Crystal structure of highly glycosylated human leukocyte elastase in complex with an S2′ site binding inhibitor

A novel binding mode of a small-molecule inhibitor of human leukocyte elastase is revealed by its co-crystal structure with the enzyme. In the structure, a comparatively large part of the N-glycan chains attached to the enzyme are visible.

Abstract

Glycosylated human leukocyte elastase (HLE) was crystallized and structurally analysed in complex with a 1,3-thiazolidine-2,4-dione derivative that had been identified as an HLE inhibitor in preliminary studies. In contrast to previously described HLE structures with small-molecule inhibitors, in this structure the inhibitor does not bind to the S1 and S2 substrate-recognition sites; rather, this is the first HLE structure with a synthetic inhibitor in which the S2′ site is blocked that normally binds the second side chain at the C-terminal side of the scissile peptide bond in a substrate protein. The inhibitor also induces the formation of crystalline HLE dimers that block access to the active sites and that are also predicted to be stable in solution. Neither such HLE dimers nor the corresponding crystal packing have been observed in previous HLE crystal structures. This novel crystalline environment contributes to the observation that comparatively large parts of the N-glycan chains of HLE are defined by electron density. The final HLE structure contains the largest structurally defined carbohydrate trees among currently available HLE structures.

1. Introduction  

Human leukocyte elastase (HLE; EC 3.4.21.37; UniProt entry P08246) is a chymotrypsin-like serine protease composed of 218 amino acid residues (Sinha et al., 1987 ▸). A synonym for the enzyme is human neutrophil elastase (HNE), reflecting the fact that it is produced by neutrophilic granulocytes, which are the most abundant cells of the innate immune system in human blood. HLE is encoded by the ELANE gene (Takahashi et al., 1988 ▸). Its expression provides a precursor protein of 267 residues that undergoes a multistep maturation process involving transport to the endoplasmic reticulum and cleavage of the corresponding signal peptide, the removal of both C-terminal and N-terminal propeptides, the formation of four disulfide bonds and N-glycosylation at three Asn side chains (Loke et al., 2017 ▸). Mature HLE is preferentially stored in peroxidase-positive (azurophilic) granules of neutrophils as part of ‘a library of innate immunity proteins’ (Borregaard et al., 2007 ▸) including the HLE relatives cathepsin G and human proteinase 3 (hPR3; Hajjar et al., 2010 ▸). The enzyme is subsequently released into the extracellular space either for direct protective functions in the context of antimicrobial defence (Hirche et al., 2008 ▸) or to perform regulatory tasks within the innate immune response (Sumer-Bayraktar et al., 2016 ▸). HLE is prohibited from causing proteolytic damage to the connective tissue, which is a particular problem in chronic obstructive pulmonary disease (COPD; Hiemstra et al., 1998 ▸), by α₁-antitrypsin, a serpin-type protease inhibitor, which is likewise produced by neutrophils (Clemmensen et al., 2011 ▸). HLE and α₁-antitrypsin are stabilized by N-glycosylation (Sarkar & Wintrode, 2011 ▸); furthermore, recent experimental evidence presented by Loke et al. (2017 ▸) suggests that N-glycosylation serves to modulate the protein–protein interaction of the two partners and thus the inhibition of proteolytic activity, as discussed below in more detail.

HLE is the target of drug development efforts owing to its key roles in inflammatory diseases and in defence against bacterial infections (Macdonald et al., 2002 ▸; Huang et al., 2008 ▸; Hansen et al., 2011 ▸; von Nussbaum et al., 2015 ▸, 2016 ▸). Significantly, 15 of the 19 structures of HLE currently available in the Protein Data Bank (PDB; Berman et al., 2000 ▸) are co-crystal structures with small-molecule inhibitors, the majority of which are drug candidates or intermediates in drug development processes.

The work presented here was also motivated by an HLE inhibitor, a 1,3-thiazolidine-2,4-dione derivative (Fig. 1 ▸ a) with antibacterial activity (Zvarec et al., 2012 ▸), that we now show to inhibit HLE with an IC₅₀ of ∼0.5 µM (see below). To elucidate the binding mode of this compound to HLE, we performed co-crystallization studies. This led, as shown below, to a novel crystal packing of HLE in which a relatively large number of carbohydrate residues are visible at two of the glycosylation sites of the enzyme. Recently, a comprehensive structural analysis of HLE glycosylation indicating novel functional roles for the carbohydrate chains has been published (Loke et al., 2017 ▸). This is developed further here, more so than in previous HLE structural publications where little attention has been given to glycosylation.

