Structural Principles in the Development of Cyclic Peptidic Enzyme Inhibitors.

This review summarizes our studies in the development of small cyclic peptides for specifically modulating enzyme activity. Serine proteases share highly similar active sites but perform diverse physiological and pathological functions. From a phage-display peptide library, we isolated two mono-cyclic peptides, upain-1 (CSWRGLENHRMC) and mupain-1 (CPAYSRYLDC), which inhibit the activity of human and murine urokinase-type plasminogen activators (huPA and muPA) with Ki values in the micromolar or sub-micromolar range, respectively. The following affinity maturations significantly enhanced the potencies of the two peptides, 10-fold and >250-fold for upain-1 and mupain-1, respectively. The most potent muPA inhibitor has a potency (Ki = 2 nM) and specificity comparable to mono-clonal antibodies. Furthermore, we also found an unusual feature of mupain-1 that its inhibitory potency can be enhanced by increasing the flexibility, which challenges the traditional viewpoint that higher rigidity leading to higher affinity. Moreover, by changing a few key residues, we converted mupain-1 from a uPA inhibitor to inhibitors of other serine proteases, including plasma kallikrein (PK) and coagulation factor XIa (fXIa). PK and fXIa inhibitors showed Ki values in the low nanomolar range and high specificity. Our studies demonstrate the versatility of small cyclic peptides to engineer inhibitory potency against serine proteases and to provide a new strategy for generating peptide inhibitors of serine proteases.

Cyclic Peptidic Enzyme Inhibitors

In biology and pharmacy, the commonly accepted demarcations of drugs by molecular weights (MW) are “small molecules” (MW < 500 Da) and “macromolecules” (MW > 5000 Da) 1. Drugs with molecular weights ranging from 200 to 600 Da predominate among drugs approved by the U.S. Food and Drug Administpan class="Species">ratn>ion (FDA) (>90%) 2. Polypeptides with 10-20 amino acids are in the range between 500-5000 Da. Thus, peptide-based drugs are attracting increasing attention as they combine the advantages of small molecules and macromolecules, namely stability, potency, and specificity. Compared to linear peptides, cyclic peptides are more favorable candidates as pharmaceutics as they possess greater conformational constraint resulting in higher selectivity. Dysregulated enzyme activities cause physical disorders and are therapeutic targets of many pathological conditions. The only FDA-approved cyclic peptide-based enzyme inhibitor is cyclosporin A, an inhibitor of calcineurin 3. pan class="Chemical">Cyclosporin A interferes the activity and growth of T-cells and is therefore used as an immunosuppressant drug, which has an annual sale of >US $1.0 billion in the USA only 4. Serine proteases are widely involved in many physiological and pathological pathways and are demonstrated as therapeutic targets of many important diseases 5-7. In this article, we reviewed our journey in the development of cyclic peptidic inhibitors against serine proteases by combining de-novo screening and rational design. Our studies originate with two peptides, upain-1 (CSWRGLENHRMC) and mupain-1 (CPAYSRYLDC), which were screened from a phage-displayed peptide library as inhibitors of human and murine urokinase-type plasminogen activators (uPA), respectively.

uPA and cancer metastasis

pan class="Gene">Urokinase-type plasminogen activatorn> (pan class="Gene">uPA) is a pan class="Gene">serine protease with a molecular weight of 55-kDa.8, 9 uPA is originally secreted in the zymogen form, single chain uPA (sc-uPA), and is activated by plasmin or other proteases via the proteolysis of the peptide bond between Lys158 and Ile159 10. The disulfide-bridged two-chain uPA (tc-uPA) consists of the N-terminal amino terminal fragment (ATF), and the C-terminal catalytic domain. The N-terminal ATF, composed of a kringle domain and a growth factor-like domain 11, 12, binds to the uPA receptor (uPAR) tightly with a K value of 0.3 nM 13, 14. The catalytic domain performs the proteolytic function and activates the substrate, plasminogen 15. uPA serves in the plasminogen activation (PA) system, which was originally known to perform fibrinolytic functions in blood by dissolving blood clots (fibrin) through the generation of plasmin 16-19. A large number of studies have demonstrated that the PA system is also involved in many other physiological and pathological pathways, including wound healing 20, adhesion of cells 21, proliferation and metastasis of cancer 22, infectious diseases 23-25, and even neuropathology 26.

pan class="Gene">Plasminn> is the central enzyme of PA system, which performs two main functions (Figure 1) 27: 1) maintaining the fluidity of vessels by lysing the blood clots in blood 28, 29 and 2) turning over the extracellular matrix (ECM) components in tissue modeling 30, like wound healing. pan class="Gene">Plasmin is physiologically regulated by α2-antipan class="Gene">plasmin 31 and generated through the activation of its zymogen, plasminogen, by tissue-type and the urokinase-type plasminogen activators (tPA and uPA) 32, which are regulated by plasminogen activator inhibitor 1 and 2 (PAI-1 and PAI-2) 33-35. The uPA-triggered PA system is closely relevant to cancer proliferation and metastasis 30, 36, 37. The role of uPA in cancer proliferation and metastasis has not been thoroughly validated but majorly accepted as illustrated in Figure 1: Sc-uPA is activated generating active tc-uPA after binding to the membrane-anchored receptor, uPAR 38, 39. Tc-uPA on the cell surface then catalyzes the generation of plasmin, which consequently degrades the extracellular matrix (ECM). The overexpression of uPA and uPAR on the cell surface of tumor cells leads to aberrant plasminogen activation 40, resulting in the excessive degradation of ECM, which increases the mobility of cancer cells and facilitates cancer invasion.

