Reagents.

The fully human mAb PF-03446962 (IgG2, κ) was generated from human IgG2 transgenic XenoMouse (Abgenix, Fremont, CA) immunized using human ALK1 extracellular domain immunogen (Kellermann and Green, 2002). The nongermline framework mutations were fixed to be consistent with the germline sequence and the resulting antibody. Human ALK1/Fc chimera (lot 002K1721), goat anti-human κ conjugated to horseradish peroxidase (lot 032K9157), tetramethylbenzidine (TMB), and TMB stop solution were obtained from Sigma-Aldrich (St. Louis, MO).

Pharmacokinetics in Cynomolgus Monkeys.

Animal experiments were carried out in accordance with Pfizer Global Research and Development Institutional Animal Care and Use Committee guidelines and with the Institute of Laboratory Animal Resources (1996). Animals were housed in compatible pairs in stainless steel cages. They were separated early in the morning for individual feeding, dosing, assessment of clinical signs, and other procedures. They remained separated throughout the day and then were paired again in the afternoon.

In the single-dose PK study, cynomolgus monkeys (3–6 kg) received a single intravenous bolus dose of PF-03446962 at 5 or 50 mg/kg. PF-03446962 formulation was prepared in a vehicle consisting of 20 mM sodium acetate, 140 mM sodium chloride, and 0.2 mg/ml polysorbate-80, pH 5.5. In the multiple-dose PK study, cynomolgus monkeys (2–8 kg) received 5 weekly intravenous doses of PF-03446962 at 2, 10, or 50 mg/kg followed by a 2-month washout period. PF-03446962 formulation was prepared in a vehicle consisting of 20 mM histidine, pH ∼5.5, 84 mg/ml trehalose dehydrate, 0.2 mg/ml polysorbate-80, 0.05 mg/ml disodium EDTA, and 0.1 mg/ml l-methionine. Single-dose PK study was conducted during the discovery stage while the multidose study was conducted as a part of the investigational new drug-enabling studies. At the time, the formulation used for the single-dose study was standard and was used for the majority of early-stage experiments. However, once the mAb reached a stage of “clinical candidate,” a more extensive formulation work was conducted to select the clinical formulation. Therefore, investigational new drug-enabling studies were conducted in the final formulation. It was assumed that formulation would not alter the PK of an intravenously administered mAb. Blood samples (∼1 ml) were collected via femoral venipuncture from all animals at predetermined time points. The serum was separated and stored frozen at approximately −20°C until samples were analyzed. In the single-dose study, only one animal was antidrug antibodies (ADA)-positive. In the multiple-dose study, two animals were ADA-positive. Only animals (n = 14, combination of the single- and multiple-dose studies) that were negative for ADA were selected for modeling to avoid confounding issues associated with the clearance of PF-03446962.

Clinical Pharmacokinetics.

Preliminary PK data from an ongoing phase I study of PF-03446962 were used to validate the PK prediction based on the model. The PK data were obtained from a phase 1, multicenter, multinational trial in patients with solid tumors. The available data from the 0.5, 1, 2, 3, and 4.5 mg/kg dose cohorts (three patients per cohort) were used to verify human PK prediction. Noncompartmental analyses (NCA) were performed to compare the projected versus observed PK parameters.

Analytical Method.

Study serum samples were analyzed using a “sandwich” enzyme-linked immunosorbent assay (ELISA). Plates were coated with human ALK1/Fc chimera (50 μl/well at 2.5 μg/ml) overnight at 4°C followed by washing with PBS buffer (containing 0.05% Tween 20) and incubation with the blocking buffer (100 μl per well) for 1 h at room temperature. After the incubation, plates were washed and incubated with 50 μl per well of study samples that were diluted in the blocking buffer, standards, or quality control (QC) samples for 1 h at room temperature. After the incubation, plates were washed and incubated with 50 μl per well of secondary antibody (diluted 2000 times in PBS) for 1 h at room temperature. The plates were then washed followed by the addition of TMB (50 μl per well). The reaction was terminated by an addition of 50 μl per well of TMB stop solution at 2 to 5 min. The resulting absorbance was measured at 450 nm using a SpectraMax M2 plate reader (Molecular Devices, Sunnyvale, CA). A calibration curve was constructed by the SoftMax Pro version 5 software (Molecular Devices) using a four-parameter logistic curve. Concentrations of study and QC samples were determined by reverse prediction from their absorbance against the calibration curve. The lower limit of quantitation was 146 ng/ml. The percentage coefficient of variation for the high, low, and medium QC ranged from 3.7 to 15.8%, whereas %RE ranged from −0.8 to 2.5%.

