I'm very pleased that Adam had one could come in person and we're having a real seminar today with real food and real people. So glad to see our atom is well-known in the circles of studying motor systems as one of sort of the avant-garde of the new, new optogenetics wave of mouse studies that is taking the world by storm. And he sort of one of the whirlwind of that storm. He got prepared for this drop through a bachelor's in biology since our biology department that our sounds great. In 1999, he got a PhD at UNC without parole. He was already interested in the spinal cord and movement. A post-doc with Thomas just self, who we are I think now from the distal Schwartz condyle textbook. So that was a Columbia and then he was a group leader at Janelia for 10 years until last year. And Janelia had this tendency to, to turnover and send their people to new and better frontiers. So he went back to you and see to this I think, very PhD department who I'm sure we're super happy to efficient back. And now he's in his own that there as Associate Professor studying how mice process movements. Without further ado. Thank you so much for coming and joining us. I'm not I want to thank theater for inviting me. I'm I'm super excited to be here. I didn't find a lot of the work that's been coming out at Georgia Tech and Emory for many years now. So I'm very excited about all the meetings today with the students and the faculty. My lab, as either mentioned, is interested in motor control. And I think at least my primary interest in here sort of stems from the fact that I think that motor control is one of the oldest and one of the most essential functions of the nervous system. It's really critical for all aspects of our survival is how we find our food, it's how we find our mates, It's how we avoid are predators. Because it's so old and so important. I think it might provide real insights to lots of other functions because perhaps are built on the architecture of the motor systems. Now, it's true that we don't actually need a motor system, nervous system to actually produce movements. Of course, we know that single cell organisms and it's very simple. Animals with no discernible nervous system can of course move through their environment. I like this example of this trick of plaques here. It's a very simple animal, has nothing that resemble the nervous system. Get it can sense things about the world, moved to those things, and then interact with them, at least Bye ingesting them. But so you can of course, make these movements without a nervous system. So why do we have a nervous system then? Well, I would argue the reason why we haven't nervous system is to make movements way better, to really upgrade our movement. It allows us to sense and respond to things from much further away. Make complicated decisions about those movements. Move faster with more control, more coordination. And to even use large effectors such as limbs and even tools for some species. And so one of the primary question to my lab is how does the nervous system afford us these upgrades? In the particular matter that we're interested in is Berlin control. How does the nervous system to control these limbs and use them for their behavioral needs? And of course, for the nervous system to do this, what it must have to do is understand the properties of these limbs, the ones that had been born with. And then given that understanding with some goals in mind, it eventually has to do kind of a simple thing, which is to generate motor neuron output and deliver it to the muscles. Give it a pattern of activity that will then hopefully move the limb in a certain way to achieve some sort of behavioral control. Okay, that's pretty simple on that side. And so you can think about the rest of the nervous system. Everything that's its upstream of these motor neurons as kind of a pattern generator, all they need to do is generate the pattern of activity in the motor neurons. And actually thinking about the nervous system as a pattern generators are really old idea in the history of neuroscience. It goes all the way back to the sort of the forefathers of the field, including Charles Sherrington, who's really interested in pattern generation for rhythmic movements such as locomotion. And actually originally, Sherrington thought that those pattern generators were actually peripherally dependent. That actually it was the sensory information coming back into the nervous system from the periphery that was actually driving the next steps of the motor output that it had the information. Now that was convincingly demonstrated to be incorrect by Sherrington's own student, Thomas Graham Brown, who took this question on and he actually showed that the pattern generators for locomotion, we're actually central origin. Actually the spinal cord could generate the basic patterns for locomotion on its own. You could deprive the spinal cord of its sensory inputs and it's descending inputs. And as long as you gave it a kick, it would generate some of the basic patterns of activity underlying the rhythms of locomotion. So my lab is actually also interested in these ideas of pattern generation and the nervous system. But not for these rhythmic movements, but actually for skilled movements. This is how we learn to interact with our world adaptively and flexibly. It is the movement that we get better with over practice it if of course it's also. The movements, we spend a lot of money just to watch other people do. And it's because of this practical importance and because of their cultural importance that neuroscience has of course been studying this really for over a 100 years. And what we've been doing as a field is sort of going through, so there's a little bit of a delay here that we've been going through all these different regions of the nervous system. And then ascii, how each region might contribute to either the learning or the production of skilled movements. Now today what I'm going to try to do is a little bit of a twist on this, where we're going to start to ask questions about how some of these areas actually interact in order to produce these skilled movements. We're going to focus on three areas. Motor cortex, thalamus and cerebellum. And we're going to start off with motor cortex because it is thought to be one of the real master commanders for skilled movements. There's lots of reasons for thinking this. One of the reasons is that damage or disease to this part of the nervous system leads to really severe consequences almost specifically and skilled movements. If you have stroke in this area, skilled movements are greatly deteriorated. This is of course, the area that if you do stimulations of particular kinds, you can get very discrete movements which are thought to be potential building blocks for skilled movements. And this of course, is the region that has neurons at both at the single neuron level and at the population level that have really rich information about skilled movements. And I think my favorite version of this right now comes from the work of Krishna Chinois lab. Crushes lab has been looking at motor cortex into the population activity within motor cortex, and then examining what the behavior of that population activity is. To each one of these traces here represents the population activity within motor cortex over different movements, where every different trace is a different movement. And this is a state-space. Probably don't need to understand it's part yet, we'll come back to it. But all I want you to really recognize here is for all of these different movements, motor cortex seems to be doing something a little bit different and there's rich information, these dynamics. So, you know, one could argue that actually it seems like we understand a lot about how motor cortex is involved in skilled movements. And I actually agree with that. I think we do understand a lot. But there are some really basic questions that still remain open. And there's going to be one that we're going to tackle the first here, which is weird. I'm sorry. Is it little bit laggy? Is it like jumping the videos? I can't quite tell. Okay. The basic question is, where do these dynamics actually come from? Okay, Like who's generating them? And there's sort of two ideas in the field, sort of sitting at the extremes or the possibility. One is much like the spinal cord. Motor cortex is generating those patterns on its own. It's like sort of an autonomous dynamical system is capable of generating these patterns of activity as long as it gets a proper trigger signal, it can generate those and it's learned how to generate those. Importantly, unlike the spinal cord over the training process. And I would argue this is probably one of the leading ideas in the field and there's good reason for that and a lot of practical good reason for that. But there is another possibility. And probably an older possibly, which is that