This work started a few years ago. We were training rats on precisely timed lever press sequences and found that they responded, after weeks of training, with beautifully choreographed little dances. 2/n
-
-
Prikaži ovu nit
-
These were particular to each animal and very stereotyped. Once learned, they were stable over months. Here are four trials each from two of our amazing rats, Jane and Maine. 3/npic.twitter.com/nWkH5fLhXE
Prikaži ovu nit -
These behaviors have some intriguing qualities that reminded us of similar over trained behaviors in humans. Here is a spectacular example of bread makers in Ashesh’s home country of India. 4/npic.twitter.com/My1rrFUeUf
Prikaži ovu nit -
But how do our rats produce the skilled behaviors acquired in our task? A few years ago we found, to our surprise, that motor cortex (MC) is not required. The paper on that story is here. 5/nhttps://www.sciencedirect.com/science/article/pii/S0896627315002202 …
Prikaži ovu nit -
But if not the MC, which circuits are responsible? In the first of our two preprints, we describe the role of the basal ganglia (BG) in the execution of these stereotyped behaviors. 6/nhttps://www.biorxiv.org/content/10.1101/827261v1 …
Prikaži ovu nit -
Based on the literature, we suspected that the BG will be involved in initiating the skilled behaviors and/or modulating their ‘vigor’ (i.e. speed/amplitude), with detailed kinematic structure being controlled from downstream circuits in brainstem-midbrain-cerebellum. 7/n
Prikaži ovu nit -
The first clue that BG may have a larger role to play came from Ashesh’s recordings in the striatum - the input nucleus of the BG. But before we look at the recordings, let’s remind ourselves of how the BG are connected to the rest of the motor system. 8/n
Prikaži ovu nit -
The BG form loops with cortex, thalamus, and brainstem. These are topographically organized, with the dorsolateral striatum (DLS) receiving input from sensorimotor cortex, and projecting to control regions of the brain. 9/npic.twitter.com/sxtEdMSbHX
Prikaži ovu nit -
Dorsomedial striatum (DMS), on the other hand, is part of the associative ‘loop’, and receives most of its input from PFC and PPC. We recorded activity from both sub-regions of the striatum in expert rats with well developed ‘dances’. 10/n
Prikaži ovu nit -
Here is a video showing our neural and behavioral recordings, and the spiking activity of two (red and blue) spiny projection neuron (SPNs) in the DLS of Hindol, another favorite rat. 11/npic.twitter.com/1jvltnvQ69
Prikaži ovu nit -
Now compare the activity of neurons in DLS and DMS. See the difference? The DLS representation is continuous and each SPN is locked to a particular phase of the behavior. DMS neurons are much less modulated. Also no prominent start/stop activity. 12/npic.twitter.com/NJ9OyqE6B4
Prikaži ovu nit -
DLS activity allowed us to decode the details of the rats' behaviors, DMS activity did not. Shown are observed and decoded aspects of kinematics and timing of the motor sequences. 13/npic.twitter.com/FqqtrAkfAD
Prikaži ovu nit -
OK, but remember that MC is NOT important for executing the learned behavior. Remember also that MC provides a major input to the DLS. If activity in the DLS reflects a role in control, then the representation should NOT depend on input from the MC. 14/n
Prikaži ovu nit -
Ashesh and Steffen looked at this. And voila: the nature of the representation in DLS did not change after removing the MC. Ashesh could decode the behavior just as well without MC input to the DLS.15/npic.twitter.com/kLTXJoAufJ
Prikaži ovu nit -
Interesting. So what role do the motor cortical inputs to DLS play? That is something we'll revisit in the second paper, so stay tuned. Now back to the DLS. Its activity tells us about the behavior, but correlation is not causation, you say, and we agree. 16/n
Prikaži ovu nit -
To look at whether DLS activity reflects a causal contribution to the learned behaviors, Steffen and Raymond Ko (a student of mine) decided to lesion it in expert animals. They also lesioned DMS as a control (units in DMS are not correlated with behavior). 17/n
Prikaži ovu nit -
DMS rats had no problem reproducing their ‘dances’, but DLS lesioned animals showed no signs of their learned behavior. They were pressing the lever after the lesions, but in ways that were strikingly similar to their initial attempts at start of training (each shade a rat).18/npic.twitter.com/7tZ6Xd4goL
Prikaži ovu nit -
Indeed, our DLS lesioned rats all reverted back to a simple lever-pressing behavior that was similar across rats. We speculate that our rats press levers by adapting a species-typical behavior generated by 'hardwired' circuits in the brainstem/midbrain. 19/n
Prikaži ovu nit -
This may be the default way for a rat to press a lever - all our rats start out this way. But, they can earn a higher rate of reward in our task if they change their movements, insert new elements etc. And that's indeed what they do, hence the ‘dances’. 20/n
Prikaži ovu nit -
We believe the demand for new task-specific behaviors recruits the BG, which modify the innate pressing behavior. 21/n
Prikaži ovu nit -
OK. That’s interesting (we think). But how do the BG learn to do what they do? And how can these circuits perform their control function without input from the MC? 22/n
Prikaži ovu nit -
These questions were the starting point for our second study, now also published in preprint form. 23/nhttps://www.biorxiv.org/content/10.1101/825810v1 …
Prikaži ovu nit -
Consider again the BG related neural circuits. DLS receives excitatory input from both thalamus and MC. We had previously shown that MC is required for learning the skills we train, but not for their executing them once acquired. 24/n
Prikaži ovu nit -
We wondered whether this ‘tutor’ function of the MC is mediated through its projections to the striatum (i.e. DLS)? To find out, Steffen silenced neurons in MC that project to the DLS before he introduced the animals to our task. 25/n
Prikaži ovu nit -
The results were unequivocal. Rats did NOT learn the task, no ‘dances’ emerged from the training. When Steffen silenced DLS-projecting MC neurons after learning, he saw no effect. We take this to mean that MC ‘tutors’ subcortical circuits through its projection to the DLS. 26/n
Prikaži ovu nit -
Let’s return to the question of how these ‘dances’ are generated. Remember that the DLS, which is an essential cog here, also receives input from the thalamus. If you didn’t know this, it may be because this connection is often omitted from textbook chapters on the BG.
Prikaži ovu nit -
However, thalamic inputs to the DLS are comparable in strength and numbers to those from cortex. Steffen probed their function through the lens of the idiosyncratic behaviors we train. 27/n
Prikaži ovu nit -
He silenced neurons in thalamus that project to the DLS in expert animals with well developed ‘dances’. Upon silencing, no more dancing. Animals reverted back to the species-typical lever pressing behaviors we see early in training and after DLS lesions. 28/n
Prikaži ovu nit -
Steffen further showed that plasticity at excitatory synapses in DLS is essential for ‘storing’ the learned behaviors. Given that MC input to the striatum is dispensable, this suggests that plasticity happens at one of the other inputs to DLS, perhaps the thalamic ones. 29/n
Prikaži ovu nit -
Much is left to figure out about how these behaviors are learned and controlled, but I hope these studies will bring us a step closer to understanding how the mammalian brain learns new motor skills. Thanks for reading to the end!
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.