Synthesis and Structure‐Activity Relationships of N‐(4‐Benzamidino)‐Oxazolidinones: Potent and Selective Inhibitors of Kallikrein‐Related Peptidase 6

Abstract Kallikrein‐related peptidase 6 (KLK6) is a secreted serine protease that belongs to the family of tissue kallikreins. Aberrant expression of KLK6 has been found in different cancers and neurodegenerative diseases, and KLK6 is currently studied as a potential target in these pathologies. We report a novel series of KLK6 inhibitors discovered in a high‐throughput screen within the European Lead Factory program. Structure‐guided design based on docking studies enabled rapid progression of a hit cluster to inhibitors with improved potency, selectivity and pharmacokinetic properties. In particular, inhibitors 32 ((5R)‐3‐(4‐carbamimidoylphenyl)‐N‐((S)‐1‐(naphthalen‐1‐yl)propyl)‐2‐oxooxazolidine‐5‐carboxamide) and 34 ((5R)‐3‐(6‐carbamimidoylpyridin‐3‐yl)‐N‐((1S)‐1‐(naphthalen‐1‐yl)propyl)‐2‐oxooxazolidine‐5‐carboxamide) have single‐digit nanomolar potency against KLK6, with over 25‐fold and 100‐fold selectivities against the closely related enzyme trypsin, respectively. The most potent compound, 32, effectively reduces KLK6‐dependent invasion of HCT116 cells. The high potency in combination with good solubility and low clearance of 32 make it a good chemical probe for KLK6 target validation in vitro and potentially in vivo.


Introduction
Kallikrein-related peptidase 6 (KLK6), previously known as protease M, zyme, neurosin, or myelencephalon specific protease, [1] is a secreted serine protease that belongs to the family of tissue kallikreins (KLKs). [2] Like all 15 KLKs, KLK6 is released into the extracellular matrix as a zymogen and activated upon cleavage of a pro-peptide, a process which can be mediated by other proteases such as KLK5, [3] plasmin, [4] urokinase (uPA), [4] and MMP-20. [5] Removal of the pro-peptide generates mature KLK6, a trypsin-like enzyme with cleavage specificity after basic P1 residues, preferably arginine. Broader sequence requirements have been reported for the flanking residues (P2, P3, P1', and P2'). [6] Relevant endogenous substrates of KLK6 have been identified in vitro and include proteaseactivated receptors (PARs), [7] α-synuclein, [7][8] and myelin basic protein. [9] Secreted proteases (e. g. KLKs and matrix-metalloproteinases) are investigated as potential therapeutic drug targets due to their role in extracellular signaling via proteolysismediated production of small signaling molecules or proteolytic activation of membrane receptors. [10] KLK6 can activate PARs, and this signaling pathway has been found to be dysregulated in cutaneous malignant melanoma. [11] In this cancer, KLK6 was found to be secreted by the keratinocytes surrounding the tumor cells in response to stimuli from the tumor, and to act in a paracrine fashion to activate PAR-1 receptors, which are overexpressed on melanocytes. This signaling cascade was found to have an effect on tumor migration and invasiveness in vitro [11] and is considered to contribute to recurrence and metastasis in melanoma patients that undergo surgery. [12] KLK6promoted migration was also observed in colon cancer, where knockdown of KLK6 reduced migration and invasion of HCT116 cells in vitro. [13] Furthermore, in an orthotopic colon cancer mouse model, mice injected with KLK6 positive HCT116 cells had significantly more metastases and worse survival than mice injected with shKLK6 HCT116 clones. [13] Nevertheless, the role of KLK6 needs further investigation in these and other types of cancers, as its role is clearly tumor-dependent. In head-andneck cancer for example, high levels of KLK6 have been associated with a better prognosis for the patients, resulting in reduced aggressiveness of the disease. [14] In addition to malignancies, KLK6 is studied in the central nervous system (CNS), where its physiological abundance might imply an important role for KLK6 in the maintenance of homeostasis in these organs. In neurodegenerative diseases such as multiple sclerosis, Alzheimer's [15] and Parkinson's, [16] as well as spinal cord injury, [17] aberrant levels of KLK6 have been reported. Potential therapeutic approaches targeting KLK6 have been investigated but require further validation, particularly with high-quality KLK6 chemical probes. [9] To date, few accounts of reversible KLK6 inhibitors have been reported, the most relevant being two sets of small molecules (e. g. 1 and 2, Figure 1) discovered by in silico highthroughput screening (HTS) supported by X-ray crystallography, [18] and a series of pseudopeptides (e. g. 3), which were reported to be highly selective over the closely related KLK5. [19] Covalent coumarin-based suicide inhibitors (e. g. 4) [20] as well as transient quiescent affinity labelers (e. g. DKFZ-251), our first disclosure of KLK6 inhibitors, [21] have also been reported.
Given the growing interest in KLK6 as a drug target and the potential benefit of being able to control its enzymatic activity to validate current biological hypotheses, we set out to find a novel series of selective reversible KLK6 inhibitors.