motor cortex is really critical for producing these patterns of activity, but actually it can't do it on its own. It's constantly relying on inputs from other areas, putting them together, building them up this pattern of activity. But on its own, it can't generate those patterns. And this is more of a distributed dynamical system. And I think it's actually, it's really important to actually distinguish between these two hypotheses. Not only does it tell us about the true nature of the dynamical system underlying skilled movements? But it also tells us how to study this going off in the future. For example, if you're in this local autonomous mode, then perhaps you really should just understand to find features of motor cortex and how it changes over the training process. But if it's more distributed dynamical system, then you're sort of forced to try to tackle the interactions of all of these other areas of input, how they all combine in motor cortex and I, and I would argue those are fundamentally different jobs to do. So how we gotten to this place where we haven't sort of answer this very basic question. I think about this system and I think one of the reasons is that most of the pioneering work in this field was actually done in primates. And that has been excellent for electrophysiology. It's been excellent for behavior. But where it is a bit limited is our ability to do causal perturbations where we affect some part of the nervous system, purposely then monitor the consequences both neurally and behaviourally. And I, and I would argue that those classes of experiments help, that, help distinguish between these two possibilities. So what I started my lab now, 12 years ago, I decided I wanted to take on this problem. I thought one way to do it was try to port these questions to another model species where we can actually take on these experiments. And the model species that we decided to work on was the mouse. There was a couple of different reasons for this. First of all, the mouse makes excellent use of all these new tools for doing causal perturbations. Tools like optogenetics and pharmacogenetic see you in part developed in the mouse and they work quite well there. But we also pick the mouse because they can do really nice versions of skilled movements in perhaps the champion of these movements, elites for the neuroscience community. Was this movement, It's a reach to grab task really developed by Ian wish on Canada. The iteration that they had was, for me, moving animal sits in this box. There's a slit at one end of the box. And when the animal wants to eat a food pellet, he throws his him through the slit, grabs a food pellet and places it in their mouth. And when you look at the fine features of this movement as I'm showing you here on the right. They seem to do it through this stepwise execution of lifting and opening their hand and then grabbing. And they'll do this double supination go to their mouth. And what I really like about it is actually, it looks a lot so sorry. It looks a lot like actually how a human would grab an object off the table. And we love that sort of analogy to our own behavior. Now, when we started to think about this behavior though many years ago, we wanted to change it in one fundamental fashion. And that we wanted these animals to perform this behavior while being head fixed. Head fixation offers a slew of experimental advantages. It allows us to isolate the movements as much as possible to a single factor. It allows us to impose a cued trial structure. And the animal, which was really critical for the work in a primate. And perhaps most importantly from my lab. And really facilitated these hopes to do recordings and manipulations from within the animal's brain. So we tried to develop this task over the last 10 years. Sorry. It's kind of a simple task. Head fixed animal sits in this rig. Comfortable, we at least think is hand-drawn. This purchase a table out in front of the animal eats in the first iterations of this. And after a tone in a prescribed delay, this table will rotate, deliver a pellet of food within the animals reach, then hopefully they'll reach out and grab the food pellet. Okay. Now there's one thing I need to tell you about this behavior, which is kind of critical, but it is basically done in the dark or blinding light to the animal. So the way that we think that they have to solve this behavior is hit the memorize the position of the food pellet, store that memory of that whole motor plan in their brain somewhere and then call it up when we ask them to do the behavior. Okay, So actually animals can do this, can do this quite well after training, printing can take anywhere from a couple of days to a couple of weeks depending on the trainer and the animal. But they can do it really nicely. They understand the tones and the prescribed delays of the task. And then they can smoothly and coordinate and really execute through all the steps of this behavior. And these animals can really achieve remarkable success at this behavior. So for example, if you look at trial after trial, which I'm showing you here, these are just consecutive trials. They can reach success rates, basically a 90% of the trials on, and this is the first reach attempt. This is about double that success rate you see in the freely moving version of the task. So these are just absolute all stars at the behavior. So we really like the skillful performance that the animals are achieving. Really like the analogy to our own behavior. But there is one sort of annoying thing about the behavior which is its best captured through video. There's, there's sort of analog read out of this. And back in the early 2010s, when we started to think about this, we thought that the, the way of an analyzing these videos which affiliated through manual annotations are going to be too concerned, time-consuming, and laborious really to really last that long the lab. And this was around the time when computer vision and machine learning was really taking off. And we thought that we had a pretty good task for computer vision. This fixed registration to the cameras and a stereotype trajectory. And so we teamed up actually with a really brilliant computer vision specialists at Janelia Kristin Branson. And we presented this problem to her and she thought it was completely solvable and she was right. And sweep over the last 10 years. It really created a suite of tools to help us analyze all the features of these videos that we're interested in. I'll just mention this briefly because these have become sort of old hat now in the community, we have one sort of toolbox to be. It's akin to convert these videos into ether grams, or graphical depictions of the movements where he basically make up classifiers. I don't know why these little things are coming up. Graphical depictions of all the different steps of the behavior. As you can see down here, we basically treat up a classifier for these. And we also have ways of actually monitoring in three-dimensions positions of any unmarked part of the body. And these things have been great. They not only help how we analyze our videos, but the also fundamentally change how we do our experiments because you can actually run these enclosed loop with minimal latency. And that allows you to detect features of the behavior and then go back and perturb the animal any which way you want. And that will be critical, as you'll see in a few slides. So we were feeling pretty good about ourselves and the state's behavior analysis method. And so we thought we could start really asking some fundamental questions about how cortex was actually pulling this behavior off. But we were presented with maybe not an unexpected problem, but an annoying one which is at the Field of a sort of losing faith. That rodent motor cortex was really doing the same thing for skilled movements as was the primate motor cortex. And a lot of this doubt really came from chronic lesion experiments such as this one. This is a very large lesion of rodent cortex, actually extending even beyond motor cortex. And while there is a persistent deficit in these lesions in the skilled reaching test, a freely moving on. Effects are much milder than you would have imagined. They kind of form of the food pellet on more trials, but there's still able to do the basic steps of the behavior. Now, I think that these chronic lesion experiments are very, very interesting. I think they could tell us a lot about the brain. I think they can particularly tell us about the compensation capabilities. That mean what can the rest of the brain do in the face of the loss in this case of motor cortex. But what I don't really necessarily tell us about is what does motor cortex do normally in an uncompensated brain? And I think if you want to get at that question, well then one way to do this is to try to find a way to minimize compensation while taking these brain regions