2
N. Andreakos et al.
In 2010 a much more flexible model was introduced that controlled for itself the storage and recall of patterns of information arriving at unpredictable rates [1]. The model
was based upon the many details were then known about the neuronal hippocampal circuit [21, 22]. The model explored the functional roles of somatic, axonic and dendritic
inhibition in the encoding and retrieval of memories in region CA1. It showed how theta
modulated inhibition separated encoding and retrieval of memories in the hippocampus into two functionally independent processes. It predicted that somatic (basket cell)
inhibition allowed generation of dendritic calcium spikes that promoted synaptic longterm plasticity (LTP), while minimizing cell output. Proximal dendritic (bistratified cell
(BSC)) inhibition controlled both cell output and suppressed dendritic calcium spikes,
thus preventing LTP, whereas distal dendritic (OLM cell) inhibition removed interference from (new and old) memory patterns trying to be encoded during recall. The mean
recall quality of the model was tested as function of memory patterns stored. Recall
dropped as more patterns were stored due to interference between previously stored
memories. Proximal dendritic inhibition was held constant as the number of memory
patterns stored was increased.
Here, we more systematically investigated the mechanisms to improve the recall
performance of [1]. In particular, we examined how selective modulation of feedforward/feedback excitatory/inhibitory pathways targeting inhibitory and excitatory cells
may influence the thresholding ability of dendritic inhibition to remove at the network
level spurious activities, which may otherwise impair the recall performance of the network, and improve its mean recall quality as more and more overlapping memories were
stored.
2 Materials and Methods
2.1 Neural Network Model
Figure 1 depicts the simulated neural network model of region CA1 of the hippocampus.
The model consisted of 100 pyramidal cells (PC), 1 axo-axonic cell (AAC), 2 basket cells
(BC), 1 BSC and 1 OLM cell. The neuronal dynamics of the model cells with respect
to a theta rhythm is depicted in Fig. 2. Model cells were simplified compartmental
models with complex ion channel dynamics. Simplified morphologies including the
soma, apical and basal dendrites and a portion of the axon were used for each cell
type. The biophysical properties of each cell were adapted from cell types reported
in the literature, which were extensively validated against experimental data in [4–7,
13]. In the model, AMPA, NMDA, GABA-A and GABA-B synapses were considered.
GABA-A were present in all strata, whereas GABA-B were present in medium and distal
SR and SLM dendrites. AMPA synapses were present in SLM (EC connections) and
SR (CA3 connections), whereas NMDA were present only in SR (CA3 connections).
The complete mathematical formalism of the model has been described elsewhere [1].
Schematic representations of model cells can be found in [12]. The dimensions of the
somatic, axonic and dendritic compartments of model cells, the parameters of all passive
and active ionic conductances, synaptic waveforms and synaptic conductances can be
found in [12]. All simulations were performed using NEURON [8] running on a PC with
eight CPUs under Windows 10.
N. Andreakos et al.
In 2010 a much more flexible model was introduced that controlled for itself the storage and recall of patterns of information arriving at unpredictable rates [1]. The model
was based upon the many details were then known about the neuronal hippocampal circuit [21, 22]. The model explored the functional roles of somatic, axonic and dendritic
inhibition in the encoding and retrieval of memories in region CA1. It showed how theta
modulated inhibition separated encoding and retrieval of memories in the hippocampus into two functionally independent processes. It predicted that somatic (basket cell)
inhibition allowed generation of dendritic calcium spikes that promoted synaptic longterm plasticity (LTP), while minimizing cell output. Proximal dendritic (bistratified cell
(BSC)) inhibition controlled both cell output and suppressed dendritic calcium spikes,
thus preventing LTP, whereas distal dendritic (OLM cell) inhibition removed interference from (new and old) memory patterns trying to be encoded during recall. The mean
recall quality of the model was tested as function of memory patterns stored. Recall
dropped as more patterns were stored due to interference between previously stored
memories. Proximal dendritic inhibition was held constant as the number of memory
patterns stored was increased.
Here, we more systematically investigated the mechanisms to improve the recall
performance of [1]. In particular, we examined how selective modulation of feedforward/feedback excitatory/inhibitory pathways targeting inhibitory and excitatory cells
may influence the thresholding ability of dendritic inhibition to remove at the network
level spurious activities, which may otherwise impair the recall performance of the network, and improve its mean recall quality as more and more overlapping memories were
stored.
2 Materials and Methods
2.1 Neural Network Model
Figure 1 depicts the simulated neural network model of region CA1 of the hippocampus.
The model consisted of 100 pyramidal cells (PC), 1 axo-axonic cell (AAC), 2 basket cells
(BC), 1 BSC and 1 OLM cell. The neuronal dynamics of the model cells with respect
to a theta rhythm is depicted in Fig. 2. Model cells were simplified compartmental
models with complex ion channel dynamics. Simplified morphologies including the
soma, apical and basal dendrites and a portion of the axon were used for each cell
type. The biophysical properties of each cell were adapted from cell types reported
in the literature, which were extensively validated against experimental data in [4–7,
13]. In the model, AMPA, NMDA, GABA-A and GABA-B synapses were considered.
GABA-A were present in all strata, whereas GABA-B were present in medium and distal
SR and SLM dendrites. AMPA synapses were present in SLM (EC connections) and
SR (CA3 connections), whereas NMDA were present only in SR (CA3 connections).
The complete mathematical formalism of the model has been described elsewhere [1].
Schematic representations of model cells can be found in [12]. The dimensions of the
somatic, axonic and dendritic compartments of model cells, the parameters of all passive
and active ionic conductances, synaptic waveforms and synaptic conductances can be
found in [12]. All simulations were performed using NEURON [8] running on a PC with
eight CPUs under Windows 10.