Abstract |
Hippocampus is engaged in memory processes, like episodic and spatial memory. Hippocampal
Dentate Gyrus (DG) is one of the two regions where adult neurogenesis occurs in mammals, and
has been suggested to underlie pattern separation, i.e., the ability to formulate distinct memories of
similar episodes. Principal neurons of the DG, granule cells (GCs), are considered to perform
pattern separation through sparsifying and orthogonalizing their inputs. We investigate the role of
newborn GCs in pattern separation using a simple computational, yet, biophysically relevant,
spiking neural network. The DG network consists of 2,000 GCs (1,800 developmentally-born GCs
(dbGCs >8 weeks-old), 100 mature adult-born (mab) GCs (6-8 weeks-old) and 100 immature (iab)
GCs (4 weeks-old)), 100 GABAergic basket cells, 80 glutamatergic mossy cells, and 40 HIPP
interneurons. Each neuronal type is simulated as a point neuron, using the adaptive exponential
integrate-and-fire (AdEx) model. GCs are simulated as multicompartmental point neurons,
consisting of a somatic compartment connected with 12- (dbGCs) or 3-dendrites (mabGCs and
iabGCs). Five different networks were used: two control networks A,B (1900 dbGCs, 50 mabGCs,
50 iabGCs and 1800 dbGCs, 100 mabGCs, 100 iabGCs for networks A,B respectively), a network
C with equal percentages of each GC subpopulation (33.3%), one network D with 50% dbGCs,
25% mabGCs and 25% iabGCs and a network Ε without adult neurogenesis (2000 dbGCs).
Moreover, we simulated two additional networks; network B without abGC-BC synapses that lead
to over-excitation of abGC population (network F) and network B without abGC-MC synapses
(network G) that did not lead to over-excitation. Study’s results showed that GC activity was
highest in the network with the highest percentage of abGCs (66% abGCs) populations (mean ± std:
3.39 ± 0.67), followed by the 50-50% network (2.97 ± 0.61), which was in turn higher than the
control networks (1.38 ± 0.41 & 1.57 ± 0.38 for networks A,B respectively). Complete lack of adult
neurogenesis resulted in a network with the lowest GC population activity. These simulations
indicate that as the population of abGCs grows, while keeping the total GC population the same, the
excitability of the DG network increases. This is because abGCs are more active than the overall
GC population, irrespectively of the network’s composition. Another set of simulations examined
DG network’s capacity of performing pattern separation in the above networks for EC Layer II
inputs that shared a degree of similarity (60%, 70%, 80% or 90%). The results indicated that the f1
scores of output patterns were decreased as the pattern separation task became more and more
difficult and that conclusion was valid for DG networks C,D,E. Hence, we deduced that the
presence of abGCs seems to aim pattern separation efficiency for easy tasks (f1(input) = 0.4, 0.3) but
does not contribute significantly for more complex tasks (f1(input) = 0.2, 0.1). Networks F,G exhibit
pattern separation but not better than control network B.
|