Good afternoon. My name is Nico Hatsopoulos. I'm from the University of Chicago. I should say up front, I'm not a neurosurgeon, although I tell my kids that I'm a monkey neurosurgeon. I am a neuroscientist. I work primarily with non-human primates, and my focus is understanding the cortical basis for motor control of the upper limb, reaching and grasping, as well as more recently on orofacial control of the jaw and tongue. But I really won't be talking much about that today. What I thought I'd do is start with some review and talk about various areas involved in cortical control of movement. And if you look here, there's vast areas, multiple areas in the cortex involved in control of movement. Over half of the cortex is involved in some way or another, beginning with the parietal cortex, posterior parietal cortex, where the initial intent to move seems to occur. There's some elegant studies with surface stimulation in posterior parietal and epileptic patients that seem to indicate that when you stimulate there, people have the, they describe a desire or an intent to move, even though they don't necessarily move when you stimulate there. As you move forward into S1, M1, and then premotor areas, area 8 for eye movement control, and then prefrontal cortex involved in high-level goal selection, goal choice, and various high-level executive functioning. You've got this very large network of nodes. I'm going to focus primarily on M1 and S1 today, critical nodes in this much larger network. And as, I'm sure this is review for most of you, but the homunculus notion that Penfield first developed way back in the, almost a century ago now, using surface stimulation in human patients, we came, he came up with this notion that there is this representation of sensory percepts on the cortical surface for leg, arm, and hand, and face down as you move laterally. And if you stimulate on the motor side, of course, you evoke movements of the leg, arm, hand, and face. And this is arguably the most famous image in neuroscience, and you'll see it in all introductory textbooks. But there's a couple problems with this notion, or at least it's not complete. Let's see. So, first of all, it's incomplete, and I'll describe what I mean by that in a moment. And then secondly, I think it's actually misleading. And in fact, many of my students who actually seem, we try to educate them, are still misled by this based on the original homunculus image. So, let me describe why it's incomplete. First of all, it ignores one dimension. It considers only the medial lateral dimension. It ignores the rostrocodal dimension, which is actually quite interesting. If you look rostrocodally, of course, somatosensory cortex has a number of distinct robin areas with functional modules, beginning with area three, A and B, one and two. And then on the motor side, you've got half of the motor cortex, at least in primates, that is buried in the sulcus, the caudal portion of motor cortex. And then more rostrally, the cortex sitting on the gyrus. And there's some argument that the more caudal portion of motor cortex sitting in the sulcus is a newer, evolutionarily speaking, a newer cortical area that developed with the, with our ability to sort of generate dexterous movement of our arms and hands. So, in terms of sensory motor cortex and the rostrocodal dimension, of course, we've got these various robin areas, but also they have various functional distinctions. So, you've got area 3A, which is the primary proprioceptive area that receives inputs from thalamus and ultimately from the body, reflecting the position and limbs of the, positions and velocities of the limb in space via various sensors in the muscles and joints. And then sort of the meat of somatosensory cortex, which is area 3B and 1, which is the focus of tactile inputs coming in. And then more caudally, area 2, which has a complex representation of both proprioception and tactile inputs with more complex, larger receptive fields. Now, in the motor cortex, as I said, there's a portion buried in the sulcus, which is believed to be, as Peter Strick talks about it, a newer area of motor cortex. So, the motor cortex in general is actually a relatively novel development in mammals. It probably emerged about a hundred million years ago. But then, even more recently, we have the evolution of this caudal portion of motor cortex buried in the sulcus. And this is a study by Peter Strick's group where they use transneuronal rabies tracers to identify corticomotor neuronal cells, that is cells, these are cell bodies now, that make monosynaptic connections with motor neurons in the spinal cord. So, very direct line to the muscles. And most of them are buried in the sulcus, in the caudal portion of motor cortex. There are very few sitting on the gyrus. And this newer motor cortex seems to be unique to the primate. Although, there might be a hint that raccoons might have such a system. If you've ever had, I've got raccoons in my backyard and they're actually quite dexterous with their upper limbs. So, there may be an argument that it goes