L5tt cells are attentional (paper summaries)

[Casey]

[1] High-frequency burst spiking in layer 5 thick-tufted pyramids of rat primary somatosensory cortex encodes exploratory touch

[2] Active dendritic currents gate descending cortical outputs in perception

Main Point
L5tt cells cause the brain to notice a stimulus. To do so, they burst, triggering detection via thalamus, SC, and striatum.

[1]
L5tt cells can burst upon touch.

This study just measured spiking while the rat whisked and touched an object.

Many L5tt cells were capable of responding to touch by bursting. Most of this paper addresses possible complications, so it’s fairly safe to take that result as-is. For example, they can burst upon touch regardless of their long-range projections.

Bursting might overcome the generally depressed corticothalamic synapses, causing thalamocortical cells to fire.

I’m mentioning this study because the other study is a bit ambiguous. It uses calcium imaging, for example.

[2]
Bursts are required to notice stimuli, at least very weak ones. (Or at least apical calcium events are required.)

This study measured the perceptual threshold, meaning the stimulus strength which the mouse noticed half the time. You can imagine trying to notice a very weak stimulus.

If the mouse notices the stimulus, the L5tt cells burst (or at least have an apical calcium signal closely associated with bursting). Otherwise, they don’t. Activating their apical dendrites reduces the perceptual threshold.

The L5tt cells in this study project to thalamus, superior colliculus, striatum, pons, and medulla. Silencing their synapses in thalamus or superior colliculus strongly impaired the mouse’s detection performance. The same goes for the striatum, but the effect was smaller. There was no effect for pons or medulla.

They repeated the experiment without the behavioral task or reward. This reduced the average calcium signal a lot.

That contradicts the other study, where they burst without a behavioral task or reward. However, the stimuli in this study were very weak, so I think the mice just didn’t notice them. There are other possible explanations, though.

[dmac]

Related to your reference #2 is this previous publication by some of the same authors:

Active cortical dendrites modulate perception

Naoya Takahashi, Thomas G. Oertner, Peter Hegemann, and Matthew E. Larkum, 2016
Link: https://www.science.org/doi/10.1126/science.aah6066

Abstract

There is as yet no consensus concerning the neural basis of perception and how it operates at a mechanistic level. We found that Ca2+ activity in the apical dendrites of a subset of layer 5 (L5) pyramidal neurons in primary somatosensory cortex (S1) in mice is correlated with the threshold for perceptual detection of whisker deflections. Manipulating the activity of apical dendrites shifted the perceptual threshold, demonstrating that an active dendritic mechanism is causally linked to perceptual detection.

---

And before that, Larkum wrote this review about apical dendrites, which describes how apical dendrites work at a mechanistic level (but does not explore their purpose).

A cellular mechanism for cortical associations: an organizing principle for...

[dmac]

Thalamocortical interactions

S Murray Sherman, 2012
https://doi.org/10.1016/j.conb.2012.03.005
Free version: Thalamocortical Interactions - PMC

Glutamatergic pathways dominate information processing in the brain, but these are not homogeneous. They include two distinct types: Class 1, which carries the main information for processing, and Class 2, which serves a modulatory role.

[…] the Classes are the same for cortical and thalamic circuitry. These are also the only two classes of glutamatergic input so far seen.

Figure 1 shows the phenomenon of paired-pulse depression & facilitation.
Class 1 synapses (depressing) are tuned to respond to individual spikes and to ignore burst-firing.
Class 2 synapses (facilitating) are tuned to respond to burst-firing and to ignore individual spikes.

---

At the time of publication of this review, it wasn’t possible to directly prove that apical dendrites have synapses with only class 2 presynaptic axons. However they did find that class 1 and class 2 synapses are predominantly proximal and distal (respectively), and apical dendrites are by definition also distal dendrites.

Therefore it appears that: apical dendrites are tuned to respond to burst-firing, and they can respond by causing their cell to burst-fire.

[dmac]

Cortical control of behavior and attention from an evolutionary perspective

S. Murray Sherman, and W. Martin Usrey, 2021
https://doi.org/10.1016/j.neuron.2021.06.021

Abstract

For animals to survive, they must interact with their environment, taking in sensory information and making appropriate motor responses. Early on during vertebrate evolution, this was accomplished with neural circuits located mostly within the spinal cord and brainstem. As the cerebral cortex evolved, it provided additional and powerful advantages for assessing environmental cues and guiding appropriate responses. Importantly, the cerebral cortex was added onto an already functional nervous system. Moreover, every cortical area, including areas traditionally considered sensory, provides input to the subcortical motor structures that are bottlenecks for driving action. These facts have important ramifications for cognitive aspects of motor control. Here we consider the evolution of cortical mechanisms for attention from the perspective of having to work through these subcortical bottlenecks. From this perspective, many features of attention can be explained, including the preferential engagement of some cortical areas at the cost of disengagement from others to improve appropriate behavioral responses

