Structure of collagenase G reveals a chew and digest mechanism of bacterial collagenolysis

Abstract

Collagen constitutes one third of the body protein in humans, reflecting its extraordinary role in health and disease. Of similar importance, therefore, are the idiosyncratic proteases that nature evolved for collagen remodeling. Intriguingly, the most efficient collagenases are those that enable clostridial bacteria to colonize their host tissues, but despite intense studies, the structural and mechanistic basis of these enzymes has remained elusive. Here we present the crystal structure of collagenase G from Clostridium histolyticum at 2.55 Å resolution. By combining the structural data with enzymatic and mutagenesis studies, we derive a conformational two-state model of bacterial collagenolysis, in which the recognition and unraveling of collagen microfibrils into triple helices as well as the unwinding of the latter go hand in hand with collagenase opening and closing.

By its mere quantity, collagen remodeling plays an outstanding physiological and pathological role in all vertebrates ^1-3. Several independent lines of collagenases have evolved in nature to tackle the highly complex task of collagen hydrolysis, among them members of the mammalian matrix metalloproteases (MMPs) and bacterial collagenases. Interestingly, the most efficient collagenases are those found in clostridial bacteria which enable them to colonize and infiltrate the host tissue ^4,5.

While considerable insight into the structure and mechanism of endogenous collagenases, belonging to the MMP family, has been obtained ^6-14, exogenous bacterial collagenases so far have resisted a crystallographic analysis, partly due to the complexity of these multi-domain proteins. Therefore, much of the mechanistic basis for their enzymatic properties remained elusive despite their manifold medical and biotechnological use ^15-17. Here we present the crystal structure and enzymatic analysis of a clostridial collagenase, resulting in a unified two-step model whereby recognized collagen microfibrils are mechanically unraveled and subsequently hydrolyzed.

RESULTS

Unexpected domain architecture of ColG

Clostridial, and other bacterial, collagenases have an approximate size of 120 kDa. Based on naturally occurring isoforms and bioinformatic analysis, their domain organization was expected to be composed of a pre-domain of variable length containing the export signal which is clipped in the mature protein; an N-terminal domain harboring the catalytic zinc (Tyr119 to Gly790; numbering corresponding to ColG from C. histolyticum); a variable number of polycystic kidney disease-like (PKD-like) domains (Ala791–Asn880); and one or more collagen binding domains (Glu881–Gly1001, Leu1002–Lys1118) ¹⁸, with structural information being available only on the latter domain ¹⁹ (Fig. 1a).

Figure 1. Domain organization and architecture of ColG.

(a) Schematic of the domain organization of ColG together with a functional annotation. The catalytic Zn²⁺ ion (yellow dot) and the catalytic residues (red stars) within the peptidase domain are indicated. The hatched domains (prepro-peptide and collagen binding domains) are not present in the crystal.

(b) Ribbon representation of the collagenase module, with identical color code as in a. The position of the PKD-like domain (yellow ribbon) at the rear of the peptidase domain is indicated in surface representation, reflecting a positional variance of up to 10 Å. The saddle-shaped collagenase is composed of an activator and a peptidase domain. The catalytic Zn²⁺ and the catalytic residues are highlighted by ball and stick representation. The seat of the saddle is formed by the distorted four-helix bundle, represented by four cylinders, and completed by the glycine-rich hinge, shown in green.

(c) Full-length model of ColG in complex with a collagen microfibril. The collagenase module (colored as in a) bound to a modeled collagen microfibril (in surface representation; grey) is shown in ribbon representation. The accessory domains are shown as surface representation. The two collagen binding domains (orange) were not present in the crystal, but oriented and positioned in order to satisfy the biochemically derived binding data, including the inferred binding epitopes. Direction of collagenase processivity is indicated at the right top.

Unexpectedly, the crystal structure discloses a distinct segmentation of the N-terminal collagenase module (Tyr119–Gly790), featuring a saddle-shaped two-domain architecture (Fig. 1b; Table 1). The smaller N-terminal saddle flap serves as an activator domain (Tyr119–Asp388), as to be discussed below, and comprises an array of 12 parallel α-helices. Starting with a distorted helix pair (Tyr119–Lys161), it continues with ten tandemly repeated HEAT motifs, ideally suited to generate a protein recognition interface ²⁰. A solvent-exposed glycine-rich linker is positioned at the twist of the saddle seat (Gly389–Tyr397; Fig. 1b).

