In this project, we gathered experimental data on the Aβ-, Aη- and AICD-induced alterations in synaptic plasticity, synaptic signal integration and intrinsic excitability of hippocampal CA1 pyramidal neurons. We built a new computational model of hippocampal synaptic plasticity in a mouse, based on NMDAR functioning, and optimized computational models of mouse CA1 pyramidal neurons under control and AD conditions. We showed that synaptic plasticity was impaired in CA1 pyramidal neurons, leading to perturbed memory storage and recall in CA1 networks. We demonstrated that partial blockade of Glu2NB-NMDAR led to normalization of hippocampal learning in AD conditions.
The project integrated experimental data with the computational modeling and enabled analysis of complex interactions at molecular, synaptic, cellular and network levels, leading to a deeper understanding of the mechanisms involved in AD. The suggested pharmacological manipulation of Glu2NB-NMDAR indicated the potential to restore cognitive functions in AD. The scientific results shed light on a complex AD pathophysiology and its possible treatment.
The activities organized within the project resulted in a productive and strong cooperation not only between the experimental and theoretical neuroscientists in Lithuania, France and Italy, but also with many leading groups external to the project and a major National Research Infrastructure. It also educated young scientists, increased the visibility of neuroscience in academic society and general audience through scientific seminars, summer schools, conferences and general outreach to the public.
Synaptic plasticity models under control and Alzheimer’s disease conditions
We built a phenomenological voltage-dependent NMDAR-based model of hippocampal synaptic plasticity at CA3-CA1 synapses and incorporated the effects of toxic AD peptides (Dainauskas et al, Front. Synapt. Neurosc., 2023; Dainauskas et al, Front. Comput. Neurosc., 2023). The model relies on the functioning of the GluN2A-NMDAR and GluN2B-NMDAR subunits as the separate mediators of LTD and LTP, respectively (Fig.1). It is assumed that LTP is mainly dependent on GluN2B-NMDAR, and GluN2A-NMDAR is the main mediator of LTD.
Fig. 1. Schematic diagram of the synapse model in control conditions (A), under the increased AICD concentration (B), and Aβ concentration (C). A) Presynaptic action potential activates NMDAR, composed of GluN2B-NMDAR and GluN2A-NMDAR subunits gNMDAGluN2B and gNMDAGluN2A, and triggers LTP and LTD functions φNMDALTP and φNMDALTD, respectively. These LTP and LTD functions φNMDALTP and φNMDALTD mutually inhibit each other. NMDAR contributes to the postsynaptic local membrane potential that is low-pass filtered, and the resulting LTP and LTD variables VLTP and VLTD are multiplied by the corresponding NMDAR-dependent functions φNMDALTP and φNMDALTD to form the LTP and LTD components of the AMPAR weight wAMPAR. NMDAR is inhibited by the activity of the nearby Ca2+-dependent K+ channels CagK, that are in turn triggered by the NMDAR activation and lead to the hyperpolarization of the membrane potential. In a similar manner, Ca2+-dependent K+ channels CagK are activated by L-type Ca2+ channels CaL. Dendritic ion channels CagK and CaL exhibit the effect on synaptic weight wAMPAR indirectly through the influence on NMDAR and membrane potential filtered values VLTP and VLTD, and are therefore presented in grey boxes. B) Elevated AICD concentrations (orange bars) lead to GluN2B-NMDAR - gated channel conductance upregulation (green arrow), increase in Ca2+ - activated K+ channel CagK conductance (green arrow) and L-type Ca2+ channel CaL conductance (green arrow) (Pousinha et al., Cell Rep. 2017, eLife 2019). The resulting high intracellular Ca2+ levels overactivate Ca2+-dependent K+ channels and cause hyperpolarization of the neuron membrane, thus leading to the failure in LTP induction. Increase in specific membrane conductance and in axonal, somatic and dendritic M -type K+ current, decrease in axonal, somatic and basal Na+ current are omitted. C) Oligomeric Aβ (light blue bars) increases NMDAR activation (green arrow) through glutamate transmission facilitation, (Abramov et al., Nat. Neurosci., 2009 ), and prevents the activation of LTP protein CaMKII expressed as a phenomenological function φNMDALTP thus leading to deficits in LTP (Opazo et al., 2018). Spine loss caused by Aβ is not indicated. Adjusted from (Dainauskas et al., Front. Comput. Neurosci., 2023).
Computational modeling of AD-related peptide effects on the synaptic plasticity at CA3-CA1 synapses and the pharmacological treatment
We used the synaptic plasticity model, embedded into the multicompartmental model of a CA pyramidal neuron, to study how elevated concentrations of AICD, Aβ, and Aη influence synaptic changes at CA3–CA1 synapses in AD. In the presence of AD-related peptides, LTP was impaired (Dainauskas et al., Front. Comput. Neurosci., 2023), and LTD was slightly enhanced under AICD, Aβ, and Aη conditions, following the experimental data (Pousinha et al., Cell Reports 2017; Opazo et al., Cell Reports 2018).