High-Throughput Screening and Validation of a Promising Hit
Using our previously published KLK6 assay format, [21] we performed a HTS with the European Lead Factory (ELF), [22] and tested~350,000 substances at 10 μM for their ability to reduce KLK6-catalyzed hydrolysis of the fluorogenic Boc-Phe-Ser-Arg-AMC peptide. About 8,000 actives were identified and re-tested. After applying a threshold cutoff of 25 % inhibition, and discarding compounds with inherent high fluorescence (> 3 times the background), 1026 compounds were selected for dose-response curve analysis ( Figure 2). These substances were tested at seven concentrations from 20 μM to 20 nM, resulting in 312 entities with a pIC 50 > 4.7. A selectivity screen against trypsin, thrombin and factor Xa, as previously reported, [21] reduced the number to 226 preliminary hits. Further validation of these hits was performed via surface plasmon resonance (SPR), resulting in 61 likely reversible KLK6 binders. LC-MS analysis eliminated 4 impure substances, and a qualified hit list (QHL) of 50 compounds was generated. These were sorted into ten structural clusters with fifteen singletons. Many of the hits contained a highly basic moiety such as amidine or guanidine, functional groups that are often found in trypsin-like serine protease inhibitors and sometimes associated with poor cellular permeability. Nevertheless, we chose to further advance with such compounds because KLK6 is a secreted protease, mindful that permeability might need to be considered at an early stage.
Compounds containing an oxazolidinone benzamidine were the largest and most promising cluster with nine members. All of the compounds in the cluster were potent enough to generate a pIC 50 in the primary dose-response assay (Table 1). Each substance also gave pK D values in the SPR assay which

Full Papers
were similar to the pIC 50 values. Furthermore, hits from this cluster exhibited on/off binding kinetics in the SPR assay consistent with a reversible mechanism of action (data not shown). In addition, hints of structure-activity relationships (SAR) could be gleaned. For example, the most potent hit (5) is the only compound in the cluster with an amide directly linked to the oxazolidinone. Compounds with an amide extended by one carbon (6) or containing other functional groups (7 and 8) are significantly less active. In addition, 5 had by far the best selectivity profile relative to trypsin, thrombin and factor Xa (vide infra).

Determination of the Active Stereochemical Series
Hit 5 was resynthesized along with all three of its stereoisomers 9-11. The activity of 5 against KLK6 was confirmed, although a consistently lower pIC 50 (6.6 vs. 7.2) was found relative to the same substance in the HTS library (Table 2). Encouragingly, the other three stereoisomers showed poor activity against KLK6, suggesting only the (R,S) configuration of the scaffold allows for productive molecular interactions between the inhibitor and KLK6.

