Synthesis and Structural Characterization of Macrocyclic Plasmin Inhibitors

Two series of macrocyclic plasmin inhibitors with a C‐terminal benzylamine group were synthesized. The substitution of the N‐terminal phenylsulfonyl group of a previously described inhibitor provided two analogues with sub‐nanomolar inhibition constants. Both compounds possess a high selectivity against all other tested trypsin‐like serine proteases. Furthermore, a new approach was used to selectively introduce asymmetric linker segments. Two of these compounds inhibit plasmin with Ki values close to 2 nM. For the first time, four crystal structures of these macrocyclic inhibitors could be determined in complex with a Ser195Ala microplasmin mutant. The macrocyclic core segment of the inhibitors binds to the open active site of plasmin without any steric hindrance. This binding mode is incompatible with other trypsin‐like serine proteases containing a sterically demanding 99‐hairpin loop. The crystal structures obtained experimentally explain the excellent selectivity of this inhibitor type as previously hypothesized.


Introduction
The trypsin-like serine protease plasmin is the key enzyme of fibrinolysis and degrades fibrin clots into soluble fragments. [1] Under physiological conditions, the fibrinolysis is well balanced with the blood coagulation cascade. Together, both systems ensure a normal blood flow in the circulation and prevent the occurrence of thrombosis and bleeding. The activity of plasmin is normally controlled by its endogenous inhibitor α 2 -antiplasmin and the broad-spectrum protease inhibitor α 2 -macroglobulin. In hyperfibrinolytic situations, enhanced plasmin activities can cause bleeding disorders, e. g., in cardiac surgery with cardiopulmonary bypass, organ transplants, or trauma after serious injuries. [2] In sub-Saharan Africa and other low-and middle-income countries, postpartum hemorrhage is still the major cause of maternal mortality, with approximately 86000 annual deaths. [3] In principle, such bleeding complications can be reduced by administering effective plasmin inhibitors. [2,4] Currently, tranexamic acid (TXA) is the most common antifibrinolytic drug. It inhibits the plasmin generation by blocking the lysine binding sites on the kringle domains present in its zymogen plasminogen and the tissue-type plasminogen activator. Hence, TXA is not a direct protease inhibitor and has no effect on activated plasmin. Due to its moderate potency, it is usually given in high doses, which can provoke seizures during cardiac surgery. For many years, the 58 amino acids long natural polypeptide aprotinin from bovine lung (Trasylol®) was clinically used to reduce severe blood loss during cardiac surgery. Aprotinin is a direct plasmin inhibitor with an inhibition constant of 0.5 nM. [5] However, due to increased mortality rates in comparison to TXA treatments, [6][7][8] it was withdrawn from the market in 2008. A few years later, aprotinin was reintroduced with warnings and precautions in several countries, but not in the United States. It was also required to establish a registry for patients treated with aprotinin. [9] In addition, various types of synthetic plasmin inhibitors have been described over the past decades, although only a few of them exhibited a significant inhibitory potency, like the elongated TXA-derived inhibitors from the Okada group [10][11] and several acyclic substrate-analogue benzamidine derivatives such as CU-2010. [12] Meanwhile, the first crystal structures of the zymogen plasminogen [13] and plasmin's protease domain in complex with different types of inhibitors are available, [14][15][16] which enabled a rational design of significantly improved plasmin inhibitors in recent years. A series of 14-mer bicyclic peptides derived from the natural sunflower trypsin inhibitor-1 with inhibition constants around 50 pM together with excellent selectivity for plasmin inhibition have been described. [17] A few years ago, we developed macrocyclic small molecule inhibitors containing a C-terminal 4-amidinobenzylamide residue in the P1 position together with an N-terminal P4 sulfonyl group. These substrate-analogue inhibitors were cyclized between the side chains of a P2 amino acid in lconfiguration and a P3 residue in d-configuration using symmetric linker segments. [18][19] We assumed that this specific macrocyclization provokes a steric clash with the so-called 99hairpin loop present in all trypsin-like serine proteases except plasmin. Indeed, these inhibitors showed an excellent potency against plasmin but not against other tested proteases. Only for trypsin, a considerably reduced but still significant inhibition was retained. A crystal structure of one inhibitor from this series in complex with trypsin has confirmed the predicted clash between its macrocycle and the 99-loop of trypsin. [20] Further optimization of this series generated inhibitor 1, which inhibits plasmin and trypsin with K i values of 0.56 nM and 1.54 μM, respectively ( Figure 1). In this inhibitor, the P1 4-amidinobenzylamide is replaced by a p-xylenediamine (p-Xda) group leading to a further improved selectivity profile while maintaining robust antifibrinolytic activity in plasma. [20] Herein we report the influence of several P4 substitutions on the potency and selectivity of inhibitor 1, the synthesis and introduction of asymmetric linker segments, as well as the first crystal structures of these inhibitors in complex with the active site mutant (Ser195Ala) of the protease domain of plasmin.