beyond just primates. But I think it, so far, the definitive evidence suggests it's unique to primates. So, the other point is, besides it being incomplete, the homuncular perspective may be misleading. Because if you look at this map, particularly on a gross level, it's certainly quite correct. But if you look at a more fine detail and you look, consider the upper limb. This map seems to indicate that there is this one zone, there's this one zone right here that if you stimulate it right here, you'll get movements, twitches of the thumb. And you move slightly more medially, you'll get twitches of the index finger and middle finger and so forth. That actually turns out to be incorrect. And in fact, Penfield didn't claim this at all. In fact, his early studies indicated that, in fact, there isn't a specific zone for each digit of the finger, of the hand, nor specific zones for the wrist or elbow or shoulder. This is a map from Penfield showing that, for example, the index finger, or let me, sorry, the thumb is, movements of the thumb are evoked at many regions along the motor strip and not in just one focused area. So it's rather, it's a highly distributed, non-contiguous representation of the different joints. If that's with surface stimulation, if you use intercortical micro-stimulation with micro-electrodes penetrating the cortex and using high stimulus frequencies around 300 hertz for short durations, 50 milliseconds, you can evoke movements. This is in a monkey. And what's shown here is a map, but what I've marked in red are zones that evoke movements of the wrist. As you can see, the wrist representations don't form a single contiguous zone, but rather are scattered all over the place. And surrounding these little islands that evoke movements of the wrist are other parts of cortex that stimulate other joints. So both based on surface stimulation as well as more precise intercortical micro-stimulation, the motor cortex doesn't seem to, at least at the level of the upper limb, doesn't seem to have a very well-structured layout or somatotopy like we may see actually in the somatosensory cortex. However, it doesn't mean that it's just a complete mishmash. There is some structure. And this is work done by Paul Chaney's group where they used intercortical micro-stimulation in the monkey and recorded EMGs from electrodes indwelling in the muscle. And what you see here is, despite the fact that you don't see particular zones for the index fingers, the thumb, and so forth, there is this central core. This is the cortex and it's actually flattened. The dotted line here is the fundus of the central sulcus. Rostral is going to the left. You can see this central core right here in blue that evokes movements of the distal appendage, that is, the wrists and the fingers. As you move out more rostrally and surrounding it is this red zone that evokes movements of the elbow and shoulder. And then in between is the purple zone that evokes movements both of proximal and distal appendages. And so this reminds people of sort of a horseshoe organization where you have a central core surrounded by a horseshoe. And in fact, that's been seen even in humans with high-resolution fMRI where you have this central core in red that responds to movements of the fingers surrounded by maybe not a perfectly beautiful horseshoe, but nevertheless a representation of the wrist and the forearm surrounding it forming that horseshoe. Okay so that's a little bit about stimulation mapping and the homunculus, but my work focuses mainly on encoding, what the motor cortex encodes in terms of movement. And this is actually recordings that I did using microelectrodes from multi-electrode arrays that we implanted in a monkey. And we trained these monkeys to make various arm movements. And this is an example of a bunch of neurons we recorded while monkeys made movements to the left. So they're basically holding a joystick and they see a cursor on a screen and they have to move the cursor to a target to receive juice. So they're making these horizontal movements to the left. And these are all simultaneously recorded from this array of electrodes and you can see some neurons tend to increase their firing, others decrease their firing at movement onset. Movement onset, by the way, is this vertical dashed red line. But they tend to modulate several hundred milliseconds, two to three hundred milliseconds before movement onset. And in classical work from Georgopoulos in the 80s found that a preponderance of neurons in the motor cortex encode direction of movement. So this is an example of one such neuron we recorded from while monkeys made movements in one of eight directions in the horizontal plane. And this particular cell preferred movements up and to the left and did not like movements down and to the right. And different cells have different preferred directions. You can represent these preferred directions as a vector in 2D space. You can expand this now to 3D space as well. But if you limit yourself to 2D, you've got a preferred direction represented as a vector. And Georgopoulos