Discovery and optimization of marstacimab, a human monoclonal antibody targeting tissue factor pathway inhibitor for the treatment of hemophilia A and B Authors: Macy Jin, Sunita Hett, Reema Jasuja, Swapnil Rakhe, Jatin Narula, Amy Tam, James R Apgar, Zong Sean Juo, Lidia Mosyak, Matthew Holsti, Alison Joyce, Susan Hurst, Robert Webster, Laura Lin, Mark Stahl, Debra D Pittman, Laird Bloom Published: 2026 ABSTRACT ABSTRACTWe report the discovery and optimization of marstacimab, a novel monoclonal antibody targeting tissue factor pathway inhibitor (TFPI), for the treatment of hemophilia A and B. Hybridoma and phage display approaches identified antibodies that blocked the TFPI: coagulation factor Xa (FXa) interaction. Antibodies bound TFPI with nanomolar affinity to multiple epitopes, including one that covered the entire FXa binding surface and others that partially overlapped the interface from different sides of the TFPI-K2 domain. Several antibodies reduced bleeding in a hemophilia A mouse injury model for up to 96 h following a single dose. Rabbit pharmacokinetic studies indicated that antibodies with ≤~1 nM TFPI-binding affinity were cleared rapidly from circulation, whereas TFPI-23 (5.72 nM) had a longer plasma residence time. Pharmacokinetic-pharmacodynamic modeling indicated that this intermediate affinity allowed circulating concentrations of antibodies such as TFPI-23 to maintain concentrations required for 50% residual activity; in contrast, higher-affinity antibodies quickly decreased to concentrations that would not effectively neutralize TFPI. The end-to-end process leading to final candidate selection incorporated rigorous assessment of biophysical properties, including thermal stability, aggregation propensity, viscosity, predicted immunogenicity, stable cell line productivity, and nonspecific binding. An optimized derivative of TFPI-23, TFPI-106 (PF-06741086, known as marstacimab), had the functional and biophysical properties with other characteristics suitable for clinical development and was selected as the final candidate. TFPI-106 elicited enhanced hemostasis in hemophilic plasma, shortened clotting time, and enhanced thrombin generation velocity. Marstacimab is now approved for patients with hemophilia A or B with or without inhibitors.KEYWORDS: Antibody optimization; developability; hemophilia, TFPI, antibody engineering Introduction Hemophilia A and B are genetic disorders characterized by a deficiency in clotting factors VIII (FVIII) and IX (FIX), respectively, leading to impaired hemostasis. Treatment for hemophilia has advanced significantly over the past few decades from traditional intravenous (IV) clotting factor replacement therapy, administered prophylactically or on demand,1 to nonfactor replacement therapies and gene therapies.2,3 For example, emicizumab is a bispecific monoclonal antibody (mAb) that aims to mimic the cofactor activity of FVIII for individuals with hemophilia A.4,5 An alternative approach is the restoration of thrombin generation through the inhibition of endogenous anticoagulant pathways. Tissue factor pathway inhibitor (TFPI) is a multivalent Kunitz-type serine protease inhibitor that regulates tissue factor (TF)-induced coagulation through factor Xa (FXa)-dependent feedback inhibition of the TF-factor VIIa (FVIIa) complex.6 TFPI is composed of three tandem Kunitz domains, K1–K3.7 K1 and K2 block the active sites of FVIIa7 and FXa, respectively, with the R107 residue within the K2 domain being the key site for FXa binding.8 Inhibition of TFPI can enhance coagulation by reversing its inhibition of FXa-dependent production of thrombin.9 This bypasses the need for FVIII and FIX replacement while maintaining the activity of other negative regulators of coagulation (Figure 1A). This, in turn, brings thrombin generation closer to normal and can restore hemostasis and may result in a reduction in the number of bleeding episodes in individuals with hemophilia A or B, with or without anti-FVIII or anti-FIX antibodies, referred to as inhibitors.10–13 (A) Intrinsic and extrinsic coagulation pathways. (B) Screening funnel. dPT = dilute prothrombin time; FIXa = activated factor IX; FVIIa = activated factor VII; FVIIIa = activated factor VIII; FV = factor V; FVa = activated factor V; FX = factor X; FXa = activated factor X; PK = pharmacokinetics; scFv = single-chain variable fragment; TF = tissue factor; TFPI = tissue factor pathway inhibitor; TGA = thrombin generation assay. Here, we report the end-to-end process of the discovery, selection, and optimization of an anti-TFPI mAb, marstacimab. We sought to identify therapeutic mAbs with the potential to provide sustained inhibition of TFPI with weekly subcutaneous (SC) dosing. Anti-TFPI antibodies previously reported were capable of potent TFPI neutralization, but exhibited a short half-life in animal studies, likely a result of target-mediated drug disposition and enhanced clearance.14,15 Our antibody discovery process was guided by a combination of in vitro affinity and neutralization assessment, in vivo pharmacokinetic (PK) studies, and PK-pharmacodynamic (PD) modeling to identify candidates that achieved hemostasis with long duration. In addition to identifying molecules with potent efficacy, the discovery process included a series of molecular assessments designed to generate a therapeutic with an optimal manufacturability profile characterized by minimal potential of posttranslational modifications and low immunogenicity risk, high conformational stability, and high solubility with low viscosity and low nonspecific binding, suitable for SC administration at a therapeutically efficacious dose.16 The resulting antibody, marstacimab, has subsequently been tested in vivo in animals and humans9–11,17–20 and received approval in the United States, Europe, and elsewhere for prophylaxis in patients with hemophilia A or B with or without inhibitors.21,22 Results Selection of anti-TFPI antibody lead candidates from phage display and hybridomas Using a combination of phage display and hybridoma technologies, a panel of anti-TFPI antibodies was identified. A nonimmunized human single-chain variable fragment (scFv) phage display library was subjected to four rounds of selection for phage that bound to a reagent containing the K1 and K2 domains of TFPI (TFPI-K1-K2). From 1984 phage clones from Rounds 2–4, 883 TFPI binders were identified and sequenced, leading to the identification of 288 unique TFPI-binding scFvs. These were tested in the form of crude periplasmic extracts for competition with TFPI-FXa binding, and 48 inhibitors were identified. Following reformatting to immunoglobulin G (IgG) and screening for blockade of TFPI-FXa binding and reversal of TFPI inhibition of TF/FVIIa/FXa (data not shown), three lead candidates, designated TFPI-21, TFPI-23, and TFPI-24, were selected and scaled for further characterization (Figure 1B). In parallel, BALB/c mice were immunized with human and mouse TFPI-K1-K2, and hybridomas were generated from the lymph nodes of seropositive mice. Supernatants from 1440 hybridoma cultures were tested for binding to human TFPI-K1-K2 by enzyme-linked immunosorbent assay (ELISA), leading to the identification of 93 TFPI-binding hybridomas. Based on their binding activity and their efficacy in initial functional assays, antibodies were selected from three hybridoma clones and designated 4D8, 6B7, and 7A4. Of these, 4D8 was selected as the lead candidate, humanized, and scaled for further characterization (Figure 1B). Functional screening assays The functional effects of TFPI blockade by TFPI-21, TFPI-23, TFPI-24, and 4D8 were quantitated in several ways (Figure 2). In an assay measuring the reversal of inhibition of FXa by TFPI (Figure 2A), all antibodies reversed inhibition of FXa. The antibodies derived from phage display (TFPI-21, TFPI-23, and TFPI-24) were less potent than the hybridoma-derived 4D8 at lower concentrations (10–50 nM), but all antibodies reached a similar maximum level of inhibition. Similarly, in the two-stage TF-FVIIa-FXa inhibition reversal assay, which tests the ability of purified anti-TFPI antibodies to restore TF/FVIIa activation of factor X (FX) in the presence of inhibitory concentrations of TFPI (Figure 2B), all antibodies were able to neutralize TFPI-K1 domain inhibition of the TF/FVIIa complex. Similarly to their performance in the FXa inhibition reversal assay, phage display-derived antibodies tended to have lower potency than hybridoma-derived mAbs, but achieved similar maximum inhibition. In a thrombin generation assay (TGA) in human FVIII-deficient plasma, all the antibodies tested had a similar ability to restore thrombin generation in plasma at concentrations above 10 nM, with the phage-derived antibodies again showing lower potency at low concentrations (Figure 2C). In vitro functional analysis of anti-TFPI antibody candidates’ ability to inhibit human TFPI-K1-K2 (A) FXa inhibition assay (B) two-stage FVIIa/FX inhibition assay and (C) thrombin generation assay. FVIIa = activated factor VII; FX = factor X; FXa = activated factor X; hIgG1 = human immunoglobulin G1; mIgG1 = murine immunoglobulin G1; TFPI = tissue factor pathway inhibitor. Binding affinity: surface plasmon resonance Surface plasmon resonance (SPR) analysis showed that all four antibody candidates bound human TFPI with affinities that paralleled their functional potency. 