs [14]. While these two molecular complexes rely on the dimeric coiled-coil topology of the M proteins, the plasminogen kringle-2 domain binds to monomeric, non-coiled segments of certain M proteins [29, 30].

In a screening of GAS strains belonging to 43 emm types, only emm3 and emm18 strains were identified by Dinkla et al. to bind collagen IV to their surface [27]. For emm3 isolates, this interaction was demonstrated to depend on direct binding of collagen to the M3 protein. For emm18 GAS strains, binding was mediated by the hyaluronic acid capsule rather than the M18 protein. Direct collagen binding has also been reported for M1 protein [31]. Based on available evidence collagen binding is not a universal property of GAS M proteins, but appears to be common among M proteins of group C and G streptococci (Streptococcus dysgalactiae subsp. equisimilis, SDSE) [32, 33], which are emerging as important human pathogens that share disease manifestations and virulence traits with GAS [34, 35].

Collagens are commonly targeted by a wide range of pathogenic bacteria for the purposes of tissue colonization and dissemination [36, 37]. They have been shown to interact directly with several bacterial adhesins, such as CNA from Staphylococcus aureus [38], YadA from Yersinia enterocolitica [39], and CNE from Streptococcus equi subsp. equi [40]. In the latter two cases, adhesin binding has been mapped using the peptide libraries applied in the present study. While the biological role of collagen binding by GAS is unknown, it has been linked to the induction of an autoimmune response associated with post-streptococcal sequelae [27, 41]. The collagen binding site in M3 was mapped to the HVR. A sequence motif that is conserved in related M proteins of group A, C and G streptococci has been identified – the “peptide associated with rheumatic fever” (PARF) [33]. This motif is found in a region predicted to deviate from the canonical coiled coil structure in M3, several other M proteins and their homologs in non-group A streptococci (Figure 1). Binding of M3 protein to other collagen types has not been reported, but the M protein FOG (aka Stg11) of SDSE was found to bind to collagens I and IV [33, 42].

Figure 1.

To gain insights into mechanism and biological role of streptococcal collagen binding, we characterised the interaction of M3 with the Collagen Ligand Collections (CLCs, triple-helical collagen peptide libraries formerly known as Toolkits), and determined crystal structures of the M3 N-terminal domain (M3-NTD), encompassing the HVR, alone and in complex with a CLC-derived collagen peptide. We find that M3-NTD folds into a novel T-shaped domain that binds promiscuously to collagen triple helices. Furthermore, we demonstrate that the M3-collagen interaction underpins biofilm formation by GAS emm3 isolates from necrotizing soft tissue infections, which can be inhibited in vitro by recombinant M3 protein and collagen peptides.

Results

Recombinant M3 protein binds to the triple-helical domain of collagens II and III

The first aim of this study was to confirm the collagen binding activity of streptococcal M3 protein, and to identify any specific binding sites in collagen. To reduce the complexity of the experimental system we chose collagens II and III, as these are homotrimers, in contrast to, for instance, heterotrimeric collagens I and IV [44]. Recombinant M3 protein, lacking the N-terminal secretion signal and the C-terminal membrane-spanning region was produced in Escherichia coli as a glutathione S-transferase (GST) fusion protein. Binding to collagens II and III was investigated using CLCs, libraries of 56 and 57 overlapping triple-helical peptides, respectively, covering the entire triple-helical tropocollagen domain of these two collagen types (1014 and 1029 amino acids) [45].

Recombinant M3 was found to bind to CLC peptides to various degrees, as quantified by an ELISA-like approach using an anti-GST antibody (Figure 2). While there were clear differences in binding between the CLC peptides, it was impossible to identify sequence features that were conserved in good binders (high apparent affinity) but absent from peptides that gave rise to signals comparable to negative controls. Neither sequence motifs, nor composition of the peptides (e.g., presence of hydroxyprolines, charged, hydrophobic or aromatic residues) were distinct in good binders. To investigate further the sequence requirements for M3 binding, we ranked the CLC peptides in decreasing apparent affinity, measured as A₄₅₀ determined in the solid-phase binding assay, and then analyzed the distribution of particular amino acids in the entire set, pooling data from CLC-II and CLC-III. After background subtraction, we defined three binding groups: high affinity, having A₄₅₀ between 0.5 and 0.75; medium affinity, A₄₅₀ from 0.25 to 0.5; and low affinity, A₄₅₀ from 0 to 0.25.

Figure 2.

We counted residues of each type in these three groups and compared their occurrence in peptides of the three binding groups using non-parametric tests (Figure 2—figure supplement 1). The outcomes are summarized in Table 1. Hydrophobic amino acids and hydroxyprolines appear to be more frequent in good binders, while prolines and acidic residues are underrepresented. The negative effect of proline on binding may explain why the [GPP]₅ flanking sequences of the CLC peptides do not dominate binding to M3, allowing marked sequence selectivity to be observed in the binding assays.

