Pharmacokinetics and pharmacodynamics of the cytolytic anti‐CD38 human monoclonal antibody TAK‐079 in monkey – model assisted preparation for the first in human trial

Abstract We are studying the fully human, IgG1λ cytolytic monoclonal antibody TAK‐079, which binds CD38. CD38 is expressed on plasma and natural killer (NK) cells constitutively and upregulated on subsets of B and T lymphocytes upon activation. TAK‐079 cross‐reacts with CD38 expressed by cynomolgus monkeys and depletes subsets of NK, B, and T cells. Therefore, safety and function of TAK‐079 was evaluated in this species, prior to clinical development, using bioanalytical, and flow cytometry assays. We pooled the data from eight studies in healthy monkeys (dose range 0.03‐100 mg/kg) and developed mathematical models that describe the pharmacokinetics and the exposure–effect relationship for NK cells, B cells, and T cells. NK cell depletion was identified as the most sensitive pharmacodynamic effect of TAK‐079. It was adequately described with a turnover model (C 50 = 27.5 μg/mL on depletion rate) and complete depletion was achieved with an IV dose of 0.3 mg/kg. Intermediate effects on T‐cell counts were described with a direct response model (C 50 = 11.9 μg/mL) and on B‐cell counts with a 4‐transit‐compartment model (C 50 = 19.8 μg/mL on depletion rate). Our analyses substantiate the observation that NK, B and T cells are cleared by TAK‐079 at different rates and required different time spans to replete the blood compartment. The models were used to simulate pharmacokinetic and cell depletion profiles in humans after applying a straightforward scaling approach for monoclonal antibodies in preparation for the first‐in‐human clinical trial.


| INTRODUCTION
TAK-079 is a fully human IgG1k monoclonal antibody with high affinity for CD38 that is developed for the treatment of multiple myeloma and autoimmune diseases. 2,3,5,14,21,22 The target molecule CD38 is a type II-transmembrane protein which functions as a cell adhesion receptor and a multifunctional ectoenzyme. 8,18,20 It is highly expressed on plasmablasts, plasma cells, NK cells, and activated T and B cells in normal healthy subjects and on malignant plasma cells in multiple myeloma patients. 18 In vitro models with human cell lines demonstrated that binding of TAK-079 to CD38 induced depletion of human B cell lines by antibody-dependent cellmediated cytotoxicity and complement-dependent cytotoxicity and, in most cases, cell lines with increased CD38 expression were more susceptible to cell lysis. 22 The amino acid sequences of human and rodent CD38 exhibit low homology, whereas the homology of the human CD38 protein with cynomolgus monkeys is considerably higher (92%). 7,13,16 Despite the high homology, the anti-human CD38 monoclonal antibodies daratumumab (Darzalex) and isatuximab do not cross-react with monkey CD38. 4,26 In contrast, TAK-079 binds to monkey CD38 and this provides the unique opportunity to study anti-CD38 cytolytic activity in nonhuman primates.
The objectives of this study were to characterize the pharmacokinetics (PK) and pharmacodynamics (PD) of TAK-079 in monkeys and to build mathematical models, which could guide dose selection for the first-in-human (FIH) clinical trial. To this end, assays were developed to measure drug concentrations and immunogenicity, and to quantify T, B, and NK lymphocytes in the blood of monkeys. We assessed these parameters in eight pharmacological and toxicological preclinical studies. Mathematical models that describe the PK and PD data of therapeutical monoclonal antibodies were recognized as useful tools to gain mechanistic and quantitative insights into the relationships between drug exposure and effect. 9,12,17 Typical PK features of IgG antibodies including distribution and elimination, physiological and genetic similarities between monkey and human could be leveraged to explain the pharmacology of TAK-079. 11,15 In addition, those models have been successfully applied to predict PK concentrations and PD effects in healthy human subjects. 12 Here we describe the derivation of unique PK and PD models of anti-CD38 activity and the first opportunity to utilize this model for guiding the design of the first in human studies of anti-CD38 therapeutics.