I highly recommend reading the whole article. It is short, informative, and novel.
Here are some interesting sections which indirectly support the claim that L5tt cells are attentional:

These links between cortex and behavior involve layer 5 corticofugal projections. The only other source of subcortical projection arises from a population of layer 6 cells: some innervate thalamus and others the claustrum, but of great importance in the context of this perspective, in no case do these innervate obvious subcortical motor centers (Sherman and Guillery, 2013; Sherman, 2016; Usrey and Sherman, 2019). It is also worth noting that cells in layers 5 and 6 that project subcortically rarely project to other cortical areas, and vice versa (Petrof et al., 2012). Some further details on differences between layer 5 and 6 corticofugal projections are provided below. It is the layer 5 cells that innervate numerous subcortical motor centers (Deschenes et al., 1994; Bourassa and Deschenes, 1995; Bourassa et al., 1995; Kita and Kita, 2012; Prasad et al., 2020) and thus appear to be the effective route whereby cortex influences behavior.

An alternative way to consider attention is from the perspective of the selection process whereby the appropriate […] circuits are brought into play to provide the best response to any environmental challenge at any given time. All of the other results described for attention can be seen as a logical and predictable consequence of this process.

And I just found this part interesting:

many aspects of sensory perception also appear to function without cortex. A classic example is the Sprague effect, in which cats without visual cortex can detect and locate novel visual stimuli on the basis of functioning of the superior colliculus (Sprague, 1966). This ability to detect stimuli strictly with subcortical circuits has also been extended to the auditory and somatosensory systems (Lomber et al., 2007; Hong et al., 2018). […] The main point here is that during pre-mammalian vertebrate evolution, brainstem centers formed the capacity for quite effective sensorimotor processing, typically reaching maximum effectiveness in evolution of circuits involving midbrain structures such as the optic tectum, which is the homolog of the mammalian superior colliculus, although older and simpler telencephalic structures also exist in primitive vertebrates such as the lamprey (Herman et al., 2018; Krauzlis et al., 2018; Gharaei et al., 2020; Basso et al., 2021). Many and perhaps all of these older circuits remain viable and in use in mammalian brains.

[thanh-binh.to]

does anybody implement L5tt in HTM.core or anyhows in C++ or Python? Thanks

[dmac]

Yes, I made a simplified model to demonstrate the concepts: A Model of Apical Dendrites

And here is a list of more detailed and realistic model of L5tt cells: ModelDB: Models that contain the Neuron : Neocortex L5/6 pyramidal GLU cell

[thanh-binh.to]

@dmac

I made a simplified model to demonstrate the concepts: A Model of Apical Dendrites

thanks. I can run your code but I am trying to understand

1. your model
2. your application results.
   Could you pls explain me your implementation, and what is expected result?
   Thanks

[thanh-binh.to]

Now I understood your L5tt, Thanks @dmac for sharing your codes!
It looks like tracking algorithm, where GNW for the new object is initialized by local SDR at object position. The attention score is based on overlapping score between GNW at a position and predictive cells.
In principle, your algorithm is comparable to the well-known grid-based object detection.

What I am thinking now: your L5tt is somehows very close to Thalamus at the folder “py/htm/advanced/algorithms” of htm.core.

[CollinsEM]

This seems to support the notion of (at least) two attention mechanisms.

1. Automatic attention: In HTM, this is typically what we are modeling through the interaction of the distal and proximal synapses. Novel (unpredicted) input patterns/sequences generate localized columnar bursting[1] behavior which in turn gates the thalamus to broadcast more of the incoming stimulus to nearby columns. The initial bursting might be considered a ‘pay-attention’ signal with the subsequent thalamic activations providing the learning context. The increased frequency of thalamic activations might then permit rapid synaptic learning of the new context via Hebbian updates.

2. Volitional Attention: Context or task specific biasing of neurons through apical dendrites and/or neuromodulators.
   a) The soup of neuromodulators released by the basal ganglia and other sub-cortical structures can enhance the response of specific types of synapses to certain types of neurotransmitters. I interpret this neuromodulatory biasing of synapses as a task-oriented signal. For example: When I’m hungry, sub-cortical structures detect a chemical/hormonal imbalance and begin emitting neuromodulators (e.g. dopamine). This, in turn, enhances the response of neurons that respond when food is present, or that remember where food might be found, or that generate goal-directed behaviors which might be associated with activities like foraging.
   b) The apical synapses respond to top-down signals from other cortical regions, thus biasing the neuron in a slightly different way than the proximal and distal inputs. Such biasing my increase the chances of the neuron firing in bursts[2] when activated via proximal inputs. My interpretation is that the apical signals provide more context-specific, top-down feedback. For example: I’m attending to an object concealed in a box. Enhance the response of neurons that are specifically sensing and/or interpreting that object. The input patterns appear to indicate the presence of something that feels like a coffee mug. Enhance the response of neurons which are usually active when we are experiencing coffee mug. Continue until we either satisfy that the object is in fact what think it is, or until some evidence is encountered which refutes, alters, or refines this hypothesis. In the latter case, a new top-down pattern may be projected onto the apical dendrites which may bias a different set of neurons associated with a different object/context.