4D8, the most potent neutralizing antibody, had the highest affinity for human TFPI (1.25 nM), whereas TFPI-24, TFPI-23, and TFPI-21 had affinities of 5.46, 9.89, and 44.0 nM, respectively, mirroring their rank order in the FXa and two-stage FXa/FVIIa assays (Table 1). These antibodies bound rabbit TFPI (75.5% identity to human TFPI) with similar affinities to their affinity to human TFPI (Table 1). All antibody candidates except for 4D8 bound to mouse TFPI (65.5% identity to human TFPI). However, TFPI-21 and TFPI-23 bound mouse TFPI with a higher affinity than human TFPI, whereas TFPI-24 bound mouse TFPI with a lower affinity than human TFPI. ELISA analysis of TFPI-21, TFPI-23, and TFPI-24 showed that these antibodies did not bind to recombinant human TFPI-2 protein, the closest homolog of TFPI, with 26.2% amino acid identity (Figure S1). Table 1.Anti-TFPI antibody binding affinity to human, mouse, and rabbit TFPI K1-K2 protein by SPR analysis.Anti-TFPI antibody ligandBinding affinity to TFPI K1-K2 protein, Kd (nM)HumanMouseRabbit4D81.25NB0.640†TFPI-2144.04.5434.9TFPI-239.891.475.72TFPI-245.4624.12.99All data represent the average of two measurements.†Represents 4D8.B1, a chimeric version of the mouse hybridoma 4D8, with mouse-derived VH and VL domains grafted to human IgG1 constant domains.Kd = equilibrium dissociation constant; NB = no binding; SPR = surface plasmon resonance; TFPI = tissue factor pathway inhibitor. Anti-TFPI antibody binding affinity to human, mouse, and rabbit TFPI K1-K2 protein by SPR analysis. All data represent the average of two measurements. †Represents 4D8.B1, a chimeric version of the mouse hybridoma 4D8, with mouse-derived VH and VL domains grafted to human IgG1 constant domains. Kd = equilibrium dissociation constant; NB = no binding; SPR = surface plasmon resonance; TFPI = tissue factor pathway inhibitor. Epitope binning/epitope mapping of anti-TFPI antibodies by SPR The distribution of TFPI binding epitopes among the antibody candidates and comparator antibody MAB2974, a commercially available anti-TFPI antibody, was further explored in a sandwich SPR assay. TFPI-21 and TFPI-23 showed similar patterns, competing with each other and with TFPI-24 for binding to TFPI-K1-K2, but not with 4D8 or MAB2974 (Table S1). Similarly, 4D8 and MAB2974 competed with one another and with TFPI-24, but not with TFPI-21 or TFPI-23. Taken together, these data indicated that TFPI-21 and TFPI-23 share an epitope, distinct from the epitope bound by 4D8 and MAB2974. TFPI-24 displayed no binding in the presence of any other antibodies tested, indicating that TFPI-24 has an epitope that overlaps the other two clusters. X-ray crystallography and modeling: lead anti-TFPI antibody candidates block FXa interaction through distinct epitopes Antigen-binding fragments (Fabs) from the antibodies TFPI-23, TFPI-24, and 4D8 were cocrystallized with the cynomolgus monkey TFPI-K2 domain,23 and the X-ray crystal structures were solved to obtain a detailed map of the interaction between the antibodies and TFPI (Figure 3 and Table S2; Protein Data Bank accession numbers 35UH, 35UI, and 35UJ, respectively). A published structure of anti-TFPI mAb 2021 (now known as concizumab; Protein Data Bank accession 4DTG) was also included for comparison.24 All TFPI-binding antibodies bound to the TFPI-K2 domain via distinct sets of contacts, with overlapping binding interactions that corroborated the epitope bins established by competitive binding. TFPI-24 (Figure 3A, upper left) closely mimicked the interaction of FXa (Figure 3B), directly covering the surface that has the critical TFPI amino acid residue R107 at its center. The antibodies 4D8, and mAb 2021 recognized overlapping epitopes (the lower surface of TFPI-K2 in Figure 3), whereas TFPI-23 occupied a nonoverlapping epitope on the opposite face. Both TFPI-23 and mAb 2021 engaged with the TFPI amino acid residue R107 (Table S2 and Figure S2), the residue central to TFPI binding to FXa. However, they contacted TFPI from opposite sides of the protein. X-ray crystal structure of TFPI blockade via multiple epitopes. (A) the complex structure of TFPI-K2 domain with anti-TFPI antibody candidates TFPI-23, TFPI-24 and 4D8 along with mAb 2021, a comparator molecule. (B) the structure of FXa in complex with the TFPI-K2 domain.23 FXa = activated factor X; R107 = arginine 107; TFPI = tissue factor pathway inhibitor. Hemophilia A (FVIII deficient) mouse acute tail clip injury model of hemostasis A mouse acute tail clip injury model of hemostasis using FVIII-deficient mice (a model for hemophilia A) was used to test the efficacy of the antibody candidates. In this severe injury model, a small portion of the tail is amputated, and the amount of blood loss and the time required for clot formation are measured. A single 6-mg/kg dose of TFPI-21, TFPI-23, and TFPI-24 decreased blood loss compared with control-dosed mice when administered up to 96 h prior to the tail clip (Figure 4). 4D8, which did not show significant binding to mouse TFPI, was not effective at reducing blood loss in this model. In vivo efficacy of anti-TFPI antibody candidates in hemophilia A mouse acute tail clip injury model of hemostasis. The effect on bleed time in hemophilia A mice after tail transection following dosing at different times before tail transection with antibodies TFPI-21, TFPI-23 and TFPI-24, and 4D8 compared with vehicle control (saline). Hemophilic mice received a single intravenous injection of 6 mg/kg of anti-TFPI antibodies via the tail vein (in a volume of 4 mL/kg) or saline vehicle. Data represent mean (±95% confidence intervals) for five animals for TFPI-21, TFPI-23, and TFPI-24, one to three animals for saline control and two animals for 4D8 measurements. TFPI = tissue factor pathway inhibitor. PK/PD of anti-TFPI antibodies in rabbits Long-lived therapeutic efficacy of anti-TFPI antibodies is expected to require a balance between binding that is sufficiently strong for effective blockade of the TFPI-FXa interaction and binding that does not lead to excessive TFPI-mediated antibody clearance (Figure 5A). To identify candidates with suitable PK, antibodies were intravenously administered to New Zealand white rabbits at 2 mg/kg. The serum concentrations of all anti-TFPI antibodies decreased faster than that of the control IgG, suggestive of target-mediated drug disposition (Figure 5B). TFPI-24 and 4D8 showed similar rapid clearance, approaching the limit of detection by 4–6 days after dosing. However, TFPI-23 displayed notably longer plasma residence time than the others and had measurable concentrations 8 days post dosing. PK/PD of anti-TFPI antibody candidates TFPI-23, TFPI-24, and 4D8. (A) schematic of the impact of target affinity on PK/PD of anti-TFPI antibody candidates. (B) affinity-dependent PK in rabbits. (C) model predictions for TFPI receptor occupancy. (D) model predictions for dose-dependence of trough TFPI receptor occupancy for SC weekly antibody dosing. (E) model predictions for serum drug exposure after repeated SC antibody administration. Panel B: Affinity-dependent PK in rabbits (N = 3, 3 mg/kg). Circles and error bars represent observed mean serum concentrations ± SD; lines indicate the corresponding 2-compartment MM model fits. Note: most samples were BLQ for 4D8 post day 4 (not shown). (C) Model predictions for TFPI receptor occupancy at 3 mg/kg SC weekly dosing based on observed PK and TFPI affinity. (D) Model predictions for trough TFPI receptor occupancy at 3 mg/kg SC weekly dosing. (E) Model predictions for serum drug exposure after repeated SC antibody administration at 3 mg/kg relative to