Synthesis
The best compound from our former publications based on its overall affinity and selectivity profile was inhibitor 1. [20] All of the four previously prepared p-Xda derivatives possess an identical P3-P2 macrocycle containing a symmetric dipropionylpiperazine linker segment. Among them, an unsubstituted benzenesulfonyl group was identified as the preferred P4 residue. Unsubstituted phenyl rings are problematic since they are prone to metabolic functionalization by cytochrome P450 enzymes such as aromatic ring hydroxylation. [21][22] This issue can be addressed by blocking the metabolization sites with suitable substituents. Furthermore, the substitution of phenyl rings is a common approach to identify even more potent analogues. Therefore, in the first approach, we have prepared a few analogues containing substituted benzenesulfonyl groups (Table 1). Inhibitors 2, 3, 4, and 7 were prepared according to Scheme 1 by coupling of the corresponding sulfonyl chlorides (see Supporting Information) to the previously described cyclic intermediate 12. [20] A second synthesis approach is shown in Scheme 2 for inhibitor 6. The P4-P3 building block 3-methoxy-benzenesulfonyl-dPhe(4-NO 2 )À OH was coupled to the previously described P2-P1 segment 13 [20] to obtain the linear precursor 14. After the reduction of the nitro groups by hydrogenation and incorporation of the symmetric piperazine-dipropionyl linker segment, [19] the cyclic intermediate 16 was obtained. The Tfa protection was removed to produce inhibitor 6. The 3-Cl analogue 3 as well as inhibitor 8, which contains the symmetric piperazine-dibutanoyl linker [23] were prepared by a similar strategy using a Cbz-protection on the p-Xda group and The symmetric piperazine-dipropionyl amide linker segment is colored in blue, the P1 p-xylenediamine (p-Xda) group in red. [20] Scheme 1. Synthesis of inhibitors 2, 4, 5, and 7. a) 1.3 equiv. sulfonyl chloride, 5 equiv. DIPEA in THF, 1 h at 0°C and overnight at room temperature; b) aqueous 1 N LiOH in 1,4-dioxane, 3 hours at room temperature. subsequent reduction of the nitro groups by zinc dust, as described in the Supporting Information (Scheme S1).
A different strategy was applied for inhibitors 9, 10 and 11, which contain asymmetric linker moieties. The synthesis is illustrated in Scheme 3 for inhibitor 11. The singly protected linker segment 17 and its analogues were prepared by solid phase synthesis on 2-chlorotrityl chloride (2-CTC) resin following a previously described strategy. [24] The resin was initially loaded with the appropriate bromoalkylcarboxylic acid followed by a substitution reaction with piperazine and alkylation with benzyl 3-bromopropanoate. Intermediate 17 was then obtained by mild acidic cleavage from the resin. In parallel, Boc-Phe(4-NO 2 )À OH was coupled to the P1 segment 18 [20] using a mixed anhydride procedure without preactivation. [25] The nitro group of intermediate 19 was reduced by hydrogenation, followed by coupling of the linker group 17 to the amine of intermediate 20. Subsequently, the Boc-group was removed to obtain compound 21, which was subsequently coupled with the P4-P3 segment. Catalytic hydrogenation of the obtained nitro derivative 22 enabled the simultaneous removal of the benzyl ester, thereby providing compound 23. After an intramolecular cyclization, inhibitor 11 was obtained through removal of the Tfa group.

