Factor XIa–specific IgG and a reversal agent to probe factor XI function in thrombosis and hemostasis

See allHide authors and affiliations

Science Translational Medicine  24 Aug 2016:
Vol. 8, Issue 353, pp. 353ra112
DOI: 10.1126/scitranslmed.aaf4331

To clot or not: Anticoagulation without bleeding

Current drugs designed to prevent blood clots, heart attack, or stroke also increase the risk of hazardous bleeding. Toward testing whether selective block of one branch of the coagulation cascade might inhibit clotting without causing bleeding, David et al. developed an antibody that occupies and blocks the active site of a critical protein called FXIa. This antibody inhibited clotting of human blood and prevented blood vessel block in animal models but did not cause bleeding. So that the antibody can be used safely in people, the authors also developed a second antibody that can reverse the action of the first.


Thrombosis is a major cause of morbidity and mortality. Current antithrombotic drugs are not ideal in that they must balance prevention of thrombosis against bleeding risk. Inhibition of coagulation factor XI (FXI) may offer an improvement over existing antithrombotic strategies by preventing some forms of thrombosis with lower bleeding risk. To permit exploration of this hypothesis in humans, we generated and characterized a series of human immunoglobulin Gs (IgGs) that blocked FXIa active-site function but did not bind FXI zymogen or other coagulation proteases. The most potent of these IgGs, C24 and DEF, inhibited clotting in whole human blood and prevented FeCl3-induced carotid artery occlusion in FXI-deficient mice reconstituted with human FXI and in thread-induced venous thrombosis in rabbits at clinically relevant doses. At doses substantially higher than those required for inhibition of intravascular thrombus formation in these models, DEF did not increase cuticle bleeding in rabbits or cause spontaneous bleeding in macaques over a 2-week study. Anticipating the desirability of a reversal agent, we also generated a human IgG that rapidly reversed DEF activity ex vivo in human plasma and in vivo in rabbits. Thus, an active site–directed FXIa-specific antibody can block thrombosis in animal models and, together with the reversal agent, may facilitate exploration of the roles of FXIa in human disease.


Thrombosis is the proximal cause of most myocardial infarctions and strokes, pulmonary embolism, deep venous thrombosis, acute limb ischemia, and other cardiovascular events. It remains a major cause of morbidity and mortality (1, 2). Current antithrombotic drugs decrease thrombosis but impair hemostasis, a trade-off that limits both efficacy and safety (3, 4). Better agents are needed.

Thrombin, the effector protease of the coagulation cascade, activates platelets and produces fibrin, which form the clots that mediate both thrombosis and hemostasis (1). The coagulation cascade that generates thrombin has two separate triggers, the extrinsic and intrinsic pathways, which converge on a common pathway that generates thrombin (fig. S1) (1, 5, 6). All currently used anticoagulant drugs act by inhibiting one or more components of the extrinsic and common pathways (3, 4, 6). The effects of genetic deficiencies in animals (713) and humans (1, 14) demonstrate that all extrinsic and common pathway components are necessary for hemostasis and normal survival. In accord, all currently used anticoagulants balance antithrombotic efficacy against bleeding risk (3, 4, 15). By contrast, substantial evidence suggests that inhibition of the intrinsic pathway might prevent thrombosis with less cost in bleeding than current agents (15, 16).

The intrinsic pathway is composed of the protease zymogens factor XII (FXII), FXI, and plasma prekallikrein (PK) (fig. S1) (1, 5, 6, 15). Assembly of these factors on negatively charged polymers or surfaces leads to activation of FXII to the active protease FXIIa. FXIIa then converts PK to its active protease form, plasma kallikrein (PKa), and zymogen FXI to its active form, FXIa. FXIa connects intrinsic pathway activation to the common pathway of coagulation by converting FIX to FIXa. Identifying the negatively charged polymers and surfaces that trigger the intrinsic pathway in different settings in vivo is an active area of study; chromatin and RNA released from damaged cells, activated neutrophils, and polyphosphates released from platelets or bacteria may all contribute (15, 17, 18).

Unlike the extrinsic and common pathways, intrinsic pathway components are largely unnecessary for day-to-day hemostasis (1, 15). Although limited by size, concentration in specific ethnic groups, and incomplete deficiency states, genetic studies suggest that FXII-deficient humans have no defect in hemostasis and that FXI-deficient humans rarely show spontaneous bleeding but can have increased bleeding with trauma or surgery (14, 15, 19). Increased risk of intracranial bleeding or joint bleeding has not been reported in FXI-deficient people. Mice that lack FXII or FXI show neither spontaneous bleeding nor increased bleeding after hemostatic challenge.

Despite its relative lack of importance for hemostasis, substantial evidence suggests that the intrinsic pathway can contribute to thrombosis [reviewed in (15)]. Mice deficient in FXII or FXI are protected against thrombosis. Antibodies that inhibit FXII or FXI function as well as FXII- and FXI-depleting antisense molecules protect against thrombosis in rabbit and nonhuman primate models. FXI deficiency in humans is associated with decreased risk of venous thromboembolism and stroke, whereas increased FXI levels appear to be associated with increased risk.

Unlike FXI deficiency, an association of FXII deficiency with decreased risk of venous thromboembolism and stroke in humans has not been detected (15). Additionally, whereas FXI and FXII deficiencies had similar effects in mouse models of thrombosis, FXI-blocking antibodies were more effective than FXII-blocking antibodies in inhibiting clot formation in a primate shunt model. In addition to being activated by FXIIa, FXI can be activated by thrombin in an amplification loop that may support thrombin generation in some circumstances (fig. S1) (15, 20), and it is possible that differences in the relative importance of this amplification loop lead to differences in the relative importance of FXII and FXI among species, models, and disease processes.

The notion that FXIa activity helps drive intravascular thrombus formation, whereas extrinsic pathway activation in the setting of vessel disruption and extravasation of blood into a high tissue factor environment is usually sufficient to drive extravascular clot formation and hemostasis, provides a plausible mechanistic rationale for the hypothesis that FXI inhibition may preferentially impair thrombosis over hemostasis. The previous studies that were outlined above and elsewhere have made a strong case for testing this hypothesis in humans. In a recent phase 2 clinical study, depletion of FXI with an antisense molecule in patients undergoing total knee arthroplasty was associated with decreased venous thrombosis compared to treatment with the common pathway inhibitor enoxaparin (21), supporting a necessary role for FXI in at least one form of thrombosis in humans. However, a requirement for a 5-week pretreatment for effective depletion of FXI may limit the utility of the antisense approach. Agents that permit facile exploration of the relative importance of FXI in human thrombosis and hemostasis are needed. Although progress has been made (22), achieving FXIa inhibition with small molecules without inhibiting the related protease PKa remains a challenge. Antibodies to FXI are an alternative strategy, and several are in preclinical or early development (15, 23, 24). We sought to develop a highly active site–dependent FXIa antibody that does not bind FXI zymogen and that provides immediate and prolonged, but readily reversible, anticoagulant activity for several reasons. By avoiding binding to FXI zymogen, such an antibody would act only during active coagulation by binding only to the small fraction of FXI converted to FXIa, and because inhibition of FXI function would not involve depletion or sequestration of FXI zymogen, dosing would not need to keep up with new FXI synthesis [predicted to be about 16 mg/day based on an estimated half-life of FXI of 50 hours (1)]. Additionally, anticoagulation could be reversed by inhibiting an FXIa-specific antibody without a need for FXI replacement.


Selection and characterization of human immunoglobulin Gs that specifically bind FXIa and block its enzymatic activity

To obtain FXIa-specific antibodies, we used biotinylated human FXIa in the presence of excess nonbiotinylated human FXI to select phage from a naïve human single-chain variable fragment (scFv) phage display library. Sequential rounds of selection with increasing stringency (fig. S2) led to identification of 13 unique clones that bound FXIa, but not FXI zymogen, by enzyme-linked immunosorbent assay (ELISA). After reformatting from scFv to human immunoglobulin G1 (IgG1), IgGs corresponding to seven of these clones bound cynomolgus macaque and rabbit as well as human FXIa. The most potent, designated D4, bound human FXIa with an apparent KD of 0.93 nM and inhibited cleavage of the fluorogenic peptide substrate SN-59 by FXIa by 50% at a concentration (IC50) of 5 nM (Tables 1 and 2).

