L5tt bursting seems to be a location-related stimulus detection signal, creating awareness via projections to superior colliculus, thalamus, and striatum. These papers could point towards a mechanism for computing displacements.
Context in HTM Theory: Displacements
Layer 5 thick-tufted cells are the cortex’s only projection to many subcortical structures and the spinal cord. They produce behavior, yet sensory input directly triggers strong sensory responses. Naively, that would constantly cause knee-jerk reactions to every stimulus. Something must be stopping that, but what’s the point? L5 doesn’t make sense unless it does something combining sensory and motor.
HTM’s idea of displacements seem like a good explanation for L5. Displacements let the cortex keep track of the sensor’s location in reference frames, and they indicate how to move from A to B.
Displacements also break the sensory/motor binary by attention. When you shift attention between locations, there’s a displacement. Stimulus detection seems like an attentional shift, except without necessarily attending anything beforehand. If layer 5 identifies displacements, it has to start somewhere.
I’ll summarize some papers which point towards aspects of cortical circuitry involved in displacements. One series of papers basically shows L5 cells burst upon perceptual detection. The other series of papers basically shows bursting is selective for something like location. In both cases, there’s just one sensory stimulus, so there are no displacements.
All of these studies are in mouse barrel cortex. That’s the part of primary somatosensory cortex for whiskers. Mice have large mobile whiskers which contact objects, generating forces at the base of the whiskers which they sense. They have many whiskers, but these studies trim all but one. For each whisker, there’s a corresponding cortical column.
Here, bursting means something different than minicolumn bursting. It’s two or more spikes in rapid succession. When a cell receives proximal input and fires a spike, it can make that a burst by quickly firing again and potentially several more times. It does that if it receives enough input on its distal apical dendrite. So a burst indicates coincident input from proximal and distal apical dendrites. Bursts have different downstream effects than single spikes.
I don’t know much about electrophysiology, so that’s probably a gross oversimplification. Also, these studies mostly detected the associated calcium event in the apical dendrite, which aren’t completely correlated with bursts.
Bursts Selective for Whisker Angle
In a task where mice look for a pole using their whiskers, touch can cause S1 L5 cells to burst. Each cell’s bursting prefers a certain whisker angle. The touch-induced bursting requires distal apical input from M1, which is probably responsible for the angle selectivity.
Nonlinear dendritic integration of sensory and motor input during an active sensing task
The mouse’s head is fixed in place and a pole is positioned at one of four places nearby. The mouse is trained to lick in response to the pole at any of those positions, but not if the pole is at a fifth location. The mice usually didn’t touch the pole at the fifth location, so the results only include trials where it touches the pole at the first four positions. Although the lick is a direct response to whisker contact, they have a couple seconds where they’ll be rewarded for licking, much longer than the immediate neural response. The mice only touch the pole with one whisker because all others are trimmed, or because all but one row are trimmed.
They looked at layer 5 pyramidal cells. There are one or two types of those besides L5 thick-tufted cells. They didn’t exclusively look at cells in the whisker’s cortical column, but mostly there, because they selected calcium imaging regions which had a response to the whisker.
Calcium signals (the kind associated with burst firing) often occurred in the distal apical dendrite directly after contact with the pole, with short latency. Each cell responded strongest to one of the pole locations, and less strongly for poles located further from that one.
They hypothesize that the location selectivity results from axons going from primary motor cortex to L1. When M1 was silenced, the calcium signals mostly disappeared (the mice were still able to perform the task although more poorly). Their reference 26 (“Activity in motor–sensory projections reveals distributed coding in somatosensation”) says some of those axons correlate with object location, many with activity persisting for seconds. That could be involved in integrating information over time for multi-touch object recognition, perhaps specifically computing displacements.
Active dendritic integration and mixed neocortical network representations during an adaptive sensing behavior
This study follows up the prior one. I’ll just list some things it adds.
- Each cell bursts more around a particular whisker angle.
- Normally, the mice learned where to look for the pole if it suddenly started appearing at a particular location. They didn’t adapt as much if apical tufts were silenced.
- Silencing apical dendrites didn’t impair object detection, unlike the results in the next series of studies I’ll summarize. However, those ones use weak stimuli at the threshold of detectability and have other differences.
Bursts for Perceptual Detection
These studies used stimuli right at the threshold of detectability, comparable to trying to hear a very quiet sound. L5tt cells burst upon stimulus detection, whereas L5st cells are less involved but seem to burst in relation to reward.
L5tt bursting triggers detection via subcortical influence. Silencing its output to the superior colliculus, thalamus, or striatum all impaired detection, especially the first two.
Active cortical dendrites modulate perception
Google Scholar and click the pdf link to the right.
They stimulated a whisker with a magnet, with a stimulus strength right at the threshold of detectability, and mice were trained to lick in response to indicate detection. The authors examined layer 5 pyramidal cells in the whisker’s corresponding cortical column.
A subset of cells burst more often when the mouse detects the stimulus. Silencing the calcium events shifted the detection threshold higher, whereas activating the distal apical dendrites reduced that threshold.
Cited studies suggest calcium events or bursting are related to conscious perception, which the results support.
Active dendritic currents gate descending cortical outputs in perception
This is a follow up study. It uses the same experiment, but now they differentiate two subclasses of L5 cell.
They use different names but they’re basically the same as L5 thick-tufted (L5tt) and L5 slender-tufted (L5st). There are many papers about them. L5tt cells project to various subcortical structures and in some regions they project to the spinal cord. L5st cells project within the cortex and to the striatum. There might also be a third type which just projects within cortex (“Three Types of Cortical Layer 5 Neurons That Differ in Brain-wide Connectivity and Function”).
About half of L5tt cells had reliable apical calcium events (so bursts) upon stimulus detection. To determine whether that’s caused by the reward or licking behavior rather than stimulus detection, they checked what happens whey they gave the mice water droplets at random intervals. Some L5tt cells had calcium events associated with the droplets, but a lot fewer.
In contrast to L5tt cells, L5st cells generally didn’t respond with calcium events. About one-tenth had more calcium events, and another tenth had fewer. Importantly, the increased calcium events usually didn’t directly follow from stimulus detection. Instead, they usually happened after the behavioral response, peaking during the reward. That’s interesting because L5st cells project to the striatum. Perhaps their mixed increased or decreased occurrence of calcium events relates to the basal ganglia’s direct and indirect pathways.
Unlike L5st cells, activating the apical dendrites of L5tt cells reduced the stimulus detection threshold. Likewise, inactivating L5tt cells reduced it.
They hypothesized that L5tt cells influence stimulus detection threshold via their projections to subcortical structures. To do so, they silenced their synaptic outputs within those structures. Silencing their outputs in superior colliculus strongly impaired their detection performance. The same goes for POm, a higher order thalamic nucleus. Silencing their output to striatum had an effect, but a smaller one. For pons and medulla, there was no effect on detection performance.
They also tested whether behavioral context is relevant. To do so, they removed the reward, and the mice didn’t need to respond to the stimulus by licking. The mice just sat there during the sensory stimulus. In these trials, L5tt cells had far less calcium response. That means L5tt cells burst upon stimulus detection either because of the behavioral context, or because of paying attention. Because there was no way to tell whether the mouse noticed the stimulus, and the stimulus was right at the threshold of detectability, I think the mice didn’t pay attention and never noticed it.