Selective inhibition in CA3: A mechanism for stable pattern completion through heterosynaptic plasticity

Engrams corresponding to distinct memories compete for retrieval in the CA3 region of the hippocampus, yet the detailed mechanisms underlying their formation remain elusive. Recent findings indicate that hippocampal inhibitory neurons display feature-selective firing patterns and diverse forms of synaptic plasticity, suggesting a crucial role in engram formation. Conventional CA3 attractor network models typically employ global inhibition, where inhibitory neurons uniformly suppress the activity of excitatory neurons. However, such models fail to capture the dynamics arising from sparse distributed coding or reflect inhibitory neurons' roles in the competition between engrams during memory retrieval. We propose a mechanism for engram formation in CA3 using a spiking neural network model that emphasizes heterosynaptic plasticity at excitatory-to-inhibitory (E-to-I) synapses. In our model, inhibitory neurons associate with specific neural assemblies during encoding and selectively inhibit competing engrams during retrieval. Driven by a simplified feedforward dentate gyrus (DG), this mechanism generates sparse, distributed engrams in CA3. This representation allows us to examine the effects of selective inhibition on pattern completion across various conditions, including partially overlapping engrams. Simulations show that selective inhibition substantially enhances recall stability and accuracy compared to global inhibition alone. Furthermore, emergent activity patterns across DG, CA3, and CA1 of the model replicate experimental signatures of pattern separation and completion. These results suggest that assembly-specific inhibition mediated by heterosynaptic plasticity could provide a parsimonious mechanism for engram formation and competition in CA3, offering testable predictions for future experiments.