CNS*2025 Florence

Multilevel Modeling of mouse CA1 pyramidal neurons under Alzheimer’s condition, using NEURON- based and Adaptive GLIF Frameworks

Paola Vitale^*1, Caterina Tribuzi², Addolorata Marasco^1,3, Michele Migliore<sup>1 ,4

1</sup> Institute of Biophysics, National Research Council, Palermo, Italy.
² Nova Analysis, Brescia, Italy.
³ Department of Mathematics and Applications, University of Naples Federico II, Naples, Italy.
⁴ SUNY Downstate Health Sciences University, Brooklyn, New York 11203, USA.

^*Email: paola.vitale@ibf.cnr.it

Introduction


Alzheimer's disease (AD) is a progressive
neurodegenerative disorder characterized by cerebral amyloid-β accumulation and
disrupted neuronal Ca²⁺ regulation, impairing neural stability. Recent findings
suggest a pathological role of the APP intracellular domain (AICD), even at
physiological levels, in altering excitability and synaptic function of
hippocampal CA1 pyramidal neurons.¹ In this study, we combine statistical
analysis of electrophysiological data with computational modeling (NEURON and
A-GLIF frameworks) to reveal AICD-induced alterations in single-neuron
dynamics.

Methods


Electrophysiological features were extracted using the
NFE tool (EBRAINS), based on the eFEL library.² Single-cell models were optimized in NEURON³ using a control cell, then extended to control and
AICD datasets using 10 features via BluePyOpt,⁴ focusing on passive properties and Na, CaL, SK, and
Km channels. A-GLIF models reproduced spike times of both datasets. To assess
whether NEURON and A-GLIF accurately replicated spike trains, we used a
multivariate Mann-Whitney U-test. Model performance was evaluated using
ST-Accuracy and ST-Fscore metrics via STSimM.⁵

Results

We analyzed Control and AICD datasets to assess AICD
effects on neural firing and to define constraints for NEURON-based models.
Significant differences emerged, especially in features related to the first 5
ISI firing frequencies, which were then used as optimization constraints.
Optimized NEURON models revealed the active and passive properties most
affected by AICD. A-GLIF models were tested across all cells and stimulation
protocols. In all cases, spike times from A-GLIF and NEURON models were
statistically indistinguishable from experimental data (Mann-Whitney U-test),
with high accuracy confirmed by STSimM metrics.⁵



Discussion


Our results offer a comprehensive view of how AICD
affects single-neuron dynamics. Detailed NEURON models combined with
experimental data highlight specific impairments caused by AD-related
byproducts. A-GLIF models, despite their reduced complexity, accurately
reproduce spike trains, firing adaptation, and firing block, while better
capturing subthreshold dynamics than standard LIF models. This makes A-GLIF a
powerful tool for simulating neuronal activity in large-scale network models
under pathological conditions.






This paper was funded by the Italian National Recovery and Resilience Plan (NRRP), M4C2, financed by the European Union - NextGenerationEU (Project IR0000011, CUP B51E22000150006, EBRAINS-Italy).




¹ https://doi.org/10.1016/j.celrep.2019.08.103; https://doi.org/10.3389/fnmol.2021.696476; https://doi.org/10.1111/acel.13778; https://doi.org/10.3389/fncom.2023.1305169.
² https://doi.org/10.3389/fninf.2021.713899.
³ https://doi.org/10.1162/neco.1997.9.6.1179.
⁴ https://doi.org/10.3389/fninf.2016.00017; https://doi.org/10.3389/fninf.2022.991609.
⁵ https://doi.org/10.1016/j.mbs.2024.109192, https://doi.org/10.1016/j.jneumeth.2024.110324.