Antithrombotic Effects of Montelukast by Targeting Coagulation Factor XIa

Current oral anticoagulants prescribed for the prevention of thrombosis suffer from severe hemorrhagic problems. Coagulation factor XIa (FXIa) has been conrmed as a safer antithrombotic target as intervention with FXIa causes lower hemorrhagic risks. In this study, by a high-throughput virtual screening, we identied Montelukast (MK), an oral antiasthmatic drug, as a potent and specic FXIa inhibitor (IC 50 = 0.17 µM). Compared with the two mostly prescribed anticoagulants (Warfarin and Apixaban), MK demonstrated comparable or even higher antithrombotic effects in three independent animal models. More importantly, in contrast to the severe hemorrhage caused by Warfarin or Apixaban, MK did not measurably increase blood loss in vivo. In addition, MK did not affect the hemostatic function in plasma from healthy individuals. In contrast, MK suppressed clot formation in clinical hypercoagulable plasma samples. This study provides a lead compound of anticoagulants targeting FXIa, and suggests the exploratory clinical researches on antithrombotic therapies using MK.


Introduction
Vascular uidity is maintained by the equilibrium among platelet adhesion, coagulation, anticoagulation, and brinolysis 1 . Thrombosis (e.g. deep venous thrombosis, myocardial infarction, ischemic stroke, etc) is caused by the excessive activation of the coagulation system overwhelming the relatively insu cient anticoagulation and brinolytic systems leading to the obstruction in vessels by thrombi. Anticoagulants, intervening with the coagulation cascade and slow down brin deposition in vessels, are widely used for the preventions and treatments of venous thromboembolism (VTE) and for the prevention against coronary artery diseases (CADs) accompanied with antiplatelets. The most prescribed oral anticoagulants are vitamin-K-antagonists (VKAs) and the directly-acting-oral-anticoagulants (DOACs).
VKAs reduce vitamin K activity required for the post-translational modi cation of coagulation factors 2 .
DOACs are orally available inhibitors of coagulation factor Xa (FXa) or thrombin 3 . A common but signi cant adverse effect of these anticoagulants is the narrow therapeutic window: over-anticoagulation results in an inherently high risk of severe bleeding, while insu cient anticoagulation leads to thrombosis 4 .
The ideal anticoagulant is considered to attenuate thrombosis while maintaining hemostatic functions. FX(a) and (pro)thrombin are simultaneously critical to thrombosis and hemostasis 5 . VKAs and DOACs directly or indirectly intervene with the activation, modi cation, or activity of (pro)thrombin and/or FX(a), thus impairing the hemostatic function and resulting in hemorrhage. Coagulation factor XI(a) (FXI/FXIa) was originally recognized to facilitate the generation of thrombin in the intrinsic cascade, a supplement of the tissue factor (TF)-triggered extrinsic cascade 6 . Clinically, patients with congenital FXI de ciency show no or mild hemorrhagic problems 7,8 . Clinical evidences have con rmed that high levels of FXI in plasma are closely associated with thrombotic pathologies, including VTE 9 , ischemic stroke 10 , and myocardial infarction 11 . The role of FXI(a) in thrombosis can be complex. First, FXI, activated by thrombin or FXIIa, ampli es thrombin generation by sequentially activating FIX and FX 12 . Besides, in CADs, rupture of an atherosclerotic plaque releases negatively charged substances activating FXII and FXI to enhance thrombin generation 13 . Furthermore, FXIa suppresses congenital anticoagulation by degrading tissue-factor-pathway-inhibitor (TFPI) 14 , the physiological inhibitor of TF:FVIIa complex and FXIa. The above functions of FXI(a) are considered to be less relevant to hemostasis but crucial in thrombosis, because the ampli cation of thrombin generation enhances thrombi's resistance against brinolysis.
Thus, compared to (pro)thrombin or FX(a), FXI(a) is considered as a safer antithrombotic target to avoid severe hemorrhages. Although several molecules targeting FXI(a) are currently in clinical trials, only BMS-986177 is potentially orally available for the long-term prevention against thrombosis. However, the potential toxicities and side-effects have not been con rmed by the post-market surveillance. Thus, retargeting prescriptive drugs to FXI(a) is a strategy to achieve FXIa inhibitors with well-studied pharmacological properties and safeties. In this study, we identi ed Montelukast (MK), an FDA-approved antiasthmatic drug, as a potent FXIa inhibitor through a combination of in silico screening and experimental validation (IC 50 = 0.17 µM).
MK alleviates asthmatic symptoms by blocking cysteinyl leukotriene receptors (cys-LTRs) and thus suppressing smooth muscle contraction in the airway 18 . MK has been prescribed for the prevention of chronic asthmatic symptoms and for the relief of seasonal and perennial allergic rhinitis 19,20 . Patients administered with MK generally demonstrate good tolerance, and only few of them showed mild sideeffects 21 . MK has never been associated with coagulation or antithrombotic treatments, though subcutaneous bruising or mild nosebleed was observed as a rare side-effect, similar to the side-effect of anticoagulants, indicating that MK avoids impairing hemostatic functions.
The inhibitory potency of MK against FXIa, identi ed in this study, suggests its antithrombotic property. In addition, MK demonstrated speci city to FXIa showing non-measurable inhibition against homogenous proteases, especially thrombin and FXa, which are critical to hemostatic functions. Next, the molecular mechanisms of MK's inhibition and selectivity were revealed using molecular dynamics (MD) simulations. After con rming that MK inhibited FXIa-induced clot formation in vitro, we used three in vivo thrombotic models to evaluate MK's antithrombotic properties and compared with those of two clinically used anticoagulants, Warfarin (a VKA) and Apxaban (Apx, a DOAC). In addition, the hemorrhagic risks caused by MK, Warfarin, and Apx were evaluated and compared in a mouse tail bleeding model. We also studied the in uence of MK on the global coagulation in plasma from healthy individuals and clot formation in clinical hypercoagulable plasma samples.