I lump those last two (2a, 2b) together because I think they may be serving similar purposes. In other words, the specific patterns and contexts that you attend to (e.g. through apical top-down projection) could themselves be the result of neurons that have also been biased (e.g. due to the influence of neuromodulators or other upstream top-down projections). It may also be that these are two independent systems that just happen to work nicely together.

Parts of the automatic attention mechanism might also interact with the volitional mechanism when, for example, a top-down projection biases neurons to generate predictions for specific sequences associated with the top-down context. If those expected sequences do not accurately predict the bottom-up stream of input, then it could generate a bursting event that forces the higher levels to consider alternative contexts or to learn a new/novel pattern.

[1] When applied to columns, bursting refers to when many local neurons fire in response to a common set of proximal inputs because none were in a predictive state.

[2] When applied to individual neurons, bursting refers to the rapid-firing pattern of the neuron. Contrast with the tonic firing pattern.

[dmac]

I’d like to clarify that the term “burst” has two different meanings:

• In the HTM literature: “bursts” are minicolumns where all of the cells activate all at once.
• Most other neurosci. literature uses the term “bursts” to mean: short high frequency trains of AP’s out of a single cell.

[CollinsEM]

Duly noted. I realize that I mixed my usage of the term in my prior response without clarifying. Hopefully my meaning was clear.

Edit: Added footnotes for clarity.

[dmac]

Cellular Mechanisms of Conscious Processing

Jaan Aru, Mototaka Suzuki, Matthew Larkum (2020)
Redirecting

Abstract

Recent breakthroughs in neurobiology indicate that the time is ripe to understand
how cellular-level mechanisms are related to conscious experience. Here, we
highlight the biophysical properties of pyramidal cells, which allow them to act
as gates that control the evolution of global activation patterns. In conscious
states, this cellular mechanism enables complex sustained dynamics within the
thalamocortical system, whereas during unconscious states, such signal propa-
gation is prohibited. We suggest that the hallmark of conscious processing is
the flexible integration of bottom-up and top-down data streams at the cellular
level. This cellular integration mechanism provides the foundation for Dendritic In-
tegration Theory, a novel neurobiological theory of consciousness.

In this article Larkum further develops their ideas about how apical dendrites are involved in consciousness.

They discuss the role of oblique dendrites. Oblique dendrites are located halfway up the apical dendrite, in between the soma and the distal tufts, which allows them to control the strength of the connection between the two ends of the apical dendrite. They hypothesis that general anesthesia works by effecting the oblique dendrites.

[Casey]

@CollinsEM

Automatic attention: In HTM, this is typically what we are modeling through the interaction of the distal and proximal synapses. Novel (unpredicted) input patterns/sequences generate localized columnar bursting[1] behavior which in turn gates the thalamus to broadcast more of the incoming stimulus to nearby columns.

L5tt cells form functional minicolumns (source), which supports that.

I guess automatic and volitional attention are sort of conceptually merged for the single-cell bursting. Reducing the threshold for bursting makes it easier to detect weak stimuli, but below a point, the stimulus is just too weak to notice even though the animal would be rewarded for noticing it. So volitional attention (specifically single-cell bursting) would be enhanced by how strong the stimulus is.

The stimulus strength and also the automatic attention you described (minicolumnar bursting) are both probably basal / proximal activity, which’d enhance (or at least enable) the volitional attention (single-cell bursting) because single-cell bursting generally requires a proximally-initiated spike to trigger the rapid subsequent couple of additional spikes which comprise a burst.

[dmac]

Apical amplification—a cellular mechanism of conscious perception?

Tomáš Marvan, Michal Polák, Talis Bachmann, and William A. Phillips (2021)
https://doi.org/10.1093/nc/niab036

Abstract

We present a theoretical view of the cellular foundations for network-level processes involved in producing our conscious experience. Inputs to apical synapses in layer 1 of a large subset of neocortical cells are summed at an integration zone near the top of their apical trunk. These inputs come from diverse sources and provide a context within which the transmission of information abstracted from sensory input to their basal and perisomatic synapses can be amplified when relevant. We argue that apical amplification enables conscious perceptual experience and makes it more flexible, and thus more adaptive, by being sensitive to context. Apical amplification provides a possible mechanism for recurrent processing theory that avoids strong loops. It makes the broadcasting hypothesized by global neuronal workspace theories feasible while preserving the distinct contributions of the individual cells receiving the broadcast. It also provides mechanisms that contribute to the holistic aspects of integrated information theory. As apical amplification is highly dependent on cholinergic, aminergic, and other neuromodulators, it relates the specific contents of conscious experience to global mental states and to fluctuations in arousal when awake. We conclude that apical dendrites provide a cellular mechanism for the context-sensitive selective amplification that is a cardinal prerequisite of conscious perception.