Kinetics measurements
All of the prepared P4 substituted derivatives are highly potent plasmin inhibitors with K i values in the low nanomolar or subnanomolar range (Table 1). Non-linear progress curves have been observed for these compounds during kinetic measurements with plasmin, as shown in Figure 2A for inhibitor 5, which reveals a slow-binding behavior and allows the determination of the association and dissociation rate constants k on and k off , respectively. [20,26] The progress curves have been fitted to Equation (1), providing the steady state rates v s and the apparent first order rate constants k obs for each curve. Equation (2) was used for K i calculation ( Figure 2B) and Equation (3) to determine k on from the slope of the linear dependence of k obs from the inhibitor concentration ( Figure 2C). The substitution by chlorine or methoxy in para position of the P4 ring results in a slightly stronger plasmin inhibition when compared with their meta and ortho analogues, although none of these compounds is more potent than the reference inhibitor 1. A reduced potency was determined for derivative 7 containing the sterically demanding 2-naphthylsulfonyl group. This is mainly caused by its strongly reduced association rate, even though its k off value is slightly lower than those of the other compounds. As expected, all inhibitors possess an excellent selectivity against the other tested trypsin-like serine proteases. The inhibition constants for thrombin, aPC, PK, fXa, and fXIa are > 15 μM (Table 1). Only for trypsin, a more pronounced inhibitory potency with K i values between 1 and 10 μM was found. However, except for inhibitor 4, these inhibition constants are more than 1000-fold increased when compared with the values for plasmin. The highest selectivity Scheme 2. Synthesis of inhibitor 6. a) 1.0 equiv. 3-methoxybenzenesulfonyl-dPhe(4-NO 2 )À OH, 1.2 equiv. HBTU, 3.0 equiv. DIPEA in acetonitrile, 15 min at 0°C and 2 h at rt; b) H 2 and Pd/C in 90 % aqueous AcOH, rt overnight; c) 1.0 equiv. piperazine-dipropanoic acid, [19] 2.4 equiv. HATU and 5.4 equiv. DIPEA, overnight at rt; d) 1 N NaOH solution in 1,4-dioxane, 2 h at rt. The k obs values were fitted as a function of the inhibitor concentrations according to Equation (3), providing a k on value of 5.39 · 10 5 M À 1 s À 1 . The k off value of 4.58 · 10 À 4 s À 1 was calculated from Equation (4). factor against trypsin was found for inhibitor 5, which is a 3882fold stronger plasmin inhibitor.
Compared to inhibitor 1, the number of ring atoms in the P2-P3 linker moiety was varied in the second series providing analogues 8-11 (Table 2). A considerably reduced plasmin inhibition lacking the slow-binding behavior was observed for inhibitor 8, which contains an elongated symmetric piperazinedibutyryl amide linker. In contrast, inhibitors 9-11 are the first analogues from this inhibitor type possessing asymmetric linker segments. A potent plasmin inhibition with K i values of 1.95 nM and 1.83 nM was retained for inhibitors 9 and 11, respectively. Both compounds exhibit a similar excellent selectivity profile as described above for the P4 substituted compounds. A nearly 10-fold reduced potency was determined for inhibitor 10, which contains a reversely incorporated linker segment compared with inhibitor 11. Although the slow-binding behavior of inhibitor 10 was less pronounced in the progress curves, it was still possible to determine its k on and k off rate constants. Compared to all compounds summarized in Table 1, the k off values of inhibitors 9-11 are enhanced to values > 10 À 3 s À 1 indicating a reduced stability of the enzyme inhibitor complex.

Crystal structures
The used microplasmin (μ-plasmin) consists of the serine protease domain (residues 540-791) and contains an active site mutation Ser195(741)Ala (the first number corresponds to the chymotrypsinogen numbering and the second number (in parenthesis) always to the full-length plasminogen numbering). This μ-plasmin was crystallized in complex with the reference inhibitor 1 and its chloro-substituted inhibitor 3, both containing the symmetric piperazine-dipropionyl amide linker, and also with inhibitors 9 and 10 possessing asymmetric linkers. In all cases, two monomer complexes (units A and B) were found per asymmetric crystallization unit. All four inhibitors reveal a similar overall binding mode ( Figure 4), with the exception of inhibitor 10, which also shows a significantly lower activity than the other inhibitors with a K i of 16.81 � 1.11 nM. Here, the amide bond on the P2 side chain of inhibitor 10 has rotated 180 degree compared to inhibitors 1, 3, and 9. In all four Table 2. Inhibitory potencies for analogues with modified size of the macrocycle. For plasmin, also k on and k off values are provided, as well as the selectivity index [SI] for trypsin (K i trypsin /K i plasmin ).