Table 1:

M3-NTD harbors the collagen binding site

Full-length M proteins are not amenable to high-resolution structural characterization due to their anisotropic shape and potential conformational dynamics. We therefore designed a construct representing the M3 N-terminal domain (M3-NTD), which comprises the HVR and a short region predicted to adopt a dimeric coiled-coil structure. This fragment would not only allow structural characterization but could also be used to confirm the previously suggested localization of the collagen binding site at the HVR [33]. M3-NTD includes the 110 N-terminal residues of mature M3 (residues 42-151 of the M3 protein precursor sequence). To stabilize a dimeric conformation, Leu151 was replaced with a cysteine for disulfide bond formation at the C-terminus. Leu151 is predicted to occupy a “d” position in the canonical coiled coil heptad pattern, forming part of the hydrophobic interface between the α-helices, a position ideally suited for disulfide bond formation [46]. A ¹⁵N isotopically labelled version of M3-NTD was made for NMR spectroscopic analysis. From a comparison of ¹H, ¹⁵N heteronuclear single-quantum coherence (HSQC) spectra it is evident that a significant structural change occurred upon oxidation (resulting in disulfide bond formation) of the protein. The higher dispersion of signals, most noticeably in the ¹H dimension, indicates disulfide-linked M3-NTD adopts a folded conformation (Figure 3A). In contrast, the reduced, monomeric form gives rise to a poorly resolved spectrum with signals falling within the random-coil chemical shift range, and big differences in crosspeak intensities that suggest dynamic behavior (Figure 3B). This demonstrates the C-terminal disulfide bond is required to stabilize a dimeric, folded structure that differs from extended unfolded or linear α-helical conformations.

Figure 3.

To test if M3-NTD harbors the collagen binding site, binding to selected CLC triple-helical peptides was studied by isothermal titration calorimetry (ITC). We selected two peptides that showed medium to high apparent affinity in the CLC screening (II-27 and II-44) and a low affinity peptide (II-16). These three peptides all interacted with M3-NTD in solution (Figure 4A-C). II-27 and II-44 binding was characterized by dissociation constants (K_D) in the low micromolar range (7 and 5 µM, respectively). II-16 had an approximately ten times lower affinity (K_D = 70 µM) for M3-NTD. These K_Ds reflect the differences in binding of full-length M3 evident from the CLC solid-phase binding assay. Fitting of the sigmoidal binding curves for II-27 and II-44 suggested binding of one triple-helical peptide per M3-NTD monomer, i.e., two indistinguishable collagen binding sites per M3 dimer. To assess if the interaction was dependent on the triple-helical conformation of the CLC peptides, we titrated M3-NTD into a monomeric version of II-44 with scrambled GPP repeat sequences at the termini [47]. No binding was observed (Figure 4D).

Figure 4.

In conclusion, the collagen binding site of M3 was confirmed to reside in M3-NTD. CLC solid-phase binding results obtained for full-length M3 align with in-solution interaction analyses by ITC using M3-NTD. Binding of M3-NTD to collagen peptides depends on their triple-helical structure.

M3-NTD adopts a folded structure deviating from dimeric coiled coil

M3-NTD crystallized in several conditions with the best crystals diffracting X-rays to a resolution of 1.9 Å at the synchrotron radiation source. Phasing and structure determination was achieved using a selenium single-wavelength anomalous diffraction (Se-SAD) dataset at 2.6 Å resolution with the final model refined against the native data set at 1.9 Å resolution (Table 2).

Table 2:

M3-NTD is a symmetrical homodimer, linked at the C-terminus by a disulfide bond (Figure 5A). The first three residues of the construct (Gly-Ala-Met), a cloning artefact, are not resolved in the structure.

Figure 5.

The fold represents a novel T-shaped architecture with no significantly similar structures identifiable by the protein structure comparison server DALI [48]. Each monomer is composed of three α-helices, H1-H3. The H3 helices comprise the C-terminal 38 residues of M3-NTD (M3 residues 114-151) and form a coiled coil stem. This, in the full-length protein, would be extended into a ∼50 nm long coiled coil characteristic of M proteins (Figure 5B). Helices H1 and H2 form hairpins that pack against each other to form the slightly kinked bar of the T-shape, effectively a three-helix bundle. A key residue is Gly113, which resides at the T-junction, separating H2 and H3 (Figure 5C). It is completely conserved in M proteins identified in our sequence similarity search (Figure 1). The T-junction structure is defined by conserved leucine and isoleucine residues that form a small hydrophobic core (Figure 5C). The T-bar structure is stabilized by a network of polar interactions of several conserved residues, most notably inter-chain salt bridges of Arg52 with Asp102 and Glu105, and an inter-chain hydrogen bond between the side chains of Glu49 and Asn101 (Figure 5D). Asn101 is one of the completely conserved residues in the PARF motif, previously implicated in collagen binding. This motif is located on H2 at the bottom of the T-bar (Figure 5A). Some but not all conserved PARF residues contribute to the formation of the T-bar structure of M3-NTD. Leu97 and Asn101 are the only completely conserved residue of PARF with a structural role. Leu97 is buried at the interface between the H1-H2 hairpin of one monomer and H1 of the other monomer, packing against the Tyr81 sidechain (Figure 5D).

