Constraining the function of CA1 in associative memory models of the hippocampus
CA1 is the main source of afferents from the hippocampus, but the function of CA1 and its perforant path (PP) input remains unclear. In this thesis, Marr’s model of the hippocampus is used to investigate previously hypothesized functions, and also to investigate some of Marr’s unexplored theoretical ideas. The last part of the thesis explains the excitatory responses to PP activity in vivo, despite inhibitory responses in vitro. Quantitative support for the idea of CA1 as a relay of information from CA3 to the neocortex and subiculum is provided by constraining Marr’s model to experimental data. Using the same approach, the much smaller capacity of the PP input by comparison implies it is not a one-shot learning network. In turn, it is argued that the entorhinal-CA1 connections cannot operate as a short-term memory network through reverberating activity. The PP input to CA1 has been hypothesized to control the activity of CA1 pyramidal cells. Marr suggested an algorithm for self-organising the output activity during pattern storage. Analytic calculations show a greater capacity for self-organised patterns than random patterns for low connectivities and high loads, confirmed in simulations over a broader parameter range. This superior performance is maintained in the absence of complex thresholding mechanisms, normally required to maintain performance levels in the sparsely connected networks. These results provide computational motivation for CA3 to establish patterns of CA1 activity without involvement from the PP input. The recent report of CA1 place cell activity with CA3 lesioned (Brun et al., 2002. Science, 296(5576):2243-6) is investigated using an integrate-and-fire neuron model of the entorhinal-CA1 network. CA1 place field activity is learnt, despite a completely inhibitory response to the stimulation of entorhinal afferents. In the model, this is achieved using N-methyl-D-asparate receptors to mediate a significant proportion of the excitatory response. Place field learning occurs over a broad parameter space. It is proposed that differences between similar contexts are slowly learnt in the PP and as a result are amplified in CA1. This would provide improved spatial memory in similar but different contexts.