Table 1.

X-ray data collection and model refinement statistics.

	Zn-depleted (Selenomethionine MAD)				Zn2+ bound	Inhibitor complexed	PKD-like domain
Data collection
Space group	P212121	P212121	P212121	P212121	P212121	P212121	C121
Cell dimensions
a, b, c (Å)	57.8, 108.8, 182.0				57.3, 108.7, 181.5	58.1, 108.8, 181.0	68.2, 59.3, 55.3
α, β, γ (°)	90, 90, 90				90, 90, 90	90, 90, 90	90, 125.6, 90
	Peak	Inflection	High-Rem	Low-Rem	Peak	Remote	Remote
Wavelength	0.97958	0.97969	0.97190	0.98783	1.28188	0.98793	0.98793
Resolution (Å)	50.3–2.75	50.3–2.75	36.6–2.75	48.3–2.55	35.6–2.80	35.7–3.25	40.5–1.18
R merge	0.116 (0.571)	0.076 (0.639)	0.084 (0.866)	0.099 (0.695)	0.094 (0.588)	0.102 (0.300)	0.066 (0.299)
I/σI	10.1 (2.5)	13.7 (2.3)	12.5 (1.9)	9.3 (1.8)	12.4 (2.7)	11.7 (4.7)	13.9 (3.2)
Completeness (%)	93.7 (70.5)	93.7 (68.9)	92.2 (67.9)	88.1 (84.4)	98.9 (98.1)	99.5 (99.4)	96.4 (78.4)
Redundancy	6.4 (5.2)	6.4 (5.1)	6.4 (5.3)	5.4 (5.0)	6.9 (6.8)	6.5 (6.2)	6.0 (2.9)
Wilson B-factor	76.7	79.9	81.3	62.2	72.6	69.0	9.7
Refinement
Resolution (Å)				38.9–2.55	34.4–2.80	34.8–3.25	39.61–1.18
No. reflections				33068	28436	18619	57689
Rwork/ Rfree				21.0 / 25.6	23.5 / 28.2	21.9 / 26.6	14.7 / 18.0
No. atoms
Protein				5434	5300	5341	2498
Solvent				181	158	137	364
B-factors
Protein				58.74	72.85	76.16	7.38
Solvent				73.44	67.75	75.20	38.77
R.m.s deviations
Bond lengths (Å)				0.011	0.0064	0.0068	0.0181
Bond angles (°)				1.152	0.908	0.915	1.805
Ramachandran plot (%)
Most favorable				93.1	91.2	91.9	97.2
Allowed				6.7	8.3	7.6	2.8
Generously allowed				0.0	0.2	0.3	0.0
Disallowed				0.2	0.3	0.2	0.0

^*Highest resolution shell is shown in parenthesis.

The peptidase domain resembles a thermolysin-like fold

The subsequent catalytic subdomain (Asp398–Gln669) adopts a thermolysin-like peptidase (TLP) fold of mixed α and β type, most closely resembling the structures of Leukotriene A4 hydrolase and F3 protease ^21,22, and is accompanied by a catalytic helper subdomain. The flanking α-helix pairs of the activator domain and the catalytic subdomain combine to form a distorted four-helix bundle and thus comprise the seat of the saddle topology, which is completed by the glycine-rich linker. These structural elements latch the relative spatial arrangement of the peptidase domain and the N-terminal activator, which will turn out to be critical for collagen binding and triple-helix unraveling.

Topologically equivalent to thermolysin, the five-stranded β-pleated sheet of the catalytic subdomain presents the edge strand mediating the canonical substrate-recognition motif, as described below in more detail. Leu517–Leu534 form the “central helix” of the TLP fold ²³ and contain the Zn²⁺-binding H⁵²³EXXH⁵²⁷ motif. The third Zn²⁺-ligand Glu555 resides on the neighboring helix, defining ColG as a gluzincin ²⁴.

Another linker segment of eight amino acids (Gly670 to Asp678) connects the catalytic subdomain with a catalytic helper subdomain (Asp679–Asp788). The four-stranded antiparallel β-sheet partly connects to and extends the β-sheet of the catalytic subdomain, whereas the two antiparallel α-helices of the helper subdomain cover the β-sheet to the solvent.