Results
In silico screening Based on the most representative structure of the catalytic domain of FXIa obtained from molecular dynamics (MD) simulations on its crystal structure (PDB ID: 4CRG 21 ), we carried out a docking-based virtual screening against a customized drug repurposing library (3,727 compounds, Figure 1A) using Libdock and CDOCKER modules implemented in Discovery Studio 2017 R2 Client. We achieved 5 potential compounds according to the consensus analysis on seven different scoring functions and visual inspection ( Figure S1). By the validation with the enzymatic assay, we found that MK inhibited the activity of FXIa.

Molecular mechanism of MK's inhibition and speci city
To reveal the molecular mechanisms of MK's inhibitory potency and selectivity, we performed molecular docking and atomistic MD simulations on the complexes of MK bound to FXIa, PK or FXa ( Figure 3). The FXIa:MK complex was structurally stable and showed no large conformational changes over 500 ns, i.e. the root-mean-square deviation (RMSD) of MK was constantly below 1.5 Å, in contrast to the RMSDs of MK bound to PK or FXa uctuating 1.5-3.0 Å ( Figure S2A), suggesting the higher stability of MK:FXIa complex. Such distinctive dynamic features were also supported by the superimpositions of time course snapshots obtained from MD simulations ( Figure 3A-C). Notably, MK stably persisted within the active site of FXIa with one dominant conformation, while the superimposed MK in PK and FXa showed varied conformational states, which aligned with the weak a nity determined by enzymatic experiments ( Figure   2B). Thus, the virtual stability analysis was in agreement with our experimental determinations. Figure 3D-F depicted the binding poses of MK to FXIa, FXa and PK, respectively. In both FXIa:MK and PK:MK complexes, MK's chloroquinoline P1 moiety deeply inserted into the S1 pocket, anchored by halogen bond and halogen-π interactions between the chloride atom and the phenolic group of Y228. In contrast, MK failed to establish such interaction with FXa. Detailed structural comparisons suggested that the failure may be due to the occlusion by the indole ring of W215 which was engaged in stacking interactions with Y99, F174 and one phenyl ring of MK ( Figure 3G and 3H). Compared to FXIa and PK, the W215 indole ring in FXa was ipped by ~120º (from -60º to 60º, Figure 3I), so that its sidechain directed toward the 2-phenylpropan-2-ol moiety of MK, making the chloroquinoline moiety stuck at the rim of S1 pocket. Intriguingly, bimodal distribution of the ipping dihedral angle (C-Cα-Cβ-Cγ of W215) was observed in the PK:MK complex. Such dihedral angle signi cantly increased from -60º to 60º at ~120 ns and then sharply decreased back to -60º at ~300 ns ( Figure S3A). The distance between the phenyl ring of 2-phenylpropan-2-ol moiety of MK and the indole ring of W215 also exhibited large uctuations at the same time periods ( Figure S3B). Therefore, the formation of stacking interaction between W215 and MK is closely related to the ipping rotation of W215 sidechain. As a result, the co-existed ipped conformation of W215 in the PK:MK complex disrupted the stability of chloroquinoline moiety at the S1 pocket ( Figure S3C). Consistently, the halogen-related interactions in FXIa:MK were stronger than those in PK:MK, as indicated by the plots of the centre-of-mass distance between the phenolic group of Y228 and the chloride atom ( Figure 3J). Besides, the reduced stability of chloroquinoline moiety may also attribute to the loss of additional hydrogen bond formed with one of the catalytic triad residue S195. Furthermore, two additional hydrogen bonds were established with K192 and Y143 in FXIa, providing the desired boost in binding a nity.