Figure 1.

The HLE inhibitor CQH, as originally described by Zvarec et al. (2012 ▸), and its effect on HLE. (a) Structural formula of CQH with atom numbering and ring labelling. (b) CQH tautomerization at ring A and delocalization of the established negative charge by mesomerism. (c) CQH in the final 2F _o − DF _c density (cutoff level 1σ) which was averaged with Coot (Emsley et al., 2010 ▸) over the two NCS copies; both CQH tautomers were drawn with the best-fitting conformation. (d) Hill plot for the inhibition of HLE with CQH. Data points are mean values from quadruplicate experiments. (c) was prepared with PyMOL v.1.7 (Schrödinger). The data in (d) were analysed and visualized with GraphPad Prism 5 for Windows v.5.04 (GraphPad Software, San Diego, California, USA).

2. Materials and methods  

2.1. Inhibitor  

The inhibitor that was co-crystallized with HLE (Fig. 1 ▸) was synthesized as described by Zvarec et al. (2012 ▸), who referred to the compound as 5h. Here, we use the designation CQH that was assigned by the PDB during the structure-deposition process. A ring-labelling and atom-numbering scheme for CQH is shown in Fig. 1 ▸(a).

2.2. Macromolecule production  

HLE from human blood was purchased from SERVA Electrophoresis GmbH, Heidelberg, Germany as a powder (Lot No. 111430) obtained by lyophilization from 50 mM acetate buffer pH 5.5. A molecular mass of 29.5 kDa was given on the certificate of analysis which, when compared with the value of 23 224 Da calculated from the amino acid sequence (Table 1 ▸), indicates a high degree of glycosylation.

Table 1. HLE production information.

Source organism	Homo sapiens
Source tissue	Human blood
Manufacturer	SERVA Electrophoresis GmbH, Heidelberg, Germany
Complete amino acid sequence of the construct crystallized	IVGGRRARPHAWPFMVSLQLRGGHFCGATLIAPNFVMSAAHCVANVNVRAVRVVLGHNLSRREPTRQVFAVQRIFENGYDPVNLLNDIVILQLNGSATINANVQVAQLPAQGRRLGNGVQCLAMGWGLLGRNRGIASVLQELNVTVVTSLCRRSNVCTLVRGRQAGVCFGDSGSPLVCNGLIHGIASFVRGGCASGLYPDAFAPVAQFVNWIDSIIQ

The following HLE-homologous serine proteases were obtained in order to perform comparative inhibition analyses: bovine trypsin from Sigma, Steinheim, Germany; human thrombin from Calbiochem, Darmstadt, Germany; and human cathepsin G from AppliChem, Darmstadt, Germany.

2.3. Inhibition experiments with HLE and other proteases  

Catalytic activity assays with suitable para-nitroanilide (pNA) derivatives as artificial substrates have been described for HLE ([MeOSuc-Ala-Ala-Pro-Val-pNA] = 100 µM; Gütschow et al., 2005 ▸) and for the HLE homologues cathepsin G ([Suc-Ala-Ala-Pro-Phe-pNA] = 500 µM; Neumann et al., 2001 ▸), trypsin ([Suc-Ala-Ala-Pro-Arg-pNA] = 200 µM; Sisay et al., 2010 ▸) and thrombin ([H-d-Phe-l-Pip-Arg-pNA] = 100 µM; Sisay et al., 2010 ▸); they were used here to determine the IC₅₀ values of CQH for these proteases. Each combination of enzyme, substrate and inhibitor concentration was measured in duplicate with the exception of the HLE experiments that were performed in quadruplicate. For the inhibition of HLE by CQH (0.4–2.0 µM), a Hill plot was generated with the apparent Hill coefficient n ^I _app as the slope and the logarithmic IC₅₀ value as the intersection with the abscissa using the double logarithmic Hill equation (1),

where v ₀ is the enzymatically catalysed initial reaction velocity at a given substrate concentration and in the absence of inhibitor, v is the enzymatically catalysed initial reaction velocity at given substrate and inhibitor concentrations, [I] is the molar inhibitor concentration, n ^I _app is the slope of the Hill plot, reflecting the apparent number of inhibitor binding sites in the enzyme, and IC₅₀ is the inhibitor concentration at which v equals 50% of v ₀.

The Hill plot for HLE (Fig. 1 ▸ d) gave an IC₅₀ of 0.494 µM and a Hill coefficient of 1.53, indicating the binding of more than one CQH ligand per HLE molecule. The inhibitory potency of CQH against other proteases is much reduced, with IC₅₀ values of 25.7 µM for cathepsin G, greater than 40 µM for thrombin and greater than 500 µM for trypsin.