Given the compact quaternary arrangement of the helper and catalytic subdomains, we refer to the combination of both subdomains as the peptidase domain. This interpretation is supported by the fact that the catalytic helper is indispensable for proper folding and full peptidase activity.

Position and hybrid structure of the PKD-like domain

Crystal packing as well as low-resolution contrast electron density analysis enabled us to position the PKD-like domain at the rear of the peptidase domain in Fig. 1b, with its exact orientation remaining flexible. We could further derive high resolution (1.18 Å) structural information from crystals of the isolated PKD-like domain. The PKD-like domain adopts a two-tiered β-barrel structure. Intriguingly, we found the conserved Trp833 buried inside the β-barrel (Supplementary Fig. 1). At first sight this orientation appears conflicting with literature data, reporting the critical role of Trp833 in collagen binding and swelling ²⁵. However, our structure can reconcile this apparent conflict, as the elimination of the indole ring of Trp833 would destabilize and collapse the barrel structure and thus destroy the structural framework necessary for specific collagen binding (Supplementary Fig. 1).

The remaining C-terminal collagen binding domains (CBD) were not present in our crystal form; fortunately, crystal structure information on the isolated C-terminal CBD has been obtained previously ¹⁹, enabling us to construct a full-length structural model of ColG (Fig. 1c).

Enzymatic assessment of the collagenase domains

The unexpected domain architecture prompted us to re-investigate the functional assignment of the individual subdomains. Firstly, the structure analysis suggested that the enzymatic activity should be contained within the peptidase domain (Asp398 to Gly790). Indeed, an activator deletion construct, Lys396–Lys1118, showed 100% activity against small peptidic substrates as compared to the full-length ColG protease, thus confirming this domain to represent the peptidase domain of clostridial collagenases (Fig. 2a). Furthermore and remarkably, the peptidase domain was completely inactive towards collagen substrates. Full collagenolytic activity is, however, contained in the segment Tyr119–Gly790, comprising the activator and peptidase domains. This segment thus represents the collagenase module of the protein (Fig. 1a and Fig. 2b).

Fig. 2. The activator domain is necessary for the degradation of collagen.

(a) Histogram showing the relative peptidolytic activities of ColG variants, Tyr119–Gly790, Tyr119–Asn880, Tyr119–Gly1001, Lys395–Lys1118, and Δ(Gly389–Val397), normalized to the wild type (WT) protein, against FALGPA. Lys395–Lys1118 represents a variant lacking the N-terminal activator domain, whereas the glycine-rich hinge is short cut in Δ(Gly389–Val397). Data are graphed as mean ± s.d. (n=6).

(b) Histogram showing the collagenolytic activities given as relative initial velocities of Tyr119–Gly790, Tyr119–Asn880, Tyr119–Gly1001, Lys395–Lys1118, and Δ(Gly389–Val397), normalized to the WT, against fluorescein-labeled collagen type I. Chymotrypsin serves as a negative control, indicating that the collagen substrate was not denatured. Data are graphed as mean ± s.d. (n=36).

While these biochemical findings are consistent with, and were predicted by, our crystal structures, they are in seeming contrast with both earlier reports on clostridial collagenases ⁵ and the situation in mammalian collagenases, see also our discussion below. In the latter case, the analogous collagen binding domain, the hemopexin-like domain, is critical for collagen binding, unwinding of the triple helix, and catalytic turnover ²⁶. The C-terminal accessory domains serve to recruit and partly to swell insoluble collagen without unwinding its triple-helical structure ^19,25. These Ca²⁺-dependent recruitment domains tend to accelerate the hydrolysis of insoluble collagen substrate at low component concentrations.

Catalytic Zn²⁺ site

We found the catalytic Zn²⁺ tetrahedrally coordinated by the side chains of His523, His527 and Glu555 as well as a water molecule which was further hydrogen-bonded to the general base Glu524 (Fig. 3a). Surprisingly, however, we also crystallized ColG in a zinc-depleted form, suggesting the zinc binding to be atypically weak ²⁷. The ligating bond distances for the zinc (in brackets: the inhibitor complex) are 2.16 (2.08) Å, 2.17 (2.09) Å and 2.03 (2.09) Å to the three proteinacious ligands His523, His527 and E555, respectively. These distances tend to be somewhat larger than the analogous distances in thermolysin and related peptidases where the average ligation lengths are found to be at a distance of 2.00 Å ± 0.03 Å ^21-23,27.

