Design and Preclinical Characterization Program toward Asundexian (BAY 2433334), an Oral Factor XIa Inhibitor for the Prevention and Treatment of Thromboembolic Disorders

Activated coagulation factor XI (FXIa) is a highly attractive antithrombotic target as it contributes to the development and progression of thrombosis but is thought to play only a minor role in hemostasis so that its inhibition may allow for decoupling of antithrombotic efficacy and bleeding time prolongation. Herein, we report our major efforts to identify an orally bioavailable, reversible FXIa inhibitor. Using a protein structure-based de novo design approach, we identified a novel micromolar hit with attractive physicochemical properties. During lead modification, a critical problem was balancing potency and absorption by focusing on the most important interactions of the lead series with FXIa while simultaneously seeking to improve metabolic stability and the cytochrome P450 interaction profile. In clinical trials, the resulting compound from our extensive research program, asundexian (BAY 2433334), proved to possess the desired DMPK properties for once-daily oral dosing, and even more importantly, the initial pharmacological hypothesis was confirmed.


■ INTRODUCTION
The formation of a network of fibrin fibers and activated platelets is a key physiological mechanism to prevent massive bleedings at sites of vessel wall leakage. 1 On the other hand, overly excessive fibrin generation may result in the formation of vessel-occluding thrombi, which can cause ischemic damages to the surrounding tissues. 2 Such events can potentially lead to heart attacks, strokes, or pulmonary embolisms, all of which are ranked high as causes of death or disability and increased healthcare costs. 3Therefore, thrombus prevention and resolution remain of key importance for patients and societies. 4ince the introduction of novel antithrombotics in the early 2000s, mainly platelet aggregation inhibiting P2Y 12 antagonists 5 and direct oral anticoagulants (DOACs), 6 significant therapeutic advances have been made.The DOACs target the final common pathway of the coagulation system via inhibition of either factor Xa (rivaroxaban, apixaban, and edoxaban) or thrombin (dabigatran).While blocking any kind of fibrin formation by addressing a target after the convergence of the intrinsic and extrinsic cascade is very efficacious, the antithrombotic efficacy of these inhibitors is linked to an increased risk of bleeding, and the identification of "sweet spots" with positive benefit/risk ratios has been a major goal in clinical trials.Nevertheless, especially in many high-risk patients, this bleeding risk is a major concern.
Recently, another serine protease in the coagulation cascade, factor XIa (FXIa), 7 has emerged as an attractive target, because it may allow for a more selective reduction of fibrin generation: FXI is activated via FXIIa after contact activation or by thrombin in the course of the positive feedback loop for thrombin amplification, but it is not involved in the tissue factor pathway, which is key for fibrin production to cover a vessel wall leakage.Thus, this form of "selective coagulation modulation" 8 by FXIa inhibition may allow for a decoupling of antithrombotic efficacy and bleeding risk and offer the opportunity for better antithrombotic protection to a much broader patient population.Several lines of evidence support this notion: (1) Individuals with FXI deficiency have a reduced risk for venous thrombosis 9 or stroke 10 and very rarely experience any abnormal bleeding events, which are not related to FXI levels.(2) Knock-out mice are protected from thrombosis and have not revealed any increase in bleeding risk versus wild-type animals. 11(3) Experiments in animal models in multiple species with irreversible or reversible inhibitors to achieve reduced FXI synthesis or FXI(a) activity have demonstrated strong anticoagulant effects without an impact on bleeding times. 12(4) Most importantly, currently, several approaches to inhibit FXI synthesis or FXIa activity are in clinical phases and in these early phases have demonstrated antithrombotic efficacy with less bleedings compared to anticoagulant standard of care or similar bleeding rates compared to placebo.
With regard to parenteral treatment, antisense approaches to reduce FXI synthesis, such as with ISIS 416858 or fesomersen, 13 and FXI-or FXIa-targeting monoclonal antibodies, such as osocimab, 14 abelacimab, 15 or xisomab, 16 are in Phase 2 clinical evaluation.
Orally bioavailable, reversible FXIa inhibitors might be particularly desirable for chronic treatment, but the research aimed at identifying such small molecule FXIa inhibitors has been challenging despite significant and lengthy efforts.Recently, clinical Phase 2 data 17−20 have been presented for two orally administered candidates, milvexian 21 (BMS-986177 or JNJ-70033093) and asundexian 8  (BAY 2433334, 80).
In this study, we describe our drug discovery efforts, resulting in the identification of asundexian (BAY 2433334, 80), a direct, potent, selective, and reversible small molecule FXIa inhibitor for oral application, which is currently being investigated in clinical Phase 3 trials.