An N-terminal T-shaped fold is predicted to be a feature of other M proteins

AlphaFold3 [49] predictions were carried out for the proteins identified in our sequence similarity search (Figure 1) to support the structural role of the conserved residues. In validation of this approach, the AlphaFold3 predicted model of M3 is almost perfectly superimposable with the experimental M3-NTD structure (Figure 6A). All other proteins included in this study are predicted to have T-shaped N-terminal domains, with exception of proteins M133 and M228, where deletions break the topology of the T-bar fold (Figure 6B, Figure 6—figure supplement 1). Some of the predictions for GAS M protein variants, such as M12 and M55, carry low confidence and may well not adopt the T-fold. HVRs of M proteins that are phylogenetically more distant to M3, such as M1 and M28, are not predicted to fold into any distinct tertiary structure deviating from coiled coil (Figure 6C). On the other hand, M proteins of SDSE and the SzM protein of Streptococcus equi subsp. zooepidemicus (SESZ) are predicted with high confidence to adopt a structure similar to M3. This analysis structurally validates our sequence alignment of M3 homologs. It suggests the T-shaped structure of the NTD is a feature of a subclass of M proteins and is common in SDSE.

Figure 6.

M3-NTD binds promiscuously to the collagen triple helix

M3-NTD is a structurally tractable fragment that retains the collagen binding activity of the full-length protein. We chose II-27, a CLC peptide with good affinity for M3-NTD, as the basis for structural characterization of an M3-collagen complex. II-27 contains an atypical α1β1 integrin-selective motif, GVOGEA [50]. To increase chances of crystallization, a shorter 24-residue peptide, JDM238, was synthesized which contains the GVOGEA motif flanked by three glycine-proline-4-hydroxyproline

(GPO) repeats at each terminus. Crystals formed in conditions containing JDM238 and M3-NTD at a 1.5:1 ratio and diffracted synchrotron X-rays to 2.3 Å resolution. Molecular replacement with M3-NTD yielded a high-quality structural model for the complex (Table 2).

In agreement with our ITC data for CLC II-27 binding to M3-NTD, the M3-NTD dimer is bound to two copies of triple-helical JDM238. While both copies occupy equivalent binding sites on opposite faces of the M3-NTD T-bar, they contact the protein with two different regions (Figure 7). One copy contacts M3-NTD through residues 15-21 of the C-terminal GPO repeats, the other is bound at its more central GPOGVO region (residues 6-12) (Figures 7 and 8A). The presence of two different binding registers in the same crystal is likely an effect of crystal packing, but it also reflects a lack of specificity of collagen binding by M3, in line with the CLC screening data. Superposition of the two binding sites highlights how the M3-NTD T-bar complements the characteristic and uniform surface topology of triple helices in both binding registers in the same fashion (Figure 8B).

Figure 7.

Figure 8.

The M3 collagen binding site includes the PARF region on H2 (residues 94-101), extending beyond it to include the C-terminal half of H1 and most of H2 (Figure 8). This confirms previous reports on the involvement of PARF in collagen binding [32, 33]. Interfaces between the binding partners largely comprise hydrophobic interactions between highly complementary surfaces. M3 residues Tyr96 and Trp103 play prominent roles: they pack against Hyp and Pro sidechains, filling grooves on the collagen triple helices and form polar contacts with the peptide backbone. The indole amino group of Trp103 interacts with the carbonyl of Ala15 or Hyp6 (depending on the register), while Tyr96 forms a water-mediated hydrogen bond with the carbonyl of Hyp9 or Hyp18. In one binding mode, M3 Gln42 hydrogen bonds with the same water molecule as Tyr96. Direct polar contacts also include M3 residues Lys35 and Arg49 which interact with the hydroxyl group of Hyp9 and the backbone carbonyl of Hyp12, respectively. In addition, in one binding mode M3 residue Gln46 forms a hydrogen bond with Gly22 of the peptide via a water molecule.

Tyr96 is a conserved residue of the PARF motif. We tested its role in collagen binding using two site-specific variants, Tyr96Ala and Tyr96Phe. When titrating M3-NTD Tyr96Ala into CLC peptide II-44, which bound to wild type M3-NTD with low micromolar K_D, no binding was observed (Figure 4E). The more conservative substitution of Tyr96 with Phe also severely reduced the affinity for II-44 (K_D = 100…300 µM) (Figure 4F), suggesting that the water-mediated interaction between Tyr96 and the peptide backbone strongly contributes to binding. This supports the critical role of Tyr96 in collagen binding, both via forming van der Waals interactions with the collagen peptide triple helix as well as forming the water-mediated polar interaction with the peptide backbone. In conclusion, we present the first structural evidence for an M-protein collagen complex. The structure indicates a general binding mechanism that might explain the promiscuity of M3 for diverse CLC peptides and explains how M3 is able to bind non-selectively to different collagens.