2.4. Crystallization  

The HLE lyophilizate as purchased was dissolved in water, resulting in the following protein solution: 5 mg ml⁻¹ HLE, 250 mM acetic acid/sodium acetate pH 5.5. No deglycosylation was performed. A 10 mM stock solution of the inhibitor CQH in dimethyl sulfoxide (DMSO) was also prepared. Prior to crystallization, 120 µl HLE solution was mixed with 6 µl CQH solution and cleared of any precipitate by centrifugation (24 000g). This HLE–CQH mixture was used for crystal screening. Typically, 1 µl HLE–CQH mixture was combined with 0.5 µl reservoir solution per crystallization drop.

Initial HLE–CQH crystals were found in a setup using 20%(w/v) polyethylene glycol monomethyl ether 5000 (PEG MME 5000), 0.2 M potassium sulfate. Repeated microseeding steps with 1:1000 and 1:10 000 dilutions of the seed suspensions were necessary to obtain usable, albeit still tiny, crystals. In the final seeding step, the reservoir composition was 20% PEG MME 5000, 0.2 M sodium sulfate. Crystallization information is summarized in Table 2 ▸.

Table 2. Crystallization.

Method	Vapour diffusion, sitting drop
Plate type	Cryschem 24-well plate
Temperature (K)	293
Protein concentration (mg ml−1)	5
Buffer composition of protein solution	0.5 mM CQH, 5%(v/v) DMSO, 250 mM sodium acetate pH 5.5
Composition of reservoir solution	20%(w/v) PEG MME 5000, 0.2 M sodium sulfate
Volume and ratio of drop	1 µl HLE–CQH solution:0.5 µl reservoir solution
Volume of reservoir (µl)	200

2.5. Data collection and processing  

HLE–CQH crystals were flash-cooled in liquid nitrogen directly from the crystallization mother liquor. X-ray diffraction data were measured at the European Synchrotron Radiation Facility (ESRF), Grenoble, France and at the Swiss Light Source (SLS), Villigen, Switzerland. The best data set (Table 3 ▸) was collected on SLS beamline X06DA. It was processed with XDS (Kabsch, 2010 ▸), scaled and merged with AIMLESS (Evans & Murshudov, 2013 ▸) and converted to structure-factor amplitudes using CTRUNCATE from the CCP4 software package (Winn et al., 2011 ▸).

Table 3. Data collection and processing.

Values in parentheses are for the outer shell.

Diffraction source	Beamline X06DA, SLS
Wavelength (Å)	1.0000
Temperature (K)	100
Detector	PILATUS 2M, Dectris
Crystal-to-detector distance (mm)	270
Rotation range per image (°)	0.15
Total rotation range (°)	360
Exposure time per image (s)	0.248
Space group	H32
a, b, c (Å)	204.56, 204.56, 62.16
α, β, γ (°)	90, 90, 120
Mosaicity (°)	0.134
Resolution range (Å)	45.6–2.70 (2.83–2.70)
Total No. of reflections	283984 (29285)
No. of unique reflections	13722 (1359)
Completeness (%)	99.94 (100)
Multiplicity	20.7 (21.6)
〈I/σ(I)〉	15.40 (1.12)†
R r.i.m. ‡	0.247
Overall B factor from Wilson plot (Å2)	62.04§

^†The high-resolution limit was set according to a recommendation implemented in AIMLESS (Evans & Murshudov, 2013 ▸) that uses a CC_1/2 limit of 0.3 (Karplus & Diederichs, 2012 ▸). AIMLESS reports that 〈I/σ(I)〉 falls below 2.0 beyond a resolution limit of 2.92 Å.

^‡Estimated R _r.i.m. = R _merge[N/(N − 1)]^1/2, where N is the data multiplicity.

^§The Wilson plot does not show any anomalies. No ice-ring-related problems were reported by phenix.xtriage (Adams et al., 2010 ▸).

2.6. Structure solution and refinement  

The structure was solved by molecular replacement using Phaser (McCoy et al., 2007 ▸) for Patterson searches with PDB entry 1ppg (Wei et al., 1988 ▸) as a search model. The subsequent refinement was performed with PHENIX (Adams et al., 2010 ▸) supported by Coot (Emsley et al., 2010 ▸) for manual model building and in particular for adding sugar residues at the glycosylation sites. The carbohydrate chains were validated with Privateer (Agirre, Iglesias-Fernández et al., 2015 ▸) from the CCP4 suite (Winn et al., 2011 ▸) in order to avoid typical errors in carbohydrate model building (Agirre, Davies et al., 2015 ▸; Agirre, 2017 ▸).

