Laskowski inhibitors regulate serine proteases by an intriguing mode of action that involves deceiving the protease into synthesizing a peptide bond. Studies exploring naturally occurring Laskowski inhibitors have uncovered several structural features that convey the inhibitor's resistance to hydrolysis and exceptional binding affinity. However, in the context of Laskowski inhibitor engineering, the way that various modifications intended to fine-tune an inhibitor's potency and selectivity impact on its association and dissociation rates remains unclear. This information is important as Laskowski inhibitors are becoming increasingly used as design templates to develop new protease inhibitors for pharmaceutical applications. In this study, we used the cyclic peptide, sunflower trypsin inhibitor-1 (SFTI-1), as a model system to explore how the inhibitor's sequence and structure relate to its binding kinetics and function. Using enzyme assays, MD simulations and NMR spectroscopy to study SFTI variants with diverse sequence and backbone modifications, we show that the geometry of the binding loop mainly influences the inhibitor's potency by modulating the association rate, such that variants lacking a favourable conformation show dramatic losses in activity. Additionally, we show that the inhibitor's sequence (including both the binding loop and its scaffolding) influences its potency and selectivity by modulating both the association and the dissociation rates. These findings provide new insights into protease inhibitor function and design that we apply by engineering novel inhibitors for classical serine proteases, trypsin and chymotrypsin and two kallikrein-related peptidases (KLK5 and KLK14) that are implicated in various cancers and skin diseases.

INTRODUCTION

Laskowski (or standard mechanism) inhibitors [1,2] regulate key events in development, homoeostasis and death by reversibly inhibiting serine proteases. On the surface, Laskowski inhibitors appear very similar to ideal protease substrates, which allows them to bind to their target protease(s) at exceptionally fast rates and leads to efficient cleavage of the inhibitor's reactive site (scissile) bond [3,4]. However, unlike ordinary substrates, Laskowski inhibitors have the capacity to reverse this cleavage event by hijacking the protease's catalytic machinery to resynthesize the reactive site bond [5,6]. This intriguing biochemical anomaly allows the inhibitor to remain bound within the protease's active site cleft without permanently disabling its catalytic activity and occurs in at least 19 families of inhibitors that are found in all forms of life [1,7]. Remarkably, despite sharing a common mode of action, the various Laskowski inhibitor families display essentially no sequence or fold conservation. Thus, Laskowski inhibitors are regarded as one of the largest known superfamilies of convergently evolved proteins [1,8].

Although each inhibitor family is structurally unique, two features have emerged that are common to all Laskowski inhibitors. First, the inhibitor's reactive site bond is located in an exposed binding loop (the canonical loop) that shows strictly conserved backbone geometry [9,10] and often displays a cleavage sequence that is favoured by a specific set of proteases. Both the sequence and the conformation of the canonical loop have been linked to the inhibitor's substrate-like association kinetics, although the relative contribution of each factor remains unclear. Second, the canonical loop is often braced by an integrated network of hydrogen and disulfide bonds. This feature has been shown to be essential in many inhibitor families [1117] and is thought to stabilize the protease–inhibitor complex once the reactive site bond is cleaved (acyl–enzyme complex), thus keeping the neo-N-terminus in place to promote efficient religation [5,18].

Understanding how the key features of Laskowski inhibitors contribute to their ability to inhibit different proteases is particularly important when adapting naturally occurring Laskowski inhibitors to new protease targets. Indeed, Laskowski inhibitors are receiving growing interest from the pharmaceutical industry and have emerged as promising engineering templates for a range of therapeutic applications, as seen by the success of Kalbitor®, an engineered Kunitz domain inhibitor that is now used in the clinic for treating hereditary angioedema [19]. During the inhibitor development phase, the most common approach for evaluating the effect of various substitution(s) on the inhibitor's activity is to determine the inhibition constant (Ki). This method provides insight into the binding affinity of the new inhibitor variant, but does not directly reveal how fast the inhibitor binds to the protease or how slowly it is released. Accordingly, the way that various modifications alter the inhibitor's association rate and dissociation rate (and thus contribute to its potency and selectivity) remains unclear. This information would better inform future engineering studies and contribute to developing systematic approaches for adapting Laskowski inhibitors to new protease targets.

In the present study, we used the 14 amino acid cyclic peptide, sunflower trypsin inhibitor-1 (SFTI-1) [20], as a model system for exploring how changing the inhibitor's sequence and structure impacts on its association and dissociation rate. The activity of SFTI-1 derives from (at least) two key features: the P4 to P2′ sequence that binds within the protease active site (Figure 1) and a series of intramolecular interactions, including a disulfide bond and a network of hydrogen bonds. To examine the role of each structural feature, we used two contrasting strategies for engineering SFTI-1 to generate novel inhibitor variants. In the first approach, we prioritized binding complementarity between the target protease and inhibitor over the intramolecular hydrogen bond network. This was achieved by substituting a favoured binding sequence for the new protease into the P4-P1 residues of the canonical loop, then performing further substitutions to restore intramolecular hydrogen bonds. Conversely, the second approach aimed to maximize the number of intramolecular hydrogen bonds (at the expense of side chain interactions with the protease) by re-configuring the inhibitor's hydrogen bond network to involve residues in the contact β-strand.

Structural features of trypsin-like serine proteases and SFTI-1

Figure 1
Structural features of trypsin-like serine proteases and SFTI-1

(A) Structure of trypsin (PDB ID: 1SFI) shown as a ribbon diagram overlayed by the enzyme's molecular surface (transparent). The serine protease catalytic triad (His57, Asp102 and Ser195) is shown in stick representation. (B) Binding subsites (S4-S3-S2-S1-S1′-S2′) that engage the protease's substrates and inhibitors are mapped on to the surface of trypsin and are labelled according to Schechter and Berger nomenclature [53]. (C) Amino acid sequence of SFTI-1 in single letter code (P4-P2′ residues are labelled) and (D) structure of SFTI-1 in stick representation. The inhibitor backbone is cyclized between Gly1 and Asp14 and an intramolecular disulfide bond is present between Cys3 and Cys11 (canonical loop residues are coloured blue). (E) Ribbon diagram of SFTI-1 (same orientation as panel D) showing the contact β-strand and non-contact β-strand. The P1 lysine residue is shown in ball-and-stick model and canonical loop residues are coloured blue.

Figure 1
Structural features of trypsin-like serine proteases and SFTI-1

(A) Structure of trypsin (PDB ID: 1SFI) shown as a ribbon diagram overlayed by the enzyme's molecular surface (transparent). The serine protease catalytic triad (His57, Asp102 and Ser195) is shown in stick representation. (B) Binding subsites (S4-S3-S2-S1-S1′-S2′) that engage the protease's substrates and inhibitors are mapped on to the surface of trypsin and are labelled according to Schechter and Berger nomenclature [53]. (C) Amino acid sequence of SFTI-1 in single letter code (P4-P2′ residues are labelled) and (D) structure of SFTI-1 in stick representation. The inhibitor backbone is cyclized between Gly1 and Asp14 and an intramolecular disulfide bond is present between Cys3 and Cys11 (canonical loop residues are coloured blue). (E) Ribbon diagram of SFTI-1 (same orientation as panel D) showing the contact β-strand and non-contact β-strand. The P1 lysine residue is shown in ball-and-stick model and canonical loop residues are coloured blue.

To evaluate inhibitors produced by each engineering strategy, we studied diverse protease targets, including the classical serine proteases, trypsin and chymotrypsin and two closely-related members of the kallikrein serine protease family, kallikrein-related peptidase (KLK)5 and KLK14 that are promising therapeutic targets in several diseases [21,22]. For example, KLK5 and KLK14 are implicated in skin diseases, including atopic dermatitis and Netherton syndrome, where they contribute to impairing the epidermal barrier and activate an array of inflammatory signalling pathways [2327]. Additionally, KLK5 contributes to tumour progression and chemoresistance in ovarian cancer [28,29], whereas KLK14 has been shown to activate cell-surface receptors that modulate the proliferation of colon cancer cells [30]. Engineered SFTI variants were subsequently screened in competitive inhibition assays to examine the effect of different substitutions on the inhibitor's binding affinity as well as its association rate and dissociation rate. A subset of inhibitors was also analysed by MD simulations or NMR spectroscopy to provide complementary structural insights.

MATERIALS AND METHODS

Reagents

Analytical grade solvents for peptide synthesis were obtained from Merck unless stated. 9-fluorenylmethyl carbamate (Fmoc) protected amino acids and reagents for amino acid coupling were obtained from Iris Biotech unless stated. Analytical grade solvents for reverse phase HPLC were obtained from Merck and trifluoroacetic acid (TFA) was obtained from Auspep. Proteases not expressed in-house were sourced from Sigma–Aldrich (trypsin, plasmin), Commonwealth Serum Laboratories (thrombin), R&D Systems (matriptase, thermolysin), Invitrogen (enterokinase) or Boehringer Ingelheim (chymotrypsin).