emm type-dependent effect of human collagen on biofilm formation by GAS strains

GAS emm3 strains are highly prevalent in necrotizing soft tissue infections [19, 20], where they have been found to form biofilm [18]. Based on our previous data demonstrating that the tissue milieu seems to promote GAS biofilm [18], and with collagen being a ubiquitous structural protein, it was of interest to assess whether the M3-collagen interaction could affect biofilm formation. For this purpose, we used crystal-violet based assay to determine biofilm formation by GAS strains from patients with necrotizing soft tissue infections on polystyrene plates, either uncoated or coated with human type I collagen. The three isolates, emm1, emm3 and emm28 strains, all formed biofilm on uncoated plates (Figure 9A). Notably, on collagen-coated plates, biofilm of isolates 2006 (emm1) and 5004 (emm28) was significantly reduced in a dose-dependent manner, while that of isolate 2028 (emm3) was significantly enhanced (Figure 9B, C). The collagen-enhancing effect of biofilm for the emm3 strain was seen even at inocula as low as 100 CFU/well (Figure 9D), where also a dose-response to collagen was evident (Figure 9D). Confocal microscopy was used to assess bacterial attachment and biofilm formation with bacteria grown on glass slides with or without collagen coating. In line with the crystal-violet assay, bacterial attachment of the emm1 (2006) and the emm28 (5004) strain was almost completely blocked by collagen whereas the emm3 (2028) strain formed strong robust biofilm in the presence of collagen (Figure 9E).

Figure 9.

Taken together the data shows a collagen-enhanced biofilm formation for the emm3, but not the emm1 and emm28 strains. To test whether this effect was linked to the M3-protein/collagen interaction, we performed competition experiments where biofilm formation of the emm3 (2028) strain was assessed in the presence of increasing concentrations of M3-NTD in the medium. The enhancing effect of collagen was almost completely negated in presence of 20 µM M3-NTD (Figure 10).

Figure 10.

These data indicate that the interaction between M3 protein and collagen enhances biofilm formation. In further support of this finding, two additional emm3 strains, GAS 5626 and 8003 showed a similar increase in biofilm formation on collagen-coated plates (Figure 10—figure supplement 1A). The expression of M3 protein by all three emm3 strains was confirmed using an M3-specific antibody (Figure 10—figure supplement 1B).

GAS M3-collagen interaction in patient biopsies and in a 3D skin tissue model

To demonstrate a potential interaction between GAS and collagen during infection, we stained tissue biopsies from patients with necrotizing soft tissue infections using specific anti-GAS and anti-collagen IV antibodies (Figure 11). In areas with high bacterial load, collagen was observed to co-localize with the bacteria in biopsies from two patients infected with emm3 strains 2028 and 5020. In contrast, in patients infected with emm1strains (2006 and 2068), no co-localization between the bacteria and collagen was evident.

Figure 11.

To further investigate this in a more controlled infection model, we used a human 3D organotypic skin model, which is based on a collagen type IV scaffold. We have previously employed this tissue model for studies of biofilm and GAS-elicited tissue pathology [51, 52]. Infection of the skin organotypic tissue with the emm1 (2006) and emm3 (2028) strains showed that both strains efficiently infected the skin model tissue and caused disruption of the epithelial layer (Figure 12). Bacterial load and tissue pathology increased over time. Collagen began to colocalize with emm3 GAS 8 h after infection and strongly colocalized with bacteria 24 and 48 h after infection. In contrast, no colocalization of collagen and emm1 GAS was observed. These data suggest that an interaction between emm3 GAS and collagen occurs in human tissue and might be an important factor for biofilm formation.

Figure 12.

Discussion

No obvious or simple connection can be made between M protein sequences and the differential abilities of the over 200 variants to interact with host factors. At least some activities of this major streptococcal virulence factor may be encoded in hidden sequence patterns, as identified for the M4-binding protein interaction with HVRs [14]. In this study, we show that the HVR of the M3 protein deviates from the canonical coiled-coil structure generally assumed to be adopted by M proteins. The novel T-shaped fold identified here is required for the collagen binding activity of M3. While this fold may be limited to a few phylogenetically close variants of M proteins in GAS, it is more abundant in SDSE. The discovery of a folded HVR may inform vaccine design. Although immunization with short and therefore monomeric M protein HVR peptides may be sufficient for most GAS serotypes, immunogenic epitopes of M3 may only be present in the folded, dimeric form of the protein.

The structures of M3-NTD alone and in presence of a collagen peptide give a rational basis for the role of the PARF motif, which was previously implicated in collagen binding [33]. The role of conserved residues within PARF is predominantly to stabilize the T-fold of M3-NTD. The motif alone, encompassing eight residues (Ala94 to Asn101) as defined by Barroso et al. [32], does not present all structural features involved in collagen binding based on our structure. This depends on a larger region spanning residues Gln92 to Leu107 on the H2 helix, with some minor contributions of residues in H1 (Figure 8). The complex structure identifies Tyr96 and Trp103 as key interface residues. They provide shape complementarity and stack against proline and hydroxyproline rings of the collagen triple helix. Our mutational analysis confirms a role in collagen binding for Tyr96. Conclusions from the crystal structure are consistent with our analysis of the binding propensities of the different amino acids in the CLC peptides. The observed prominent roles for hydrophobic residues and for hydroxyproline in the solid-phase binding assays would be predicted from the complex structure. The negative effects on binding rank of glutamate and of proline are less easy to explain. Proline is restricted to the X position of the GXY triplet, which is also the position usually occupied by glutamate. The effect of both residues may be to exclude more productive hydrophobic amino acids from the X position. The positive role of hydroxyproline in the Y position is amply demonstrated in the crystal structure. In addition to its ability to form hydrogen bonds, this may derive from its greater extension from the helix axis than proline. Our findings are reminiscent of those obtained for YadA [39], in that we also observed promiscuous binding to the CLC peptides, which was strongly dependent on hydrophobic residues. However, in the case of YadA, the number of GPO triplets and number of both Pro and Hyp also correlated with affinity.