Figure 3. Mapping the peptidase active site.

(a) Ribbon representation of the active site. Three protein residues (His523, His527, Glu555; shown in red) and a water molecule (blue sphere) tetrahedrally coordinate the catalytic Zn²⁺ (yellow sphere). Glu524 hydrogen-bonds to the water molecule. The non-primed substrate recognition sites are formed by the antiparallel edge strand extending from Gly494 to Glu498 (orange). The S1′ oxyanion site is formed by the amides of Gly493–Gly494 (bright orange).

(b) Accessible surface representation of the ColG active site containing the inhibitor isoamyl-phosphonyl-Gly-Pro-Ala in a stick representation, with identical orientation and color code as in a. The inhibitor is overlaid with the experimental 2F_o-F_c omit electron density, contoured at the 1 σ over the mean (light grey). Substrate recognition is governed at the non-primed sites by the antiparallel edge strand; by the catalytic Zn²⁺ (dark red with yellow sphere); and by the secondary S1′ oxyanion pocket. A prominent wall delimits the S3′ site and explains the tripeptidyl-carboxypeptidase activity of ColG.

The partial zinc occupancy opens an unexpected route to tune enzyme activity; this regulation mechanism appears particularly sensible for a protease class that lacks a prodomain which would confer zymogen-like latency. In line with these observations, we found that His₆-tagged protein variants show reduced activity, which can be explained by partial Zn²⁺ extraction ²⁸. Importantly, the Zn²⁺ site was fully occupied in the presence of a substrate analogue inhibitor, consistent with the additional zinc coordination by the inhibitor. This finding suggests that substrates contribute to the stabilization, and correct positioning of the catalytic Zn²⁺, an effect which is expected to be k_cat-dependent, consistent with our observations ²⁸. Such substrate-assisted activation is indeed observed with substrates extending from P3 to P3′ residues, but not for the synthetic furylacryloyl-Leu+Gly-Pro-Ala (FALGPA) peptide lacking the non-primed edge strand stabilization ²⁹.

Belonging to the metzincin family, MMPs contain a strictly conserved methionine residue that serves as a “hydrophobic basement” to the catalytic zinc. Positioned one helical turn downstream to the catalytic Glu555, Ala558 of ColG is located at a 3D position corresponding to the conserved metzincin methionine, with its Cβ side chain pointing towards the zinc at ~4.25 Å distance. Ala558 is conserved within clostridial collagenases. From the structure it seems unlikely that larger hydrophobic residues can be tolerated at this position.

Peptidic substrate recognition and processing

The comparison of apo, unliganded, and complexed active sites show a more than 3 Å contraction movement of the two TLP-half domains upon Zn²⁺ loading and inhibitor binding. The peptidomimetic inhibitor isoamyl-phosphonyl-Gly-Pro-Ala binds to the “edge strand” in an antiparallel orientation, consistent with TLP-like peptidase-inhibitor recognition. Preceding the edge strand (Leu495-Glu498) of the TLP β-sheet, two glycine residues form the entrance to the canonical antiparallel substrate recognition strand motif (Fig. 3a). This Gly493–Gly494 motif forms the S1′ recognition sites and thus serves as the shaping die for the conformation of the P1′ residue which can only be adopted by a glycine at P1′ position ^29,30.

The complex structure of ColG with the peptidomimetic inhibitor revealed two additional noteworthy consequences resulting from the unique double glycine S1′ site in clostridial collagenases: First, the amides of both Gly493 and Gly494 point towards the P1′ carbonyl oxygen and thereby form a secondary S1′ oxyanion pocket (Fig. 3a). Second, the unusual topology of the strong S1′-P1′ oxyanion interaction dominates, yet misaligns the substrate binding geometry at the Zn²⁺ and the non-primed recognition sites. This will result in an unproductive binding of short substrates lacking P2 and P3 residues and explains why ColG starkly prefers substrates containing P2 and P3 residues ^29,30. By virtue of the multi-valent interactions with the edge strand Gly493 to Glu498, N-terminally extended substrates can be bent and thus assist their hydrolysis.

The importance of the secondary oxyanion pocket (Gly493–Gly494) is witnessed by a serendipitous G494V mutation in the closely related collagenase H from C. histolyticum which completely inactivated the protease ²⁸.