Table 1. Kinetics of anti-FXIa antibody binding to human FXIa.

Rate constants were measured for binding of anti-FXIa antibodies to human FXIa by surface plasmon resonance (Biacore). The density of FXIa bound was set at a low level, and a 1:1 binding model was used to fit the association (ka) and dissociation (kd) constants and the calculated affinity (KD) for IgG as well as Fab. Results are means ± SEM of three to four experiments. C24 and DEF were studied as the Fab forms at 37°C, because slow dissociation rates of the IgG forms at 25°C precluded accurate measurements.

View this table:
Table 2. Inhibition of FXIa fluorogenic substrate hydrolysis by FXIa-specific antibodies and species cross-reactivity.

The IgG forms of the anti-FXIa antibodies D4, B11, C24, and DEF or a control IgG were incubated at varying concentrations in the presence of 200 pM human, cynomolgus macaque (cyno), or rabbit FXIa and 100 μM SN-59 fluorogenic peptide substrate. IC50 values were calculated from Vmax determinations recorded over nine threefold serial dilutions of each antibody from 77 to 0.012 μg/ml and no-antibody controls. Results are means ± SEM of at least three experiments. The control IgG had no effect on FXIa activity in any of these assays.

View this table:

To increase the affinity and potency, we generated a D4-based phage display library using error-prone polymerase chain reaction (PCR) and selected for phage binding using pM concentrations of biotinylated FXIa and conditions that favored fast binding and slow off-rate (see Materials and Methods). The resulting unique clones were converted from scFv to human IgG1, and IgGs corresponding to B11 and eight other clones showed higher potency for FXIa binding compared to D4 (Tables 1 and 2 and fig. S2). Testing of combinations of the amino acid substitutions found in these clones resulted in the generation of the antibody C24. The KD values for B11 IgG1 and C24 Fab binding to human FXIa were 0.19 and 0.015 nM, respectively (Table 1). IC50 values for inhibition of FXIa-mediated SN-59 hydrolysis by B11 and C24 IgG1 were 1.23 and 0.22 nM, respectively (Table 2). Data for an effector null derivative of C24, designated DEF (see below), are also shown.

Multiple lines of evidence suggested that D4, B11, and C24 impaired FXIa active site function by direct occupancy of that site. As expected from the selection and optimization process, these antibodies all recognized FXIa but not FXI zymogen (Fig. 1A), and their binding to FXIa was mutually exclusive (Fig. 1B), consistent with binding to a shared epitope that requires active-site formation. Despite the high affinity of C24 for FXIa, incubation of FXIa with phenylmethylsulfonyl fluoride, which adds only a phenylmethyl group to the active-site serine, was sufficient to prevent C24 binding (Fig. 1, C and D). Antibody-FXIa complexes were labile to SDS (fig. S3). Together, these results indicate that C24 binding to FXIa involves a direct, noncovalent interaction with the FXIa active site.

Fig. 1. Binding properties of anti-FXIa IgGs.

(A) Binding of the indicated IgGs to immobilized FXIa or FXI zymogen. IgGs were added in the amounts (microgram per well) indicated on the x axis, and the binding was measured by ELISA. Means ± SEM (n = 2). OD, optical density; hFXIa, human FXIa. (B) Competitive binding of D4 antibody to the indicated FXIa-antibody complexes. Biotinylated FXIa was captured on Octet Streptavidin Biosensors; each was saturated with anti-FXIa antibodies (D4, B11, C24, DEF, and AHXI-5061) and then dipped in 500 nM D4 antibody, and the binding response was measured. (C and D) Equal amounts [~400 response units (RU)] of biotinylated FXIa that had been incubated in the absence (C) or presence (D) of phenylmethylsulfonyl fluoride (PMSF) were captured in surface plasmon resonance flow cells, and then the sensorgrams for C24 Fab (0.1 to 5 nM) binding were obtained. Curves shown are from a single experiment. The experiments in (C) and (D) were replicated four times and twice, respectively, with similar results.

C24 binding to FXIa was not only of high affinity but was also specific to FXIa over other coagulation proteases (fig. S4). C24 showed no binding to thrombin, FXa, FVIIa, FXIIa, PKa, or FIXa, even at concentrations >100 times as much as those required to saturate FXIa in an ELISA-type assay. Similarly, at >80 times as much as the concentration required to completely block small-substrate cleavage by FXIa, C24 did not inhibit small-substrate cleavage by thrombin, FXa, FVIIa, FXIIa, PKa, or activated protein C (APC). Similarly, C24 did not inhibit FIXa-mediated activation of FX. Data for DEF are also shown. These results suggested that C24 could serve as a useful tool to specifically probe the function of FXIa.

Anti-FXIa IgGs inhibit FXIIa-induced thrombin generation and intrinsic pathway–triggered clot formation in human plasma and whole blood

We next tested the ability of antibodies against FXIa to inhibit intrinsic pathway–mediated activation of the coagulation cascade. In human plasma, the concentrations of D4, B11, and C24 required for half-maximal inhibition of peak thrombin generation triggered by FXIIa addition were >512 μg/ml (~3.5 μM), ~50 μg/ml (345 nM), and ~5 μg/ml (34 nM), respectively (Fig. 2A). Similar relative results were obtained when lag time to onset of thrombin generation, time to peak, and endogenous thrombin potential—which integrates total thrombin activity generated over time—were used as end points (Fig. 2B and fig. S5). Similar results were also seen when we assessed intrinsic pathway–dependent clotting using activated partial thromboplastin time (APTT), a standard FXI-dependent clinical assay that measures time to clotting in recalcified human plasma after addition of a negatively charged polymer. Addition of D4, B11, and C24 to human plasma resulted in half-maximal prolongation of APTT at concentrations of about 1 mg/ml (6.9 μM), 100 μg/ml (690 nM), and 10 μg/ml (69 nM), respectively (Fig. 2C). Last, C24 inhibited intrinsic pathway–triggered clotting in whole human blood with a potency similar to that obtained in the plasma-based assays; the concentration required for half-maximal prolongation of clotting time was ~3 μg/ml (21 nM) (Fig. 2D).

Fig. 2. Effect of anti-FXIa antibodies on FXIIa-induced thrombin generation, APTT in human plasma, and intrinsic pathway–triggered clotting in whole human blood.

(A and B) FXIIa-triggered thrombin activity as a function of time was determined in the presence of the indicated concentrations of anti-FXIa antibodies D4 (blue), B11 (red), C24 (green), or control IgG1 (black). Peak thrombin activity (A) and lag to onset of thrombin generation (B) are shown (means ± SEM; n = 3 to 5). (C) APTT assay as a function of antibody concentration (means ± SEM; n = 2). Control IgG1 had no effect in this assay. (D) Effect of C24 or control IgG1 on intrinsic pathway–triggered clotting of whole blood. Time to clot is shown (means ± SEM; n = 3 to 4).

Even at 30 μg/ml, C24 did not alter thrombin generation in human plasma triggered by the concentration of tissue factor and phospholipid typically used in this assay, but C24 did inhibit thrombin generation triggered by lower concentrations of tissue factor (fig. S6). The latter activity is unlikely to be a result of any direct effect of C24 on the extrinsic or common pathways because C24 is unable to bind coagulation factors other than FXIa (fig. S4). Rather, such inhibition could be a result of inhibition of a small amount of FXIa generated during plasma preparation (25) and/or inhibition of the function of FXIa in the thrombin-mediated positive feedback loop, which may help amplify thrombin generation when the stimulus is weak (fig. S1) (20). Regardless, these data suggest that coagulation can proceed normally when FXIa is inhibited by C24 when concentrations of tissue factor/phospholipid are high.

The observation that the potency order for D4, B11, and C24 is the same for FXIa binding, inhibition of small substrate cleavage, and inhibition of thrombin generation and clot formation in plasma (table S1) suggests that occupancy of the FXIa active site is the common mechanism by which these antibodies act. The need for higher absolute concentrations of antibody for activity in the plasma-based assays is unlikely to be due to the presence of plasma protein in general because addition of albumin at physiological concentrations had no effect on inhibition of FXIa activity by C24 (fig. S7). Rather, FXIa-specific antibodies likely need to rapidly bind newly formed FXIa with high fractional occupancy and in competition with substrate to prevent thrombin generation in plasma (see Discussion).

