Neural mechanisms of wakefulness
Solo journal club post (was curious about thing and read/wrote about it)
Are there neurons, circuits, patterns of brain activity, etc. whose activation is necessary for being awake?
I review research on a candidate answer — layer 5 pyramidal neuron rapid firing and the role of excitatory input from neurons that project from the thalamus to L5 neurons. 1 L5 neurons are neocortical neurons that are thought to integrate sensory data and “priors”.
Much of what I cover has recently been synthesized into the Dendritic Integration Theory (DIT), introduced in 2020 — reflecting a growing cohort of neuroscientists who think these cellular mechanisms are causally significant for wakefulness.
There are some promising signs that these findings and theories are relevant to identifying neurostimulation targets for treating disorders of consciousness (using tech like transcranial focused ultrasound (tFUS) and deep brain stimulation (DBS)) as well as understanding mechanisms of anesthesia. The research I discuss also may help generate some intuitions for theories of consciousness.
I focus specifically on wakefulness rather than awareness of a stimulus. These operate as two somewhat distinct research questions, although they’re closely related. A scientist studying wakefulness may try to reverse the effects of anesthesia with brain stimulation while one studying awareness could be observing what happens in the brain when a subject becomes aware of an object in their visual field.
I explain experimental evidence, neuroanatomy, and theory + included a tldr. I also discuss proposed cellular mechanisms of dreaming, clinical relevance, and the generality of these findings.
I’m hoping this is comprehensible to a motivated reader with some biology context but detailed enough to not offend most neuroscientists!
Outline
Tldr
Layer 5 pyramidal neurons as an “on/off switch” for wakefulness
What’s the evidence for L5 coupling as a neural correlate of wakefulness?
Assuming L5 neurons are an on/off switch for wakefulness or an “NCC”, does that mean you need feedforward and feedback activity integrated for conscious experience?
What’s the evidence that apical drive could be the cellular mechanism of dreaming?
What’s the role of the thalamus?
A note on cell specific mechanisms in the thalamus: matrix cells in the CL
Compatible electromagnetic field theories?
How general are these findings?
T
ldr:
Layer 5 pyramidal neurons are a cell type in the neocortex that connect incoming sense data (e.g. sights, sounds) to neural signals from diverse regions associated with functions like memory and arousal (these are hypothesized to play a role in encoding expectation or priors).
When these signals meet in an L5 neuron, the neuron fires rapidly (called bursting or coupling). Some scientists theorize that this bursting activity could enable one’s waking conscious experience. That means when coupling is inhibited, one would become unconscious. In the past few years, there’s been experimental evidence that L5 coupling is disabled across various anesthetics.
“Non-message passing” (aka not relaying sense data) neurons projecting from a section of the thalamus to these L5 neurons seem to play a key role in wakefulness. It’s hypothesized (and increasingly supported experimentally) that input from these thalamic projections help L5 feedforward and feedback signals meet. When scientists stimulate these thalamic projections in anesthetized non-human primates, they regain wakefulness. Wakefulness is typically measured with behavior i.e. does an animal appear awake?
Layer 5 pyramidal neurons as an “on/off switch” for wakefulness
The cortex consists of 6 layers of neurons that each vary in their inputs, outputs, and sensitivity to different neurotransmitters. L5 neurons span across the cortical layers and are the only cortical neurons with dendrites in every cortical layer.
L5 apical dendrites, the dendrites that reside in Layer 1 (see above figure), integrate neural signals from all over the brain (higher perceptual regions, lateral interactions within the same region, “higher-order” thalamus, amygdala, the limbic system, claustrum, and various types of inhibitory interneurons). 90% of synaptic inputs to L1 come from “long-range feedback connections”. Inputs to apical dendrites are what I referenced in the tldr as priors or expectations.
When feedback activity, coming down the axon from apical dendrites, and feedforward activity, coming up the axon from the basal dendrites and previously primarily the thalamus, meet in the middle in the “coupling compartment”, the region of the axon between the apical dendrites, the soma of L5 cells then burst fires, or fires rapidly successively.
L5 neurons “BAC fire” (backpropagation activated calcium firing) where ions from feedforward signals can travel up the axon (the opposite direction ions travel to trigger action potentials at the cell body) and lower the threshold for dendritic spiking in the apical compartment.
When there are large spikes in the apical dendrites, it can lead to burst firing in the soma2. The literature on this question often emphasizes that L5 coupling “amplifies” apical activity when the apical inputs are relevant to incoming sensory information. There’s some evidence that during waking that excitatory input to L5 neurons from the thalamus is key to enabling burst firing in most cases.
If you’re reading this as a non-neuroscientist— action potentials don’t solely occur at the cell body. Dendrites, the spindly things coming out of opposing ends of the above neuron, can also spike. In a typical undergrad neuro class or just reading online, you probably won’t learn about dendritic action potentials.
What’s the evidence for L5 coupling as a neural correlate of wakefulness?
