X-ray data

Like all chymotrypsin-like serine proteases, hPR3 and hNE adopt a fold consisting of two β-barrels made each of six anti-parallel β-sheets (Fig. 2B). The structures of the mature forms of the human enzymes were revealed by X-ray crystallography in the late 1980s for hNE, and some years later for hPR3. To date, there are seven structures of hNE deposited in the Protein Data Bank (PDB code: 1B0F [29], 1H1B [30], 1HNE [31], 1PPF [32], 1PPG [33], 2RG3 [34] and 2Z7F [35]) and one of hPR3 (1FUJ [36]). No structures of NE or PR3 from other species are available (only computational models of the murine species have been described) [11,14], nor are structures of the proforms. hNE crystals all contain monomers (1B0F, 1HNE, 1PPF, 1PPG, 2RG3 and 2Z7F) or dimers (1H1B), whereas hPR3 was crystallized as a tetramer, which can be regarded as a dimer of dimers; two monomers in a dimer are oriented so that their active sites face each other, preventing the binding of large substrates. Moreover, the hole in the middle of the tetramer is lined with hydrophobic residues. This characteristic of the crystals of hPR3 has not been observed in the case of hNE, although it might be related to an increased propensity of monomeric hPR3 to interact with hydrophobic environments through this region. Both hPR3 and hNE contain the same four disulfide bridges between cysteine pairs 42–58, 136–201, 168–182 and 191–220.

The overall structural differences between hNE and hPR3 are very small. The rmsd, calculated after structural alignment with stamp [37], on C-alpha atoms of hPR3 and all seven available hNE structures is below 1 Å and the structural difference between the seven hNE structures is in the range 0.3–0.6 Å. The loop between extended sheets 1 and 2 (amino acids 36–39) is the only region showing a significantly larger structural variation as a result of the insertion of three residues (NPG) in hPR3 (Fig. 2A).

Different glycosylation sites

Most structures of hNE (1B0F, 1H1B, 1PPF, 1PPG and 2Z7F) reveal the presence of two sugar moieties on both Asn109 and Asn159, whereas one of the latest structures shows only one occupied site (Asn159 for 2RG3). In the structure of hPR3, only one glycolysation site (Asn159 and not Asn113) is occupied by a sugar. All three glycosylation sites (Asn109, Asn113 and Asn159) are remote from the catalytic triad and the ligand binding sites (3, 4). According to Specks et al. [38], both sites are occupied in neutrophil hPR3, although they have different functional significance; glycosylation on Asn159 influences hPR3 thermostability and increases significantly the catalytic activity measured on a small peptidic substrate (N-methoxysuccinyl-Ala-Ala-Pro-Val), whereas glycosylation on Asn113 appears to be important (but is not an absolute requirement) for an efficient N-terminal processing of hPR3. Interestingly, of both glycosylation sites, Asn113 is the furthest away from the N-terminal end of hPR3, as observed from the structure of the mature form. Unfortunately, no structure of any of the proforms of hPR3 or hNE is available, although X-ray structures of proforms of homologues have been reported (pro-granzyme K, 1MZA [39], chymotrypsinogen, 2CGA [40], and trypsinogen, 2TGT [41]), where it can be clearly seen that the N-terminal end (residues 14–17) freely extends out of the core of the protein along the extended sheet containing residue 159 (extended sheet numbered 8 in Fig. 3). Residue 159 in chymotrypsinogen and pro-granzyme K is thus very close to the extended N-terminal segment.

Different surface properties

Earlier studies have questioned the relationship between the net charge difference and the functions of PR3 and NE [36,42]. Indeed, at neutral pH, hNE has a net charge of +10 (it contains 19 arginines and only nine acidic residues), whereas hPR3 has a much lower net charge of +2, although it contains approximately the same number of charged amino acids as hNE. The sequence of the mature form contains 13 arginines, two lysines, ten aspartic acids and four glutamic acids. Fujinaga et al. [36], as well as subsequent in silico pK_a calculations [10,43], suggest Asp213 to be protonated. The availability of the enzymes atomistic structures allows the calculation of the electrostatic surface potential (i.e. the electrostatic potential created by all the amino acids of the enzyme in its vicinity) (Fig. 5). It is a critical determinant of its surface properties and it is more relevant to its structure–activity relationship because it reflects not only the net charge, but also the charge distribution. Electrostatic interactions are known to play a key role in macromolecular interactions (e.g. with partner proteins, ions or membrane binding); thus, they might explain protease-specific functions. Interestingly, in the case of hNE, there is an omnipresence of electropositive potential that covers most of the surface of the enzyme, except at the substrate-binding site. This is not the case for hPR3, where electropositive clusters (or ‘patches’) alternate with negative and neutral patches. This pattern has been shown to be particularly suited for peripheral membrane binding of other proteins [44–46]. Moreover, the electrostatic surface properties of hPR3 and hNE around the active site plays an important role in the substrate specificity of the enzymes [10,11] and in their propensity to be part of multiprotein complexes.