The results described above indicate that C24 impaired intrinsic pathway–triggered thrombin generation in plasma and clotting in whole blood at a concentration of tens of nanomolar. Combined with its specificity, this potency suggested that C24 was suitable for inhibiting FXIa function in vivo.

C24 inhibits FeCl3-induced arterial thrombosis in an FXI-humanized mouse

We used a mouse model known to be FXI-dependent for an initial determination of whether C24 could alter thrombosis in vivo. FXI-deficient mice are protected against FeCl3-induced carotid artery occlusion (26, 27), a commonly used assay of injury-induced arterial thrombosis. Because our antibodies did not cross-react with mouse FXIa, we established an FXI-humanized mouse analogous to that reported by Geng et al. (28) but by administration of human FXI protein rather than hydrodynamic transduction. Administration of human FXI [0.25 mg/kg intravenously (iv)] to FXI-deficient mice rescued FXIIa-driven thrombin generation in plasma to wild-type values and provided a concentration of human FXI in plasma of ~1.5 μg/ml, ~30% of the level in human plasma, for the duration of the thrombosis protocol, as measured by ELISA.

Reconstitution of FXI-deficient mice with human FXI restored carotid occlusion after application of FeCl3 (4% w/v; 250 mM) (Fig. 3A). Four of four, zero of four, and four of four carotids were occluded by the end of the protocol in wild-type, FXI-deficient, and FXI-humanized mice, respectively. Median time to occlusion was similar in wild-type and FXI-humanized mice (850 s versus 740 s). In FXI-humanized mice that received control IgG1 at 35 mg/kg iv, the highest dose tested, seven of eight carotids were occluded by the end of the protocol and median time to occlusion was 750 s, a rate indistinguishable from that seen in wild-type or FXI-humanized mice. By contrast, only 3 of 19 carotids occluded in FXI-humanized mice that received C24 at 2 mg/kg iv or above (Fig. 3B). At 0.5 mg/kg, C24 prolonged median time to occlusion to 1100 s, and at 2 mg/kg and above, median time to occlusion was greater than 1380 s, the end of the protocol. Achieving substantial inhibition of carotid occlusion at C24 dose of 2 mg/kg (Fig. 3B) supported further exploration of its activity in vivo.

Fig. 3. Effect of C24 on FeCl3-triggered carotid arterial thrombosis in human FXI–reconstituted FXI-deficient mice.

(A) Blood flow after carotid injury with 250 mM FeCl3 (v/v) in FXI-deficient mice injected with vehicle (blue) or human FXI at 0.25 mg/kg (red) or in age-matched wild-type (WT) mice from the same colony (black). Percentage of vessels remaining open as a function of time after injury. KO, knockout. (B) Human FXI–reconstituted FXI-deficient mice were injected with C24 at 0.5 (blue), 2 (red), 4 (green), 12 (orange), and 35 (gray) mg/kg iv or with the same doses of control IgG1. The percentage of vessels remaining open as a function of time after injury was determined as in (A). Carotids in mice injected with all doses of control IgG1 had median occlusion times similar to WT mice; only IgG1 at 35 mg/kg data are shown (black) to avoid clutter. The rate of occlusion was decreased in human FXI–reconstituted FXI-null mice treated with C24 at 2 mg/kg and higher doses when compared to the rate in mice treated with control IgG1 by log-rank analysis (Mantel-Cox) (P = 0.01).

Anti-FXIa antibody decreases venous thrombus formation in a rabbit model in a dose-dependent manner that correlates with APTT prolongation

Encouraged by its activity in the FXI-humanized mouse, we modified C24 to improve its drug features by replacing an in silico–predicted isomerization residue and mutating the Fc region to render it “effector null,” that is, unable to activate complement or engage Fc receptors or thereby activate platelets and other cells. The resulting molecule, designated DEF, was indistinguishable from C24 in terms of affinity for FXIa, IC50 for SN-59 hydrolysis by FXIa, inhibition of FXIIa-triggered thrombin generation in human plasma, and selectivity for FXIa over other active coagulation proteases and zymogen FXI (Tables 1 and 2 and fig. S4).

Because DEF cross-reacted with rabbit FXIa and because rabbits were used as preclinical models for recently developed anticoagulants (2931), we next probed the effect of DEF on thrombosis and hemostasis in the rabbit. To select a dose range, we determined DEF levels in plasma, APTT, and prothrombin time (PT) over 14 days after a single intravenous injection of DEF at 0.03, 0.1, 1, and 10 mg/kg (Fig. 4, A and B, and fig. S8). DEF exhibited pharmacokinetics typical of human IgGs with a relatively long half-life (Fig. 4A). APTT was prolonged in a concentration-dependent manner beginning at a DEF level of 1 μg/ml (~6.9 nM) and increasing through ~10 μg/ml (~69 nM) (Fig. 4B). These concentrations were achieved by administration of DEF at 0.1 and 1 mg/kg, respectively (Fig. 4A), and are in good agreement with the concentrations required to impair FXIIa-triggered thrombin generation and prolong APTT in spiking studies in human plasma (Fig. 2). Together, these data suggest that DEF inhibited FXIa function in rabbits at easily achieved doses. By contrast, DEF did not impair extrinsic/common pathway function as measured by PT, even at a dose of 10 mg/kg and a plasma concentration of 200 μg/ml.

Fig. 4. Pharmacokinetics and pharmacodynamics of DEF in rabbits.

(A) DEF antibody concentration in plasma at the indicated time points. Rabbits were given the indicated doses of DEF intravenously. mAb, monoclonal antibody. (B) APTT (dots) and PT (triangles) as a function of antibody concentration across the time course and doses examined in (A). Note that APTT prolongation was detected at DEF concentrations above 1 μg/ml (B), which was achieved at a dose of 0.1 mg/kg (A), and was maximal at a DEF concentration achieved at 1 mg/kg (see also fig. S8). Three rabbits were treated and followed at each dose as described above.

We next determined whether inhibition of FXIa function was associated with proportional antithrombotic effects and whether APTT might be useful as a pharmacodynamic marker for DEF dose and activity sufficient to achieve such effects using a rabbit model of venous thrombosis. In this model, which was used in the preclinical development of the FXa inhibitors rivaroxaban and apixaban (2931), threads are deployed from a catheter in the inferior vena cava to trigger local intravascular thrombus formation, and the extent of thrombus formation is quantitated by weight. Compared to rabbits that received control IgG1 (10 mg/kg), rabbits that received DEF showed a dose-dependent decrease in clot weight, with no effect at 0.03 mg/kg, a partial effect at 0.1 mg/kg, and maximal effect at 1 mg/kg and above (Fig. 5A). Analysis of plasma samples collected at the end of the protocol revealed that DEF prolonged APTT (Fig. 5C) in a dose-dependent manner that paralleled the reduction in clot weight (Fig. 5A). No effect of DEF on PT was detected (Fig. 5E).

Fig. 5. Effect of DEF versus rivaroxaban in a rabbit venous thrombosis model.

Thrombus formation in the rabbit inferior vena cava. APTT and PT were measured in rabbits treated with DEF or control IgG1. There were six rabbits per treatment group; clot weight in each rabbit is shown by a single symbol as is pre- and posttreatment APTT and PT. Means ± SEM (n= 6). (A) Weights of thread-induced clot in rabbits treated with either DEF (0.03, 0.1, 1, 3, or 10 mg/kg iv) or control (ctrl) IgG1 (10 mg/kg iv). (B) Weights of thread-induced clot in rabbits treated with either rivaroxaban (0.045 and 1.5 mg/kg iv) or its vehicle. (C and D) Pre- and posttreatment APTT for the rabbits examined in (A) and (B). (E and F) Pre- and posttreatment PT clotting times for the rabbits examined in (A) and (B). One-way analysis of variance (ANOVA) and Tukey’s multiple comparisons test were used to determine significance of clot weight changes: *P < 0.05 versus IgG1 (A) or vehicle control (B). Two-way ANOVA and Sidak’s multiple comparisons test were used to determine significance of clotting time changes: *P < 0.05 pretreatment versus posttreatment. Note DEF dose-dependent reduction in clot weight paralleled prolongation of APTT clotting times, with effects beginning at 0.1 mg/kg.