No. n n'  complexes, the polar contacts in the S1 pocket and of the inhibitor backbone are nearly identical as previously described for a P1 benzamidine derived macrocyclic inhibitor in complex with trypsin (PDB: 5EG4). [20] The binding mode of inhibitor 1 in both monomers of the asymmetric unit is identical and well-defined by the electron density ( Figure S1). The p-Xda amino group interacts with Asp189(735) oxygen atoms at the bottom of the S1 pocket, the side chain OH and carbonyl oxygen of Ser190(736), as well as with the carbonyls of Trp215(761) and Val227(773) via a conserved bridging water molecule ( Figure 5A). This water has been found in most crystal structures of trypsin-like serine proteases in complex with substrate-analogue inhibitors containing a benzamidine function at the P1 position. The P1 amide nitrogen binds to the carbonyl of Ser214(760) and the P3 backbone forms the typical antiparallel β-sheet interaction with Gly216(762), while one of the sulfonyl oxygens binds to the NH of Gly219(764) ( Figure 5A).
The macrocycle adopts a compact conformation in the open S2-S3/4 subsites of μ-plasmin and is stabilized by numerous intra and intermolecular interactions. These include the edge-to-face contact between the P2 and P3 phenyl rings ( Figure 5B). Furthermore, both rings are involved in face-to-face interactions with either the imidazole of His57(603) or the indole of Trp215(761) ( Figure 5A). However, the orientation of the imidazole ring of His57 (603) is different compared to other plasmin structures, e. g. 5UGG.pdb, [15] where it forms the Hbond with the side chain of the active-site serine, which is missing in the alanine mutant. The amide groups on the P2 and P3 side chains as well as the connecting piperazine linker segment are surrounded by a complex network of water molecules ( Figure 5B). These six water molecules mediate intramolecular interactions between the side chain amides and the piperazine group of the inhibitor, and furthermore, they contribute to intermolecular contacts with plasmin residues Asp102(646), Arg175(719), Gln177(721), Glu180(724), and Ser214(760). All these interactions stabilize the plasmin-bound The μ-plasmin from the complex with inhibitor 1 is shown as white transparent surface with the residues of the catalytic triad (Ser195 is mutated to Ala) given as green sticks and labeled with both chymotrypsinogen and plasminogen numberings. The carbon atoms of inhibitor 1, 3, 9, and 10 are colored in wheat, cyan, gray, and pink, respectively. conformation of the macrocycle. The P4 phenysulfonyl group occupies a shallow subpocket above the disulfide bridge between residues Cys191(737) and Cys220(765) (cf. Figure 6) leading to a horseshoe-like backbone conformation of this inhibitor type, similar to several other structures of trypsin-like serine proteases reported. [28] In the case of inhibitor 3, the N-terminal phenylsulfonyl group adopts different conformations ( Figure 6). In monomer A, the 3-chloro substituent is solvent exposed, whereas in monomer B the phenyl ring is rotated by~180 degrees. Here, the 3-chloro substituent faces the protein surface and forms a halogen-bonding interaction with the sulfur atom of Cys220(765). [29] The hydrogen bonds formed by the six water molecules around the linker segment and the P2 and P3 side chain amides are virtually identical as described above for the complex with inhibitor 1.
Despite the methylene insertion between the P3 side chain amide and the piperazinyl residue (n = 3 in Table 2), inhibitor 9 adopts the same binding mode to μ-plasmin in both monomers in the crystal structure, as found for inhibitors 1 and 3. Furthermore, the placement of the linker segment including the surrounding water network is very similar as described above for these complexes. Notably, the side chain of His57(603) from the catalytic triad is rotated away from Asp102(646), thereby disrupting the typical H-bond normally found between these two residues ( Figure 7). Instead, a water molecule is placed close to the usual position of the imidazole nitrogen in this complex, which mediates a polar contact between the carbonyl of Ser214(760) and the side chain of Asp102(646).
