## Introduction

During percutaneous coronary intervention (PCI), balloon angioplasty is frequently followed by stent implantation. The procedure causes extensive endothelial disruption and injury, leading to an intense burst of thrombin generation1,2,3. As such, patients are routinely pre-treated with dual antiplatelet therapy (DAPT) and administered an injectable anticoagulant, commonly unfractionated heparin (UFH), during the PCI procedure3,4. The limitations of UFH include major bleeding, heparin-induced thrombocytopenia (HIT) and the need for coagulation monitoring due to its unpredictable pharmacokinetics. However, UFH remains one of the most widely used parenteral anticoagulants due to low cost (~$4–$10 USD per PCI5) and the wealth of clinical experience accumulated in more than eight decades of use6. To overcome some of UFH’s disadvantages, an injectable direct thrombin inhibitor (DTI), bivalirudin, was developed as an alternative to UFH in PCI3,4,7. Although bivalirudin is more expensive (~$400 to$600 per PCI without post-procedural infusion5), initial randomised trials showed that bivalirudin was associated with similar antithrombotic efficacy but less bleeding when compared with a combination of UFH and a platelet glycoprotein IIb/IIIa inhibitor8,9,10. However, protocol-mandated use of glycoprotein IIb/IIIa inhibitors in the UFH arm of these trials could have contributed to the higher bleeding rates among patients randomised to UFH. More recent trials with balanced use of glycoprotein IIb/IIIa inhibitors in both bivalirudin and UFH arms have shown less favourable results for bivalirudin11,12,13,14. Therefore, in the 2018 European Society of Cardiology/European Association for Cardio-Thoracic Surgery guidelines on myocardial revascularization, routine use of UFH received a higher recommendation (class I) than bivalirudin (class IIb) for peri-PCI anticoagulation15. Bivalirudin has been more widely adopted in the United States than in the rest of the world. One study estimated that 47.1% of PCI cases between July 2009 and December 2014 in the US used bivalirudin (52.9% UFH) but also noted a decline in bivalirudin use after 201316. With increasing use of potent platelet P2Y12 antagonists such as ticagrelor, prasugrel and cangrelor, there remains an even greater unmet is a need for safer peri-PCI anticoagulants to adequately improve the efficacy-safety balance of antithrombotic therapy during PCI7,17. Combination antiplatelet and anticoagulant therapy has become more common especially among patients with atrial fibrillation or venous thromboembolism undergoing PCI15,18.

Haematophagous animals such as leeches, mosquitoes, ticks, tsetse flies and others are rich sources of antithrombotic agents19. Molecules such as hirudin from the medicinal leech20, anophelin from mosquitoes21,22, madanin from ticks23 and tsetse thrombin inhibitor from tsetse flies24,25 are potent and selective thrombin inhibitors that are highly amenable to customisation as synthetic inhibitors with improved properties25,26,27,28. We have previously characterised a new class of DTIs, variegin and other variegin-like peptides, from the salivary gland of the tropical bont tick (Amblyomma variegatum)29,30. Variegin, a 32-residue peptide, binds to both thrombin’s active site and exosite-I. Variegin inhibits thrombin more potently than bivalirudin by sixfold and demonstrates superior selectivity for thrombin29,31,32. Bivalirudin has a half-life of 25 min8 and must therefore be administered as a continuous infusion during PCI as the majority of these procedures last ~30 min to an hour. Variegin has a half-life of 52.3 ± 4.4 min33 and may potentially be given as a single bolus for peri-PCI anticoagulation.

In this study, we hypothesise that a high-affinity, high-specificity bivalent DTI with non-covalent binding to thrombin like variegin would sufficiently prevent thrombus formation during PCI when administered in low doses in combination with aspirin and ticagrelor (DAPT); this low-dose approach to peri-PCI anticoagulation would then preserve the ability to regenerate thrombin when bleeding occurred. Through iterative design and optimisation, we develop a picomolar affinity DTI, named ultravariegin. We perform global coagulation experiments in plasma to better understand the antithrombotic efficacy versus preservation of haemostatic capacity by taking into consideration the interaction of DTIs with platelets and DAPT. In dose-ranging studies using rodent models of carotid artery thrombosis and tail bleeding, we compare the therapeutic indices of variegin and ultravariegin with UFH and bivalirudin. Then, in an ex vivo porcine model of coronary stent thrombosis and venous bleeding, we test the efficacy and safety of these anticoagulants in the absence and presence of DAPT. Finally, we identify specific antidotes for ultravariegin through a screen against a naïve human antibody library and test them against ultravariegin in vitro and in vivo.

## Results

### Design of ultravariegin, a picomolar thrombin inhibitor, from the tropical bont tick A. variegatum

Variegin-like thrombin inhibitors in Amblyomminae are synthesised as larger precursor proteins containing multiple repeats that are post-translationally processed into shorter peptides30. Potential thrombin inhibitor precursors in A. variegatum were identified from its salivary gland transcriptome34 (Fig. 1). Proteomic data and sequence alignment suggest that these peptides have serine residue at the amino terminus29,30. However, the post-translational cleavage site remains unidentified29,30. Considering the variability in sequences, there is uncertainty about the identity of carboxyl-terminal amino acids of these inhibitors (Fig. 1). We synthesised three peptides representing repeats from one of the precursors (GenBank accession number DAA34688.1 repeat 1, 1B and 1C) with different possible C-termini (Supplementary Figs. 114 and Supplementary Table 1). These peptides potently inhibited thrombin’s amidolytic activity on chromogenic substrate S2238 with significant inhibition at equimolar concentrations of inhibitor and thrombin with linear inhibition progress curves (Fig. 2a). This is consistent with the previously reported fast- and tight-binding behaviour of variegin-like peptides29,30. Truncation of amino acids from the C-terminus resulted in progressively stronger inhibition as indicated by the corresponding inhibitory constant (Ki) values (Table 1). The Ki value for DAA34688.1 repeat 1C is 6.5 ± 1.3 pM, which is 42-fold and 96-fold stronger than synthetic variegin and avathrin, respectively (Table 1). We hypothesised that mutating the non-conserved Thr22 in DAA34688.1 repeat 1C to Glu would enrich negative charges needed for binding to thrombin exosite-I. Consistent with this design, the Ki improved to 4.0 ± 0.5 pM (Table 1 and Fig. 2b). We named this mutant ultravariegin, which has a Ki for thrombin that is 445-fold greater than bivalirudin (Table 1). The apparent Ki ($${K}_{{{{{{\mathrm{i}}}}}}}^{{\prime} }$$) of ultravariegin increased linearly with increasing concentration of thrombin chromogenic substrate, indicating that ultravariegin was a competitive inhibitor of thrombin with respect to thrombin’s active site (Fig. 2c).