■ RESULTS AND DISCUSSION
To establish a starting point for a program toward orally bioavailable, reversible FXIa inhibitors, we performed several high-throughput screening campaigns using the Bayer AG inhouse proprietary compound library, which contained up to 4.3 million compounds.However, a viable hit was not identified. 22We therefore pursued a protein structure-based de novo design approach with the aim of generating a novel starting point for lead modification toward a reversible, active site FXIa inhibitor.Two structurally distinct compound series discovered by Bristol Myers Squibb were known as FXIa inhibitors at the start of our drug discovery program, a charged 4-aminomethyl-trans-cyclohexyl-bearing series and a neutral series containing chloroaryl and 3-aminoindazole moieties.We studied both series in detail to understand their interaction with FXIa.
Exploration of Distinct Chemical Classes of FXIa Inhibitors: Zwitterionic/Charged FXIa Inhibitors.To explore FXIa inhibitors, we synthesized tetrazole 1 (inspired by chemical matter published by Bristol Myers Squibb in 2007 23 ) which showed good potency (FXIa IC 50 = 9 nM, Figure 1).A docking experiment of this compound in the published X-ray structure of FXIa 24 indicated that the 4aminomethyl-trans-cyclohexyl residue would occupy the S1 pocket.The tetrazole moiety appeared to point toward the S2′ pocket.An X-ray cocrystal structure of tetrazole 1 in complex with human FXIa then revealed the binding mode with the key hydrogen bonds and hydrophobic contacts (Figure 1).The 4aminomethyl-trans-cyclohexyl moiety was indeed found to be located in the S1 pocket where its primary amine forms a direct ionic interaction with the carboxylate of Asp189 and a hydrogen bond with the backbone carbonyl of Gly218.The S1 pocket is the only deep pocket of the FXIa active site, and ligand 1 extends from the S1 pocket into the S1′ and S2′ pockets.We hypothesized that this orientation in the binding mode toward the prime sites might facilitate the required selectivity versus other serine proteases such as FXa and thrombin, as potent inhibitors of the latter use the more druggable pockets of their nonprime site (S1 and S2; S4). 25 The cyclohexyl-bound amide carbonyl oxygen forms two hydrogen bonds with the protein backbone forming the oxyanion hole (OAH in Figure 1).The ester binding pocket (EBP) is not occupied with substituents of 1. Located in the S2′ pocket, the tetrazole moiety of 1 forms a hydrogen bond with the hydroxyl group of Tyr143 and a water-mediated interaction with the carbonyl group of His38.The docking binding mode experiment did not indicate the potential of this interaction since Lys192 was kept rigid in that particular X-ray structure.In the S1′ pocket, the phenyl ring of the benzyl side chain participates in a hydrophobic contact with the Cys42-Cys58 disulfide bridge.
However, the low membrane permeability observed for tetrazole 1 made it an unfavorable starting point for our program.We speculated that the rigid molecular structure, which prevents close intramolecular contacts between the counter charged functional groups, had a negative effect on permeation (driven by separated charged moieties).
Neutral FXIa Inhibitors.Neutral P1 groups such as chloroaryl, chlorothienyl, or methoxyaryl have been successfully exploited in the discovery of oral inhibitors of thrombin 26 and FXa. 27In our experience, whether serine proteases of the coagulation cascade (e.g., thrombin, FVIIa, FIXa, FXa, and FXIa) can be addressed by neutral P1 groups depends mainly on the amino acid at position 190.Thrombin and FXa possess an Ala190, flanking Asp189 at the bottom of the S1 pocket and offering no interaction sites with bulk water.Ligands with neutral, lipophilic P1 groups can expel the deeply buried water and release it into bulk solvents, which is associated with a huge gain in entropy.In contrast, serine proteases such as FVIIa and FIXa with Ser190 offer hydrogen bond acceptor and donor sites to the bulk water and directly bind that water molecule.Lipophilic residues alone are not sufficient to release the more tightly bound solvent molecules, and ligands require rather basic P1 groups for potent inhibition.As FXIa belongs to the Ala190 "class", with an otherwise completely conserved S1 pocket compared to thrombin or FXa, we proposed at the start of our project that neutral P1 groups such as chloroaryl should also work in the design of our new FXIa inhibitors.This proposal was supported by compound 2 (FXIa IC 50 = 4 nM) disclosed by Bristol Myers Squibb in 2008. 28The binding mode of this compound could be modeled based on the cocrystal structure of 1, using a docking approach.Based on these data, we hypothesized that the chloroaryl residue would make similar contacts to Tyr228 as in homologous serine proteases and occupy the S1 pocket.In theory, the 3aminoindazole moiety can be regarded as cyclized benzamidine and could have functioned as the P1 residue as well.Serine protease inhibitors carrying this P1 motive are known. 29The X-ray cocrystal structure of ligand 2 in complex with human FXIa showed that a basic P1 binder as in 1 can be replaced by a nonbasic chloroaryl moiety.The chlorine substituent forms an edge-to-face interaction with Tyr228 at the bottom of the S1 pocket (Figure 2).
Hit Finding.In 2010, based on our in-house X-ray cocrystal structure of compound 1 in complex with human FXIa (Figure 1), and the shape and properties of the FXIa binding pocket, we defined for our program the most important interactions which an oral FXIa inhibitor should fulfill: (1) most importantly, the S1 pocket should be occupied by a lipophilic aromatic moiety (preferably a chloroaryl substituent), (2) the oxyanion hole (OAH) should be occupied by a carbonyl group being able to form one or two strong hydrogen bonds with the backbone NH groups of Lys192 and Gly193, (3) the solventexposed backbone NH of Gly216 at the rim of the S1 pocket should be contacted by a strong hydrogen bond acceptor, (4) a trapped water molecule (between Leu39 and Ser195) functioning as an indirect hydrogen bond acceptor on the protein side should be contacted by a strong hydrogen bond donor, such as an NH group, on the ligand side, and (5) a lipophilic/aromatic spacer carrying a neutral or acidic func-tional group that is capable of accepting two hydrogen bonds (from Tyr143 and another trapped water) should be present (Figure 3).Since this design concept involved a relatively high number of polar functional groups, we aimed to connect them with primarily lipophilic spacers.
We then de novo designed a novel chemotype that we hypothesized would allow for all of these interactions.The design was based on the previously undescribed idea of incorporating the central pharmacophoric elements into one heteroaromatic core, allowing decoration with substituents carrying the above-noted functional groups.Various different aromatic cores were enumerated, docked into the active site, and energy-minimized.Besides the steric and electrostatic complementarity, synthetic accessibility of a similar core that was previously used in an unrelated, internal Bayer AG project 30 played a role in the final selection for the synthesis of the first compound.This structure-based de novo design approach resulted in hit 3 with micromolar potency in FXIa inhibition and favorable selectivity versus other serine proteases such as thrombin, FXa, and especially important, trypsin (all IC 50 values >50 μM).The subsequent comparison of the (pre-synthesis) docked complex versus the X-ray cocrystal structure of ligand 3 in complex with human FXIa confirmed our design concept (Figure 3).Since 3 already had favorable physicochemical properties, we had fortunately achieved a promising hit with a novel chemotype with the synthesis of just one compound, which we could then use as the starting point for further modification.