Kinetic studies on hNE and hPR3

Many studies have been devoted to the understanding of hNE specificity [47–54]. Substrates with a hydrophobic side chain at P1 are efficiently cleaved by hNE and P1-valine is preferred over an alanine or a phenylalanine. Substrates with P1-Ile can also be hydrolyzed [48] by hNE. Peptides with P1-Met residues can also be cleaved, whereas oxidation of the methionine decreases the binding to hNE [50]. Extension of the peptide chain results in significant increases in catalytic efficiency and P1-specificity becomes broader with decreasing chain length. Compared to hNE and other serine proteases, there are relatively few available results on the specificity of PR3. The earliest studies revealed the preference of PR3 for ‘small hydrophobic amino acids’ such as alanine, serine and valine [55]. Brubaker et al. [56] emphasized that norvaline is preferred to valine, which itself is better than an alanine. Substrates with a methionine at P1 would also be efficiently cleaved by PR3 [57,58]. These studies focused on the P1 amino acid and described differences mostly resulting from the size or volume of the S1 pocket. Most importantly, they could not rely on structural data because no structure of PR3 was then available.

Regarding hNE, most studies used a substrate with chromophoric leaving groups (i.e. these do not extend with amino acids in P′ sites), which explains why the importance of the P′–S′ interaction has been overlooked for many years. Recently, however, the importance of S′–P′ interactions has been clearly established for hPR3 [54,58]; substrates extending beyond P1′ have a systemic favorable effect on PR3 catalysis.

Insights from the abundant structural X-ray data on hNE

The first NE structure solved in 1986 was co-crystallized with the third domain of the turkey ovomucoid inhibitor (OMTKY3) (1PPF) [32], which is a canonical protein inhibitor of the Kazal family. It is bound noncovalently and provides structural insights into how peptidic substrates bind to NE in the Michaelis complex. Recently, the crystal structure of hNE with another proteic inhibitor was released (2Z7F) [35]; the secretory leukocyte protease inhibitor is a secreted inhibitor that protects epithelial tissues from serine proteases. Additional NE structures were solved with synthetic inhibitors in the binding site, including chloromethyl ketone peptides MeO-Suc-Ala-Ala-Pro-Val (1PPG) [33] and MeO-Suc-Ala-Ala-Pro-Ala (1HNE) [31] and an orally active peptidyl pentafluoroethyl ketone (1B0F) [29]. The structure of hNE with low- molecular weight inhibitors has also been reported; MacDonald et al. [30] reported the structure of NE with a nonpeptidic inhibitor, a pyrrolidine trans-lactame which also binds covalently to Ser195 (1H1B). Later Huang et al. [34] reported the structure of NE with another mechanism based (suicide) inhibitor, a derivative of a 1,2,5-thiadiazolidin-3-one 1,1-dioxide (2RG3). This scaffold has been described as particularly suited for the design of specific NE-inhibitors [59]. Mechanism-based inhibitors are covalently linked to the active site (either Ser195 or both Ser195 and His57) and mimic either the conformation of a tetrahedral intermediate (tetrahedral adduct) or of the acyl enzyme.

Thus, unlike hPR3, hNE has been crystallized with several ligands (see list in Table 1); these structures provide information about the interactions possible between the enzyme and different types of substrates, and thereby about its specificity much more extensively than in the case of hPR3. The different X-ray structures show that pockets S1–S4 are mostly made of hydrophobic amino acids [29,30,32,33], with the exception of the side chain of Arg217, which can stabilize polar or acidic groups at P4 [32] and even at P5 [33]. Interestingly, the benzyl of the N-protecting group in 1B0F forms a π–π stacking interaction with Phe215. An additional feature of the ‘unprimed subsites’ is that residues 214–216 form an antiparallel β-sheet with backbone atoms of residues P1–P3 of protein and peptidyl inhibitors which is known as the ‘canonical conformation’ [60]. The primed substrate binding sites are less well defined; the low molecular weight compounds do not extend far into the prime sites, and the same applies for the peptidyl inhibitors. Proteic inhibitors reveal some contacts in the prime site but, because of their size and rigidity, it is possible that the contacts they make with the enzymes are not exactly comparable with the contacts small flexible peptides would make. An exhaustive list of amino acids generally described as forming the binding sites is provided elsewhere [53,60].