The electron density allowed the easy identification and fitting of two copies of the inhibitor CQH (Fig. 1 ▸ c). For refinement, CQH was parametrized using the phenix.elbow routine (Moriarty et al., 2009 ▸) in PHENIX (Adams et al., 2010 ▸). The CQH conformation in the zone from ring B to C is unambiguously indicated by electron density (Fig. 1 ▸ c), while in the ring A regions of both copies of CQH modelling was more difficult: in the standard tautomer of CQH (Fig. 1 ▸ a) a double bond exists between C atoms 5 and 8 which implies planarity of all ring A atoms with C atoms 8, 9, 12 and 15. This, however, is not supported by the electron density and leads to steric clashes in the environment (Fig. 1 ▸ c). Therefore, an alternative tautomeric form of CQH was used for modelling in which a single bond exists between C atoms 5 and 8 (Fig. 1 ▸ b). This allows rotation around this bond and leads to a better fit of ring A to the electron density in both copies of CQH (Fig. 1 ▸ c). Nevertheless, in this conformation the ring A region also refines with comparatively high temperature factors for the corresponding atoms, indicating some disorder, which is also reflected by the electron density.

In the final model the amino acid residues are numbered according to the classical chymotrypsinogen-based sequence-numbering scheme introduced by Hartley & Shotton (1971 ▸), i.e. with the histidine, the aspartate and the serine of the catalytic triad being numbered 57, 102 and 195, respectively. The refined structure and the structure-factor amplitudes have been deposited in the PDB as entry 6f5m. Refinement statistics are summarized in Table 4 ▸.

Table 4. Structure solution and refinement.

Values in parentheses are for the outer shell.

Resolution range (Å)	45.553–2.700 (2.8423–2.7000)
Completeness (%)	100.0
σ Cutoff	F > 1.37σ(F)
No. of reflections, working set	12675 (1814)
No. of reflections, test set	1046 (129)
Final R cryst	0.1758 (0.2979)
Final R free	0.2336 (0.3568)
Cruickshank DPI (Å)	0.325
No. of non-H atoms
Protein	3272
Ion	32
Ligand	297
Water	44
Total	3645
R.m.s. deviations
Bonds (Å)	0.002
Angles (°)	0.542
Average B factors (Å2)
Protein	61.04
Ion	91.95
Ligand	100.24
Water	49.96
Ramachandran plot
Most favoured (%)	96.76
Allowed (%)	3.24

3. Results and discussion  

3.1. Noncrystallographic symmetry and quaternary structure  

Calculation of the Matthews parameter led to a value of 2.55 ų Da⁻¹ and to a probability of 0.98 (Kantardjieff & Rupp, 2003 ▸) that the asymmetric unit is occupied by two HLE chains. A full self-rotation map with polar angles was calculated in order to analyse the noncrystallographic symmetry (NCS) relation between the two protomers (Figs. 2 ▸ a, 2 ▸ b and 2 ▸ c). Consistent with the rhombohedral crystallographic symmetry (Table 3 ▸), the highest peaks in this map are found in the κ = 120° section (along the c axis; Fig. 2 ▸ a) and the κ = 180° section (along the a axis; Fig. 2 ▸ b). Strong twofold NCS peaks (with a height of 90% of the crystallographic maximum) are apparent in the direction of the crystallographic c axis as well as perpendicular to the c axis, namely both parallel to the b* axis and in directions deviating 60 and 120° from the b* axis (Fig. 2 ▸ b). The twofold NCS operation in the c direction combines with the crystallographic threefold symmetry to give a sixfold NCS rotation (Fig. 2 ▸ c), resulting in pseudo-point group 622. Such a crystalline arrangement has not been observed in any of the 19 existing crystal structures of HLE.

Figure 2.

Noncrystallographic symmetry (NCS) and dimeric quaternary structure. (a, b, c) Significant κ sections of a full self-rotation map with contouring from 50 to 100 (relative maximum) in intervals of 10; NCS peaks are indicated with red marks. (d) Asymmetric unit of the HLE–CQH co-crystals containing a dimeric HLE₂–CQH₂ complex that is probably stable as assessed by PISA (Krissinel & Henrick, 2007 ▸); the view is along the twofold NCS dyad (indicated by a twofold-axis symbol). (e) Equivalent dimers to those in (d) formed by human proteinase 3 (hPR3) as part of a probably stable D2 tetramer (Fujinaga et al., 1996 ▸). (f) Hydrophobic core of the HLE₂–CQH₂ complex embedded in the final 2F _o − DF _c electron density (cutoff level 1σ) with different colours for the inhibitor and the protein part and viewed along the twofold NCS dyad; the affiliations of the amino acid residues to HLE protomers A or B are indicated, as are hydrogen bonds between the CQH molecules and the protein (black dotted lines). (d), (e) and (f) were prepared with PyMOL v.1.7 (Schrödinger).

