Gianluca Selvaggio 1, Moriah Pellowe 1, Erik Sjögren 1
1 Pharmetheus (Uppsala, Sweden)
Objective:
Immunogenicity prediction is a critical component of monoclonal antibody (mAb) development. The emergence of anti-drug antibodies (ADA) remains a primary bottleneck in drug development, as it can significantly hamper mAb pharmacokinetics, efficacy while posing serious safety risks. Thus, in silico prediction of ADA incidence is essential for de-risking candidates. While mechanistic models, such as the multiscale framework by Chen et al. [1,2], provide a robust basis for immune dynamics, their ability to reproduce clinical incidence across diverse mAbs sequences requires validation.
Building on previous MoBi implementation [3] of the Chen model extended to accommodate other routes of administration, this work aims to benchmark and refine the model using a library of publicly available mAb sequences and literature-reported clinical incidence.
Methods:
A workflow was established to integrate sequence-based epitope prediction with multiscale modelling:
• Sequence analysis and processing. The mAb sequences were partitioned into specific regions (CDR, FR) using abYsis [4] to facilitate targeted analysis. Sequences were then processed via NIH IgBLAST [5] to identify non-self regions and exclude endogenous sequences that would not contribute to an immunogenic signal.
• Epitope prediction. The putative epitopes and their specific binding affinities were predicted using NetMHCIIpan 4.3i [6]. To maintain high stringency, only the top 2% of elution rank binders were retained for downstream simulation.
• HLA virtual population generation. A high-resolution virtual population was generated using the UPHD 9-loci HLA frequency dataset from the 17th International HLA and Immunogenetics Workshop [7].
• Epitope selection and mapping. Two epitopes were selected per HLA allele per individual. The criteria were based on epitope-MHC binding strength and the total number of receptors recognizing the sequence.
• Pharmacokinetic modelling. mAb disposition was characterized using a population pharmacokinetic framework. Intravenous (IV) administration followed parameters from Betts A et al. [8] while for subcutaneous (SC) administration, were derived from Davda JP et al [9].
• Model refinement and sensitivity analysis. Investigations focused on the core Chen et al. [1,2] multiscale model structure. To address the observation of overly strong immune response in standard configurations, the threshold for B-cell activation was systematically reduced. Specifically, the number of B-cells activated per functional T-cell—traditionally 10 naive and 100 memory cells—was modified to evaluate the impact on predicted clinical incidence without requiring the addition of complex regulatory T-cell (Treg) dynamics [10].
Results:
Preliminary simulations indicate that standard B-cell activation parameters may overestimate immune response strength. A streamlined sequence selection and epitope-MHC identification process for the generation of representative virtual populations were established. The standard B-cell activation parameters were identified as key parameters to address a systematic overestimation of immune response strength. Systematic reduction of the B-cell-to-T-cell activation ratio allowed for a more nuanced calibration to match literature-reported immunogenicity incidence. This modification suggests that modulating the proliferative capacity of the B-cell pool is a viable alternative to incorporating complex regulatory Treg dynamics for reproducing clinical observations [10].
Conclusion:
This effort demonstrates a structured path toward standardizing mechanistic immunogenicity models within the OSP framework. By aligning sequence-derived epitope burdens with refined B-cell activation thresholds, the model provides a mechanistic driven scalable tool for early stage assessment of ADA risk.
References:
[1] Chen X, et al. CPT:PSP (2014) 3, e133.
[2] Chen X, et al. CPT:PSP (2014) 3, e134.
[3] Pellowe M, et al. PAGE (2022) 31, 10245.
[4] Swindells MB, et al. J Mol Biol. (2017) 429, 356–364.
[5] Ye J, et al. Nucleic Acids Res. (2013) 41, W34–W40.
[6] Reynisson B, et al. Nucleic Acids Res. (2020) 48, W449–W454.
[7] Creary LE, et al. Hum Immunol. (2021) 82(7), 505–522.
[8] Betts A et al. JPKPD (2018) 45, 507–523.
[9] Davda JP, et al. mAbs (2014) 6(4), 1094-1102
[10] Hamuro L, et al. AAPS J. (2019) 22, 14.
Reference: PAGE 34 (2026) Abstr 12256 [www.page-meeting.org/?abstract=12256]
Poster: Methodology - New Modelling Approaches