Lead Finding.The subsequent lead finding was, among other aspects, guided by consequent considerations of physicochemical properties like clogD 7.5 , 31 corrected molecular weight (MW corr ), 32 topological polar surface area (tPSA), 33 ligand efficiency (LE) 34 (with MW corr ), and ligand lipophilicity efficiency (LLE) 35 (with clogD 7.5 ), as well as being designdriven by WaterMap, 36 a molecular dynamics approach characterizing solvent structure and energetics within a ligand binding pocket.The system interaction energy and excess entropy for each water molecule in the substrate binding pocket are calculated using inhomogeneous solvation theory. 37n principle, the less energetically (enthalpically and/or entropically) favorable expelled water molecules are, the more favorable is their contribution to the binding free energy if replaced.As shown in Figure 4, the energetically most favorable hydration sites with the color-coded modeled highest free energy (ΔG) upon dis-or replacement are shown (for detailed values, see Figure S1, Supporting Information).The analysis indicated that the energetically least favored water molecules reside in the S1 pocket (site 16, ΔG = 5.2 kcal/mol and site 37, ΔG = 3.2 kcal/mol), followed by the S2′ pocket (site 106, ΔG = 4.0 kcal/mol) and the S1′ pocket (site 57, ΔG = 3.1 kcal/mol).Other important hydration sites are located at the rim of the S1 pocket (site 47, ΔG = 3.0 kcal/mol), the distant part of the S1′ pocket (site 49, ΔG = 2.8 kcal/mol), and in the EBP (site 68, ΔG = 2.6 kcal/mol).Our internal analysis led us to the conclusion that further hydration sites lie between the EBP and the S2′ site and the S3′ site but cannot be addressed with ligands of molecular weight compatible with good absorption properties.It is worth noting that in FXIa, the S2 pocket is so shallow that it cannot be detected by this computational/protein structural analysis as such.
Compound 3 displaces water molecules corresponding to sites 16, 37, 106, 93, and 47.Aiming to improve the potency of our hit 3, we first concentrated on adding additional substituents to displace energetically unfavorable water molecules in the EBP and S1′ pocket.The aforementioned WaterMap calculations helped to guide our structural modifications.
We hypothesized that an additional ortho substitution at the P1 chloroaryl moiety could further improve potency by filling the EBP (and thus displacing an energetically unfavorable water molecule, site 68) and by conformationally preorganizing the dihedral angle between the pyridinone and chloroaryl ring.We proposed that this angle, which was derived from the X-ray complex of FXIa with ligand 3, should ideally be approximately −66°to fit the S1 pocket with a minimal entropy loss upon binding.To facilitate synthetic access, we first considered a series of simplified nonannulated pyridinones.We aimed to explore the EBP with relatively small, not too polar substituents, composed of one or two non-hydrogen atoms.As a quick decision criterion, we looked at the difference of the dihedral angle between the pyridinone and P1 aryl moiety in the active conformation (energy-minimized in the FXIa binding site) and the unbound ligand in water and the steric filling of the EBP.We hypothesized that the smaller this difference, the more preformed the compounds would be, having a lower entropy penalty upon binding.Energy calculations were carried out with the OPLS2.0force field 38 and implicit continuum solvent model for water.Virtual derivatives were enumerated, docked, and prioritized.Cyano as a substituent was clearly favored as it showed an excellent fit in the EBP and a relatively flat torsion profile with a still not ideal energy minimum at around −43°in the unbound energyminimized state.Several derivatives (like compounds 5−7) were synthesized and, indeed, an ortho-cyano group turned out to be a local optimum among smaller ortho substituents ().The cocrystal structure (data not shown) for compound 7 in complex with human FXIa revealed a torsion angle of 68°of the P1 phenylpyridinone moiety, with only Lys192 giving way to the larger CN group in comparison to 3 and enlarging the EBP.
The inactivity (>50 μM) of the unsubstituted derivative 4 is interesting and shows the importance of conformational preformation and rigidification in that series.Had this derivative, fulfilling 80% (4 out of 5) of the pharmacophoric requirements (Figure 3, see de novo strategy outlined above), been synthesized and tested first, without recognizing the importance of the dihedral angle (P1 phenyl/pyridinone, −18°), namely, the presence of at least one appropriate ortho substituent and the conformational prearrangement, the project would have taken a different turn.
Following the guidance from the WaterMap calculations and a superimposition of the chemically distinct compound 1, we speculated that a larger lipophilic substituent protruding toward the S1′ pocket would displace energetically unfavorable water molecules and lead to an increase in binding affinity.The first derivative designed to explore this pocket was compound 8 carrying an additional benzyl group.The observed FXIa IC 50 value of 8 nM for the racemate exceeded our expectations (Table 1).Parts of the 70-fold affinity increase (compared to unsubstituted 7) were speculated to arise from the rigidification of the flexible hinge in the middle part of the molecule.The additional stereocenter leads to a productive conformational prearrangement.We hypothesized from this analysis that a single methyl group (P1′ substituent, see Table 1) would also lead to a significant potency gain without occupying the S1′ pocket.The said methyl derivative 9 indeed displayed a strong increase in IC 50 (53 nM) compared to 7.This was an important finding since with two derivatives, the potency of the compound class was increased by more than 10fold with a minimal increase in molecular weight and lipophilicity.Additional derivatives (like compounds 10 and 11) with a sterically more demanding P1′ residue showed a general trend of increasing affinity with increasing size, which is compatible with the water displacement theory.The n-butyl derivative 12 does not obey that general trend, which may be explained by higher flexibility and a relatively high gain in entropy upon binding as a potential reason for the somewhat lower measured binding affinity.
The next step in the design of novel FXIa inhibitors would be the combination of the annulated cyclopentanone (cf.compound 3) and the ortho-cyano P1 moiety.However, since The P1 dihedral angle from force field calculations with implicit water is shown to monitor the progress in adjusting an ideal angle around 66°d erived from the X-ray complex of FXIa with ligand 3. b n.d.: not determined.the cyclopentanone moiety (especially the pyridinone C6 substituent) was hypothesized to lead to an intramolecular steric clash with a P1′ substituent that was considered necessary to achieve high potency, such a structural modification was deprioritized at that point in time. 39nstead, we chose to further explore the nonannulated pyridinone as the central core and to address the important interaction with the backbone Gly216:NH with a single substituent at the C5 position.We were looking for C5 substituents that lead to a dihedral angle of around −66°and ideally act as a hydrogen bond acceptor to form strong interactions with Gly216.For P2′, we changed the acidic function from tetrazole to carboxylate.From structurally related matched pairs, we could see the trend that Caco-2 permeation was better here.Combinations of various C5 substituents with different EBP residues were enumerated, characterized, and ranked in silico.Compound 15 with a C5chloro substituent, allowing for a favorable dihedral angle of −59.8°, was prepared and led to 6-fold improved potency (FXIa IC 50 = 27 nM) compared to C5-unsubstituted compound 13 (Table 2).This was especially promising as the C5-chloro substituent would not form a hydrogen bond and the potency increase was primarily arising from the conformational prearrangement compared to the C5-unsubstituted compound.Based on our work, we anticipated that C5-methoxy substitution would be highly beneficial, as we designed it to not only adopt a more favorable dihedral angle (energy minimum of unbound compound at −54°) but also form an additional hydrogen bond with the backbone NH of Gly216.Indeed, the resulting compound 19 showed a 29-fold improved potency (FXIa IC 50 = 5.9 nM, racemic mixture) relative to 13.We observed that sterically more demanding