While the non-primed substrate recognition sites are defined by the backbone interaction with the edge strand (Gly494–Glu498), the wide cleft at the S3 to S1 sites allow for slightly alternative binding geometries, depending on the nature of the P3 to P1 substrate residues. In particular, prolines and hydroxyprolines, often found in collagen, cannot adopt an ideal antiparallel β-strand with the edge strand, but can still be accommodated, as indicated in figure 3b.

A loop segment Gln511–Phe515 preceding the central helix forms a prominent wall-like feature at the primed substrate recognition sites (Fig. 3b). By its geometric distance to the catalytic zinc, the wall enforces a major kink at the P4′ position of longer substrates, making its binding energetically less favorable. The wall thus serves as molecular ruler that explains the observed tripeptidyl-carboxypeptidase activity of clostridial collagenases ²⁹.

Collagen recognition, unwinding, and processing

The integration of the crystal structure with the observed enzymatic properties of the different ColG variants (Fig. 2) suggests a two-step mechanism, whereby the N-terminal activator domain cooperates with the peptidase domain in both collagen triple helix and microfibril recognition and processing. The activator and peptidase domains, forming the two “saddle flaps”, have a distance of ~40 Å, whereas the diameter of the collagen triple helix is only approximately 15 Å ¹ (Fig. 4a). Naively, one should therefore expect that the activator domain is neither involved nor required in triple helix recognition and processing – which is not the case (Fig. 2b). Consequently, one must conclude that the collagenase can adopt two preferred conformational states: In addition to the crystallized open state, there exists a closed state which allows the collagen triple helix to contact both the activator and peptidase domain (Fig. 4b). The closed state is latched by two major contacts at the bottom of the saddle and by an alternative 4-helix-bundle arrangement at the saddle seat. Only now are the activator HEAT-repeats able to interact with triple-helical collagen, and initiate the unwinding of the triple-helix α-chains which are processively cleaved ^31,32. Consistent with these data we postulate that the activator and peptidase domains remain mostly closed during collagen cleavage, but relax to the open ground state, once the collagen is degraded, which constituted the source of the attractive interaction for collagenase closing (Fig. 4c).

Figure 4. Unified processing model of triple-helical and microfibrillar collagen.

(a) A collagen triple helix (green) initially docks to the peptidase domain of collagenase (colored as in Fig. 1). In the open state, the activator (dark blue) cannot interact with the substrate, with no hydrolysis occurring.

(b) Step 2, closed conformation, showing the activator HEAT repeats interacting with the triple helix, which is a prerequisite for collagen hydrolysis.

(c) Step 3, semi-opened conformation, allowing for exchange and processive degradation of all three α-chains ^31,32. Once the triple helix is completely cleaved, the collagenase can relax back to the open ground state conformation, as found in our crystals, and thus complete the catalytic cycle.

(d) Collagenase with a docked collagen microfibril (grey; based on pdb entry 4clg). The micro-fibril typically consists of five triple-helical molecules ¹⁹; the triple helix analogous to a is indicated in green.

(e) Step 2, closed conformation with all triple helices but one (green) being expelled from the collagenase. The microfibril wound caused by the triple-helix stripping is indicated in red.

(f) Step 3, semi-opened conformation allowing for completely processing the triple helix, indicated in green. After that, the collagenase will relax back to the open state and only then allow the remaining part of the microfibril to enter the collagenase.

To test this chewing mechanism of collagen degradation, we constructed a variant where the glycine-rich hinge is short cut, Δ(Gly389–Val397). This variant should interfere with the closed conformation (Fig. 4b), and consequently with collagen chewing and digesting. While fully active towards small peptidic substrates (Fig. 2a), this variant exhibits only approximately one third of the wild type activity towards collagen substrates (Fig. 2b), in full agreement with our chew & digest mechanism.

Recognition, unwinding, and processing of microfibrils

One of the most glaring findings of the crystal structure analysis is the perfect match of the observed ground state conformation with the dimensions of collagen microfibrils, having a diameter of approximately 40 Å ^33,34 (Fig. 4d). Albeit a bit more speculative, our model provides a strikingly simple mechanism how collagen microfibrils are proteolytically processed. Upon transition from the microfibril-loaded open state to the closed state, ColG will crimp the microfibril with its pliers, with only one triple helix remaining within the collagenase pliers (Fig. 4e). Analogous to the triple-helical processing, one triple helix remains embraced by the activator and peptidase domains until it is completely clipped, after which the collagenase will relax to the open conformation, allowing the remaining microfibril to enter the collagenase (Fig. 4f).