Recombinant protein expression

Recombinant KLK5 and KLK7 were expressed in zymogen (pro-) form in Pichia pastoris strain X-33, as described previously [26,31]. Recombinant proteins were purified from the culture supernatant by cation exchange chromatography and fractions containing pro-KLK5 or pro-KLK7 were identified by SDS/PAGE (gels were post-stained with Coomassie Brilliant Blue) and pooled. Purified pro-KLKs were activated using enterokinase (EK Max) at 37°C for 2 h (1 unit EK Max per 50 μg of pro-KLK). Enterokinase was subsequently separated from active KLKs following a second round of cation exchange chromatography. Recombinant KLK4 and KLK14 were expressed in zymogen form in stably transfected Sf9 insect cells (Spodoptera frugiperda), as previously described [32,33]. Prior to purification, clarified culture supernatant was dialysed against 50 mM Na2HPO4, pH 8.0, containing 0.5 M NaCl for 72 h at 4°C. Subsequently, recombinant pro-KLK4 and pro-KLK14 were purified from conditioned culture supernatant by nickel agarose affinity chromatography, using Ni-NTA superflow resin (Qiagen). Fractions containing pro-KLK4 or pro-KLK14 were identified by SDS/PAGE (gels were post-stained with Coomassie Brilliant Blue) and pooled. Purified pro-KLKs were activated using thermolysin at 37°C which was inhibited with 25 mM EDTA after optimal KLK activity had been achieved. For KLK4, it was also possible to separate active KLK4 from thermolysin by anion exchange chromatography. Concentrations of active KLKs were determined by active site titration, using either α2-antiplasmin (Sino Biologicals; KLK5), α1-antitrypsin (Sigma–Aldrich; KLK7) or 4-methylumbelliferyl-p-guanidinobenzoate (MUGB, Sigma–Aldrich; KLK4 and KLK14). Glycerol was added to active KLKs to a final concentration of 20% (v/v) allowing active proteases to be stored at −80°C with minimal losses of activity.

Solid phase synthesis of SFTI variants and peptide substrates

SFTI variants were synthesized on 2-chlorotrityl resin (1.55 mmol/g, Iris Biotech) using a Discover SPS Microwave System (CEM Corporation), see Supplementary Methods for full details. Peptide para-nitroanilide (pNA) substrates were synthesized on para-phenylenediamine derivatized 2-chlorotrityl resin, as previously described [33,34]. Following synthesis, inhibitors and substrates were purified by reverse phase HPLC using a Jupiter Proteo 90 Å (1 Å=0.1 nm) C-18 column (Phenomenex). Fractions were analysed by MALDI–TOF/MS using a Biorad ProteinChip System and high accuracy mass analysis was performed using a Q-TOF LC–MS system (Agilent Technologies), see Supplementary Table S1 and Supplementary Figure S1.

Inhibition assays

SFTI-1 and engineered SFTI variants were evaluated by determining the inhibition constant (equilibrium dissociation constant, Ki) in competitive inhibition assays. Proteases were incubated with serial dilutions of inhibitor (eight concentrations per variant) in 96-well low-binding assay plates (Corning) and allowed to reach equilibrium prior to addition of a constant amount of peptide-pNA substrate. All assays were performed at 298 K (see Supplementary Methods for full assay conditions). The degree of inhibition at each concentration of inhibitor was determined by analysing kinetic rates describing cleavage of the peptide-pNA substrate compared with controls without inhibitor (measured by the change in absorbance, λ=405 nm) using a Biorad Benchmark Plus microplate spectrophotometer (reading interval: 10 s, assay time course: 300 s). Assays were performed three times in triplicate and where the IC50 was less than 5000 nM, the inhibition constant was determined in GraphPad Prism 5 by non-linear regression using the Morrison equation for tight binding inhibitors.

Inhibitor binding kinetics

To assess inhibitor binding kinetics, the dissociation rate constant (koff) was determined using the method derived by Baici and Gyger-Marazzi [35]. This assay examines the inhibitor's ability to block proteolytic activity under two conditions: pre-incubating the protease with inhibitor (before adding substrate) and adding inhibitor simultaneously with substrate. Assays were performed three times in triplicate using a constant concentration of protease and 150 or 200 μM peptide–pNA substrate in 96-well plates. Substrate cleavage was monitored using a Biorad xMark microplate reader (λ=405 nm, interval: 10 s) and proceeded until steady state inhibition was attained (where assays exceeded 15 min, a reading interval of 20 or 30 s was used). Absorbance readings at each time-point were exported to Microsoft Excel and reaction curves were generated by plotting substrate cleavage over time for each condition (simultaneous addition of inhibitor and substrate or pre-incubating protease with inhibitor; see Supplementary Figure S2). These plots were used to calculate koff using the formula koff=(| τ | + | τ* |)−1 where τ and τ* represent the x-intercept of the steady state (linear) reaction curve from the simultaneous addition and pre-incubation assays, respectively. Association rate constants (kon) were subsequently derived from known Ki and koff values using the formula kon=koff ÷ Ki.

MD simulations of protease–inhibitor complexes

MD simulations were performed to allow virtual screening of inhibitor variants in complex with different proteases. SFTI-1 was positioned in the active site of KLK5 by overlaying the KLK5 crystal structure (PDB ID 2PSX) [36] with the crystal structure of the trypsin–SFTI-1 complex (PDB ID 1SFI) [20] in DeepView v4.10 [37]. As a crystal structure for KLK14 is not currently available, a homology model was generated using SWISS-MODEL [37] with KLK5 as the template (PDB ID 2PSX, Cα RMSD 0.39 Å), see Supplementary Methods for full details. KLK14–SFTI complexes were subsequently generated by overlaying the KLK14 model with the trypsin–SFTI-1 structure (as above). Amino acid substitutions to generate inhibitor variants were performed in YASARA Dynamics [38]. Complexes were solvated with TIP3P water and neutralized by adding Na+ and Cl counter-ions using VMD 1.9.1 [39]. Prior to simulation with full MD, each protease–inhibitor complex was equilibrated using a stepwise relaxation procedure (see Supplementary Methods). Subsequently, 10 independent 1 ns production runs were performed using NAMD 2.9 [40]. Coordinates were saved every 100 simulation steps producing 5000 frames per trajectory. Analyses were performed using VMD 1.9.1 [39] with hydrogen bond lengths and angles set to 3.3 Å and 40° respectively [11]. Trajectories yielding Cα RMSD values more than two S.D.s outside the mean (averaged across 10 runs) at any residue were replaced to reduce impacts of random seeding (exclusion rate: 18.9%). All structural figures were produced using CCP4MG [41].

NMR spectroscopy

Samples for NMR spectroscopy were prepared by dissolving ∼3 mg of variant 15 in 500 μl of 10% 2H2O/90% H2O (v/v) at pH 3.7 or ∼3 mg of variant 15NMe in 500 μl of DMSO-d6. 2H2O and DMSO-d6 were both obtained from Cambridge Isotope Laboratories. NMR experiments including 1H, TOCSY, NOESY and double quantum filtered COSY (DQF-COSY) were recorded at 298 K using a Bruker Avance 600 MHz NMR spectrometer. Chemical shifts were referenced to internal 2,2-dimethyl-2-silapentane-5-sulfonate (DSS) at 0 ppm. Mixing times of 80 and 200 ms were used for the TOCSY and NOESY spectra respectively. Processed spectra were analysed and assigned using the program CCPNMR with the sequential assignment protocol [42].

NMR structure calculations

Distance and dihedral angle restraints were derived from the NOESY spectrum using peak volumes and 3JHN-Hα coupling constants from 1D 1H NMR spectrum respectively. The ϕ angles were constrained to −120±40° for 3JHN-Hα > 8.0 Hz. Initial structures were generated using CYANA [43] based on torsion angle dynamics and refined within CNS [44] using protocols from the RECOORD database [45] as described previously [46]. The 20 potential solutions with the lowest energy were selected to represent the NMR structure of variants 15 and 15NMe (Supplementary Table S3 and S4 for NMR statistics and 1H chemical shifts). Coordinates for each set of NMR structures have been deposited into the Biological Magnetic Resonance Data Bank (BMRB) via the Protein Data Bank Japan (PDBj) (accession codes: 21056 and 21057).

RESULTS AND DISCUSSION

Engineering the canonical loop to display a favoured binding sequence improves the inhibitor's association rate

The first strategy for engineering SFTI-based inhibitors for KLK5 and KLK14 was to substitute preferred binding sequences for each protease into the contact β-strand of SFTI-1 (compound 1, Table 1). This involved substituting the P4, P2 and P1 residues (Arg2, Thr4 and Lys5 in SFTI-1; Figure 1), but not the P3 residue (Cys3) as it forms a critical intramolecular disulfide bond [13]. For KLK14 inhibitors, substitutions were based on the sequence of favoured tetrapeptide substrates (YASR, YAVR, WAVR) that we previously identified in a non-combinatorial peptide library screen [34]. An additional substitution was made at residue 14 as KLK14 favours asparagine over aspartic acid at this position [11]. These modifications yielded three KLK14 inhibitors (2–4), with 2 being the most potent (Ki=2.2 nM; Table 1) and 4 being the most selective (Supplementary Figure S3). To engineer inhibitors for KLK5, SFTI β-strand substitutions were based on an optimized P4-P1 sequence (FCHR) from a larger, naturally occurring KLK5 inhibitor, SPINK9 [47] and then Asp14 was replaced with asparagine. This variant (5) showed a 2.2-fold improvement in activity against KLK5 compared with 1 (Table 1).