Journal of Medicinal Chemistry
EBP substituents than cyano, such as trifluoromethyl (18)  versus difluoromethyl (17) and trifluoromethoxy (21) versus difluoromethoxy (20), can result in significantly less potent compounds (Table 2), emphasizing the importance of choosing the right combination of both EBP and C5pyridinone substitution.
Initial Lead Modification.The goal of the initial lead modification was to increase potency, specifically, to increase the anticoagulant activity of the compound series significantly, without changing the favorable physicochemical properties excessively.Following the WaterMap calculations (Figure 4) and our initial design of substituents in the S1′ pocket, we explored suitable P1′ groups in more depth (Table 3).Direct branching by extending the P1′ ethyl group to isopropyl was unfavorable (compound 26).Both FXIa potency and anticoagulant activity improved going from isopropylmethyl (27) to cyclopropylmethyl (28) and then to cyclobutylmethyl (29).Additional terminal branching like tert-butylmethyl (30)  resulted in loss of potency.We observed a gain in anticoagulant activity upon the introduction of a methoxyethyl substituent, namely, compound 31 (human aPTT EC 150 = 0.12 μM).Based on the limited potency previously obtained for the P1′ n-butyl derivative 12 (Table 1), this was surprising and appeared to result from an additional hydrogen bond interaction of the ether oxygen with a water molecule, held in place by Ser195.This hydrogen bond was subsequently observed in a cocrystal structure with one of the later described derivatives (34).By integrating the P1′ methoxyethyl into a five-or six-membered aliphatic ring, we focused our activities on cyclic ethers due to their balanced physicochemical properties and potentially improved solubility compared to, for example, a benzyl substituent.This resulted in highly potent tetrahydropyran 34 (FXIa IC 50 = 0.5 nM, human aPTT EC 150 = 0.18 μM).From the X-ray cocrystal structure of 34 in complex with human FXIa, the absolute stereochemistry of both chiral centers was determined to be the (S)-configuration.The additional interaction between a water molecule, binding to the hydroxy functions of Ser195 and the ether oxygen of the P1′ tetrahydropyran, is nicely revealed (Figure 5).Moleculardynamics-based relative binding free energy calculations later confirmed the choice of (S)-tetrahydropyran within the typically observed error range (calculated ΔG = −12.21± 0.21 kcal/mol versus experimental ΔG = −12.77kcal/mol).Further efforts to reduce the number of carbon atoms resulted in the design of methyl-substituted 36 [preferred isomer: (S)-Me, FXIa IC 50 = 0.7 nM, human aPTT EC 150 = 0.08 μM] which can be regarded as a seco-pyran derivative.The P1′ 4hydroxycyclohexylmethyl residue was designed to replace an energetically unfavorable water molecule (site 49, Figure 4) and form an additional hydrogen bond with the backbone carbonyl oxygen of Cys58, and derivative 37 marked one of the most potent compounds (FXIa IC 50 = 0.3 nM, human aPTT EC 150 = 0.12 μM) but lacked Caco-2 permeation and in vivo metabolic stability.Also, in binding free energy calculations, 37 showed high potency with a calculated ΔG of −13.69 ± 0.30 kcal/mol (experimental ΔG = −13.07kcal/mol).
Characterization of Carboxylic Acid 34.At that point in the project, carboxylic acid 34 was regarded as the best compound overall and was profiled in depth: 34 inhibited human FXIa directly, reversibly, potently, and selectively with an IC 50 of 0.5 nM in buffer.Plasma kallikrein, the protease closest related to FXIa, was inhibited with an IC 50 of 4.9 nM.Compound 34 showed selectivity for FXIa inhibition versus other proteases, especially of the hemostatic system, including thrombin, FXa, FVIIa, FIXa, plasmin, tissue plasminogen activator, and trypsin (all inhibited at a more than 1000-fold lower potency).While the 50% clotting time prolongation (EC 150 ) in the aPTT assay during the compound screening phase was 0.18 μM (final assay concentration as described in the Supporting Information, corresponding to 0.54 μM in plasma), the in-depth characterization with repeated measurements yielded an EC 150 of 0.25 μM in human plasma.In plasma samples of various other species, compound 34 prolonged the clotting time as well, resulting in an aPTT EC 150 of 2.8 μM (rabbit), 0.28 μM (dog), and 0.25 μM (pig).However, no clotting time prolongations were observed in mouse or rat plasma (aPTT EC 150 > 30 μM).
The effect of 34 on thrombus formation in vivo was evaluated in ferrous-chloride-induced injury models at the rabbit carotid artery or jugular vein.When administered prophylactically by intravenous injection, 34 reduced the thrombus weight dose-dependently in both models without interfering with hemostasis, which was checked simultaneously by measuring the ear bleeding time (see Figure 8A−E).Ex vivo experiments in plasma of the respective animals demonstrated a close correlation between the aPTT and the thrombus weight reduction.Oral administration of a solution of 10 or 20 mg/kg 34 in a PEG/ethanol/water vehicle to rabbits resulted in 16  CYP3A4 inhibition: IC 50 without preincubation/IC 50 after 30 min preincubation.c tPSA value does not reflect the actual polar surface area as the algorithm simply sums up the polar surface of precalculated functional groups and does not take into account intramolecular hydrogen bonds.and 39% reduction of thrombus weight, respectively, in a ferrous chloride-induced injury model (see Figure 8F).
The P2′ carboxylic acid moiety turned out to be favorable for a noncritical drug−drug interaction (DDI) profile with respect to inhibition and induction of CYP3A4.Despite this carboxylic acid, we observed glucuronidation as only a minor pathway in human hepatocytes, relatively high free fractions (fraction unbound, f u ) in animal species (13−15%) and still moderate f u in man (5%), and moreover, moderate to low in vivo clearance (CL) and moderate oral bioavailability (53%/ 55%) in rats and dogs (see Tables 7 and 8).
With additional favorable results in preclinical safety studies, 34 was chosen for evaluation in a first-in-human study.In this trial, however, 34 turned out to have lower oral exposure and a shorter dominant half-life in humans than expected.Based on data from animal studies, we apparently had underestimated the more complex absorption behavior of 34 in humans and the extent of transport via P-gp and OATP, most probably due to the carboxylic acid moiety.
Modification of P2′ Substitution.We further evaluated a broad range of compounds with P2′ carboxylic acids, benzoic acids, and several other derivatives (e.g., acids 38−40, Table 4), but it became evident to us from additional in vivo PK experiments that nonacidic P2′ moieties displayed longer halflives and lower clearances than acidic P2′ moieties and therefore could be more promising for improving the PK profile in humans.
We did not identify a nonaromatic P2′ group leading to compounds with sufficient potency (e.g., 41 is less potent than 42, Table 4).Moreover, the NH of the central amide bond is essential for the water-mediated interaction with Leu39 (Figure 5) so that, for example, cyclization to circumvent a secondary amide was not an option.Therefore, we had to stick with the resulting secondary anilide structure having the potential to be cleaved by metabolism with liberation of an aniline derivative.To avoid potentially genotoxic metabolites, only Ames negative anilines (like 4-aminobenzoic acid) qualified for use, and we evaluated those by GLP Ames screening studies.Any P2′ group had to fulfill the complex interaction pattern (addressing His38, Arg37D, and Tyr143 concurrently) which is necessary as anchor in the S2′ pocket for high FXIa affinity but further restricts the selection pool of aromatic amines substantially.
With these prerequisites in mind, we evaluated a range of nonacidic P2′ groups.The selection of P2′ groups for synthesis was supported by binding free energy calculations using FEP +. 40 Compared to the corresponding carboxylic acids, the nonacidic P2′ derivatives were overall less potent in FXIa inhibition and showed less anticoagulant activity.This is rationalized by the missing ionic interaction of the acidic moiety with Arg37D.Nonacidic P2′ groups leading to more potent compounds often required several hydrogen bond donors/acceptors resulting in compounds with low Caco-2 permeation and insufficient exposure after oral administration, correlating with high tPSA values.Moreover, many of these derivatives showed critical inhibition of CYP isoforms.These overall issues hold true for many derivatives with different P1′ groups, and a few examples comparable with carboxylic acid 31 are shown in Table 4.
Dihydroindazolone 43 was designed to interact via its aromatic NH group with the His38 backbone carbonyl and replace a bridging structural water always observed for the P2′ benzoic acids.It was hypothesized that the dihydroindazolone carbonyl function would accept a hydrogen bond from the hydroxy function of Tyr143.Compound 43 displayed good potency but, matching its high tPSA value of 133 Å 2 , it showed almost no permeation in the Caco-2 assay and a very high efflux ratio.To try to improve permeation, we aimed for carbamoyl substituents where one hydrogen of the NH 2 functionality could be involved in an intramolecular hydrogen bond.Carbamoyl derivatives 44 and 45 were indeed reasonably permeable, as the ortho-fluoro group or orthopyridinyl nitrogen apparently ended up masking one of the highly polar carbamoyl hydrogens.Quinoxalinyl amide 46 and substituted indazol-5-yl amide 47 with reasonable potency showed more favorable Caco-2 permeation due to lower tPSA values.However, CYP3A4 inhibition and moreover timedependent inhibition (TDI) emerged as further parameters for closer consideration in the course of our project.For many nonacidic P2′ derivatives, we still observed high in vivo CL in rat, unfortunately not closely correlating with in vitro CL in rat hepatocytes.
Reassessment of P1′ Substitution.To try to improve metabolic stability, we investigated a diverse set of derivatives with different P1′ substituents more closely for their in vitro metabolism in rat and human hepatocytes.The P1′ residue turned out to be the main soft spot for metabolism for many compounds.Therefore, our amended strategy to achieve higher metabolic stability was to keep P1′ groups rather small and minimize the number of sites for metabolic attack; for example, P1′ groups like ethyl were preferred.
Reassessment of EBP Substitution.Derived from in vivo pharmacology studies, our potency target for a single stereoisomer was inhibition of FXIa activity with a low single-digit nanomolar IC 50 and prolongation of plasma clotting time in the aPTT assay with an EC 150 of less than 0.5 μM in the final volume of the assay (as described in the Experimental Section).As potency was lost by abandoning charged P2′ carboxylic acids and extended P1′ alkyl ether groups, we needed to identify other options to regain potency.Essentially, the only pocket left for further experimentation was the EBP.The WaterMap analysis (Figure 4) showed an energetically unfavorable water molecule (site 68) that could be displaced by larger substituents to gain potency.In addition, previously described compounds having a five-membered heteroaromatic ring as EBP moiety 41,42 have shown to be highly potent.Especially, an uncharged N-substituted tetrazole led to highly potent compounds.However, in our series, such derivatives were associated with low permeability and unfavorably high efflux ratios in the Caco-2 assay, explainable by the increased tPSA values (e.g., see compounds 54 and 59, Table 5).
Therefore, we evaluated a set of five-membered heteroaryl and heterocyclyl rings for EBP substitution.Figure 2 shows the details of the interaction of an EBP tetrazole residue with FXIa.The aromatic carbon (C5) points toward a small groove at the rim of the S1 pocket formed by Gly216, Glu217, and Gly218.N4 forms a water-mediated hydrogen bond with the Gly218:NH.N2 and N3 are surrounded by the aliphatic carbon side chain of Lys192.We aimed at exploring the role of the aromatic carbon (C5) and hydrogen bonding features of N4, with further substitution at position 4 to increase interactions and directly contact Gly218:NH via hydrogen bonds.We wanted to keep the tPSA low for increased permeation and to try to improve the interactions with FXIa.Results for a small set of compounds (with P1′ a methoxyethyl

