Stefan Van Den Berg 1,2, Thomas Dorlo 3, Theo Rispens 1,2
1 Sanquin Research, Department of Immunity and Inflammation (Amsterdam, The Netherlands), 2 Amsterdam UMC, Molecular Cell Biology and Immunology (Amsterdam, The Netherlands), 3 Uppsala University, Department of Pharmacy (Uppsala, Sweden)
Introduction:
Therapeutic monoclonal antibodies (mAb) share a high similarity in physiological and pharmacokinetic (PK) properties, suggesting that an underlying generic population PK model may describe IgG homeostasis [1]. Their primary clearance results from non-specific pinocytosis followed by lysosomal degradation, unless rescued by binding to the neonatal Fc receptor (FcRn), which recycles them back into circulation [2]. Other elimination pathways are target‑mediated drug disposition (TMDD) and anti‑drug antibodies (ADA) [3]. Simultaneous processes hinder parameter estimation at both population and individual levels, ultimately making its use in therapeutic drug monitoring more challenging. We hypothesized that a generic IgG model adequately captures the key IgG homeostatic processes of absorption, distribution, degradation, allowing the identification of additional non-linear processes.
We demonstrate this using adalimumab, widely prescribed for the treatment of autoimmune diseases. Adalimumab binds to tumor necrosis factor (TNF). Given the low circulating TNF concentrations, TMDD is expected to be marginal [4]. However, adalimumab is highly immunogenic, resulting in ADA formation that complexes with the drug and accelerates its neutralization and clearance, with substantial between-subject-variability in both the timing and magnitude of these effects [5].
Objectives:
– To assess whether a generic IgG PK model describes adalimumab PK
– To evaluate the ability of the generic model to identify the impact of ADAs on adalimumab elimination
Methods:
The pooled dataset consisted of three cohorts: healthy volunteers which received a single 40‑mg subcutaneous dose with frequent sampling; a plaque psoriasis cohort received 80-mg initially, then 40-mg after one week and biweekly thereafter; and a rheumatoid arthritis cohort received 40-mg biweekly in combination with methotrexate. ADAs were measured with a drug-tolerant electrochemiluminescent assay.
A two-compartmental population PK model with absorption depot was applied with generic IgG parameter estimates as derived previously [1]. All structural and stochastic parameters were fixed, including the variances of inter-individual variability (IIV) on clearance (CL), central volume (Vc) and absorption rate (ka), and body weight as covariate with an exponent on CL, Vc, Vp, Q. The only estimated parameters were the additive and proportional residual errors.
Individuals were grouped according to their ADA titer measured at the final visit (ADA_final). For each group, the mean EBEs of the parameters were examined to assess whether they aligned with the expected mean of zero or showed statistically significant deviations.
The generic model was extended to capture the effect of ADAs with a time-dependent Hill function, parameterized by the amplitude and time to half‑maximal amplitude, both including IIV. This function was applied in three different mechanisms: I) time-varying clearance, II) time-varying nonlinear elimination effectively of zero-order and III) ADA dynamics with ADAs produced by a time-varying production rate and complex formation and elimination. Clearance was not fixed but estimated, informed by the generic estimate as prior.
Results:
The dataset contained 26098 observations of 1784 individuals. Generic population predictions versus observations showed underprediction in the absence of ADAs, while the degree of overprediction increased with increasing ADAs. The mean EBE_CL for individuals without ADAs, translated into a clearance of 0.18 L/day for a 70-kg subject, which is 18% lower than the generic estimate. The EBEs of CL, Vc, and Ka showed significant Spearman correlations with ADA_final (p<0.001).
Three methods to incorporate the effects of ADAs were evaluated, and all three decreased the Spearman coefficient between EBEs and ADA_final. The largest reduction in OFV was achieved with model III and abolished the correlation between EBE_Vc and ADA_final. The estimate of linear CL was 0.17 L/day. Mean EBEs for Vc and ka showed slight deviations of 7.4% and –4.6%, respectively, from the generic estimates. Among psoriasis patients, 58% showed ADA-mediated elimination of at least 0.25 mg/d, with a median(IQR) of 1.5 (0.55-2.4) mg/d. In rheumatoid arthritis patients receiving methotrexate, 38% had >0.25 mg/d of drug removed by ADAs, with a median(IQR) of 0.99 (0.56-1.9) mg/d.
Conclusion:
The generic parameter estimates closely matched the mean parameters of the population receiving adalimumab, whether no ADAs were present or ADA‑related processes were accounted for. Except for clearance, which was lower than the generic estimate obtained by averaging 160 population PK models. This bias may be due to models in which linear clearance absorbs additional processes. A generic model can identify nonlinear clearances that can subsequently be integrated into the model.
References:
[1] Van Den Berg, S. P. H., Adolfsen, P. E. A., Dorlo, T. P. C. & Rispens, T. Does one model fit all mAbs? An evaluation of population pharmacokinetic models. mAbs 17, 2512217 (2025).
[2] Ghetie, V. & Ward, E. S. Transcytosis and Catabolism of Antibody. IR 25, 097–114 (2002).
[3] Gibiansky, L., Gibiansky, E., Kakkar, T. & Ma, P. Approximations of the target-mediated drug disposition model and identifiability of model parameters. J Pharmacokinet Pharmacodyn 35, 573–591 (2008).
[4] Berkhout, L. C. et al. Formation and clearance of TNF–TNF inhibitor complexes during TNF inhibitor treatment. British J Pharmacology 181, 1165–1181 (2024).
[5] Borrega, R. et al. Systematic Review and Principal Components Analysis of the Immunogenicity of Adalimumab. BioDrugs 35, 35–45 (2021).
Reference: PAGE 34 (2026) Abstr 11925 [www.page-meeting.org/?abstract=11925]
Poster: Methodology - New Modelling Approaches