The positive control and comparator rivaroxaban at 1.5 mg/kg prolonged APTT and decreased thrombus/clot weight to an extent comparable to that obtained with DEF (Fig. 5, B and D). As expected for an FXa inhibitor that blocks the common pathway of coagulation, rivaroxaban also prolonged PT (Fig. 5F). Control IgG1 and the rivaroxaban vehicle [dimethylamine (DMA)/polyethylene glycol, molecular weight 400 (PEG-400)] had no effect on clot weight, APTT, or PT (Fig. 5, A to F).

DEF treatment did not increase cuticle bleeding in rabbits or cause spontaneous bleeding in cynomolgus macaques

We next determined whether the antithrombotic effects of DEF in the rabbit were accompanied by impaired hemostasis by measuring blood loss after cuticle transection as well as APTT and PT before and after DEF treatment. In this design, in addition to being part of a specific treatment group, each rabbit served as its own control (Fig. 6A). Treatment with DEF at 10 mg/kg—a dose 10-fold that is required for maximal inhibition of intravascular thrombus formation (Fig. 5A) and near maximal prolongation of APTT (Fig. 5C)—did not increase blood loss after cuticle transection (Fig. 6A). Rivaroxaban treatment at 1.5 mg/kg was associated with increased blood loss (Fig. 6A), confirming that cuticle hemostasis was dependent on the common pathway of coagulation under our assay conditions. Control IgG and rivaroxaban vehicle had no effect on blood loss. Analysis of plasma samples confirmed that, as expected, DEF treatment prolonged APTT but not PT, whereas rivaroxaban prolonged both (Fig. 6, B and C). These data suggest that FXIa inhibition at a level sufficient to maximally impair intravascular clot formation (Fig. 5A) and prolong APTT (Fig. 4) does not impair hemostasis as assessed by blood loss after cuticle transection.

Fig. 6. Effect of DEF and rivaroxaban on cuticle bleeding in rabbits.

(A to C) Blood loss after cuticle transection (A), APTT (B), and PT clotting times (C) were measured before (dots) and after (triangles) treatment with DEF (10 mg/kg iv), rivaroxaban (1.5 mg/kg iv), or their cognate control IgG or vehicle. Ten rabbits were studied per treatment group; cuticle blood loss before (open circle) and after treatment (filled triangle) for each rabbit is shown as a single point as is pre- and posttreatment APTT and PT. Horizontal lines indicate mean values. Two-way ANOVA and Sidak’s multiple comparisons test were used to determine significance of blood loss and clotting time changes: *P < 0.05 before versus after treatment.

To probe for effects on spontaneous hemostasis and potential toxicity in a species closely related to humans, cynomolgus macaques were treated on days 1, 8, and 15 with DEF at dosages 0, 20, 75, and 266.5 mg/kg iv and 266.5 mg/kg subcutaneously (sc). Note that these doses were 20- to 266-fold higher than those required for maximal effects on APTT and venous thrombosis in rabbits (Figs. 4 and 5) and that DEF had a similar potency for inhibition of rabbit, cynomolgus macaque, and human FXIa (Table 1). Blood samples were collected for the determination of antibody concentration, APTT, and PT at the indicated times (fig. S9) and for hematology and clinical chemistry parameters before necropsy on day 16. The plasma concentration of DEF was above 150 μg/ml. APTT was maximally prolonged at all doses and at all posttreatment time points examined, and PT was unaffected at any dose (fig. S9). Even at the achieved high exposures, animals showed no DEF-related clinical signs, effect on body weight, change in hematocrit or other hematology or clinical chemistry parameters, or macroscopic or microscopic findings in any organ examined (see Materials and Methods).

Development of an agent for rapid reversal of anti-FXIa mAb activity

Although increased bleeding associated with FXI deficiency or FXIa inhibition in animals has not been reported and was not detected in our studies, FXI deficiency in humans can be associated with bleeding, albeit bleeding that is usually minor and linked to trauma and some types of surgery (14, 15, 19). Because our goal was to provide a tool for exploring the roles of FXIa in humans, we sought to create a reversal agent for DEF to permit rapid control of anti-FXIa activity.

We used biotinylated DEF to select from a human scFv phage display library and ultimately isolated a single clone, designated revC4, that exhibited selective binding to DEF over other human IgGs (Fig. 7A). When reformatted to IgG1, revC4 had an apparent 3 nM KD for binding to DEF Fab (Fig. 7, B and C). At a 10:1 ratio to DEF IgG1, revC4 reversed inhibition of FXIa-mediated small-substrate cleavage by DEF (Fig. 7D). When added to human plasma containing DEF (16 μg/ml) before initiation of coagulation, revC4 IgG (10 μg/ml) partially reversed DEF-mediated inhibition of FXIIa-induced thrombin generation, and revC4 (20 μg/ml) completely reversed this inhibition (fig. S10, A and B). Thus, under conditions that likely mimic the in vivo situation (formation of a revC4-DEF complex without competition from already formed FXIa), revC4 appears to reverse DEF at ~1:1 stoichiometry.

Fig. 7. Characterization of DEF binding and reversal of DEF activity by revC4, an anti-DEF IgG.

(A) Binding of revC4 scFv to streptavidin-captured biotinylated control IgG (Mab8.8, negative control), human serum (Hu-serum) IgG, and DEF IgG detected by ELISA. (B) Surface plasmon resonance sensorgrams for the DEF Fab (2 to 100 nM) binding to immobilized revC4 IgG. Data were fit with a 1:1 binding model, and curve fits are shown overlaid on the sensorgrams. (C) Surface plasmon resonance kinetic rate constants for the association (ka) and dissociation (kd) and calculated affinity (KD) for the revC4 IgG–DEF Fab interaction. (D) revC4 blocks inhibition of FXIa-mediated SN-59 hydrolysis by DEF. DEF antibody (10 nM) was preincubated with the indicated concentrations of C4 IgG (4 to 500 nM) before addition of 0.7 nM FXIa and then SN-59. Results in (B) to (D) were replicated once with similar results. (E) Five rabbits received bolus injections of DEF (1 mg/kg) at time 0, control IgG (3 mg/kg) at 30 min, and revC4 IgG (3 mg/kg) at 60 min. Plasma samples taken before each IgG injection (0, 30, and 60 min) and 30 min after the last injection (90 min) were used to determine APTT and PT. Each APTT measurement in each of the five rabbits is shown as a single point. Horizontal lines indicate mean values. No treatment altered PT (fig. S11). APTT was prolonged after DEF treatment and reversed after revC4 treatment (*P < 0.0001, one-way ANOVA with Tukey’s tests).

Encouraged by these data, we next evaluated the ability of revC4 to reverse the anti-FXIa activity of DEF in vivo in rabbits. In a pilot study, addition of DEF (25 μg/ml) to rabbit plasma prolonged APTT to 210% of control, addition of revC4 at 25 μg/ml partially reversed this prolongation to 157% of control, and addition of revC4 at 75 μg/ml fully normalized APTT. We therefore tested the effect of a3:1 ratio of C4 to DEF in vivo (Fig. 7E). Administration of DEF (1mg/kg iv) to rabbits prolonged APTT as before (Fig. 7E). This prolongation was unchanged 30 min after treatment with control IgG1 (3mg/kg) but was fully reversed 30 min after subsequent administration of revC4 (3 mg/kg). In the absence of revC4, DEF (1 mg/kg) sustained a plasma concentration sufficient to maximally prolong APTT for days (Fig. 4, A and B). Together, these results suggest that revC4 reversed the anticoagulant effect of DEF within 30 min of dosing. As expected, none of the treatments had any effect on PT (fig. S11).

An ideal reversal agent for DEF would be rapidly cleared to permit reanticoagulation when appropriate and would be unable to form potentially toxic immune complexes. Fab fragments of antibodies have a half-life in plasma of 12 to 20 hours and form (Fab)2(antigen)1 complexes, which are generally considered nontoxic. Like revC4 IgG1, the Fab form of revC4 reversed the ability of DEF to block FXIa enzymatic activity (fig. S12).


Thrombosis is a major cause of morbidity and mortality, and anticoagulant medications aimed at preventing thrombosis are used by millions of patients each year (1, 2). Recently developed FXa and direct thrombin inhibitors have achieved better efficacy versus safety in several indications compared to warfarin or heparins, likely because of better pharmacokinetics and more predictive dosing (3, 4). However, like their predecessors, these anticoagulants target common pathway proteases required for hemostasis and still trade antithrombotic benefit for bleeding risk.