Journal of Medicinal Chemistry
or ethyl residue and P2′ methylindazolyl as a standard) are shown in Table 5. Prioritization and selection of residues for synthesis were supported by relative binding free energy (FEP +) calculations using the OPLS2.1 force field 43 (see the Supporting Information).
The importance of the water-mediated hydrogen bond (tetrazole N4, see Figure 2) is further supported by the two regioisomeric oxazoles 51 and 52.The nitrogen at (tetrazole)position 4 is a significantly better hydrogen bond acceptor than oxygen.In addition, the polar isolated nitrogen at position 2 (Figure 2) is detrimental for binding as it is forced into a primarily lipophilic environment (Lys192 side-chain carbons).Both aspects together explain the 13-fold difference in potency between these two isomers.
1,3,4-Oxadiazol-2-yl derivative 53 is less potent than the corresponding oxazole 51, forcing a nitrogen into a lipophilicity preferring surrounding (tetrazole C5, Figure 2).Imidazole 56 which possesses the minimum EBP pharmacophore features (five-membered heterocycle with one strong hydrogen bond accepting atom at position 4, see Figure 2) combined high potency with low tPSA.However, this compound exhibited low Caco-2 permeability and a high efflux ratio, and therefore was not pursued further.1,2,4-Triazol-4-yl derivative 55 again lost affinity due to the lower hydrogen bonding acceptor strength of the two neighboring nitrogens (due to the interaction of the nitrogen lone pairs) in the aromatic ring.
Besides tetrazole, the best results regarding potency and anticoagulant activity were achieved with 1,2,3-triazol-1-yl groups with an electron-withdrawing substituent, such as chloro, difluoromethyl, or trifluoromethyl.The difluoro-or trifluoromethyl substituent is able to displace the structural water near the EBP observed in the complex of 2 with FXIa (see Figure 2) and form a hydrogen bond with one of the fluorine atoms and the backbone Gly218:NH.In contrast to their corresponding tetrazole, triazole derivatives 60, 62, and 64 showed significantly higher Caco-2 permeation and lower efflux ratios.1,3,4-Oxadiazol-2-yl, 1,3,4-thiadiazol-2-yl, and 1,2oxazol-3-yl groups with electron-withdrawing substitution (66−68) resulted in compounds with reasonable potency but insufficient anticoagulant activity (human aPTT EC 150 > 1 μM), albeit as racemic mixtures.1,3-Oxazol-5-yl derivative 69 is 26-fold less potent than the corresponding unsubstituted 51.We hypothesized that the additional difluoromethyl substituent forms a hydrogen bond to Gly218:NH, thereby forcing the polar isolated aromatic nitrogen into a primarily lipophilic environment.In contrast, compound 51 can adopt a 180°fl ipped orientation presenting the nitrogen to a water-mediated hydrogen bond contact (see position N4 in the corresponding tetrazole, Figure 2).The corresponding oxadiazole derivative 66 is again more potent than 69, apparently because the hydrogen bond acceptor strength and polarity of both aromatic nitrogens is reduced due their direct vicinity within the fivemembered heterocycle.A similar effect is observed for the pair 57 and its corresponding triazole 58, rendering the latter by over a factor of three more potent.
One hypothesis, that an acidic hydrogen (CH or NH) at position 5 (tetrazole C5, Figure 2) could lead to the formation of a direct hydrogen bond with the backbone carbonyl of Gly218 and thus an increased affinity, was disproved with compound 70.The tetrazole C5 surrounding is best occupied with nonpolar ligand atoms.
We concluded from our study of the EBP moiety that none of the prepared alternative five-membered heteroaryl deriva-  tives showed overall more convincing properties than the compounds with a 1,2,3-triazol-1-yl EBP residue.
Reassessment of P2′ Substitution.Having identified chloro-, trifluoromethyl-, or difluoromethyl-substituted triazole as preferred EBP options and small alkyl substituents like methyl or ethyl as favorable P1′ options, we next returned to our preferred set of P2′ substitutions (Table 6).
Quinoxalinyl amide 71 is highly potent, with an FXIa IC 50 of 0.7 nM and human aPTT EC 150 of 0.24 μM.As the quinoxaline moiety is associated with a rather low increase in  However, relevant TDI of CYP3A4 activity in human hepatocytes was not observed.After intravenous administration in rat, 71 already showed a reasonable in vivo CL (Table 7).Aiming to further improve metabolic stability, we investigated the metabolism of 71 in hepatocytes of different species and learned that in human and monkey hepatocytes, additional metabolism by aldehyde oxidase (which shows low or no expression in rat and dog) occurred.Although we introduced additional substitution at the quinoxalinyl moiety to block metabolism by aldehyde oxidase (e.g., compound 72), a compound with improved in vivo CL could not be identified.Moreover, genotoxicity studies revealed that quinoxalin-6amine, a potential metabolite of amide 71, is Ames positive.
Compared to its des-fluoro analogue 74 in the Caco-2 assay, compound 78 showed a significantly higher P app (A−B) value (88 vs 12 nm/s) and lower efflux ratio (11 vs 70).This difference could also be confirmed by rat in vivo exposure studies using oral administration, emphasizing the favorable impact of this ortho-fluoro substitution by masking one Nhydrogen of the carbamoyl group due to the formation of an intramolecular hydrogen bond.
Candidate Selection.In the final modification cycle, we studied a 3 times 3 matrix of compounds as shortlist (single enantiomers 75−83, Table 6), having chloro-, difluoromethyl-, or trifluoromethyl-substituted triazole as EBP group, methyl, ethyl, or n-propyl as P1′ group, and 2-fluorobenzamide as P2′ group.On the one hand, P1′ methyl derivatives were more potent FXIa inhibitors than their P1′ n-propyl counterparts.On the other hand, the P1′ n-propyl derivatives were favored due to their higher Caco-2 permeation and significantly lower efflux ratios compared to the P1′ methyl derivatives.Overall, the P1′ ethyl derivatives displayed properties somewhat in between.Furthermore, initial data on inhibition and induction of CYP isoforms for the P1′ methyl and ethyl derivatives did not reveal such a difference allowing for prioritization.As our main goal was to identify an FXIa inhibitor with high oral bioavailability and a human dose allowing for once-daily dosing, we focused on comparing all nine derivatives on the basis of their PK profiles in rats.Evaluation of the concentration-to-time graphs and the overall exposure (AUC norm ) after oral administration (Figure 6) revealed that the P1′ ethyl substitution was superior to methyl and propyl for each triazole substitution.Based on the concentration-totime graphs of each P1′ group, in trend, trifluoromethyl substitution at the triazole was more favorable than the other two substitutions.Focusing then on the P1′ ethyl set, difluoromethyl-substituted 79 was least potent and showed a significantly higher efflux ratio.Finally, it was decided to evaluate compound 80 (BAY 2433334, asundexian) in more depth and, after having gained favorable preclinical safety data, asundexian was moved into clinical Phase 1 studies.
At this point, a remark on the importance of the C5 methoxy group of our compound series is pertinent.Omitting the C5 methoxy group at the pyridinone core as for 84 and 85 led to a 20-to 23-fold drop in FXIa inhibitory potency compared to the corresponding 78 and 80 (Table 6), emphasizing the significant impact of this C5 methoxy group, a design element already introduced early on in the project during lead finding.
Characterization of the Clinical Candidate Asundexian (80).Asundexian was cocrystallized with human FXIa (Figure 7).Special binding features include the formation of a π-cation interaction of the P1 chloroaryl with Tyr228 displacing an energetically unfavorable water molecule, a water-mediated hydrogen bond with the central amide NH, a hydrogen bond between one of the protons of the terminal P2′ amide function and the phenolic group of Tyr143, a hydrogen bond between the said amide carbonyl function and Arg37D and a water-mediated hydrogen bond to the backbone oxygen of His414, and a hydrogen bond between one of the fluorine residues of the substituted triazole and the Gly218 backbone NH.The central ethyl group points toward the S1′ pocket and undergoes lipophilic contacts.Strong hydrogen bonds are observed between the methoxy oxygen and Gly216 and for the pyridinone carbonyl group in the OAH (Gly193, Ser195).
The in-depth evaluation 8 of asundexian revealed a potent and reversible inhibition of FXIa in buffer (IC 50 = 1.0 nM) and after contact activation in human plasma (IC 50 = 0.14 μM).Asundexian inhibited the closest homologue of FXIa, human plasma kallikrein, with IC 50 values of 6.7 nM in buffer and 1.23 μM in human plasma.This potential reduction of both enzyme activities after contact activation may have intriguing features and is currently under further pharmacological evaluation.Asundexian showed favorable selectivity of greater than 1000fold versus serine proteases associated with the hemostatic system, including FVIIa, FIXa, FXa, FXIIa, thrombin, urokinase, tissue plasminogen activator, activated protein C, or plasmin, and other proteases of potential importance related