Table 1
Inhibition (Ki) and rate constants (kon, koff) for SFTI-1 and engineered SFTI variants
Variant Sequence1 Protease Ki (nM) koff (× 10−3 s−1Fold kon (× 106 M−1 s−1Fold 
GRCTKSIPPICFPD Trypsin 0.02±0.002 0.05±0.01 – 2.0±0.3 − 
  KLK5 143±3.7 12±0.3 – 0.09±0.02 − 
  KLK14 25.1±1.3 5.5±0.7 – 0.2±0.04 − 
1NMe GRCTKSNMeIPPICFPD Trypsin 22.3±0.9 0.5±0.07 −102 0.02±0.003 −1002 
GYCSRSIPPICFPN KLK14 2.2±0.2 7.2±0.7 – 3.3±0.3 − 
GYCVRSIPPICFPN KLK14 14.3±0.6 – – – − 
GWCVRSIPPICFPN KLK14 28.4±0.7 43±3.2 – 1.5±0.1 − 
GFCHRSIPPICFPN KLK5 58.7±1.7 115±24 – 2.0±0.4 − 
GWCVRSIPPICEPN KLK14 26.8±0.6 – – – − 
GWCVRSIPPICQPN KLK14 13.5±0.5 17±3.6 2.53 1.3±0.3 −1.23 
GWCVRSIPPICTPN KLK14 5.2±0.2 13±2.4 3.33 2.5±0.5 1.73 
GWCVRSIPPICDPN KLK14 3.5±0.1 12±2.5 3.63 3.3±0.7 2.23 
10 GWCVRSIPPICNPN KLK14 2.2±0.1 6.4±0.4 6.73 2.9±0.2 1.93 
11 GWCIRSIPPICNPN KLK14 2.0±0.1 6.2±0.3 6.93 3.1±0.1 2.13 
12 GFCHRSIPPICRPN KLK5 215±6.4 – – – − 
13 GFCHRSIPPICWPN KLK5 31.5±0.9 37±2.7 3.14 1.2±0.1 −1.74 
14 GFCHRSYPPICWPN KLK5 6.2±0.2 14±2.1 8.24 2.2±0.3 1.14 
15 GTCTRSIPPICNPN Trypsin 0.7±0.07 3.9±1.1 −782 5.9±1.6 3.02 
  KLK5 2.0±0.1 3.8±0.2 304 1.9±0.1 < 1.14 
  KLK14 0.4±0.02 1.0±0.2 433 2.6±0.4 1.73 
15CHO SIPPICNPNGTCTRCHO Trypsin 17.4±0.4 1.8±0.2 2.25 0.1±0.01 −595 
  KLK14 44.8±2.5 3.1±0.04 −3.16 0.07±0.001 −376 
15NMe GTCTRSNMeIPPICNPN Trypsin 114±5.1 3.8±0.3 < 1.15 0.03±0.003 −1975 
  KLK14 41.7±2.0 0.8±0.05 1.36 0.02±0.001 −1306 
16 GTCTFSIPPICNPN Chymotrypsin 0.5±0.03 3.4±0.3 – 6.8±0.7 − 
  KLK7 8.0±0.2 42±2.9 – 5.3±0.4 − 
Variant Sequence1 Protease Ki (nM) koff (× 10−3 s−1Fold kon (× 106 M−1 s−1Fold 
GRCTKSIPPICFPD Trypsin 0.02±0.002 0.05±0.01 – 2.0±0.3 − 
  KLK5 143±3.7 12±0.3 – 0.09±0.02 − 
  KLK14 25.1±1.3 5.5±0.7 – 0.2±0.04 − 
1NMe GRCTKSNMeIPPICFPD Trypsin 22.3±0.9 0.5±0.07 −102 0.02±0.003 −1002 
GYCSRSIPPICFPN KLK14 2.2±0.2 7.2±0.7 – 3.3±0.3 − 
GYCVRSIPPICFPN KLK14 14.3±0.6 – – – − 
GWCVRSIPPICFPN KLK14 28.4±0.7 43±3.2 – 1.5±0.1 − 
GFCHRSIPPICFPN KLK5 58.7±1.7 115±24 – 2.0±0.4 − 
GWCVRSIPPICEPN KLK14 26.8±0.6 – – – − 
GWCVRSIPPICQPN KLK14 13.5±0.5 17±3.6 2.53 1.3±0.3 −1.23 
GWCVRSIPPICTPN KLK14 5.2±0.2 13±2.4 3.33 2.5±0.5 1.73 
GWCVRSIPPICDPN KLK14 3.5±0.1 12±2.5 3.63 3.3±0.7 2.23 
10 GWCVRSIPPICNPN KLK14 2.2±0.1 6.4±0.4 6.73 2.9±0.2 1.93 
11 GWCIRSIPPICNPN KLK14 2.0±0.1 6.2±0.3 6.93 3.1±0.1 2.13 
12 GFCHRSIPPICRPN KLK5 215±6.4 – – – − 
13 GFCHRSIPPICWPN KLK5 31.5±0.9 37±2.7 3.14 1.2±0.1 −1.74 
14 GFCHRSYPPICWPN KLK5 6.2±0.2 14±2.1 8.24 2.2±0.3 1.14 
15 GTCTRSIPPICNPN Trypsin 0.7±0.07 3.9±1.1 −782 5.9±1.6 3.02 
  KLK5 2.0±0.1 3.8±0.2 304 1.9±0.1 < 1.14 
  KLK14 0.4±0.02 1.0±0.2 433 2.6±0.4 1.73 
15CHO SIPPICNPNGTCTRCHO Trypsin 17.4±0.4 1.8±0.2 2.25 0.1±0.01 −595 
  KLK14 44.8±2.5 3.1±0.04 −3.16 0.07±0.001 −376 
15NMe GTCTRSNMeIPPICNPN Trypsin 114±5.1 3.8±0.3 < 1.15 0.03±0.003 −1975 
  KLK14 41.7±2.0 0.8±0.05 1.36 0.02±0.001 −1306 
16 GTCTFSIPPICNPN Chymotrypsin 0.5±0.03 3.4±0.3 – 6.8±0.7 − 
  KLK7 8.0±0.2 42±2.9 – 5.3±0.4 − 

*(P4-P1 sequence underlined)

†relative to variant 1 (SFTI-1)

‡relative to variant 4

§relative to variant 5

∥relative to variant 15 (trypsin values)

¶relative to variant 15 (KLK14 values)

The inhibition constant (Ki) is the most commonly used parameter for describing the activity of an inhibitor against a given protease and expresses the inhibitor's binding affinity. However, Ki does not provide direct insight into the rate at which the inhibitor engages or releases from the protease active site. Therefore, we also studied inhibitor association and dissociation rates. Determining the dissociation rate constant (koff) then calculating the association rate constant (kon) for 1 and trypsin revealed that the high affinity of SFTI-1 for trypsin (Ki=0.02 nM) derived from fast association (kon=2×106 M−1 s−1) and slow dissociation rates (koff=5×10−5 s−1). By contrast, the weaker affinity of 1 for KLK14 (Ki=25 nM) was due to 10-fold slower association and 110-fold faster dissociation rates (Table 1). Substituting the sequence of a favoured KLK14 substrate into the SFTI contact β-strand improved kon to approach that of 1 with trypsin, as seen with the most selective variant (4) where kon increased by 7.5-fold. However, this gain was offset by 7.8-fold decrease in koff. These effects were more pronounced with KLK5. Here, the weak affinity of 1 for KLK5 (Ki=143 nM) was due to 22-fold slower association and 240-fold faster dissociation rates compared to trypsin (Table 1). Again, kon was improved by optimizing the P4-P1 binding sequence (22-fold increase, variant 5), but these substitutions led to a 9.6-fold decrease in koff. Collectively, these data demonstrate the importance of studying the inhibitor's binding kinetics in addition to its binding affinity. Whereas variation in Ki was within 2-fold for 1 and 4 with KLK14 or 1 and 5 with KLK5, the way that each inhibitor interacted with its target protease was dramatically different. Indeed, these findings indicated that optimizing the sequence of the binding loop had generated variants with improved association rates, but at the cost of faster dissociation rates.