Unlike the extrinsic and common pathways, the intrinsic pathway is relatively unimportant for hemostasis and was thought to be unimportant for thrombosis as well. The latter view was challenged by several concordant lines of evidence [reviewed in (15, 16)]. Case-control studies in humans suggested that FXI deficiency is associated with less venous thromboembolism and stroke. Pharmacological studies with antibody, small-molecule, and antisense inhibitors of FXI demonstrated protection against thrombosis in multiple animal models and species. Genetic studies in mice with FXII and FXI deficiencies revealed protection against arterial thrombosis (15, 16). Thus, there is a strong case for testing the possibility that inhibition of FXI might impair thrombosis with less bleeding than conventional anticoagulation. The FXI–antisense oligonucleotide phase 2 trial (21) reported during the course of our study demonstrated efficacy of FXI inhibition in at least one form of human thrombosis but was not powered to test safety. The relatively long treatment period required for reduction of FXI levels by the antisense treatment may limit the utility of the antisense approach for exploring FXI inhibition in several indications, but antibody- and small molecule–based approaches are being pursued (15, 16, 22).

We report the generation and characterization of an active site–directed, FXIa-specific, effector-null (unable to bind Fc receptors), fully human IgG1—DEF—and a second IgG—revC4—that rapidly reverses DEF activity. The active-site specificity of DEF and absence of reactivity with FXI zymogen may be useful clinically in that DEF will act only during activation of coagulation and will not deplete FXI zymogen, thereby avoiding a need for a dosing schedule that accounts for new FXI synthesis and allowing reversal of DEF anticoagulation without a need for FXI replacement. Given the amplification, positive feedback, and robustness of the coagulation cascade and the fact that active-site blockers must compete with substrates, it was far from a foregone conclusion that an antibody that binds a coagulation protease only after it is activated would be able to block newly created active sites with sufficient speed to inhibit coagulation. Our data suggest that DEF achieves this at practical doses in animals and is likely to do so in humans, and are consistent with those described in patent WO2013167669 A1 by Wilmen and colleagues (24), published during the course of our studies.

The specificity of DEF for FXIa over FXI zymogen and over other active coagulation proteases had two notable features. The protease domains of human FXIa and PKa are 65% identical and 79% similar, and developing small-molecule active-site inhibitors specific for FXIa over PKa has been difficult, although progress has recently been made (22). We focused on an antibody—rather than a small molecule—inhibitor to have a better chance of separating FXIa inhibition from PKa inhibition, and DEF was unable to bind or inhibit PKa. In addition, the observation that phenylmethylsulfonyl fluoride treatment, which adds a single phenylmethyl group to the FXIa active-site serine, blocked DEF binding suggests a high degree of active-site dependence not previously described among FXI antibodies.

Because our aim was to develop a tool suitable for clinical investigation and a possible therapeutic, we were guided by two constraints beyond specificity. First, to block clotting in vivo, we expected that we would need a high-affinity antibody with an adequate on-rate to rapidly achieve high fractional inhibition of FXIa activity (32). In the plasma-based assays used herein, only a few percent of the normal level of FXI was sufficient to support intrinsic pathway–triggered thrombin generation and clotting (figs. S13 and S14). As expected and analogous to the observations with small-molecule FXa active-site inhibitors (29), the concentration of DEF required to prolong clotting in plasma and whole blood and to inhibit thrombosis in rabbits was indeed well above its KD for FXIa (Figs. 4 and 5 and Table 1). However, with a KD of ~30 pM and a ka of ~1.6 × 106 s−1 (compared to a ka of FIX for FXIa of 9.7 × 106 s−1), DEF was still able to inhibit coagulation and thrombosis at easily achieved doses. Second, we would need to achieve anticoagulation and antithrombotic effects at a dose of ~2 mg/kg or less to allow formulation for subcutaneous delivery, which would be necessary for some indications in humans. In spiking studies, C24 and DEF inhibited thrombin generation and prolonged APTT in plasma from rabbits and humans at concentrations achieved at a dose of 1 mg/kg in rabbits (Figs. 2, 4, and 5, fig. S8, and table S1). Inhibition of thrombosis in FXI-humanized mouse and rabbit models was achieved at the same dose, and there was a clear relation between prolongation of APTT and clot inhibition in the rabbit model (Figs. 4 and 5). The magnitude of APTT prolongation in rabbit plasma at antibody doses required for inhibition of thrombus formation in rabbits in vivo was similar to that seen in human plasma with >90% depletion of FXI (Fig. 5 and fig. S14). In the recent phase 2 trial of FXI depletion by antisense oligonucleotide, depletion of 85% or more was associated with decreased venous thrombosis compared to enoxaparin (21). These results suggest that DEF may be sufficiently potent for use in humans and that APTT may serve as a useful pharmacodynamic marker for initial dose finding in any future clinical development.

The observation that DEF inhibited FXIIa-triggered thrombin generation in plasma and prolonged APTT but did not inhibit thrombin generation that was triggered by standard concentrations of tissue factor and did not prolong PT is consistent with DEF blocking intrinsic but not extrinsic pathway function. The observation that DEF did not increase blood loss after cuticle transection in rabbits even at a dose 10-fold higher than the dose required for maximal inhibition of thrombus formation and APTT prolongation suggests that FXIa inhibition by DEF does not substantially impair hemostasis, at least as assessed by that model. The lack of clinical signs of spontaneous bleeding or decreased hematocrit in cynomolgus macaques exposed for 15 days with DEF at a dose of 266.5 mg/kg weekly, which was more than 100-fold the dose expected for inhibition of thrombosis, further supports a lack of substantially impaired hemostasis. The common pathway (Xa) inhibitor rivaroxaban did increase cuticle blood loss in the rabbit model. Together, these observations are consistent with the view that extrinsic and common pathways can support hemostasis in the absence of FXIa and intrinsic pathway function in these species and models (15, 33).

Animal studies examine relatively small numbers of individuals for a limited duration. The relative importance of FXI for hemostasis may differ across mammalian species, and underlying conditions that predispose to bleeding in patients (for example, genetic or other hemostatic system defects, structural or functional vascular abnormalities, inflammation, and concurrent drug treatment) are usually not present. Humans with FXI deficiency have increased risk of excessive bleeding with trauma or surgery, especially involving tissues with high fibrinolytic activity such as the oral and nasal mucosa and genitourinary tract, and can exhibit rare, usually mild, spontaneous bleeding episodes (34). Bleeding in FXI-deficient patients correlates poorly with functional FXI levels. Whether this is due to the presence or absence of other hemostatic abnormalities or other variables is unknown. Regardless, anti-FXIa treatment in a population of human patients might well lead to increased bleeding in some, particularly in the setting of trauma or some types of emergency surgery. Although recombinant FVIIa, fibrinolysis inhibitors, and FEIBA (factor VIII inhibitor bypassing activity) can be used as emergency prohemostatic agents in patients with natural FXI deficiency (3537), we anticipated the desirability of a specific reversal agent for DEF and generated revC4, an anti-idiotype IgG and Fab, which rapidly reversed the anticoagulant effect of DEF in human plasma ex vivo and in rabbits in vivo. An ability to rapidly reverse DEF activity with revC4 may support development of DEF in humans, particularly given the development of reversal agents for FXa and thrombin inhibitors (38, 39).

Human genetics and the recent FXI antisense trial indicate that FXI plays a role in venous thrombosis in humans. An important role for FXI in arterial thrombosis in animal models is well supported (26, 27, 4044), and our demonstration that C24 prevented carotid occlusion after FeCl3 injury in FXI-humanized mice provides additional confirmation. Such models have been considered to be tissue factor–dependent (43, 44), but tissue factor dependence and FXI dependence are not mutually exclusive. An antibody to FXI zymogen prevented thrombus growth in an arteriovenous shunt model in which thrombus formation was initiated by tissue factor, and FXI inhibition was more effective than FXII inhibition in this model (45). Thus, FXI activation by FXIIa- and FXIa-dependent amplification of coagulation via thrombin-FXI positive feedback (fig. S1) may contribute to propagation of thrombi even when they are initiated by tissue factor (15). However, whether FXI inhibition might prevent or interrupt thrombus formation in, for example, acute coronary syndromes and why a decreased risk of myocardial infarction has not been detected in FXI-deficient humans are unknown.