offline. You might want rapid reversible, unpredictable forms, a perturbation. And of course, optogenetics is ideally suited for this. So back in the day we really, for the first time ever, we wanted to use optogenetics to turn motor cortex off during these Reaching assays and see what happened. We've done this in many different ways now, but the original way we turned off motor cortex was actually by leveraging a Nelson made by group Anfang. Who made this mouse that expresses channelrhodopsin in the inhibitory neurons of the cortex. Expression of that opsin and exposure to blue light activates the inhibitory neurons. And they dumped gaba all over motor cortex and shut it down locally. And the nice thing about this is you could do this on random intermittent trials, so the animals will compensate for it less and you can do it or whatever temporal epoch you want. So at first what we're going to do is just sort of do it over the duration of an entire trial, but on random interment and trials. And because we wanted to map out sort of the full cortical space here. We can do this over many cortical regions. And actually most cortical regions silencing it give us absolutely no effect. And I'll just show you one example of that. This is ipsilateral sensory motor cortex. The lasers on here, you've delivered the food pellet and the animals able to perform that behavior really an altered in any way that we could detect. Okay, and I think this serves as an important control that if you ever do see an effect with blue light, entities, cantered off things breath in animals. It's not from the light or some non-specific effect of shutting off a very large swath of the brain. Okay, so this is ipsilateral sensory motor cortex. You wouldn't imagine, maybe it would have a huge effect. But you might imagine on the contralateral side to the arm, you'd have many areas that were involved. Actually, we found very little effect except for in motor cortex. And here the effect was absolutely dramatic. And that if you turned off motor cortex on the contralateral side, the animal simply could not engage in that behavior whatsoever. So for as long as we hold this laser on, you're going to see the animal simply not going to move in. Our animals never miss an opportunity to reach for food pellet. They are fairly hungry. Now, we were a little bit worried at the time that this had actually nothing to do with the movements of the arm and just that the animal didn't understand this kinda weird trial structure we have set up for the animals. So we decided to quickly probe that by putting these same animals in a context where the trials structure is exactly the same, but they could actually do a simple ballistic movement of their tongue in order to consume that food pellet, just a simple task. And you can see here that turning off and he's seeing the animals, this part of motor cortex has no effect on their ability to solve the task with a simple ballistic movement. As you can see here. Now, to notice something about this video though, and I can replay it because I think it's really interesting. I think this is like a general statement I want to make about like looking at all the videos because there's rich information here. While the animals are able to sort of weld tying the task here, they are also able to actually move this normally affected hand. This is the hand that they would normally reach out for the normal animal. The normally the Him, they can normally not initiate the movement at all. But in this context, they are able to initiate a hand movement. It's actually not that different. They're going to live, they're going to supinate, they're going to go to their mouth. They're going to help to push the food into their mouth. But they are able to initiate that movement with the motor cortex off. This in many other experiments suggested to us that there was something very specific about this learned skill behavior and maybe to learn skilled interaction with the world, which is the major difference between this task and the other one for the hand. That really makes you sort of depended on the cortex. All right, so this really was suggesting that cortex was absolutely required for the initiation, specifically in these learned skill behaviours. But it is at all it was doing, was it just important for initiation or did you actually require motor cortex pass the initiation for the continuous execution of these movements? And actually this was a very longstanding question. And they feel the motor cortex and motor control. Yep. Yep. Just a question. This looks pretty bimanual to me. Did you find that the kind of movements that survive motor cortex being silenced were more likely to be by manual then I don't know. I don't have enough data to like a real like a Olympics of our movements and like really test it out. I will say, you know, movements are definitely survive. Our movements like scratching, locomotion, this one. Now, when he's initiating that movement from the perch up that he's had her hands are actually doing something very different. So it's not a coordinated sort of synchronous movements of the two arms. And it's still affected. In many people have tried to silence it through simple joystick tasks and you don't stop the movement. But those are also frequently by manual that in actuality they present it like a single, but they're animals usually doing it. So by nato movements. There might be interesting cases for bimanual movements when they're doing interactions. We haven't probed it. Good enough. Yeah. That's interesting question. There's been lots of hope of on a transfer task that that might be really critically dependent, but we don't know yet. Yep. Okay, so the question is, is do you require cortex for the continuous execution? The reason why this hasn't been probed even in the monkey is you need to do this kind of special experiment. Let the animal start the behavior unperturbed and then detect the movement and then perturb them are midway through the behavior. That's pretty hard to do. But with the rapid sort of on, of optogenetics, the closed-loop nature of our experiments were able to try to do this experiment for the first time. And this was, you know, sort of gave us like superhero level status over these animals. Kids, we could essentially freeze or pause the animal almost anywhere along their path. And I'll show you two examples of this. The first is we're going to turn off motor cortex during this ballistic form of the movement where they basically lifting off the perch and going out to the food pellets thought to be too quick for sensory information to guide it. So it's thought to be largely preplanned. And what we're going to see is as soon as we turn off motor cortex here, the animals arm is going to stop in midair. So right there. And you can see that trial over trial over trial. They just cannot complete the task. And we've never seen an animal stop midway through a reach before. Alright, so motor cortex seems to be required for this sort of ballistic component. But what about when you get sensory information? Can sort of sensory feedback sort of take you home through subcortical systems. For that, all we're going to do is hold back the perturbation till after they're getting rich sensory information. In this case, actually we don't let them grab the food pellet and have the food, pull it in their hands. And we're going to see is again, they cannot complete the task. So the animal reaches out, grabs a food pellet, then we turn off cortex in his hand gets stuck here and holds position. Now again, I want you to notice thing about this video. There's the animals not disengaged from the behavior. He's actually trying to pull up his affected hand, what is unaffected arm? And he's trying to complete that task, but he's still that the hand is stuck. So obviously, motor cortex seems to be critically important for moving this hand. But I think it's important to think about, is that the rest of the brain still ticking away and trying to find a solution to this problem. And I think we should just remember that in all of these other genetic experiments. And I will just mention that again, this is also very selective. The interruption of an ongoing movement, selective to this skilled movement, if you do it during grooming, for example, animal is unaltered. We're sort of building up a nice necessity story for us, for cortex, it seemed to be necessary for the initiation and the continuous execution of the movement. But in these same experiments, we also started to gain some evidence that perhaps manipulation of cortical activity was sufficient to at least evoke the movement. And the evidence for this really came from watching what happened right after we stopped inactivating the cortex, sorry for that double negative. But what we saw is when we flip from laser on to laser off and laser has been on for two seconds. I have clipped it. Every time we do