The molecular complex described here explains how GAS of the M3 type achieve very strong binding to collagens. While the affinity for the triple helical peptides measured here by ITC is moderate, the presence of two binding sites on the HVR that is found at the distal end of each M3 protein suggests cooperative binding in the context of larger collagen assemblies. Most likely the bacteria do not encounter isolated triple-helices, but higher-order collagen structures made of bundles of triple helices. It can be envisaged how M3 proteins probably intercalate between triple helices in such higher-order assemblies. Given the high density of M proteins on GAS surfaces, this would generate a highly polyvalent host-bacteria interface with the HVR acting as an anchor binding M3 perpendicularly to collagen fibrils/fibers. It is also at this higher-order structural level that an explanation for the ability of M3 to induce an anti-collagen autoimmune response might be found. In the rare autoimmune disease Goodpasture syndrome, an anti-collagen response is thought to be caused by collagen IV neoepitopes, with infections being suggested as one possible cause for their exposure [53]. It is conceivable that such neoepitopes could be generated as a result of collagen structural changes caused by M3 protein binding. However, such structural rearrangements could not be observed in our simplified molecular system, and our data cannot provide evidence backing the anti-collagen hypothesis of post-streptococcal rheumatic sequelae [41] beyond providing the molecular basis for collagen recognition by M3 protein.

The structural data presented here allow us to conclude that collagen binding by M proteins relies on the presence of a T-shaped helical bundle domain combined with key residues to provide shape complementarity to tropocollagen triple helices. But even if all GAS M protein variants included in our sequence analysis bound collagens in a similar manner to M3, this would still indicate that M3-like collagen binding mechanism is rare among GAS M protein variants. Based on data published for M55, which was found to bind comparably weakly to collagen IV [54], it seems likely that M3 and closely related serotypes have a unique ability to bind collagens in the way described here. AlphaFold3 predictions support this, yielding converging structural models for the M3-collagen peptide complex that closely resembled our experimental structure. This was also true for M31 and M222. However, predictions with M12 and M55 resulted in low-confidence, non-converging structural models with poorly defined binding sites, and/or disruption of the T-fold (Figure 6—figure supplement 1). On the other hand, highly confident structural predictions together with conservation of key residues suggests the M3-like collagen binding mechanism is shared by SDSE and SESZ M proteins (Figure 6—figure supplement 1). This has previously been experimentally validated for the FOG (or Stg11) protein of SDSE, which was found to bind to collagens I and IV with high affinity [33, 42]. SDSE, in the past often associated with animal infections, is increasingly recognized as a highly prevalent human pathogen closely resembling GAS in terms of virulence traits and pathologies, including invasive infections, post-infection autoimmune sequelae [35] and biofilm formation [55]. Collagen binding by M proteins may be therefore a far more common virulence mechanism in human streptococcal infections than the prevalence of emm3 GAS alone would suggest. It should be noted, too, that a new emm3 variant, emm3.93, is currently emerging in the UK and the Netherlands, with a high prevalence in invasive infections [56].

The interaction between M protein and collagen has implications for several biological functions, including bacterial attachment and biofilm formation. In this study, we investigated biofilm formation due to the high prevalence of emm3 strains in necrotizing soft tissue infections [20], in which biofilm is complicating feature [18, 57]. Our findings reveal that the M3 protein-collagen interaction promotes biofilm in emm3 strains. Notably, for other emm types the presence of collagen appears to have an opposing effect, reducing bacterial attachment and biofilm in vitro for other emm types. This type-specific difference is further supported by observations within infected tissue where emm3 but not emm1 strains co-localize with collagen fibers. These findings highlight the importance of collagen for emm3 biofilm development in the tissue setting, while other GAS emm types likely utilize other mechanisms. Further studies are warranted to explore the underlying mechanisms of biofilm formation in streptococcal tissue infections, including not only GAS but also SDSE strains.

Materials and Methods

Cloning and site-directed mutagenesis

M3-NTD DNA insert was amplified from a plasmid containing full M3 DNA sequence using primers 5’-GCTAGCCATGGATGCTAGGAGTGTTAATGG-3’ and 5’-CTAGGGATCCCTAGCAGTCCTGATATTCCTTTTC-3’, digested with NcoI and BamHI and ligated into the pEHISTEV vector [58], pre-digested with the same enzymes. The final M3-NTD construct, after digestion with TEV protease, comprised 113 amino acids starting with Gly-Ala-Met, an artefact of the purification tag. Site-directed mutagenesis was performed by PCR amplification of pEHISTEV-M3NTD plasmid using mutagenic primers listed in Table 3. For selenomethionine labelling, two methionine residues were introduced by substituting Ile60 and Ile141 using two sequential rounds of site-directed mutagenesis.