dPT assay potency. Dashed horizontal lines indicate dPT assay EC50values as follows: 4.16 nM (0.63 mg/mL), 1.02 nM (0.15 mg/mL) and 0.530 nM (0.08 mg/mL) for TFPI-23, TFPI-24, and 4D8, respectively. Kdvalues were: 5.72 nM, 2.99 nM and 0.64 nM for TFPI-23, TFPI-24, and 4D8, respectively. Panel E: Data for TFPI-23, TFPI-24 and 4D8 are presented. Note: data for TFPI-24 and 4D8 are overlapping so only 4D8 is visible. BLQ = below limit of quantification; hIgG = human immunoglobulin G (control); IC50 = half maximal inhibitory concentration; MM = Michaelis Menten; PD = pharmacodynamic; PK = pharmacokinetics; dPT = dilute prothrombin time assay; SC = subcutaneous; TFPI = tissue factor pathway inhibitor. The PK parameters of the antibodies were characterized using a two-compartment model with Michaelis-Menten (MM) elimination, a commonly used mathematical model for antibodies with target-mediated clearance.25 Modeled PK for all antibodies fit the observed profiles closely. Based on the model fits, the estimated MM constants (Km) for the antibodies, i.e., the plasma drug concentrations at which the target (TFPI)-mediated drug elimination is half-saturated, were dependent on the in vitro measured target affinity. This is consistent with previously reported analyses of target-mediated drug disposition pharmacology, which shows that the concentration dependence of in vivo clearance is sensitive to the drug-target affinity.26 Specifically, with reference to Kd values for binding of antibody candidates to rabbit TFPI, the estimated Km for TFPI-23 was 10× higher than that for 4D8. Clearance (CL) tracked approximately with Kd (Figure S3). Vmax and other PK parameters for all tested anti-TFPI antibodies were comparable (data not shown). Model simulations were also used to compare the projected TFPI receptor occupancy (RO%) for the antibody candidates with varying affinities with weekly dosing. As shown in Figure 5C, for weekly SC dosing in rabbits, the predicted TFPI RO% was dependent on the affinity of the individual antibody for rabbit TFPI. TFPI-23, which had a higher Kd value (i.e., lower binding affinity), was projected to maintain higher TFPI occupancy at trough than antibodies with higher affinity (Kd values < 1 nM). Similarly, dose-response predictions for trough TFPI RO% (Figure 5D) showed that TFPI-23, with a weaker binding affinity for rabbit TFPI, was able to maintain higher TFPI RO% at steady state with SC weekly dosing at the same dose as the 4D8 antibody with subnanomolar binding affinity. Model predictions for serum drug exposure after repeated SC administration of antibody candidates at 3 mg/kg are shown in Figure 5E, relative to their potency to inhibit endogenous TFPI in rabbit plasma in the dPT assay (dashed horizontal lines, EC50 values shown in Table S3). 4D8 rapidly decreased below its EC50 value at trough levels, leading to low RO%. In contrast, TFPI-23 stayed above its EC50 at trough levels and maintained a higher RO% than 4D8. Overall, consistent with previous analyses of optimal affinity for antibody therapeutics,27 these results highlight the trade-off of increased clearance versus potency related to the affinity of anti-TFPI antibodies for TFPI and indicate that lower affinity for TFPI can offer an advantage in terms of maintaining target occupancy with longer dosing intervals. Optimization of anti-TFPI antibody lead candidates TFPI-23 and TFPI-24 were selected for further development based on their mid-range affinity. Both antibodies were fully human, isolated as scFv from a phage display library derived from B cells from nonimmunized human donors. The variable region frameworks of TFPI-23 have only three amino acid differences from the closest human germlines. The TFPI-23 variable light (VL) framework is identical to IGL1-40*01 (Figure S4A), and the TFPI-23 variable heavy (VH) framework has only three differences from the IGHV3-23*01 human germline framework sequence. Two variants that restored either two or all three of these germline residues were tested for binding (Figure S4B). TFPI-106, which contained Q1E and V5L mutations (Kabat numbering), retained full binding to human TFPI-K1-K2 as measured by SPR (Figure S4C), whereas TFPI-107, which contained these two mutations as well as I94K, immediately adjacent to complementarity-determining region (CDR) H3, lost all binding to human TFPI. A similar strategy was used for TFPI-24, which had 12 framework residues different from the closest human germlines, IGLV1-47*01 for the VL and IGHV3-9*01 for the VH (Figures S5A and S5B). Four VL variants and three VH variants of TFPI-24 were paired with parental sequences or with one another for expression and purification. All variants retained binding to human TFPI-K1-K2 by SPR (Figure S5C) and were further tested in multiple functional and biophysical assays (see below). Biophysical triage To identify antibody candidates with an optimal developability profile, a staged series of molecular assessment experiments was conducted (Figures S6). Early triage conducted with small quantities of protein excluded antibodies with undesirable characteristics that could affect stability, aggregation propensity, and expression level (Figure S6). Among variants of TFPI-24, several could be eliminated from further consideration based on their reduced thermal stability in a differential scanning calorimetry (DSC) assay. TFPI-124 had a Tm1 of 65.0°C, compared to 68.7°C for TFPI-24 (Figure S6A), whereas TFPI-115 and TFPI-118 showed enhanced stability (Tm1 of 73.3°C and 72.5°C, respectively). Similarly, in an assay of aggregation conducted over a range of temperatures, several antibodies showed near-complete aggregation at 60.0°C (TFPI-121, TFPI-122, and particularly TFPI-123, which was >50% aggregated at 54.0°C), whereas TFPI-115 and TFPI-118 showed little aggregation at 60°C, better than TFPI-24 (Figure S6B). TFPI-118 was selected for further evaluation, in addition to TFPI-106, the optimized derivative of TFPI-23. Functional analysis of TFPI-106 and TFPI-118 The two optimized antibodies were analyzed by SPR to determine their binding kinetics to TFPI from different species followed by an evaluation of the hemostatic efficacy in functional assays. TFPI-106 and TFPI-118 bound human TFPI-K1-K2 with Kd values of 3.70 and 9.61 nM, respectively (Table S4), comparable to 9.89 and 5.46 nM for TFPI-23 and TFPI-24, respectively, prior to optimization. TFPI-106 and TFPI-118 bound with several-fold higher affinity to cynomolgus monkey TFPI-K1-K2 (1.22 and 1.80 nM, respectively) than human TFPI, and their binding to rat and rabbit TFPI were in a similar range to their binding to human TFPI (Table S4). They diverged more significantly in their binding to mouse TFPI K1-K2, with a Kd of 0.575 nM for TFPI-106 and 45.6 nM for TFPI-118. In vitro hemostatic activities of the antibody candidates and their optimized counterparts were examined in human FVIII-deficient plasma shortening of dilute prothrombin time (dPT). As observed with nonoptimized TFPI-23 and TFPI-24, TFPI-106 and TFPI-118 led to a similar clotting time in FVIII-deficient human plasma and were similar to nonoptimized TFPI-23 and TFPI-24 (Figure 6A). Similar to their nonoptimized counterparts TFPI-23 and TFPI-24, TFPI-106 and TFPI-118 were effective at reducing bleeding in the tail clip model of hemophilia A at a dose of 6 mg/kg (Figures 6B,C). TFPI-118 reduced bleeding for at least 96 h, whereas TFPI-106 retained a significant reduction out to 189 h. TFPI-106 was also tested in a tail clip model of hemophilia B and reduced bleeding for at least 72 h (Figure 6D). In vitro and in vivo functional analysis of TFPI antibody candidates. (A) effect of anti-TFPI antibodies on dPT assay. (B) TFPI-106 in the hemophilia A mouse acute tail clip injury model. (C) TFPI-118 in the hemophilia A mouse acute tail clip injury model. (D) TFPI-106 in the hemophilia B mouse acute tail clip injury model. Panel A: in vitro concentration-response curves of anti-TFPI antibodies on shortening of dPT in human FVIII-deficient plasma. Samples were run in duplicate. Panels B – D: animals were dosed intravenously at 6 mg/kg at the indicated time points (hours before injury). Total volume of blood loss (mL) was then measured after tail transection. Vehicle (saline) treated hemophilia A mice (panels B and C) or hemophilia B mice (panel D) served as a control. Error bars represent 95% confidence intervals. N = 4–5/group. * indicates significant difference from saline control, p < 0.05 from Dunnett’s test. dPT = dilute prothrombin time assay; TFPI = tissue factor pathway inhibitor. Manufacturability assessment TFPI-106 and TFPI-118 were further assessed for manufacturability. Both antibodies retained low viscosity at high concentration (apparent viscosity of 7 and 8 cP, respectively, at 162 mg/mL, and 16 and 20 cP, respectively, at 200 mg/mL), as measured using a dynamic light scattering (DLS) tracer bead-based method (Figure 7A). By DSC, TFPI-106 and TFPI-118 both showed high thermal stability, with onset of melting >60.0°C and Tm1 at 69.57°C and 69.59°C, respectively (Figure 7B). TFPI-106 and TFPI-118 demonstrated good stability following storage up to 8 weeks at room temperature, as evidenced by minimal aggregate formation (Figure 7C) and retention of binding activity in an extended stress test at high concentration (50 and 150 mg/mL) (Figures 7D,E). TFPI-106 showed low-level nonspecific binding (comparable to the negative control) to ELISA plates coated with the negatively charged substrates DNA, insulin, and lipopolysaccharide, which correlates with normal clearance from circulation in humans.28 TFPI-118 showed moderately higher levels of nonspecific binding (1.5- to 4-fold higher than TFPI-106) (Figure S7). Finally, both antibodies were produced in a stable Chinese hamster ovary (CHO) culture system. In 1.5-L cultures, TFPI-106 was expressed at 899 mg/L, whereas TFPI-118 was expressed at 497 mg/L (Table S5). Biophysical characterization and molecular assessment of anti-TFPI antibody candidates TFPI-106 and TFPI-118. (A) apparent viscosity. (B) thermal stability. (C) aggregate formation following storage. (D) binding activity of TFPI-106 following storage and (E) binding activity of TFPI-118 following storage. Panel A: apparent viscosity assessed by dynamic light scattering bead-tracer method. Data represent mean ±1 SD; N = 15. (B) thermal stability assessed by differential scanning calorimetry. (C) aggregate formation assessed at W0 and W8 following storage (25°C; antibody concentration, 150 mg/mL). (D and E) binding activity assessed by ELISA at W0 and W4 and W8 following storage (5–40°C; antibody concentration, 50 and 150 mg/mL). HMMS = high molecular mass species; LMMS = low molecular mass species; TFPI = tissue factor pathway inhibitor; W = Week. Immunogenicity assessment: TFPI-106 and TFPI-118 have low immunogenicity potential Based on in silico analysis and in vitro cell-based assays, TFPI-106 and TFPI-118 were predicted to have low immunogenicity. In silico analysis of the TFPI-106 sequence identified four predicted T-cell epitopes: three overlapping epitopes in CDR H3 and one epitope in CDR L3 (Table S6). The amino acid mutations in the Fc region did not create a predicted T-cell epitope. None of the four predicted T-cell epitopes was presented by dendritic cells in a mass spectrometry–based presentation assay. The overall CD4 T-cell response to TFPI-106 was low, with 5% (n = 2/40) of donors responding to the whole molecule presented in a dendritic cell–T-cell (DC-T) coculture assay. In silico analysis of the TFPI-118 sequence identified nine predicted T-cell epitopes: two overlapping epitopes each in CDR H1 and CDR H2; three overlapping epitopes in CDR L2, and single epitopes in CDR L1 and CDR L3 (Table S6). The amino acid mutations in the Fc region did not create a predicted T-cell epitope. None of the predicted T-cell epitopes were presented by dendritic cells, measured by mass spectrometry. The overall CD4 T-cell response to TFPI-118 was low, with 10% (n = 4/40) of donors responding to the molecule. Selection of the final clinical candidate TFPI-106 and TFPI-118 exhibited nanomolar binding affinity to human TFPI, enabling them to bind TFPI tightly enough to block its activity while avoiding rapid clearance from the circulation. Based on the totality of the data–binding kinetic properties, performance in the series of molecular assessment assays, CHO expression, in vitro hemostatic assays, and in vivo mouse efficacy–TFPI-106 was selected as the clinical candidate and designated as PF-06741086, now known as marstacimab. Discussion Interaction of TFPI with the TF-FVIIa complex and FXa via K1 and K2 domain results in the physiological inhibition of the extrinsic and intrinsic coagulation pathways.6 Neutralization of TFPI activity may increase coagulation activity from intrinsic and extrinsic pathways and is therefore an attractive therapeutic target for the management of bleeding in patients with hemophilia A or B who have deficiencies in coagulation FVIII or FIX. When this antibody discovery effort was undertaken, a substantial challenge in hemophilia treatment was the absence of prophylactic medicines for patients with either hemophilia A or B, with or without inhibitors, that could be self-administered at a convenient interval.29 Although antibodies targeting TFPI to enhance the extrinsic clotting pathway in hemophilia A and B have been effective,10–13 initial experience in preclinical studies and in the clinic has shown that these antibodies experience rapid target-mediated clearance,14,15 which necessitates frequent dosing. Here, we have described the discovery and optimization of a therapeutic anti-TFPI antibody, now known as marstacimab, with a desirable manufacturability profile along with potent, long-lasting hemostatic efficacy in models of hemophilia A and B. Phage display and hybridoma technology were successfully used to obtain panels of antibodies that bound to TFPI in its first or second Kunitz domains (K1 and K2). Antibodies that blocked the interaction of purified TFPI-K1-K2 and FXa or a FXa/FVII complex and that enhanced thrombin generation from hemophilia A plasma were narrowed down to a set of four: TFPI-21, TFPI-23, and TFPI-24 derived from phage display and 4D8 from mouse hybridoma. The hybridoma-derived antibody 4D8 bound to TFPI with relatively high affinity (1.25 nM for human TFPI), whereas phage-derived antibodies bound with lower affinity (5.46–44.0 nM), and binding affinity generally paralleled their rank order in functional assays. Studies have shown that the R107 residue within the TFPI K2 domain is critical for FXa binding.8 X-ray crystal structure analysis demonstrated this panel of antibodies and a set of comparator antibodies achieved blockade of the TFPI-FXa interaction by binding to distinct epitopes on the K2 domain of TFPI. Although TFPI-24 occupied the same face of the TFPI K2 domain bound by FXa, centered on the critical amino acid R107 residue of TFPI, the other K2 domain-specific antibodies formed two separate clusters of epitopes on opposite faces of the TFPI molecule, each blocking FXa binding sterically, while partially overlapping portions of the TFPI-24 epitope. Indeed, TFPI-23 and the comparator mAb 202124 interacted with an almost completely nonoverlapping set of TFPI residues, except for R107, which they approached from opposite sides of the TFPI molecule. TFPI-21, TFPI-23, and TFPI-24 demonstrated efficacy in the hemophilia A mouse acute tail clip injury model up to 72 h post dose, demonstrating that TFPI binding antibodies exhibit functional activity in a physiologically relevant hemostatic bleeding model. Higher TFPI binding affinity was expected to increase inhibition potency and target-mediated clearance, with a trade-off of increased clearance versus potency, whereas mid-range TFPI affinity was expected to offer an advantage in terms of maintaining target occupancy with longer dosing intervals.17,18 Therefore, the PK of TFPI-23, TFPI-24, and 4D8, which span a range of nine-fold in affinity for rabbit TFPI, was examined in rabbits to identify an appropriately balanced affinity. TFPI-23, with an affinity of 5.72 nM to rabbit TFPI, had a significantly longer plasma residence time and lower clearance compared with the other antibodies. TFPI-23 was also projected to maintain higher TFPI occupancy at trough levels in rabbits than the higher-affinity antibodies because it remains in circulation at concentrations above its EC50, while the higher-affinity antibody 4D8 rapidly fell below detectable concentrations. The final candidate, TFPI-106, was derived from TFPI-23 by replacement of two non-human-germline amino acids in the VH framework, leaving only a single non-germline framework residue that was required for retention of binding. Similarly, a series of variants of TFPI-24 replaced subsets of its 12 non-germline framework residues with germline sequences, and through functional and biophysical assessment (see below), the final variant