### Novel sequences located at the C-terminal of ultravariegin impart high affinity towards thrombin

As reported earlier, the N-terminal segment of variegin (residues 1–7) interacts with thrombin through long-range electrostatic steering. Residues 8–14 of variegin target thrombin’s active site and the C-terminal segment binds to thrombin’s exosite-I29,30,31,32. Thrombin-inhibiting peptides found in A. variegatum (Fig. 1) show the highest sequence variability within the C-terminal region. In UV003, UV004 and UV005, we substituted ultravariegin residues in segments to the sequence of variegin to investigate which local segment of ultravariegin is most likely responsible for improved activity over variegin. Substitutions in the C-terminal segment, predicted to target thrombin’s exosite-I, is most intolerable. Ki of ultravariegin increased fourfold to 16 pM in UV005 (Table 1). In contrast, replacement of the N-terminal segment (UV003) and thrombin’s active site binding segment (UV004) in ultravariegin with corresponding sequences of variegin resulted in minimum changes in Ki. The sensitivity of ultravariegin activity to changes in C-terminal residues was also demonstrated by UV012, whereby the A27E mutation increased the Ki > fivefold. Overall, ultravariegin was the most potent peptide (Table 1 and Fig. 2b).

We have previously reported that thrombin cleaves variegin between the Lys10-His11 scissile bond but the cleavage product C-terminal to the scissile bond (sequence: MHKTAPPFDFEAIPEEYLDDES) is a non-competitive inhibitor of thrombin’s active site with a Ki of ~14.1 nM31. We synthesised the equivalent cleavage product of ultravariegin UV011 and showed that it inhibits thrombin (Table 1 and Fig. 2d). Assuming the same non-competitive mode of inhibition, the Ki of UV011 is 1.66 ± 0.76 nM. Bivalirudin is also cleaved by thrombin upon binding31,35 and we synthesised the equivalent cleavage product of bivalirudin (BV001) (Table 1). Surprisingly, BV001 was not only unable to inhibit thrombin but instead activate thrombin’s active site by up to 20% when tested at concentrations higher than 1 µM (Fig. 2d).

Ultravariegin was then screened for selectivity against 11 serine proteases involved in blood coagulation and fibrinolysis, as well as trypsin and chymotrypsin. At 0.1 nM, ultravariegin inhibited thrombin at around 27%, whereas even at 100 µM, ultravariegin did not inhibit any other serine protein by more than 20%, indicating at least 1,000,000-fold selectivity in preference for thrombin over other serine proteases (Fig. 2e). Thus, ultravariegin appeared to be a viable lead molecule with a substantially optimised Ki at 4 pM and enhanced selectivity for thrombin.

### Haemostatic capacity of blood is largely preserved with variegin or ultravariegin in combination with DAPT

We then compared variegin and ultravariegin with UFH and bivalirudin. The activated partial thromboplastin time (aPTT) was dose-dependently prolonged by all four compounds (Fig. 3a). Molar estimates of the concentrations of UFH used were derived to allow for comparisons with the other three compounds on the same axis. Among the three DTIs, ultravariegin appeared to be most potent, followed by variegin and bivalirudin, consistent with their respective affinity towards thrombin (Table 1). We then performed clot waveform analysis (CWA), in which successive derivatives of the clotting curves indicate activity of individual coagulation enzymes or complexes. The first derivative (min1) represents thrombin activity (i.e., thrombin burst) and bleeding risk (low absolute value is associated with greater bleeding risk)36,37, the second derivative (min2) represents the activity of the prothrombinase complex and the third derivative (min3) represents activity of the tenase complex38. All three derivatives dose-dependently decreased with each of the four anticoagulants tested. However, the rates declined very sharply above 25 nM of UFH while a decline only occurred when variegin, ultravariegin and bivalirudin were above 1 µM (Fig. 3b).

In the thrombin-generation test (TGT), lag-time (LT) and time-to-peak (TTpeak) for thrombin generation generally increased with increasing concentrations of all four anticoagulants in platelet-poor plasma (PPP) or platelet-rich plasma (PRP), with or without DAPT (Supplementary Fig. 15). However, key differences were observed in the rate of increase in LT and TTpeak with increasing concentrations of anticoagulants. The dose–response changes of variegin and ultravariegin were gradual compared with UFH and bivalirudin (Fig. 3c and Supplementary Fig. 16). Across all conditions, gradients of LT for variegin and ultravariegin were 1.9- to 6.4-fold lower than for UFH and bivalirudin; gradients of TTpeak for variegin and ultravariegin were 2.3- to 13.7-fold lower than for UFH and bivalirudin (Supplementary Table 2). Relative to UFH and bivalirudin, increases in variegin and ultravariegin concentration result in a more moderate inhibition of thrombin by gradually delaying the thrombin burst.

Endogenous thrombin potential (ETP) for variegin, ultravariegin and bivalirudin were largely maintained around the baseline, indicating the capacity to generate thrombin was not fully abrogated by the three DTIs (Supplementary Fig. 16). The peak thrombin and velocity index (VI) were increased for variegin across a 100-fold concentration range (0.01–1 µM) in PPP and PRP without DAPT, but this increase was attenuated in PRP with DAPT (Fig. 3d, Supplementary Fig. 16 and Supplementary Table 2). For ultravariegin, the thrombin peak and VI showed the most gradual change across a 1000-fold concentration range (0.001–1 µM) in PPP and PRP (Fig. 3d, Supplementary Fig. 16 and Supplementary Table 2). In contrast, ETP, peak thrombin and VI were severely impaired in plasma spiked with UFH, indicating inhibition of thrombin production that did not recover within the timeframe of the experiment (Supplementary Fig. 16). The dose–response curves for UFH were steep, with near-complete inhibition achieved within the small (eightfold) concentration range tested (0.025–0.2 µM) (Fig. 3d, Supplementary Fig. 16 and Supplementary Table 2). ETP, peak thrombin and VI increased within the small dose range (fourfold) of bivalirudin tested (1.5–6 µM) under all conditions, suggesting a potential risk for excessive thrombin generation on the rebound (Fig. 3d, Supplementary Fig. 16 and Supplementary Table 2). Taken together, these results suggest that thrombin-generation capacity for haemostasis was best preserved when ultravariegin, and to a lesser extent, variegin, was used in combination with DAPT.