Surface Plasmon Resonance Studies.

Surface plasmon resonance (SPR)-based binding studies were carried out using a Biacore 3000 instrument (GE Healthcare, Chalfont St. Giles, Buckinghamshire, UK) at 25°C in 150 mM NaCl, 25 mM HEPES, pH 7.5, 2 mM MgCl₂, and 0.005% P20 surfactant. PF-03446962 fragment antigen-binding(Fab) fragment was immobilized on a CM5 chip by standard 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide hydrochloride/N-hydrosuccinimide amine coupling chemistry at 25°C using 1 μM in 10 mM sodium acetate, pH 5.0. B and C anti-human ALK1 Fab was injected at concentrations of 97, 195, 390, and 781 pM and of 1.56, 3.125, 6.25, 12.5, 25, 50, 100, and 200 nM. A human ALK1 extracellular domain (ECD) (180 μl) was injected at concentrations of 976 pM and 1.95, 3.9, 7.81, 15.6, 31.25, 62.5, 125, 250, and 500 nM using a Kinject at a flow rate of 50 μl per min. The association and dissociation times were measured for 3.6 and 23 min, respectively. Regeneration was performed using a 1 min pulse of 10 mM H₃PO₄. Data analysis was performed using the Scrubber2 software (BioLogic Software, Pty., Campbell, ACT, Australia). Injections were referenced to a blank surface and by a buffer blank. Kinetic characteristics were obtained from a fit to a simple 1:1 kinetic binding model using the Scrubber2 program software (BioLogic Software, Pty.).

SPR biosensor studies were carried out to determine the direct binding properties of PF-03446962 to recombinant human and cyno-ALK1 ECD. To obtain meaningful kinetic parameters the analyte molecule in solution, we ensured that the monomeric and its concentration were accurately known. The human and cyno-ALK1 ECD were immobilized on the sensorChip surface, and the binding of the monomeric Fab form of PF-03446962 was observed. This experimental format provided the best means to compare the binding of PF-03446962 to the human and cyno-ALK1 ECD forms given that the association rate depends only on the concentration determined for the analyte molecule present in solution.

Fluorescence-Activated Cell Sorting Analysis of PF-03446962-ALK1 Complex Internalization in Human Umbilical Vein Endothelial Cell and Monkey Endothelial Cells.

Human umbilical vein endothelial cells (HUVEC) and monkey endothelial cells (smUVEC-cyno) were seeded on a 24-well plate with 300 μl of Complete HUVEC growth medium (EGM-2; Lonza Walkersville, Inc., Walkersville, MD) in each well and grown at 37°C with 5% CO₂ overnight. Three microliters of PF-03446962 (1 mg/ml stock) were added to the cells and incubated at 37°C with 5% CO₂ for 0, 0.5, 2, 6, or 24 h to for allow the internalization activity to take place. At the end of the time points, samples were placed on ice for 1 h. The cells were then washed with PBS and detached with 0.025% trypsin-EDTA. Cells were washed twice with ice-cold PBS-SA (0.04% sodium azide, NaN3) by centrifugation and blocked with PBS-SA containing 2% normal goat serum on ice for 15 min. The cells were then incubated with R-phycoerythrin-conjugated goat anti-human κ secondary antibody (Southern Biotechnology Associates, Birmingham, AL) for 45 min in the dark. Cells were then washed three times, resuspended in 200 μl of PBS-SA, and analyzed by GUAVA Flow Cytometer CytoSoft ExpressPro (GUAVA Technologies, Inc., Hayward, CA) using yellow fluorescent channel. PF-03446962 bound to the cell surface receptors was determined as the fluorescence intensity. The experiment was repeated to obtain replicates from which the averages of the replicates were calculated and used to determine the rate of antibody-complex internalization. The internalization rate (k_int) was assumed to be the initial rate of mAb-target complex disappearance and was calculated by taking the regressed slope of percent receptor remaining values of the first three time points.

PK Modeling.

TMDD model (Mager and Jusko, 2001; Mager et al., 2003) was used to simultaneously fit the monkey single- and multiple-dose serum PK data of PF-03446962. The model diagram is shown in Fig. 1. PF-03446962 was assumed to undergo first-order elimination (k_el) in the plasma compartment and distribute into the peripheral compartment as characterized by k₁₂ and k₂₁. The antibody was assumed to interact with the ALK1 receptor via reversible binding constants k_on and k_off. Because the ALK1 receptor is predominantly expressed on endothelial cells, drug-target binding was designated to take place in the central instead of the peripheral compartment. As verified by in vitro experiments, binding to target receptor leads to internalization of the antibody-receptor complex. The internalization rate of the drug-receptor complex was assumed to be first-order and was represented by k_int. The turnover kinetics of the ALK1 receptor was represented by k_syn, rate of synthesis, and first-order k_deg, rate of degradation.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Model diagram of the TMDD model used to describe the in vivo PK of PF-03446962 in monkey and to predict its human PK.