DISCUSSION

Collagenases are optimized to cleave glycine-rich substrates

One distinct difference between MMPs and ColG relates to the glycine residue which binds to the S1 and S1′ site, respectively. Consistent with transition state theory of enzyme catalysis, the substrate is bound to the active site in a highly distorted conformation (Fig. 3a). We found that Gly on P1′ versus P1. The distortion is counteracted by the flexibility of glycine residues, ubiquitously present in collagen. In ColG, this challenge is addressed by strong binding of the glycine P1′ carbonyl to the secondary oxyanion pocket.

Additionally, the position and orientation of the P1′ amid-nitrogen is fixed within the peptide plane by the P1 carbonyl via the classical oxyanion pocket formed by the Zn²⁺. The ColG solution to the flexible glycine-challenge may thus allow a more efficient substrate distortion as compared to the MMPs, because both dihedral angles (φ and ψ) are fixed upon substrate binding.

Entropically driven collagenolysis

Unraveling collagen (micro)fibrils and unwinding triple-helical collagen are prerequisites for collagen cleavage, both for mammalian MMPs and clostridial collagenases. There are two possible solutions how to store the required mechanical energy, namely either in the (i) protease or the (ii) substrate – or both. The protease-oriented view is supported by the crystal structure of the collagen binding domain of clostridial collagenase G. Here, a Ca²⁺-triggered conformational rearrangement ¹⁹ may implement the energy reservoir which could contribute to the triple-helicase activity of these enzymes.

However, our model suggests and emphasizes the substrate as the energy source powering the molecular machinery (Fig. 4). The required free energy is stored in the substrate itself: Triple-helical collagen is hydrated by at least two ordered water layers that are released upon its cleavage ³⁵. With the accompanying entropy release the substrate thus assists its own catalytic turnover. Importantly, the release of water, and entropy, is independent of the exact mechanism of the triple-helicase and peptidase activity, and as such should hold true for all collagenases, including MMPs.

Transferring mechanistic insights from ColG to MMPs

Naturally, the question arises whether and to which extent the structural insights into the clostridial collagenase (Supplementary Fig. 2a) can be transferred to the mechanism of action in MMPs. It is particularly tempting to speculate which MMP domain may take the role of the activator domain in ColG. To us there are two interesting and alternative hypothetical models. In the first scenario, the MMP hemopexin domain takes the role of the activator domain (Supplementary Fig. 2b), reflecting their strict requirement for collagenase activity^26,36.

The second scenario refers to the observation that MMPs form dimers ^37-42. In this analogy, each the activator and peptidase domains of ColG correspond to a separate MMP catalytic domain, whereas the MMP hemopexin-like domains correspond to the PKD and collagen binding domains (Supplementary Fig. 2c). This model explains the necessity for hemopexin-like domain for collagenase activity by its critical role in MMP dimerisation ^38,43,44. The latter model should be testable by chimeric MMP constructs, e.g. by replacing the hemopexin domain of an MMP collagenase with a suitable dimerisation motif such as GST.

Processivity of clostridial collagenases

Clostridial collagenases were traditionally divided into classes, class I and class II. Whereas the latter group is highly active towards peptidic substrates, class I enzymes such as ColG have a particular preference for collagen substrate degradation in a processive manner ^4,17,28. For fibrillar collagen substrates, processivity includes two dimensions, namely the complete and the flush cut of a (micro-)fibril at one site of the substrate. For triple-helical substrates, processivity implies multiple cleavage events along the substrate, a process described as inch-worming by Overall and colleagues ⁴⁵. The structure suggests how the accessory domains assist both aspects of processivity. When cleaving microfibrillar collagen, the activator and peptidase domains have to open to allow for the remaining microfibril to enter the collagenase active site (Fig. 3d-f). Here, the accessory domains avoid an inadvertant shift of the substrate. On triple helical collagen, but also on microfibrils, the accessory domains help to direct the collagenase module along the substrate. Importantly, this directed movement implies a polarity with the accessory domains leading and paving the way of processing (Fig. 1c). Furthermore, given the tri-carboxypeptidase activity of ColG, we suggest that the collagen processing occurs from the C-terminus of the collagen substrate to its N-terminus, consistent with the antiparallel substrate recognition by the edge strand (Fig. 2).