Optimizing the intramolecular hydrogen bond network contributes to slowing the dissociation rate

A distinctive feature of SFTI-1 is the coordinated network of hydrogen bonds present within the inhibitor [20]. In MD simulations of SFTI-1 in complex with trypsin, this amounts to 7.1±0.1 intramolecular hydrogen bonds per frame (averaged across the trajectory; Figure 2A), replicating the inhibitor's hydrogen bond network observed in the trypsin–SFTI-1 crystal structure [20]. Stabilizing intramolecular interactions (including hydrogen bonds) are regarded to be important for promoting efficient religation in Laskowski inhibitors, which in turn leads to slow rates of dissociation [1118]. This trait was evident in assays with trypsin and also with KLK14 and KLK5 as 1 displayed considerably slower dissociation rates than either engineered variant (4 or 5; Table 1). Equivalent simulations of KLK14 bound with 4 revealed that changing the sequence of the canonical loop had significantly diminished the intramolecular hydrogen bond network (2.8±0.3 intramolecular hydrogen bonds; Figure 2B). A similar effect was observed in simulations of KLK5 bound with 5 (3.6±0.2 intramolecular hydrogen bonds; Figure 2C). These findings indicated that restoring intramolecular hydrogen bonds in 4 and 5 could contribute to slowing the dissociation rate and lead to further improvements in Ki.

Intramolecular hydrogen bond analysis and inhibitory activity for SFTI-1 and engineered inhibitor variants

Figure 2
Intramolecular hydrogen bond analysis and inhibitory activity for SFTI-1 and engineered inhibitor variants

(AF) For each inhibitor, the inhibition constant (Ki) determined from three independent assays performed in triplicate is shown next to the variant name, along with the intramolecular hydrogen bond frequency (AvH) as the mean ± S.D. from 10 independent 1 ns trajectories. Average simulation structures are shown in stick representation (same orientation as SFTI-1, Figure 1) and are used to display the distribution of intramolecular hydrogen bonds (dashed cylinders: coloured by frequency as shown in the figure key). Inhibition plots show residual proteolytic activity (y-axis) at various inhibitor concentrations (x-axis) and were used to calculate Ki. Data represent kinetic rates (mean ± S.E.M.) from three independent experiments performed in triplicate. Simulations were performed using NAMD 2.9 [40], average structures were generated using VMD 1.9.1 [39] and structural graphics were produced using CCP4MG [41].

Figure 2
Intramolecular hydrogen bond analysis and inhibitory activity for SFTI-1 and engineered inhibitor variants

(AF) For each inhibitor, the inhibition constant (Ki) determined from three independent assays performed in triplicate is shown next to the variant name, along with the intramolecular hydrogen bond frequency (AvH) as the mean ± S.D. from 10 independent 1 ns trajectories. Average simulation structures are shown in stick representation (same orientation as SFTI-1, Figure 1) and are used to display the distribution of intramolecular hydrogen bonds (dashed cylinders: coloured by frequency as shown in the figure key). Inhibition plots show residual proteolytic activity (y-axis) at various inhibitor concentrations (x-axis) and were used to calculate Ki. Data represent kinetic rates (mean ± S.E.M.) from three independent experiments performed in triplicate. Simulations were performed using NAMD 2.9 [40], average structures were generated using VMD 1.9.1 [39] and structural graphics were produced using CCP4MG [41].

To guide optimization of the intramolecular hydrogen bond network, we used MD simulations to screen virtual inhibitor libraries, as previously described [11]. For both 4 and 5, substitutions were performed at Phe12 (located on the non-contact β-strand), which was replaced by each of the naturally occurring amino acids (excluding cysteine to prevent formation of disulfide bond isomers) to generate 18 new variants for each inhibitor. The tendency for each variant to form intramolecular hydrogen bonds when bound to its target protease (KLK14 or KLK5) was then calculated from 10 independent 1 ns simulation trajectories.

Analysing simulation trajectories for each inhibitor indicated that substituting residue 12 had altered the hydrogen bond network within most variants (Supplementary Table S2). Substitutions that improved the intramolecular hydrogen bond network were often associated with new interactions formed between the side chain of the substituted residue and neighbouring backbone or side chain atoms. For KLK14 inhibitors, five second-generation variants covering a range of hydrogen bond tendencies higher than Phe12 were synthesized and evaluated in competitive inhibition assays: Glu12 (6), Gln12 (7), Thr12 (8), Asp12 (9), Asn12 (10). All variants were found to be more effective KLK14 inhibitors (Table 1) with the improvement in Ki generally following the increased tendency to form intramolecular hydrogen bonds (Figure 2, Supplementary Figure S4), except for 6. The most potent second generation variant was 10 (Ki=2.2 nM; Figure 2D), which showed 13-fold improvement from the starting variant, 4. Similarly, substituting Phe12 in 5 to restore intramolecular hydrogen bonds led to the design of a second generation KLK5 inhibitor (13) that showed improved activity (Ki=31.5 nM; Figure 2E). However, an exception was identified for each set of inhibitors (6 and 12) where an increase in intramolecular hydrogen bonds did not produce an improvement in activity, suggesting that other factors, including the location of hydrogen bonds and subsequent interactions with the protease, may also need to be considered.

Next, we examined how optimizing the intramolecular hydrogen bond network had changed the inhibitor's binding kinetics. For KLK14, comparing 10 to 4 revealed that the increase in binding affinity derived from 6.7-fold improvement in koff and 1.9-fold increase in kon (Table 1). Examining variants 7–9 also revealed that the increase in activity against KLK14 was predominantly due to improvements in koff. Similarly, 13 was a more effective KLK5 inhibitor than 5 as koff was 3.1-fold slower (overcoming a slight decrease in kon). Collectively, these findings illustrate the importance of the intramolecular hydrogen bond network when engineering Laskowski inhibitors, such that restoring these interactions after optimizing the binding sequence can provide further improvements in Ki. In a previous study, we observed a similar effect with engineered SFTI variants and a different target protease, KLK4 [11]. We have extended this finding in the present study by demonstrating that the improvement in Ki predominantly results from slowing the dissociation rate. Indeed, analysing the first generation variants for KLK14 revealed that the most potent variant (2) also displayed a prominent network of intramolecular hydrogen bonds (Supplementary Figure S4F) and thus the dissociation rate was 6-fold slower for 2 compared with 4 (Table 1).

Engineered inhibitor selectivity mainly derives from variation in the dissociation rate

An additional aim of engineering the inhibitor's binding loop is to generate variants with improved selectivity for the target protease. Accordingly, the newly developed inhibitors for KLK14 and KLK5 were screened against an array of off-target serine proteases to examine their selectivity. Initial assays revealed that 10 was an equipotent inhibitor of KLK14 (Table 1) and KLK4 (Ki=1.5 nM). Therefore, Val4 was replaced with isoleucine as previous specificity analyses have shown that KLK14 tolerates both valine and isoleucine at P2, whereas KLK4 prefers valine [48]. This led to a third generation variant (11) that showed similar activity against KLK14, but 6-fold lower activity against KLK4 and more than 180-fold selectivity over remaining off-targets, including KLK5 and trypsin (Table 2). Initial screening also showed that 13 was an equipotent inhibitor of KLK5 (Table 1) and KLK14 (Ki=31.5 nM). Therefore, we substituted Ile7 with tyrosine (the corresponding P2′ residue in SPINK9) [47]. This variant (14) showed a 5-fold increase in activity against KLK5 (Ki=6.2 nM), a 5.6-fold decrease in activity against KLK14 (Ki=176 nM) and at least 20-fold selectivity over remaining off-targets, except KLK4 (Table 2).

Table 2
Selectivity analysis for engineered variants 11 and 14
Protease Variant 111 Fold selectivity Variant 141 Fold selectivity 
KLK4 Ki=9.0±0.9 4.5 Ki=0.65±0.05 NS2 
KLK5 Ki=362±14 181 Ki=6.2±0.2 – 
KLK7 IC50 > 7500 NI3 Ki=143±5.3 23 
KLK14 Ki=2.0±0.1 – Ki=176±7.1 28 
Plasma kallikrein IC50 > 10000 NI3 IC50 > 10000 NI3 
Trypsin Ki=379±9.2 190 Ki=489±11 79 
Matriptase IC50 > 10000 NI3 IC50 > 10000 NI3 
Chymotrypsin IC50 > 10000 NI3 IC50 > 10000 NI3 
Plasmin IC50 > 10000 NI3 IC50 > 10000 NI3 
Thrombin IC50 > 10000 NI3 IC50 > 10000 NI3 
Protease Variant 111 Fold selectivity Variant 141 Fold selectivity 
KLK4 Ki=9.0±0.9 4.5 Ki=0.65±0.05 NS2 
KLK5 Ki=362±14 181 Ki=6.2±0.2 – 
KLK7 IC50 > 7500 NI3 Ki=143±5.3 23 
KLK14 Ki=2.0±0.1 – Ki=176±7.1 28 
Plasma kallikrein IC50 > 10000 NI3 IC50 > 10000 NI3 
Trypsin Ki=379±9.2 190 Ki=489±11 79 
Matriptase IC50 > 10000 NI3 IC50 > 10000 NI3 
Chymotrypsin IC50 > 10000 NI3 IC50 > 10000 NI3 
Plasmin IC50 > 10000 NI3 IC50 > 10000 NI3 
Thrombin IC50 > 10000 NI3 IC50 > 10000 NI3 

