Fabian Cardozo 1, Mukti Chowkwale 1, Brett Matzuka 1, Vincent Hurez 1, Anasuya Hazra 1, Lutz Harnisch 1, Jason Chittenden 1
1 Regeneron Pharmaceuticals (Tarrytown, US)
Introduction:
Cytokine release syndrome (CRS) is a systemic inflammatory response triggered by several factors such as infections and certain drugs, such as T cell-engaging cancer immunotherapies [1]. Symptoms range from mild, flu-like symptoms to severe, life-threatening conditions like multi-organ system failure. Linvoseltamab is a BCMAxCD3 bispecific antibody, currently approved by the FDA for relapsed/refractory multiple myeloma [2]. While it has demonstrated durable efficacy and generally manageable safety, CRS occurs in 46% of linvoseltamab-treated patients [3]. Most of the CRS events are low grade but severe CRS events with significant clinical effects can occur. Predictive biomarkers for patients at risk of severe CRS are not established yet. Thus, understanding the mechanistic drivers of CRS occurrences will have substantial clinical impact. The levels of cytokine release and the clinical response to a given cytokine level highly vary between patients and are likely a function of patient and disease characteristics, drug physiochemical and pharmacological properties, and dosage of the drug. This study identifies the relationship between linvoseltamab concentrations and released cytokines, inhibitory or stimulatory effects among released cytokines in patients, and tolerance induced by continued linvoseltamab exposure using a precursor-pool model.
Objectives:
The objective of this study is to model the dynamics of cytokines relevant to CRS in individual patients with relapsed/refractory multiple myeloma treated with linvoseltamab.
Methods:
A precursor-dependent model of indirect pharmacodynamic response [4] is used to describe dynamics in cytokine release after linvoseltamab administration. The pharmacokinetic component of the model distributes the IV-administered drug in the central and peripheral compartments, and the precursor pool model uses drug and cytokine concentrations in the central compartment as inputs. The cytokines tumor necrosis factor α (TNFα), interferon γ (IFNγ), interleukin 6 (IL-6), interleukin 10 (IL-10), and C-reactive protein (CRP) were modeled as they have been linked to CRS. Cytokine release was modeled as a function of (1) total linvoseltamab concentration in the central compartment, (2) tolerance induced by continued drug exposure, (3) inhibitory or stimulatory interactions between cytokines, and (4) total concentration of concomitant medications, if any. These effects were determined by iteratively adding one interaction at a time and fitting cytokine concentrations to available clinical trial data. Literature curation was performed to account for known cytokine interactions in oncology and inflammatory diseases, while the iterative addition process accounted for unknown interactions. The models were run using Monolix and model fits were compared using corrected Bayesian Information Criteria (BICc). If a model iteration had a lower BICc value than a previous iteration (indicating better fit), it was selected for comparison with the next model iterations.
Results:
After 183 iterations of predicting cytokine interactions in addition to the linvoseltamab concentration effect, the best model comprises stimulatory effects for TNFα, IFNγ, IL-6, and CRP and inhibitory effect of IL-10. Eight cytokine interactions (5 stimulatory and 3 inhibitory) were included in the model and are listed below.
IL-6 –> TNFα
IL-6 –> IFNγ
IL-10 –> IL-6
CRP –> IL-6
IL6 –> IFNγ
TNFα –| IL-10
IFNγ –| IL-10
IL-6 –| IL-10
The model accurately predicted the median and 10th percentile of TNFα data, and all the data for IL-6, IL-10, and IFNγ. The model underpredicted CRP concentrations (median and 10th percentile). We found that incorporating crosstalk between cytokines significantly improved model accuracy compared to the precursor pool model without any cytokine effects (ΔBICc = 1387). Moreover, we found that the model improved significantly upon adding concomitant medications as covariate effects on IL-6 signaling and CRP production.
Conclusions:
In this study, we developed a precursor-pool model to characterize the heterogenous kinetics of cytokine release after linvoseltamab administration in patients with multiple myeloma. The model accurately predicts the kinetics of TNFα, IFNγ, IL-6, IL-10, and CRP, along with the effects of concomitant medications used to manage CRS symptoms. This model will be applied to predict the occurrence of CRS events in patients treated with linvoseltamab.
References:
[1] Shimabukuro-Vornhagen, Alexander et al. Journal for immunotherapy of cancer, 2018
[2] Lee, Hans C et al. Clinical lymphoma, myeloma & leukemia, 2026
[3] Bumma, Naresh et al. Journal of clinical oncology, 2024
[4] Sharma, A et al. Journal of pharmaceutical sciences, 1998
Reference: PAGE 34 (2026) Abstr 12180 [www.page-meeting.org/?abstract=12180]
Poster: Drug/Disease Modelling - Oncology