A combined physiologically-based pharmacokinetic and quantitative systems pharmacology model for modeling amyloid aggregation in Alzheimer's disease

Abstract

Antibody-mediated removal of aggregated β-amyloid (Aβ) is the current, most clinically advanced potential disease-modifying treatment approach for Alzheimer's disease. We describe a quantitative systems pharmacology (QSP) approach of the dynamics of Aβ monomers, oligomers, protofibrils, and plaque using a detailed microscopic model of Aβ₄₀ and Aβ₄₂ aggregation and clearance of aggregated Aβ by activated microglia cells, which is enhanced by the interaction of antibody-bound Aβ. The model allows for the prediction of Aβ positron emission tomography (PET) imaging load as measured by a standardized uptake value ratio. A physiology-based pharmacokinetic model is seamlessly integrated to describe target exposure of monoclonal antibodies and simulate dynamics of cerebrospinal fluid (CSF) and plasma biomarkers, including CSF Aβ₄₂ and plasma Aβ₄₂ /Aβ₄₀ ratio biomarkers. Apolipoprotein E genotype is implemented as a difference in microglia clearance. By incorporating antibody-bound, plaque-mediated macrophage activation in the perivascular compartment, the model also predicts the incidence of amyloid-related imaging abnormalities with edema (ARIA-E). The QSP platform is calibrated with pharmacological and clinical information on aducanumab, bapineuzumab, crenezumab, gantenerumab, lecanemab, and solanezumab, predicting adequately the change in PET imaging measured amyloid load and the changes in the plasma Aβ₄₂ /Aβ₄₀ ratio while slightly overestimating the change in CSF Aβ₄₂ . ARIA-E is well predicted for all antibodies except bapineuzumab. This QSP model could support the clinical trial design of different amyloid-modulating interventions, define optimal titration and maintenance schedules, and provide a first step to understand the variability of biomarker response in clinical practice.

FIGURE 1

Schematic representation of the PBPK (left) and the QSP (right) model. The PBPK model is a simplified version of the whole‐body PBPK model with a particular emphasis on the brain and CSF compartments. The QSP model includes synthesis of both Aβ₄₀ and Aβ₄₂ (upper and lower branches, respectively) from APP and C99; enzymatic degradation of the monomeric forms; formation of higher order oligomers by nucleation (forward reactions); monomers splitting off from higher order oligomers (backward reactions); formation of plaques from higher order oligomers; identification of a protofibril population in an off‐pathway structure; secondary nucleation on the surface of plaques; breakdown of large protofibrils in two pieces; and microglia‐dependent clearance of oligomers, protofibrils, and to a lesser extent plaques. The antibodies interact with each of the amyloid peptide forms according to their affinity and bind to the γ‐subunit of immunoglobulin Fc receptor on microglia cells that activate microglia cells via proliferation and switch to a high phagocytosing phenotype. Clinically relevant readouts include SUVR from amyloid imaging, CSF Aβ₄₂ levels, and plasma Aβ₄₂/Aβ₄₀ ratio. Aβ, β‐amyloid; APP, amyloid precursor protein; C99, 99 amino acid C‐terminal fragment of APP; CSF, cerebrospinal fluid; DAM, disease‐associated microglia; IV, intravenous; PBPK, physiologically‐based pharmacokinetic; QSP, quantitative systems pharmacology; SC, subcutaneous; SUVR, standardized uptake value ratio

FIGURE 2

Schematic representation of the perivascular compartment involved in the generation of the ARIA‐E adverse effects. The basic readout is the amount of macrophage‐dependent clearance of antibody‐Aβ complexes in plaques. Recruitment and activation of macrophages leads to inflammatory reactions that is assumed to be visualized with magnetic resonance imaging (ARIA‐E). This is linked to the clinical incidence of ARIA‐E via a time‐to‐event model. σ, reflection coefficient; Aβ, β‐amyloid; ARIA‐E, amyloid‐related imaging abnormalities with edema; CL, clearance; ISF, interstitial fluid; k_on, association rate; PBPK, physiologically‐based pharmacokinetic; PVS, perivascular system; Q_glym, blood flow; QSP, quantitative systems pharmacology