### Variegin and ultravariegin showed a wider therapeutic index than UFH and bivalirudin in a rodent model

In a rat model of FeCl3-induced carotid artery thrombosis, the time taken for complete carotid artery occlusion increased dose-dependently with all four anticoagulants. A single 5 mg/kg IV bolus injection of variegin resulted in an occlusion time close to 60 min (maximum observed duration) (Fig. 4a). Consistent with its lower Ki value for thrombin inhibition, a lower dose of ultravariegin at 2 mg/kg fully prevented carotid artery occlusion (Fig. 4b). This maximum level of antithrombotic efficacy was similarly observed with clinically approved doses of UFH and bivalirudin. For UFH, this therapeutic dose was a bolus IV injection of 432 U/kg which translates to a human equivalent dose of 70 U/kg (Fig. 4c). For bivalirudin, it was 10.8 mg/kg/h of IV infusion, which translates to a human equivalent dose of 1.75 mg/kg/h (Fig. 4d).

In the tail bleeding model, bleeding time increased dose-dependently for all four compounds. The occlusion time and bleeding time curves for variegin and ultravariegin were well-separated, suggesting a wider therapeutic index (TI) than either UFH or bivalirudin. With the estimated therapeutic dose of variegin (5 mg/kg) and ultravariegin (2 mg/kg), the bleeding time was only at around half of the maximum bleeding time (~30 min). In contrast, the therapeutic dose of UFH and bivalirudin resulted in maximal bleeding (Fig. 4c, d). The dose–response curves for antithrombotic efficacy (occlusion time) and safety (bleeding time) largely overlapped for UFH, indicating a very low therapeutic index, while the same set of curves for bivalirudin showed minor separation. We then estimated the dose of each compound that resulted in a 50% maximum response time (RT50; 30 min) for occlusion time and bleeding time experiments (Fig. 4). The TIs for each compound were calculated as the dose at RT50 for bleeding time divided by occlusion time. The TIs for UFH and bivalirudin were 1.0 and 1.3, respectively, while the TIs for variegin and ultravariegin were 3.8 and 6.7, respectively, around three- to sevenfold better than UFH and bivalirudin (Supplementary Table 3). The wider TIs observed with variegin and ultravariegin are consistent with the preceding in vitro human blood experiments in which inhibition of coagulation is effective but gradual, and haemostatic capacity is preserved.

We further validated the results of the tail bleeding model with a different rodent bleeding model, the saphenous vein bleeding model39,40. We tested variegin, ultravariegin, UFH and bivalirudin at their respective doses that resulted in a 50% maximum response time of occlusion in the carotid artery thrombosis model (RT50; 30 min). In the saphenous vein bleeding model, a higher number of haemostatic events indicate better preservation of haemostatic capacity39,40. We showed that the number of haemostatic events ranked as follow: ultravariegin > variegin > bivalirudin > UFH (Fig. 4e). The average time per bleeding event had an identical rank order as tail bleeding time experiments: UFH > bivalirudin > variegin > ultravariegin (Fig. 4f). The bleeding time from both models showed strong, positive correlation (r = 0.978) (Fig. 4f, insert), indicating consistent results from both bleeding models.

### Variegin or ultravariegin is more effective in preventing stent thrombosis than UFH and bivalirudin

We then compared the efficacy of the four anticoagulants in an ex vivo porcine model of stent thrombosis in the absence and presence of DAPT (Supplementary Fig. 17). This model investigates thrombosis when blood is exposed to thrombogenic tissue under flow conditions, and has been commonly employed for antithrombotic drug testing41,42,43,44,45. Simultaneously, we compared the safety of these compounds by performing bleeding time experiments on a superficial ear vein of the same animals. Without anticoagulation (saline), the stents were completely occluded with thrombus within 60 min of blood flow through the extracorporeal circuit (Fig. 5a, b). With 1 mg/kg of variegin, the therapeutic dose translated based on rodent experiments, thrombus formation was reduced by ~83% (Fig. 5a, b). Near-complete thrombus reduction was achieved with pigs receiving 0.25 mg/kg of ultravariegin (Fig. 5a, b). The administration of therapeutic doses of UFH and bivalirudin reduced thrombus formation by ~38% and 61%, respectively (Fig. 5a, b). Compared with saline, all anticoagulants increased bleeding time by 1.9- to 2.4-fold although only the difference between variegin and saline achieved statistical significance (4.2 min vs 10.3 min, multiplicity-adjusted P value = 0.0376) (Fig. 5c).

### Low doses of variegin or ultravariegin in combination with DAPT are highly efficacious in preventing stent thrombus formation without increasing bleeding time

When DAPT (300 mg aspirin and 180 mg ticagrelor loading dose) was orally administered 16 h prior to the experiments, in the absence of anticoagulant, thrombus formation on both the aortic strip and stents was reduced significantly compared with saline (10.8 g vs 83.8 mg, two-tailed Student’s t test P ≤ 0.0001), implying that DAPT was partially effective in preventing PCI-related stent thrombosis (Figs. 5a, b and 6a, b). Given the high affinity of variegin and ultravariegin for thrombin, we hypothesised that their combination with DAPT should provide enough antithrombotic efficacy at a much reduced dose of either variegin or ultravariegin, thereby avoiding excessive inhibition of haemostasis. We, therefore, reduced the dose of variegin and ultravariegin by 10-fold to 0.1 mg/kg and 0.025 mg/kg, respectively, for pigs pre-treated with DAPT. Consistent with our hypothesis, the low dose of variegin and ultravariegin reduced thrombus formation from DAPT alone by 87% and 75%, respectively (Fig. 6a, b). In comparison, the clinically approved doses of UFH and bivalirudin reduced thrombus formation from DAPT alone by 66% and 76%, respectively (Fig. 6a, b).

Low-dose variegin (7.7 ± 2.6 min) or ultravariegin (6.2 ± 1.3 min) with DAPT did not increase the bleeding time compared to DAPT without anticoagulation (6.4 ± 3.7 min). In contrast, UFH or bivalirudin with DAPT both resulted in the maximally observed bleeding time of 45 min. Figure 6c shows that, in the presence of DAPT, UFH or bivalirudin resulted in at least fivefold higher bleeding times than either variegin or ultravariegin. Taken together, on a background of DAPT, low-dose variegin or ultravariegin have marginally greater efficacy in preventing stent thrombosis than UFH or bivalirudin but with substantially less bleeding.