The final structure was analysed with PISA (Krissinel & Henrick, 2007 ▸) in order to explain this NCS pattern. The program predicts two potentially stable complexes, namely a dimer of the two HLE chains A and B in the asymmetric unit (Fig. 2 ▸ d) and a tetramer established from this dimer by a crystallographic twofold operation. The orientations of the point symmetry axes of these complexes within the crystal packing are in fact fully consistent with the NCS patterns of Figs. 2 ▸(b) and 2 ▸(c): the twofold axis visible in Fig. 2 ▸(d) coincides with the NCS peak in the b* direction of Fig. 2 ▸(b); combined with a crystallographic dyad in the a direction, a 222 system arises with the conspicuous 180° NCS peak in the c direction (Fig. 2 ▸ b).

It is noteworthy that at 331 Ų the size of the interface between the two HLE chains within the central AB dimer is far too small for a stable protein–protein interaction (Jones & Thornton, 1996 ▸). Rather, as revealed by a comparative PISA calculation, dimerization requires the presence of the two symmetrically organized CQH ligands in between (magenta-coloured spheres in Fig. 2 ▸ d). With the exception of C atoms 5, 8, 9 and 10, and the ring A atoms S1 and N3 (Fig. 1 ▸ a), all atoms of each CQH molecule are involved in attractive interactions, either with one of the two HLE subunits (total contact surfaces of 230 and 295 Ų, respective­ly) or with the other CQH molecule (interface size of 47 Ų). With their terminal nonpolar 2-propylphenyl groups, assisted by hydrogen bonds formed by their two peptide groups to the backbone peptides of each HLE protomer, the two CQH molecules integrate several side chains from both HLE subunits into a hydrophobic core (Fig. 2 ▸ f) and in this way stabilize the whole HLE₂–CQH₂ assembly.

Further work is needed to clarify experimentally whether this HLE₂–CQH₂ complex exists in solution and whether it is functionally significant. Yet, some preliminary notions about function are already possible: in a seminal review, Hajjar et al. (2010 ▸) compared the available structures of HLE with that of the closely related serine protease hPR3 and noticed interesting differences with respect to their quaternary structures. HLE is a monomer in its crystal structures, with the exception of PDB entry 1h1b (Macdonald et al., 2002 ▸), where HLE dimerizes within the crystal via an extension of a β-sheet, an assembly that a PISA analysis (Krissinel & Henrick, 2007 ▸) does not report to be stable in solution. In contrast, hPR3 exists as a tetramer in the form of a dimer of dimers in its only reported crystal structure (PDB entry 1fuj; Fujinaga et al., 1996 ▸). The two active site surfaces within the basic hPR3 dimer are packed against each other (Fig. 2 ▸ e). This dimeric architecture is incompatible with the binding of substrate proteins (Hajjar et al., 2010 ▸) and thus represents an inactive state. Remarkably, the novel CQH-mediated HLE dimers (Fig. 2 ▸ d) are equivalent to the inactive hPR3 dimers (Fig. 2 ▸ e) identified by Hajjar et al. (2010 ▸). This suggests a potential mechanism of HLE inactivation by CQH that may supplement its inhibitory binding mode described below.

3.2. Tertiary structure and inhibitor binding  

The three-dimensional structures of the two HLE protomers are nearly identical (with an r.m.s.d. on main-chain atoms of 0.19 Å and an r.m.s.d. on all atoms of 0.46 Å) and are very similar to previously published HLE structures. Like all chymotrypsin-type serine proteases, HLE is mainly composed of two six-stranded, antiparallel and twisted β-barrels (Fig. 3 ▸ a). The oxyanion hole required to stabilize the transition state of the reaction is defined by the peptide N atoms of Gly193 and Ser195 in the C-terminal domain. Both domains contribute to the catalytic triad (His57, Asp102 and Ser195) and to an array of subsites that are critical for substrate binding and recognition (Fig. 3 ▸ a). These subsites are structurally well characterized by three HLE structures in complex with proteinaceous inhibitors [PDB entries 1ppf (Bode et al., 1986 ▸), 2z7f (Koizumi et al., 2008 ▸) and 4nzl (Stapels et al., 2014 ▸)]. According to an established convention (Schechter & Berger, 1967 ▸) the specificity sites are designated S3, S2 and S1 at the N-terminal side of the substrate peptide bond to be hydrolysed and S1′, S2′ and S3′ at the C-terminal side (Fig. 3 ▸ a).