Journal of Medicinal Chemistry
to the oral administration route, such as trypsin, chymotrypsin, and caldecrin (chymotrypsin c).
After asundexian was added to human plasma, the clotting time in the aPTT assay was prolonged with an EC 150 of 0.20 μM (calculated in the final assay volume of 150 μL) and of 0.61 μM when calculated for the concentration in plasma (50 μL, see the Supporting Information for details).While the aPTT was prolonged in rabbit, dog, pig, and guinea pig plasma samples as well, with EC 150 values of 4.5, 4.8, 1.5, and 6.4 μM, respectively, no prolongation of clotting time was observed in plasma samples from the mouse or rat (EC 150 > 30 μM).
In vivo, the antithrombotic effect of asundexian was determined in various thrombosis models.After FeCl 2 -induced injury of the rabbit carotid artery, the compound reduced the thrombus weight dose-dependently versus control animals when given intravenously in a prophylactic mode up to almost completeness at the highest doses, with an ED 50 of 380 mg/L (Figure 8A).In the simultaneously conducted ear bleeding time measurements, no impact was observed in any of the asundexian groups (Figure 8B).This profile of strong antithrombotic efficacy without increasing the bleeding time was confirmed when asundexian was coadministered with antiplatelet drugs (aspirin and ticagrelor, Figure 8C,D) and are supported by studies in venous vessels (FeCl 2 -induced injury of the rabbit jugular vein, Figure 8E) and on artificial surfaces in arteriovenous shunt models (data not shown).When administered orally to rabbits at doses of 10 and 30 mg/kg in a PEG/ethanol/water solution, asundexian reduced the thrombus weight by 30 and 91%, respectively (Figure 8F).Table 8.In Vivo Pharmacokinetic Properties of Selected Compounds in Dogs (0.3 mg/kg iv infusion and 1.0 mg/kg po) One parameter which was critical for some earlier compounds was reversible and time-dependent CYP inhibition.Coincubation of asundexian did not affect CYP1A2, CYP2A6, CYP2B6, CYP2C19, CYP2E1, CYP2J2, and CYP3A4 activities up to the highest test concentration (IC 50 > 41 μM).Weak inhibitory potential on CYP2C8 (IC 50 = 3.6 μM), CYP2C9 (IC 50 = 17 μM), CYP1A1 (IC 50 = 13 μM), and CYP2D6 (IC 50 = 19 μM) was observed.Furthermore, after preincubation (30 min) of asundexian with NADPHsupplemented human liver microsomes, a slight increase in the inhibitory potency of asundexian was only shown toward CYP3A4 (IC 50 = 17 μM).This was further investigated in human hepatocytes, and relevant TDI of CYP3A4 activity was not observed.
In Caco-2 cells, asundexian showed a high permeability [P app (A−B) = 143 nm/s] and a moderate efflux ratio of 7. The fraction unbound of asundexian in plasma was determined using the equilibrium dialysis method with 3 H-labeled drug substance and is moderate in rat (2.4%), monkey (6.2%), and human (6.4%) plasma and is high in dog (10%) and rabbit (14%) plasma.
First Results on Clinical Studies with Asundexian (80).In the Phase 1 program, no clinically relevant bleeding events or impact on bleeding times was observed.Asundexian showed a consistent PK/PD relationship, with dose-dependent changes in pharmacodynamic parameters such as anticoagulant activity (aPTT) and FXIa inhibitory activity, while bleeding times were consistent across all dose cohorts and similar to those of the placebo.−46 Therefore, asundexian is a promising clinical candidate as a once-daily oral anticoagulant and has been investigated for its efficacy and safety in three Phase 2 dose finding studies, namely, PACIFIC-AF 18 (NCT04218266) in patients with atrial fibrillation, PACIFIC-STROKE 19  Currently, the OCEANIC program is ongoing with two Phase 3 studies investigating the efficacy and safety of asundexian in the prevention of stroke events in patients with atrial fibrillation as well as in patients with a noncardioembolic ischemic stroke or high-risk transient ischemic attack, involving up to 30,000 patients (OCEANIC-AF and OCEANIC-STROKE).The U.S. Food and Drug Administration (FDA) has granted Fast Track designation for asundexian as a potential treatment for secondary prevention in patients with a noncardioembolic ischemic stroke.
Chemistry.The synthesis route of asundexian (80) used during the research phase comprises eight linear steps and a chiral separation as the last step (Scheme 1).Starting from commercially available 2,5-dimethoxypyridine, deprotonation with LDA, addition of triisopropyl borate, and subsequent ester hydrolysis resulted in boronic acid 86 in 72% yield.Boronic acid 86 was converted into bromide 87 in the presence of copper(II) bromide in 65% yield.As direct bromination of 2,5-dimethoxypyridine resulted in undesired regioisomeric mixtures, this two-step approach was used to obtain the desired bromide regioisomer.The methoxy group adjacent to the pyridyl nitrogen was selectively deprotected with pyridine hydrobromide to give 88 in 69% yield.Alkylation of 88 with tert-butyl 2-bromobutanoate afforded racemate 89 in 53% yield.Best to use in this step is the tert-butyl ester, as the methyl or ethyl esters are too prone to saponification due to activation by the pyridinone substitution.The benzyl ester also works, and both the tert-butyl and benzyl ester can be already deprotected under basic conditions (e.g., with lithium hydroxide).Pinacol boronic ester formation (→ 90) with subsequent Suzuki coupling of 1-(2-bromo-4-chlorophenyl)-4-(trifluoromethyl)-1H-1,2,3-triazole (91) 47 resulted in compound 92 in 59% yield for the two steps.After ester deprotection, acid 93 was coupled with 4-amino-2-fluorobenzamide in the presence of propylphosphonic anhydride (T3P) to yield compound 94 in 84% yield.A last step, chiral separation of racemate 94 resulted in eutomer 80 (asundexian).