TFPI-118, with only three non-germline residues remaining, was selected. The binding affinities to human TFPI, including dissociation rates, were largely unaffected by these germline sequence substitutions. Hemostasis assays, including the dPT assay, which evaluated extrinsic and common pathways of coagulation, and TGA, which measured thrombin generation, a primary physiological event initiating clotting, have been tested to assess TFPI neutralization activity in the presence of anti-TFPI antibodies.30,31 As FVIIIa and FIXa form the same enzymatic complex crucial for activating FX to FXa in the coagulation pathway, FVIII-deficient models can be used to predict activity for hemophilia A and B. We used human FVIII–deficient plasma for a relative comparison between the antibody candidates; no significant difference in hemostatic activities between TFPI-23 and TFPI-24, and their optimized counterparts TFPI-106 and TFPI-118, was observed. The success of biotherapeutics in clinical development depends not only on their biological activity, but also a set of molecular properties described collectively as “developability”,16 including biophysical properties (stability, solubility, aggregation propensity, and viscosity), characteristics affecting production (expression and aggregates produced during expression), and sequence-related liabilities (nonspecific binding and sequences potentially leading to immunogenicity). TFPI-106, TFPI-118, and other variants of TFPI-23 and TFPI-24 were subjected to a series of developability assessments to determine their suitability for clinical development, staged from simple high-throughput assays that could be conducted on limited amounts of sample material from multiple candidates and progressing to high-concentration studies that required >100 mg of the most advanced molecules. Assessment of thermal stability by DSC and aggregation propensity at elevated temperature can be done on <1mg of protein. These assays were able to rapidly rule out derivatives of TFPI-24 with lower melting temperatures or with greater propensity for the onset of aggregation at lower temperatures, leading to the selection of TFPI-118 as the best of the TFPI-24 derivatives.32 A test of nonspecific binding to negatively charged substrates, which has been correlated with the probability of rapid clearance in vivo due to nonspecific interactions,28 can be performed on microgram quantities of protein and demonstrated that TFPI-106 showed only background levels of binding and therefore had minimal risk, whereas TFPI-118 demonstrated binding slightly above background. Among assays requiring high concentrations of protein, the assessment of viscosity offers a strong indication of suitability for SC injection. TFPI-106 and TFPI-118 demonstrated low viscosity (<20 cP, even up to 200 mg/mL). Similarly, protein behavior at high concentration for extended periods can reveal liabilities that could limit storage life or stability. Under the stress of an extended period at high concentration at room temperature, both antibodies showed minimal increases in high- and low-molecular mass material. Following 8 weeks at 25°C at 150 mg/mL or 4 weeks at 40°C at 50 mg/mL, both antibodies retained full binding activity, indicating high stability. The induction of an antidrug antibody response to protein therapeutics can arise from T-cell epitopes within the protein sequence itself, which are predicted to be presented on class II major histocompatibility complex (MHC) for recognition by corresponding T-cell receptors.33 TFPI-106 and TFPI-118 were predicted to have a low probability of inducing immunogenicity based on in silico predictions, mass spectrometric assessment of peptides presented on MHC on dendritic cells, and in vitro T-cell stimulation in a DC-T coculture assay. Taken together, the molecular properties of TFPI-106 and TFPI-118 determined by the suite of developability assays indicated that both were suitable to move forward for clinical development, with TFPI-106 showing superiority in stable CHO cell productivity, as well as in its nonspecific binding profile and projected immunogenicity risk. TFPI-106 (later named marstacimab) was selected as the clinical candidate, supported by its projected ability to maintain TFPI inhibition significantly longer than higher-affinity TFPI antibodies. These conclusions were borne out in subsequent Phase 1b/2 clinical studies of marstacimab, in which treatment-related changes suggested hemostatic pharmacology for all PD biomarkers, and effective targeting of the TFPI pathway with a weekly dosing regimen.20 The moderate affinity of marstacimab leads to a PK-PD profile that differs from those of higher-affinity TFPI antibodies that have been tested in humans. It is possible that the moderate affinity of marstacimab enables normalization of coagulation pathway activity while avoiding over inhibition of the coagulation pathway, as seen in clinical ex vivo results.34 The comparator antibody mAb 2021, known as concizumab, has a reported affinity to human TFPI of 25 ± 8pM,24 while befovacimab (formerly known as BAY-1093884) has a reported binding affinity of <10pM.35 The clinical development of befovacimab was halted due to safety concerns.36 Concizumab (Alhemo®) has been shown to reduce annualized bleeding rate (ABR) compared with prior prophylaxis in patients with hemophilia A or B with or without inhibitors 12,13 and is approved for prophylactic treatment for patients with hemophilia A or B with or without inhibitors.37,38 However, concizumab is approved for once-daily SC dosing. The safety and efficacy of a once-weekly SC dose of marstacimab for people with severe hemophilia A (FVIII < 1%) or moderately severe to severe hemophilia B (FIX ≤ 2%) was demonstrated in the pivotal Phase 3 BASIS trial (NCT03938792).10,11 Once-weekly marstacimab resulted in a significant reduction in ABR up to 12 months compared with previous factor replacement therapy in participants without inhibitors.11 The results for the cohort of participants with inhibitors have also demonstrated once-weekly marstacimab significantly reduced ABR for up to 12 months compared with previous bypassing agents in participants with severe hemophilia A or B with inhibitors.10 On the basis of these results, marstacimab (Hympavzi®) was approved for prophylactic treatment of patients with hemophilia A or hemophilia B with or without inhibitors, with weekly dosing.21,22 Materials and methods Protein reagents Recombinant protein reagents used for immunization (for hybridoma screening), phage display selection, and characterization of anti-TFPI antibodies are listed in Table S7. Recombinant reagents included a C-terminal AviTag™ (Avidity LLC) for site-specific biotinylation followed by a 10-residue histidine tag for purification. All TFPI variants were transiently expressed from human embryonic kidney cells (HEK293), and the resultant proteins were isolated from the supernatant and purified by immobilized metal affinity chromatography followed by size exclusion chromatography. Human FXa was purchased from Haematologic Technologies, Inc. (catalog no. HCXA-0060). The commercially available research antibody MAB2974 was obtained from R&D Systems (catalog no. MAB2974). Human hemophilic plasma was obtained from George King Biomedical (catalog no. 0800). Normal rabbit plasma was obtained from Bio-IVT-Bioreclamation LLC (catalog no. RAB00PL32NC-0102879). Animals These studies were conducted in accordance with current guidelines for animal welfare and approved by the Institutional Animal Care and Use Committee. BALB/cAnNCrl mice for the generation of hybridomas were obtained from Charles River Laboratories. Male hemophilia A mice weighing 30–35 g on average were obtained from a Pfizer proprietary line maintained at Charles River laboratories. Male hemophilia B mice weighing 30–35 g were from a Pfizer proprietary line maintained at Taconic Biosciences. Male drug-naïve New Zealand white rabbits for the in-life PK/PD study in were obtained from Charles River Laboratories Inc. Phage display and ELISA screening scFv antibody fragments that bound to recombinant human and mouse TFPI-K1-K2 were identified following four rounds of phage display using the proprietary WyHN5 phage display library, composed of scFv antibody fragments derived from nonimmunized human donors. Selection strategies incorporated multiple branches for identification of anti-TFPI antibodies that have desirable properties, such as thermostability and cross binding to human and