that, the animal immediately reaches out for the food pellet. Okay. And so we thought maybe he remembered the food pellet or maybe he read, sensed it during this inactivation period. But that didn't really explain all of the cases. And there was something kind of weird about the timing of these responses. If you looked at the latency from when the laser went off to the production of that movement, the laser the latency was shorter and tighter in its jitter then was the normal behavior in relationship to the tone, which is what the animal learned like. That's, that's the thing that they had learned. And this started to suggest to us that maybe the actual perturbation of cortex here and it's released from that perturbation, which actually causing the movement to happen. So we wanted to test that. We thought one way to test this was to take an animal, put them in a completely unmotivated state for reaching no food pellet know tones, nothing. All we're going to do is perturb their cortex and then release that perturbation and see if that causes reaching and lo and behold, this perturbation and actually now many other perturbations was effective at causing the animal to reach out where the food pellet should have been. And actually, if you slip a food pellet in the position, that actually complete the entire task. Okay, So perturbation of motor cortex and unreleased that somehow affects neural activity. Instead of going back to rest, it actually goes back to the movement. So perturbation of motor cortex is at least sufficient to evoke this entire motor program. So I thought, yep, yeah, question. That's that only work in wild-type mice after how many days of training does this affect? Yep, so we've tried to map this is a little bit difficult. The question is, does this only work in well trained animals? It only works in well trained animals. It actually seems to only work once the animals had no, not only how to reach well, but also when they're, they have minimal latency and variance on their normal initiation time. Like when it becomes like so hardwired in there, they're basically going at max reaction R min reaction time. Sorry, that's when we start to see this. But then their reaction time to the laser, it actually How about a 100 milliseconds faster than that, but like so it's, you know, some, some animals will go and 50 milliseconds after the end of the laser. Any other questions? Yep. Text dependent, is it if you put them in a different room so that they don't know that they might not they might be presented with treat. Would they still be able to evoke this? Yup. So we have put, the question is how context-dependent it in a specific experiment if you put them in a different room, so in their little cradle because you don't have any other way to do it. But in their cradle, we are taking them and put them in the two photon, which is a very different arena, ticking the table out. So now they've never seen this arena and there's no table, the event oriented, they will reach in midair. We've never done it freely moving or things like that. But there are other, there are other weird vagaries of this particular behavior. So for example, if you tried to do this within 20 seconds of the last successful trial, where success is measured by just a pellet touching the tongue. Even if they don't consume it, then you can not have okay. If you take them off food depth for one day, they will still reach on to the queue on every trial, almost every trial, but they will not do it to the laser. There is some weird dependencies here that we don't fully understand. Yeah. Okay. So anyway, the basic, this is just to get to the point where now I think at least for us, we feel confident that motor cortex in the road and it seems to be pretty important for skilled movements. And that might seem like a small one, but with a big win at the time. And that gave us some confidence that we could start to ask these questions. Now about how motor cortex is actually doing this, how it, how it was involved in these behaviors. And maybe that would have some relevance for the work in the primate. This brings us back to this question about where are these beautiful patterns of activity coming from the motor cortex? Well, before we get to that question, you need to confirm that there are beautiful patterns that are highly related to the movement in motor cortex of eroded. So we're going to leverage the head fixed nature of our experiments to establish electrophysiological recordings when the animals are doing the behavior. We wanted numerous simultaneous recordings from many units. We can get an assessment of the population activity. They're going to put down these silicone electrodes with a high-density electrode arrays deep into the cortex. We're going to target Layer five, which are the output layers of the cortex. And you can see down here are the animals are performed this movement. We have subsample the electrodes here, and I've colored in units that we think are individual units. And you can see nice recordings here in some relationship of the firing properties of these neurons to the actual behavior. But let's dig in a little bit more. There's just a few trials. So here are four representative units of things that we see in motor cortex. Histograms of their firing rates are on the top here, then raster plots are on the bottom. And these raster plots, every one of these rows is a trial and every little tickets a spike. Okay? So you can see are a real sort of x2 of neurons. You even in these four here, you have Sumner who really like to have an increase in firing rates relationship the behavior, this is all lined up to the production of the movement when the animals first lifting your hand off. Some neurons interestingly turn off right before the behavior or right during the behavior, and then other neurons have more complicated relationships. So real diversity cross at least this very small population. But I want to make an important note here that within a neurons, they are more remarkably consistent across trials. They really are doing very similar things across all these trials. But this is four units. Let's show you a bunch more units. This is 600 units, which used to be a lot. Now it's a small fraction of the population that we have. Every, this is a heat map of the average activity profile of these 600 units. Every row here is a neuron lined up to the production of the movement. And then what I've done, It's a separate these neurons out into three major categories. Those that increase their firing rates, those that decrease, and then still other ones that at least by this statistical tests are not modulated. So the first thing to notice here is even without hunting for neurons like we just put the electrode, didn't we get, well, we get 70% of the neurons, the ends of our electrodes really care about this movement. This area fundamentally cares about the behavior. A little bit over half increase their firing rate and a little bit under half decrease their firing rates. And it kinda interesting way. But let's see if there's a pattern of activity here. Okay, Let's concentrate on these neurons and increase their firing rates. And what I've done is I've sorted these rows, which are the neurons by the peak of their firing rates. And what you'll see is a really nice sort of tuning of neurons to different aspects of the behavior. Some have their peak fire rates before the behavior starts and then others sort of tile over this whole behavioral experience. And so perhaps these are involved in initiation and these are involved in that sort of continuous execution and are possibly why we can stop the movement midway through. Okay? So this is the single neuron level, but of course, we know we really want to examine this also at the population level. This is really inspired by the work from Christian Chinois lab. Population activities a little bit difficult to visualize because it sits in high dimensional space. It's related to the number of neurons you're recording from. And now we're recording from three hundred, four hundred neurons at a time. So what most people do for visualization purposes is to do a dimensionality reduction. You can think about that have like finding like sort of super neurons. And then you plot the lowest dimensions of that dimensionality reduction against one another. Then you can sort of map the trajectory of the super neurons to neural state space. So that's what I'm gonna show you here. On the left is the three dimensions explained most of the variance in mouse motor cortex. Every one of these little dots here is going to be a different trial for this particular session. And how the neural trajectory unfolds through the state space and on the right for those seem trials, what's happening to the arm in three-dimensions in the real-world. And then these little great things are just the projections on the back fields of the, of the, of the