Hi all,
HTM theory suggests a cortical microcircuit with some differences from the canonical microcircuit. If people would be able to share some of the paper references for the HTM information pathways, it would be greatly appreciated. Specifically, the recent suggestions that the main FF pathway up the hierarchy is from layer V and not III, and that the information for the coordinate transform is long range connections into layer VI (from where, laterals in layer VI?). Yes there are pathways diagrammed in the Thomson papers and the HBP database, but have targeted studies suggested/confirmed these hypotheses? Thanks in advance!
Hi Alex,
Regards to L5 being the main feed forward pathway going up the cortical hierarchy: This idea originated with Murray Sherman at the U Chicago. He has been promoting it for many years. He has written multiple papers on this topic and it is featured heavily in his most recent book. I know I read at least one other paper from another lab that specifically tested this hypothesis with a positive conclusion. Unfortunately I canât recall the authors. I suspect with a bit of literature searching you can find some relevant papers. You can always email Murray and ask him the state of support.
Everyone agrees that both L2/3 and L5a project to the next region. The question is what are their relative roles. Sherman argues that the L5a pathway, which goes through relay cells in the thalamus, is the main âdriverâ pathway. Although the number of axons is numerically smaller than L2/3, those axons are large and form large synapses on the recipient layers L4 and L6a. Hence they can make the recipient neurons fire. The axons from L2/3 are numerically greater but form smaller synapses that are not proximal, therefore they play a âmodulatoryâ role. Another supporting piece of evidence is that primary sensory regions only get input via the thalamus; there is no equivalent of a feed forward L2/3 path. If the primary driver input to primary sensory regions is via the thalamus then it would be very odd if that wasnât true of higher regions which exhibit the exact same pathway and anatomy. Another piece of evidence is that the L5a/thalamic neurons seem to only target the higher region, whereas L2/3 cells project many places, not just to the next higher region.
We did some literature searches on the question of long range connections into L6 between âwhatâ and âwhereâ regions. Thomson stated these exist in her review paper. I read the papers she referenced and a bunch of others. The anatomical evidence was not conclusive. There seems to be a general level of agreement that what and where (motor and sensory) regions are highly interconnected. There seems to be evidence that most of these connections arrive in the lower layers. But I couldnât find out which L6 cells receive this input. I also found evidence that these connections originated in multiple layers on the âwhereâ/motor side, which somewhat contradicts what Thomson said. As is often the case, the different studies used different animals and protocols. This makes it extremely difficult to reach definitive answers.
I am sorry that I canât cite specific papers, but as you know I have no memory for authors. It isnât hard to find these papers with a bit of searching on Google Scholar or elsewhere.
In neuroscience circles, the paper by Constantinople and Bruno [1] has been very influential in lending credence to Shermanâs hypothesis that L5 is a primary FF pathway. Through their experiments, they argue convincingly that L2/3 cannot be the primary pathway. The L5/6 pathway is at least equally important. They show, for example, that L2/3 cannot be driving L5/6, which is the classical assumption. Their detailed timing studies show that L5/6 might actually become active slightly before L2/3. A quote from their paper:
L5 neurons make up a major output of the cortex, as they have the most substantial axonal innervation of subcortical and cortical structures.
Larkum has also argued for a while that the L5 pathway is highly significant for complex cortical processing [2]. At a conference in September he showed very convincing evidence that selectively disabling L5 cells in sensory areas prevented decisions and/or motor responses controlled by higher level and motor areas. This was convincing (to me at least) that L5 is required for higher level processing.
At some level arguing that these pathways âprimaryâ vs âsecondaryâ is purely a matter of semantics. Both pathways play important roles. What is clear though is that the L5/6 pathway is at least equal in importance to the L2/3 pathway. This evidence completely contradicts the classical âone pure feed-forward hierarchical pathwayâ model of recognition. As far as I know there are no published computational models that explain the role of L2/3 in combination with L5/6.
Regarding the suggestion that L5 projects FF info and not L3, Jeffâs succinct summary of the evidence + the Constantinople and Bruno paper make a strong case. If I reach out to Sherman Iâll be sure to share his thoughts here.
Itâll be interesting to see the specifics of this Larkum study; I donât think itâs any of his publications here.
âprimaryâ vs âsecondaryâ is purely a matter of semantics
This is a very good point. So much of the literature tries to fit observed connectivity into concise pathways because we favor simplicity, a bit too much perhaps. Similarly, Thomson has an interesting note on avoiding the terms âfeedforwardâ and âfeedbackâ because theyâre ânonsensicalâ â the connectivity implies more complex circuits than one-way processing â but here I am preaching to the choir
On that page, I believe it is the first one, authored by Naoya:
Naoya Takahashi, Thomas Oertner, Peter Hegemann, Matthew E. Larkum. (2016) Perceptual Detection depends on Dendritic Mechanism in Cortical Pyramidal Cells. Science; In press
Looks like it got accepted to Science - cool! It should be coming out soonâŚ