| Animal studies
The studies were conducted in cynomolgus monkeys (Macaca fascicularis). A summary of them is shown in Table 1 in chronological order. The single dose studies 2, 7, and 8 were primarily conducted to evaluate PK and PD of intravenously (IV) and subcutaneously (SC) administered TAK-079. The repeated dose studies were performed to evaluate safety, PK, and PD, including two 4-week studies (studies 1 and 3) and three 13-week studies under Good Laboratory Practice (GLP) conditions (studies 4, 5, and 6). In study 4 seven doses of 3, 30 or 80 mg/kg were administered every other week (Q2W). According to the protocol the majority of animals were terminated after 98 days (14 days after the last dose) for detailed toxicological assessments. 4 animals of each group were assigned to a recovery group and kept for additional 98 days. In the other 13-week studies 5 and 6, the animals received weekly doses (QW). In study 5, in which a dosing error occurred, animals in the lowest dose group received 0.01 mg/kg instead of the intended 0.1 mg/kg at one occasion (the second dose) and then continued with 0.1 mg/kg. These data were added to the data set with the correct information of the actual administered dosing amounts. Study 6 repeated the low dose of 0.1 mg/kg QW group of study 5. All animal studies were carried out in accordance with the Guide for the Care and Use of Laboratory Animals as adopted and promulgated by the U.S. National Institutes of Health.

| Bioanalytics
PK was analyzed using a validated method developed and performed by Charles River Laboratories (Reno, NV). Briefly, the concentration of TAK-079 was measured in monkey serum, using an indirect enzyme-linked immunosorbent assay (ELISA). A 96-well plate was coated with anti-idiotypic antibody against TAK-079. Blanks, standards, and quality control (QC) samples containing TAK-079 at various concentrations were added to the plate, and incubated for 5565 minutes at room temperature (RT). After washing the plate, the detection antibody (Peroxidase AffiniPure Mouse Anti-Human IgG, Fcc Fragment Specific; Jackson ImmunoResearch) was added, and incubated on the plate for an additional 5565 minutes. The plate was washed, and tetramethylbenzidine was added to the wells to generate a chromophore, and the reaction stopped by the addition 2N sulfuric acid. Absorbance at 450 nm was measured using a SPECTRAmax â 190 microplate reader (Molecular Devices) and the TAK-079 concentrations calculated using a 4-parameter logistic weighted (1/y 2 ) standard calibration curve. In the first study (Table 1), the lower limit of quantification (LLOQ) of TAK079 in serum was 0.061 lg/mL and in all other studies it was 0.05 lg/mL.

| Determination of anti-TAK-079 antibodies (immunogenicity)
Anti-drug antibodies (ADA) screening of monkey serum was done, using qualitative electrochemiluminescent (ECL) method, validated and performed by Charles River Laboratories (Reno, NV). Briefly, undiluted serum samples were mixed acid dissociated (300 mmol/L acetic acid) then incubated in a mixture of biotinylated TAK-079, TAK-079 labeled with SULFO-TAG (Meso Scale Diagnostics, labeled at Charles River) and 1.5 mol/L Trizma base to neutralize the acid and form an immune complex. This complex was added to a streptavidin-coated MSD plate (Meso Scale Diagnostics) and allowed to bind. After washing, the complex was detected by the addition of MSD read buffer T (Meso Scale Diagnostics) to the plate and subsequent excitation was read, using the MSD Sector 6000 (Meso Scale Diagnostics).

| Characterization of blood cells
To compare the level of TAK-079 binding between humans and monkeys, blood samples from each were collected into sodium heparin tubes and an aliquot (100 lL) was mixed with appropriate volume of each antibody (Table S1)  In studies outlined in Table 1, cells were stained and analyzed, using a validated method developed and performed by Charles River Laboratories (Reno, NV). Monkey blood samples were collected into sodium heparin tubes before and at multiple times after TAK-079 treatment and specific lymphocyte populations measured, using FACSCanto TM II Flow Cytometer (BD Biosciences). Commercial antibodies and a CD38 antibody (TSF-19) were titered to optimal concentrations for staining. T-cell, B-cell and NK-cell populations were identified and lymphocytes quantified, using CD45TruCount TM tubes (BD Biosciences). Blood cells (100 lL aliquots) were mixed with antibodies added at the indicated volume (Table S1) (Table S1). Although TSF-19 binds to a different epitope, the results were very similar and are therefore not presented separately.