Docking Study Predicts Key Binding Interactions
Compound 5 was docked into an X-ray crystal structure of KLK6 (PDB ID 4D8N), allowing us to model the main interactions with the target ( Figure 3A). As expected, the highest scoring poses predict that the amidine group makes critical hydrogen bonds (H-bonds) with Asp189 and Ser190 in the S1 pocket and is likely responsible for a significant amount of the binding energy. A secondary H-bond network is also formed by the amide: the carbonyl of the amide interacts with the backbone of Gly193 in the anionic hole and the amide NH interacts with His57 of the catalytic triad. The more solvent exposed S1'/S2' regions are occupied by the lipophilic α-methylbenzyl amine substituent. Close inspection suggested that binding could be increased through more effective filling of the S1'/S2' pockets lined by Leu40, Leu41, and Phe151. Comparison of our KLK6 model and X-ray structures of trypsin, [23] KLK4, [24] and KLK8 [25] suggested that by exploiting subtle differences between the enzymes' S1' and S2' pockets, we could modulate selectivity for KLK6. For example, the S1' pocket appears more compact in KLK6 with Lys60, Leu41, and the Cys42/Cys58 disulfide bond forming a tight lipophilic space. In trypsin, the S1' pocket features Lys60, Phe41 and Cys42/58; however, the Lys60 side chain appears pulled back by an intra-residue H-bond with Tyr39, creating a more open pocket which may prefer to accommodate larger groups ( Figure 3B). We also predicted that it could be possible to improve binding by adding an additional peptide bond to the inhibitor scaffold, thereby adding an H-bond to the Leu41 backbone carbonyl and the Gln192 side chain ( Figure 3C). On the basis of our modeling, initial medicinal chemistry efforts were invested into increasing affinity toward the S1'/S2' pockets, which showed the highest potential for exploration and structural expansion. Furthermore, the model was tested by synthesizing and testing substances which lacked specific features that were predicted to be key for binding.

Synthesis of N-Benzamidine-Oxazolidinone Derivatives
Compounds for this study were synthesized as depicted in Scheme 1A, beginning with conversion of 4-aminobenzonitrile (12) to the corresponding benzyl carbamate 14.
The key enantiomerically enriched oxazolidinone 17 was prepared by [a] pIC 50 measured in the enzymatic inhibition assay. [b] pK D measured in the SPR binding assay.   chromatography. To obtain non-commercial enantioenriched chiral amines 39a-39c for the above amide coupling, we utilized Ellman's t-butylsulfinamide chemistry (Scheme 1B). [26] Conversion of the cyano group in 21a-21m to an amidine was performed in a one-pot, three-step process: addition of hydroxylamine, yielding an amidoxime, which was acetylated and then reduced with zinc dust or with Pd/C and H 2 to give 5, 11, and 23-33. Compounds 9 and 10 (not depicted in Scheme 1) were synthesized in analogy to 5 and 11, but using (S)-glycidyl butyrate ((S)-16) in the second step. Compound 34 was prepared similarly, starting with 2-cyano-5-aminopyridine (13) instead of 12. Benzyl amine 35 was prepared from the corresponding nitrile via hydrogenation with Raney nickel. The synthesis of three additional substances is depicted in Scheme 2. In the first example, 12 and 2-methylenesuccinic acid (40) were combined to make lactam 41 as a racemate, which was advanced to 43 via amide coupling with 42 followed by amidine formation. Amide 43 was separated from the diastereomer arising from the undesired enantiomer contained in rac-41 via chromatography. The synthesis of 47 began with an S N Ar reaction between 4-fluorobenzonitrile (44) and methyl (R)pyrrolidine-3-carboxylate (45) to give after saponification pyrrolidine 46, which was advanced to 47 as above. Compound 49 was synthesized starting with conversion of alcohol 17 to amine 48, via activation of 17 as a mesylate and displacement with amine 42. Conversion of the cyano group in 48 to an amidine gave 49.