MK inhibited the FXIa-induced clot formation in vitro
We next evaluated MK's effect on clot formation in human plasma with an addition of recombinant FXIa ( Figure 2C). Citrate-stored plasma was pre-incubated with 5 nM recombinant FXIa and MK (0-25 μM) for 15 min. Clot formation was monitored by real-time determining the absorbance at 405 nm (A405) after the addition of 2 mM CaCl 2 . The time to reach half of the plateau (T 1/2 ) was used to evaluate the anticoagulative effect of MK ( Figure 2D). T 1/2 of normal plasma (the control group) was 9.5±1.3 min, which was signi cantly shortened till 4.6±0.1 min by the addition of FXIa. The presence of 1, 5, or 25 μM MK delayed clot formation, and T 1/2 was prolonged to 5.4±0.1, 7.5±0.6, or 8.3±0.8 min, respectively. This result demonstrates the inhibition of the FXIa-induced coagulation by MK.

MK suppressed the FeCl 3 -stimulated carotid artery thrombosis in vivo
The in vivo antithrombotic properties of MK were evaluated in three thrombotic mouse models in vivo. For comparison, we parallelly studied the effects of two clinically prescribed oral anticoagulants, Warfarin and Apx ( Figure 1B). First, we studied the antithrombotic effects of these anticoagulants in an FeCl 3stimulated carotid artery thrombosis mouse model ( Figure 4A-B). Saline, MK, warfarin, or Apx were administered (p.o.) 3-h before the aesthetic mice's carotid arteries were exposed and treated with a lter paper saturated with 6% FeCl 3 to trigger clot formation. Blood ow was real-time imaged and recorded by a laser speckle imaging instrument. The time-to-occlusion (TTO) was recorded to quantitatively compare the vascular occlusion among different groups. TTO in the untreated group was 287 s, which was prolonged to 363 s (P<0.05) or 678 s (P<0.01) by the administration with 2 mg/kg or 10 mg/kg MK, respectively. Administration with 4 mg/kg warfarin or 10 mg/kg Apx increased TTO to 512 s (nonstatistical signi cance) or 642 s (P<0.05), respectively. Warfarin was administered at a low dose (4 mg/kg) because of its high toxicity to mice 23 .