*Values are in nM units

†NS denotes not selective

‡NI denotes no substantial inhibition

The kinetic basis for inhibitor selectivity was subsequently examined by focusing on a subset of three proteases (KLK5, KLK14 and trypsin) and studying association and dissociation rates for 11 and 14. Here, the degree of complementarity in each protease/inhibitor interaction will be different as the active site cleft of each protease has a unique biophysical signature (Figure 3). Despite this, the association rate appeared to be rapid for each protease/inhibitor pair, irrespective of the inhibitor's binding affinity. For 11, steady state inhibition was quickly reached for KLK5 and trypsin when the inhibitor was added prior to or simultaneously with substrate, indicating fast association and fast dissociation rates (Figures 3A and 3C). By contrast, onset of the steady state was noticeably delayed for KLK14, consistent with slow dissociation of protease/inhibitor complexes (Figure 3B). Similar effects were observed for 14, where selectivity for KLK5 over KLK14 was based on 10.4-fold variation in koff and 2.8-fold difference in kon (Table 1, Figures 3D–3F) and also 1, where higher affinity for trypsin compared with KLK5 and KLK14 was due to up to 22-fold change in kon, but up to 240-fold variation in koff (Table 1). Taken together, these data indicate that the sequence of the binding loop influences inhibitor selectivity by modulating both the association rate and the dissociation rate, but the latter is the predominant factor. However, kon would probably become more a prominent influence (on inhibitor selectivity) when the inhibitor's P1 residue does not match the specificity profile of the protease of interest. Indeed, previous studies have shown that replacing P1 lysine with alanine in aprotinin decreases kon with trypsin by 230-fold [49]. This effect appears largely confined to the P1 residue as substituting other residues in the binding loop with alanine had only modest effects on kon [49].

Selective SFTI variants show divergent dissociation rates

Figure 3
Selective SFTI variants show divergent dissociation rates

Upper panels show the molecular surface of KLK5 (PDB ID: 2PSX), KLK14 (homology model) and trypsin (PDB ID: 1SFI) coloured by electrostatic potential (blue: positive, red: negative). For each protease, surface electrostatic potential was calculated using built-in functions in CCP4MG [41]. Variant 11 is shown in stick representation (black) bound to the active site of each protease. (AF) Reaction progress plots (0–60 s) illustrating substrate cleavage (y-axis) over time (x-axis) following addition of substrate to pre-formed protease/inhibitor complexes () or simultaneously with inhibitor (). The concentration of protease for each assay was 10 nM KLK5, 1 nM KLK14 and 1 nM trypsin, the concentration of substrate was 150 μM (Ac-YRSR-pNA for KLK5, Ac-YANR-pNA for KLK14 and Ac-YASR-pNA for trypsin) and the concentration of inhibitor is given in the top left corner of each graph. To demonstrate a shift from the uninhibited rate, reaction curves from control assays without inhibitor are also shown (▲). Note that the reaction in panels (B and D) has not reached steady state by 60 s (koff was determined using plots covering an extended time course).

Figure 3
Selective SFTI variants show divergent dissociation rates

Upper panels show the molecular surface of KLK5 (PDB ID: 2PSX), KLK14 (homology model) and trypsin (PDB ID: 1SFI) coloured by electrostatic potential (blue: positive, red: negative). For each protease, surface electrostatic potential was calculated using built-in functions in CCP4MG [41]. Variant 11 is shown in stick representation (black) bound to the active site of each protease. (AF) Reaction progress plots (0–60 s) illustrating substrate cleavage (y-axis) over time (x-axis) following addition of substrate to pre-formed protease/inhibitor complexes () or simultaneously with inhibitor (). The concentration of protease for each assay was 10 nM KLK5, 1 nM KLK14 and 1 nM trypsin, the concentration of substrate was 150 μM (Ac-YRSR-pNA for KLK5, Ac-YANR-pNA for KLK14 and Ac-YASR-pNA for trypsin) and the concentration of inhibitor is given in the top left corner of each graph. To demonstrate a shift from the uninhibited rate, reaction curves from control assays without inhibitor are also shown (▲). Note that the reaction in panels (B and D) has not reached steady state by 60 s (koff was determined using plots covering an extended time course).

Maximizing the intramolecular hydrogen bond network generates high potency, multi-target inhibitors

Having engineered selective inhibitor variants by customizing the sequence of the canonical loop, we next focused solely on optimizing the intramolecular hydrogen bond network. To design this variant, we started with 10 and added to the existing favourable interactions involving Asn12 (Figure 2D) by substituting Thr2 at P4 to form a new internal hydrogen bond. Additionally, Val4 was replaced with threonine to regain the binding loop hydrogen bonds present in SFTI-1 (Figure 2A). Thus, both the P4 and the P2 side chains were now directed inward (Figure 2F). Despite having fewer side chain binding determinants than earlier engineered variants (11 and 14), the new variant (15) was a potent inhibitor of KLK5 (Ki=2.0 nM), KLK14 (Ki=0.4 nM) and trypsin (Ki=0.7 nM; Table 1). Additionally, replacing P1 arginine with phenylalanine yielded a potent inhibitor (16) of chymotrypsin (Ki=0.5 nM) and KLK7 (Ki=8.0 nM; Table 1).

Most engineering studies based on Laskowski inhibitors focus on increasing the number of favourable binding interactions in order to improve the template inhibitor's activity against a given protease target. With 15 and 16, we show that effective inhibition can be achieved by maximizing the intramolecular hydrogen bond network and leaving relatively few side chain contacts. For example, 15 showed higher activity against KLK5 and KLK14 than either of their selectively engineered variants (14 and 11; Table 1). Further, although 11 and 14 contained highly preferred binding sequences, these variants had similar kon values to 15 (Table 1). Rather, the increase in binding affinity for 15 was based on slower rates of dissociation (due to the inhibitor's extensive network of intramolecular hydrogen bonds). This difference in koff exceeded 30-fold when compared with first generation variants (4 and 5) that were deficient in hydrogen bonds (Table 1; Figures 2C–2D).

Interestingly, although the dissociation rate for 15 and trypsin was slow, koff was 78-fold slower for 1 (Table 1). These two inhibitors are almost identical in the binding loop (residues 3–11), but are very different in the side loop, including at the P4 residue. In simulations of 1 or 15 bound to trypsin, the P4 residue of each inhibitor (Arg2 and Thr2 respectively) was found to contribute to the intramolecular hydrogen bond network, but for 1, Arg2 also formed a hydrogen bond with trypsin (via Asn97). In silico energy calculations for 15 bound to trypsin revealed that the relative interaction energy between the protease and inhibitor was −196 kcal/mol, with Thr2 contributing only −11 kcal/mol as it lacked side chain contacts with trypsin. By contrast, the relative interaction energy was higher for 1 bound to trypsin (−219 kcal/mol) and Arg2 contributed −46 kcal/mol. As both inhibitors have a well-configured intramolecular hydrogen bond network, the additional interaction formed between Arg2 and trypsin may contribute to 1 having a slower koff compared with 15 (for trypsin). Taken together, these findings suggest that the inhibitor's intramolecular and intermolecular interactions both influence koff and that finding an optimal balance between each feature will allow inhibitors to achieve the slowest dissociation rates.

Backbone modified variants highlight the advantages of the Laskowski mechanism as a strategy for reversible serine protease inhibition

The quintessential feature of the Laskowski mechanism is the religation reaction, which allows the inhibitor to adopt the form of an efficient substrate to engage the protease, but protects it from rapid hydrolysis. However, there are several alternate strategies for generating substrate-like inhibitors that are hydrolysed at slow rates (or not at all), including transition state analogues and non-cleavable peptides. Therefore, to investigate why Laskowski inhibitors have expanded prolifically by convergent evolution, we modified the reactive site chemistry of the broad-range variant 15 to change its mode of action. First, a peptide aldehyde variant was synthesized to produce a transition state analogue of 15. This variant (15CHO) is acyclic at the reactive site (Figure 4) and can form a covalent (hemiacetal) complex with the protease but cannot be religated. For trypsin and KLK14, the koff obtained for 15CHO was similar to 15 (Table 1), indicating that forming a stable covalent complex with the protease could replicate the slow dissociation kinetics of a Laskowski inhibitor. Second, a non-cleavable variant (15NMe) was synthesized by replacing Ser6 with N-methyl-L-Ser (Figure 4). This inhibitor does not form a covalent complex with the protease but once bound, it cannot be hydrolysed (analogous to complete religation efficiency). Synthetically protecting the reactive site from hydrolysis also preserved the inhibitor's ability to remain at the active site (Table 1), illustrating that slow dissociation kinetics could be attained without forming a covalent complex. However, both 15CHO and 15NMe showed dramatically lower binding affinity compared with 15 (25–160-fold change in Ki) due to substantial decreases in kon (59- and 39-fold for 15CHO and 135- and 197-fold for 15NMe). Similarly, N-methylating SFTI-1 at the reactive site (1NMe) markedly impaired its activity against trypsin, mainly due to a large decrease in kon (100-fold; Table 1).