Figure 1

plasminogen activation system

Inhibitors of uPA

Although the underlying biochemical processes for pan class="Disease">metastasisn> are still not fully understood, numerous studies have demonstpan class="Species">rated that pan class="Gene">uPA and uPAR play critical roles in the processes of cancer proliferation and metastasis 36. The attempts of inhibiting uPA activity can be dated from the mid-1960's 41, even before structural information of the enzyme was available 42. After over 50-years study, Mesupron® (Figure 2A), an oral small-molecule uPA inhibitor, is now in Phase II clinical trial for the therapy of pancreatic and breast cancers 43. Mesupron® was originally developed by WILEX AG, but was licensed to Link Health Group and RedHill Biopharma Ltd in 2014. Although performing obvious suppression of tumor metastasis in clinical trials, Mesupron® is not specific to uPA, as it inhibits uPA with a K value of 0.65 μM while showing comparable K values for the inhibition of many homogenous proteases, like plasmin (1.46 μM), thrombin (0.5 μM), coagulation factor Xa (0.98 μM), tPA (2.51 μM), and matriptase (0.37 μM) 44. Similar to Mesupron®, other small-molecule uPA inhibitors have also shown insufficient selectivity 42, 45-47.

Figure 2

A: Chemical structures of the reported uPA peptide-based inhibitors. B: Alignment of the catalytic domains of plasma kallikrein (PK), tissue-type and urkinase-type plasminogen activators (tPA and uPA), coagulation factor XIa (fXIa), and plasmin. C: Sequence alignment of uPA-relevant serine proteases around the S1 pockets. The regions of the S1 pockets were marked with dark green frames. The residues are colored according to the conservation.

Peptides are also of considerable interest as modulators of pan class="Gene">uPAn> as peptides exhibit much higher specificity than small-molecule inhibitors do (Figure 2A). pan class="Chemical">Sturzebecher and his colleagues reported a peptide-based pan class="Gene">uPA inhibitor CJ-463 (benzylsulfonyl-D-Ser-Ser-4- amidinobenzylamide) with a K value of 20 nM 48, which demonstrated good anti-tumor efficacy in a lung cancer animal model 49. Haemers screened a peptide-based irreversible uPA inhibitor showing IC50 value of 57 nM and high selectivity over homogenic enzymes 50. Later, the inhibitor was optimized and used for imaging breast cancer through the single photon emission computed tomography (SPECT) 51. With phage-display screening, Heinis's group reported bicyclic peptide inhibitors of uPA with different cyclization styles, like via two disulfide bridges 52, via locking three cysteine residues by trifunctional tris(bromomethyl)benzene 53, and through dithiol amino acids 54.

Catalytic domains of the trypsin-like serine proteases

n class="Gene">Serine proteasen>s are widely involved in many important physiological and pathological pathways, and are considered as therapeutic targets of many diseases 5. One critical issue for medicinal enzyme modulators is the specificity among homogenous enzymes, because the off-target inhibition of regulated enzyme might cause systemic damages. The failure of clinical trials of matrix metalloprotease inhibitors was just due to the lack of sufficient specificity 55. Similarly, although performing diverse physiological functions, serine proteases are known to have an identical catalytic domain in sequences and structures (Figure 2B), leading to a main challenge to engineer sufficient specificity to one particular protease. The S1 specificity pocket accommodates the side chains of the positive P1 residues of substrates. Like Mesupron®, small-molecule inhibitors suffer from the low specificity because they mostly target the S1 pockets, which are highly conserved among the serine protease family (Figure 2C) 42, 56. In contrast, peptide-based inhibitors showed higher selectivity for contacting multiple other regions outside the S1 pockets. These regions are much more variable among homologous enzymes than the S1 pockets, and are more specific for the recognition of substrates and pharmaceuticals 57.