■ CONCLUSIONS
Using protein structure-based de novo design, we identified a novel micromolar hit as FXIa inhibitor with attractive physicochemical properties.The main parameters we attempted to improve upon were potency, permeability, metabolic stability, and cytochrome P450 interaction profile.A careful design approach focused on understanding the compound's conformational behavior and introducing functional groups that (1) undergo strong interactions with the target protein, (2) rigidify the structure to offer the FXIa preformed conformations, exactly fitting in the active site without losing entropy upon binding, and (3) keep the polar surface low.Only a focus on the most important hydrogen bond donor and acceptor interactions enabled a balanced combination of potency and absorption.
The first compound that reached clinical trials was a highly potent inhibitor with a carboxylic acid moiety.The inclusion of that moiety solved many problems with cytochrome P450 interaction but the compound ultimately suffered from low oral exposure and a short half-life in humans.Active transport issues were suspected to be related to the acidic moiety.
In a second attempt, the carboxylic acid moiety was replaced by carboxamides capable of forming an intramolecular hydrogen bond.The challenge was to regain potency and to simultaneously attempt to improve the metabolic stability and cytochrome P450 interaction profile on top of high potency and permeability.Several learning cycles were needed to identify the most promising substituents for addressing each of the protease pockets, S1, EBP, S1′, and S2′, within one compound.The resulting compound from our extensive research program, asundexian (BAY 2433334, 80), combines high potency and selectivity with excellent oral bioavailability, a long half-life, and a favorable CYP interaction profile.Asundexian shows antithrombotic efficacy in arterial and venous thrombosis models in prevention and intervention settings, without increasing bleeding.Completed clinical Phase 1 trials 45,46 and the PACIFIC Phase 2 program (PACIFIC-AF, 18 PACIFIC-STROKE, 19 PACIFIC-AMI 20 ) on asundexian have confirmed the desired DMPK properties and the initial pharmacological hypothesis.The OCEANIC Phase 3 studies to investigate asundexian in patients with atrial fibrillation (OCEANIC-AF) or after stroke (OCEANIC-STROKE) have

