Figure 12. Simulation of learning-dependent plasticity with and without Purkinje cell (PC) feedback.

(a) Net changes in the strength of the 12,000 synapses between granule cells and PCs in a simulation with PC feedback to PC-MLIs and trained at an interstimulus interval (ISI) of 500ms. Negative changes indicate net LTD (left), while positive changes indicate net LTP (right). The vertical scale of the histogram is cropped at 60 synapses because the vast majority of synapses showed little or no change. The synapses that underwent net LTD are most likely to support expression of the conditioned responses in the simulations. (b) Same analysis as in a, for a simulation lacking PC feedback to PC-MLIs. (c) Superimposition of a and b indicates a larger number of synapses underwent robust LTD in the absence of PC feedback. (d) Difference between a (with feedback) and b (without feedback). The simulation lacking PC feedback to PC-MLIs had a larger number of synapses undergoing robust LTD and fewer synapses undergoing a small degree of net LTD. (e) To compare the differences between simulations with and without PC feedback across several ISIs, difference plots such as the one shown in d were converted to a cumulative histogram. Across the three ISIs shown, the simulations lacking PC feedback collaterals had more synapses undergoing robust LTD than those with feedback. This was true for ISIs where the simulations learned approximately equally well (ISI 500 in blue) and for ISIs where simulations with collaterals learned much better (ISIs 150 and 800). MLIs, molecular layer interneurons.