mouse TFPI species. Phage selections were performed in solution using streptavidin beads with various concentrations of biotinylated (site-specific: AviTag) human TFPI-K1-K2 or mouse TFPI-K1-K2. The streptavidin-bound phages were eluted and then expanded by infecting Escherichia coli strain ER2738 to generate phage particles for successive rounds of panning. Crude scFv-containing periplasmic material (peripreps) for use in ELISA assay was used to identify TFPI scFv binders using a binding ELISA and neutralizers of the TFPI-FXa interaction using a competition ELISA. TFPI binding scFvs were sequenced to identify unique clones. The VH and VL domains from 48 antibodies were cloned into mammalian expression vectors and transfected into HEK293 cells to produce small quantities of IgG. Functional screening was performed, measuring the effect of clones on TFPI inhibition of TF/FVIIa-mediated FX activation and thrombin generation. Immunization and ELISA screening A cohort of five BALB/c mice were each immunized subcutaneously with a mixture of human and murine TFPI-K1-K2 proteins for the generation of hybridomas. Mice with positive serum titers for anti-TFPI antibodies were sacrificed, and the draining lymph nodes (axillary, inguinal, and popliteal) were harvested. Fusion of pooled lymph node cells and P3X63.Ag8.653 cells (a B lymphoblast cell line) used electro-cell fusion. Screening of the resulting hybridoma cell supernatants for binding to human TFPI-K1-K2 was performed by ELISA. Parental hybridomas were subcloned by limiting dilution to obtain monoclonal hybridoma cell lines. Positive subclones were identified by ELISA screening for reactivity with human TFPI-K1-K2 and expanded. Purified antibodies from one subclone of each hybridoma were characterized further by functional screening, measuring the effect of clones on TFPI inhibition of TF/FVIIa-mediated FX activation and thrombin generation. RNA from selected hybridomas was prepared and the variable region DNA sequences from the expressed antibodies were obtained by reverse transcriptase (RT)–polymerase chain reaction (PCR) cloning. Variable region cDNAs derived from selected hybridomas were subcloned into mammalian expression vectors to generate mouse-human chimeric antibodies that were then tested for binding to both human and rabbit TFPI-K1-K2 proteins by protein binding ELISA. The human-mouse chimera sequence was humanized by CDR grafting onto human acceptor framework sequences and were also tested for binding to human and rabbit TFPI binding in an ELISA binding assay and human TFPI binding in a SPR binding assay. Functional screening assays TFPI inhibition was assessed in three primary in vitro functional assays: FXa and TF/FVIIa/FXa inhibition reversal assays and the TGA.9 The assays are outlined below, and further details are provided in the Supplementary Methods. The results of all anti-TFPI antibody candidates were compared with a control (human or murine IgG1) in these analyses. FXa inhibition reversal assay The ability of purified anti-TFPI antibodies to restore FXa activity in the presence of inhibitory concentrations of TFPI was assessed in vitro. Anti-TFPI antibodies diluted in phosphate-buffered saline (PBS) at concentrations ranging from 1–500 nM were preincubated with 10 nM of recombinant human TFPI-K1-K2 or 10 nM of recombinant rabbit TFPI-K1-K2 proteins in activity buffer (20 mM HEPES, pH 8.0, 150 mM NaCl, 5 mM CaCl2, 0.5 mg/mL bovine serum albumin (BSA) for 30 min at 37°C in wells of a 96-well microplate. Human plasma-derived FXa (2 nM; Haematologic Technologies, catalog no. HCXA-0060) was introduced, and the reactions were incubated at 37°C for 30 min. The chromogenic substrate Spectrozyme FXa (Sekisui Diagnostics LLC, catalog no. 222) was added to reach a final concentration of 500 µM to each well for a final reaction volume of 100 µL. Control reactions included reactions without FXa to control for assay background, without TFPI, allowing for maximal generation of FXa (100% activity) or without anti-TFPI antibody (PBS alone). Absorbances of the reactions were immediately read on a SpectraMax M5e multimode plate reader (Molecular Devices LLC) at 405 nm at 2-min intervals over a period of 60 min. Raw data were exported and analyzed in Microsoft Excel (Version 2020). The percentage activity relative to the no TFPI control was determined for each reaction and plotted versus concentration. Two-stage TF-FVIIa-FX inhibition assay The ability of purified anti-TFPI antibodies to restore FVIIa-TF activity in the presence of inhibitory concentrations of TFPI was assessed in vitro. Anti-TFPI antibodies at concentrations ranging from 1–500 nM were preincubated with 10 nM of recombinant TFPI proteins in activity buffer (20 mM HEPES, pH 8.0, 150 mM NaCl, 5 mM CaCl2, 0.5 mg/mL BSA) for 30 min at 37°C in wells of a 96-well microplate. Lipidated tissue factor, 1 pM (Innovin, catalog no. 23–044-776) and 1 nM recombinant FVIIa (NovoSeven, Novo Nordisk) were added to the reactions and incubated at 37°C for 5 min. Human FX, 150 nM, (Haematologic Technologies, Inc. catalog no. HCX-0050) was introduced into the reactions. The chromogenic substrate Spectrozyme FXa (Sekisui Diagnostics LLC, catalog no. 222) was added to a final concentration of 500 µM to each well for a final reaction volume of 100 µL. Control reactions included reactions without FVIIa, without TF, without FX, without TFPI or without anti-TFPI antibody (PBS alone). Absorbances were immediately read on a SpectraMax M5e multi-mode plate reader (Molecular Devices, LLC) at 405 nm at 2-min intervals over a period of 60 min. Raw data were exported and analyzed in Microsoft Excel (Version 2010). The percentage activity for each reaction was calculated and plotted versus concentration. Thrombin generation assay The ability of purified anti-TFPI antibodies to restore thrombin generation in human hemophilic plasma (with attenuated FVIIIa activity) was assessed in TGA, with TGA reagents Calibrated Automated Thrombogram including the Fluoroskan Ascent fluorescent plate reader (Thrombinoscope BV), and Thrombinoscope software (version 5; Diagnostica Stago) as described previously.9 The thrombin generation reagents, PPP-Reagent LOW (a mixture of 4 µM phospholipids and 1 pM tissue factor), thrombin calibrator, FluCa-kit (comprised Fluo-Buffer and Fluo-Substrate containing the fluorogenic substrate solubilized in dimethyl sulfoxide), were obtained from Diagnostica Stago. Anti-TFPI antibodies were tested at concentrations ranging from 1–500 nM in the assay. The thrombin generation velocity index parameter for each antibody concentration was calculated as the ratio of the peak thrombin concentration to the quantity (time to peak – lag time) and plotted versus concentration. Cloning, purification, and expression of anti-TFPI antibodies RNA from selected hybridomas was prepared and the variable region DNA sequences from the expressed antibodies were obtained by RT–PCR cloning. RNA derived from 1,000,000 hybridoma cells was prepared using a Qiagen RNAeasy Mini kit. The first strand cDNA was generated using a SuperScriptII RT kit (Invitrogen, catalog no. 18064–022). Double-stranded cDNAs encoding the variable regions of each hybridoma were generated and amplified by PCR using primers specific for the mouse IgG1 heavy chain and kappa light chain constant regions. After confirming by gel electrophoresis that cDNA bands of the correct size had been generated, the PCR products were cloned into the TOPO-TA cloning vector (Invitrogen, catalog no. K459501). The clones were then sequenced by conventional methods. VH and VL fragments with unique sequences from antibodies that demonstrated binding to TFPI and inhibition in a TFPI/FXa competition ELISA were selected for subcloning into human IgG-3 M cloning vectors. The VH and variable kappa (VK)/VL paired expression vectors corresponding to the scFv were then used for transient mammalian expression in HEK293 cells by standard methods. Binding affinity: SPR SPR experiments were carried out to assess the affinity of the lead anti-TFPI antibody candidates for TFPI-K1-K2 proteins across species and TFPI2. An anti-human Fc sensor chip was prepared by amine coupling anti-human IgG antibody (GE Healthcare, catalog no. BR-1008-39) to all four flow cells of a carboxymethylated dextran coated sensor chip (CM5) (GE Healthcare, catalog no. BR-1005-30) according to the manufacturer’s protocols. The flow cells were activated by injecting a 1:1 mixture of 400 mM 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride and N-hydroxysuccinimide for 7 min at a flow rate of 10 µL/min. Anti-human IgG antibody was diluted to 25 µg/mL in 10 mM sodium acetate pH 5.0 and injected over all flow cells for 7 min at 10 µL/min. All flow cells were blocked with 1 M ethanolamine-HCL (ETH) for 7 min at 10 µL/min. Final immobilization level of the capture antibody was approximately 10,000 resonance units (RU). The running buffer for immobilization and kinetics was 10 mM HEPES pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.05% (v/v) Tween-20 (HBS-EP+). To characterize the binding of anti-TFPI antibodies to human TFPI, the antibodies were diluted to 0.5 µg/mL in HBS-EP+ and captured by the anti-human IgG immobilized on flow cells 2, 3, and 4 for 30 s to 1 min at a flow rate of 5 μL/min to achieve a capture level of 70–300 RU. Flow cell 1 was used as a reference surface. After antibody capture, the flow rate was increased to 50 μL/min and buffer or human TFPI ranging in concentration from 0.2–200 nM in HBS-EP+ was injected over all flow cells for a 1.0-min association and allowed to dissociate for 10–15 min. Buffer cycles collected for each captured antibody were used for double referencing.39 At the end of each cycle, the entire anti-IgG surface was regenerated by a 60-s pulse of 3 M MgCl2. Kinetic assays were conducted at 25°C at a collection rate of 10 Hz on a BIAcore T200 instrument (GE Healthcare). Rate constants and affinities were determined by fitting the data to a 1:1 model in BIAcore T200 Evaluation software version 1.0 (GE Healthcare). Binding affinity: Octet Binding of rationally designed variants of lead anti-TFPI antibodies to human TFPI K1-K2 was assessed using an OctetRED384 biolayer interferometry instrument (Sartorius). Antibodies at 10 µg/mL were captured on an anti-human Fc AHC biosensor tip (Sartorius Cat. 18-5060), and subsequent binding of 100, 33, and 1111 nM human TFPI K1-K2 in Kinetics PBS buffer (Sartorius Cat. 18-1105) was assessed, with an association time of 300 s and a dissociation time of 900 s. X-ray crystallography X-ray crystallography was used to identify the binding epitopes in the complex structure of anti-TFPI antibodies bound to the cynomolgus monkey TFPI-K2 domain. The generation of cynomolgus TFPI K2 domain and anti-TFPI antibody Fab/cynomolgus TFPI K2 complexes for structural studies is described in the Supplementary Methods. The K2 domain of cynomolgus monkey TFPI-K1-K2 was isolated with human neutrophil elastase40 and the K2 domain separated from the K1 domain by anion exchange chromatography. Cynomolgus monkey TFPI-K1-K2 was chosen over human TFPI-K1-K2 for structural studies because the monkey protein was expressed at a higher level and had lower occupancy of the single N-linked carbohydrate site on the K2 domain, allowing more efficient separation of the K2 domain on anion exchange chromatography post cleavage with human neutrophil elastase. There are only two residue differences between monkey and human in the K2 domain, and both residues are at the opposite end of the domain from the region of the K2 domain that binds to FXa. Human neutrophil elastase cleaves TFPI specifically between the K1 and K2 domains in human and monkey TFPI. TFPI antibody Fab was produced by cleaving the individual anti-TFPI antibody with papain and then separating the Fab from the Fc portion of the antibody by protein A chromatography. A 1:1 complex of the anti-TFPI antibody Fab and cynomolgus monkey TFPI-K2 domain was purified by size exclusion chromatography and the resulting complex was crystallized and the structure was solved using X-ray diffraction data. Epitope binning/epitope mapping of anti-TFPI antibodies by SPR Sandwich SPR assay was used to map the epitopes of the generated anti-TFPI antibodies (Figure S2). Epitope binning of anti-TFPI antibodies was conducted using an SPR assay (Biacore-T200, GE Healthcare). In this assay, the first antibody was captured on the surface of a CM5 chip and recombinant human TFPI was bound by the first antibody (see section ‘Binding affinity: SPR’ for details). Subsequently, a second antibody was flowed over the chip, and the subsequent change in binding signal was measured. If the paired antibodies had different epitopes, binding of the second antibody would be observed following binding of TFPI. If the paired antibodies shared overlapping epitopes, no binding signal would be observed for the second antibody. Hemophilia mouse acute tail clip injury model A mouse acute tail clip injury model of hemostasis using factor VIII-deficient mice (hemophilia A) or FIX-deficient mice (hemophilia B) was used to test the efficacy of lead anti-TFPI antibody candidates. Mice were injected with one of the anti-TFPI antibody candidates (6 mg/kg) or control (saline) via the tail vein. The procedure and methods for determining blood volume lost have been described previously9 and bleeding time was also determined from the time bleeding stopped post tail transection. Blood loss was analyzed using a linear model to assess the profile across time. The model included fixed effects for categorical time points and the general form of the model was: where Y represents blood loss for time i and mouse j, µ is the overall mean, τ is the ith time point, and ϵ is the error associated with the jth mouse at time i, assumed normally distributed with constant variance. Saline is the vehicle control and is included in the model as a time effect treatment (τ0). An analysis of variance (ANOVA) was performed to test for overall differences among time points. Pairwise comparisons were conducted using Dunnett’s test to compare mean blood loss at each time point to the saline vehicle control. Adjusted p-values and significance determined by p < 0.05 level were reported for each comparison. PK/PD of anti-TFPI antibodies in rabbits To obtain in vivo PK/PD, anti-TFPI antibody candidates and a control human IgG were intravenously administered to New Zealand white rabbits. Twelve male drug-naïve New Zealand white rabbits weighing 3.1–3.5 kg were intravenously administered 2 mg/kg of anti-TFPI antibody (dosing vehicle: PBS, 1 mL/kg). The whole blood samples were collected into tubes containing 3.2% sodium citrate at room temperature. Plasma was obtained by centrifugation of the blood at 2500 g at room temperature for 10 min and stored at −70°C. The collection time points were 5 min and 1, 6, 24, 48, 96, 120, 144, 192, and 240 h after administration of test agents. Anti-TFPI antibody concentration measurement A total human IgG quantitative ligand-binding assay was used to quantify plasma concentration of different anti-TFPI candidate antibodies. In the assay, donkey anti-human IgG (H+L) biotin was captured onto the streptavidin-coated beads on the affinity capture column of the Gyrolab Bioaffy microstructure (Gyros Protein Technologies). Anti-TFPI was captured onto the affinity capture column with biotin-conjugated donkey anti-human IgG (H+L) (Jackson ImmunoResearch, Inc. catalog no. 709-065-149). Bound anti-TFPI was detected with Alexa 647 labeled donkey anti-human IgG (H+L) (Jackson ImmunoResearch, Inc. catalog no. 709-605-149). The fluorescent signal on the column allowed for detection of the bound anti-TFPI construct. Response units were read by the Gyrolab instrument at 1% Photomultiplier tube setting. Sample concentrations were determined by interpolation from a standard curve that is fit using a five-parameter logistic curve fit with 1/y 2 response weighting. The standard points in assay buffer containing 5% pooled sodium citrated rabbit plasma range from 0.78–891 ng/mL, and the range of quantitation in 100% rabbit plasma matrix is 90–5500 ng/mL. Anti-TFPI samples at 0.45, 8.0, 47.15, 153, and 275 ng/mL in assay buffer containing 5% pooled sodium citrated rabbit plasma served as quality control measurement. Anti-TFPI PK/PD data analysis and modeling Noncompartmental analysis using the NonCompart package in R was used to analyze and compare the observed PK of anti-TFPI antibody candidates. A two-compartment model with MM elimination (2CT-MM) was developed to describe the antibody concentration data. In this model, the anti-TFPI antibody candidates were assumed to be metabolized in the plasma through a target-mediated mechanism with Vmax-Km, and the antibody was also assumed to be able to distribute to the other tissues (represented by the peripheral compartment). Model equations are described below: Cp and Cper represent the drug concentration in central and peripheral