arena here. And so we're going to see here when I play this movie that plays HD, you really get this really beautiful pattern of activity in this state space plot on consistent over trial over trial. And it's heavily related to the information on the right matter of fact, I can decode what's going to happen on the hand from what's happening on the neural trajectories. And this dynamics is really important if it happens and we don't have time to show you all this data. If this dynamic happens, it's going to reach like that. So it's sort of if this thing is going on, the animal is reaching. It basically will look like this no matter how we ask the animal to reach. So for example, the blue traces are actually these laser a vote reaches of the animal has ever even heard of a sound or initiated. And that same way, if the trajectory is the same, and if the trajectory fails to get up this hill, it around 100, the animals not going to reach. Okay, So these dynamics seem really important. And this then finally allows us to get back to this being a question about exactly where are these dynamics coming from? Are they autonomist and coming from motor cortex? Are they from this more distributed system? Now the critical difference between these two hypotheses is really the role of the inputs in this local autonomous mode, all the inputs are doing is triggering motor cortex. And motor cortex goes in, does its thing. And in this distributed system, the inputs are constantly important because they're helping build up this pattern of activity. So what we wanted to do is manipulate these inputs and ask when and how they were important. So there's lots of inputs to motor cortex. What the major input, of course, comes from motor thalamus. This is just a parts of thalamus that project to motor cortex. So we decided to manipulate those first. We've done this many different ways, but the original way we did it was by actually co-opting they're inhibitory input, which comes from a nucleus called the thalamic reticular nucleus. They put channelrhodopsin into that nucleus. You put fiber over motor thalamus, shine light through it that activate. So it's inhibitory terminals dumped gaba all over motor thalamus and then shuts it down. That's been shown by us and by others, that that works quite nicely. And so then the first thing we did was just shut off motor thalamus who on random intermittent trials throughout the whole trial, they really have a hard time initiating the reach altogether. You can see the lift probability here is significantly decreased. And this really suggests that motor thalamus is actually really necessary, but this behavior, but actually does not distinguish between these two hypotheses. Because motor thalamus could be triggered either one of these cases. But what you really want to do now is hold back that limit perturbation till after initiation and then see if you can stop it midway through the reach. That's what I'm going to show you here. This is the hand position, at least along one of the dimensions on a controlled trial here in magenta. And when we do the limbic activation, that's going to be this white line that turns into green line when we do the Atlantic inactivation, you can see as soon as we turn that clinic inactivation on, the animals arm stops and either holds in a certain position or regresses back to the starting position. Okay? So this is suggesting that thalamus is required here throughout the continuous execution, this movement. But why, why is it required? You can imagine two different possibilities. One possibility which kind of boring. Which is that actually thalamus is not part of the pattern generator, but it does provide a tonic permissive signal to motor cortex. And without this tonic permissive signal, motor cortex slips below its threshold and it really is the pattern generator, but because it's below threshold, can exhibit it. Motor cortex is just silent. Okay? Now that's an opposition to a more interesting possibility, which is that motor thalamus is critically part of this whole dynamical system. Part of the pattern generator has got rich information about the pattern of motor cortex because it's building it. And without this limbic information, it's not that cortex go silent, but actually you just generate a bad pattern of activity. Alright, so let's see which one of these is true. While we got to do is actually record from motor cortex while doing the Islamic perturbations and ask what regime or in that's we're going to do here. This is now trialed average PCA, sort of like those dimensionality reduction plots I showed you before for the trials, we just sort of average over all the trials. And this is what happens in the control situation. The neurotic. I just briefly mentioned that we doing this from this period of time where we actually evoke it from the laser. We did that. So we can always get the neurons in the same initial state. But what normally happens is the cortical activity jumps out here to the right, bends around this turn does this squiggly line, which is the technical term. And then that's where the animal produces the behavior. Okay, that's what happens normally. Now, what happens when we Science motor thalamus is cortex just go silent. It just can't do anything or did it generate a bad pattern of activity? So that's what I'm gonna show you here. You can see a little movie of this over here as well. It's quite interesting and motor cortex certainly doesn't go silent. Matter of fact, motor cortex kinda does the right thing initially jumps out here to the right and the state space plots. But then it gets up to this particular quadrant of the state-space plot where things start to fall apart for the neural dynamics. And it dives down and sort of this neural of this down here or persist for basically as long as the perturbation is on. And this is not a period of inactivity. Actually, the neurons are actually more active and this configuration than they are. They are actually control configuration. But it's sort of a neural not, it's just not doing anything. And then interestingly, we can talk about this later. When you turn off the perturbation, you jump back out to the right, do the squiggly line, and actually the animal will read reach for the food pellet. Okay, so this really suggests that this motor thalamus is necessary to generate correct pattern of activity without at motor cortex. That's something very different. And you could potentially think about what's happening in here, about what is autonomist to motor cortex. Okay? So if I were to question each class crushed. This is from the link. I was wondering if in case you looked into what the pieces are actually encoding in the two conditions, have you had a look at what the, because I imagine that in both of these cases, the PCs where did encoding must be different greatly. What is PC1? Pc1 and green? Though? Maybe the first question is, how do we compare between the 2? Second is, do you know what they didn't actually encoding? Yes, you can do you can do the principal components anyway, this, this particular one was we did it. We did the weights in the control and then we plot it in the laser trials into it. So the PC1 and PC2 are the same in this configuration. You can actually do it the other way as well. You can actually do your PCA on both trial types, but this one is done actually on control. Now, what they encode, it's a little bit difficult, especially when we only have one trajectory that gets hard for us to really like pull this out. I can, in the monkey stuff, they can, you know, because they have many behaviors in many, a much variability they can really, I've asked interesting decoding questions were a bit more limited there. We know that PC is 12, can decode aspects of the movement, such as the velocity of the movement and things like that. But we, I don't know what each one does. Pc1 in the monkey and probably in the mouse has been, is likely a condition invariant component largely locked to the triggering of the movement, like the big sort of push in order to produce the movement. In general. That's what has been in the monkey and I think it's probably similar in the mouse. But going beyond that, I'm unable to provide like a probably a satisfactory answer. Right now we are. Yeah. Go ahead. Yeah. Okay. We are doing with the current version of this behavior on the animals are now doing two different reached targets in two different control policies, all in the same animal. And so we could potentially start to ask and untangle what the neural activity does, different dimensions in behaviors like that. I just think we're a little underpowered here. Okay? So it's necessary to generate these correct patterns of activity, but actually is it sufficient? Can you actually use the limbic information to steer around cortical activity? For that, We did sort of the analogous experiment. Instead of silencing motor cortex, we're