### Reversal agents for ultravariegin were identified from a naive human antibody library

The availability of a reversal agent would provide an additional safeguard against bleeding complications from anticoagulation. We, therefore, explored the potential of generating ‘active and specific’ reversal agents for ultravariegin. Screening against a naive human antibody library, seven antibodies were found to bind to biotinylated ultravariegin. These clones were expressed and purified as IgG antibodies. These antibodies dose-dependently reversed ultravariegin’s inhibition of thrombin’s amidolytic activity (Fig. 7a). Almost complete reversal of 0.5 nM ultravariegin occurred with two antibodies (Ab1282 and Ab1283). The binding affinity between biotinylated ultravariegin and Ab1282 or Ab1283, measured by biolayer interferometry (BLI), was 1.25 and 1.40 nM (mean of two independent experiments), respectively (Fig. 7b, c and Supplementary Fig. 18). In contrast, Ab1283 did not show binding to the control peptide with a scrambled non-functional ultravariegin sequence, suggesting binding of Ab1283 to ultravariegin is sequence-specific (Supplementary Fig. 19). Next, the reversal activity of Ab1283 was tested in vivo. Rats that were injected with 0.75 mg/kg ultravariegin and followed by 10 mg of Ab1283 showed a statistically significant reduction in tail bleeding time (12.7 min vs 16.8 min, P = 0.0078) compared to those receiving saline in place of Ab1283 (Fig. 7d).

## Discussion

Despite the availability of direct oral anticoagulants (DOACs), anticoagulation for the acute coronary syndrome (ACS) and PCI still require parenteral anticoagulants that are effective and safe7. Concerns about an unfavourable efficacy-safety profile had led to reduced interest in thrombin as an anticoagulant target6. Our study revisits thrombin as a therapeutic target and identifies drug affinity-dose relationships as a means of optimising the therapeutic window of DTIs. Using complementary approaches including activity-based purification29, cDNA amplification and data mining of tick saliva transcriptomes30, we identified multiple potent and specific inhibitors of thrombin from A. variegatum. Starting with variegin, we progressively improved and developed ultravariegin, a 30-amino acid peptide with Ki = 4.0 pM with at least 1,000,000-fold selectivity for thrombin over other serine proteases. Compared with UFH and bivalirudin, variegin, and particularly ultravariegin, showed more gradual inhibition of thrombin generation across a broad range of drug concentrations in human blood, suggesting a capacity to regenerate thrombin for haemostasis when needed. In a series of efficacy and safety experiments in small and large animals, comparing variegin and ultravariegin with UFH and bivalirudin, we established that variegin and ultravariegin had better antithrombotic efficacy with a shorter bleeding time when used at low doses in combination with DAPT. Importantly, variegin and ultravariegin achieved this therapeutic advantage with a single bolus injection at a fraction of the molecular weight-adjusted dose of a bivalirudin infusion. Finally, two specific reversal agents showing nanomolar level binding affinity to ultravariegin were identified. In vivo reversal experiments of rats anticoagulated with ultravariegin demonstrated that Ab1283 reduced bleeding time compared with control, representing a potential lead for antidote development.

Our results have several implications for peri-PCI anticoagulant drug development. First, we confirm that potent DAPT (aspirin and ticagrelor) without anticoagulation cannot sufficiently protect against stent thrombosis in an ex vivo porcine model. Second, thrombin remains a viable target for peri-PCI anticoagulation in the presence of DAPT. Here we showed that short, peptidic DTIs could be structurally optimised to inhibit thrombin with Ki values 6- (variegin) and 445-fold (ultravariegin) better than bivalirudin, respectively. The strong affinity and specificity of variegin and ultravariegin allow for aggressive reduction of DTI dose in the presence of potent DAPT.

For any given therapeutic, high selectivity and affinity for the intended target are often crucial in achieving efficacy with minimum side effects46. In this regard, variegin and ultravariegin are substantially more advantageous than UFH and bivalirudin. Mechanistically, UFH is a heterogenous compound, indirectly inhibiting thrombin through antithrombin-III and is non-selective, resulting in unpredictable pharmacodynamics and pharmacokinetics47. These limitations are not present in specific DTIs including bivalirudin, variegin and ultravariegin. Previously, we have reported the novel class of DTIs from A. variegatum have some similarities to bivalirudin in structure but with several distinct features. Several structural–functional properties of variegin-like peptides may account for these differences from bivalirudin in pharmacokinetic-pharmacodynamic relationships. For example, variegin-like peptides contain specific amino acid residues within the active site and extended amino- and carboxyl-termini leading to stronger binding of thrombin. In contrast, bivalirudin has a non-specific, flexible tetra-glycyl linker between the active site and exosite-I binding sequence, resulting in loss of activity upon cleavage by thrombin29,30,31,32. Using a pair of peptides, UV011 and BV001, we have demonstrated continual, albeit weaker inhibition of thrombin after cleavage of ultravariegin, in contrast to complete loss of inhibition by bivalirudin after cleavage (Fig. 2d). In historical clinical trials, bivalirudin has previously been shown to increase the risk of myocardial infarction and stent thrombosis compared with UFH while the reduction in bleeding with bivalirudin might have been a result of unbalanced use of glycoprotein IIb/IIIa inhibitors with more patients in the UFH group receiving glycoprotein IIb/IIIa inhibitors than in the bivalirudin group48. More contemporary trials suggest that acute stent thrombosis may be mitigated with a prolonged infusion of full-dose bivalirudin post-PCI12,14,49,50. These more recent trials also had more balanced use of glycoprotein IIb/IIIa inhibitors across treatment groups and higher use of platelet P2Y12 antagonists. With the exception of VALIDATE-SWEDEHEART, these trials showed that bivalirudin was associated with slightly less major bleeding compared with UFH with no difference in ischaemic events, indicating that thrombin remains an appropriate target for peri-PCI anticoagulation. The issues with bivalirudin appear to be related to efficacy and hence the dose during post-PCI infusion and concurrent use of antiplatelets are critical.

