Not being a neural scientist and generally not familiar with how neural science works. How do we know that L5 sends motor commands. If that’s the case why do need a motor cortex dedicated to controlling? Also why do we need a Cerebellum if we have the Neocortex and Motor cortex doing the job? Is there a way I can look to clear my confusions?
I have also been thinking quite a lot about this very question during the last months. Many structures project to motor centers, so what is the added value of cortical projections to motor centers? I don’t have a definitive answer, but my personal intuition is that those cortical projections have a teaching role, not a controlling role.
It may seem counter-intuitive because we like to think about the cortex as the smartest part of the brain which can directly send complex and specialized motor commands to execute complex actions. But in nominal operating mode, those projections are not enough to activate motor centers by themselves. I speculate that they only act as learning signals for subcortical structures.
To elaborate a bit, there are two main types of corticofugal projection cells from L5:
Pyramidal Tract cells (PT cells) which project ipsilaterally to cortex, striatum, thalamus & motor centers
Intra Telencephalic cells (IT cells) which project bilaterally to cortex & striatum
PT cells are remarkable in that they send axon collaterals to many subcortical structures (including motor centers). They are in a good position to coordinate a complex process distributed over several different structures. I posit that this process is dedicated to learning and I was glad to read a recent paper which stated that a mouse could execute an already learned behavior (but not learn a new behavior) while corticostriatal projections were desactivated, in line with my intuition: https://twitter.com/mthiboust/status/1189972405395107842
IT cells help the coupling between cortical areas (to converge towards compatible action maps) and send relevant information to the striatum in regards to action selection. Those messages could be of two kinds: valence of actions (the corticostriatal connections which terminate in striosomes) & specification of actions (terminate in matrix).
My view was first inspired by the work of Paul Cisek and Sten Grillner (action valences & specifications), and by the “3 visual streams” paper by Randall O’Reilly (learning signal via corticothalamic loops)
This to me is testimony of the fact that what the nervous system does - is meant to do - has been meant to do for so long - is first and foremost controlling behaviour. Actions. Motor.
Making sense of the world ? not so much.
However, we do. Thanks to them cortical structures, but that seems like an afterthought to the initial design…
I fully agree, the whole nervous system is first and foremost action-focused in order to control behavior (“external” actions) & internal physiology (“internal” actions).
Our brains are full of redundancy, with nearly every structure projecting to some extend to nearly all other structures. Projections to motor centers are no exception!
However, it is probable that some connections specialize in action selection, others in action execution, and others in action learning (and to go further, there are different levels of action selection, execution & learning…) In my current mind framework, I like to think of cortical projections to motor centers as a learning/teaching guide.
To go further, I recommend the papers coming from the lab of Bence Ölveczky:
By means of motor cortical lesions and high-resolution behavioral analysis, we show that motor cortex is not required for executing the learned skills we train, implying that they can be stored and generated subcortically. Intriguingly, however, we find that motor cortex is essential for learning the very same skills, suggesting that it plays an essential role in guiding plasticity in downstream control circuits during skill learning.
They talked about that in Jeff leads Numenta Research Meeting - Sep 4, 2019
L5 doesn’t necessarily represent motor commands like “throw ball”. It’s probably tied into sensory processing too. One example where motor/sensory are arbitrary distinctions is target selection. The superior colliculus (SC) produces saccadic eye movements and is sheet-like. In the fraction of a second before a saccade, there’s a bump of activity on part of the SC. That part of the SC corresponds to where the eye will move in visual space. Is the sheet a map of visual space or a map of motor commands? Maybe both.
Another thing which complicates motor vs. sensory is, where does sensory end and motor begin? The deeper layers of the SC send motor commands, whereas the superficial layers do not do so directly. V1 projects to the superficial layers of SC, whereas other regions project to deep layers. Does L5 of V1 send motor commands, or does it send visual information, such as potential saccadic targets? Is attention sensory or motor? Maybe L5 sends attention signals, like paying attention to throwing a ball.
The motor cortex probably has a lot of control over behavior, but most or all neocortical regions can produce behavior.
The cerebellum probably helps fine-tune behavior, which is important in nature but not so much for us. 3/4 of the brain’s neurons are tiny cells in the cerebellum. The circuitry is simpler than in the neocortex, but repeated a lot.
I think they probably can produce behavior, although their influence on learning is probably important. Some cortical regions project to the spinal cord, which has central pattern generators for things like walking. I don’t think those can do much learning, so the cortex must be able to produce behavior (so long as subcortical structures don’t strongly disagree.)
In fact, the main counter-example to my speculations is that lesions to the corticospinal tract impair the execution of skilled behaviors. But interestingly, there seems to be a gradation of the level of impairment along the phylogenetic tree, with the highest level seen in human (paresis or flaccid paralysis).
One possibility could be that some L5 PT cells evolved progressively from learning signals to controlling signals (I am thinking about the Betz cells in primates but not mammals). But it is just adding some speculations on speculations…
From the different articles in your old post, the paper by Henry Yin on basal ganglia has deeply attracted my attention:
He argues against the classic theory of BG which states that BG select specific actions by disinhibiting downstream structures. His results are informative and convincing. His theory is intriguing. But I still don’t understand why his papers don’t have been cited by more researchers.
Behavior should be interpreted as a continuous process (not a sequence of discrete events as neuroscientists usually do)
High firing rates of BG output should not be interpreted as increased behavioral inhibition as traditionally assumed. SNr neurons increase firing for movement in one direction and decrease firing for movement in the opposite direction.
What matters is the delta with the baseline firing rate (can be less, can be more)
The system works as negative feedback control of input. This means that the signals in selected input channels will approximate the relevant internal reference signals, in a closed loop manner.
There is a hierarchy of those control loops, the one implicating the BG being at the top
SPN activity in sensiromotor striatum highly correlates with velocity
SNr activity (output of BG) highly correlates with position
The transformation from velocity to position underlines the “integrator” function of the BG which mainly acts on transitions
The descending signals in the motor system are not commands that specify behavioral outputs, but orders to request specific inputs (those orders are described as reference signals)
I end with an interesting extract regarding his interpretation of predictive signals:
The transition control model explains action selection as a process by which competing cortical representations alter the reference signals of transition controllers. This account explains what are commonly called value signals or prediction errors. Such signals appear to be predictive to the external observer, but their actual function is to control. In other words, they do not modify open loop state-behavior mappings, as assumed in reinforcement learning, but represent reference signals that request specific inputs in a closed loop.
Doesn’t really help directly with my speculation about the role of L5 PT cells, but really interesting and intriguing info!