found that using a very simple decoding algorithm where you take the preferred directions of a bunch of cells and represent them as these vectors and by adding them up and weighting them by the firing rate of each individual cell, you can get a very good estimate of the actual direction of movement the monkey made. And it's actually still being used today in brain machine interfaces work that Andy Schwartz, for example, does at the University of Pittsburgh. So it's a very powerful approach. And it shows things, it actually expands our understanding of motor cortex because one can actually predict movement direction even before the movement actually occurs. So this is a task where a monkey is given a cue to tell him where to go, but he's trained not to move until a go signal comes on. So you can look during this very early period, well, you know, almost a second before the movement actually occurs. And the population vector as denoted by these vectors here actually do a pretty good job of predicting the actual movement direction the monkey's going to make. Very early in the movement, before the movement occurs. So that's all fine and good, but the idea that cells encode a static preferred direction has come into some, various research has indicated this may not be complete and may need to be augmented. For example, if a monkey makes movements in different parts of the workspace, the preferred directions of these cells change. Or if the monkey makes movements with different postures of the arm, such as with his elbow sticking up or his elbow sticking down, the preferred directions of the cells change. So it's not this invariant representation that we thought occurred at the level of single neurons and motor cortex. And there was a study that Michael Graziano did with a different kind of stimulation paradigm that led me to refine this very, this static representation of preferred directions. So Michael Graziano at Princeton found that if you use very long electrical stimulation trains, these are from microelectrodes, but you now stimulate instead for 50 milliseconds for on the order of 500 to 1,000 milliseconds. And typically use very high currents. Instead of using currents around 30 microamps or less, you use currents up to 100 microamps. What you can get, and what he found was, you can get very complex movement trajectories of the arm and hand. And they, in fact, appear to be very goal-directed. So for example, if you stimulate the red zone here, you'll get movements that look like reaching and grasping objects. Or if you stimulate another zone, you'll get a climbing behavior. Or you stimulate right here, down more laterally, you'll get movements of the arm that approach the hand as if he's trying to feed himself. So this led me to consider perhaps the motor cortex single cells don't represent static preferred directions, but rather represent movement trajectories. And so we set out to investigate this with these multi-electrode arrays. So these are, this is a Utah array. It's composed of 100 electrodes arranged in a 10 by 10 grid. The electrodes are about a millimeter or a millimeter and a half long. That's my fingertip to give you a sense of the scale. And we implanted it in various cortical areas, including the primary motor cortex, as well as in the premotor cortex. And I'll just focus on the primary motor cortex right here. And we trained these monkeys to do various tasks. So the first task was this straightforward Georgopolous task where you make, you have the monkey make movements in one of eight different directions. And we said, well, oh actually I have a video, but I better not, how much more time do I have? Ten minutes. I better not play that video because it seemed to crash on PowerPoint. So I'll, it was just a video of our recordings. We can record with these Utah arrays for, and record up to 200 single units from one array in these monkeys. And in our best case, we can record for many, many years. We had got one monkey going now for close to nine years with the same array implanted. So we wanted to explore this idea of trajectory encoding. So we had the monkeys do this very simple center out task that is the Georgopolous task, making movements in one of eight directions. And what we did instead of what typically people do is where they look at a very large time window and compute the preferred directions. We said, let's look at very fine time periods. If you calculate the preferred direction at very fine, maybe 50 millisecond intervals, this is one neuron. You can see, and zero is when the movement begins. And he's making one of eight directions of movement. You can see this particular neuron prefers movements down and to the right. And then over time in the trial, it changes direction. It starts pointing up and to the right. This cell prefers movement down and to the left. And then over time, it shifts its preferred direction and prefers movements up and to the left. And likewise for these two cells down here. So they're not static representations. So that presents a problem for the standard directional tuning model