Figure 2. Effect of GluN2B-NMDAR blockade on somatic EPSP change in control and AICD conditions. GluN2B-NMDAR synaptic conductance was increased by a factor of 4 in AICD conditions. (A) LTP induction protocol. Presynaptic inputs were stimulated at 100 Hz for 1 s. Partial blockade of GluN2B-NMDAR leads to LTP impairment and prevention in control conditions (grey bars) and restores LTP in AICD conditions (red bars). Inset shows the experimental results modified from Pousinha et al. (Cell Reports 2017). Average LTP magnitude normalised to pre-LTP baseline values during 0 to 5 μM in vitro bath application of idenprodil in control conditions (black bars) and in AICD conditions (white bars). Courtesy of Pousinha et al. (Cell Reports 2017). (B) LTD induction protocol. Presynaptic inputs were stimulated at 1 Hz for 500 s. Blockade of GluN2B-NMDAR does not affect LTD in control conditions (grey bars) and AICD conditions (red bars). From (Dainauskas et al., Front. Comput. Neurosci., 2023).
We analyzed how synaptic plasticity was affected by a partial GluN2B-NMDAR blockade in the presence of increased concentrations of AICD, Aβ, and Aη, for high- and low-frequency stimulations. We successfully replicated the bell-shape effect of GluN2B-NMDAR downregulation in the presence of the elevated AICD concentration for high-frequency stimulation and LTP rescue by the optimal blockade of GluN2B-NMDAR (Pousinha et al., Cell Reports 2017). We qualitatively reproduced experimental findings that GluN2B-NMDAR downregulation restores LTP in oligomeric Aβ conditions. Increased AICD levels enhanced intracellular Ca2+ concentration via GluN2B-NMDAR and L-type Ca2+ channels, which activated the nearby Ca2+-dependent K+ channels and led to hypoexcitability of a CA1 pyramidal neuron. Oligomeric forms of Aβ caused synapse loss and prevented CaMKII activation via Glu2NB-NMDAR-dependent pathway. Optimal blockade of GluN2B-NMDAR normalized Ca2+-dependent K+ channel functioning, restored excitability and allowed CaMKII to be phosphorylated to induce LTP. Our modeling results show that LTP induction mechanisms at CA1 pyramidal neuron synapses can be restored by optimal partial blockade of synaptic NMDA GluN2B channel conductance in AICD, Aβ, and AICD & Aβ conditions (Fig. 2-3). Partial GluN2B-NMDAR blockade did not rescue LTP in Aη conditions.
Figure 3. Optimal blockade of GluN2B-NMDAR restores synaptic functionality. (A) Somatic EPSP change for LTP induction protocol. GluN2B-NMDAR synaptic conductance was increased by a factor of 4 in AICD conditions and AICD conditions & Aβ conditions. Left to right columns: (Control) Somatic EPSP change is 183 % (grey bar); (AICD) Somatic EPSP change is 94 % (red bar); (AICD restored) Active GluN2B-NMDAR proportion is 1 of its elevated value of 4, i.e. the active fraction is 0.25; somatic EPSP change is 152 % (light red bar); (Aβ) Somatic EPSP change is 105 % (blue bar); (Aβ restored) Active GluN2B-NMDAR proportion is 0.6; somatic EPSP change is 144 % (light blue bar); (AICD & Aβ) Somatic EPSP change is 60 % (violet bar); (AICD & Aβ restored) Active GluN2B-NMDAR proportion is 0.72 of its elevated value of 4, i.e. the active fraction is 0.18; somatic EPSP change is 122 % (pink bar). (B) Somatic EPSP change for LTD induction protocol. Left to right columns: (Control) Somatic EPSP change is 40 % (grey bar); (AICD) Somatic EPSP change is 42 % (red bar); (AICD restored) Active GluN2B-NMDAR proportion is 1 of its elevated value of 4, i.e. the active fraction is 0.25; somatic EPSP change is 45 % (light red bar); (Aβ Somatic EPSP change is 36 % (blue bar); (Aβ restored) Active
GluN2B-NMDAR proportion is 0.6; somatic EPSP change is 36 % (light blue bar); (AICD & Aβ) Somatic EPSP change is 39 % (violet bar); (AICD & Aβ restored) Active GluN2B-NMDAR proportion is 0.72 of its elevated value of 4, i.e. the active fraction is 0.18; somatic EPSP change is 40 % (pink bar). From (Dainauskas et al., Front. Comput. Neurosci., 2023).
Full-scale model of the hippocampus CA1 area under AD peptide conditions
We use a full-scale model of the hippocampus CA1 the Bianchi et al. 2014 network (Figure 4) integrating the synaptic plasticity models developed in for the CA3-CA1 synapses. The synapses undergo LTP resulting in a high quality recall, and in AD conditions, LTP is prevented, synaptic weight decreases and recall of the pattern is impaired.
Figure 4. Diagram of the hippocampal CA1 microcircuit A). Schematic representation of cell types and their connectivity; arrows and small ovals represent excitatory and inhibitory connections, respectively; EC: entorhinal cortex input; CA3: Schaffer collateral input; AA: axo-axonic cell; B: basket cell; BS: bistratified cell; OLM: oriens lacunosum-moleculare cell; SEP: Septal GABA input; active CA3 inputs are represented by a red outline.
Implementation of a full-scale model of the hippocampus CA1 area under control conditions
We have contributed to implementing the first scaffold model (soma positions and connectivity) of a human hippocampus CA1 area (Gandolfi et al, Nature Comput Sci., 2023a) under control conditions (Fig. 5).
Figure 5. 3D positioning of the excitatory (PCs, pink) and inhibitory neurons. Interneurons are divided into seven classes according to positioning and morphological features (Gandolfi et al, Nature Comput Sci., 2023a).