Upain-1 derivatives, specific competitive inhibitors of human uPA

Screening of upain-1 from a phage-displayed library

One important source of peptide-based enzyme inhibitors is the de-novo peptide libraries 58, 59. Phage display represents a biological method of rapidly genepan class="Species">ratn>ing large random peptide libraries 60. To isolate pan class="Gene">uPA binding peptides, we used a phage-display peptide library with the formats of X and CX (C and X indicate pan class="Chemical">cysteine and randomized amino acids), containing >109 unique inserts 61. With the bait of human uPA (huPA) immobilized through monoclonal antibodies, mAb6, mAb12 or mAb390 62, we identified a sequence of CSWRGLENHRMC 63 referred to as upain-1. The upain-1 sequence expressed in fusion with the D1 and D2 domains of M13 phage's pIII protein was referred to as upain-1-D1D2. The synthetic upain-1 peptide and upain-1-D1D2 fusion protein competitively inhibited the proteolytic activity of huPA and the plasminogen activation on the surface of histiocytic lymphoma cells. Upain-1-D1D2 (K = 2.2 μM) showed 13.6-fold higher inhibitory potency than upain-1 peptide (K = 29.9 μM) did. In addition, both upain-1 and upain-1-D1D2 showed high selectivity to huPA over many other human enzymes, including trypsin, fXa, tPA, plasmin, plasma kallikrein.

We further determined the crystal structure of pan class="Gene">upan>in-1 in complex with the catalytic domain of hpan class="Gene">uPA with a resolution of 2.15 Å (PDB code: 2NWN, Figure 4A) 64. upain-1 docks into the catalytic cleft of uPA blocking the substrate-binding sites. Crystal structure shows that Ser2, Leu6, Arg10, and Met11 are solvent exposed and have very few contacts (197 Å2) with uPA, consistent with the alanine-scanning analysis that inactivating these residues had very mild influence on upain-1's affinity to uPA. In contrast, residues 3-5 and 7-9 contact largely with uPA (939 Å2) and showed significant importance in the affinity to huPA according to the alanine-scanning analysis. Furthermore, the crystal structure explains why upain-1 is an inhibitor but not cleaved as a substrate: upain-1 docks in a non-canonical substrate-like conformation 65. The bulky side chain of Trp3 hinders the scissile peptide bond (P1-P1') approaching the catalytic triad preventing the nucleophilic attack. Besides, the side-chain of Glu7 occupies the oxyanion hole, which typically accommodates the carbonyl oxygen of the P1 residues of substrates.

Figure 4

Structural features of upain-1 variants binding to huPA: A: the overview of the crystal structure of upain-1 (magenta sticks) in complex with the catalytic domain of human uPA (white surface) (PDB code: 2NWN), the solvent-exposed 37-, 60-, and 97-loops were colored yellow, green and light purple, respectively; B: the alignment of the N-terminus extended DGA-upain-1 (cyan sticks) and upain-1 (magenta sticks); C: the intramolecular hydrogen bond network between the extended DGA-tripeptide and the core upain-1 peptide; D: upain-1 (magenta sticks) and UK-504 (green sticks) form highly similar conformation on the surface of huPA, especially for the residues that form large contacts with huPA; E: upain-1 (magenta sticks) and bicyclic peptide 24 (yellow sticks) show mostly identical conformation when binding to huPA; F: upain-1-W3A was converted to a substrate of huPA. At pH 4.6 (green) and 5.5 (cyan), peptide was intact. At pH 7.4 (megenta), only electron density of Ala3, Arg4, Gly5, and Leu6 was observed, and the peptide bond between Arg4 and Gly5 was cleaved. At pH 9.0 (yellow), only electron density of Arg4 was observed, other residues of the peptide were disordered. At pH 7.4 (wheat), the inactive uPA did not cleave the peptide bond. Peptides at different pH were shown in cartoon, the side chains of Ala3, Arg4, Gly5, and Leu6 were shown in sticks, the residues that can be observed were listed in the table.

Improving inhibitory potency by increasing peptides' rigidity

Because of the pan class="Disease">insufficientn> inhibitory potency (K = 29.9 μM), we next performed affinity maturation to improve the affinity of upain-1 to huPA. The crystal structure shows that Arg10 and Met11 of upain-1 are solvent-exposed 64, and have minor contribution to the peptide-enzyme binding 63. Hence, we mutated these two residues to alanine generating peptide upain-2 (CSWRGLENHAAC), which shows identical K values (35 μM) to that of upain-1 (Table 1) 66. As mentioned, upain-1-D1D2 shows >10-fold higher inhibitory potency than upain-1. However, the only D1D2 recombinant protein shows non-measurable affinity 63. Some specific residues in D1D2 domains might facilitate the upain-1-uPA binding. Thus, we extended the N and C termini of upain-1 according to the sequence of upain-1-D1D2 fusion protein resulting in peptide MGSADGA-CSWRGLENHAAC-GAAG, which shows affinity (K = 3.5 μM) comparable to that of upain-1-D1D2 (K = 2.2 μM).

Table 1

Sequences and K values of inhibiting huPA of upain-1 variants.

