P024 Analysis of autoassociative dynamics in the hippocampus through a full-scale CA3-CA1 model
Giulia M. Boiani1,2*, Serena Giberti2, Lorenzo Tartarini3, Giampiero Bardella4, Sergio Solinas5, Stefano Ferraina4, Michele Migliore2, Jonathan Mapelli3, Daniela Gandolfi1
1Dipartimento di Ingegneria "Enzo Ferrari", Università degli Studi di Modena e Reggio Emilia, Italy
2CNR, Istituto di Biofisica, Palermo, Italy
3Dipartimento di Scienze Biomediche, Metaboliche e Neuroscienze, Università degli Studi di Modena e Reggio Emilia, Italy
4Dipartimento di Fisiologia e Farmacologia, Sapienza, Università di Roma, Roma, Italy
5Dipartimento di Scienze Biomediche, Università di Sassari, Sassari, Italy
*Email: giuliamaria.boiani@unimore.it
IntroductionThe hippocampus is a key brain structure for memory formation and spatial navigation. We present a full-scale point-neuron realistic model of the mouse CA3-CA1 network [1]. The structural validity of the network has been assessed by applying graph theory, whereas functional validation has been performed by incorporating a parameterized point-neuron and a custom-developed synapse with short- and long-term synaptic plasticity. We demonstrated the ability of the modeled CA3 to operate as an autoassociative network that can reconstruct complete memories from partial clues [2]. These results confirm the role of CA3 in pattern completion and provide a benchmark to investigate information processing in the hippocampal formation.Methods
The network connectivity has been obtained by adopting a morpho-anatomical strategy based i) on the intersection of abstract geometrical morphologies and ii) by extending the tubular structures of CA3 pyramidal cell (PC) axons to target CA1 PCs while accommodating the hippocampal anatomy [3]. The custom-developed synapse was implemented through NESTML and included short-term dynamics and long-term STDP[4]. The autoassociativity was investigated by applying a theta-gamma stimulation protocol to train a subset of 400 out of 4000 CA3 PCs. The network’s tendency to balance local interconnectedness (clustering) and efficient information routing (short path lengths) was assessed using a key graph-theoretic metric: the small-world coefficient.
Results
The CA3-CA1 network (Fig 1A-B) showed an outdegree (Fig 1C) distribution compatible with experiments. Interestingly, CA1 and CA3 exhibited distinct connectivity profiles: a hub-like organization potentially facilitating the integration of information in the CA1, and a nearly fully connected hub-less architecture in the CA3 consistent with its role in pattern completion. Moreover,CA3 showed a high clustering coefficient (Fig 1D), while both regions exhibited small-world properties, with CA3 having a higher value (Fig 1E).Autoassociativity test showed that CA3 (Fig 1F) can indeed retrieve complete memories upon presentation of degraded inputs and complete retrieval occurred when at least 20% of trained neurons were stimulated (Fig 1G).
Discussion
These results validate the accuracy of the model. The network can perform pattern completion effectively exploiting autoassociativity. The analysis of the network's topology suggests that CA1 acts as a hub-like connector, while CA3 shows signatures of small-worldness with an efficient architecture balancing local segregation and global reach. Our biologically realistic network exhibits a non-trivial topology allowing the emergence of functional properties, which could be altered in pathological conditions together with topology [5]. These results offer insights into the functions of hippocampal circuitry, paving the way to the use of computational models to investigate physiological and pathological conditions.
Figure 1. A CA3-CA1 scaffold B Simulation activity snapshots C CA3 and CA1 outdegree distributions. D Clustering coefficients of CA3 and CA1 networks compared to equivalent Erdős–Rényi (ER) and Watts–Strogatz (SW) null models. E Small-World Coefficients. F Neuronal activation over time in recall tests with varying fractions of stimulated neurons. G Evaluation of recall performance.
Acknowledgements
The University of Modena and Reggio Emilia FAR-DIP-2024-GANDOLFI E93C24000500005 to DG. The Italian National Recovery and Resilience Plan (NRRP), M4C2, funded by the European Union – NextGenerationEU (Project IR0000011, CUP B51E22000150006, “EBRAINS-Italy” and Project PE0000006, “MNESYS” to JM), the Ministry of University and Research, PRIN 2022ZY5RXB CUP E53D2301172 to JM.
References
[1]https://doi.org/10.1038/s41598-022-18024-y
[2]https://doi.org/10.1007/s10827-024-00881-3
[3]https://doi.org/10.1038/s43588-023-00417-2
[4]https://doi.org/10.1038/ncomms11552
[5]https://doi.org/10.1016/j.clinph.2006.12.002
