Gilles Tiraboschi1, David Fabre1, John T O’ Malley2, Karl Yen3, Charlotte Bernigaud4, Fabrice Hurbin1
1Sanofi, 2Sanofi, 3Sanofi, 4Sanofi
Introduction & Objectives: Amlitelimab is a fully human, non-depleting, anti-OX40L monoclonal antibody that binds to OX40L on antigen-presenting cells (APCs), preventing interaction with OX40 on activated T cells. Amlitelimab is currently developed in several indications including atopic dermatitis (AD), also known as atopic eczema, a common chronic/relapsing inflammatory skin disorder, which has a significant impact on the health and quality of life of individuals. Recent 2b clinical trial has demonstrated the safety and efficacy of amlitelimab in adults with moderate-to-severe AD with data supporting the evaluation of 250 mg Q4W with 500 mg loading dose (+LD) in phase 3 trials. A modeling strategy has been developed with the following objectives. Characterize the pharmacokinetic of amlitelimab in AD patients using population pharmacokinetic (popPK) Characterize the pharmacokinetic/pharmacodynamic relationships with Eczema Area and Severity Index (PopPK/PD-EASI) Simulate alternative dosing regimens and patient populations to support the choice of dose regimens for future phase 3 clinical studies Methods: A sequential approach was used to first develop the popPK model, and then the PopPK/PD-EASI model, using the predicted amlitelimab concentrations. Models were developed in NONMEM 7.5.1 [1] using FOCEI for optimization. For the qualification steps Prediction-Corrected Visual Predictive Check [2] and Sampling Importance Resampling iterative [3] methods were used. For the computation of exposure parameters and simulations, the mrgsolve package [4] embedded in an in-house shiny application [5] was used. Results: A PopPK model for amlitelimab was developed pooling data from three Phase 1 and two Phase 2 (Phase 2a and 2b) trials. 439 participants (78 healthy adults; 361 adults with AD) were included in the PopPK analysis (bodyweight [ 40.5-148kg]). The final 2-compartment model with linear and non-linear clearances, and bioavailability, lag-time and first order rate to characterize subcutaneous route included bodyweight allometric factors as well as a baseline EASI increasing effect on linear clearance. A limited effect of albumin on the bioavailability was also identified. The PopPK/PD-EASI model was developed using interim data from 269 amlitelimab-treated patients with AD from the completed Phase 2a and ongoing 2b trials. A final turnover response model (type I – loss of induction) with an Imax drug effect equation was selected. Both Phase 2 studies were a 2-part randomized, double-blinded placebo-controlled trials. In Part 1, all patients were treated, and only clinical responders were selected for Part 2. In Part 2, clinical responders were randomly reallocated to withdraw amlitelimab or to continue their previous dose regimen). To robustly predict and simulate EASI clinical endpoint improvements, a categorical clinical responder covariate was tested and selected in the final PopPK/PD-EASI model, showing its superiority to a mixture model. Simulations indicated that responders at 250mg Q12W +LD had a percent change from baseline in EASI in the range observed with Q4W dosing regimens in Phase 2b (62.5 mg to 250 mg). Simulations performed in patients from Phase 2b based on the final PopPK model, showed that exposures following an extended dose of 250mg Q12W (with 500mg +LD), predicted exposures between the range observed in the Phase 2b for the 62.5mg Q4W and 250mg Q4W +LD regimens. Simulations of virtual patients identified a 2-fold dose reduction for patients <40kg to achieve amlitelimab exposures within the range observed in patients =40kg on 250mg Q4W or Q12W. Conclusions: The simulations support the Q12W extended dose regimen for Phase 3 studies and proposed an adaptation of the dose of amlitelimab for patients <40kg.
[1] Beal S, et al. NONMEM 7.5 Users Guides. ICON Dev Solutions. 2020 [2] Bergstrand M, et al. AAPS J. 2011;13(2):143-51 [3] Dosne A-G, et al. J Pharmacokinet Pharmacodyn. 2017;44(6):509-20 [4] Baron KT. mrgsolve: Simulate from ode-based population pk/pd models. 2019 [5] Chang W, et al. shiny: Web Application Framework for R. 2020
Reference: PAGE 33 (2025) Abstr 11702 [www.page-meeting.org/?abstract=11702]
Poster: Clinical Applications