Considering the important role of platelets in arterial thrombosis, this approach of combining a low-dose, high-affinity anticoagulant with DAPT may achieve high antithrombotic efficacy while minimising bleeding risk during PCI. In the porcine model, UFH plasma levels corresponded to ~0.6 µM based on clotting time assays (Supplementary Figs. 20 and 21). The plasma concentration of bivalirudin was reported to be around 2.7 µM for PCI51. For 0.1 mg/kg variegin used in combination with DAPT, the average plasma concentration was ~30 nM33. Since ultravariegin was given at a fourfold lower dose compared with variegin, the plasma concentration of ultravariegin was estimated to be 7.5 nM. At low doses, variegin and ultravariegin appeared more potent than UFH and bivalirudin (Fig. 3a) without impacting the rate of synthesis of various enzymes/complexes important to haemostasis, thus causing less bleeding (Fig. 3b). In contrast, 0.6 µM UFH and 2.7 µM bivalirudin resulted in severe impairment of coagulation (e.g., min1 was less than half of the control); a decrease in the absolute value for min1 is clinically associated with an increase bleeding risk (Fig. 3b)36. The TGT also showed that the anticoagulant intensity of UFH and bivalirudin increased sharply over a very narrow concentration range (Fig. 3, Supplementary Fig. 16 and Supplementary Table 2), suggesting a tendency to excessively suppress haemostasis with just small increases in their plasma concentration. By comparison, variegin or ultravariegin produced a more moderated increase in anticoagulation response over a wider concentration range. In in vitro assays, cleavage of ultravariegin results in two species of thrombin inhibitors: ultravariegin with higher affinity (Ki = 4.0 pM) and UV011 with 415-fold weaker affinity (Ki = 1660 pM). Hypothetically, UV011 may ‘compete’ with ultravariegin for thrombin binding and hence an overall moderated increase in anticoagulation response was observed. In contrast, bivalirudin’s cleavage product does not bind to thrombin, and any increase or decrease in the concentration of bivalirudin is sharply reflected in its anticoagulation intensity (Fig. 3, Supplementary Fig. 16 and Supplementary Table 2). Furthermore, in our rodent models, variegin/ultravariegin also achieved a three- to sevenfold wider therapeutic index compared with UFH/bivalirudin.

Our study has limitations. First, the experiments using human blood were performed in vitro. Although we used thrombin-generation parameters that are widely validated to predict bleeding in clinical settings52, the in vitro findings may not fully reflect thrombin regeneration capacity in vivo. Second, we used established small and large animal models to test the efficacy and safety of the four compounds but the results will need subsequent verification in human trials. Pre-clinical animal bleeding models may not always predict clinical bleeding risks in human trials because of a variety of reasons53,54. Here, we used three different bleeding models (tail, saphenous vein and superficial ear vein) to increase confidence in the results. Third, we did not perform dose optimisation of UFH and bivalirudin when used together with DAPT, since we aimed to validate variegin/ultravariegin in comparisons with clinically approved doses of UFH and bivalirudin. It remains possible that dose optimisation will result in a more balanced efficacy-to-bleeding profile of UFH or bivalirudin. Fourth, although the two antibodies were able to fully reverse the thrombin inhibition effect of ultravariegin in vitro, the in vivo studies showed that large doses of Ab1283 may be required for complete reversal, hence the molecule may require further optimisation as an antidote.

In conclusion, we discovered a unique class of bivalent, non-covalent DTIs from tick saliva transcriptomes and showed that optimisation of drug affinity-dose relationships can vastly improve the efficacy-safety balance of DTIs in ACS and PCI. Compared with UFH and bivalirudin, this new class of DTIs achieved greater efficacy at preventing ex vivo stent thrombosis but caused far less bleeding. We also demonstrated proof-of-concept in antidote development by identifying two antibody antidotes that effectively reversed the effect of ultravariegin in vitro. These results require corroboration in clinical trials to determine if the salutary effects of this relatively simple affinity-dose optimisation approach is translatable to humans.

## Methods

### Study approvals

All animal experiments were approved and conducted in accordance with the guidelines of the National University of Singapore (NUS) Institutional Animal Care and Use Committee (R16-008 and R15-0165). All human studies were approved by the NUS Institutional Review Board (B-15-094) and in compliance with the declaration of Helsinki. All participants gave written informed consent prior to study participation.

### Peptide synthesis and purification

Peptides were synthesised using an automated microwave peptide synthesiser (CEM, NC, USA). The C-terminal amino acid was loaded to Cl-MPA ProTide resin (CEM, NC, USA). Subsequent coupling of amino acids (0.2 M) were performed using 0.5 M N,N’-diisopropylcarbodiimide as activator and 0.1 M N,N’-diisopropylethylamine in 2 M Oxyma as activator base. Fmoc deprotection was achieved with 10% w/v piperazine in ethanol:N-methyl-2-pyrrolidone (1:9). Cleavage of the synthesised peptides were performed at room temperature for 3 h in the following cocktail: trifluoroacetic acid (TFA)/triisopropylsilane/water/dioxa-1,8-octane-dithiol (92.5:2.5:2.5:2.5). Peptides were precipitated using cold diethyl ether. Purification of peptides was performed on Jupiter® 4 µm Proteo 90 Å (250 × 21.2 mm) reversed-phase column (Phenomenex, CA, USA). The purity and masses of peptides were determined on AerisTM 3.6 µm Widepore XB-C18 100 Å (150 × 4.6 mm) reversed-phase column (Phenomenex, CA, USA) and by electrospray ionisation mass spectrometry (ESI-MS) using an LCQ Fleet Ion Trap MS and Thermo Xcalibur 2.2 software (Thermo Fisher Scientific, MA, USA) (Supplementary Figs. 114 and Supplementary Table 1). Mass spectra were deconvoluted using ProMass for Xcalibur 3.0 software. Concentrations of peptide solutions were estimated using UV absorbance at 280 nm and the extinction coefficient was calculated from the peptide sequence. For the two peptides without Tyr (avathrin and UV004), we measured absorbance at 205 nm and estimated the concentrations based on standard curves constructed using peptides of identical length and similar sequences with known concentrations (e.g., ultravariegin and UV003). Ultravariegin and a peptide with scrambled ultravariegin sequence were synthesised with an additional Cys at the N-terminal for conjugation to biotin using EZ-linkTM iodoacetyl-LC-biotin (Thermo Fisher Scientific, MA, USA) using standard protocols recommended by the manufacturer.

### Inhibition of thrombin amidolytic activity

Assays for the inhibition of thrombin (Haematologic technologies, VT, USA) amidolytic activity on chromogenic substrate S2238 (Chromogenix, NY, USA) by peptides were used to estimate half-maximal inhibitory concentration (IC50) and inhibitory constant (Ki). All assays were performed in 50 mM Tris buffer (pH 7.4), 100 mM NaCl, and 1 mg/ml BSA at room temperature using 0.83 nM thrombin, varying concentrations of peptides and 100 µM S2238. The rates of formation of product p-nitroaniline were followed at 405 nm for 10 min with a InfinitePro M200 microplate reader using Tecan Magellan 7.0 software (Tecan, Mannedorf, Switzerland). Dose–response curves were fitted using Prism 6.0 software (GraphPad, CA, USA) to calculate IC50 values with a logistic sigmoidal equation:

$$y={A}_{2}+({A}_{1}-{A}_{2})/(1+{[x/{x}_{0}]}^{H})$$
(1)

where y is the percentage of inhibition, A2 and A1 are the right and left horizontal asymptote, respectively, x is log10 of the inhibitor concentration, x0 is the point of inflection and H is the slope of the curve.

