kaleidoscopik

Pyramidal cells are a special class of neurons that feature prominently in neocortical and hippocampal circuits. They are characterized by their elaborate dendritic structure. The apical dendrite grows like a shoot out of the top of the cell body with multiple perpendicular offshoots sort of like a pine tree. Unlike a pine tree, the apical dendrite often gets bushy at the top, fraying into an ‘apical tuft’ of split branches. On the other end, the basal dendrites extend from the sides and bottom of the soma and are much shorter than the apical dendrite. Below is an idealized pyramidal neuron from Spruston (2008).

Dendrites are, of course, the input structure of a neuron. Thousands of other neurons connect to a given pyramidal cell along these dendritic structures and influence the cell to fire or stay quiet. Spruston’s review is a brief discussion of the features of pyramidal neurons followed by a larger collection of data pertaining to the function of pyramidal dendritic structure and processsing. That aim is a little broad, so we don’t end up with much integration. I found myself lost in numerous experimental results with dubious generality.

Nevertheless, I extracted a couple principles to carry with me in future reading.

The first principle is input segregation within dendrites. There are clear examples of this in the hippocampus and neocortex, and it means you can begin to predict things about an input based on the synapse location along the apical shaft. Anatomical distance is mirrored in apical dendrite distance. Input from neurons close by (local) will synapse closer to the cell body. Inputs from other regions of the brain connect to the distal apical tuft. I currently care more about the hippocampus than the neocortex, so I’ll point to the clear example from there.

CA1 neurons receive input to the distal tuft from the entorhinal cortex through the perforant path and from the thalamus, whereas the remainder of the dendrites receive input from CA3 through the schaffer collaterals. Furthermore, CA3 neurons that are distant from CA1 project primarily to apical dendrites, whereas CA3 neurons that are closer to CA1 project more heavily to basal dendrites. The functional significance of this arrangement remains mysterious. - Spruston (2008)

I’m really curious about how this arrangement develops. Is there some competition between inputs for space along the apical trunk? It seems obvious to propose spike-timing dependent plasticity during development to allow the CA3 input, which has a low conduction delay to fire closer to the cell body. Actually, now that I think about it though, I can’t imagine that the coincidence is ever between NT-release event 1 and an action potential caused by that NT-release event. Rather, NT-RE 1 must cause an AP that coincides with NT-RE 2. I wonder if inputs fire rhythmically at a rate that corresponds to the length of time it would take for a postsynaptic potential to reach the soma and the action potential to backpropagate back to the input site. Of course, one neurotransmitter release event at one synapse counts for nothing, but I could generalize to multiple synapses. But then the synapses would have to be organized according to their firing rhythm along the dendritic tree. That would be pretty, but there’s no evidence for anything like that that I’ve ever seen. The other outcome of this line of thinking is that rhythmic input from the further reaches of the brain should be slower to allow for the longer and more complicated post-synaptic propagation process. The truth is I need to pay more attention to the time scales on potential plots. I have no idea how long any of this stuff takes.

The second important principle is that dendrites are active elements and that their activity is compartmentalized. This allows intermediate levels of neuronal organization between the synapse and the whole cell. For instance, imagine a stretch of dendritic branch with perhaps a couple dozen synapses. Activity at a few of these synapses can reduce the driving force and allow less ion flux through synapses activated just like it. I imagine this a little like stealing someone’s bounce on a trampoline. I recognize that that analogy might not resonate with all cultures, but I grew up with yards and trampolines. The result is that synapses close together on a tree can produce less than the sum of their parts. On the other hand, they can produce more than expected by causing the firing of dendritic spikes. There are voltage-gated ion channels in dendrites (I think I read somewhere that they were clustered near branchpoints, but I can’t find that now). These can be brought to activate by synchronous or spatially clustered dendritic input and produce a large ion flux that looks something like a miniature action potential in the dendrites. Spatially clustered processing is a growing area of interest in the literature. There was a neat little discussion of clustered plasticity from Tonegawa and associates a while back and Magee just published a paper in Nature (with comments by Spruston) entitled “Compartmentalized dendritic plasticity and input feature storage in neurons.“

The idea of a dendritic branch as a processing unit inspires thoughts of the the branch as a memory unit. There is an overarching concern in the field about what the unit of memory is. The whole cell? The individual synapse? The answer theoretically puts some limit on the storage capacity of the system which we then have to square with our experience of life-long detailed records. Karim Nader mentioned this issue at the recent Day of Memory meeting at NYU. Upon reflection, I have started thinking about this problem as a continuum. The question is not ‘What is the unit of memory?’ but instead ‘What type of information is stored at each level of neuronal processing?’. The most straightforward answer is that more detailed information is stored as more refined modifications to neural function. If I think about this long enough I can imagine behavioral tasks along a range of complexities that might correspond to different plasticity requirements, but it is hard to guess what the limits should be. The most complex and detailed tasks should require the smallest unit of plasticity, but they will be difficult to train a rat to do and will almost certainly take a long time which precludes fine analysis of the dynamics of memory.

Take these home:

• Compartmentalized input and compartmentalized processing
• Active dendrites
• Intermediate units smaller than the cell and larger than the synapse