Schematic diagram showing the reactive site chemistry for variants 15, 15CHO and 15NMe

Figure 4
Schematic diagram showing the reactive site chemistry for variants 15, 15CHO and 15NMe

The sequence of 15 is shown in three letter amino acid code (residue number in subscript), except for the P1 residue (Arg5) and P1′ residue (Ser6) which are shown by structural representation (the P2-P2′ segment is enclosed by a dashed rectangle). The inhibitor backbone is cyclized between Gly1 and Asn14 and an intramolecular disulfide bond is present between Cys3 and Cys11. Changes in the reactive site chemistry are illustrated for the transition state analogue (peptide aldehyde, 15CHO) and N-methylated scissile bond (15NMe) variants by showing the T4-I7 (P2-P2′) segment of each inhibitor.

Figure 4
Schematic diagram showing the reactive site chemistry for variants 15, 15CHO and 15NMe

The sequence of 15 is shown in three letter amino acid code (residue number in subscript), except for the P1 residue (Arg5) and P1′ residue (Ser6) which are shown by structural representation (the P2-P2′ segment is enclosed by a dashed rectangle). The inhibitor backbone is cyclized between Gly1 and Asn14 and an intramolecular disulfide bond is present between Cys3 and Cys11. Changes in the reactive site chemistry are illustrated for the transition state analogue (peptide aldehyde, 15CHO) and N-methylated scissile bond (15NMe) variants by showing the T4-I7 (P2-P2′) segment of each inhibitor.

Pseudo-substrate binding efficiency requires the canonical loop conformation

To explore the structural basis for the decrease in association efficiency (particularly for 15NMe), solution structures of 15 and 15NMe were characterized by NMR spectroscopy (Figure 5; Supplementary Table S3 and S4). Overlay of the 20 lowest energy structures for 15 revealed that the inhibitor showed close overall similarity to SFTI-1 [50], including across the P2-P2′ residues of the canonical loop (see Supplementary Table S5 for detailed flexibility and conformational analyses). In contrast, the corresponding segment adopted a different geometry in structures of 15NMe. Accommodating the N-methyl group required rotations of up to 90° in either direction from the position of the unmodified amide, causing rearrangement of the binding loop and departure from the canonical conformation. Other structural features such as the contact β-strand and side loop were largely unaffected.

Solution structures for SFTI-1 and engineered variants 15 and 15NMe

Figure 5
Solution structures for SFTI-1 and engineered variants 15 and 15NMe

Structures for SFTI-1 (1, blue, PDB ID: 1JBL) [50], 15 (green) and 15NMe (orange) are represented by an overlay of the 20 lowest energy NMR structures (backbone trace) for each variant (left panels). The P2-P2′ residues for each inhibitor are labelled (on the corresponding Cα atom) and NMR statistics and 1H chemical shifts for 15 and 15NMe are reported in Supplementary Tables S3 and S4. Right panels show an overlay of the average structure from MD simulation in solution (coloured) or bound to protease (silver) for each variant. The unmodified amide or N-methyl group is coloured magenta.

Figure 5
Solution structures for SFTI-1 and engineered variants 15 and 15NMe

Structures for SFTI-1 (1, blue, PDB ID: 1JBL) [50], 15 (green) and 15NMe (orange) are represented by an overlay of the 20 lowest energy NMR structures (backbone trace) for each variant (left panels). The P2-P2′ residues for each inhibitor are labelled (on the corresponding Cα atom) and NMR statistics and 1H chemical shifts for 15 and 15NMe are reported in Supplementary Tables S3 and S4. Right panels show an overlay of the average structure from MD simulation in solution (coloured) or bound to protease (silver) for each variant. The unmodified amide or N-methyl group is coloured magenta.

Anticipating that the different geometry of 15NMe might affect protease binding, we performed MD simulations to compare the conformation of SFTI-1, 15 and 15NMe in solution to the inhibitor's conformation in complex with protease. For all three variants, the average structure from simulating the inhibitor in solvent (without protease) superimposed with one of the possible solutions derived from the respective NMR dataset with 0.5 Å RMSD (all backbone atoms). Overlay of each simulation structure (bound and unbound) for 15 revealed that the average structure in solution was essentially the same as the average structure bound to protease (Figure 5), indicating that 15 was pre-organized for efficient protease binding. This effect was also seen for SFTI-1 (1), but not 15NMe which appeared to require a conformational change to reach a favoured binding state. An earlier study exploring the efficient binding of Laskowski inhibitors suggested that formation of the protease–inhibitor complex is based on a dynamic selection process where the protease selects a matching conformer from within the inhibitor's ensemble of solution structures generated by fast timescale oscillations [10]. Our data for 15NMe indicate that departure from the conserved geometry of the canonical loop decreased the proportion of productive binding states within the inhibitor's ensemble of solution structures and thus dramatically lowered its association efficiency. Interestingly, whereas previous studies have focused on the inhibitor's contact β-strand as the major driver of protease interaction [51], our data with 15NMe shows that the broader geometry of the canonical loop is important as the contact β-strand was largely unaffected by N-methylating the reactive site bond.

CONCLUSIONS

In the present study, we have paired structural analyses with detailed functional assays to explore how engineering the sequence or backbone chemistry of the model Laskowski inhibitor, SFTI-1, influences its association and dissociation kinetics. We demonstrate that the conserved conformation of the canonical loop pre-organizes the inhibitor for binding within the protease active site and is a major contributor to rapid association rates. Further, we show that the inhibitor's sequence (including both the binding sequence and its intramolecular scaffolding) also influences kon, but has a much greater impact on koff by modulating the degree of complementarity with the target protease and providing stabilizing interactions that are regarded to support the religation reaction, including hydrogen and disulfide bonds [1117]. Guided by these findings, we successfully designed highly potent inhibitors for several proteases that were customized either for selective or multi-target inhibition starting from a single engineering template (SFTI-1).

By studying the binding kinetics of engineered variants, in addition to their affinity for the target protease, we were able to pin-point the impacts on kon and koff that accompanied changes in the inhibitor's potency or selectivity. This is important as only considering Ki when characterizing engineered inhibitor variants can overlook key details relating to the inhibitor's activity. For example, variants 4 and 15NMe are less than 1.5-fold different in Ki when assayed against KLK14, but differ by 75-fold in kon and by 54-fold in koff (Table 1). This level of detail was critical for uncovering the kinetic basis for the selectivity of variants 11 and 14 and understanding the advantages of Laskowski inhibitors compared with other types of reversible protease inhibitors. In the present study, we demonstrated that slow dissociation could be achieved either by the Laskowski mechanism or by chemically modifying the reactive site (both of which obstruct completion of the catalytic cycle). However, for the first time, we showed that only the Laskowski mechanism provides an ideal balance between fast association and slow dissociation kinetics. As SFTI-1 operates in the same way as protease inhibitors from 18 additional families, these findings provide insight into why Nature has arrived at this strategy for reversible serine protease inhibition at least 19 times by convergent evolution [1,8]. Indeed, a similar mechanism appears to be evident in a previously characterized reversible metalloprotease inhibitor [52].

Our findings on Laskowski inhibitor function also extend to engineering inhibitor variants and are not confined to SFTI-1 but are likely to be relevant to diverse Laskowski inhibitors. Since the major features that contribute to rapid association rates (the conserved geometry of the binding loop and the P1 residue) offer limited opportunities for optimization, the most productive way to modulate an inhibitor's potency and selectivity is to alter its dissociation kinetics. This phenomenon can be exploited to develop broad-range or selective inhibitors. For broad-range inhibition, the binding surface must be designed to match the active site of all intended targets. In the case of 15 and 16, several residues protruding from the inhibitor scaffold were substituted to maximize the intramolecular hydrogen bond network and minimize the number of selectivity determinants. Conversely, for selective inhibition, the binding sequence must be preferred by relatively few proteases. This was achieved for 11, 14 and in a previous study [11,33] by modifying the sequence of the SFTI contact β-strand then refining the intramolecular hydrogen bond network. In conclusion, Laskowski inhibitors are often described as mimicking the classical lock and key model for enzyme–substrate interactions [1,10]. By systematically engineering the inhibitor's inter- and intra-molecular interactions, we have been able to produce separate keys for different locks and one key that fits several locks.

AUTHOR CONTRIBUTION

Simon de Veer and Joakim Swedberg contributed to the study design, synthesized peptides, performed kinetic assays, analysed data and wrote the manuscript. Muharrem Akcan and Johan Rosengren performed NMR experiments and calculated NMR structures. Maria Brattsand contributed to the study design and developed Pichia strains for recombinant protein expression. Jonathan Harris and David Craik supervised the study, contributed to the study design and wrote the manuscript.