There have been several experiments testing the role of L5 neurons in generating consciousness, mostly in mice3:
Suzuki and Larkum found that anesthetics, with varying neuropharmacological mechanisms, decoupled communication between apical and basal dendrites. They also found that optogenetically stimulating apical “compartments” of L5 neurons resulted in them “high-frequency firing”, but when anesthetized this stimulation fizzled out before reaching the cell body (soma).
The exact biophysical mechanism of the decoupling isn’t certain, but Suzuki and Larkum theorize that input from higher-order thalamic nuclei which project to L5a and L1 may be the key mechanism for enabling coupling between apical and basal dendrites. Across anesthetics, they took recordings at 50 μm intervals along the L5 neuron and found that feedback signals would still propagate along the apical dendrites but would fizzle out (not effect the voltage) in the apical shaft, near the soma, consequently not reaching the soma. Relatedly, when they blocked the receptors that the thalamic nuclei project to in the cortex, L5 coupling no longer occurred.
L5 neuron synchronization “coincides with the loss and recovery of consciousness” across several anesthetics — this effect was not found for any other cortical cell types. Interestingly, basal dendrite activity stays in step with L5 firing whereas apical dendrite activity, while synchronized, occurs at a different frequency.
Takahashi et al. 2016 modulated L5 activity after training a mouse to lick for a water reward when its whiskers were stimulated. They found that when suppressing apical activity a stronger stimulus was required to elicit a behavioral response whereas increasing apical activity lowered the detection threshold — even inducing some false negatives where the mouse would lick for water when there was no stimulus.
Assuming L5 coupling is an on/off switch for wakefulness or an “NCC”, what does that mean about the relative roles of feedforward and feedback activity in producing conscious experience?
In trying to interpret what I’ve read, I think neuroscientists working on this question suspect that incoming information, in the form of feedforward activity, doesn’t become consciously perceived until it generates burst firing4. In this view, feedback activity is the causally relevant neural correlate of experience and one could have experience generated by just feedback activity e.g. if feedforward was inhibited. So, the waking conscious experience we are familiar with depends on feedback waves that are regularly integrating data from sense organs — checking and combining predictions and context with incoming sense information.
This question is interesting to explore through theorizing about the neural mechanisms of dreaming. It’s hypothesized that feedback activity that’s strong enough to generate somatic bursting, described as “apical drive” could be the cellular mechanism of dreaming (and waking hallucinations!).
What’s the evidence that apical drive (somatic bursting driven by just feedback activity) could be the cellular mechanism of dreaming?
Aru et al. 2020 outline this theory and review much of the L5 literature, but the main point I took away is that really high Acetylcholine (ACh), an excitatory neurotransmitter, levels during dreaming causes increased spiking in apical dendrites to drive somatic firing and generate outputs from L5 cells without needing feedforward input.
Aru et al.’s cite studies examining ACh like:
Layer 1 of the cortex, where L5 apical dendrites reside, has the highest sensitivity to ACh of any cortical layers, and Williams and Fletcher 2019 found that ACh release in layer 1 leads to apical activity that alone can drive somatic firing.
ACh peaks during REM (dreaming) sleep — activity in ACh neurons were recorded as more than twice as active, with some neurons being 11x more active, during REM than waking. Conversely, during NREM (non-dreaming) sleep ACh levels become very low.
The previously mentioned anesthesia paper (Larkum’s 2020) found that ACh receptor antagonists “abolish” apical drive.5
In the same section, they also review how changing norepinephrine (NA) levels effect apical activity. Apparently, this is unclear — there’s evidence that NA both opens and closes HCN channels, a type of ion channel that when open reduces apical drive. If NA opens HCN channels, then it would imply apical drive is reduced during NREM sleep when NA is high (and vice versa for REM when NA is low).
This story has more parts to it— I’d recommend reading Aru et al. 2020 in full if interested.
What’s the role of the thalamus?
I referenced some evidence for thalamic nuclei activity contributing to wakefulness (Suzuki and Larkum’s findings in the L5 evidence section). Projections from the thalamus, particularly the central lateral thalamus, a region of the higher order thalamus, to L5 pyramidal neurons appear critically involved in of wakefulness — excitatory input from these neurons enables distal and apical spikes to meet and cause somatic bursting.
A couple of recent papers have generated more evidence for this proposal including, namely from Michelle Redinbaugh, including:
Redinbaugh 2020 demonstrates in macaques that thalamic deep brain stimulation (DBS) of the central lateral thalamus (CL), a region of the HO thalamus, reverses the effects of general anesthesia
More recently Redinbaugh 2022 found DBS of CL, in the waking state also reduces levels of consciousness:
“DBS induced episodes of abnormally long, vacant staring with low-frequency oscillations here termed vacant, perturbed consciousness (VPC).”
There’s also ample evidence of DBS of the CL inducing similar states in other animals or lesions in the CL in humans being associated with disorders of consciousness like “spatial neglect and coma”.
The effect of DBS on reducing versus restoring consciousness varies by frequency. Lower frequency DBS i.e. 10 Hz has been found to reduce consciousness while higher frequency can restore it i.e. 50 Hz. (Redingbaugh 2022)
Redinbaugh and I met at a conference, and I was trying to better contextualize her data. While I knew that CL activity was relevant to consciousness, my assumption was that if you inject enough current into any part of the brain, you’d see changes in wakefulness.
