If you’re at all interested in long-term potentiation (LTP) and long-term depression (LTD), you’re going to want to read this review/opinion by Govindarajan, Kelleher, and Tonegawa in July’s Nature Reviews Neuroscience. This team has previously written one of the best comprehensive reviews of neuronal translation control mechanisms. They do a brief reprise on that business, broadly suggesting which signaling molecules are involved and emphasizing that protein synthesis happens local to the stimulated synapses. Then they turn their attention to synaptic tagging/capture experiments for the bulk of the paper.
As some here must be sick of hearing by now, memories are likely to be stored as changes in synaptic weights. LTP and LTD are increases and decreases respectively in synaptic weight. The long-lasting (late) phases of LTP and LTD require protein synthesis. Govindarajan et al make the case that this is achieved by increased efficiency in general translation machinery. Synaptic tagging and cross-tagging experiments have shown that the proteins generated by a given L-LTP or L-LTD inducing stimulus can be captured by synapses that have only received E(early)-LTP or E-LTD inducing stimuli, thus producing late phase plasticity.
Govindarajan et al note that transport speeds are known for important plasticity-related proteins, like AMPA receptor subunits, and that it would take them about an hour to move the length of your average dendritic branch. In tagging experiments, if L-LTP and E-LTP stimuli are separated by around this length of time, the capture of plasticity proteins does not occur. So they are suggesting that plasticity proteins are probably only shared within a dendritic branch, producing ‘clusters’ of synapses that have undergone long-term change. I’m not sure the transport speed issue is central to their theory or if you could simply argue that the generated proteins are degraded in that hour, but one prediction they didn’t make that I think follows is that in cell populations with shorter dendritic branches you should see greater spread of these induced proteins or maybe a longer time window for capture, and vice versa.
The clustered plasticity model carries advantages over a dispersed model where synapses are changed at more random sites across the neuron in part because stimulated synapses in the same dendritic branch sum supralinearly. The whole of the stimulation is greater than the sum of its parts. To get cells to fire in the dispersed model would take a lot more network activation than that in the clustered model. The authors argue that this has implications for ease of memory recall.
I have an interest in trying to understand the relationship between broad-scale network activities like theta oscillations and sharp-wave ripples and the small scale on which plasticity is studied. There is a little bit about that in this review.
…assuming that connectivity between the set of presynaptic neurons and postsynaptic neurons is random, clustered plasticity would be advantageous compared with dispersed plasticity only if the density of active inputs is high enough to enable the setting of potentiation and depression tags at multiple synapses within at least one dendritic branch in the postsynaptic neuron. In support of this, 30-50% of hippocampal cells are active in a given environment, hippocampal activity resembles theta-burst stimulation, which has been used as a robust inducer of plasticity, and 45-75% of synapses are capable of undergoing plasticity. Therefore, in an episode (a sequence of related events; the hippocampus is important for acquiring memory of such sequences), it is probable that there is sufficient activation so as to result in many dendritic branches in the hippocampus containing multiple tagged synapses. The probability of this being the case is even higher when sharp waves are considered. In rats, it has recently been shown that sharp wave activity during exploration carries information about the environment explored by the animal. Furthermore, sharp wave-type activity would lead to high enough activity to enable activation of multiple synapses in a dendritic branch.
I don’t think this quite does it, but at least folks are starting to turn their attention to it. It’s weird to try to make actual predictions about memory based on this hour timeframe of integration. Would a given dendritic branch represent events in about a two-hour window with a peak in the very middle? When a salient or unexpected event occurs that might drive neuromodulation and increase protein synthesis occurs, do events an hour on either side of that event benefit from the upregulation? How much time worth of events is integrated in a single sharp-wave ripple or compressed in a single theta oscillation? All I know now is about distances that are integrated and it seems like it is not an hour’s of exploration.