In case of analogue 10, a shorter acetamide group is attached to the P3 side chain (n = 1 in Table 2). Although this leads to a 10-fold reduced plasmin inhibition, the electron density of the whole inhibitor structure is well-defined and the compound adopts an identical conformation in both complexes of the crystallographic unit ( Figure S1). The amide group on the P3 ring resembles the orientation found for the other inhibitors and its carbonyl oxygen is bound to the guanidine of  Arg175(719) via a bridging water molecule. In contrast, the amide on the P2 side chain is rotated by approximately 180 degrees and its carbonyl group is pointing toward the enzyme surface. It is connected to residues Asp102(646), Gln177(721), Thr179(723), Glu180(724), and Ser214(760) via connecting water molecules, respectively. This leads to a significant shift of the piperazine moiety toward the protein surface and to an altered water network around the linker segment of inhibitor 10, when compared with the other structures. Whereas in the inhibitor 9 complex, most water molecules of this network involved in contacts to the residues of the μ-plasmin segment Arg175(719)À ValÀ GlnÀ SerÀ ThrÀ Glu180(724) are localized below the piperazine moiety, this is impossible in the complex with inhibitor 10. Here, the shift of the piperazine group closer to the protein surface is preventing the placement of water molecules between the piperazine ring and the protein surface. Instead, the bridging water molecules involved in contacts to the carbonyl of Arg175(719) and the side chain of Gln177(721) are rather placed above the piperazine ring ( Figure 7). We assume that these differences in the arrangement of the water molecules around the linker segment together with the rotated amide bond at the P2 side chain are responsible for the reduced inhibitory potency of compound 10. Notably, in the inhibitor 10 complex the imidazole ring of His57(603) is placed in its usual position making polar contacts to the side chain of Asp102(646).

Discussion
In this study, two new series of plasmin inhibitors have been prepared for future efficacy studies in vivo. An approximately threefold reduced inhibitory potency against plasmin was obtained for the meta-and ortho-substituted P4 analogues in comparison to the reference compound 1. Initially we assumed that these substituents are directed upwards into the solvent lacking beneficial contacts with plasmin. Indeed, this was found for the complex of inhibitor 3 in unit A. Surprisingly, an opposite orientation of the P4-phenyl ring was observed in the complex of unit B, where the chlorine points towards the plasmin surface and interacts with a sulfur atom of a conserved disulfide bridge found in all trypsin-like serine proteases. The substitution of the N-terminal benzenesulfonyl group in para position leads to several plasmin inhibitors with inhibition constants close to 1 nM and excellent selectivity profiles. The high potency inhibitors 2 and 5 possess K i values of 0.86 and 0.85 nM, respectively, which are comparable with inhibitor 1. Importantly, a halogen substitution at the P4 group like in inhibitor 2 could be beneficial to reduce a potential metabolism at the phenyl groups, which are sometimes prone to oxidation. [30] So far, we have not yet performed any metabolic studies with these benzylamine-derived inhibitors to verify this hypothesis. Our previous studies with analogous P1 benzamidine-derived plasmin inhibitors showed a negligible inhibition of the five tested CYP-450 isoenzymes. [19] Otherwise, in metabolism studies with the benzylamine containing fXa inhibitor DPC-423 in rodents, numerous metabolites and conjugates at the benzylamine group were identified, which might raise some concern for its use in drug development. [31][32][33] However, a structurally very similar inhibitor of plasma kallikrein was recently approved for the oral treatment of hereditary angioedema. [34] Notably, berotralstat contains an identical benzylamine group attached to a similar substituted pyrazole ring as previously used in DPC-423. This suggests that a benzylamine could be a useful P1 residue to address trypsin-like serine proteases, especially for drugs which are used for shortterm or single treatments which would be the case for plasmin inhibitors to reduce bleeding in certain surgeries or severe trauma.