Structure-Activity Relationship Studies
Guided by the docking studies, we set out to establish SAR by dissecting the main molecular features of the compound class. In addition to trypsin, thrombin, and factor Xa, we included KLK4, KLK7, and KLK8 as additional targets for selectivity analysis (Table 3). Having already determined the active stereoisomer 5, we assessed the importance of the oxygen atoms in the oxazolidinone ring by testing lactam 43 and pyrrolidine 47. While 43 showed similar potency and selectivity toward KLK6 relative to 5, 47 slightly lost potency and selectivity against KLK6, indicating some binding role for the carbonyl. A much larger effect was observed with 49, which lacks the amide carbonyl and results in a 40-fold KLK6 potency loss. This data is largely consistent with the docking model, where the amide makes H-bonds, while the oxazolidinone heteroatoms make no strong interactions with surrounding residues.
We next examined variations of the amide N-substituents, which are hypothesized to bind in the S1'/S2' pockets. As mentioned above, docking models suggested these were regions where subtle differences between KLK6 and the related proteases could be harnessed to boost selectivity. In particular, the small methyl group of 5 was predicted to fit into the S1' pocket. This pocket is shallower in KLK6 than in related proteases, but our model suggested that it could accommodate a slightly larger group, strengthening this interaction. Consistent with the model, we found that removal of this methyl group (cmpd. 23) is detrimental for activity, while replacement with an ethyl group (cmpd. 24) results in an almost 10-fold gain in potency for KLK6 and an increase in selectivity. A larger cyclopropane substituent (cmpd. 25) does not improve the potency against KLK6 relative to 24, and is deleterious for selectivity against trypsin and KLK8. It may be that the cyclopropyl group can better interact with the larger S1' pocket of trypsin and KLK8. The appropriate stereochemistry is essential to direct the P1' substituent towards the S1' pocket, as testified by the loss of potency of diastereoisomer 11. Interestingly, a dimethyl group, which presumably still directs one methyl group to the S1' pocket, is also tolerated and appears to be beneficial for selectivity (cmpd. 26).
The SAR of the P2' substituent of the inhibitor class was investigated next. Compound 28 was synthesized to validate the docking prediction, which was indeed supported by an increase in potency compared to parent compound 27, suggesting that the additional amide moiety might form further H-bonds with the active site of KLK6. Moreover, 28 exhibited excellent selectivity over KLK4 and trypsin. In parallel, larger lipophilic substituents were introduced at P2'. Compound 29, with a 1-naphthyl P2', was more potent than 28 (7.9 vs. 7.5 pIC 50 ). Aiming to minimize the peptidic-nature of the compounds, we continued building on 29. Interestingly, the 1naphthyl regioisomer (29) was preferred over the 2-naphthyl (30), which showed a 10-fold potency loss. While a dimethyl P1' substituent showed promise in the case of 26, the gemdimethyl derivative 31 was~6 fold less potent than 29. Fortunately, the optimal P1' (ethyl) and P2' (1-naphthyl) substituents showed additive effects, with compound 32 approaching the potency limit of the KLK6 assay (pIC 50 = 8.6).
This compound retained a similar selectivity profile to the original hit 5, while having almost 100-fold higher potency against KLK6. Interestingly, introduction of a nitrogen atom in the naphthalene bicycle resulted in a dramatic loss of potency as measured for compound 33, further highlighting the steep SAR observed within the P2' substituent.
With a potent compound such as 32 in hand, we examined the possibility of lowering the basicity of the P1 amidino group, which could have an effect on permeability. Previous attempts to introduce less basic functionalities, e. g. benzylamine, amide and aminoisoquinoline within the 5 scaffold, resulted in significant to complete loss of activity against the target protease (data not shown). When the benzylamine replacement was made on the improved scaffold of 32, it resulted in~100fold potency loss (cmpd 35). However, good activity was still detected against KLK6 (pIC 50 = 6.5) with an altogether unchanged selectivity profile, which provides a good starting point for further development of amidine-free KLK6 inhibitors. Interestingly, introduction of a nitrogen atom in the benzamidine ring (cmpd 34), which is predicted to reduce the pKa of the amidine from~11 to~9, resulted in a compound of slightly reduced potency for KLK6, but an improved selectivity profile against the related enzymes. An increase in selectivity over trypsin was also observed when the indole of DKFZ-251 was replaced with a 7-azaindole. [21] The structural reasons for these effects and whether they are connected is currently under further investigation. a All values are reported as pIC 50 . *8.6, i. e. 2.5 nM, represents the assay limit in the KLK6 assay. SAR highlight of each compound is given by the dashed orange box. An entry of "À " means the compound was not measured in that assay.
When analyzing the compounds in a lipophilic efficiency (LipE) plot (LipE = pIC 50 -logP), a number of observations could be made (Figure 4). The value of an S1' ethyl vs. methyl is clear when comparing compounds 5 (LipE = 4.79) and 24 (LipE = 5.16). The two highly potent compounds 32 and 34 show improved LipE values of 5.34 and 5.76, respectively. Interestingly, 28 shows the highest LipE value of 6.20, due to its relatively low lipophilicity. While the additional amide in 28 might be expected to pose a problem for cell permeability, particularly in conjunction with an amidine moiety in the same molecule, further medicinal chemistry optimization could focus on mimicking the amide with heterocycles or other H-bond donors/acceptors.