Mapping the substrate binding sites of hPR3: the contribution of in silico studies

The release of the first X-ray structure of hPR3 [36] and its comparison with already known structures of hNE has revealed some of the basis for the difference specificity between the two enzymes, despite the lack of a structure of a complex between hPR3 and a substrate. Indeed Fujinaga et al. [36] describe the S1 pocket, but also report a list of the amino acids constituting the neighboring binding sites, which most probably form the basis for the differential substrate specificity of hPR3 and hNE. The Ala213 to Asp213 substitution in PR3 reduces the size of the S1 pocket (from 152.7 to 98.6 ų), making P1 more restrictive in hPR3 than hNE. The Asp213 is observed to be hydrogen-bonded to the carbonyl of Gly197 and its protonation state (and increased pK_a value) is probably a result of the hydrophobic environment. Indeed, using molecular dynamics simulations and pK_a calculations, we predicted a significant increase of the pK_a value of Asp213 compared to its value in solution [43]. Fujinaga et al. [36] suggest that the substitution of Leu99 to Lys99 makes the S2 pocket of hPR3 deeper and more polar compared to hNE, and that it increases the polarity in S4. The substitutions of Arg217 to Ile217 and Gly218 to Trp218 in hPR3 compared to hNE should render the S5 binding site more hydrophobic. For the primed subsites of hPR3, the most significant substitution is Asn61 to Asp61. Consequently, hPR3 should prefer basic residues at P1′ and P3′. The substitution of Ile151 to Pro151 and Ile143 to Arg143 will create a basic S2′ site favoring the binding of acidic residues.

However, because there is no X-ray structure of hPR3 with a substrate in its active site, there is no accurate description of the ligand–enzyme interactions such as there is for hNE. In an attempt to fill this gap, we performed molecular dynamics simulations of hPR3 and hNE complexed with peptides of varying sequences and sizes [10–12]. Indeed, molecular dynamics simulations are very well suited and widely used [61–63] for providing both structural and dynamical information, and thus neatly complement experimental techniques such as X-ray and NMR. These simulations have provided structures at the atomic level of detail of hPR3, a map of the S and S′ sites (Fig. 6) and a description of the interaction scheme for six different peptidic ligands. The simulations show no direct interaction between Asp213 and the P1 amino acids, suggesting that a polar amino acid in S1 is not something that hPR3 can accommodate better than hNE. Relatedly, Asp213 does not appear to have a role in specificity, in agreement with that suggested by Fujinaga et al. [36]. The specificity difference between the two enzymes lies in S2, S1′, S2′ and S3′, which are all more polar in hPR3 than in hNE. S2 in hPR3 is clearly suited to accommodate a negatively-charged amino acid, such as Asp, with Lys99 and Arg60 acting as potential hydrogen bond donors. It is interesting to note that Lys99 contributes to both the hydrophobic S4 site and to the polar S2 site, playing two different roles. Similarly S1′ and S3′ are inter-connected sites of polar character because they are made of Asp61 (as well as other amino acids). It participates in the two interaction sites and interacts with lysine or arginine side chains of the ligand. S2′ is characterized by Arg143, which can interact through hydrogen bonds with a negatively-charged amino acid at the P2′ position of the substrate sequence. Ile190 of hPR3 is not observed to be in close contact with the P1 side chain but Val216 constitutes the main interaction partner in the S1 site, together with the CβH₂ group of Ser195. The S4 and S3 sites are hydrophobic in both hPR3 and hNE. Another important conclusion derived from the molecular dynamics simulations is that the seven sites are shown to be interconnected; in particular, S4 and S2 share Lys99, which plays both a hydrophilic and a hydrophobic role, with the latter being achieved as a result of the CH₂ groups of its long sidechain. S1′ and S3′ overlap and share Asp61, and the lack of a P′–S′ interaction modifies the network of S–P interactions. This is an important result in the context of drug design and the investigation of enzyme–ligand binding sites by directed mutagenesis experiments. At least as important is the finding that different amino acids of the enzyme can alternatively participate in one given recognition site or another depending on the nature of the sequence of the ligand. This illustrates the adaptability of the enzyme; most of the recognition sites are not rigid predefined pockets. Rather, they are regions at the surface of the enzyme that have a significant degree of flexibility and can adapt to the substrate.