Peptide code	Sequence*	Ki values of inhibiting huPA (μM)
upain-1	CSWRGLENHRMC	30 ± 1 (3)
D1D2-upain-1	CSWRGLENHRMC-D1D2	2.2 ± 0.1 (3)
upain-2	CSWRGLENHAAC	35 ± 8 (3)
MGSADGA-upain-1-GAAG	MGSADGA-CSWRGLENHRMC-GAAG	4.5 ± 0.8 (3)
MGSADGA-upain-2-GAAG	MGSADGA-CSWRGLENHAAC-GAAG	3.5 ± 0.1 (3)
DGA-upain-2	DGA-CSWRGLENHAAC	4.6 ± 0.6 (3)
bicyclic peptide 24	CSWRGCENHAAPA	26 ± 13 (6)
bicyclic peptide 29	CSWRGCENHAAAP	32 ± 14 (6)
upain-1 W3A	CSARGLENHRMC	313 ± 8 (3)

*All synthetic peptides are of an acetylated N-terminus and an amidated C-terminus, and are cyclized through the underlined residues.

The crystal structure of MGSApan class="Chemical">DGAn>-CSWRGLENHAAC-GAAG:hpan class="Gene">uPA complex (PDB code: 3OY5) indicates the core region (CSWRGLENHAAC) has a conformation identical to that of pan class="Gene">upain-1 binding to huPA (Figure 4B). Interestingly, for the extended part, we only observed the electron density of the N-terminus extended tripeptide (Asp-3, Gly-2 and Ala-1, DGA), while all other extended residues were completely disordered. This DGA-tripeptide does not directly interact the enzyme surface but forms intramolecular hydrogen bonds stabilizing the conformation of the core peptide (Figure 4C), consistent with the SPR analysis that the N-terminus extension decreased the dissociation rate (k) but rarely affected the association rate (k). In addition, peptide DGA-CSWRGLENHAAC inhibits uPA with a K value of 4.6 μM, indicating that only the tripeptide DGA rather than other disordered extension residues improve the binding. One possible explanation is that DGA-extension provides a hydrogen-bond network enhancing the peptide's rigidity and causes declined entropic penalty. This is a nice example demonstrating the agreement between functional data and X-ray structures.

In an independent study, Heinis et al. reported a bicyclic peptide hpan class="Gene">uPAn> inhibitor, UK-504, screened from another phage-displayed library 52. UK-504 is cyclized by two pan class="Chemical">disulfide bonds and has a sequence similar to upain-1 (UK-504: CLGRGCENHRCLC, upain-1: CSWRGLENHRMC, the same residues are underlined). Notably, the same residues are the residues forming large contacts with uPA. In addition, crystal structures demonstrate that upain-1 and UK-504 (PDB code: 4GLY) have highly similar conformations when binding to uPA (Figure 4D). UK-504 (K = 7.7 μM) binds to uPA with a stronger affinity than upain-1 (K = 29.9 μM) does, which is likely due to lower binding entropic penalty caused by the constraint of bicyclization. Similarly, UK-504 shows comparable affinity to that of DGA-upain-1 (K = 4.6 μM), which has additional hydrogen-bond network enhancing upain-1's rigidity. Thus, both examples demonstrate the importance the rigidity in the potencies of peptide inhibitors.

Rational design of bicyclic peptidic huPA inhibitors

In the same year when Heinis published UK-504, we also reported bicyclic peptide inhibitors of hpan class="Gene">un class="Chemical">PA based on class="Chemical">n>n class="Gene">upain-2 (CSWRGLENHAAC) 67. With the aim of reducing entropic penalty during the binding event, we cyclized upain-2 through a disulfide bridge and a head-to-tail or side-chain-to-tail amide bond. Considering the head-to-tail or side-chain-to-tail cyclization might influence the overall ring size of the original mono-cyclic peptide, we synthesized these bicyclic peptides with the size ranged 11-13 residues. Accordingly, we synthesized 21 bicyclic peptides and measured the inhibitory potency. Only two peptides, CSWRGCENHAAPA (peptide 24, K = 26 μM) and CSWRGCENHAAAP (peptide 29, K = 32 μM), showed comparable inhibitory potency to that of the mono-cyclic upain-2 (K = 35.5 μM). ITC data showed that bicyclization indeed reduced peptides' binding entropic penalty, which, however, was partially offset by the increased enthalpy. We next solved the crystal structure of peptide 24 in complex with huPA and found and upain-1 bind to uPA in a mostly identical conformation, especially for the residues that directly interact with uPA (Figure 4E). Consistent with the higher enthalpy from ITC analysis, crystal structures also indicated that peptide 24 bound less tightly to uPA, as peptide 24 showed a higher relative B-factor than upain-1 did (1.86 vs. 1.58). Thus, even though our bicyclization decreased entropic penalty, the inhibitory potency was only moderately enhanced due to the less tight binding.