Table 3.

Recombinant protein production

Recombinant M3 protein was produced as an GST fusion construct as previously described [59] from a pGEX6P-1 vector, a kind gift from Dr Susanne Talay, in E. coli BL21 (DE3) cells. The cells were grown in LB media containing 100 mg mL^-1 ampicillin at 37 °C for 3 h after growth to an optical density at 600 nm of 0.6, and induction with 1 mM isopropyl β-D-1-thiogalactopyranoside. The fusion protein was purified from bacterial cell lysate using a GSTrap 4B column according to the protocol provided by the manufacturer (Cytiva). The recombinant protein comprised GST fused to a proteolytic cleavage site (not used in this study) and the N-terminus of M3 (UniProt entry A0A0H2UWN1), lacking N-terminal secretion signal, cell wall anchor and membrane spanning region (residues 42-546).

M3-NTD constructs were overexpressed from pEHISTEV plasmid in E. coli soluBL21 cells (Genlantis), in a standard LB medium supplemented with 50 mg mL^-1 kanamycin. An overnight culture was used to inoculate flasks containing 1 L of the growth medium which were incubated at 37 °C with shaking. When the optical density at 600nm was 0.5-0.8 expression was induced by adding IPTG to a final concentration of 1mM. After 3-4 h of further incubation with shaking the cells were harvested and frozen. For selenomethionine labelling the pHT-M3-NTD [Ile60Met, Ile141Met] plasmid was introduced into the methionine auxotroph strain of E. coli, B834, which was obtained from Dr Clarissa Melo Czekster. The cells from the overnight culture were washed twice with M9 minimal media, resuspended in 0.2x the initial volume and used to inoculate (10 mL per L) of SelenoMet medium (Molecular Dimensions) supplemented with 0.04 mg L^-1 L-selenomethionine (ThermoFisher). Following induction of expression, the cells were incubated with shaking for 5 h.

Cell pellets were resuspended in 50-100 mL 50 mM Tris-HCl pH 8, 500 mM NaCl lysis buffer supplemented with one cOmplete protease inhibitor tablet (Roche) and 1 mg DNAse I (Merck). The suspension was passed twice through a Cell Disrupter (Constant Systems) at 30 kpsi and the lysate was clarified by centrifugation at 20,000 x g for 25 min at 4 °C. The supernatant was applied to a HisTrap column (Cytiva) pre-equilibrated in the lysis buffer and the bound protein was washed with 10 column volumes of the wash buffer (50 mM Tris-HCl pH 8, 500 mM NaCl, 20 mM imidazole) and eluted in a linear gradient (20-250 mM imidazole). To remove imidazole and the N-terminal purification tag, the eluate was mixed with TEV protease (produced in-house) at approximately 30:1 stoichiometric ratio and dialyzed overnight at room temperature against PBS supplemented with 1 mM DTT. The digested protein solution was passed through the HisTrap column again, with PBS as the mobile phase, and M3-NTD collected in the flow-through. Following concentration in 10 kDa Amicon Ultra centrifugal filters (Millipore) the protein was further purified by size exclusion chromatography using Superdex75 10/300 column (Cytiva) pre-equilibrated in PBS. Purity and oxidation state were verified by SDS-PAGE. If some monomeric protein was still evident on the gel the fractions from size exclusion chromatography were left overnight at 4 °C. Fully oxidised protein was concentrated, aliquoted and flash frozen.

Collagen peptides

The Collagen Ligand Collection (CLC) peptides (formerly known as Collagen Toolkits) were obtained as C-terminal amides from Triple Helical Peptides Ltd, Cambridge, UK. They were synthesised on TentaGel R-Ram resin using Fmoc/tBu chemistry, either on an Applied Biosystems Pioneer peptide synthesiser as described previously [60], or a CEM Liberty or Liberty Blue microwave-assisted peptide synthesiser. Fractions containing homogeneous product were identified by analytical HPLC on an ACEphenyl300 (5 mm) column, characterised by MALDI-TOF mass spectrometry, pooled and freeze-dried. In the CLCs, the variable 27-residue primary collagen structure (guest sequence) was flanked by GPC[GPP]₅- and -[GPP]₅GPC (host) peptides, to ensure stable triple-helical form. In JDM238, the guest sequence is GVOGEA and the host sequences [GPO]₃.

Solid phase CLC binding assay

CLC peptide solid-phase binding assays were performed following a previously published protocol [45, 47]. Briefly, CLC peptides, collagen II (positive control), GPP10 and BSA (negative controls) (10 μg/ml in 0.01 M acetic acid) were immobilized on Immulon2 HB 96-well plates (Nunc, Langenselbold, Germany) overnight at 4 °C. All subsequent incubation steps were for 1 h at 25 °C. The assay volume was 100 µl per well. Wells were washed three times with adhesion buffer (0.1% (v/v) Tween-20 and 1 mg/ml BSA in PBS) between incubation steps. The wells were blocked with 50 mg/ml BSA in PBS prior to the addition of recombinant GST-M3 at a concentration of 10 µg/ml in adhesion buffer. Bound protein was detected with biotin goat anti-GST antibody (Abcam) at a 1:500 dilution in adhesion buffer, followed by streptavidin-fused HRP and 3,3’,5,5’-tetramethylbenzidine liquid substrate system (Sigma), and plates read at 450 nm. Including Tween in the washing steps reduced background signal but did not change the overall outcome for CLC-II. It was omitted for the CLC-III assay.