Regarding the derivatives with asymmetric linker moieties, the more potent inhibitors 9 and 11 possess K i values of 1.95 nM and 1.83 nM, respectively. Despite a reduced activity in comparison to inhibitor 1, their inhibitory potency is still in the same range as previously described for the acyclic inhibitor CU-2010 (K i = 2.2 nM), [12] which had reached clinical phase II development. In spite of the relatively high plasminogen levels between 1.1 and 2 μM in the circulation, clinical studies with different dosages of CU-2010 showed that an inhibitor concentration of approximately 0.5 μM in blood is sufficient to achieve the intended antifibrinolytic efficacy in vivo. [35] Since this concentration would be approximately 250-fold higher than a K i value of 2 nM, an effective inhibition of the activated plasmin in the circulation could be expected. Since these compounds are also very selective plasmin inhibitors, their effective concentration is rather not expected to be affected by targeting other targets. Furthermore, Copeland stated in numerous publications that the pharmacological efficacy of a drug is not solely dependent on the binding affinity per se, which is usually described by the equilibrium constants K i or K d , but also from the stability of the formed binary drug-target complex indicated by its lifetime or residence time. [36][37] Although the residence time is also influenced by the association rate constant k on , it mainly depends on the first-order dissociation rate constant k off and approximately corresponds to the half-life time of the complex (τ = ln2/k off or even further simplified as τ = 1/k off ). [36] Due to the observed slow-binding mechanism the individual rate constants for most of our inhibitors could be determined and allowed the calculation of their residence time. The most potent inhibitors shown in Table 1 contain the same symmetrical linker segment that was used for the reference inhibitor 1 and possess k off values close to 4 · 10 À 4 s À 1 , thereby corresponding to a half-life of approximately 30 min for the ligand-target complex. Interestingly, considerably faster k off values and therefore reduced residence times (approximately 10 min in case of k off = 1.2 · 10 À 3 s À 1 ) were determined for inhibitors 9 and 11, which contain asymmetrical linker segments. This tendency was even more pronounced for the less potent inhibitor 10, for which the highest k off value was determined. Together with the inhibition constants, these data suggest to maintain the preferred symmetric piperazinedipropionamide linker segment in this type of macrocyclic plasmin inhibitors.
The determined crystal structures confirmed the postulated overall binding mode of our macrocyclic plasmin inhibitors, although in our previous publications the molecular modelling was performed in absence of water molecules. [18][19] As found for many other trypsin-like serine proteases in complex with substrate-analogue inhibitors possessing a sulfonyl group in P4 position, the backbone interactions are completely conserved. The open active site of plasmin is lacking the 99-loop, which allows an efficient binding of the linker segment to various residues of the protease via bridging water molecules ( Figures 5  and 7). Besides plasmin, all other trypsin-like serine proteases contain a sterically demanding 99-loop provoking a steric clash with the relatively rigid macrocycle. This becomes obvious when comparing the binding mode of inhibitor 1 in the mutated μ-plasmin superimposed with the previously determined complex of a structurally related benzamidine-derived inhibitor (#31 in reference [20] ) in trypsin (Figure 8). The drastic change in the conformation of its macrocycle explains the significantly reduced inhibitory potency in trypsin and for all other trypsin-like serine proteases possessing the 99-loop. This leads to the excellent selectivity profile of our cyclic plasmin inhibitors.
Despite their differences in the macrocycle, all four crystallized inhibitors possess a similar overall binding mode. The water network around the piperazine-dicarbonic acid amide linker of inhibitors 1, 3, and 9 is nearly identical in all three complexes and both structures of each unit cell. Most likely, this complex water network contributes to the enhanced potencies of these inhibitors when compared with the less affine analogue 10, whose linker segment is surrounded by com-pletely differently arranged water molecules. This is probably caused by the reverse arrangement of the amide group on the P2 side chain of inhibitor 10.

Conclusion
In summary, two series of highly active and selective macrocyclic plasmin inhibitors have been synthesized. Substitutions at the N-terminal P4-phenylsulfonyl moiety in para position were well tolerated but could not improve the plasmin inhibition. Such modifications could be useful to reduce the metabolism of the inhibitors, which often occurs at unsubstituted phenyl rings. For the first time, asymmetric linker moieties were selectively incorporated to further modify the size of the macrocycle. Thereby, it was shown that the symmetric piperazine-dipropionyl amide moiety used in the reference inhibitor 1 is superior to the asymmetric linker segments. Compared with aprotinin, a stronger antifibrinolytic activity was determined in human plasma for inhibitors 5 and 11, the best compounds of both series. Furthermore, the first crystal structures of these macrocyclic inhibitors in complex with mutated μ-plasmin have been determined. These structures prove their previously postulated binding mode and explain the excellent selectivity as plasmin inhibitors. In vivo studies will be important to determine how different efficacies and half-lives of the benzylamine-derived inhibitors perform in disease models.  [20] From trypsin, only the backbone of its 99 hairpin loop (residues 94-101) is shown as orange cartoon, which would induce a steric clash with the binding mode of inhibitor 1 in μplasmin. In the trypsin complex, the 99 loop pushes the macrocycle of inhibitor #31 away, thereby preventing its P3 residue from binding into the distal S3/4 binding pocket above Trp215(761), which is usually addressed by P3 residues in d-configuration.