MK suppressed electricity-stimulated carotid artery thrombosis in vivo
The antithrombotic properties of MK, Warfarin, and Apx were also evaluated in the electricity-stimulated carotid artery thrombosis mouse model ( Figure 4C). Similarly, all anticoagulants were administered (p.o.) 3-h before the surgical exposure of the carotid artery and the stimulation with 0.1 mA electricity. The blood ow was real-time recorded by an infrared detector immediately after stimulation. TTO was recorded when the blood ow went below 5% and continuously kept for >5 s. The vessels of the untreated mice were occluded at 82 s after stimulation. With the administration of 4 mg/kg warfarin, 2 mg/kg, or 10 mg/kg Apx, the TTO was prolonged to 160 s (P<0.01), 145 s (non-statistical signi cance), or 210 s (P<0.05), respectively. Administration with 2 mg/kg or 10 mg/kg MK signi cantly delayed the vascular occlusion by increasing TTO until 135 s (P<0.05) or 202 s (P<0.01), respectively. Thus, in both carotid artery thrombotic models, MK performed antithrombotic effects comparable to those of the two mostly prescribed anticoagulants.

MK ameliorated the LPS-induced pulmonary micro-thrombus formation in vivo
We next evaluated the antithrombotic effects of MK, Warfarin, or Apx on in ammation-induced pulmonary micro-thrombus formation in vivo. The thrombosis in pulmonary vessels was stimulated by intraperitoneal injection with LPS in mice. Four hours before LPS-stimulation, mice were orally administrated with saline, MK (2 or 10 mg/kg), Warfarin (4 mg/kg), or Apx (10 mg/kg), respectively. One group without LPS-stimulation was set as the normal group for comparison. Four hours after stimulation, mice were sacri ced, and the lungs and blood were harvested for histopathological analysis and the determination of cytokine levels, respectively ( Figure S4). Pulmonary histopathological analysis demonstrated that LPS-stimulation caused severe in ammation and thrombosis in lungs from the salinetreated group ( Figure 5A). Treatments with MK, Warfarin, or Apx all ameliorated the micro-thrombi formation ( Figure 5B-E). All the treated groups showed signi cantly reduced numbers and sizes of thrombi compared to the saline-treated group ( Figure 5F). We also evaluated the effects of MK, Warfarin, or Apx on the in ammation levels by determining the plasma levels of IL-1β or TNF-α ( Figure 5G-H).
Consistent with the known anti-in ammatory effects of MK 24,25,26 , administration with 10 mg/kg MK reduced the levels of both IL-1β and TNF-α to the normal levels. However, despite Warfarin and Apx ameliorated pulmonary thrombosis, they did not show anti-in ammatory effects. The decreased TNF-α level by Warfarin was by impairing the transcription of TNF-α rather than by anti-in ammatory property 27,28 . Thus, based on its anti-in ammatory and antithrombotic properties, MK might exhibit a dual therapeutic effects on pathologies combining in ammation and thrombosis, e.g. sepsis-induced disseminated intravascular coagulation (DIC).

MK caused mild hemorrhage in vivo
Next, we evaluated the potential hemorrhage caused by MK, Warfarin, and Apx using a tail-truncation mouse model (Figure 6). Both bleeding time and hemoglobin loss were recorded. Compared to the salinetreated mice, Mice treated with 4 mg/kg warfarin showed a doubled bleeding time and a 3.7-fold enhanced hemoglobin. In contrast, neither MK nor Apx increased the bleeding time at the dose of 10 mg/kg. At the concentration of 50 mg/kg, MK and Apx prolonged the bleeding time for 1.46-fold and 1.35-fold, respectively. However, although Apx-and MK-treated mice showed similar bleeding time ( Figure  6A), MK caused much lower hemoglobin loss than Apx did ( Figure 6B): At the concentrations of 10 and 50 mg/kg, MK only caused around 1.3-fold enhanced hemoglobin loss compared to the saline-treated group. In contrast, at the same concentrations, Apx caused 2.9-and 3.5-fold enhanced hemoglobin loss, respectively. This result indicates that, compared to VAKs or DOACs, MK, as a FXIa inhibitor, does not impair the hemostatic function and causes much milder hemorrhage, which is also consistent with the fact that mild bleeding is a rare side-effect of MK.