Lead Compound Pharmacokinetic Properties
The most potent inhibitor, 32, was tested in a number of computational and experimental ADME profiling assays ( Table 4). The free base of 32 was calculated to have an ALogP of 3.26, and its conjugate acid a pK a of 11.3. The trifluoroacetate salt of 32 showed excellent solubility in both kinetic and thermodynamic assay formats, presumably a result of the basic amidine. Clearance measurements with both mouse and human microsomes was relatively low. Clearance in mouse hepatocytes correlated well with the microsomal clearance (2.3 vs 2.4 mL/ min/g), suggesting no significant contribution from phase II pathways. Measurement of cytochrome P450 (CYP) activity with five different isozymes showed no significant inhibition by 32. Only the 3A4 isozyme was inhibited with an IC 50 of 4.8 μM. The compound showed no instability in mouse plasma over 3 h, and was found to bind protein plasma to an extent of 89 %.

Compound 32 Reduces Invasion of HCT116 Cells
We have previously shown that knockdown of KLK6 reduces migration and invasion of HCT116 cells in vitro. [13] In order to test whether this effect could be recapitulated with small molecule KLK6 inhibitors, we measured the ability of 32 to reduce invasion of HCT116 cells. Because the enantiomer of 32 was never prepared, compound 9, the enantiomer of hit 5, was used as an inactive control substance. While compound 9 had no significant effect on the invasion of HCT116 cells after 24 h at multiple concentrations, compound 32 induced a significant reduction in invasion at both 50 nM and 500 nM ( Figure 5). HCT116 cell growth and viability was not altered upon treatment with compounds 9 and 32 at the tested concentrations (data not shown).

Conclusions
After a high throughput screen of the European Lead Factory compound collection, we discovered a validated hit cluster of N-(4-benzamidino)-oxazolidinones that showed consistent inhibitory activity against KLK6. Docking-guided optimization of this scaffold, with a focus on potency against KLK6 and selectivity against up to six different related proteases resulted in compounds with single digit nanomolar potency and good to excellent selectivity. Compound 32 was found to be the most potent inhibitor, while compounds 28 and 34 exhibited the highest selectivity against trypsin. ADME profiling of 32 showed that it has reasonable properties for pre-clinical biological experiments. Finally, 32 was found to reduce invasion of HCT116 cells in a dose-dependent manner, while 9, a control substance from the inactive enantiomeric series, showed no such effect even at the highest tested concentrations.
These compounds show promise as useful chemical probes for the study of KLK6 biology. One particular benefit is the availability of inactive enantiomeric control compounds. While the enantiomers of 28, 32, and 34 were not synthesized as part . Lipophilicity plot of the compounds in this manuscript. The size of each dot represents selectivity over trypsin, with a larger dot indicating better selectivity. Trypsin data was used for this plot because trypsin data was generated for each compound and because trypsin activity varied significantly between derivatives. ALogP was calculated using Pipeline Pilot. Blue = hit substance. Yellow = stereoisomers of the hit. Red = potential leads.