FIGURE 3

Simulation of the natural life cycle (from 20–100 years) of amyloid aggregation for an individual virtual Alzheimer's disease patient as in an observational study. Age‐related pathology is implemented as a linear decrease in monomer degradation, resulting in an increased amount of monomers being pushed into the aggregation pathway. There is a steep transition around the age of 60 years for the SUVR to increase followed by a slow saturation at higher ages. This transition corresponds to the steep decline of Aβ₄₂ CSF and the plasma Aβ₄₂/Aβ₄₀ ratio. Aβ, β‐amyloid; APOE, apolipoprotein E; CSF, cerebrospinal fluid; ISF, interstitial fluid; SUVR, standardized uptake value ratio

FIGURE 4

PK profile simulation of plasma levels (top) and ISF concentrations (bottom) for different doses of subcutaneous gantenerumab and intravenous lecanemab. Measured plasma levels in the single subcutaneous gantenerumab 300 mg dose are overlaid on the top left figure. The top right shows a comparison between five different doses of intravenous lecanemab (symbols) and the simulated concentrations (lines). The bottom left figure shows the anticipated ISF concentration for doses used in phase III studies: 1200 mg subcutaneous for gantenerumab and 10 mg/kg once every 2 weeks for lecanemab. Their corresponding CSF concentrations are shown in the bottom right figure. carr, carrier; CSF, cerebrospinal fluid; ISF, interstitial fluid; IV, intravenous; mAb, monoclonal antibody; PBPK, physiologically‐based pharmacokinetic; PK, pharmacokinetics; Q2W, once every 2 weeks; Q4W, once every 4 weeks; SC, subcutaneous

FIGURE 5

Calibration of changes in SUVR with different antibodies and timepoints for readout showing a robust correlation between simulated and actual values in different antibodies (a total of 23 data points) (top left). Predicted and actual values of changes in SUVR with lecanemab treatment 10 mg/kg once every 2 weeks (top right). Predicted and actual values of changes in CSF Aβ₄₂ with lecanemab treatment 10 mg/kg once every 2 weeks (bottom left). Predicted and actual values of changes in plasma Aβ₄₂/Aβ₄₀ ratio with lecanemab treatment 10 mg/kg once every 2 weeks (bottom right). Aβ, β‐amyloid; CSF, cerebrospinal fluid; LEC, lecanemab; Obs, observed; PET, positron emission tomography; Q2W, once every 2 weeks; Q4W, once every 4 weeks; SIM, simulated; SUVR, standardized uptake value ratio

FIGURE 6

Calibration of ARIA‐E using the clinical data of six different antibodies using the interaction between macrophages and bound antibody‐Aβ plaque levels (a) and bound antibody‐Aβ protofibril levels (b). The model suggests a much higher correlation when considering the amount of antibody bound to plaques (r ² = 0.72) compared with the amount of antibody bound to protofibrils (r ² = 0.06). In both cases, the model does not fit the data obtained by bapineuzumab. The antibody‐plaque complex in the perivascular space correlated best with ARIA‐E. Aβ, β‐amyloid; ARIA‐E, amyloid‐related imaging abnormalities with edema; carr, carrier; IV, intravenous; Q2W, once every 2 weeks; Q4W, once every 4 weeks; Q13W, once every 13 weeks; SC, subcutaneous

FIGURE 7

Predicted changes in soluble oligomeric, protofibril, and plaque levels corresponding to measurable changes in SUVR for the six different antibodies dosed at clinically relevant doses for 18 months. Because all forms are linked by well‐described reactions, the model allows for simulations of the intermediate forms of aggregated Aβ peptides starting from clinically observed PET imaging values. The simulated profiles suggest different profiles for changes in oligomers and protofibrils even with very similar SUVR changes (compare, for instance, lecanemab and aducanumab). Aβ, β‐amyloid; IV, intravenous; PET, positron emission tomography; Q2W, once every 2 weeks; Q4W, once every 4 weeks; Q13W, once every 13 weeks; SUVR, standardized uptake value ratio