actually going to do stimulate, sorry, not motor cortex, motor thalamus. You're going to put channelrhodopsin in motor thalamus. And then we're going to put blue light onto motor cortex. And we're going to do this at different frequencies to sort of get like a dose-response curve to thalamic perturbations to motor cortex. Now what I'm showing you here again on the left is the neural activity averaged over the trials and on the right, what's happening in the hand? And these are the different frequencies. So the white line is what's happening in control. The pink magenta line is what's happening at four hertz. And then what's happening at 10 hertz is this purply line, and then the cyan line is what's happening at 14. So what we really see here in the neural space, that's sort of dose-response effect that we get different effects of the neural trajectories given these different thalamic input. So we see to be able to steer around the cortical activity using motor thalamus. And it actually is not just junk neural activity that's important for driving movements. You can see really similar effects on the arm over here and this graded fashion. So you do, you are able to sort of steer motor cortex with thalamus, is that's what's happening in real life though. What a hypothesis that you might make that if thalamus is really giving this pattern of activity to motor cortex and that there should be rich information in motor thalamus about the activity patterns in motor cortex. So in order to answer that question. We've tried to do simultaneous recordings from both motor thalamus and motor cortex. Now, this ends up being a really tough experiment because you need to find the right parts of motor thalamus. We were able to do this by leveraging these neural pixel probes just give us a lot of good chances to find the right units. And actually using optogenetics to anti-drug quickly identify the neurons along this path. Okay? So what we did was he found the units that corresponded to motor thalamus, that projected motor cortex. And then we looked at the activity profiles at the population level of thalamus and cortex. And what you can see here at the qualitative level is an amazing similarity between what's happening at the population level and thalamus and the population and Cortex. And I don't have time to tell you about this today, but we can actually show quantitatively that there's a lot of information here. For example, I can use thalamus activity not to predict the behavior, but to predict the future state of cortex. It actually often, I can actually predict the future state of cortex from thalamus better than I can do that for, from cortex, which didn't really suggest as rich information thalamus and cortex with an autonomous dynamical system, it should certainly be true that I should be able to predict its future state better from itself, which is not always the case. All right, so at this point, what we had concluded was that unlike those sort of role, the spinal cord for things like locomotion, motor cortex does not seem to be an autonomist pattern generator capable of doing all of the interesting dynamics that are required for skilled movements. But it seems to be critically relying on inputs from these other areas for motor thalamus. And remember, motor thalamus is actually made up of many different parts receiving inputs from many other parts of, of, of the brain. And a lot of the work that we're doing in a lab right now as dissecting out the specific roles of all these limbic channels. What each one contribute to the pattern generation that we see evident in motor cortex. And they usually I give this talk and talk about that. But today I thought I'd talk about another side of this story which is equally as complicated, which is on the output side. So we would really like to know for these really important dynamics that are sitting in motor cortex. How do they actually contribute to movement? Like how, how did these dynamics or she produce anything that resembles a skilled movement? This is complicated by the fact that motor cortex goes to actually many places. It goes to basal ganglia, it goes back to thalamus, it goes to the brainstem. And famously of course, it goes to the spinal cord, which is an old love of mine. And so we would like to understand how these things are receiving this information, processing it, and then using it for movement. This is further complicated by the fact that not only does motor cortex go to all these areas, individual neurons within motor cortex go to all these areas. Either a reconstruction of a single learn from now our mouse-like project at Janelia, you can see this neurons axons going to many different targets. And right now as a field we, we don't have a really good toolbox for dissecting out the collateral pass from individual systems like this. Now there's one target of the cortex where we can attack and try to understand. And, and that's because of it's sort of Anatomical Gift we were given from this area called the pontine nucleus or the pontine gray. It provides a dice synaptic relay between motor cortex and the cerebellum. So it receives most of its information from the cortex, including motor cortex. And then it only delivers its information to the cerebellum. And so therefore, by understanding this, you might be able to understand what motor cortex is telling the cerebellum and why. Okay, so the first thing we wanted to do is know what kind of information was sitting in the pontine nuclei. There have been very few recordings from this nucleus, especially during movement. That's because it's sort of on the wrong side of the brain here, 10 millimeters down. But we thought that we could get there by leveraging again these neuro pixel probes and by using optogenetics to help identify the pons, we know that the pontine nuclei isn't the densest termination of cortical output. And therefore, if we put chin or UPS and in the output of motor cortex and ping did off. And then looked for the heaviest recipient field for that, that we have pretty good evidence that we were in the ponds, which we then later confirmed but histology. Okay, So we can find these right area. Then we can look at what these neurons do. How much time do I have left? Sorry. Okay. Alright, I'm gonna go quick. So basically, I'll skip this, but we see a diversity of neurons, but here's a big collection of these neurons. And you can see that a lot of the neurons and pons actually have very transient activation patterns. And they're highly related to the queue, which we see those in motor cortex. But we also see tons of activity a tiling over the whole behavior. And you can get further sort of evidence for this by using a GLM to try to understand the firing properties of these neurons. And if you make a filter to either the queue or to the reach, while you can find neurons that have some heavyweights from the reach filter, actually the heaviest weights are from the queue. So this is starting to suggest that actually the palm see, the pontine gray has an unusual biased information about initiation and preparation of the movement. Okay, So that's starting to be a little interesting. We now wanted to see what the fate of this information is as it sort of goes through the cerebellum. And so we thought one way to do that was actually take control of Pantene firing patterns, put in set frequencies of information, and then monitor how they got transformed throughout the entire loop. So the way we got selective control is who made a transgenic line expresses Cre recombinase only in the pons. And we put channelrhodopsin selectively. And those neurons, we've done this for a variety of different frequencies. But today I'm just came out one of stimulation frequency we put in, which is 40 hertz for two seconds. So we're just gonna go through, take a tour through the cerebellum of what happens when we put this 40 hertz of Ponto cerebellar information in the first sort of stop at the grandma cell layer. These are extracellular field recordings from the granule cell layer. The pons projects directly to the granule cells have mossy fibers. These are these really simple neurons that receive these mossy fiber inputs. And actually what you see in the greenhouse layer is a nice entrainment at the 40 hertz that we're putting in. What is the expected? We think that the mossy fiber granule synapse is pretty, has high fidelity and this is what we expect. So let's move on now. From the granules out to the Purkinje cells, which is the next station. Green hair cells provide these parallel fiber excitatory inputs to the Purkinje cell. And if you look at the responses to this 40 hertz simulation, we see lots of neurons that are driven at 40 hertz seem to be nicely and trained, but we're starting to get some neurons will show very different response properties. For example, we start to see these pauses in the activity patterns at the onset of this 40 hertz