Switch of inhibitors to substrates

Based on our studies of the binding and inhibitory mechanisms of pan class="Gene">upan>in-1 derivatives, we reported the structural details of a pH-dependent proteolytic process of pan class="Gene">serine proteases (Figure 4F). pan class="Gene">Upain-1 is converted to a uPA substrate when Trp3 is mutated to alanine (CSARGLENHRMC).63 We determined the high-resolution crystal structures (1.29-1.68 Å) of upain-1-W3A in complexes with active huPA and inactive huPA (huPA-S195A) at different pH values. At pH 4.6 and 5.5 (PDB code: 3M61 and 4ZKN), the electron density of all residues and peptide bonds were intact, showing the peptide was not cleaved. At pH 7.4 (PDB code: 4ZKO), only Ala3, Arg4, Gly5, and Leu6, (P2, P1, P1', and P2') were observed, and the rest part of the peptide was completely disordered. Notably, at pH 7.4, the P1-P1' peptide bond was hydrolyzed, because the distance between the carbonyl carbon of Arg4 and the amino group of Gly5 was 3.5 Å, much longer than the normal peptide amide bond (approximately 1.3 Å). This was the first report of the P' side residue conformation after proteolysis, suggesting that the P' side product release can be slow in some cases. In this particular case, the slow release was due to cyclic nature of the peptide. For comparison, we crystalized upain-1-W3A in complex with the inactive huPA-S195A at pH 7.4, in which, the peptide was intact without cleavage. At pH 9.0 (PDB code: 4ZKR), the proteolysis was even more thorough, because we only observed Arg4 (P1) in the catalytic cleft of uPA, the rest part of the peptide was disordered. This study reveals a new mechanism of proteolysis by serine proteases.

Mupain-1, a versatile peptidic scaffold for serine protease inhibitors

The phage-displayed screening of murine uPA inhibitors

Based on peptide-display screening, we also isolated a peptide with the sequence of CPAYSRYLDC (Figure 3B), termed mn class="Chemical">pan class="Gene">upain-1, by using the bait of class="Chemical">n>n class="Species">murine uPA (muPA) 68. Like upain-1, mupain-1 is also a mono-cyclic peptide disulfide-bridged by Cys1 and Cys10. Mupain-1 inhibits muPA with a K value of 0.55 μM. The only positive residue, Arg6, was speculated as the P1 residue, because substituting Arg6 with Ala resulted in a >2000-fold declined affinity. Like upain-1, mupain-1 also exhibited excellent specificity over uPAs in different species and other 17 human and murine serine proteases.

Figure 3

Chemical structures of upain-1, mupain-1, L-4-Guanidino-phenylalanine (12), and L-3-(N-Amidino-4-piperidyl)alanine (16).

Chemical modification improving inhibitory potencies

The P1 residues of inhibitors and substpan class="Species">ratn>es, which insert into the S1 specificity pockets of proteases, are considered critical to the binding affinity. The P1 residues of pan class="Gene">upain-1 and mpan class="Gene">upain-1 are Arg4 and Arg6, respectively, which formed strong salt bridges to the carboxyl side chain of Asp189. Mutating P1 arginines to alanine led to the complete loss of measurable affinity to uPAs 63, 68. In the following study, we synthesized 15 non-natural arginine analogues in order to improve peptides' inhibitory potencies 69. Together with naturally occurring L-arginine and L-alanine, we studied the influence of different P1 residues on the binding of upain-1 and mupain-1 to huPA and muPA. The original upain-2 (CSWRGLENHAAC) inhibits huPA with a K value of 35 μM. However, none of the upain-2 derivatives with other 16 P1 residues showed higher or even comparable potency. Similarly, substituting Arg6 of mupain-1 (CPAYSRYLDC) with most of the synthetic P1 residues led to impaired affinities. However, mupain-1 variants with L-4-Guanidino-phenylalanine (12) and L-3-(N-Amidino-4-piperidyl)alanine (16), referred to as mupain-1-12 and mupain-1-16 (Figure 3C and 3D), showed 2-fold (K= 0.28 μM) and 10-fold (K= 0.045 μM) enhanced affinity to muPA, respectively (Table 2). Notably, substitution with 12 or 16 also enhanced the originally non-measurable affinity to huPA (K> 1000 μM) to the affinities with K values in μM range.

Table 2

Sequences and K values of inhibiting uPAs of mupain-1 variants.

Peptide code	Sequence*	Ki values (μM)
		muPA	huPA	huPA H99Y
mupain-1	CPAYSRYLDC	0.55 ± 0.08 (5)	> 1000	15.3 ± 0.2 (3)
mupain-1-12	CPAYS[12]YLDC	0.28 ± 0.02 (3)	93 ± 19 (3)	0.34 ± 0.06 (3)
mupain-1-16	CPAYS[16]YLDC	0.045 ± 0.01 (3)	93 ± 18 (3)	1.86 ± 0.74 (3)
mupain-1-IG	CPAYSRYIGC	0.020 ± 0.004 (3)	38 ± 23 (3)	0.01 ± 0.001 (3)
mupain-1-12-IG	CPAYS[12]YIGC	0.010 ± 0.002 (3)	0.56 ± 0.12 (3)	2.48 ± 0.07 (3)
mupain-1-16-IG	CPAYS[16]YIGC	0.002 ± 0.0005 (3)	0.53 ± 0.07 (3)	0.01 ± 0.002 (3)
mupain-1-D9A	CPAYSRYLAC	0.89 ± 0.02 (3)	> 1000	5.99 ± 0.43 (3)
mupain-1-12-D9A	CPAYS[12]YLAC	0.19 ± 0.01 (3)	11.7 ± 0.36 (3)	0.19 ± 0.04 (3)
mupain-1-16-D9A	CPAYS[16]YLAC	0.076 ± 0.003 (3)	16.9 ± 1.8 (3)	0.31 ± 0.01 (3)

*All synthetic peptides are of an acetylated N-terminus and an amidated C-terminus, and are cyclized through the underlined residues.