Potential impact

The presented structural and functional data provide, for the first time, an integrated mechanistic model of collagen recognition and processing by bacterial collagenases; this model may also prove as a valuable model for mechanistic studies on vertebrate collagenases. Given the use of collagenases in diverse biotechnological processes, including islet transplantation ^16,46, the presented results open a structure-based strategy to optimize these processes, for instance by engineering substrate preferences or designing hyper-active collagenases, e.g. by combining key elements responsible for the catalytic properties of class I and class II collagenases. Another important opportunity resulting from this study is in the structure-based design of collagenase inhibitors ⁴⁷. This approach is particularly attractive, as it is expected to enable a step change in “anti-biotics” development: Rather than attacking the pathogen directly, clostridial collagenase inhibitors block the colonization and infiltration of the host by the pathogens only, thereby escaping the Darwinian selection pressure which is encountered in classical antibiotics development.

METHODS

Protein preparation

Collagenase cloning, expression, purification, crystallization, and diffraction data collection, were performed as described previously ^18,48. The PKD-like domain (Ile799–Asn880) was cloned, expressed, and purified analogously. Selenomethionine labeled ColG (Tyr119–Asn880) was expressed in E. coli B834 (DE3) cells overnight at 25 °C in ready-to-use selenomethionine medium (Molecular Dimensions), and purified as described for unlabelled protein.

Biochemical characterization

The peptidolytic assay was performed with FALGPA (Fluka, Sigma Aldrich) as substrate at a final concentration of 0.4 mM, as described previously ²⁸. Collagenolytic activity was assayed with the EnzChek Gelatinase/Collagenase Assay Kit (Invitrogen), using fluorescein-labeled collagen type I from bovine skin according to the manufacturer’s protocol. In short, 0.4, 0.8, and 1.0 nM protein were incubated with 0.2 μg ml⁻¹ substrate and fluorescence was measured in 30 s intervals using the microplate reader Infinite 200 (Tecan) (485 ± 9 nm; 530 ± 20 nm) at room temperature. For α-chymotrypsin from bovine pancreas (Fluka, Sigma Aldrich), 1.0, 10.0 and 100.0 mM were tested accordingly. Initial velocities were calculated using nonlinear regression ⁴⁹.

Crystallization and data collection

Crystals of native, Zn²⁺ depleted, and inhibitor-complexed ColG (Tyr119–Asn880) were grown by sitting-drop vapor diffusion from 7.5 mg ml⁻¹ protein solution, 20–23% PEG 3350, 175–225 mM Na₃C₆H₅O₇, pH 8.3, at 20 °C. Crystals of the PKD-like domain were obtained analogously from 70 mg ml⁻¹ protein solution, 2.5 M C₃H₂O₄Na₂, 10 mM CaCl₂, pH 7.5. Crystals were flash-frozen in a stream of nitrogen gas at 100 K, in which ColG crystals were supplemented with 30% glucose to the mother liquor for cryoprotection. Diffraction data were collected at beamline 14-1 at BESSY (Berlin), and 14-4 at ESRF (Grenoble), and processed within the CCP4 suite ⁵⁰. Data collection and processing statistics are summarized in Table 1.

Structure determination

Using the Auto-Rickshaw four wavelength MAD protocol ⁵¹, heavy-atom positions were located by using the program SHELXD ⁵² and refined with MLPHARE ⁵⁰, and the density improved by solvent flattening using DM ⁵³. The dataset resulted in an initial model that was iteratively improved by a combination of refinement using REFMAC5 ⁵⁴ with the translation libration screw (TLS) description of the anisotropic rigid body motion ⁵⁵, and manual model building in COOT ⁵⁶. Electron densities at uninterpretable loop regions were sharpened by canonical electron densities, leading to effectively improved optical resolution, enabling their interpretation. The refinement was monitored throughout by using an Rfree calculated with 5% of the unique reflections. Refinement and Ramachandran statistics are summarized in Table 1. Molecular figures were created with PyMOL (http://pymol.sourceforge.net) ⁵⁷.

Structure validation

The quality of all models was checked using the programs PROCHECK ⁵⁸, WHATCHECK ⁵⁹, and NQ-Flipper ⁶⁰.