stimulation. Okay, so moving quickly to the next station, the Purkinje cells project and inhibitory production to the deep cerebellar nuclei. Here you start to see even more of a transformation. You start to see a lot of transient activations, which makes sense since it's an inhibitory projection, but much less tonic neurons. Here you do get some neurons that aren't trained, but it's less than the population, unless it is to the final station, which is all way back to the cortex. And here we got a kind of a big surprise. On the right. Here's what I'm showing you is what happens in the cortex. Now almost all the inhibitory responses are gone. Almost all the tonic activity has gone. There's no one really in treatment. It's mostly justice, transient activation. And that's quite different than what we see even in the output of the cerebellum, the deep cerebellar nuclei. Okay, So this is really interesting. We put in 40 hertz of two seconds of information. What comes out by having good luck to the cortex is a transient activation period. We're really interested in who are these neurons that are receiving this transient activation? So an h2 for us beads, what did they do in the behavior? And if you compare those transiently activated neurons to their neighbors who are not transiently activated. And look at what they do during the behavior, they're actually fairing much earlier in the behavior relative to these neighboring neurons. And this is starting to build up this argument that maybe the pons, it's carrying all this initiation signal. No matter what I put to the cerebellum, I get like a transient activation and maybe that's helping to kick off cortex. Okay, So what it's doing when now know what the Ponto cerebellar pathway is doing for the movement. So let's perturb it. We've done it many different ways, but I'm going to show you this perturbation, 40 hertz for two seconds because we can leverage that in a minute. So when you do that, actually you get a really nice effect on behavior. And as you can see, what we normally see is dysmetria, which then this targeting of the food pellet position here you can see attempt over attempt while the lasers on there just overreaching in this particular case quite wildly actually. And then as soon as we turn this laser off on this particular trial, they're going to immediately we correct and target the food pellet. So it seems like it's having this nice effect on the targeting and the food thought you could see this right here without ever going back to the person now can correctly target the food pellet. You can see this sort of dysmetria if you look at the success rates of these animals and of the standard deviation of the hand position at the grabbed position. And you can also see other kinematic features affected. For example, one thing that's relatively consistent has an effect on speed. Okay? So we really wanted to understand really why this was happening and sweet. But back to our neural recordings and we said, okay, what's going on at the neural level, both in the cerebellum, in the cortex while the animals are receiving this laser and then this targeting the food pellet position. And as you can see here, neuroactivity actually looks pretty normal. It's remarkably similar. This is the cerebellar nuclei and this is the cortex. I mean, they are shockingly similar. And remember these are the same neurons I just showed you that could be robustly affected by the laser alone. So this really suggests to us that when the animals are producing behavior, they can quickly sort of digest away all these sort of effects of the pod to cerebellar pathway and the behavior and the dynamics underlying the behavior is quite robust even to that very powerful perturbation. And we would love to understand the mechanisms of that robustness. Now there are some differences between these neuroactivity patterns. And I'll do this rather quickly. If you out. What is different between the neuroactivity patterns explain the behavioral differences. You can. So this is a decoder either based on the DCN or on motor cortex and what's happening to the hand. And you can observe these reductions in speed from either. So most of the dynamics are robust, but what is not robust to these, to this laser perturbation seems to be really important for setting fine features like the endpoint accuracy and the velocity profiles. Okay? But I promise I'm almost done this. I put you in a really weird place though. I told you like, I've got all these signals, only initiation signals, and I just showed you a bunch of execution deficits to like, why, why did I tell you this? We were really surprised by this two and it's not true that we don't see initiation deficits. Some animals actually took during the PN stimulation cannot initiate. And you can see a couple of those animals over here. If you, if you do a pharmacological inactivation of the pons, you can actually affect the reaction time. And actually some animals will weirdly actually stimulate, it will trigger the whole behavior off of the laser alone. So there is some evidence that this Ponto cerebellar pathway could be involved in initiation. But why weren't we seeing it more robustly? And we thought, thought about this for a little bit and we thought maybe if this really is a trigger signal, maybe it's actually very easy to adapt away from this perturbation. It can be very simple. And what I'm showing you here is like animals that have seen these permissions for many, many trials. So we decided to go back. We train animals, never show them the laser and give them a block of trials. The first block they've ever seen. And there you see a big, big consequence. These pink trials down here are the first trials the animals ever seen with this perturbation. And they cannot initiate the behavior. But then 15 to 20 trials later they pop right back to normal behavior. So we were really shocked by this. This is not the normal sort of timing of motor learning and cerebellum, which usually take hundreds and thousands of trials. This is happening over 15 trials where you go from absolutely no behavior back to behavior. So we wondered how this happened. We looked at the neural activity again, both in the cerebellum and the cortex. What I'm showing here, the cerebellum. This is the first three PCs that explain most of the variance. You can see these nice patterns of activity over these trials. And then what I'm showing down here is the first 15 trials where they see the perturbation from the pons. And you can see the neural activity is not normal here, something, something really egregious is happening. But also you can start to see this is even truer at single neuron level, which and I'm trying to tell you about the reshaping of this neural activity. And then eventually it pops back. And then the animals doing the behavior again. So the cerebellum seems to be able to quickly adapt. Wild perturbation that we're putting in 40 hertz over like the entire population of Pompeii neurons and get back to the normal dynamics underlying movement basically. Now what is the cortex doing over these 15 trials? This was a big shock to us. They're basically silent over these 15 trials. So somehow while the cerebellum is working out this horrendous insult that it's dealing with. It's actually disassociated itself from communicating from the cortex. And I think this is a really interesting potential mechanism that when you're doing this adaptation, maybe you actually don't want to fully communicate alone on this loop so we can preserve all the hard one plasticity happening in the cortex. But the mechanism of this dissociation is not clear to us. So this is, you know, I think a real interesting evidence of resilience in these circuits. We would love to understand this resilience because it is not clear to us how you get back. They'll good behavior, good neural behavior, and good actual behavior so quickly. And what are the learning and teaching mechanisms for this? How does it know which direction to go? And actually for this question and many other types of perturbations in the brain are becoming sort of obsessed with that notion. So when the end there, where my lab is trying to do is try to understand skilled movements. But trying to understand how all these parts interact in order to generate these pattern generators that are so critical. And then I have to thank the people who did this work, j and Jeremy who have come down will be to you and see you. I did almost all the experiments I showed you today. And Mateo in Britain who are both moved on. I can't find Britain's name in here or there it is. They did all of the awesome analysis. So thank you. B, two more minutes. So please feel free to ask questions over there below. Really nice talk. Adam. You showed a really interesting what do you call dose-response curve with the motor cortex with different frequencies. Did you try that in the ponds into GZ things at the