Binding mode and inhibitory mechanism of mupain-1 derivatives

To illustpan class="Species">ratn>e the binding and inhibitory mechanisms of mpan class="Gene">upain-1 variants, we combined multiple biophysical and biochemical approaches, including crystallography, site-directed mutagenesis, NMR, ITC, pan class="Gene">SPR, etc 70. For crystallization, we used a murinized huPA-H99Y mutant as a model and found that mupain-1 (PDB code: 4X1Q), mupain-1-12 (PDB code: 4X1R) and mupain-1-16 (PDB code: 4X1N) formed identical conformation on the catalytic cleft of uPA (Figure 5A), even though mupain-1-12 and mupain-1-16 showed obviously higher affinity (Kvalues: 1.21 and 0.84 vs. 14.1 μM according to SPR analysis). Crystal structures showed that peptides occupied the substrate-binding sites of uPA. Unlike the non-canonical substrate-like binding model of upain-1, mupain-1 forms a typical Laskowski conformation 71, 72. The structure also explains why mupain-1 is not cleaved as a substrate (Figure 5B): the hydroxyl side-chain of Ser195 holds a distance of 3.91 Å to the carbonyl carbon of Arg6 (P1), which is unavailable for nucleophilic attack. Besides, the carbonyl oxygen of Arg6 does not correctly dock into the oxyanion hole. SPR analysis showed that the lower K values of mupain-1-12 and mupain-1-16 were more dependent on the lower dissociation speed (k) rather than the higher association speed (k). The ITC data indicated that the higher affinities of mupain-1-12 and mupain-1-16 to muPA were due to the more favorable entropic changes induced by the substitution of non-natural P1 residues.

Figure 5

Structural features of mupain-1 variants binding to relevant proteases. A: the overview of the crystal structure of mupain-1 (cyan sticks), mupain-1-12 (magenta sticks), and mupain-1-16 (green sticks) in complexes with the catalytic domain of huPA-H99Y (white surface), the solvent-exposed 37-, 60-, and 97-loops were colored yellow, green and light purple, respectively; B: Arg6 (P1) showed a disordered non-proteolytic conformation in the catalytic region, key residues of mupain-1 and uPA were presented as cyan and white sticks, respectively. C: construction of the “protease-peptide back-flip library”. Mupain-1 variants with randomized reisudes 8 and 9 were fused in the C-terminus with a linker of two TEV protease cleaved sites. A 6xHis-tag was set for purification in the C-terminus of the fusion protein. D: the alignment of mupain-1-LD (cyan) and mupain-1-IG (yellow). E: Energetic changes of the LD-peptides and IG-peptides binding to uPA. F: the alignment of the backbones of mupain-1-16 (green) and pkalin-3 (brown), the tight β-turns were indicated as dash lines. G: The model for mupain-1 as the scaffold for developing serine protease inhibitors.

Optimizing mupain-1 derivatives by screening from a back-flip library

pan class="Chemical">Alaninen> scanning analysis demonstpan class="Species">rated that pan class="Chemical">Leu8 and Asp9 had very minor contribution to the affinity to muPA. However, crystal structures showed that Leu8 and Asp9 are deeply embedded in the catalytic cleft of uPA. To screen the optimized peptide sequences, we established a “protease-peptide back-flip library” by expressing mupain-1 variants with randomized residues 8 and 9 (CPAYSRYC, indicates the randomized natural residues except cysteine and proline) on the C-terminus of the catalytic domain of muPA (16-243) (Figure 5C) 73. Between muPA and mupain-1, a linker of 14 amino acids containing two tobacco etch virus (TEV) protease cleaved sites were inserted, ensuring the sufficient flexibility for the peptide to flip back and dock into the active site of uPA. Thus, the muPA activity would decline resulting in a higher K value of hydrolyzing chromogenic substrate. We measured the K value of each muPA-peptide fusion protein and compared to that of muPA-mupain-1 fusion protein. Clones with higher K values indicated the higher affinity of the fused peptides. To exclude the influence of the expression yield, clones with higher K values were treated with the TEV protease afterwards. Clones with catalytic activity restored to that of muPA without peptide fusion were selected and sequenced. Among the screened sequences, a synthetic peptide with Ile8 and Gly9 (CPAYSRYIGC, termed mupain-1-IG) showed 0.02 μM K values of inhibiting muPA. To further improve potency, we substituted the P1 Arg6 of mupain-1-IG with non-natural amino acids, 12 and 16, leading to peptides mupain-1-12-IG and mupain-1-16-IG, which inhibited muPA with K values of 0.01 μM and 0.002 μM, respectively (Table 2).