Experimental Section
High Throughput Screen: The European Lead Factory library of 350,000 compounds was used for high throughput screening at the Pivot Park Screening Center facilities in Oss, The Netherlands. HighRes Biosolutions robotic infrastructure using Cellario software was programmed to perform screening of 281 1536-well plates. Using the primary KLK6 fluorescent intensity (FI) assay (see below), 7794 compounds having a Z-score �-4 were selected for further follow-up. An additional set of compounds was added based on Bayesian modeling to include potential false negatives in the hit confirmation. Of the tested 8706 compounds, 1026 showed > 25 % inhibition when re-tested in the KLK6 FI assay. When tested in serial dilutions (20 μM to 20 nM; 7-points p 10 diluted), 312 compounds showed a pEC 50 � 4.7. This set was subsequently tested in Trypsin, Thrombin and FXa assays (see below) resulting in a selection of 226 compounds showing KLK6 pEC 50 > 6.0 OR 5.0 < KLK6 pEC 50 < 6.0 selectivity > 0.1 OR 4.7 < KLK6 pEC 50 < 5.0 selectivity > 1. Upon testing of these compounds using SPR and LC-MS, a qualified hit list (QHL) was registered containing 50 compounds. For SPR, Biacore T200 was used and KLK6 was immobilized onto a CM5 chip. 226 test compounds were screened at 4 concentrations: 20, 4, 0.8 and 0.16 μM in running buffer: 50 mM Tris-HCl, 1 mM EDTA pH 7.5, 150 mM NaCl, 0.05 % Tween 20, 3 % DMSO at 25°C. Control injections of 5 μM 1 were used throughout the screen to ensure that KLK6 was stable and had not been blocked by potential irreversible binders/denaturants. An 8 point 1 in 3 dilution series of 1 (50 μM to 0.02 μM) was also applied at the beginning and end of each screening day.
Docking Studies: The Schrodinger Suite was used to prepare the KLK6 structure (PDB ID 4D8N) and to perform the docking experiments (Glide). Molecules were standardized and 3D conformers generated with the Ligprep module from the Schrodinger Suite using standard settings. Input stereochemistry was retained. The docking was performed using the Extra Precision (XP) method and the other parameters are assigned to their default values. A postdocking refinement step was applied using the MM-GBSA module that approximates the free energy of binding.
Chemistry: (General) Chemicals and solvents were from commonly used suppliers and were used without further purification. Silica gel 60 F254 analytical thin layer chromatography (TLC) plates were from Merck (Darmstadt, Germany) and visualized under UV light and/or with KMnO 4 stain. Automated chromatography was performed with a Biotage Isolera Purification system (Uppsala, Sweden). Deuterated solvents were obtained from Cambridge Isotopes. All NMR spectra were recorded using a Bruker Avance 400 MHz spectrometer, and the residual solvent peak was used as internal reference ( 1 H NMR: CHCl 3 (7.26 ppm); DMSO (2.50 ppm); MeOH (3.31 ppm); 13 C NMR: CHCl 3 (77.16 ppm); DMSO (39.52 ppm); MeOH (49.00 ppm)). Chemical shifts (δ) are reported in ppm and coupling constants (J) in Hz. The following abbreviations were used for multiplicities: s = singlet, d = doublet, t = triplet, q = quartet, p = pentet, m = multiplet, br = broad. Preparative HPLC was carried out on a Waters HPLC comprising a Waters 2767 Sample Manager, Waters 2545 Binary Gradient Module, Waters Systems Fluidics Organiser, Waters 515 ACD pump, Waters 2998 Photodiode Array Detector, using a Waters XBridge Prep OBD C18, 5 μm, 19 mm × 50 mm i.d. column and a flow rate of 20 mL/min. The general method used a gradient of 5 % acetonitrile/95 % water to 100 % acetonitrile (with 0.1 % trifluoroacetic acid in both phases). UV detection (254 nM) was used for the collection of fractions from HPLC. All final compounds were found to have � 95 % purity, controlled by analytical (LC/MS) and confirmed by 1 H NMR.

Method A (amide coupling):
To a mixture of carboxylic acid (1 equiv) and HATU (1.5 equiv) in DMF is added amine (1.2 equiv) and i-Pr 2 NEt (3.0 equiv). After stirring for 24 h, the mixture is partitioned between EtOAc and brine/water. The layers are separated and the aqueous phase is extracted with EtOAc (2 ×). The combined organics are washed with brine (3 ×), dried (Na 2 SO 4 ), filtered, concentrated under reduced pressure, and purified by chromatography.

Method D (Grignard addition to sulfinyl imine):
To a solution of 37a or 37c (1 equiv) in CH 2 Cl 2 is added a solution of EtMgBr (2 equiv) in Et 2 O at À 78°C under argon. After 4 h, the mixture is allowed to warm to rt. After 18 h, the reaction is quenched with H 2 O, diluted with EtOAc, and the two layers are separated. The aqueous layer is further extracted with EtOAc (2 ×), the combined organics are washed (brine), dried (MgSO 4 ), filtered, concentrated, and purified by chromatography.