Experimental Section
General Amino acid derivatives, reagents and solvents were purchased from Acros Organics, Alfa Aesar, Bachem, BLDpharm, Carbolution, Fisher Scientific, Fluorochem, Iris Biotech, Merck KGaA or Roth and used without further purification. Analytical HPLC was performed on a Hitachi Primaide system (Hitachi Europe GmbH, Düsseldorf, Germany) consisting of a Primaide 1110 pump, an 1210 auto injector, a 1430 diode array detector, and a 1310 column oven using a NUCLEODUR C 18 column ec, 5 μm, 100 Å, 4.6 mm × 250 mm (Macherey-Nagel, Düren, Germany). Water (A) and acetonitrile (B), both containing 0.1 % TFA, were used as eluents with a linear gradient (increase of 1 % B/min) at a flow rate of 1 mL/min (detection at 220 nm). Intermediates and inhibitors were purified using a Varian preparative HPLC system (pumps: Varian PrepStar model 218 gradient system; detector: ProStar model 320; fraction collector: Varian model 701; column: NUCLEODUR C 8 ec, 5 μm, 100 Å, 32 mm × 250 mm, Macherey-Nagel, Düren, Germany) at a flow rate of 20 mL/min and a linear gradient with an increase of 0.5 % B/min using identical solvents as described for the analytical HPLC (detection at 220 nM). All final inhibitors were purified to more than 95 % (based on detection at 220 nm) and obtained as lyophilized TFA salts. The masses of intermediates and inhibitors were determined using a QTrap 2000 ESI mass spectrometer (Life Technologies GmbH (Darmstadt, Germany). 1 H-and 13 C-NMR spectra were measured in DMSO-d 6 on two different spectrometers from Jeol: 1. ECX400, frequencies: 1 H: 400 MHz, 13 C: 101 MHz; 2. ECA500, frequencies: 1 H: 500 MHz, 13 C: 126 MHz. The DMSO-D 6 signal was used as reference ( 1 H-NMR: 2.50 ppm, 13 C-NMR: 39.52 ppm). [38] Chemical shifts δ are given in the unit ppm, coupling constants J in the unit Hz. Multiplicities are indicated as s (singlet), d (doublet), dd (doublet of doublet), ddd (doublet of doublet of doublet), t (triplet), td (triplet of doublet), q (quartet) or m (multiplet). Overlapping signals that produced pseudo multiplicities are highlighted with the prefix p (e. g. pd = pseudo doublet). The prefix b indicates broad signals with insufficient resolutions for multiplicity assignment. The water content as well as the solvent itself produced signals that could overlap with those of the analyte. Furthermore, piperazine moieties as well as hydrogen atoms within the macrocycle of cyclic derivatives produced broad, overlapping signals impeding analysis, whereas the region between approx. 4.5 and 2 ppm was strongly affected. If a distinct assignment was not reasonable for all protons, only clearly distinguishable signals are specified. Affected signals that are overlapping with broad mixedsignals are listed without integrals.

methyl)benzyl)amino)propan-2-yl)carbamate (20)
Compound 19 (3.70 g, 7 mmol) was hydrogenated analogously to derivative 15 using 370 mg of 10 % Pd/C and 250 mL of 90 % aqueous acetic acid. After removal of the catalyst by filtration, the residue was dissolved in a mixture of ethyl acetate and saturated aqueous NaHCO 3 solution. The organic layer was washed three times with saturated aqueous NaHCO 3 solution and once with brine, dried over anhydrous MgSO 4

Enzyme kinetics
All measurements were conducted with triplicates from three independently weighted samples using black 96-well plates and fluorogenic substrates.