Although FXa inhibitors and direct thrombin inhibitors represent a significant advance over previous generation anticoagulants such as warfarin, an unmet need for safer anticoagulation remains. For example, patients with atrial fibrillation who are on dialysis or otherwise at high risk for bleeding are often not anticoagulated and remain at risk for thromboembolic stroke. Similarly, lack of a positive risk-benefit has not supported long-term use of the new FXa and direct thrombin inhibitors in patients with mechanical heart valves or acute coronary syndromes nor, at least to date, for prolonged prophylaxis in medically ill or surgical patients at risk for venous thromboembolism. DEF or similar anti-FXIa molecules, together with a rapid reversal agent, may enable development of anticoagulants that are safer and more effective than current agents.


Experimental design

Biochemical, plasma-based, and in vivo techniques and assays were used to discover a potent active site–specific FXIa antibody and characterize its interactions with FXIa and its effects on coagulation and models of thrombosis and hemostasis. In vitro experiments were generally repeated two or more times as detailed in the figure and table legends. Mouse experiments were intended to provide a preliminary assessment of whether C24 was sufficiently potent to exhibit in vivo activity; the sample size (n = 2 to 9) was chosen on the basis of previous experience with FXI knockout mice showing complete protection against arterial thrombus formation under the conditions of this assay.

Rabbit venous thrombosis studies were intended to determine whether DEF could inhibit thrombosis in a model previously used for preclinical testing of other anticoagulants. Samples size (n = 6 per treatment group) was chosen to provide >80% power to detect a 50% decrease in thrombus weight with P < 0.05 and SD of 25% of the mean.

Rabbit cuticle bleeding studies were intended to determine whether DEF treatment could impair hemostasis. Sample size (n = 10 per treatment group) was chosen to provide >80% power to detect a doubling in blood loss with P < 0.05 and SD of 50% of the mean.

Phage display

Anti-FXIa scFvs were selected from a naïve human phage display library. The antigen human FXIa (Haematologic Technologies Inc.) was biotinylated with Sulfo-NHS-LC-Biotin (Pierce) according to the manufacturer’s protocol. Aliquots of phage display library were incubated with varying concentrations of biotin-tagged FXIa always in the presence of excess untagged zymogen human FXI (300 nM) (fig. S2). Phage bound to biotin-tagged FXIa were recovered with streptavidin-coated magnetic Dynabeads M-280 (Invitrogen). Four sequential rounds of such selection were performed with decreasing concentrations (150, 75, 30, and 5 nM) of biotin-tagged FXIa. Six thousand clones from the third and fourth round outputs were screened by FXI/FXIa ELISA; 166 clones bound FXIa but not FXI. After sequencing, 13 unique clones (including D4) that bound human FXIa and inhibited SN-59 hydrolysis by FXIa were identified. Eleven of the 13 clones cross-reacted with cynomolgus macaque FXIa by ELISA. After reformatting to IgG1, seven clones retained binding selectivity to FXIa and cynomolgus macaque cross-reactivity.

FXIa binding assay

Wells were coated with either 0.02 μg of FXIa or 0.02 μg of FXI, washed, and blocked. Anti-FXIa mAbs were added at various concentrations. After washes, bound antibody was detected with an anti-human IgG horseradish peroxidase (HRP) secondary antibody (Southern Biotech). After development, specific binding signal was measured as OD at 450 nm. AHXI-5061 (Haematologic Technologies Inc.), an active site–independent FXI mAb, was used as a comparator in some assays. For binding assays involving additional coagulation proteases, APC, and PKa, wells were coated with 0.01 μg of each protease, washed, and blocked. Biotinylated C24 or DEF was added at various concentrations. After washes, bound antibody was detected with Pierce High-Sensitivity Streptavidin-HRP (5 ng per well), washed, and specific signal was measured as OD at 450 nm. Proteases were listed in the activity assay section below, with exception of FIXa (HCIXA-0050, Haematologic Technologies Inc.), which was used for binding assays only.

IgG affinity maturation

Error-prone PCR-based random mutagenesis was performed on the D4 scFv gene. The amplified mutated D4 scFv DNA was cloned into a phage display vector, and a 2 × 1010 scFv phage display library was generated. Three sequential rounds of selection designed to select for fast-on- and slow-off binders were performed. Phage were captured by incubation with 900 pM biotin-tagged FXIa on streptavidin-magnetic beads for 1 hour at room temperature in round 1 and 90 pM biotin-tagged FXIa in solution for 1 hour followed by streptavidin magnetic bead capture in round 2. In round 3, phage were incubated with 9 pM biotin-tagged FXIa on streptavidin magnetic beads for 5 min; the beads were then washed and incubated with excess soluble FXIa overnight. Five hundred clones from the third round were picked and tested for direct FXIa binding by ELISA. A total of 367 clones were ELISA-positive, and 87 clones were identified as likely to have higher affinity than parental D4 based on competition ELISA and homogeneous time-resolved fluorescence (HTRF) assays using D4 scFv as competitor. Of these 87 clones, 70 clones were unique and were reformatted to human IgG1. After assessing relative affinity for FXIa by HTRF assay with D4 IgG1 competition and IC50 for inhibition of SN-59 hydrolysis, nine clones were selected and their sequences were used to design 32 IgGs with different combinations of five mutations. These were characterized for FXIa binding by ELISA and Biacore and for inhibitory activity in the SN-59 hydrolysis assay. The clone with the highest affinity (~20 pM) for FXIa was designated C24. C24 had a possible deamination site in the CDRH2 at position 54S. This position was changed to 54E, and effector-null mutations in the Fc region were introduced to generate a new clone named DEF. DEF showed the same affinity and potency as C24 in in vitro, ex vivo, and in vivo assays (Tables 1 and 2, fig. S4, and table S1).

Anti-FXIa binding kinetics. Blood-derived FXIa and FXI (Haematologic Technologies Inc.) were biotin-labeled via primary amines and immobilized on a CAP chip using a Biacore T200 instrument (GE Healthcare). Binding versus time for D4 IgG, B11 IgG, C24 Fab, and DEF Fab was measured over a series of IgG or Fab concentrations. After each antibody injection, the chip surface was regenerated with a mixture of 6 M guanidine HCl and 0.25 M NaOH, and new FXIa/FXI was captured. Kinetic studies were performed using a flow rate of 50 μl/min in 0.01 M Hepes (pH 7.4), 0.15 M NaCl, and 0.005% (v/v) surfactant P20 (HBS-P buffer). Characterization of initial antibody leads was performed using IgG at 25°C. After affinity maturation, the off-rate at 25°C was too slow to be accurately measured for either IgG or Fab. Accordingly, off-rates for high-affinity antibodies were determined using Fab fragments at 37°C. All data were analyzed using the Biacore T200 Evaluation Software. Kinetic constants for at least three experiments were obtained and reported as the mean. Table of values for affinity measurements for IgGs and Fabs shows overall affinity of Fab or IgG for human FXIa target, as well as the rate constants. No binding was seen with immobilized human FXI in similar experiments, consistent with the FXIa/FXI ELISA data. Representative background-subtracted Biacore sensorgrams overlaid with the kinetic curve fits are shown in Fig. 1C.

Inhibition of FXIa activity as assessed by fluorogenic substrate hydrolysis

Human, cynomolgus macaque, and rabbit FXI were cloned, expressed and purified, and activated using bacterial thermolysin. Varying concentrations of antibody were preincubated for 5 min at 37°C with 200 pM human, cynomolgus macaque, or rabbit FXIa in 50 mM tris-HCl (pH 7.4), 250 mM NaCl, and 1 mM EDTA. This was followed by addition of 100 μM SN-59, a fluorogenic peptide substrate (Haematologic Technologies), to start the reaction. The plate was read at 1-min intervals in a SpectraMax fluorescence plate reader at 37°C for 30 min at an excitation wavelength of 353 nm and an emission wavelength of 470 nm. Instrument-determined Vmax values taken from the linear part of each reaction curve were then plotted to determine the IC50 values. In a subset of assays, bovine serum albumin (up to 40 mg/ml, ~human plasma levels) was added to determine whether plasma proteins might nonspecifically effect antibody function.