lower frequencies that seem to be more effective in the motor cortex. Yeah. Excellent question I'll repeat. It just gets someone didn't mention here it. Did we replicate a dose-response or do we, we try a dose-response curve and the Ponto cerebellar manipulations. Not as carefully. We did, We have tried it. It's something interesting about it. Sometimes lower frequencies actually perturb the system worse and the higher frequencies and that system, and we haven't done enough of it. It's actually I'd board for future work. There is something there. And so it would be really interesting to see monitored this pathway both at the sort of short timescale and this more adoption adaptation timescales for different frequencies. Because as you probably thinking in the background, there's really interesting literature about coupling oscillations between cortex and cerebellum. And are there sort of sweet spots for that? We are really tough experiment and I have a hard time convincing J or Jeremy to do these experiments ever again. But that is really, it's on our sheet from for things to do. Yeah, Great question. We have a question from Linux hanging the virtual audience. Yep. Hey, that was awesome. I had a thought about these frequencies also because in the lower limb CPG, you can stop it by blocking the hip stretch information. So you can basically change the transitions from one state to another by manipulating the sensory feedback you get from the angles since there's different phases. And so I'm wondering if the four Hertz was effective in the arm because it's on the order of the frequency of the movement. And so there's something about having that time transient input to kind of keep that, that rotational trajectory going to complete the reach. My second thought is that in the PN stimulation you have this tonic activity. And it looks out, looks very much like cerebellar adaptation. Or maybe you have a huge loud noise and you can't hear it. And then you basically learn to filter it out and you can hear it again pretty, pretty soon. That's anyway, those are my thoughts about the two types of frequencies. Yeah, So I kinda went over this a little bit quickly, but the four hertz in our thalamus, cortical or transmission, it will cause a bubble in the movement. So it basically Hill hay and then go and continue on. He doesn't forget to do the action, but it will effect, it causes a bubble in the neuroactivity, it motor cortex, and it causes above all of that trajectory in the arm as well. So it says that 10 and 40 puts you in such a bad state that you are locked into a different position and are unable to fight through it. But if you think about four hertz over the length of one of these movements, which is like 300 milliseconds like you basically getting one stem in the movement. So four hertz is still pretty effective. It's just that 10 and 40, like pull you to a part of the state space that you could not break through from. So I think actually all the frequencies are potent. For the thumb, a cortical stimulation. Now, the, the pontine stimulation. Whether they, you know, when, whether they interpret that as sort of like, as I've just said, giant increase in noise. I need to think about that. I don't have a coherent thought about what that really, how to distinguish that between alternative hypotheses there. You know, it is clear that, you know, the interesting thing is actually most of the action is in PC3 there. And why, what PC3 is actually related to the movement would be interesting to maybe explore to try to figure out exactly what is going on there. I don't know. I don't have a good answer for that second question. Adam. Yep. On kind of popping up a level. Here, it's very striking that the dynamics of the movement are almost perfectly matched by the dynamics of the brain. And do you think, is the movements get more and more complex? That there might be more of a dissociation where there's maybe a limit on the complexity of the dynamics that you can generate and motor cortex or vice versa, yeah, right? Maybe that might also tease apart some differences between thalamus and motor cortex and some of these other brain area. Yes. Yeah. So I mean, I think that the nature of your question is if we drive up the complexity of the behavior, Could we get activity that actually is not so well, basically abstracted from the movement. And I think that's possibly true. I think you'll still have dimensions and motor cortex that will basically represent the movement. There's almost no way around it. Even if you're not maybe driving the movement, you'll, you'll reflect the sensory inputs. But you might actually gain some specific dimensions that and the neuroactivity that are quite different and quite abstracted. That's certainly the case. So that's one of the reasons why we have these animals. Do these two different control policies, are there like sort of dimensions that are just about which control policy you're in, which is kind of a version of that will probably know in the next few weeks. But I think that's possible. I think if you look at the monkey data and maybe the monkey, those sort of more conceptual dimensions might be in other areas, not in motor cortex, but in the mouth. I think it's fair game because like everything might be squished into this one region. Hi, great talk. Thanks so much. And I was wondering with that thalamic inactivation experience experiment and whether you think that any kind of disruption to thalamic activity statistics would have driven an impairment in that behavior or that it's specific to inactivation. Well, for 10, 40 is like another version of that. That's a stimulation and that was clearly effective. We've done a lot to Unpublish experiments where we inactivate for like just little short burst. It actually you could like slow the arm down. Even. It's like, it's like perfectly yoked to motor cortex. So almost everything you see in thalamus and every perturbation you do, you can get interesting and disruptive affected motor cortex. But yeah, I mean they are well couple, they're sort of take premotor cortex out of the equation where a lot of people think planning takes place. To some degree, there's still an ongoing interaction during execution. So going back to your piano player that you showed, I mean, they have to solve problems at different timescales and complexities, right? Yeah, playing each note with the fingers, but also planning how intense they are going to play it. And then they have thinks about melodies and rhythm and really high level concept. So where do you think motor cortex comes into that and where does that, say premotor cortex is come into that? I can't find premotor cortex for the, for the arm and mouse maybe like you know, in the experiment was talking about before we're doing different control policies in different choices. We have inactivated everywhere outside of motor cortex and none of those areas seem to affect the choice probabilities or any of those a higher level aspects of the decision. So one possibility is that they, like in the mouse, like the all sort of got wrapped into motor cortex or more importantly, and probably more probably in these loops including the basal ganglia and the cerebellum that they're all sort of wrapped. But then they basically are all sort of centered on motor cortex. But I think in the mouse, right now we have no. And we've tried hard to look at prefrontal areas and parietal areas for any other thing that can affect even find features of the movements. And for cortical areas, we can't find anything to Athena and the CFA it. So you don't see those rostral and caudal forelimb areas where we see them. I mean, we, you know, we are likely affecting both in almost all of these experiments. And so when we have tried to, you know, it's a light actually spreads considerably when we do these experiments. It's hard to limited in the experiments like with a transgenic animal. We have gone back and tried to do RFA for CFA experiments for the reach task. And actually you don't really block movement there. And I think it's because while there are some separations between RFA and CFA, they are highly similar networks and actually may be redundant with one another. And so if you take one out, the other one can very quickly overtake the whole function. So we actually don't get much phenotype if we restrict to one area or the other. And we've done it in both orders. Though. I don't know. You know, there are stories out there in the literature that RSA and CFA, or potentially doing different things with the CFAA is doing the reach and the RFA is doing the manipulation of the food pellet. We haven't seen any evidence for that. We have seen evidence of sensory cortex being really critical and evaluating whether the animals actually got the food pellet in their hand. But that's quite different thing for the questions. Then let's thank Adam again for really inspiring talk.