| PK model development
During PK model development, 1-, 2-, and 3-compartment models were investigated. Note that the effect of ADA was not described in the model but the affected data points were excluded during modeling and parameter estimation. The two-compartment model was clearly superior to the one-compartment model, as judged by goodness-of-fit (GOF) plots and decrease in objective function value (OFV). Based on visual inspections of diagnostic plots, the introduction of a third compartment was not necessary to describe the data adequately. The bioavailability (F) was modeled using the logit transformation F = exp (PAR)/(1 + exp (PAR)), where PAR designates the model parameter, to ensure that the estimates are bounded between 0 and 1. The nonlinear PK at low concentrations was modeled with the QSS approximation model of the targetmediated drug disposition (TMDD) process. 10 A schematic representation and the differential equations of the final PK base model are provided in Figure 2. For the QSS approximation, we assume that the steady-state concentrations of the free drug C, the target R, and the drug target complex RC are established very quickly compared to all other processes. This implies that the binding process is balanced with the dissociation and internalization processes and that the following equation holds in the appropriate units: The between-subject variability (BSV) was investigated for all parameters and modeled with exponential models of the following type: PAR i = TVPAR * e ETAPAR i , where PAR i is the individual and TVPAR the estimate of the typical value (or point estimate) of the parameter and ETAPAR i is the estimate of the deviation of individual i. The ETAPAR i values were assumed to follow a normal distribution with mean zero. The residuals were described with a combined additive and proportional error model. 1 The following characteristics that could be potential covariates of the PK of TAK-079 were available in the data set: body weight, sex, dose, route of administration, and study. Note that the actual dose of each animal was calculated based on the dose level (in mg/kg) and its predose body weight. The covariates were investigated by correlating their individual levels with the individual deviations of each of the PK parameters. Most of the correlations were negligible so that it was unlikely that the covariate level could explain significant parts of the between subject variability of the PK parameter.
Only the potential effects of the route of administration were tested systematically in a stepwise inclusion procedure.

| PK-PD model development
For each of the three cell types, PK-PD model development was performed separately, during which the PK model and parameter estimates were kept fixed. Note that for model development measurements close to the drug administration (<8 hours postdose) were not utilized because they were influenced by a nonspecific drug-independent effect. Turnover, transit compartment and direct response models of various forms were tested. 9,17 In the turnover models, the drug effect was introduced on the cell elimination rate in form of an E max type model with or without Hill factors. In our nota-

| Pharmacokinetics of TAK-079
The PK data set was pooled from all eight studies in healthy monkeys excluding the placebo groups (Table 1). In total, the set con-   Figure S2) provide evidence for nonlinearly augmented clearance at concentrations below 0.5 lg/mL likely caused by target-mediated mechanisms (TMDD). 15 The PK data after single dose SC administration were generated in preparation for the first in human clinical trial using another formulation of TAK-079 ( Figure 1C). The data revealed that C max was 70%-80% lower in the SC versus IV groups of the same dose, and that there were no systematic differences in elimination or AUC between the groups.
No systematic differences in PK parameters between male and female monkeys were observed. The results of these initial analyses were used as the starting point for model development.

| PK model development
Model development started with single IV dose data and then the initial model was gradually extended utilizing more complex data.
Similar to other therapeutic antibodies, the PK grossly follows a linear 2-compartment model. 15 The nonlinear elimination component (TMDD) describing the accelerated clearance at low concentrations was modeled with the QSS approximation. 10 The assumption that the drug-target association process is much faster than the processes of drug dissociation, distribution and elimination, and of target and drug-target complex elimination leads to the simplified TMDD model ( Figure 2, Table 2). The amount of data at low concentrations was relatively small and we did not manage to estimate all the parameters successfully in a single estimation run of the software program. Therefore, we chose to define the model structure  Table 2).
We obtained estimates for the absorption rate (K A ) and the bioavailability (F) when we added the data of the SC groups. Note that all SC data come from four lower single dose groups from studies 7 and 8. These lower doses (≤1 mg/kg) cover the estimated clinically relevant range but may limit the generalizability of the parameter estimates for higher doses.
During the covariate analysis, we searched for potential relationships between body weight, sex and route of administration, and PK parameters. We identified an effect of the route of administration   Figure S4).