When an enzyme was inhibited by an equimolar concentration of inhibitor, the binding of the inhibitor to enzyme caused a significant depletion in the concentration of free inhibitors. To determine the apparent inhibitory constant, $${K}_{{{{{{\mathrm{i}}}}}}}^{{\prime} }$$, the following tight-binding equation was considered:

$${v}_{s}=({v}_{o}/2{E}_{{{{{{\mathrm{t}}}}}}})({\{{[{K}_{{{{{{\mathrm{i}}}}}}}^{{\prime} }+{I}_{{{{{{\mathrm{t}}}}}}}-{E}_{{{{{{\mathrm{t}}}}}}}]}^{2}+4{K}_{{{{{{\mathrm{i}}}}}}}^{{\prime} }{E}_{{{{{{\mathrm{t}}}}}}}\}}^{1/2}-[{K}_{{{{{{\mathrm{i}}}}}}}^{{\prime} }+{I}_{{{{{{\mathrm{t}}}}}}}-{E}_{{{{{{\mathrm{t}}}}}}}])$$
(2)

where vs is the steady-state velocity in the presence of inhibitor, vo is the velocity observed in the absence of inhibitor, Et is the total enzyme concentration, It is the total inhibitor concentration and $${K}_{{{{{{\rm{i}}}}}}}^{{\prime} }$$ is the apparent inhibitory constant.

For competitive inhibition, the inhibitory constant, Ki, is related to $${K}_{{{{{{\rm{i}}}}}}}^{{\prime} }$$ by this equation:

$${K}_{{{{{{\mathrm{i}}}}}}}^{{\prime} }={K}_{{{{{{\mathrm{i}}}}}}}(1+S/{K}_{{{{{{\mathrm{m}}}}}}})$$
(3)

where $${K}_{{{{{{\mathrm{i}}}}}}}^{{\prime} }$$ increases linearly with S, Ki is the inhibitory constant, S is the concentration of substrate and Km is the Michaelis–Menten constant for S2238.

For non-competitive inhibition, the inhibitory constant, Ki, is equal to IC50.

### Serine protease specificity

The selectivity profile of ultravariegin was examined against 13 serine proteases: fibrinolytic serine proteases (plasmin, TPA and urokinase), anticoagulant serine protease-activated protein C (APC), procoagulant serine proteases (FXIIa, FXIa, FXa, FIXa, FVIIa, kallikrein and thrombin) and classical serine proteases (chymotrypsin and trypsin). Effects of ultravariegin on these serine proteases were determined by inhibition of their amidolytic activities assayed using chromogenic substrates specific for the respective enzymes. The final concentrations of proteases and substrates used are given in parentheses in nM and µM, respectively, unless mentioned otherwise: α-thrombin/S2238 (0.81/100), trypsin/S2222 (0.87/100), fIXa/Spectrozyme® fIXa (333/0.4), fXIa/S2366 (0.125/1000), fXa/S2765 (0.24/650), chymotrypsin/S2586 (1.2/0.67), tPA/S2288 (36.9/1000), fVIIa/S2288 (460/1200), plasmin/S2251 (3.61/1200), APC/S2366 (2.74/600), kallikrein/S2302 (0.93/1100), urokinase/S2444 (32 U/ml/650), fXIIa/S2302 (20/1000). The activity of thrombin was tested at lower concentrations of ultravariegin (10 nM, 1 nM and 0.1 nM) compared with other proteases (100 μM, 10 μM and 1 μM).

### Estimation for the molar concentration of UFH

Three vials of 5000 unit/ml UFH from Leo Pharma (Ballerup, Denmark) were individually lyophilised to measure the average dry weight of UFH after subtracting stated additives. The average dry weight of UFH obtained after subtraction of the weight of excipients, was 136.3 mg, which is equal to 183.5 unit/mg UFH, consistent with reported values in the literature55. Assuming the average molecular weight of 10900 g/mol as reported56, the equivalent molar concentrations of UFH were converted from U/ml in order to compare UFH with variegin, ultravariegin and bivalirudin, where data were plotted on the same axis.

### Plasma preparations

Blood was collected by venipuncture from three healthy male donors into 3.2% trisodium citrate tube. Platelet-poor plasma (PPP) was obtained by centrifugation at 2000 × g for 10 min. Platelet-rich plasma (PRP) was obtained by centrifugation of the blood within an hour of sample collection at 180 × g for 10 min. The platelet count of PRP was adjusted to 150,000–200,000 platelets/µL with autologous PPP.

### Activated partial thromboplastin time (APTT) and clot waveform analysis (CWA)

APTT was performed on PPP using the Sysmex CS-5100TM (Kobe, Japan) using Actin FSL APTT reagent (Siemens Healthcare Diagnostics, IL, USA). PPP were spiked with variegin, ultravariegin, UFH, or bivalirudin for APTT according to the standard conditions recommended by the manufacturer. The final concentrations of aspirin and ticagrelor were 333 µg/ml and 3.33 µg/ml, respectively. Changes in light transmittance were recorded to determine the clot waveform57. Clotting times were as reported and the clot waveform was differentiated to derive the first, second and third derivatives (min1, min2 and min3, respectively)38,57 using CWA analysis software IPU version 18.

### Thrombin-generation test (TGT)

Thrombin generation was determined using the Calibrated Automated Thrombogram (Diagnostica Stago, Asnières-sur-Seine, France) method in accordance with the manufacturer’s instructions. In all, 980 µl of PPP or PRP were spiked with 10 µl of variegin, ultravariegin, UFH, or bivalirudin, and 10 µl of either DAPT (final concentrations 333 µg/mL aspirin and 3.33 µg/mL ticagrelor) or vehicle (5% dimethyl sulfoxide) accordingly. Preliminary dose titrations of aspirin and ticagrelor for inhibition of platelet aggregation in whole blood were performed using the Multiplate® Analyzer (Roche Diagnostics, Basel, USA). The lowest concentrations of aspirin and ticagrelor to produce more than 90% inhibition of platelet aggregation were selected for TGT. Thrombin generation of PRP and PPP was triggered using 1 pM and 5 pM tissue factor, respectively. Thrombin-generation curves were calculated using the Thrombinoscope 5.0 software (Thrombinoscope, Limburg, Netherlands).

Change in various thrombin-generation parameters with increasing doses differed among the four anticoagulants. The dose–response was plotted with double logarithmic axes to demonstrate linear relationships in anticoagulant intensity (represented by various TGT parameters) with changes in dose. Data were fitted by regression in Prism 6.0 according to the following equation:

$$Y={10}^{\wedge }[m\ast \,\log (X)+C]$$
(4)

where Y is either LT, TTpeak, peak, ETP or VI, X is the concentrations of anticoagulants, m is the slope and C is the Y-intercept. The slopes calculated from curve fittings are recorded in Supplementary Table 2.