that Georgopolous postulated. We then had monkeys do another kind of task. And so there's a little monkey sitting in his monkey chair. And he's holding onto the joystick right here. And he looks down on the screen. And his job is to move a cursor to a target. But in this task, it's called the random target pursuit task. And again, I had a video to show you. But basically, it's very simple. He moves to a target. And the target then jumps to a random location, a new random location. He goes to that. And then he hits that target. Then the target moves to a new random location and so forth. And then after he hits about seven of these targets, he gets a little bit of juice. And then he continues. And so what this does is it creates a movement that never really, never stops. He's continuously moving. And he's moving in a random fashion. It's almost like he's scribbling. And he just samples a large part of this workspace right here. So he's just moving all over the place in a random fashion. And then we said, OK, well, let's compute these preferred directions under this task condition. But of course, in this task condition, we don't have an event to align our data on. For example, start of movement, because the movement's always happening. So instead what we did, we said, well, let's look at the relationship between the neuron and the movement at different time leads and lags. So these red arrows indicate time periods when the neuron's activity led the kinematics, the movement of the hand. The blue arrows are periods of time where the neuron's firing lagged the movement of the hand. So the red arrows represent motor representations. And the blue arrows represent sensory representations, because they reflect the movement that happened in the past. OK. So this is motor cortex. So we're arguing that the motor cortex has a sensory representation, not just a purely motor representation. And if you do this calculation at different time leads and lags, you get this kind of pattern, where you have a preferred direction down to the right. And then it shifts and moves to up and to the right, just like we saw in the center out task, but using a slightly different method. And the task is quite different. So if you now add these vectors head to tail, what you get is a trajectory, OK? And these trajectories maps out a particular complex movement. So we would argue that these cells, instead of representing a particular movement direction, actually represent a particular trajectory that can be quite complex in its shape. And so this cell prefers this kind of sweeping motion, this cell, this one. This one shows almost a sharp right turn. It prefers movements forward and then movements to the right, and so forth. So we can look at the array now and represent these preferred directions. We actually came up with a mathematical model to compute these preferred trajectories, which I won't go into detail in. But here's the map, here's the array, the Utah array. It's 4 by 4 millimeters. Each panel here represents the preferred trajectory of one neuron we recorded from during this task. And you can see different cells have different shapes. They seem to represent a whole variety of different complex shapes. Although what you might notice is here on the right, cells that are nearby each other seem to share similar preferred trajectories. And in fact, there is some sort of topographic organization. It's not completely random. We've shown this notion of trajectory encoding, not only for 2D reaching, but also for complex 3D reaching and grasping. So this notion of a trajectory seems to capture the response of neurons more accurately. One can also look at these preferred trajectories in different time periods to see how stable they are. I mean, an aspect of encoding would imply that this representation doesn't vary at a very fine time scale. So in fact, what we did was we computed the preferred trajectories for, let's say, the first half hour of the task the monkey engaged in and the last half hour the monkey engaged in. And we found, so the curve on the left, the red and blue one, shows you the preferred trajectory we calculated for the first half of the experiment. The magenta and sort of cyan trajectory represents the second half of the experiment. And you can see these actually remain relatively stable and have relatively similar shapes in their representation. So these are stable representations in time. They also seem to be more stable in space. If we have the monkey make movements in different parts of the workspace, they seem to also be quite stable. So what this leads me to believe is that, and we're pursuing this further in further research, is that motor cortex doesn't represent the static movement parameters such as movement direction or movement force, but rather complex trajectories in space and time. So with that, I'd like to thank my lab, various graduate students and postdocs that were involved in the research. Thank you for your attention. Thank you.

Nico Hatsopoulos

motor control

motor cortex

somatotopic representation

trajectory encoding

neurons