1-(4-cyanophenyl)-5-oxopyrrolidine-3-carboxylic acid (41):
A microwave vial containing a mixture of 4-aminobenzonitrile (12) (2.0 g, 17 mmol) and 2-methylenebutanedioic acid (40) (2.2 g, 17 mmol) was heated to 160°C for 1.5 h, then allowed to cool to rt. The resulting gel was sonicated in 2 M NaOH and filtered to remove a pink solid. The filtrate was acidified with 5 M HCl and extracted with EtOAc (x3). The combined organic extracts were dried (Na 2 SO 4 ), filtered and concentrated under reduced pressure. The residue was taken up in TFA (10 mL) and stirred at rt for 3 h. The mixture was heated to 55°C for 18 h, then quenched with a saturated aqueous solution of NaHCO 3 . 2 M NaOH was added and the aqueous phase was extracted using CH 2 Cl 2 , then acidified with 5 M HCl, and extracted with EtOAc (x4). The combined organic extracts were dried (Na 2 SO 4 ), filtered and concentrated under reduced pressure. The oily residue was triturated with EtOAc/Et 2 O to produce a sticky solid, which upon trituration with CH 2 Cl 2 produced a filterable off-white solid. This was filtered, washed with CH 2 Cl 2 , and dried in vacuo to give 1.  (3R)-1-(4-cyanophenyl)pyrrolidine-3-carboxylic acid (46): A suspension of 44 (0.26 g, 2.2 mmol), 45 (0.28 g, 2.2 mmol) and Na 2 CO 3 (804.2 mg, 7.59 mmol) in DMSO (3.0 mL) was heated to 85°C overnight in a sealed vial. The reaction mixture was diluted with EtOAc and filtered through celite. The combined organic extracts were washed with a saturated aqueous solution of NaHCO 3 and brine, dried (Na 2 SO 4 ) and concentrated under reduced pressure. The crude material was purified by flash column chromatography

4-((S)-5-((((S)-1-(3-methoxyphenyl)ethyl)amino)methyl)-2-oxooxazolidin-3-yl)benzonitrile (48):
To a solution of 17 (100 mg, 0.46 mmol) in THF (1.1 mL) was added Et 3 N (0.08 ml, 0.6 mmol) followed by methanesulfonyl chloride (0.05 ml, 0.6 mmol). A suspension resulted which was stirred at rt for 16 h. The resulting mixture was quenched with H 2 O and concentrated. The residue was taken up in CH 2 Cl 2 and passed through a hydrophobic phase separator, further rinsing with CH 2 Cl 2 . The filtrate was concentrated under reduced pressure. The residue was taken up in DMF (0.5 mL) and transferred to a microwave vial containing 42 (104 mg, 0.69 mmol). The resulting mixture was subjected to microwave irradiation at 120°C for 60 min and then 130°C for 90 min. Thermal heating was then used at 100°C for 16 h. The resulting mixture was partitioned between EtOAc and brine/water. The aqueous phase was further extracted with EtOAc and the combined organic extracts washed with brine (x2), dried (Na 2 SO 4 ), filtered and concentrated, and purified by column chromatography (Telos 12 g, EtOAc in DCM, 0-80 %), then (SNAP Ultra-10 g, EtOAc in DCM, 0-
Human thrombin was purchased from Biopur. Human trypsin was purchased from Polymun Scientific. Human Factor Xa was purchased from Enzo Life Sciences. Thrombin (20 nM), trypsin (0.000015 %) or Factor Xa (10 nM) were assayed in a similar manner to the KLKs but were incubated with compound for 20 min at room temperature in assay buffer (50 mM Tris HCl, pH 7.6, 100 mM NaCl, 1 mg/ml CHAPS, 0.05 % bovine serum albumin (BSA), 1 mM dithiothreitol (DTT)) before the addition of 5 μM, 2.5 μM or 1.5 μM Rhodamine 110 substrate (Cambridge bioscience ltd) respectively. Fluorescence intensity was then read at ex/em 485/535 following incubation for 1 h at room temperature (thrombin and factor Xa) or 20 min (trypsin).