For evaluation of the inhibitory activity, the steady state velocity v s of each measurement was determined. In case of linear progress curves (classical behavior), v s was calculated from the slope of the curves. If a slow-binding behavior was observed, v s was determined by fitting the progress curves over 20 minutes to Equation (1) (v 0 : initial velocity in absence of inhibitor; k obs : apparent first-order rate constant; d: displacement of the fluorescence signal from zero at t = 0). In all cases, the obtained v s values were plotted against the inhibitor concentration for K i determination with Equation (2) using V max and K m values from Michaelis-Menten plots determined on the same plate. For slow-binding inhibition, Equation (3) was used to calculate k on from plots of k obs values as function of the inhibitor concentration. The dissociation rate constant k off was calculated according to Equation (4) using the values for K i and k on determined before.

Fibrinolysis assay
Measurements have been performed as described previously [20,27] with minor modifications over a period of 30 min at 37°C in transparent 96-well-plates (Brand, Wertheim, Germany) using a Tecan Spark® microplate reader (Tecan Group AG, Männedorf, Switzerland) at 405 nm. Inhibitor stock solutions (10 mM in DMSO) were further diluted with 0.9 % aq NaCl. The reference inhibitor aprotinin (Trasylol® with 500000 KIE/50 mL corresponding to a concentration of 1.4 mg/mL or 215 μM) was obtained from Nordic Pharma GmbH (Ismaning, Germany) and dissolved and further diluted in 0.9 % aq NaCl without DMSO. Tissue factor (Dade® Innovin®-reagent, Siemens Healthcare Diagnostics, Eschborn) was dissolved in 10 mL of water, the stock solution was stored at 4°C and further diluted (1/100, v/v) using 0.9 % aq NaCl directly before the measurements. Recombinant tPA preparation (Actilyse®, Boehringer Ingelheim, Biberach, Deutschland) was dissolved in water at a concentration of 1 mg/mL and was further diluted (1/2.5, v/v) with 0.9 % NaCl directly before the measurements. Moreover, a 125 mM CaCl 2 solution was used. The total assay volume was 200 μL consisting of 20 μL inhibitor, 20 μL CaCl 2 (12.5 mM in assay), 20 μL Actilyse solution, 40 μL Innovin reagent (stock 1/500 diluted in assay), and 100 μL citrated human plasma (German Red Cross, Kassel, Germany). Mixtures of inhibitor, CaCl 2 , Actilyse, and Innovin reagent were preincubated at 37°C for 10 min in the plate reader. The measurement was started by addition of plasma, which was preincubated at 37°C in a water bath.

Crystallography
Protein preparation. The serine protease domain of human plasminogen (μ-plasminogen) was expressed as the active-site mutant Ser195(741)Ala, activated by tPA into μ-plasmin and purified as previously described. [17] Method of crystallization. The purified μ-plasmin mutant was concentrated to 10 mg/ml and mixed with excess inhibitors at 1 : 1.5 molar ratio for crystallization experiments. The crystal trays were set up by hanging drop vapor phase diffusion method at 20°C, and crystals were obtained within a few weeks in the presence of 13-18 % PEG 4000, 150 mM ammonium sulfate and 100 mM 2-(N-morpholino)ethanesulfonic acid (pH 4.5-5.5).
Structure determination. Single crystals of the Ser195(741)Ala μplasmin/inhibitor complexes were frozen in liquid nitrogen with 15 % glycerol as the cryoprotectant, and diffracted at the Australian Synchrotron using MX1 or MX2 beamlines. Dataset were processed using XDS, [41] and the complex structures were built by molecular replacement in COOT [42] using the PDB ID 5UGG as the searching model for μ-plasmin. The macrocyclic inhibitors were generated by eLBOW and fit into the electron density map using LigandFit in Phenix. [43] The final structures were further refined by Phenix and Coot. The data collection and refinement statistics are summarized in Table S1. The structures were deposited in the Protein Data Bank with accession codes 8F7V (1), 7UAH (3), 7THS (9), and 8F7U (10). Figures were generated using PyMOL. [44] Abbreviations

Supporting Information
The synthesis and analytical data of additional intermediates and inhibitors and a table on data collection and refinement statistics are provided in the Supporting Information.