MK did not affect the global coagulation
The in uences of MK on global coagulation were evaluated by determining prothrombin time (PT), activated partial thromboplastin time (APTT), and thromboelastographic parameters in plasma from heathy individuals (Table 1). MK showed non-measurable effect on all of the tested parameters even at a high concentration up to 1000 μM (>5000-fold IC 50  showed an APTT EC2× concentration >3000-fold higher than its K i value 34 . Similarly, a pyrimidine-based FXIa inhibitor showed an APTT EC1.5× concentration ~5000-fold higher than its IC 50

MK suppressed coagulations in clinical plasma samples
Thrombophilia is dangerous because blood has a high tendency to coagulate in vessels. We evaluated the effects of MK on the coagulation of 20 clinical hypercoagulable plasma samples which showed shortened APTT and enhanced D-dimer levels (Table S3). These plasma samples were collected from individuals in different situations, e.g. late pregnancy, hypotension, acute myocardial infarction (AMI), etc. CaCl 2 was used to trigger the coagulation in the samples with or without 25 μM MK. Coagulation was monitored by real-time recording A405 ( Figure 7A-B). The amount of brin generation was quantitatively represented by the plateaued A405 subtracting the starting A405 (A 405 (+MK) or A 405 (-MK)) ( Figure 7A).
The time when clot formation reached half of the plateau was recorded as T 1/2 (+MK) or T 1/2 (-MK). The  Figure S5), because the inhibition of FXa directly inhibited the generation of thrombin ( Figure S4). In contrast, MK, a selective FXIa inhibitor, only reduced the amount of brin generation, which is consistent with the less important role of FXI(a) in normal coagulation (hemostasis). In such in vitro setup, the inhibition of FXIa only reduced the ampli cation of thrombin generation while had no effect on existing thrombin. Besides, the inhibition of TFPI by FXIa does not affect the already generated thrombin either. Thus, FXIa is not essential in the in vitro coagulation, while can be critical for thrombosis, because thrombosis needs resistance against anticoagulation and brinolysis in vivo 38 . Thus, although MK did not completely suppress clot formation like Apx did in vitro, our animal models demonstrated the close antithrombotic e cacies of MK and Apx.