### Rats

Male Sprague-Dawley rats (InVivos, Singapore) were housed in microisolator cages and were kept on a 12-h light/dark cycle with constant temperature and humidity. Rats were anaesthetised with a mixture of ketamine:xylazine (75:10 mg/kg body weight) and maintained with inhaled 1% isofluorane in oxygen throughout the experiment. Catheters (23 G) were inserted into the left femoral vein and artery, for drug injection and blood collection, respectively.

### Rat carotid artery thrombosis model

Rat carotid artery thrombosis model58 was performed as followed: five minutes after an i.v. bolus injection (saline, variegin, ultravariegin, or UFH) or 15 min after initiation of continuous i.v. infusion (bivalirudin), a 2 × 5 mm piece of filter paper (soaked in 4 µl 50% FeCl3 solution) was placed on the surface of the carotid artery for 10 min. Blood flow through the common carotid artery was monitored with a Doppler flow probe (Model MA1PRB, Transonic System Inc., NY, USA) and recorded using LabChart 7 Pro (ADInstruments, CO, USA). Occlusion time was defined as the time taken after FeCl3 application for the blood flow to reach zero. The experiment was terminated after 60 min. Occlusion time was recorded as the maximal time of 60 min if no occlusion occurred by this time.

### Rat tail incision bleeding model

Bleeding time59 was measured in anaesthetised rats as followed: five minutes after an i.v. bolus injection (saline, variegin, ultravariegin, or UFH) or 15 min after initiation of a continuous i.v. infusion (bivalirudin), a spring-loaded blade device (Surgicutt Adult bleeding time device, ITC, USA) was applied longitudinally on the ventral surface of the tail to make an incision (1 mm depth × 5 mm length) at 9–9.5 cm from the tip of the tail. The filter paper was used to blot blood from the side of the wound (without touching the wound) every 15 s. Bleeding time was defined as the time after incision until the cessation of bleeding on eight consecutive blots. The experiment was terminated at 60 min after tail incision. Bleeding time was recorded as 60 min if bleeding did not stop by this time.

This model was also used for testing the in vivo reversal activity of Ab1283. Four minutes after rats were injected with an i.v. bolus of 0.75 mg/kg ultravariegin, an i.v. bolus of 10 mg of Ab1283 or saline was injected. One minute later, an incision was made on the tail and bleeding time was measured as described above.

### Rat saphenous vein bleeding model

The saphenous vein bleeding model40,60 was performed in rats as followed: the right saphenous vein was exposed and covered with normal saline to prevent drying. Five minutes after an i.v. bolus injection (saline, variegin, ultravariegin, or UFH) or 15 min after initiation of continuous i.v. infusion (bivalirudin), a 23-G needle was used to pierce a hole in the right saphenous vein followed by a longitudinal incision of ~1 mm. Blood was gently wicked away every 15 s until haemostasis occurred. The clot formed was then gently removed using a 30-G needle to restart bleeding. The process of clot disruption was repeated after every incidence of haemostasis. The number of haemostasis events (i.e., clot formation) after repeated clot disruptions within a 30 min period were recorded. The average time per bleeding event were calculated by dividing 30 min with the number of clots (i.e., bleeding time).

### Pigs

Both male and female SPF pigs (40–70 kg) were locally farmed for research purpose and obtained from the SEMC (SingHealth Experimental Medicine Centre, Singapore). All pigs were initially sedated with a mixture of ketamine (12 mg/kg), midazolam (0.5 mg/kg) and atropine (44 µg/kg). Endotracheal intubation was performed and anaesthesia was initiated with 5% isofluorane for three min, followed by 2% isoflurane for the duration of the experiment. Animals were ventilated using an Aespire ventilator machine (GE Healthcare, UK). Plasma replacement (6% volume) was given as a continuous infusion for the duration of the experiment. For the establishment of the extracorporeal loop, the carotid artery and jugular vein were surgically isolated and cannulated with 8 and 9 French percutaneous sheath introducers, respectively. Saline, variegin, ultravariegin, UFH, or bivalirudin, was administered through the cephalic vein cannula and blood were collected from the femoral artery. Real-time monitoring of ECG, heart rate, body temperature, respiration rate, mean arterial blood pressure, blood oxygen saturation and carbon dioxide levels was performed on a LifeWindow Lite multiparameter physiologic monitor (Digicare Biomedical Technology Inc., FL, USA).

### Preparation of the ex vivo perfusion chamber

The descending aorta was removed from each terminated pig. The fascia and connective tissue were removed from the tunica externa. The aorta was cut into rectangular strips of 7 mm width × 27 mm length. Sharp forceps were then used to peel the smooth endothelial layer away from the rest of the vessel, exposing the tunica media. The thickness of the aortic strip was maintained at 0.5–0.6 mm, as measured using a calliper.

Perfusion chambers and stainless-steel connectors were custom manufactured (Sunway Precision Engineering, Singapore, Singapore). The perfusion chamber consists of two Plexiglas blocks, a bottom block (9  × 16  × 50 mm) with a 2-mm diameter tubular channel, and a top block (5  × 16  × 50 mm). A 28-mm section of the tubular channel was exposed in the middle section of the bottom block. A cobalt-chromium coronary stent (2.5 mm diameter × 23 mm length) was expanded to an outer diameter of 2 mm, pre-weighed and placed inside the tubular channel. A 7  × 27 mm strip of endothelium-denuded porcine aorta was placed on top to cover the exposed surface of the coronary stent. This simulated the contact area between the coronary stent and the coronary artery during PCI. The top block was then used as a cover for the bottom block and secured with a clamp. To connect the perfusion chambers to the tubing used in the extracorporeal loop, stainless-steel connectors (with a 2-mm diameter tubular channel) were screwed into the bottom Plexiglas block, directly in line with the tubular channel of the Plexiglas perfusion chamber.

