An age-dependent QSP model of Nf trafficking to optimize treatment in neurodegenerative diseases
Alessio Paris (1), Pranami Bora (1), Silvia Parolo (1), Michael Monine (2), Xiao Tong (2), Toby Ferguson (2), Alexander McCampbell (2), Danielle Graham (2), Drew MacCannell (2,3), Nick van der Munnik (2,4), Satish Eraly (2,5), Zdenek Berger (2), Stephanie Fradette (2), Eric Masson (2,6), Enrico Domenici (1,7), Ivan Nestorov (2), Luca Marchetti (1,7)
(1) Fondazione The Microsoft Research - University of Trento Centre for Computational and Systems Biology (COSBI), Rovereto, Italy (2) Biogen, Inc., Cambridge, Massachusetts, USA, (3) current institution: Dyne Therapeutics, Waltham, Massachusetts, USA, (4) current institution: GSK, Greater Boston, Massachusetts, USA, (5) current institution: Alnylam Pharmaceuticals, Cambridge, Massachusetts, USA, (6) current institution: Arvinas, New Haven, CT, USA, (7) Department of Cellular, Computational and Integrative Biology (CIBIO), University of Trento, Trento, Italy
Introduction: Neurofilament (Nf) has been identified as a potential biomarker for different neurodegenerative diseases, including SMA and ALS [1,2], and as an indicator of both disease progression and potential treatment efficacy. The evaluation of Nfs as a biomarker would benefit from modeling Nf trafficking from the nervous system to the detection sites, typically cerebrospinal fluid (CSF) or blood.
Objectives:
- Present a quantitative systems pharmacology (QSP) framework of Nf trafficking, developed as a general and extendible Ordinary Differential Equation (ODE) model for different Nf subunits (NfL-Nf Light and pNfH- phosphorylated Nf Heavy).
- Report practical examples of the application of the model to the case of untreated and treated Spinal Muscular Atrophy (SMA) and SOD1 Amyotrophic Lateral Sclerosis (SOD1-ALS) using clinical trial data.
- Present virtual simulation scenarios of drug administration for SMA and SOD1-ALS.
Methods: The Nf platform was originally developed as an ODE model of Nf trafficking in healthy adult subjects [3]. This common platform was extended to address the modified pNfH and NfL trafficking in the presence of pediatric SMA or SOD1-ALS, respectively. For SMA, the extension required ad-hoc modifications to the equations for reproducing the effects of volume growth and neuronal development with age [4]. In the case of SOD1-ALS, distinct compartments identifying non-motor and motor neurons in the central nervous system were required, because only the latter are affected by the disease. Additional reactions describing the effect of the disease and of the treatment on the leakage of Nf were included in each case. The integration with pharmacokinetics (PK) models of antisense oligonucleotides (ASOs) nusinersen [5] or tofersen [6] allows the modeling of individual drug administration protocols for SMA and SOD1-ALS, respectively. The indirect and delayed effect of the ASOs on Nf was simulated by introducing transit compartments. The parameters of the adult healthy model were estimated on literature data to reproduce the cross-sectional variability of Nf levels at different ages, and the average Nf increasing with age from 20 to 90 years [3]. In the case of SMA [4], pNfH concentrations in CSF and blood from several nusinersen studies in participants with SMA ([7-11]) were used to calibrate and validate the model (doses from 1 to 12 mg). In the case of SOD1-ALS, NfL data from a tofersen study were used [12, 13], during which ascending-dose (10, 20, 40, 60 mg) and fixed-dose (100 mg) treatments were administered.
Results: The model could fit with good accuracy the Nf data of the training group of patients. For SMA, the median of the normalized mean root square deviations (NRMSD) of pNfH timeseries was 50% in CSF and 60% in blood, while for SOD1-ALS the median NRMSD of NfL timeseries was 18% in CSF and 22% in blood. To validate the SMA model, we compared the experimental pNfH timeseries of 25 treated patients to the model-predicted 95% CL bands of untreated and treated virtual populations, observing a substantial overlap of the timeseries with the treated band and a clear distinction from the untreated band. The validation for SOD1-ALS followed the same approach, by considering individual NfL timeseries of 41 treated SOD1-ALS patients. On average, 91% of each experimental timeseries overlapped with the predicted 95% CL bands. The calibrated model platform was then used to predict in-silico virtual scenarios, exploring higher doses and different administration schedules of nusinersen or tofersen. The model was also extended with a function reproducing the increase of NfL levels during the onset of SOD1-ALS. This opened the possibility of simulating the administration of tofersen at an early stage of the disease or even during the pre-symptomatic phase.
Conclusions: We present a QSP framework for Nf trafficking that was extended to the case of SMA and SOD1-ALS. The framework successfully reproduced Nf timeseries from clinical studies and demonstrated predictive power when applied to validation cases. It was also employed to predict drug response that could be helpful in the optimization of treatment protocols. With the same approach used for SMA and SOD1-ALS, our model can be easily extended to other neurodegenerative diseases for which Nf is identified as a relevant biomarker. Current work is underway also to integrate the model with data from different neurodegeneration biomarkers.
References:
[1] Bridel C et al. JAMA neurology, 2019; 76.9: 1035-1048.
[2] Darras BT et al. Ann Clin Trans Neurol, 2019; 6(5):932-944.
[3] Paris A et al. CPT Pharmacometrics Syst Pharmacol., 2022; 11.4: 447-457.
[4] Paris A et al. CPT Pharmacometrics Syst Pharmacol. 2022; 00:1-11.
[5] Biliouris K et al. CPT: Pharmacometrics & Systems Pharmacology, 2018; 7.9: 581-592.
[6] Monine M et al, Journal of Pharmacokinetics and Pharmacodynamics, 2021; 48.5: 639-654.
[7] Castro D et al. Neurology. 2020; 94(15):1640.
[8] De Vivo DC et al. Neuromuscul Disord. 2019;29(11):842-856.
[9] Acsadi G et al. Muscle Nerve. 2021;63(5):668-677.
[10] Finkel Ret al. Eur J Paediatr Neurol. 2017;21:e14-e15.
[11] Mercuri E et al. New England Journal of Medicine. 2018;378(7):625-635.
[12] Miller TM et al. New England Journal of Medicine, 2020, 383.2: 109-119.
[13] Miller TM et al. New England Journal of Medicine, 2022; 387.12: 1099-1110.