The PK model was described by the following set of differential equations: where A_p and A_t represent the drug amount in the serum and peripheral/tissue compartment, respectively; R represents the concentration of the unbound ALK1 receptor; and Cplx represents the concentration of drug-ALK1 receptor complex. V_c is the volume of distribution in the central (plasma) compartment. The receptor binding kinetics is governed by the association (k_on) and dissociation (k_off) rate constants. The parameters k_syn (zero-order) and k_deg (first-order) describe the intrinsic turnover kinetics of the ALK1 receptor.

In the model fitting, k_on and k_off were fixed to values experimentally obtained from the SPR study (as described above); k_int was fixed to the value obtained from the FACS experiment (as described above); and k_deg was fixed based on half-life, as estimated from the human ALK1 protein expression data from Shao et al. (2009). Because the half-life of monkey ALK1 was not available in the literature, it was assumed that it was the same as that of human. Under the initial condition, before drug administration, the free ALK1 receptor concentration, R₀, was assumed constant. On the basis of eq. 3, at the initial condition, R₀ = k_syn/k_deg. Because k_deg was known, the equation can be rearranged to k_syn = R₀k_deg. R₀ was estimated as a parameter in the model. The total initial abundance (R_T) may be derived as R_T = R₀ + R_bound, where R_bound is the initial target-ligand complex. During model fitting, we assumed the initial R_bound to be insignificant, thus, setting it to zero at the initial condition. This inherently assumes that any initial endogenous occupation of the receptor was negligible.

Initial NCA of the monkey PK data using WinNonlin (Pharsight, Mountain View, CA) (data not shown) indicated that there was a discrepancy between the clearances of PF-03446962 when comparing single-dose data with multiple-dose data. This may have resulted from the studies being conducted at different times and with different sets of monkeys. Thus, a between-study covariate term was added to the k_el parameter (Table 1). Model fitting and simulations were performed with NONMEM VII (ICON Development Solutions, Ellicott City, MD), and the first-order approximation method was used. Although the FOCE-INTERACTION method may have been preferable, attempts to use this method were not successful. The between-subject variability, assumed to be log-normally distributed, was estimated for volume of distribution (V_c) and k_el. A proportional error model was used for the estimation of the residual variability. The parameter standard errors were directly obtained from the NONMEM output.

Human Pharmacokinetic Prediction.

To simulate and predict the human PK profiles of PF-03446962 from the model, physiological PK parameters (k_el, k₁₂, and k₂₁) were allometrically scaled based on the power coefficient equation P = aBW^b, where P is the human predicted parameter, a is the monkey parameter, BW is the human body weight (70 kg), and b is the power coefficient that was assumed to be 0.25 (Mahmood, 2009). However, no scaling was performed for the mechanistic parameters (k_on, k_off, k_int, and k_deg). Instead, human values of k_on, k_off, and k_int (such as those of monkeys) were experimentally measured as described in the previous section. The parameter k_deg was separately estimated from the human ALK1 protein expression data of Shao et al. (2009)). The human V_c was the assumed plasma volume (2.8 liters in a 70-kg human body weight) based on typical volume of IgGs (Mahmood, 2009). With the consideration that the ALK1 receptor was predominantly expressed on endothelial cells, the receptor abundance (R₀), expressed in units of nanomoles (amount in molar unit normalized to liter of plasma), was assumed the same between monkey and human. This essentially assumed that total target amount of targets per plasma volume is the same between monkey and human.

On the basis of the set of human PK parameters (derived as mentioned above), population PK profiles of 100 individuals were simulated. In the absence of prior information on the variability, we arbitrarily chose 20% BSV for some model parameters. For simplicity, all subjects were assumed to have a body weight of 70 kg. Because data were available from only a limited number of subjects, we did not consider an alternative method, which was to simulate based on the distribution of the body weights of the studied subjects. For graphical comparison, the simulated individual PK profiles were overlaid with the observed clinical data in patients in an ongoing clinical trial. To compare the PK simulation against observed PK data in patients (WinNonlin 5.2; Pharsight) was used to determine the noncompartmental parameters of the mean predicted versus observed clinical data. In addition, predicted population PK profiles were graphically overlaid with observed PK profiles for comparison.