Redinbaugh later explained over email that she used low-current manipulations (an order of magnitude less than other papers) and custom DBS electrodes fitted to the CL that are less likely to spread and effect neighboring nuclei. Moving the electrodes as small as .5 – 1 mm away from the CL caused the effect to disappear!
This is explained in more detail in thalamic stimulation which includes Redinbaugh’s email responses to my questions.
There are also a few cases where central thalamic stimulation in humans has shown promise for treating disorders of consciousness using both transcranial focused ultrasound (tFUS) and implanted electrodes— usually referred to as deep brain stimulation (DBS). It also looks like there have been efforts to transition tFUS thalamic stimulation into the clinic in treating DOCs although the clinical trial I found was terminated “due to COVID-19 imposed restrictions”.
A note on cell specific mechanisms in the thalamus: matrix cells in the CL
A couple years ago, I was very interested in Diffuse neural coupling mediates complex network dynamics through the formation of quasi-critical brain states. In it, Müller emphasizes matrix cells, a thalamic cell type that projects diffusely throughout the cortex to layers 1 and 5b and whose wiring “is not compatible with the traditional notions of “message passing’” (or its inputs don’t appear to be not relaying sense data that’s representing features in the environment) (Müller 2020). He speculates that matrix cells are “a temperature dial for computation in the brain”.
After reading Redinbaugh’s papers, I was curious whether her findings connected to Müller’s theory and simulation results. So I was really excited when Müller and Redinbaugh recently published a paper together, where they highlight that the central thalamus consists almost entirely of matrix cells!
And another piece in the puzzle — A paper that came out a few months ago presents evidence that higher-order thalamic projecting to the thalamus conveys “behavioral state information”, measured by locomotion, facial motion, and pupil dilation. The authors found that thalamic projections to a region of cortex respond more strongly to changes in behavioral state than corticocortical inputs. Interestingly, they mention that most higher-order thalamic nuclei receive their strongest input from the cortex. I’m interpreting these results to support the idea that matrix or higher-order projection activity may correlate with arousal level.
How general are these findings?
It’s possible that the discussed mechanisms could turn out to be the necessary conditions for typical wakefulness of neuroanatomically normal mammals. But what about in non-normal cases? Like individuals born without a neocortex? How confident are we about evaluating their potential experience based on the reviewed data?
I’m uncertain about how much these claims, like DIT, generalize outside of neuroanatomically normal mammals. Even if DIT turns out to be true for the specific cases discussed, it’s not clear that L5 burst firing tells us something about the necessary conditions for conscious experience in general. For example, many scientists believe that children born without a neocortex, therefore lacking L5 neurons, are conscious.6
I also think there are some additional considerations regarding DIT’s claims that I haven’t seen specified. For example, if DIT is true, what proportion of L5 neurons burst firing would be required for an animal to experience wakefulness? Or inversely how many would need to be inhibited to lead to a loss of consciousness (or decreased arousal level)? What about the number of thalamic nuclei exciting L5 neurons for these effects? It seems weird/wrong to think that making a small percent of L5 neurons burst fire would cause an animal to wake up?
That said, it seems like these experimental results have practical medical significance. Central thalamic stimulation, like in Redinbaugh’s work, could turn out to be an effective tool in treating disorders of consciousness.7
Thanks to Jaan Aru, Michelle Redinbaugh, Hunter Ozawa Davis, and Laura Deming for helpful feedback and conversations.
The thalamus is a structure in the middle of the brain whose neurons project mostly to the cortex. It has diverse functions, namely gating incoming sense info and helping control attention and wakefulness. ↩︎
BAC firing is hypothesized to occur when there’s “a ‘match’ between feature-specific [feedforward, sense info] and contextual information [feedback, prediction], performing a vital cognitive role” (Aru et al 2020
↩︎There’s evidence that L5 bursting has been associated with sensory input such as Kock et al., 2021 which measured L5 high-frequency firing when whiskers are stimulated. A friend suggested this paper as a counter to DIT, suggesting evidence that L5 neurons code for phenomena beyond awareness of a stimulus. DIT’s claims go a step further than these previous papers. I think the DIT creators would say Kock et al. doesn’t necessarily challenge their theory. Kock et al. don’t control whisker stimulus intensity since the mouse is exploring, so the bursting could still be the correlate of being consciously aware of the stimulus.
↩︎I’ll discuss research on how information enters awareness in detail in later writing. ↩︎
A neuroscientist friend outside this field noted that increased ACh will have other effects — not just changes in L5 activity.
↩︎I’m told that this view is not common though. ↩︎
A closing aside — a couple of the creators of Dendritic Integration Theory (what basically this whole essay has been about) were authors of Does brain activity cause consciousness? A thought experiment, an interesting, short paper that generates an intuition for electromagnetic field theories of consciousness that could also be compatible with DIT. ↩︎