NMR

M3-NTD was isotopically labeled by expression in M9 minimal media (6.8 g/L Na₂HPO₄, 3.0 g/L KH₂PO₄, 0.5 g/L NaCl, 1.0 g/L ¹⁵NH₄Cl, 2 mM MgSO₄, 0.1 mM CaCl₂, 1% (w/v) glucose, 1 mM MgSO₄, 2.125 g/L BDTM Difco^TM Yeast Nitrogen Base without Amino Acids and Ammonium Sulfate (Thermo Fisher) and 50μg/ml kanamycin) and purified as described above for unlabeled protein. The NMR sample contained 0.4 mM protein in 10 mM phosphate, 50 mM NaCl, pH 6.5, 1.5% (v/v) D₂O, without or with addition of 10 mM DTT to generate monomeric M3-NTD. ¹H,¹⁵N HSQC spectra were recorded on a Bruker Ascend 700 MHz spectrometer equipped with a Prodigy TCI probe and controlled by Bruker Topspin 3 software at 30 °C. A standard Bruker pulse sequence with gradients and water flip back pulse (hsqcetfpf3gpsi) was used with 20 transients and spectral resolutions or 14.5 Hz and 41.2 Hz in the direct (¹H) and indirect (¹⁵N) dimension, respectively. Spectra were processed with NMRPipe [61] and visualized with CCPN Analysis 2 [62].

Isothermal titration calorimetry

The experiments were performed using MicroCal PEAQ-ITC instrument (Malvern Panalytical) at 25 °C in PBS, with the reference power set to 3 µcal/s and stirring speed to 750 rpm. The cell contained the CLC peptide at 40-90 µM (trimer), reconstituted in PBS, and the injector syringe contained M3-NTD at 0.9-1.1 mM, dialysed in the same buffer. For control experiments determining the heats of dilution, the cell contained PBS buffer only. Titration involved injection volumes of 2 µl except for the first injection of 0.4 µl.

Protein crystallisation and structure determination

Crystallization trials were conducted using the vapor diffusion sitting drop method and several sparse matrix screens. M3-NTD protein construct was freshly purified by size-exclusion chromatography and concentrated to ∼700 µM. For co-crystallization with JDM238 peptide, the protein was concentrated to 1 mM and then diluted by adding reconstituted peptide so that the final concentration of M3-NTD was ∼700 µM with triple-helical peptide at ∼1.1 mM (1:1.5 ratio). The crystals of M3-NTD grew at 20 °C within 1-2 weeks in 40-50% MPD, and the best one diffracted to 1.92 Å resolution. They were used for streak-seeding the selenomethionine-labelled protein, which resulted in shard-like crystals diffracting to 2.67 Å with the data collected at the selenium K absorption edge. Phasing of the SAD dataset was conducted automatically by the Diamond Light Source (DLS) pipeline FastEP and initial model building was performed using the ARP/wARP software package. The crystals of M3-NTD in complex with the collagen peptide were obtained in 15% PEG 10K, 0.1 M Tris-HCl pH 8.5, 0.29 M MgSO₄ and diffracted to 2.32 Å resolution. Indexing, scaling and merging of data was performed using the autoPROC pipeline at the DLS. The complex structure was solved by molecular replacement with PHASER [63] using M3-NTD and collagen peptide (PDB 3P46) structures and as search models, with all non-proline residues in the collagen peptide model substituted with alanine. For both structures, iterative model building and refinement was performed using Coot [64] and Refmac5 [65]. MolProbity [66] was used for model quality assessment. Figures of protein structure models were produced using PyMOL (Schrödinger, LLC).

Accession codes

The M3-NTD and M3-NTD/JDM238 protein structures and the data used to derive these, have been deposited at the PDBe with accession codes 8p6k and 8p6j, respectively.

Bacterial strains

GAS strains 2006 (emm1), 2028 (emm3), and 5004 (emm28) are isolates from patients with necrotizing soft tissue infections from the INFECT project [18]. GAS 5262 and 8003 were provided by Donald E. Low (Mount Sinai Hospital, Toronto, Canada) and used as further emm3 isolates from patients with necrotizing soft tissue infections [67, 68]. All isolates were cultured in either Todd-Hewitt broth supplemented with 1.5% yeast extract or brain heart infusion broth at 37 °C under a 5% CO₂ atmosphere.