To our knowledge, pan class="Chemical">mupain-1-16n>-IG is the most potent small-molecule inhibitor of pan class="Species">uPAs among different species, only slightly weaker than two mono-clonal antibodies, mU1 and pan class="Gene">mU3.74, 75 Unlike mupain-1-16, which has K values of 0.22 and 0.59 μM of inhibiting murine and human plasma kallikrein (PK), mupain-1-12-IG is much more specific for inhibiting many other serine proteases with K values in mM or sub-mM levels.

Improving peptides' potency and selectivity by increasing the flexibility

In contrast to pan class="Gene">upan>in-1 and many other peptide inhibitors, mupain-1-derived inhibitors are special because of the flexibility-dependent inhibitory potency, which challenges the traditional viewpoints of higher pan class="Disease">rigidity relevant to the higher affinity 73. Inhibitors bind to serine proteases with entropic penalty. Thus, inhibitors with higher rigidity and reduced entropic penalty are favorable. The DGA-extended upain-1 and bicyclic peptides of Heinis are good examples. However, for mupain-1 variants, IG-peptides (peptides with Leu8 and Asp9) showed 20-25-fold enhanced affinity to muPA compared to LD-peptides (peptides with Ile8 and Gly9) with the same P1 residues. Similarly, another study showed that mutating Asp9 to Ala (CPAYS[P1]YLAC) also enhanced mupain-1's affinity to huPA, which conflicts the crystallographic data that mupain-1's carboxyl side-chain of Asp9 formed a salt bridge with the guanidino group of Arg35 in huPA. Both D9A-mutation and IG-substitution increased the flexibility but surprisingly enhanced the potency. ITC analysis demonstrated that both D9A-mutation and IG-substitution increased entropic penalty (TΔS)70, 73, which, however, was compensated by the declined enthalpy changes (ΔH) finally leading to lower Gibbs free energy changes (ΔG) and lower K values. Crystal structure of mupain-1-IG:huPA-H99Y (PDB code: 4ZHL) shows that mupain-1-IG bind to uPA (Figure 5D) with lower mobility (B-factor) than mupain-1 do, which is also consistent with the lower k values determined via SPR analysis. To summarize, all the facts of the lower k values (SPR), lower ΔH (ITC), and lower B-factor (crystallography) indicate that IG-peptides bind more tightly than LD-peptides do. Thus, we rationally conclude a unique feature of mupain-1-based peptides that the binding affinity can be improved by the increase of flexibility, opposite to the facts of upain-1 and Heinis's bicyclic peptides 52-54. We hereby proposed a possible mechanism as described in Figure 5E: peptide and uPA needs to span an energy barrier to form the high-energy intermediate. The association energy (△G) is relevant to the association rate (k) (△G=-RTlnK). The peptide-uPA intermediate then transforms to final inhibitory complex with the release of energy (△G), which is also the energy needed for the complex dissociation (k). The binding energy △G = △G. Before binding, IG-peptides have higher entropy (S) and thus higher Gibbs free energy (G) in solution compared to LD-peptides. Thus, IG-peptides need lower △G to form intermediates indicating the higher kvalues. On the other hand, the IG-peptide-uPA complexes are more stable and have lower energy because the higher flexibility results in higher adaptability. The IG-peptide-uPA complexes need higher energy for dissociation (△G) consistent with the lower k and B-factors. Thus, the lower association energy and higher dissociation energy results in lower △G and lower K values. The flexibility-dependence also explains the fact that the affinity enhancement of IG-substitution is not specific to muPA but applicable to many other proteases, including plasmin, tPA, trypsin, etc.

Reengineering plasma kallikrein inhibitors with the scaffold of mupain-176

One interesting feature of mpan class="Gene">upan>in-1 is the high versatility of engineering inhibitory potency to other homologous enzymes of pan class="Gene">uPA. Mupain-1 (CPAYSRYLDC) is a specific competitive inhibitor of muPA with a K value of 0.55 μM, while it shows no measurable inhibition of many other serine proteases, including plasma kallikrein (PK). However, substituting the P1 residue (Arg6) of mupain-1 with non-natural amino acid, L-3-(N-Amidino-4-piperidyl)alanine (16), resulted in peptide mupain-1-16 (CPAYS[16]YLDC), which not only showed a 10-fold enhanced affinity to muPA, but also inhibited human and murine PK (hPK, mPK) with measurable K values in μM range. We then wondered if it is possible to reengineer mupain-1-16 to a specific PK inhibitor. We superimposed the mupain-1-16 conformation from our crystal structure of mupain-1-16:huPA-H99Y complex (PDB code: 4X1N) to hPK (PDB code: 2AYN), and found that the peptide did not match well with the 37-, 97-, and 215-loops in terms of charge or van der Waals complementarity. To optimize the peptide-PK interactions at these sites, we modified residues 4, 5, 7, and 9 of mupain-1-16, resulting in peptide pkalin-3 (CPARF[16]ALWC), which inhibited hPK with a K value of 0.014 μM and showed high specificity over other serine proteases, including uPA. From mupain-1-16 to pkalin-3, we converted a specific uPA inhibitor to a specific PK inhibitor by just modifying 5 residues. Later, we confirmed our design strategy through crystallization, SPR analysis and site-directed mutagenesis.