FUNDING

This study was supported by the Australian National Health and Medical Research Council (NHMRC) [grant numbers 1059410, 1026501 (to D.J.C.), 631420 (to J.R.) and 1069819 (to J.E.S)].

Abbreviations

     
  • KLK

    kallikrein-related peptidase

  •  
  • pNA

    para-nitroanilide

  •  
  • SFTI-1

    sunflower trypsin inhibitor-1

References

References
1
Laskowski
M.
Jr
Qasim
M.A.
What can the structures of enzyme-inhibitor complexes tell us about the structures of enzyme substrate complexes?
Biochim. Biophys. Acta
2000
, vol. 
1477
 (pg. 
324
-
337
)
[PubMed]
2
Laskowski
M.
Jr
Kato
I.
Protein inhibitors of proteinases
Annu. Rev. Biochem.
1980
, vol. 
49
 (pg. 
593
-
626
)
[PubMed]
3
Radisky
E.S.
Koshland
D.E.
Jr
A clogged gutter mechanism for protease inhibitors
Proc. Natl. Acad. Sci. U.S.A.
2002
, vol. 
99
 (pg. 
10316
-
10321
)
[PubMed]
4
Finkenstadt
W.R.
Laskowski
M.
Jr
Peptide bond cleavage on trypsin-trypsin inhibitor complex formation
J. Biol. Chem.
1965
, vol. 
240
 (pg. 
962
-
963
)
[PubMed]
5
Zakharova
E.
Horvath
M.P.
Goldenberg
D.P.
Structure of a serine protease poised to resynthesize a peptide bond
Proc. Natl. Acad. Sci. U.S.A.
2009
, vol. 
106
 (pg. 
11034
-
11039
)
[PubMed]
6
Finkenstadt
W.R.
Laskowski
M.
Jr
Resynthesis by trypsin of the cleaved peptide bond in modified soybean trypsin inhibitor
J. Biol. Chem.
1967
, vol. 
242
 (pg. 
771
-
773
)
[PubMed]
7
Glotzbach
B.
Schmelz
S.
Reinwarth
M.
Christmann
A.
Heinz
D.W.
Kolmar
H.
Structural characterization of Spinacia oleracea trypsin inhibitor III (SOTI-III)
Acta Crystallogr. D Biol. Crystallogr.
2013
, vol. 
69
 (pg. 
114
-
120
)
[PubMed]
8
Otlewski
J.
Jelen
F.
Zakrzewska
M.
Oleksy
A.
The many faces of protease-protein inhibitor interaction
EMBO J.
2005
, vol. 
24
 (pg. 
1303
-
1310
)
[PubMed]
9
Bode
W.
Huber
R.
Natural protein proteinase inhibitors and their interaction with proteinases
Eur. J. Biochem.
1992
, vol. 
204
 (pg. 
433
-
451
)
[PubMed]
10
Gaspari
Z.
Varnai
P.
Szappanos
B.
Perczel
A.
Reconciling the lock-and-key and dynamic views of canonical serine protease inhibitor action
FEBS Lett.
2010
, vol. 
584
 (pg. 
203
-
206
)
[PubMed]
11
Swedberg
J.E.
de Veer
S.J.
Sit
K.C.
Reboul
C.F.
Buckle
A.M.
Harris
J.M.
Mastering the canonical loop of serine protease inhibitors: enhancing potency by optimising the internal hydrogen bond network
PLoS One
2011
, vol. 
6
 pg. 
e19302
 
