(W-006) Application of identifiability analysis to obtain reliable parameters of the QSP model describing mechanism of action GEN1042 (BNT312) and acasunlimab

Ivan Borisov, n/a – n/a, InSysBio CY; Sergey Smirnov, n/a – n/a, InSysBio CY; Oleg Demin, n/a – n/a, InSysBio UK; Gaurav Bajaj, n/a – Senior Director, Clinical Pharmacology and Quantitative Sciences, Genmab US, Inc; Jordan Blum, n/a – n/a, Genmab US, Inc; Maria Jure-Kunkel, n/a – n/a, Genmab US, Inc; Beesan Tan, n/a – n/a, BioNTech US; Tim Maiwald, n/a – n/a, BioNTech US; Manish Gupta, n/a – n/a, Genmab US, Inc.; Craig Thalhauser, n/a – n/a, Genmab US, Inc

Objectives: GEN1042 (BNT312; DuoBody^®-CD40x4-1BB) is an investigational bispecific antibody (bsAb) that combines targeting and conditional activation of CD40 and 4-1BB on immune cells. Acasunlimab (BNT311; DuoBody^®-PD-L1×4-1BB) is an investigational bsAb providing conditional 4-1BB activation with simultaneous and complementary PD-L1 blockade. To properly describe changes in clinically measured outcomes resulting from GEN1042 and acasunlimab therapy implemented in an NSCLC QSP model [1], parameters responsible for effect of trimeric states of the bsAbs (CD40-GEN1042-4-1BB, PDL1-acasunlimab-4-1BB) on cell dynamics should be fitted against limited set of available in vitro data. We aimed to develop a model describing in vitro experiments conducted to characterize the effects of GEN1042 and acasunlimab, to fit parameters responsible for the effect of the bsAbs on T-cell proliferation, and to perform identifiability analysis to understand whether the available dataset allows us to estimate key model parameters within finite confidence intervals.

Methods: Data: (i) dependency of expansion index on GEN1042 dose measured at 96 hours, (ii) Four dependencies of expansion index on acasunlimab dose measured at different levels of PD-1 expression.

The Model describes myeloid dendritic cell-dependent T-cell activation and proliferation, formation of trimers in the immunological synapse, stimulation of T-cell proliferation via trimers, and inhibition with PD-1–PD-L1 pathway.

Fitting and identifiability analysis: Parameters describing effects of GEN1042 and acasunlimab on T-cell proliferation were fitted together or separately against in vitro experimental data. Identifiability analysis of parameters was performed per the algorithm described in [2].

Results: Fitting of acasunlimab dataset (ii) allows us to obtain the parameter values responsible for the stimulation of proliferation via PD-L1-acasunlimab-4-1BB trimer formation (1046_trimer_Emax, 1046_trimer_EC50). An identifiability analysis performed on the estimated parameters confirmed they have finite confidence intervals. However, the GEN1042 dataset (i) does not allow us to unequivocally identify parameters responsible for stimulation of proliferation via CD40-GEN1042-4-1BB trimer formation (1042_trimer_Emax, 1042_trimer_EC50). To improve identifiability, pseudo-experimental data describing GEN1042 dose dependencies at different E:T ratios, expression levels of CD40L, and PD-1 were generated; however, fitting them together with (i) did not improve identifiability. We next assumed that 1042_trimer_EC50 = 1046_trimer_EC50 (the same number of 4-1BB-engaging trimers would lead to the same effect), fitted all parameters against combined datasets (i) and (ii), and found that model parameters were identifiable.

Conclusions: Parameter estimation together with identifiability analysis allows us to find reliable parameter values responsible for the effects of GEN1042 and acasunlimab on T-cell proliferation.





Citations: [1] Bajaj G, et al, poster at ACoP2024

[2] Borisov I, et al, PLoS Comput Biol. 2020;16(12):e1008495