Table 2 summarizes the preferred amino acid types at each P or P′ site and it can be used to design peptidic substrates specific of one enzyme or the other. Using these data, we have designed the peptide sequence VADVKDR and demonstrated that it is highly specific for hPR3 versus hNE [10]. We find that the cleavage site of p21/waf1 (QEA-RER) [64] also conforms to the pattern listed in Table 2. Knowledge of the sequence pattern binding to hPR3 can also be used to help identify novel endogenous substrates of the proteinase.

Divergence of hypotheses from experimental studies

Several studies have attempted to characterize the association of PR3 and NE with the neutrophil membrane. It has been shown that only PR3, and not NE, is already present at the plasma membrane of inactivated PMN [65,66]. We have previously reported a specific association of PR3 with the plasma membrane, which is stronger than simply an ionic interaction [67], and some studies argue in favor of a weak charge-dependent mechanism similar for the two proteases [42].

A number of proteins have been experimentally identified as potential partners of PR3 at the membrane, which might be of critical importance for its functions, as well as for our understanding of its involvement in Wegener’s granulomatosis [68,69]. Aviram and colleagues provided evidence of colocalization of PR3 with the integrin CD11b/CD18 (b2 integrin) [70], the Fcgamma receptor FcgRIIIb and the p22^phox subunit of cytochrome b558 in the membrane, and PR3 is demonstrated to be localized in lipid raft domains [71,72]. More recently, Bauer et al. [73] reported the expression of membrane hPR3 (mbPR3) and CD177 (NB1) on the same subset of neutrophils, whereas von Vietinghoff et al. [74] proposed that PR3 membrane expression could be mediated through CD177 binding. Interestingly, we demonstrated that PR3 could be externalized at the plasma membrane during the very early stages of apoptosis in the absence of degranulation, thus strongly suggesting an extragranular pool of PR3 [75]. Most interestingly, we reported a co-localization of PR3 with phospholipid scramblase 1, a myristoylated membrane protein present in lipid rafts and involved in the redistribution of membrane lipids during neutrophil apoptosis [76]. Co-immunoprecipitation studies performed on neutrophil lysates provided evidence indicating that both PR3 and phospholipid scramblase 1 were associated within the same protein complex. However, to our knowledge, there is no evidence of a physical interaction between mbPR3 and any of these potential partners, highlighting the lack of biophysical and structural data on the membrane binding mechanism of PR3. It should be noted that the diversity of the potential PR3 partners might reflect specific functions and especially regulatory functions in neutrophil activation or apoptosis, which are currently being unravelled.

By contrast, a direct interaction between hPR3 and lipid vesicles has been demonstrated by Goldman et al. [77]. They investigated the interaction of hPR3, hNE and human myeloperoxidase (an enzyme also present in neutrophil azurophilic granules), with reconstituted lipid bilayers containing zwitterionic and anionic phospholipids; pure dimyristoylphosphatidylcholine, dimyristoylphosphatidylcholine/dimyristoylphosphatidylglycerol (70 : 30, 50 : 50 or 30 : 70) and pure dimyristoylphosphatidylglycerol. Their results show that the molar affinity of hPR3 is better than the one of hNE and myeloperoxidase for all mixtures with anionic lipids, and that only hPR3 anchors to neutral bilayers. According to their findings, hPR3 associates with the mixed bilayer (50 : 50) with strong hydrophobic interactions (i.e. by inserting hydrophobic amino acids in the lipid bilayer), whereas hNE and myeloperoxidase only have weak hydrophobic interactions with the lipids. Interestingly, they report that bilayer-bound hPR3 has a reduced catalytic efficiency, whereas its inhibition by α₁-antitrypsin is more important than that observed for the soluble form of the enzyme. The latter is in contradiction to other studies [42,78]. It has been shown that hPR3 enzymatic activity is not inhibited by its inhibitors α₁-proteinase inhibitor (α1-PI, 52 kDa) and elafin (6 kDa) when bound to the outer cell surface of neutrophils, but only by a low-molecular-weight protease inhibitor (phenylmethansulfonyl fluoride). This suggests that α1-PI and elafin are not active on mbPR3 as a result of steric hindrance.

On the sole basis of visual inspection of the 3D structure, Goldman et al. [77] suggest that the region constituted by Phe166, Ile217, Trp218, Leu223 and Phe224 might be involved in the insertion of hPR3 into the lipids. This study has shown that hPR3 can bind to lipid bilayers directly (i.e. with direct physical interaction between amino acids of the protein and lipids) and is stable without any other partner. It strongly suggests that it can bind with the same mechanism to cellular membranes.