Kinetic aqueous solubility:
The aqueous solubility of the test compounds was measured using laser Nephelometry. Compounds were subjected to serial dilution from 10 mM to 0.5 mM in DMSO. An aliquot was then mixed with Milli-Q water to obtain an aqueous dilution plate with a final concentration range of 250-12 μM, with a final DMSO concentration of 2.5 %. Triplicate aliquots were transferred to a flat bottomed polystyrene plate which was immediately read on the NEPHELOstar (BMG Lab Technologies). The amount of laser scatter caused by insoluble particulates (relative nephelometry units, RNU) was plotted against compound concentration using a segmental regression fit, with the point of inflection being quoted as the compounds aqueous solubility (μM).
Thermodynamic solubility: Test compound (3 mg) is accurately weighed into a Micronics tube. Extra tubes for each control compound, Warfarin (high solubility control) and Cinnarizine (low solubility control) were also accurately weighed. To each tube is added 1 mL of Milli-Q water. All samples are vortexed for 20 s and placed in a bench top incubator shaker set at 1300 rpm at 37°C for 5 h. After the incubation, samples are centrifuged at 10,000 rpm for 3 min. 100 μL aliquots of the supernatant in duplicate were placed in a 96-well plate. All samples, calibration standards and QC samples are analyzed using a validated HPLC method on the Dionex Ultramate 3000 HPLC system. The concentration determined from the supernatant was then reported as the solubility of each compound in μg/mL. Microsomal Intrinsic clearance (CLint) assay: Test compound (0.5 μM) was incubated with female CD1 mouse liver microsomes (Xenotech LLC TM; 0.5 mg/mL 50 mM potassium phosphate buffer, pH 7.4) and the reaction started with addition of excess NADPH (8 mg/mL 50 mM potassium phosphate buffer, pH 7.4). Immediately, at time zero, then at 3, 6, 9, 15 and 30 min an aliquot (50 μL) of the incubation mixture was removed and mixed with acetonitrile (100 μL) to stop the reaction. Internal standard was added to all samples, the samples centrifuged to sediment precipitated protein and the plates then sealed prior to UPLC-MS/MS analysis using a Quattro Premier XE (Waters Corporation, USA). XLfit (IDBS, UK) was used to calculate the exponential decay and consequently the rate constant (k) from the ratio of peak area of test compound to internal standard at each timepoint. The rate of intrinsic clearance (CLint) of each test compound was then calculated using the following calculation: CLint (mL/min/g liver) = k × V × Microsomal protein yield. Where V (mL/mg protein) is the incubation volume/ mg protein added and microsomal protein yield is taken as 52.5 mg protein/g liver. Verapamil (0.5 μM) was used as a positive control to confirm acceptable assay performance.
Hepatocytes Intrinsic clearance (CLint) assay: Cryopreserved vials of mouse cryopreserved hepatocytes, supplied by Life Technologies, were thawed according to manufacturer's instructions and cells re-suspended in Williams Medium E (WME) containing cell maintenance supplement pack (CM4000, Life Technologies). Hepatocytes were incubated in suspension (0.5x10 6 cells/mL) in 48 well non-collagen coated cell culture plates for 10 min at 37°C, 95 % O 2 5 % CO 2 . Upon addition of an equal volume of supplemented WME containing 1 μM test compound, an aliquot of incubation solution was removed to acetonitrile containing internal standard (final concentration 0.5 μM test compound and a cell density of 0.25 × 10 6 cells/mL). Similarly, aliquots were removed at 3, 6, 9, 15, 30, 45, 60, 90 and 120 min. 100 μL of 80 : 20 H 2 O/MeCN was added to all samples and the analysis plate was centrifuged for 10 min at room temperature prior to injection and analysis of samples by UPLC-MS/ MS. The response (area ratio of test compound to internal standard) was plotted against time using an exponential decay model and rate of disappearance calculated. Hepatocyte CLint (mL/min/10 6 cells) was scaled to in vivo CLint (mL/min/g liver) using the hepatocellularity scaling factor of 120 × 10 6 cells/g of liver.