| Pharmacodynamics
The level of TAK-079 binding on human and monkey blood NK cells,    Table 2. The differential equations specify the PK model in terms of amount of drug (cen and per) and total amount of receptor (rtot) The estimates and standard errors for the TMDD parameters were gained from a separate run that focused on the data of the low dose groups (residual variability of the separate estimation: additive 0.005, proportional 0.067), and were then fixed for the final estimation of the other PK parameters.  Figure 4B). In concordance with these results, NK function was also tested in a subset of animals in study 7 (n = 3/group; B cells and T cells were depleted to a lesser extent as compared to NK cells, which is consistent with their lower CD38 expression levels ( Figure 3). For example, at 0.3 mg/kg IV TAK-079, B cells had a median maximal level of depletion to 45% of baseline, and T cells were depleted to 43% of baseline ( Figure 4D and G). At this dose level, a 50% reduction from baseline of B cell counts was not achieved in all animals. Only at the highest doses of ≥30 mg/kg were the B cells almost completely depleted ( Figure 4D). T cells were depleted to an extent similar to B cells, but the recovery was faster ( Figure 4E-I).
In the two studies 7 and 8, we compared IV and SC dosing (

| PK-PD models
We developed separate PK-PD models to describe the effects of TAK-079 exposure on NK, B, and T cells. During PK-PD modeling the PK parameters were kept fixed and a variety of PD models were evaluated (see Materials and Methods for detail). The NK cell population in the peripheral blood was adequately described with a turnover model and the depleting drug effect was linked via the PK concentration with an E max type model to the rate of depletion. In this model the E MAX represents the maximum rate of additional NK cell depletion (in addition to the base line rate of elimination K OUT ) and the C 50 the concentration at which the rate of additional NK cell depletion is half-maximal. The structural PK-PD model for NK cells was of the following form:  (Table 3, Figure S7).
The transit compartment model was superior to direct response or turnover models to describe TAK-079-induced B-cell depletion.
Four transit compartments turned out to be adequate and the drug effect was described with an Emax-type model on the depletion rate. Like in the NK cell depletion model, the E MAX represents the maximum rate and the C 50 the concentration at which the rate is given by the following five equations:  (Table 3). Note however, that the between subject variability on E MAX was nearly 70%. In this model, different from the NK and Bcell depletion models, the C 50 represents the concentration at which the depletion of T cells was half-maximal. A special situation was observed in the 3 mg/kg group. Although the data at later time points are fitted adequately, the depletion after the first dose was underestimated ( Figures 4I, S8). This is in accordance with observations from the repeated dose studies that, despite continuous treatment, T cells recover after initial depletion. In summary, the T-cell model describes the data of the lower (clinically relevant) doses and of the repeated higher doses well but not the initial strong depletion after a first high dose.
Like for the NK cells, model evaluation of the final PK-PD models for B and T cells based on residual errors, OFV, standard errors, GOF plots and individual curve fits corroborates that they adequately described the available monkey data (Table 3, Figure S7).

| Simulation of human PK and cell depletion
The monkey PK and PK-PD models were used as the starting point At an IV dose of 0.3 mg/kg, we predicted NK cell depletion to 17% of baseline within 3 hours. After the end of infusion and recovery cells increased to more than 50% after 11 days (Figure 5). At the T A B L E 3 PK-PD modeling results: parameter estimates and standard errors in percent (%SE)

IV infusion, 2 h Subcutaneous injection
F I G U R E 5 Simulated human PK and NK cell, B-cell and T-cell depletion profiles of TAK-079. Based on the scaled monkey PK and PK-PD models 5 single IV and SC dose PK and cell depletion profiles were simulated (from 0.0003 to 1 mg/kg). The left plots show the data after IV and the right plots after SC administration. The 2 plots in the first row display the PK profiles. The y-axis is log scaled and the LLOQ of 0.05 lg/mL is indicated by a horizontal dashed line. The PK of the lowest dose was completely superimposed by noise and only at doses of 0.03 mg/kg the PK reaches levels above LLOQ that over time more and more animals developed ADA and the levels increased. TAK-079 is a fully human monoclonal antibody, and therefore we expect lower ADA levels in human subjects compared to what we observed in monkeys. Consequently, the information about the developing immunogenicity and its effects on drug elimination and potentially efficacy that can be gained in this animal model is limited.
With the emerging human data, it will be interesting to compare human and monkey PK and PD data in detail. The construction of a PK model based on human data and a comparison to the monkey The rich pharmacological data and the PK and PK-PD models enabled us to characterize exposure-effect relationships in monkeys.