Discussion
In this study, we employed an in silico structure-based approach to screen a drug repurposing library and successfully identi ed MK as a potent and selective inhibitor of FXIa. MK was originally developed for treating asthma, a paroxysmal air ow obstruction caused by chronic airway in ammation. Historically, asthma was considered irrelevant to thrombosis, which normally occurs within blood vessels. As the cross-talk between coagulation and in ammation is now being revealed, increasing evidence has demonstrated that brin formation in pulmonary compartments and the upper airway can be important in asthmatic development 39 . First, brin assembles with mucus forming plugs obstructing the airway. In addition, brinogen cleavage products (FCPs), generated from the cleavage by either fungal proteases or thrombin, can trigger the toll-like receptor 4 signaling pathway in epithelial cells leading to in ammation and thus boosting the generation of eosinophilia and mucus in the airway 40 . Besides, patients with asthma showed 2-5-fold higher risk of pulmonary embolism compared to the general population 41 . FXI(a) is responsible for the generation of thrombin and thus may also associate with the aggravation of asthmatic symptoms. The marketed antiasthmatic drug, MK, was originally developed to block leukotriene receptors and inhibit smooth muscle contraction. Thus, MK's antiasthmatic function might also be relevant to its anticoagulative effect revealed in this study.
Because of the interactions with foods and drugs, VKAs have caused the highest incidence of drugrelated life-threatening events 5,42 . In contrast, no drug interactions have been reported for MK suggesting a more versatile drug/food combination. After its approval in 1998, MK show generally mild adverse effects according to the post-marketing surveillance 43 . However, leukotriene modulators are associated with increased incidences of agitation, aggression, anxiousness, irritability, or even suicidal behaviors 44 .
Accordingly, US-FDA required a boxed warning for MK to strengthen an existing warning about the risk of neuropsychiatric events 45 . Thus, although MK overcomes the main problems of typical anticoagulants, hemorrhage and interactions with drugs/foods, the neuropsychiatric risk might hinder the long-term administration of MK. However, its well-studied pharmacology and toxicology may be bene cial to exploratory researches on the off-label antithrombotic use of MK.
It seems to be a coincidence that MK can simultaneously inhibit FXIa and cys-LTRs. The crystal structures of cys-LTR-1 in complexes with two leukotriene antagonists were recently reported 46 . Based on the large difference in the structures of cys-LTR-1 and FXIa, MK can be engineered to enhance its inhibitory potency against FXIa promoting antithrombotic e cacy, and to reduce its a nity to cys-LTRs ameliorating neuropsychiatric risk. Thus, MK can be a promising lead compound for generating more potent oral anticoagulants with low hemorrhagic and neuropsychiatric risks.
Our in silico models suggested that the neutral chlorinated aromatic ring at the P1 moiety is pivotal for MK's inhibitory activity. Similar moieties have been also successfully applied in the structures of oral inhibitors of FXIa 47 , FXa 48, 49 , and thrombin 50 . Based on structural comparisons, we found that the sidechain orientation of W215 may be a major determinant for the selectivity of MK. The exibility of W215 segment is an intrinsic nature of trypsin-like protease and the conformational state of W215 determines the accessibility of substrates or ligands to the S1 pocket 51,52 . Previous crystallographic analysis and biochemical assays indicated that W215 may exist in two conformational states (open and closed), which can be allosterically regulated by distal residues 53 . Consistent with this scenario, we observed that the binding of speci c compounds, like MK investigated here, may selectively turn on the switch of W215 conformation, which then inversely controls the stability of compound binding. It seems that the extent of conformational change is dependent on particular amino acids around W215 as well as the tendency for interaction between the compound and W215. Therefore, it provides a novel strategy for the design of FXIa-targeted anticoagulants by selectively regulating the conformation of key active site residues.
In this study, by a high-throughput virtual screening, we identi ed MK as a potent and selective inhibitor of FXIa, a safer antithrombotic target compared to (pro)thrombin and FX(a). MD simulations elucidated its mechanisms of the inhibition and speci city.  (Table S1).
In with 250 μg/kg LPS and 500 mg/kg D-(+)-Galactosamine in ICR mice (male, 6-8 weeks) 55 . Mice were randomly divided into 6 groups (6 mice per group). One group without stimulation was set as the normal group. The rest 5 groups were administered (p.o.) with saline, 2, 10 mg/kg MK, 4 mg/kg Warfarin or 10 mg/kg Apx 3-h before LPS-stimulation. 4-h after stimulation, mice were anesthetized and sacri ced. The blood and lungs were harvested. Blood concentrations of IL-1β and TNF-α, were determined by enzymelinked immunosorbent assay (ELISA) kits. Histopathological sections of lungs were stained with hematoxylin and eosin (H&E) and photographed at a magni cation of 200× imaged by a Leica DMi8 inverted microscope. The numbers of microthrombi were counted in 6 histopathological sections from 3 mice in each group, and the areas of the microthrombi were determined by the software of Image J.
Tail transection mouse model. Tail bleeding time was determined as described previously with slight modi cations 56 . Mice were p.o. administered with saline, 10, 50 mg/kg MK, 4 mg/kg Warfarin, 10 and 50 mg/kg Apx (6 mice per group), 3 h before the tails were transected 10 mm from the tip. The transected tails were immediately immersed in 10 ml isotonic saline at 37 °C. The bleeding time was recorded once the bleeding stopped. The blood loss was quanti ed by measuring the haemoglobin content collected in the 10 ml isotonic saline. After centrifugation, erythrocytes were collected and lysed with 2 mL lysis buffer (8.3 g/L NH 4 Cl, 1 g/L KHCO 3 , and 0.037 g/L EDTA). The A575 of each sample was determined.
Statistical analysis. The statistical signi cance was analyzed by using 1-way ANOVA followed by Bartlett test. A P value of less than 0.05 was considered statistically signi cant.       represented as mean ± SEM. *P < 0.05, **P < 0.01, versus the saline-treated group (N=6).