Assays for inhibition of other proteases. The following conditions were used for all the tested enzymes: 50 μl of diluted enzyme, 92 μl of standard assay buffer, and 50 μl of test IgG (DEF or C24) or buffer (no IgG wells) were preincubated for 5 min at 37°C. IgGs were tested at the indicated concentrations (fig. S4, A and B). The following fluorogenic substrates were then added at a final concentration of 100 μM to start the reaction: SN-59 for FXIIa, FXIa, APC, and PKa; SN-17A for thrombin and FVIIa; and SN-7 for FXa. The plate was read, and the Vmax values were determined as described above. Human enzymes and substrates were from Haematologic Technologies Inc., unless otherwise specified. Final enzyme concentrations were as follows: FXa (HCXA-0060) at 2 μg/ml, thrombin (HCT-0020) at 5 μg/ml, and FVIIa (HCVIIA-0031) at 5 μg/ml with tissue factor (RTF-0300) at 0.5 μg/ml and added phospholipid at 12 μM. FXIIa (HFXIIa1212a, Enzyme Research), FXIa (HCXIA-0160), APC (HCAPC-0080), and PKa (KLKB1; 2497-SE, R&D Systems) were at 0.9, 0.2, 1.2, and 4 μg/ml, respectively. The latter was purchased as PK zymogen and activated with thermolysin according to R&D’s instructions. Thermolysin alone had no detectable activity in the SN-59 hydrolysis assay. Because of the low efficiency of small peptide–substrate cleavage by FIXa, the activity against FIXa was assessed by using a coupled assay involving FIXa activation of FX in the presence of FVIIIa and phospholipid and the FXa chromogenic substrate SXa-11 (BIOPHEN FIXa, reference no. A221812, Aniara; used according to the supplier’s protocol). Incubation of FIXa with zymogen FX led to robust FXa substrate cleavage; FIXa or FX alone showed no detectable activity. Preincubation of FIXa with DEF or C24 as described above had no effect on FIXa activity, as measured by FX activation.

Ex vivo thrombin generation and plasma and whole-blood clotting assays

Thrombin generation in platelet-poor plasma was measured using a fluorogenic thrombin substrate and a multiwell automated fluorescent plate reader (ThrombinoSCOPE). Anti-FXIa or IgG1 control (Mab 8.8) antibody solution (5 μl) was mixed with 20 μl of phosphate-buffered saline (PBS) containing 60 nM human FXIIa (Enzyme Research Laboratories) and phosphatidylcholine/phosphatidylserine (Phospholipid-TGT, Diapharma) in a 96-well plate. Citrated platelet-poor human plasma (75 μl) (Triclinical Reference Plasma, TCoag) was added, and coagulation was triggered with the addition of 20 μl of calcium chloride buffer and fluorogenic thrombin substrate according to ThrombinoSCOPE’s protocol. Thrombin activity as a function of time was measured as the velocity of fluorogenic peptide hydrolysis. Total thrombin activity generated (commonly called endogenous thrombin potential), peak activity, and time to onset of thrombin generation (lag time) and time to peak activity were measured. The final concentrations of the antibodies tested (D4, B11, 24, DEF, and IgG1 control) ranged from 5 to 443 μg/ml. Because of lot-to-lot variability, the concentration of each lot of phospholipid/phosphatidylserine reagent was adjusted to achieve a ~15-min lag to onset and ~150-nM peak of thrombin activity. APTT was measured using standard pooled human platelet-poor plasma and a TriniCLOT aPTT kit according to the manufacturer’s instructions.

For whole-blood clotting, whole blood was collected into 3.8% citrate using a blood/citrate ratio of 9:1 (v/v). Five microliters of C24, control IgG1, or saline control was added to 300 μl of citrated whole blood. Samples were preincubated at 37°C for 5 min, and clotting was initiated by addition of 7 μl of in-tem reagent (ellagic acid/phospholipid, TEM Systems Inc.) and 20 μl of star-tem reagent [0.2 M CaCl2 in Hepes buffer (pH 7.4)]. Time to clotting was measured using a semiautomated coagulation analyzer (KC4 Delta, Tcoag).

FeCl3-induced carotid thrombosis in “FXI-humanized” mouse

FXI-knockout mice received an intravenous bolus injection of purified human FXI (0.25 mg/kg) and C24 (0.5, 2, 4, 12, and 35 mg/kg) or control IgG1 (same concentrations as C24) via tail vein. Fifteen minutes later, the mice were anesthetized. The left common carotid artery was exposed, and a flow probe (model MA0.5PSB, Transonic Systems Inc.) was placed around the artery, proximal to the bifurcation (46, 47). Filter papers soaked in a 250 mM ferric chloride (FeCl3) solution were placed above and below the artery for 3 min and then were removed. Arterial flow was measured continuously using a TS420 flow meter (Transonic Systems Inc.) connected to an ADInstruments PowerLab 4/30 and Chart software. Monitoring was continued until the artery was occluded (defined as no flow for ≥1 min) or for 20 min if no occlusion occurred. Human FXI preparations had no detectable (<1%) FXIa activity in the SN-59 hydrolysis assay.

Pharmacokinetics and pharmacodynamics in rabbits

Three rabbits in each dosing group received an intravenous bolus injection of DEF IgG (10, 1, 0.1, or 0.03 mg/kg) or control IgG (1 or 0.1 mg/kg). Blood samples (0.5 ml) were collected before and 0.02, 0.5, 1, 2, 4, 8, 24, 48, 72, 120, 168, and 336 hours after dosing and processed to serum and platelet-poor plasma. Serum samples were analyzed for DEF concentration by ELISA. Plasma samples collected before dosing and 30 min, 24 hours, 7 days, and 14 days after dosing were analyzed for drug effect and specificity by APTT and PT coagulation assays by Covance using standard assay formats.

Thread-induced venous thrombosis in rabbits

Rabbits were anesthetized, and six animals in each dosing group received an intravenous bolus injection of either DEF IgG (10, 3, 1, 0.1, or 0.03 mg/kg), control IgG (10 mg/kg), rivaroxaban (0.045 or 1.5 mg/kg), or its vehicle control (10% DMA/30% PEG-400/60% water). For rivaroxaban treatment, a bolus injection (0.6 mg/kg iv) was followed by a continuous intravenous infusion at 0.9 mg/kg per hour to maintain drug levels. For all treatment groups, 30 min after the bolus injection, a sheath was placed in the left femoral vein, and then a wire with eight 3-cm-long strands of cotton thread attached to its tip was inserted about 14 cm into the femoral vein (2931). Fluoroscopic imaging was used to guide this device in the inferior vena cava. The device remained in the inferior vena cava for 90 min, during which clots formed on the cotton threads. After 90 min, the device was removed by surgical dissection and the clot-bearing cotton threads were blotted to remove free blood and weighed. Before IgG dosing and again, just before removal of threads, blood samples were drawn for preparation of serum for measurement of antibody levels and for preparation of plasma for measurement of APTT and PT.

Rabbit cuticle bleeding

After anesthesia and cannulation of an ear vein for drug administration, both front paws were shaved to remove fur. Pre- and postdose bleeds were performed on the middle digits of the left and right front paws, respectively. There were four dosing groups of 10 animals each. After predose bleeds were completed, each dosing group received an intravenous bolus injection of DEF IgG (10 mg/kg), control IgG (10 mg/kg), and a bolus infusion of rivaroxaban or its vehicle as described above.

Cuticles were transilluminated with white light to visualize the quick and cut with a razor blade so as to transect the cuticle about 1 mm proximal to the apex of the quick. The nail was then immediately immersed in 37°C saline in a 10 mm–by–75 mm 3-ml polystyrene (clear) tube, and time to cessation of blood flow from the quick was measured. If bleeding did not stop, the procedure was terminated at 20 min. At the end of the blood collection, the tube was mixed and centrifuged at 250g for 15 min, and the blood pellet was resuspended in 3 ml of erythrocyte lysis buffer [NH4Cl (8.3 g/liter), KHCO3 (1.0 g/liter), and EDTA (0.037 g/liter) in water]. After at least 15 min of lysis time, hemoglobin concentration was measured as OD at 575 nm. OD values were converted to amount of loss using a standard curve generated with known volumes of blood. Antibody concentration, APTT, and PT were determined as described above using blood samples obtained before dosing and at the end of the study.