[PubMed]
12
Schirra
H.J.
Guarino
R.F.
Anderson
M.A.
Craik
D.J.
Selective removal of individual disulfide bonds within a potato type II serine proteinase inhibitor from Nicotiana alata reveals differential stabilization of the reactive-site loop
J. Mol. Biol.
2010
, vol. 
395
 (pg. 
609
-
626
)
[PubMed]
13
Colgrave
M.L.
Korsinczky
M.J.
Clark
R.J.
Foley
F.
Craik
D.J.
Sunflower trypsin inhibitor-1, proteolytic studies on a trypsin inhibitor peptide and its analogs
Biopolymers
2010
, vol. 
94
 (pg. 
665
-
672
)
[PubMed]
14
Zakharova
E.
Horvath
M.P.
Goldenberg
D.P.
Functional and structural roles of the Cys14-Cys38 disulfide of bovine pancreatic trypsin inhibitor
J. Mol. Biol.
2008
, vol. 
382
 (pg. 
998
-
1013
)
[PubMed]
15
Dasgupta
J.
Khamrui
S.
Dattagupta
J.K.
Sen
U.
Spacer Asn determines the fate of Kunitz (STI) inhibitors, as revealed by structural and biochemical studies on WCI mutants
Biochemistry
2006
, vol. 
45
 (pg. 
6783
-
6792
)
[PubMed]
16
Radisky
E.S.
Lu
C.J.
Kwan
G.
Koshland
D.E.
Jr
Role of the intramolecular hydrogen bond network in the inhibitory power of chymotrypsin inhibitor 2
Biochemistry
2005
, vol. 
44
 (pg. 
6823
-
6830
)
[PubMed]
17
McBride
J.D.
Brauer
A.B.
Nievo
M.
Leatherbarrow
R.J.
The role of threonine in the P2 position of Bowman-Birk proteinase inhibitors: studies on P2 variation in cyclic peptides encompassing the reactive site loop
J. Mol. Biol.
1998
, vol. 
282
 (pg. 
447
-
458
)
[PubMed]
18
Shaw
G.L.
Davis
B.
Keeler
J.
Fersht
A.R.
Backbone dynamics of chymotrypsin inhibitor 2: effect of breaking the active site bond and its implications for the mechanism of inhibition of serine proteases
Biochemistry
1995
, vol. 
34
 (pg. 
2225
-
2233
)
[PubMed]
19
Cicardi
M.
Levy
R.J.
McNeil
D.L.
Li
H.H.
Sheffer
A.L.
Campion
M.
Horn
P.T.
Pullman
W.E.
Ecallantide for the treatment of acute attacks in hereditary angioedema
N. Engl. J. Med.
2010
, vol. 
363
 (pg. 
523
-
531
)
[PubMed]
20
Luckett
S.
Garcia
R.S.
Barker
J.J.
Konarev
A.V.
Shewry
P.R.
Clarke
A.R.
Brady
R.L.
High-resolution structure of a potent, cyclic proteinase inhibitor from sunflower seeds
J. Mol. Biol.
1999
, vol. 
290
 (pg. 
525
-
533
)
[PubMed]
21
de Veer
S.J.
Furio
L.
Harris
J.M.
Hovnanian
A.
Proteases: common culprits in human skin disorders
Trends Mol. Med.
2014
, vol. 
20
 (pg. 
166
-
178
)
[PubMed]
22
Prassas
I.
Eissa
A.
Poda
G.
Diamandis
E.P.
Unleashing the therapeutic potential of human kallikrein-related serine proteases
Nat. Rev. Drug Discov.
2015
, vol. 
14
 (pg. 
183
-
202
)
[PubMed]
23
Briot
A.
Deraison
C.
Lacroix
M.
Bonnart
C.
Robin
A.
Besson
C.
Dubus
P.
Hovnanian
A.
Kallikrein 5 induces atopic dermatitis-like lesions through PAR2-mediated thymic stromal lymphopoietin expression in Netherton syndrome
J. Exp. Med.
2009
, vol. 
206
 (pg. 
1135
-
1147
)
[PubMed]
24
Descargues
P.
Deraison
C.
Prost
C.
Fraitag
S.
Mazereeuw-Hautier
J.
D'Alessio
M.
Ishida-Yamamoto
A.
Bodemer
C.
Zambruno
G.
Hovnanian
A.
Corneodesmosomal cadherins are preferential targets of stratum corneum trypsin- and chymotrypsin-like hyperactivity in Netherton syndrome
J. Invest. Dermatol.
2006
, vol. 
126
 (pg. 
1622
-
1632
)
[PubMed]
25
Furio
L.
de Veer
S.
Jaillet
M.
Briot
A.
Robin
A.
Deraison
C.
Hovnanian
A.
Transgenic kallikrein 5 mice reproduce major cutaneous and systemic hallmarks of Netherton syndrome
J. Exp. Med.
2014
, vol. 
211
 (pg. 
499
-
513
)
[PubMed]
26
Stefansson
K.
Brattsand
M.
Roosterman
D.
Kempkes
C.
Bocheva
G.
Steinhoff
M.
Egelrud
T.
Activation of proteinase-activated receptor-2 by human kallikrein-related peptidases
J. Invest. Dermatol.
2008
, vol. 
128
 (pg. 
18
-
25
)
[PubMed]
27
Cork
M.J.
Danby
S.G.
Vasilopoulos
Y.
Hadgraft
J.
Lane
M.E.
Moustafa
M.
Guy
R.H.
Macgowan
A.L.
Tazi-Ahnini
R.
Ward
S.J.
Epidermal barrier dysfunction in atopic dermatitis
J. Invest. Dermatol.
2009
, vol. 
129
 (pg. 
1892
-
1908
)
[PubMed]
28
Loessner
D.
Quent
V.M.
Kraemer
J.
Weber
E.C.
Hutmacher
D.W.
Magdolen
V.
Clements
J. A.
Combined expression of KLK4, KLK5, KLK6, and KLK7 by ovarian cancer cells leads to decreased adhesion and paclitaxel-induced chemoresistance
Gynecol. Oncol.
2012
, vol. 
127
 (pg. 
569
-
578
)
[PubMed]
29
Prezas
P.
Arlt
M.J.
Viktorov
P.
Soosaipillai
A.
Holzscheiter
L.
Schmitt
M.
Talieri
M.
Diamandis
E. P.
Kruger
A.
Magdolen
V.
Overexpression of the human tissue kallikrein genes KLK4, 5, 6, and 7 increases the malignant phenotype of ovarian cancer cells
Biol. Chem.
2006
, vol. 
387
 (pg. 
807
-
811
)
[PubMed]
30
Gratio
V.
Loriot
C.
Virca
G.D.
Oikonomopoulou
K.
Walker
F.
Diamandis
E.P.
Hollenberg
M.D.
Darmoul
D.
Kallikrein-related peptidase 14 acts on proteinase-activated receptor 2 to induce signaling pathway in colon cancer cells
Am. J. Pathol.
2011
, vol. 
179
 (pg. 
2625
-
2636
)
[PubMed]
31
Brattsand
M.
Stefansson
K.
Hubiche
T.
Nilsson
S.K.
Egelrud
T.
SPINK9: a selective, skin-specific Kazal-type serine protease inhibitor
J. Invest. Dermatol.
2009
, vol. 
129
 (pg. 
1656
-
1665
)
[PubMed]
32
Ramsay
A.J.
Dong
Y.
Hunt
M.L.
Linn
M.
Samaratunga
H.
Clements
J.A.
Hooper
J.D.
Kallikrein-related peptidase 4 (KLK4) initiates intracellular signaling via protease-activated receptors (PARs)
J. Biol. Chem.
2008
, vol. 
283
 (pg. 
12293
-
12304
)
[PubMed]
33
Swedberg
J.E.
Nigon
L.V.
Reid
J.C.
de Veer
S.J.
Walpole
C.M.
Stephens
C.R.
Walsh
T.P.
Takayama
T.K.
Hooper
J.D.
Clements
J.A.
Buckle
A.M.
Harris
J.M.
Substrate-guided design of a potent and selective kallikrein-related peptidase inhibitor for kallikrein 4
Chem. Biol.
2009
, vol. 
16
 (pg. 
633
-
643
)
[PubMed]
34
de Veer
S.J.
Swedberg
J.E.
Parker
E.A.
Harris
J.M.
Non-combinatorial library screening reveals subsite cooperativity and identifies new high efficiency substrates for kallikrein-related peptidase 14
Biol. Chem.
2012
, vol. 
393
 (pg. 
331
-
341
)
[PubMed]
35
Baici
A.
Gyger-Marazzi
M.
The slow, tight-binding inhibition of cathepsin B by leupeptin. A hysteretic effect
Eur. J. Biochem.
1982
, vol. 
129
 (pg. 
33
-
41
)
[PubMed]
36
Debela
M.
Goettig
P.
Magdolen
V.
Huber
R.
Schechter
N.M.
Bode
W.
Structural basis of the zinc inhibition of human tissue kallikrein 5
J. Mol. Biol.
2007
, vol. 
373
 (pg. 
1017
-
1031
)
[PubMed]
37
Guex
N.
Peitsch
M.C.
SWISS-MODEL and the Swiss-PdbViewer: an environment for comparative protein modeling
Electrophoresis
1997
, vol. 
18
 (pg. 
2714
-
2723
)
[PubMed]
38
Krieger
E.
Koraimann
G.
Vriend
G.
Increasing the precision of comparative models with YASARA NOVA–a self-parameterizing force field
Proteins
2002
, vol. 
47
 (pg. 
393
-
402
)
[PubMed]
39
Humphrey
W.
Dalke
A.
Schulten
K.
VMD: visual molecular dynamics
J. Mol. Graph.
1996
, vol. 
14
 (pg. 
33
-
38
27–38
[PubMed]
40
Phillips
J.C.
Braun
R.
Wang
W.
Gumbart
J.
Tajkhorshid
E.
Villa
E.
Chipot
C.
Skeel
R.D.
Kale
L.
Schulten
K.
Scalable molecular dynamics with NAMD
J. Comput. Chem.
2005
, vol. 
26
 (pg. 
1781
-
1802
)
[PubMed]
41
McNicholas
S.
Potterton
E.
Wilson
K.S.
Noble
M.E.
Presenting your structures: the CCP4mg molecular-graphics software
Acta Crystallogr. D Biol. Crystallogr.
2011
, vol. 
67
 (pg. 
386
-
394
)
[PubMed]
42
Wüthrich
K.
NMR of Proteins and Nucleic Acids
1986
New York
Wiley–Interscience
43
Guntert
P.
Mumenthaler
C.
Wuthrich
K.
Torsion angle dynamics for NMR structure calculation with the new program DYANA
J. Mol. Biol.
1997
, vol. 
273
 (pg. 
283
-
298
)
[PubMed]
44
Brunger
A.T.
Adams
P.D.
Clore
G.M.
DeLano
W.L.
Gros
P.
Grosse-Kunstleve
R.W.
Jiang
J.S.
Kuszewski
J.
Nilges
M.
Pannu
N.S.
Read
R.J.
Rice
L.M.
Simonson
T.
Warren
G.L.
Crystallography & NMR system: a new software suite for macromolecular structure determination
Acta Crystallogr. D Biol. Crystallogr.
1998
, vol. 
54
 (pg. 
905
-
921
)
[PubMed]
45
Nederveen
A.J.
Doreleijers
J.F.
Vranken
W.
Miller
Z.
Spronk
C.A.
Nabuurs
S.B.
Guntert
P.
Livny
M.
Markley
J.L.
Nilges
M.
Ulrich
E.L.
Kaptein
R.
Bonvin
A.M.
RECOORD: a recalculated coordinate database of 500+ proteins from the PDB using restraints from the BioMagResBank
Proteins
2005
, vol. 
59
 (pg. 
662
-
672
)
[PubMed]
46
Conibear
A.C.
Rosengren
K.J.
Harvey
P.J.
Craik
D.J.
Structural characterization of the cyclic cystine ladder motif of theta-defensins
Biochemistry
2012
, vol. 
51
 (pg. 
9718
-
9726
)
[PubMed]
47
Brannstrom
K.
Ohman
A.
von Pawel Rammingen
U.
Olofsson
A.
Brattsand
M.
Characterization of SPINK9, a KLK5-specific inhibitor expressed in palmo-plantar epidermis
Biol. Chem.
2012
, vol. 
393
 (pg. 
369
-
377
)
[PubMed]
48
Borgono
C.A.
Gavigan
J.A.
Alves
J.
Bowles
B.
Harris
J.L.
Sotiropoulou
G.
Diamandis
E.P.
Defining the extended substrate specificity of kallikrein 1-related peptidases
Biol. Chem.
2007
, vol. 
388
 (pg. 
1215
-
1225
)
[PubMed]
49
Castro
M.J.
Anderson
S.
Alanine point-mutations in the reactive region of bovine pancreatic trypsin inhibitor: effects on the kinetics and thermodynamics of binding to beta-trypsin and alpha-chymotrypsin
Biochemistry
1996
, vol. 
35
 (pg. 
11435
-
11446
)
[PubMed]
50
Korsinczky
M.L.
Schirra
H.J.
Rosengren
K.J.
West
J.
Condie
B.A.
Otvos
L.
Anderson
M.A.
Craik
D.J.
Solution structures by 1H NMR of the novel cyclic trypsin inhibitor SFTI-1 from sunflower seeds and an acyclic permutant
J. Mol. Biol.
2001
, vol. 
311
 (pg. 
579
-
591
)
[PubMed]
51
Tyndall
J.D.
Nall
T.
Fairlie
D.P.
Proteases universally recognize beta strands in their active sites
Chem. Rev.
2005
, vol. 
105
 (pg. 
973
-
999
)
[PubMed]
52
Arolas
J.L.
Botelho
T.O.
Vilcinskas
A.
Gomis-Ruth
F.X.
Structural evidence for standard-mechanism inhibition in metallopeptidases from a complex poised to resynthesize a peptide bond
Angew. Chem. Int. Ed. Engl.
2011
, vol. 
50
 (pg. 
10357
-
10360
)
[PubMed]
53
Schechter
I.
Berger
A.
On the size of the active site in proteases. I. Papain
Biochem. Biophys. Res. Commun.
1967
, vol. 
27
 (pg. 
157
-
162
)
[PubMed]

Author notes

1

These authors contributed equally to this study

2

Current address: Department of Biochemistry, Dumlupinar University, Kütahya, Turkey