### Porcine ex vivo stent thrombosis model

For UFH, 100 U/kg was injected initially and 30 U/kg UFH top-up was used if needed to maintain the activated clotting time between 250 s and 350 s as per clinical practice guidelines61. Bivalirudin was administered through an i.v. bolus injection of 0.75 mg/kg followed by a continuous infusion of 1.75 mg/kg/h. For variegin, either 1 mg/kg (without DAPT) or 0.1 mg/kg (with DAPT) was injected as a single bolus without additional doses. For ultravariegin, either 0.25 mg/kg (without DAPT) or 0.025 mg/kg (with DAPT) were injected as a single bolus. DAPT (300 mg aspirin and 180 mg ticagrelor) was administered orally 16 h prior to surgery. The extracorporeal loop41,62 was set up as followed: platinum-cured silicone tubing with 3.18 mm internal diameter (ID) × 6.35 mm outer diameter (OD) was connected to the carotid arterial (in) and jugular venous (out) cannulas (Supplementary Fig. 17). The loop was filled with saline prior to the experiment. Two minutes following drug or vehicle administration, blood was perfused through the extracorporeal loop with three consecutive perfusion chambers for 60 min at a flow rate of 70 ml/min driven by a peristaltic pump. After 60 min, the peristaltic pump was stopped and saline was perfused through the extracorporeal loop at a flow rate of 10 ml/min for 2 min to wash out residual blood and unbound debris. The perfusion chambers were disassembled, the aortic strips and stents were removed and their pictures were taken. The stents are washed with saline and gently dried using wipes. Stent thrombus was measured by weighing the stents with attached thrombus and subtracting the dry weight of the stent before the experiments. Using a scalpel blade, aortic thrombus was gently removed, washed with saline and dried with wipes, and then weighed again. Total thrombus weight was calculated as the sum of weights of the stent and aortic thrombi.

### Pig ear vein bleeding test

The ear vein bleeding test was performed 5 min after administration of saline or anticoagulant41. An 18-G needle was used to puncture the ear vein and blood was dabbed from the wound every 20 s with a sterile gauze swab until bleeding has stopped. Dabbing of the wound was continued for 3 min after the last time-point that bleeding has stopped (no blood observed on gauze). The time between injury and cessation of bleeding was noted as the bleeding time.

### Screening of ultravariegin binders against phage-displayed naive human antibody library

Biotinylated ultravariegin was screened against CIIDRET-PDHAL1, a naive human antibody library (10 billion clones) in the single-chain variable fragment (scFv) format63,64. In the first round of panning, ~3 × 1013 phages (3 ml in 2% BSA-PBS) were pre-adsorbed against streptavidin-coated MyOneT1 magnetic dynabeads. Biotinylated ultravariegin was added to a final concentration of 100 nM, followed by incubation for 2 h on a rotator at 5 rpm at room temperature. After binding, 100 µl streptavidin-coated M280 beads were added per ml of panning mixture (total 300 µl beads per 3 ml), followed by incubation for 30 min on a rotator at 5 rpm at room temperature. The beads were washed ten times each with PBST (PBS containing 0.05% Tween 20) and PBS and the bound phages were eluted using 100 mM Triethylamine, followed by neutralisation using 1 M Tris-HCl, pH 7.5. An aliquot of bound phages was titrated in E. coli TOP10F’ cells, and the remaining phages were subjected to infection in TOP10F’ cells. For the next round of panning, phages were rescued using AGM13 in 1200 ml volume and purified using double PEG precipitation. The second and third rounds of panning were performed as described above with appropriate modifications; the amount of antigen was reduced to 50 nM and 10 nM in the second and third round of panning, respectively, and the number of washes was increased to 15 each with a round of PBST and PBS washing. After the third round of panning, individual phage clones were analysed using phage ELISA and DNA sequencing and seven phage clones specific to biotinylated ultravariegin were identified.

### Cloning, expression and purification of antibodies against biotinylated ultravariegin

Variable light and heavy chain genes encoded by selected anti-ultravariegin scFv(s) were sub-cloned in pcDNA3.4 based pVCLC102 (Kappa Light Chain; KLC) or pVCLC202 (Lambda Light Chain; LLC) and pVCHC302 (Heavy Chain; HC) vectors for expression of antibodies in IgG format. Purified LC and HC plasmid DNA were mixed 1:1 and transfected in 50/100 ml ExpiCHO cells using the Max Titer expression protocol as per manufacturer’s instructions (Thermo Fisher Scientific, MA, USA). Culture supernatant was harvested 8 days after transfection, purified using 1 ml HiTrap MabSelect SuRe columns (GE Healthcare Life Sciences, IL, USA) and eluted with a linear gradient comprising 0.1 M citrate, pH 3.0 followed by neutralisation of pH using 2 M Tris-HCl. Concentrations of each protein were estimated by absorbance and extinction coefficient values at 280 nm, assuming molar extinction coefficient of human IgG to be 210,000 M−1 cm−1.

### Binding affinity measurements by biolayer interferometry

The binding affinities of biotinylated ultravariegin and scrambled ultravariegin for the two reversal agents, Ab1282 and Ab1283, were measured by biolayer interferometry on an Octet RED96 System (Pall FortéBio, CA, USA) and data analysis performed using FortéBio Data Analysis 9.0 software. Biotinylated peptides were loaded onto Streptavidin (SA) biosensors. The experiments were carried out in 50 mM Tris buffer (pH 7.4) containing 100 mM NaCl and 1 mg/ml BSA at 27 °C. Typically, the biosensors were pre-equilibrated in the buffer for 750 s, loaded with biotinylated peptide for 900 s, re-equilibrated to baseline for 300 s before the association phase of 2400 s in wells containing 20 nM, 6.7 nM, 2.2 nM, 0.74 nM, 0.25 nM, 0.082 nM and 0.027 nM of Ab1282 or Ab1283. Reference wells contained no antibodies. Biosensors were transferred into new wells containing assay buffer for a dissociation phase lasting 3600 s. Binding kinetics were calculated using the FortéBio Data Analysis 9.0 software. The association (kon) and dissociation (koff) rate constants were obtained by fitting the association and dissociation data to a 1:1 model. The equilibrium dissociation constant, KD, was estimated by fitting the steady-state binding isotherm of response at equilibrium (Req) against the concentrations of antibodies.

### Statistical analysis

All statistical analyses and curve-fitting by non-linear regression were performed on Prism 6.0 (GraphPad, CA, USA). Dose–response fits for Table 1 and Fig. 7a, tight-binding fits (Morrison equation) for Fig. 2b–d and the log-log line fit for Supplementary Table 2 were used ascertained with goodness-of-fit analyses as implemented in Prism 6.0. The Pearson correlation coefficient, r, was used to calculate the correlation between bleeding times obtained in the tail bleeding and saphenous vein bleeding models (Fig. 4f, insert). One-way analysis of variance (ANOVA) and post hoc Tukey’s multiple comparisons tests were used to compare the quantitative endpoints between treatment groups in Figs. 5 and 6. Two-tailed, unpaired t test was performed for comparison between treatment (Ab1283) and control (saline) groups in Fig. 7d. Multiplicity-adjusted P values for each comparison were reported.

### Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.