Selection of an anti-idiotype antibody to DEF as a reversal agent

DEF was biotinylated with Biotin-LC-NHS (Pierce, catalog no. 21347) according to the manufacturer’s protocol and used with streptavidin-coated Dynabeads M-280 (Invitrogen, catalog no. 11206D) to select binders from an scFv antibody phage display library. Before each round of selection, phage antibody library was absorbed to streptavidin-coated Dynabeads M-280 to deplete streptavidin binders. The phage selection was performed in the presence of 500 nM human serum IgG, using decreasing concentrations of the antigen (DEF) and increasing number of washes with PBS containing 0.1% Tween 20 (PBST), as follows: first round, 100 nM/5× PBST/2× PBS; second round, 10 nM/10× PBST/5× PBS; and third round, 1 nM/15× PBST/5× PBS. A total of 3000 clones were picked from the third round output and tested in scFv format by ELISA as follows: biotinylated control antibody (Mab 8.8) (1 μg/ml) and human serum IgG or DEF were added on to ELISA plate previously coated with streptavidin (10 mg/ml) by overnight incubation. After blocking and washing, scFv was added to each well and incubated at room temperature for 1 hour. After washing and treating with anti-myc-HRP to detect bound scFv, enzyme substrate (tetramethylbenzidine) was added to develop the color. The signal was read at 450 nm after stopping the reaction by adding 0.16 M sulfuric acid. A single specific anti-DEF clone, designated revC4, was identified, sequenced, and reformatted into a fully human IgG1 and was shown to retain DEF binding specificity.

Kinetics of DEF binding to revC4

revC4 IgG was captured by anti-human IgG (Fc) antibody amine-coupled to a CM5 chip using a Biacore T200 instrument (GE Healthcare). The anti-human IgG capture chip surface was prepared using Biacore Human Antibody Capture Kit according to the manufacturer’s directions (GE Healthcare). DEF Fab binding experiments were performed at 25°C using a flow rate of 30 μl/min in 0.01 M Hepes (pH 7.4), 0.15 M NaCl, and 0.005% (v/v) surfactant P20 (HBS-P) buffer. After each cycle, the chip surface was regenerated with 3 M MgCl2, and new revC4 antibody was captured. DEF Fab samples ranging from 2 to 100 nM were injected over the surface for 3 min, and the dissociation was monitored for a further 20 min. Data were analyzed with the Biacore T200 Evaluation Software, and the results were reported as the mean of two experiments.

Reversal of DEF activity by revC4 in vitro

Activity of C24 was initially assessed by preincubating 10 nM DEF IgG and revC4 IgG (4, 20, 100, or 500 nM) for 20 min in the FXIa assay buffer. Next, 0.7 nM FXIa was added and incubated for a further 5 min. FXIa activity was then assessed by adding SN-59 substrate and read as described above. In some experiments, the Fab form of revC4 was used and behaved similarly to the IgG form. For experiments involving human plasma, 5 μl of anti-FXIa DEF antibody (16 μg/ml) was first mixed with different amounts of the revC4 IgG (1, 5, 10, 20, 40, 60, and 80 μg/ml), and FXIIa-triggered thrombin generation was measured as described above.

Reversal of DEF activity by revC4 in rabbits

Five anesthetized rabbits were treated as follows: at time 0, each rabbit received an intravenous bolus injection of DEF IgG (1 mg/kg). Thirty minutes later, each rabbit received a bolus injection of control IgG (3mg/kg). After an additional 30 min, each rabbit received a bolus injection of revC4 IgG (3 mg/kg). Just before each bolus injection, and 30 min after the final injection (revC4 IgG), blood was drawn to make plasma for APTT and PT clotting time assays.

Pharmacokinetics and pharmacodynamics and exploratory toxicology in cynomolgus macaques

Two animals (one female and one male) in each dose group intravenously received DEF at 20, 75, or 266 mg/kg or subcutaneously received DEF at 266 mg/kg on days 1, 8, and 15, with necropsy occurring on day 16. Clinical signs and food consumption were evaluated daily. Body weight was determined weekly. Blood collection for changes in hematology and clinical chemistry parameters occurred before necropsy on day 16. Serum was collected 0.08, 6, 24, 168, 168.08, 174, 192, 240, 280, 288, and 336 hours after the first dose on day 1 for measurement of serum DEF levels by human IgG–specific ELISA. Platelet-poor plasma was collected on day −9; just before dosing and 1 hour after dosing on days 1, 8, and 15; and just before necropsy on day 16. Samples obtained on day −9 and just before dosing on day 1 were used to determine baseline values for APTT and PT; samples drawn after dosing were used to determine APTT and PT as pharmacodynamic markers of DEF action.


The number of replicates in individual experiments and the number of times an experiment was repeated are stated in the figure and table legends. The Kaplan-Meier–type data in Fig. 3 were analyzed using log-rank test (Mantel-Cox). The data in Figs. 5 (A and B) and 7E were analyzed using one-way ANOVA with follow-on Tukey’s test for multiple comparisons. Data in Figs. 5 (C to F) and 6 were analyzed using two-way ANOVA with follow-on Sidak’s test for multiple comparisons.

Study approvals

All animal studies were approved by the Institutional Animal Care and Use Committee of University of California, San Francisco (UCSF) for mouse studies, PMI for rabbit studies, and Pfizer for cynomolgus macaque studies.


Fig. S1. Schematic of coagulation cascade with emphasis on FXI and the intrinsic pathway.

Fig. S2. Schematic of phage selection and anti-FXI human mAb screening and optimization.

Fig. S3. FXIa-C24 binding is SDS-labile.

Fig. S4. C24 and DEF bind and inhibit FXIa but not other coagulation proteases, APC, or PKa.

Fig. S5. Effect of anti-FXIa antibodies on FXIIa-induced thrombin generation in human plasma.

Fig. S6. Effect of C24 on tissue factor–activated thrombin generation in human plasma.

Fig. S7. Addition of albumin at up to 40 mg/ml (human plasma levels) does not alter C24 or DEF inhibition of human FXIa–mediated small-substrate hydrolysis.

Fig. S8. Single-dose rabbit pharmacokinetics and pharmacodynamics study: Antibody levels and APTT.

Fig. S9. Pharmacokinetics and pharmacodynamics of DEF exposure during exploratory toxicology in cynomolgus macaques.

Fig. S10. Reversal of DEF inhibition of FXIIa-induced thrombin generation in human plasma by revC4 IgG1.

Fig. S11. Effect of DEF and revC4 IgG on PT clotting time.

Fig. S12. revC4 IgG and revC4 Fab block inhibition of FXIa-mediated SN-59 hydrolysis by DEF.

Fig. S13. FXIIa-triggered thrombin generation in FXI-deficient human plasma reconstituted with normal plasma or FXI.

Fig. S14. APTT in FXI-deficient human plasma reconstituted with normal human plasma or human FXI.

Table S1. Summary of potency data for D4, B11, C24, and DEF IgGs across biochemical, plasma- and blood-based, and in vivo assays.


Acknowledgments: We thank the following people for technical assistance in carrying out this work: N. Liu, J. Min-debartola, C. Concengco, Y. Srinivasan, D. Duong, D. Ha, N. Kim, M. Wolf, M. Drever, R. Groth, B. Huang, A. Wu, A. Valera, S. Kumar, S. Singh, T. Hickling, P. Vicini, L. Zhu, J.Vekich, M. Horn, and R. Roach. We also thank the following people for helpful discussions: A. Chew, A. Barry, M. Leach, J. Dal Porto, R. Lindberg, and T. S. Nissen. Funding: This work was funded by Pfizer. Author contributions: T.D., Y.C.K., and L.K.E. contributed to the design of these studies and to the execution of the in vitro and mouse experiments. H.G. and I.R. performed antibody selection and optimization; P.O. performed mouse and rabbit pharmacokinetics and pharmacodynamics; M.W.B. performed cynomolgus macaque pharmacokinetics and pharmacodynamics and exploratory toxicology; E.A.F. and J.L.G. performed rabbit thrombosis and bleeding studies; and A.J.C. performed additional critical review of the data and oversight. T.M. and S.R.C. designed and supervised the studies and wrote the manuscript. Competing interests: The following authors are listed as inventors on a patent application covering DEF and revC4: T.M., T.D., L.K.E., H.G., Y.C.K., I.R., and S.R.C. (“Antibodies to coagulation factor XIa and uses thereof," PCFC-973-001 Provisional Patent Application, filed by Pfizer Inc.). S.C. is on the Scientific Advisory Board of Portola Pharmaceuticals and is a consultant for Novartis. Data and materials availability: Materials are available via a material transfer agreement with UCSF.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article