Structural determinants predicted by in silico studies

Although the application of NMR and X-ray methods to peripheral membrane proteins remains challenging, theoretical approaches have proven extremely valuable and reliable for studying their structure and membrane-binding mechanisms [79–85]. We used molecular dynamics simulations with implicit membrane representation to allow for a cost-effective investigation of the binding of PR3 and NE to the lipid bilayers, in the presence of different types of lipid bilayers that vary with respect to their anionic character. The results obtained show that, unlike hNE that binds only to negatively-charged membranes, hPR3 is able to bind both anionic and neutral membranes but with a preference for negatively charged lipids; the calculated binding energies are −10.87 and −3.07 kcal·mol⁻¹ for anionic and neutral membranes, respectively. Furthermore, the simulations of the binding mechanism reveal a unique membrane-binding site and binding mechanism of PR3. It involves a few basic amino acids that orient PR3 towards the membrane in the correct direction to allow it to insert a hydrophobic patch comprising F165, F166, F224, L223, F184 and W218. These residues are carried by surface loops (3, 4) and in silico mutations abolish membrane association. NE follows a different binding mechanism, where different regions of its highly electropositive surface (Fig. 5B) are able to interact with anionic lipids, but without insertion of hydrophobic amino acids, and thereby in a much shallower way than PR3.

A mutagenesis analysis performed in rat basophil leukaemia (RBL) cells transfected with wild-type PR3 or PR3 mutated within the hydrophobic patch demonstrated that the latter was essential for membrane insertion [69]. The mechanism of membrane anchoring described by Goldman et al. [77] and ourselves (i.e. electrostatically driven attraction by basic amino acids and insertion of hydrophobic amino acids directly in the membrane) is not incompatible with the presence of a partner protein such as NB1. Such interactions might simply reinforce or perpetuate the membrane anchorage and/or facilitate the function(s) of membrane-bound PR3.

Substrate specificity differs from the human enzymes

Specks and coworkers [100] demonstrated that the mouse and human PR3 (mPR3 and hPR3, respectively) have different physicochemical properties. Using inhibitors and short hydrophobic peptides (not extending in the P′ sites), they show that they have different substrate specificities. Unfortunately, the size and the nature of the peptides limit the conclusions that can be drawn with respect to the P1 sites. The lack of structural information on the murine enzymes and on their complexes with cleavable peptides makes it difficult to determine the basis for their substrate specificity. We used homology modeling to build structural models of the murine enzymes and performed molecular dynamics simulations of 14 different enzyme–peptide complexes [11], and their analysis provides an inventory of the amino acids forming the substrate binding sites of the murine enzymes. The surface of mPR3 is globally more electronegative than the one of hPR3 (Fig. 5), whereas the differences between mNE and hNE are less significant on the overall protein surfaces. In particular, substitutions of several basic amino acids around the substrate binding sites of hPR3 change its surface properties: R60Q, R63bQ, R74L, N98E, K99N and G219E. Unlike hPR3, mPR3 is thus unlikely to bind substrates with acidic groups (Asp, Glu) on the S side. Consequently, very efficient substrates of hPR3 might be poor substrates of mPR3. The specificity difference between mPR3 and mNE lies in S1′ and S3′. The S2 and S2′ sites in both enzymes might accommodate the same type of amino acids, although the differences between these sites in the human enzymes were used for designing specific substrates. Table 2 summarizes the types of amino acids preferred at sites P6–P4′, and also for the human and mouse enzymes. Such an inventory will benefit the development of highly specific substrates of each of the neutrophil serine proteases [14].

Membrane binding site: poor sequence conservation between species

Half of the amino acids predicted to be involved in the membrane binding of hPR3 are not conserved in mPR3; F166L, F184L, K187A, W218R, R222L and L223Q. The mouse form of PR3 lacks three aromatic and one basic amino acid compared to the human form. The former are known, from integral membrane proteins, to be found at the membrane interface, whereas the latter provides electrostatic interactions with the polar lipid heads. In addition, we have seen that the mouse form is globally more electronegative than the human form. As a consequence of these differences in sequence and structural properties, mouse PR3 is likely to be able to bind to the plasma membrane using the same region as human PR3, although less strongly and specifically, as observed by Wiesner et al. [100].

In conclusion, molecular studies of the comparison between human and mouse PR3 clearly point to major differences in their catalytic properties and their ability to interact with membranes, which is a key determinant of the accessibility to a specific substrate and, ultimately, their function.