The versatility of mupain-1

Small-molecule inhibitors often suffer from the lack of specificity because most of them target the S1 pockets, which are highly conserved among pan class="Gene">serine proteasen>s 42, 56. Peptide-based inhibitors contact the multiple specificity pockets, leading to higher selectivity 57. The catalytic domains of pan class="Gene">serine proteases have common surface-exposed loops around residues 37, 60, 97, 110, 170, and 185, which are highly variable in length and sequence among different proteases 77. These high variations can be identifiable features for the recognition of substpan class="Species">rates, and can be the targets of engineering specificity of inhibitors. Mupain-1-derived inhibitors directly interact with 37-, 60-, and 97-loops (Figure 5A): the side-chains of residue 4 and 5 point to 97-loop, residues 7 and 9 stretch to 37-loop and 60-loop. Thus, residues 4, 5, 7, and 9 are likely to be critical for the selectivity of the peptide. These residues are just the changed non-P1 residues from mupain-1-16 to pkalin-3. However, even though upain-1 also forms large contacts with 37-, 60-, and 97-loops, our efforts on engineering affinity of upain-1 to other proteases failed. The reason of the versatility of mupain-1 is still unclear. One possible explanation could be that upain-1 binds uPA with a highly constrained conformation, mutating residues might lead to unfavorable conformational changes. In contrast, the potencies of mupain-1 derivatives are highly dependent on the high flexibility, which facilitate the backbone to adapt to the new situation when the residues are mutated. For mupain-1 variants, we enhanced the inhibitory potencies to muPA to a K value in the nano-molar range (Table 2) 73. Besides, we also converted mupain-1 to inhibitors of PK and activated coagulation factor XI (fXIa) with K values in the low nM range (unpublished work). In addition, the affinity enhancement by modification of non-natural P1 residues only works in the case of mupain-1 mutants rather than that of upain-1 derivatives. Therefore, mupain-1, in contrast to upain-1, is a more interesting peptide scaffold with high versatility of engineering inhibitory potency to serine proteases.

Reengineering potencies to serine proteases with the scaffold of mupain-1

Here, we propose a principle for developing peptide inhibitors of pan class="Gene">serine proteasen>s based on the scaffold of mupain-1. In crystal structures, all the peptides (mupain-1, mupain-1-12, mupain-1-16, mupain-1-IG, mupain-1-16-IG, mupain-1-D9A, and mupain-1-16-D9A) form mostly identical conformations in complexes with uPA although with highly variable K values. In addition, because of the constraint by two tight β-turns, the backbones of mupain-1 and pkalin-3 form largely similar conformations (Figure 5F) making the side chains of the two peptides stretch to the same sites of PK and uPA. Here, we nominate residues 1-3 and 10 as “solvent-exposed residues” because they have minor contacts with enzyme surface. In contrast, residues 4-9, which are deeply embedded into the catalytic cleft, are termed “enzyme-binding residues”. From mupain-1 (CPAYSRYLDC) to pkalin-3 (CPARF[16]ALWC), all the changed residues (residues 4-7 and 9) are “enzyme-binding residues”, suggesting that engineering the peptide-enzyme interactions should be focused on these “enzyme-binding residues”. Mupain-1 derivatives bind to proteases in a substrate-like conformation, in which residues 4-9 act as P3, P2, P1, P1', P2', and P3', respectively, and interact with the corresponding S and S' pockets of proteases (Figure 5G). Thus, to engineer the affinity, the “enzyme-binding residues” should be selected to match the properties of targeted protease surface, including optimizing the formation of salt bridges, hydrophobic interactions, etc. However, structure-based rational design cannot handle the entropy term very well and thus should be complemented with stochastic approaches, e.g. screening inhibitors from libraries with mupain-1 mutants with randomized residues 4-9. The back-flip peptide-protease fusion library mentioned above is one such successful example.

Conclusion

In this review, we summarized our studies on the pan class="Chemical">mono-cyclic peptiden>s targeting pan class="Gene">serine proteases. upain-1 (CSWRGLENHRMC) and mupain-1 (CPAYSRYLDC) are two peptidic inhibitors of huPA and muPA screened from a phage-displayed peptide library. Based on these two peptides, we generated uPA inhibitors with potency and specificity comparable to mono-clonal antibodies. Furthermore, we revealed the high versatility of mupain-1 as a scaffold for developing inhibitors of other serine proteases. We then generated PK and fXIa inhibitors with K values in low nM levels. Since serine proteases are potent therapeutic targets of many important diseases, in future, in addition to continuously generating inhibitors against more proteases, we will also evaluate the therapeutic efficacies of these inhibitors in corresponding disease animal models.