Figure 3.

Tertiary structure of HLE and substrate–inhibitor recognition. (a) Overview of an HLE protomer in complex with a CQH ligand (black C atoms), with different colours for the N- and C-terminal domains and with labels for the consensus β-strands; the side chains of the catalytic triad (Ser195/His57/Asp102) are shown as well as a section of the proteinaceous inhibitor from the HLE structure with PDB code 1ppf (Bode et al., 1986 ▸; green C atoms; main chain as ribbon representation) in order to indicate the substrate-recognition sites. (b) Occupation of the S2′ site and partial occupation of the S3′ site of HLE, as indicated by the Tyr20 and Arg21 side chains of the proteinaceous inhibitor in the HLE structure with PDB code 1ppf, by the central Leu side chain and the terminal propyl group of CQH, respectively. C–C contact distances below 3.8 Å between CQH rings B and C and its anchor regions on HLE are indicated by black dotted lines. For comparison, the equivalent zones for bovine trypsin (light blue C atoms; PDB entry 1zzz; Krishnan et al., 1998 ▸), human thrombin (magenta-coloured C atoms; PDB entry 1ppb; Bode et al., 1989 ▸) and cathepsin G (brown C atoms; PDB entry 1cgh; Hof et al., 1996 ▸) are drawn after superimposition of the protein matrices. Both parts of the figure were prepared with PyMOL v.1.7 (Schrödinger).

The S1 pocket provides the most critical interactions with a substrate peptide, with a strong preference for small hydrophobic P1 residues in the substrate. The S1 pocket (plus typically S2 and often further nonprimed pockets) is occupied by a small-molecule inhibitor in all corresponding HLE structures. HLE inhibitors extending beyond the S1 pocket to occupy the S1′ and S2′ sites have been described (Sisay et al., 2009 ▸), with a structure of such a type of inhibitor with porcine pancreatic elastase being available (Matern et al., 2003 ▸). However, none of the published HLE structures contains a synthetic HLE inhibitor bound to the S2′ and S3′ sites, and in the case of the S1′ site there is only one structure (PDB entry 1b0f; Cregge et al., 1998 ▸) with an inhibitor that protrudes into the S1′ site to some extent.

CQH is novel in that it does not bind in the region of the S1 pocket and the active site serine; rather, it bypasses the cleft between the two main domains in such a manner (Fig. 3 ▸ a) that its internal leucine side chain is well positioned to block the S2′ site (Figs. 3 ▸ a and 3 ▸ b). In addition, its terminal propyl chain extends into the S3′ site region (Figs. 3 ▸ a and 3 ▸ b). This binding mode requires that the two phenyl rings B and C are well accommodated at nonpolar anchor points. In the case of ring C of CQH this is Gly39, with which CGH is packed such that the C^α atom lies centrally over the ring plane with distances of between 3.6 and 3.8 Å to all ring C atoms (shown as dotted lines in Fig. 3 ▸ b). This arrangement is spatially challenging and is hardly compatible with larger side chains at the equivalent position as present in thrombin (Glu39) or bovine trypsin (Tyr39) or with a significantly longer β1–β2 loop as found in cathepsin G with three more amino acids than in HLE (Fig. 3 ▸ b).

The binding region for CQH ring B at the C-terminal domain is also spatially demanding, albeit not quite as strongly as the ring C binding patch: here the phenyl ring is packed against Ile151 with a nearest C–C distance of 3.8 Å (Fig. 3 ▸ b). The equivalent residues are Gln151 in thrombin, Tyr151 in bovine trypsin and Gly151 in cathepsin G, which are either too bulky or too small to serve as a good anchor point for CQH. These structural observations are remarkably compatible with our preliminary data on the inhibitory selectivity profile of CQH.

In summary, CQH has a presumed capability to promote the assembly of inactive HLE dimers, while also being directly substrate-competitive, but in a mode that differs from other HLE inhibitors. Furthermore, the HLE–CQH complex structure provides evidence of a complicated type of inhibition, which is supported by the Hill coefficient of 1.53 (Fig. 1 ▸ c). To underpin these indications and to elucidate this inhibitory mechanism more precisely, a detailed kinetic analysis is required.