compartments. V1, V2, and CL2 represent the central and peripheral compartment volumes and intercompartment distributional clearance, respectively. The nonlinear clearance in the central compartment was described by MM kinetics using the maximum elimination rate (Vmax) and the concentration at half-saturable elimination as MM constant (Km) parameters. All model parameters were estimated by fitting the observed PK of anti-TFPI antibodies using a nonlinear mixed effects modeling approach using the nlmixr2 package in R. Model selection/acceptance was based on goodness of fits, visual assessment of residual plots, and coefficient of variation. Anti-TFPI target RO and inhibition effects (Inh EFF) were modeled as direct-effect instantaneous functions of the central compartment drug concentration using the equation: Kd and IC50 represent the in vitro measured TFPI binding affinity and dPT assay potency in rabbit plasma for the relevant anti-TFPI antibody candidates. Once PK model parameters were identified for all the tested antibody candidates, the model was used to simulate dose- and time-dependent TFPI RO% and Inh EFF% in rabbits. For SC dosing simulations, identical typical SC absorption parameters were assumed for all anti-TFPI antibody candidates. Model simulations were conducted using the mrgsolve package in R. Molecular assessment/biophysical characterization Capillary differential scanning calorimetry Antibody thermal stability was analyzed using Malvern’s capillary VP-DSC equipped with an autosampler (Malvern Panalytical). The antibody sample prepared at 0.02 mM in PBS or histidine-sucrose buffer was placed in the sample cell with the respective PBS or histidine-sucrose buffer in the reference cell and heated from 10°C to 100°C at a scan rate of 100°C per h. The heat capacity difference between the sample cell and reference cell was recorded and analyzed using Origin 7.0 software (OriginLab Corporation). A buffer-only baseline thermogram was generated, and the data were used to subtract any system heat not associated with protein denaturation. Forced aggregation analysis Details of the assay for forced aggregation analysis are provided in the Supplementary Methods. DLS-based viscosity analysis Antibody samples formulated in the histidine-sucrose buffer were prepared at 0–200 mg/mL. Viscosity analysis was performed on a 1536-well plate (Greiner Bio-One) with 8 μl of antibody sample and 0.5 μl of a diluted nanosphere tracer bead mixture added to each well. The 300-nm diameter nanosphere bead (Thermo Fisher Scientific) was diluted 10-fold v/v into the histidine-sucrose buffer prior to use. After a brief centrifugation to remove air bubbles, the sample plate was placed in a DynaPro Plate Reader (Wyatt Technology) for DLS measurement. Nanosphere size measurements were analyzed using the Wyatt Dynamics software v7 as previously described.41 Antibody stability assessment Antibody stability was assessed between 50 and 150 mg/mL in histidine-sucrose buffer up to 8 weeks at 5°C, 25°C, and 40°C. Samples were buffer exchanged into histidine-sucrose buffer using 10K molecular weight cutoff dialysis cassettes (Thermo Fisher Scientific). The dialyzed protein solutions were concentrated to the final target protein concentration with Amicon 50 kDa MWCO UltraCentrifugal Filter (Millipore) and protein concentrations were determined using a Nanodrop2000 Spectrophotometer (Thermo Fisher Scientific). At each time point, % high molecular mass species and % low molecular mass species of the incubated samples were analyzed using Agilent 1200 HPLC system with YMC-Pack Diol-200, 300 × 8 mm ID (YMC America) analytical size exclusion chromatography column with PBS supplemented to 400 mM NaCl as mobile phase buffer. Anti-TFPI candidates demonstrating the preferred activity and manufacturability profiles were selected for scaling and characterization of in vivo hemostatic activity. Recombinant CHO cell line development and expression analysis Genes encoding lead anti-TFPI antibody candidates were cloned into the expression vector pRY19-GA-Q designed for stable integration using recombinase-mediated cassette exchange into a single site in CHO-SV cells.42 Cells were transfected using electroporation in three parallel cultures utilizing positive and negative selection pressure with hygromycin-B and ganciclovir. CHO pools went through a 3-week recovery phase and were banked when 95% viability was achieved. Stable pools showed similar production in 12-day fed-batch and were combined for further expansion. Scale-up was performed with 1.5 L in 5-L Thomson shake flasks (Thomson) for 12-day fed-batch platform expression process in CD-CHO media (Thermo Fisher Scientific, catalog no. 10743029). The conditioned medium was harvested and clarified using centrifugation at 3000 × g for 10 min followed by a 0.2‑µm PES bottle-top filter (Corning). Antibody was purified by chromatography with Protein A (MabSelect SuRe, Cytiva). Immunogenicity assessment Presentation of potential T-cell epitopes by dendritic cells was assessed using a DC-T assay and dendritic cell mass spectrometry conducted by ProImmune (Oxford, UK). In vitro hemostatic activity characterization: dPT assay The ability of the anti-TFPI antibodies to inhibit endogenous TFPI in normal rabbit plasma or human FVIII deficient plasma (hemophilia A) was studied using the dPT assay, which has been described previously.43,44 Abbreviations Two-compartment model with Michaelis-Menten elimination Bovine serum albumin Complementarity-determining region Chinese hamster ovary Clearance Dendritic cell–T-cell Dynamic light scattering Dilute prothrombin time Differential scanning calorimetry Enzyme-linked immunosorbent assay Factor IX Factor VIIa Factor VIII Factor X Factor Xa Human embryonic kidney cells Immunoglobulin G Inhibition effects Intravenous Equilibrium dissociation constant MM constant Monoclonal antibody Major histocompatibility complex Michaelis-Menten No binding Not determined Phosphate-buffered saline Pharmacodynamic Pharmacokinetic Receptor occupancy Reverse transcriptase – polymerase chain reaction Subcutaneous Single-chain variable fragment Surface plasmon resonance Tissue factor pathway inhibitor Thrombin generation assay Variable heavy Variable light Supplementary Material MARS early development_SA_22April2026_clean_tc03_27May2026.docx Acknowledgments We thank Chuenlei Parng for early work in the development of the marstacimab antibody and for PK modeling; Sophie Tourdot for immunogenicity work; Susan Benard and Kerry Kelleher for SPR data; Christopher Francis for CHO cell line development; Mark Krebs for biophysical analysis; May Tam, Edward Lavallie, Liz DiBlasio, Chris Corcoran, and Karl Malakian for reagent generation; Beth Leary, Katherine Wright, and Mania Kavosi for PK support; Joel Bard for structure solution and refinement; Kristine Svenson, Chao Tu, and Ching-Shin Huang for crystallography support; Jane Guo, Xiaotian Zhong, Matthew McKenna, Qingcong Lin, Peter LaPan, and Caryl Meade for antibody production; Nicole Piché-Nicholas, Simran Singh, and Oleg Kovalenko for data verification; Greg Steeno and Yao Zhang for statistical review and figure editing; and Will Somers for project advocacy. Medical writing and editorial support were provided by Marion James, PhD, of Engage Scientific Solutions and was funded by Pfizer. Funding Statement This study was sponsored by Pfizer. Disclosure statement MJ, SH, RJ, SR, JN, AT, JRA, SJ, LM, MH, AJ, SH, RW, LL, MS, DP, and LB are/were employees and stockholders of Pfizer at the time of study conduct. Data availability statement Upon request, and subject to review, Pfizer will provide the data that support the findings of this study. Subject to certain criteria, conditions, and exceptions, Pfizer may also provide access to the related individual deidentified participant data. See https://www.pfizer.com/science/clinical-trials/trial-data-and-results for more information. Supplementary Information Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2026.2685362 References Associated Data Supplementary Materials MARS early development_SA_22April2026_clean_tc03_27May2026.docx Data Availability Statement Upon request, and subject to review, Pfizer will provide the data that support the findings of this study. Subject to certain criteria, conditions, and exceptions, Pfizer may also provide access to the related individual deidentified participant data. See https://www.pfizer.com/science/clinical-trials/trial-data-and-results for more information.