Biofilm formation assay on a polystyrene surface

Biofilm formation was evaluated by crystal violet staining as described previously, with minor modifications [69]. Overnight cultures of the bacteria being tested were washed with PBS and diluted to approximately 10⁶ CFU/mL. Bacterial suspensions (100 µL) were seeded into 96-well polystyrene plates (Thermo Fisher Scientific, Waltham, MA, USA). The plates were then incubated for 24 h at 37 °C. After incubation, planktonic bacteria were removed by washing with PBS. Plates were then stained with a 0.1% crystal violet solution (Invitrogen, Waltham, MA, USA) for 30 min. Excess crystal violet was removed by washing with PBS, and then the crystal violet that was associated with bacterial cells was eluted with absolute ethanol. The amount of crystal violet (and by association, biofilm) was evaluated by measuring the absorbance at 590 nm using a spectrophotometer. Collagen-coated plates were prepared by incubating plates with 1-10 µg/mL of collagen I overnight at 4 °C.

Confocal microscopic analysis of biofilm formation

The biofilms were formed on an 8-well chamber slide (Lab-Tek, Thermo Fisher Scientific) in the same manner as for the biofilm assay. After removing planktonic bacteria by washing with PBS, the biofilms were fixed with 10% formalin and stained with wheat germ agglutinin (WGA)-Alexa Fluor 488 conjugate (Invitrogen) and Nile red (Invitrogen). The slides were mounted using ProLong™ Gold Antifade Mountant with DAPI (Invitrogen).

Biofilm competition assay

Recombinant M3-NTD was added at 10 and 20 µM concentrations in PBS to a collagen-coated 96-well plate (0.5 µg/well) and the plate was incubated for 1 h at room temperature. After removal of the protein solution, 100 μL of strain 2028 culture (ca. 10³ CFU/mL) was seeded and incubated for 24 h. Biofilms were stained with crystal violet, as described above.

Immunofluorescent staining of GAS

A few colonies were suspended in PBS on a glass slide and then fixed in 3.7% formaldehyde in PBS for 15 min. The bacteria were stained with anti-M3 specific antibodies (kindly provided by Prof. Gunnar Lindahl, Lund University), followed by WGA-Alexa Fluor 488 conjugate antibodies (Invitrogen). The slides were mounted using ProLong™ Gold Antifade Mountant with DAPI (Invitrogen).

Three-dimensional organotypic skin model

The 3D skin models were constructed as described previously [18, 51, 52]. Briefly, 4.0 ξ 10⁴ normal human dermal fibroblasts (NHDF) in a collagen matrix (Pure Col, Advanced Biomatrix, Carlsbad, CA, USA) were seeded onto a polymerized cell-free collagen layer in a 6-well filter insert (Corning, Corning, NY, USA). After culturing in DMEM for one week, 1.0 ξ 10⁶ human keratinocyte cells N/TERT-1 were seeded onto the NHDF layer. The models were cultured in EpiLife medium (Invitrogen) for three days and exposed to air for one week. For the infection assay, the models were infected with 1 × 10⁶ CFU of bacteria for 8, 24, or 48 h.

Immunofluorescent staining of patient tissue biopsies and tissue models

Snapfrozen tissue biopsies from patients with necrotizing soft tissue infections caused by either emm1 GAS strains (patients 2006 and 2068) or emm3 GAS (patients 2028 and 5020) were available from the INFECT patient biobank [18]. Cryosectioning and staining were performed as previously described [52]. Briefly, the biopsies and models were embedded in an optimum cutting temperature compound (Sakura, Torrance, CA, USA) and frozen in liquid nitrogen. Cryosectioning (8 µm) was performed using a Leica CM3050 cryostat (Leica, Nußloch, Germany). Sections were then fixed in 3.7% formaldehyde in PBS for 15 min and stained with both anti-GAS (goat polyclonal antibody, Abcam, Cambridge, UK) and anti-collagen IV antibodies (Abcam, Cambridge, UK), followed by WGA-Alexa Fluor 488 and 546 conjugate antibodies (Invitrogen).

Statistical Analysis

To compare the propensity of specific CLC amino acids to contribute to M3 binding, the A₄₅₀ value for M3 binding to BSA was subtracted from all values obtained from the solid-phase binding assays. This allowed the CLC peptides to be ranked in descending order of A₄₅₀ and divided into three absorbance groups: high, from 0.75 to 0.5; medium, from 0.5 to 0.25; and low, from 0.25 to zero. Data from both CLCs were pooled, and the amino acid abundance in the three groups was compared. For each peptide from CLC-II and CLC-III, the number of occurrences of each amino acid of interest was noted. The non-parametric Kruskal-Wallis test was used to determine whether the amino acid abundance differed between the three groups. For the biofilm experiments, One-way ANOVA with the Tukey’s post-hoc test was used to determine statistically significant differences.

Acknowledgements

M3-specific antibodies were kindly provided by Professor Gunnar Lindahl, Lund University. We are grateful to Dr Susanne Talay for the M3-encoding pGEX6P-1 vector. The authors thank Dr Conny Yu for preparing a batch of M3-NTD for biofilm experiments. We thank the i04 and i24 beamline staff at the Diamond Light Source synchrotron for their help with data collection.

Additional information

Funding

This research was funded by the Medical Research Council (MR/N009681/1), grants from European Union FP6 ASSIST (032390) and FP7 INFECT (305340), the Swedish Research Council (2022-01-202), and Region Stockholm, Center for Innovative Medicine (FoUI-975603).