3.3. Glycosylation  

Both of the HLE chains in the asymmetric unit are N-glycosylated at Asn109 and Asn159 (Figs. 4 ▸ a and 4 ▸ b). A total of nine sugar moieties in HLE monomer A and ten in monomer B were modelled and refined with preferred conformations and with partly low, but overall acceptable, real-space correlation coefficients according to the quality criteria of the carbohydrate-validation program Privateer (Agirre, Iglesias-Fernández et al., 2015 ▸; Supplementary Fig. S1). For illustration, the two N-glycans of HLE monomer B, which particularly benefit from the proximity of crystallo­graphic symmetry-equivalent HLE chains, are displayed with the corresponding electron-density sections (Figs. 4 ▸ c and 4 ▸ d).

Figure 4.

N-glycosylation of HLE. (a) Structural overview of the four N-glycans attached to the HLE₂–CQH₂ complex. (b) Schematic illustration of the HLE glycosylation as drawn by Privateer (Agirre, Iglesias-Fernández et al., 2015 ▸). In the legend, the three-letter PDB notations for the carbohydrate entities are given. (c, d) The two N-glycans of HLE chain B with parts of the protein environment and a sulfate ion (c); the carbohydrate chains are embedded in the final electron density (cutoff level 0.8σ). (a), (c) and (d) were prepared with PyMOL v.1.7 (Schrödinger).

The resulting carbohydrate substructures (Fig. 4 ▸ b) contain the typical core of eukaryotic N-glycans with two N-acetyl­glucosamine residues followed by a β-linked mannose unit that serves as a branching point (Stanley et al., 2017 ▸). Depending on the outer substitution of this core, three major types of eukaryotic N-glycans are distinguishable, which are referred to as ‘complex’, ‘hybrid’ and ‘oligomannose’. In the N-glycans of HLE an α-l-fucose moiety is attached to the first N-acetylglucosamine (Figs. 4 ▸ c and 4 ▸ d), which is characteristic of the ‘complex’ type of N-glycans (Stanley et al., 2017 ▸). These fucose residues are relatively well ordered owing to stabilizing hydrogen bonds and hydrophobic interactions with the nearby protein matrix (Figs. 4 ▸ c and 4 ▸ d). It is noteworthy that the glycan chains in Fig. 4 ▸(b) are likewise consistent with a fourth and more recently established major type of eukaryotic N-glycans called ‘paucimannose’ (with the stoichiometry mannose_1–3fucose_0–1 N-acetyl­glucosamine₂Asn), the biosynthesis of which requires the activity of specific glycosidases in order to trim larger precursors (Thaysen-Andersen et al., 2015 ▸). The existence of paucimannosylated proteins in mammals remained controversial for a long time (Schachter & Boulianne, 2011 ▸), but has been proven by recent results (Thaysen-Andersen et al., 2015 ▸); they have important roles in the innate immune system (Loke et al., 2016 ▸), with HLE being one of them (Loke et al., 2017 ▸).

No N-glycosylation at Asn72 is apparent in the HLE–CQH complex structure. Like Asn109 and Asn159, Asn72 has the sequence environment N-X-S/T (X ≠ P) required for the attachment of an N-glycan (Stanley et al., 2017 ▸). HLE glycosylation at Asn72 was in fact observed in a glycoproteomics study of liver tissue (Chen et al., 2009 ▸). Even so, the absence of Asn72 glycosylation here is unsurprising since according to a recent comprehensive analysis (Loke et al., 2017 ▸) HLE glycosylation is subject to high macro-heterogeneity, with Asn72 only being glycosylated to 4%. Furthermore, the most frequent HLE isoform in granules of human leukocytes that prevails in commercially available batches and as has been established by HLE crystallography since the first structure determination (Bode et al., 1986 ▸) was found to be glycosylated only at Asn109 and Asn159 (Sinha et al., 1987 ▸).

The observed structures of N-glycans (Figs. 4 ▸ a and 4 ▸ b) are amongst the largest to be reported per HLE protomer, being comparable with those of the first HLE structure (PDB entry 1ppf; Bode et al., 1986 ▸) and a follow-up structure (PDB entry 1ppg) from the same research group (Wei et al., 1988 ▸). In these early HLE structures the carbohydrate ‘trees’ are even longer than here; however, all sugar moieties beyond the first mannose in PDB entry 1ppf and beyond the second N-acetylglucosamine in PDB entry 1ppg have atomic occupancy factors of zero, indicating their disorder and a lack of underlying electron density. Unfortunately, a more detailed assessment of the carbohydrate chains in PDB entries 1ppg and 1ppf is not possible since no experimental structure factors were deposited together with the structures, so that no electron density maps are available. Most of the subsequent HLE structures contain only one or two N-acetylglucosamine residues per glycosylation site plus, in most cases, an associated α-l-fucose. However, a detailed inspection revealed that in several HLE structures the N-glycans can be supplemented or contain anomalies (Agirre, Davies et al., 2015 ▸) that can be relieved. Supplementary Fig. S2 documents some such cases.