Journal of Medicinal Chemistry
been initiated.The U.S. FDA has granted Fast Track designation for asundexian as a potential treatment for secondary prevention in patients with a noncardioembolic ischemic stroke.

■ EXPERIMENTAL SECTION
We describe the synthesis of compound 80 (asundexian, BAY 2433334) here in detail, referring to a route used during the research phase.Meanwhile, a process synthesis has also been disclosed. 48The synthesis of all other derivatives is described in the Supporting Information.
General Procedures.All commercial reagents and catalysts were used as provided by the commercial supplier without purification.Solvents for synthesis, extraction, and chromatography were of reagent grade and used as received.Moisture-sensitive reactions were carried out under an atmosphere of argon, and anhydrous solvents were used as provided by the commercial supplier.Preparative normal-phase flash chromatography was performed using Biotage Isolera chromatography systems with Biotage silica cartridges or silica gel 60 (230−400 mesh) in combination with glass columns/frits.Preparative reversed-phase (RP) chromatography was performed on 125/250 mm × 20/30/40 mm HPLC columns packed with YMC gel ODS-AQ S-5/15 μm and UV detection.Gradients or isocratic mixtures used as eluents are indicated.All compounds tested in biological assays were of ≥95% purity, as determined by HPLC, LC/ MS, or NMR data.
1 H NMR and 13 C NMR spectra were recorded at RT with Bruker Avance spectrometers.Chemical shifts (δ) are reported in ppm relative to TMS as an internal standard.The descriptions of the coupling patterns of 1 H NMR signals are based on the optical appearance of the signals and do not necessarily reflect the physically correct interpretation.In general, the chemical shift information refers to the center of the signal.In the case of multiplets, intervals are given.
Analytical mass spectrometry was performed on HPLC/MS (Waters, Agilent, Thermo Fisher) or GC/MS (Waters, Agilent, Thermo Fisher) systems using Waters Time-of-Flight, Waters/ Micromass Single Quadrupole, or Thermo Fisher Scientific Orbitrap mass spectrometers.Ionization methods were electrospray ionization (ESI) positive/negative or electron ionization (EI).LC/MS and GC/ MS analyses were performed using the respective method 1−3 or 21, as noted.
Assessment of optical rotation [α] was performed using an Anton Paar polarimeter MCP2000 with parameters (wavelength, temperature, optical pathway, solvent, and concentration) as indicated.

Figure 2 .
Figure 2. X-ray cocrystal structure of compound 2 28 in complex with human FXIa (PDB code 8BO6) and the corresponding 2D sketch.The magnification shows the S1 site occupied by the chloroaryl residue and the interactions of the neutral tetrazole in the EBP site.

Figure 3 .
Figure 3. Structure-based de novo design concept leading to hit 3 and comparison of the docked compound (magenta, pre-synthesis) and the X-ray cocrystal structure in complex with human FXIa (green, PDB code 8BO5) after experimental verification.

Figure 4 .
Figure 4. WaterMap hydration sites in the substrate binding pocket of FXIa (PDB code 8BO5).Sites are color-and size-coded with regard to the modeled hydration free energy (high ΔG, reddish; low ΔG greenish).The depicted compound in green is the basic framework of the lead series; the white arrows show the possibilities for substitutions to increase interactions with FXIa.

Figure 5 .
Figure 5. X-ray cocrystal structure of compound 34 in complex with human FXIa (PDB code 8BO7) and the corresponding 2D sketch.

a
The heteroaromatic EBP substituents are drawn corresponding to the supposed orientation with FXIa, see Figure2.b n.d.: not determined.

Figure 8 .
Figure 8. Antithrombotic effects and impact on hemostasis of compound 34 and asundexian (80) in vivo in rabbit models: (A) dose-dependent thrombus weight reduction after intravenously administered 34 (blue), asundexian (green), and rivaroxaban (magenta) vs control (gray) following FeCl 2 -induced damage to the carotid artery (arterial thrombosis model) with aPTT prolongation factors given for each dose group, and (B) simultaneously determined impact on bleeding time prolongation after a defined ear incision; (C and D) antithrombotic effects and impact on bleeding time prolongation compared to apixaban (red bars) in the models described in (A) and (B), with the additional presence of aspirin (ASA) and ticagrelor; (E) dose-dependent thrombus weight reduction following FeCl 2 -induced damage to the jugular vein (venous thrombosis model); (F) dose-dependent impact of orally administered compound 34 and asundexian on the thrombus weight compared to control in the model as described in (A); n = 4−8, mean + SD, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.
(NCT04304508) in patients with a n o n c a r d i o e m b o l i c s t r o k e , a n d P A C I F I C -A M I 2 0 (NCT04304534) in patients with acute myocardial infarction.

Table 1 .
Modification of EBP and P1′ Substitution

Table 5 .
Modification of EBP Substitution

Table 6 .
Candidate Selection

Table 7 .
In Vivo Pharmacokinetic Properties of Selected Compounds in Rats (0.3 mg/kg iv bolus and 1.0 mg/kg po) a Oral bioavailability.b n.d.: not determined.