In summary, the status of the N-glycans in current HLE structures is restricted with respect to quality and quantity. This reflects the objective difficulty in modelling carbohydrate chains into electron-density maps that are of poor quality owing to the significant conformational flexibility of sugars and the existing microheterogeneity of the N-glycans in HLE (Loke et al., 2017 ▸), in addition to the limited interest in N-glycosylation as a phenomenon. This indifference is not surprising given the fact that the N-glycan chains are distal to the primary functional sites of HLE which are responsible for catalysis and inhibition by small molecules such as CQH (Fig. 4 ▸ a).

However, recent experimental evidence (Loke et al., 2017 ▸) has demonstrated that the outer mannose residues of the N-glycans of HLE, as visible here (Figs. 4 ▸ c and 4 ▸ d), are important for the interaction of the enzyme with so-called ‘mannose-recognizing C-type lectin receptors’. These are typically membrane proteins and are critical components of the innate immune system (Loke et al., 2016 ▸). A soluble representative is the ‘mannose-binding lectin’ (MBL; Ip et al., 2009 ▸), a serum protein that is primarily synthesized in the liver and is overexpressed after infection. Loke et al. (2017 ▸) found that in spite of its primary role in recognizing carbohydrate structures on the cell surfaces of infectious bacteria, MBL interacts with HLE in vitro and that HLE glycoforms rich in paucimannose-type N-glycans as found in the HLE–CQH complex structure (Figs. 4 ▸ a and 4 ▸ b) are preferred in this context.

Furthermore, Loke et al. (2017 ▸) revealed that glycosylation plays a role in the physiologically most important inhibitory interaction of HLE: that with α₁-antitrypsin. This study determined the glycosylation state of a covalent inhibitory complex of the two proteins in comparison to the two unbound partners. The basic finding was that the interaction, either the affinity or the assembly process, is dependent on the glycosylation states of both partners. In contrast to HLE, α₁-antitrypsin mainly contains N-glycans of the complex type with negatively charged N-acetylneuraminic acid residues (or its derivatives, summarized as sialic acid) at their termini. In the HLE–α₁-antitrypsin complex, however, a shift in the reverse direction was observed by Loke et al. (2017 ▸): here, a HLE glycoform containing partially complex-type N-glycans was enriched, whereas on the α₁-antitrypsin side an undersialylated glycoform was preferred. The mechanistic basis and the functional meaning of these findings are unknown. Based on a trypsin–α₁-antitrypsin complex structure (Huntington et al., 2000 ▸) it can be concluded that the N-glycan chains of HLE on the one side and those of α₁-antitrypsin on the other side are too distant for a direct interaction. However, the formation of an α₁-antitrypsin complex with a protease is a complex process (Huntington et al., 2000 ▸) that includes large conformational changes and may include glycan-dependent steps. The notions of Loke et al. (2017 ▸) point in this direction since for both partners they assume that the detected alterations in N-glycans facilitate the interaction process owing to reduced steric problems. In particular, in the case of HLE they speculate that the negatively charged sialo form of an N-glycan, which is enriched, may interact with its own protein surface, which then causes less steric hindrance for the interaction with α₁-antitrypsin. The sulfate ion seen in Fig. 4 ▸(c) and coordinated by HLE in the proximity of an N-glycan is compatible with this idea.

Taken together, the N-glycans of HLE have a subtle impact on its protein–protein interactions with MBL and with α₁-antitrypsin and thus on its functionality (Loke et al., 2017 ▸). It is possible that these and further examples will direct attention towards N-glycosylation of the enzyme. In this case, the structure that we present here might become a particularly valuable addition to structural knowledge of HLE.

4. Related literature  

The following reference is cited in the Supporting Information for this article: Lechtenberg et al. (2015 ▸).

Supplementary Material

PDB reference: human leukocyte elastase, 6f5m

Funding Statement

This work was funded by Deutsche Forschungsgemeinschaft grants NI 643/4-1 and NI 643/4-2.