 Thank you, Sonia, for inviting me to this workshop. I'd like to first acknowledge my collaborator, Taka Takahashi, who's a postdoc in the lab who has been involved for the past few years on this work. So I'm actually a behavioral, I guess I'd call myself a behavioral neuroscientist, neurophysiologist. So what we want to really understand is the neural correlates of behavior and what representations in the brain that tell us something about the behavior of the animal, and particularly my focus is on motor behavior of the upper limb, that is the arm and the hand. So that's what I've been studying for the past 15 years or so. And I'd like to start by talking about what I sort of see as the beginning of a paradigm shift in behavioral electrophysiology. So historically from the late 60s, when people tried to study behavior and neurobiology, typically one designs an experiment which typically involves presenting a stimulus or having an animal do some behavior, some motor behavior or make a decision of some sort. We then, after that training process takes place, which could take months in some cases to train an animal to do something, we then stick electrodes in and historically one would stick a single electrode in, record from a single neuron, extract data from that single neuron, then repeat the same experiment, move the electrode to a different location, record from a different neuron, and repeat this over and over again for many months to collect a large data sample of many neurons. And then the study ends. And then you analyze each neuron data samples individually, separately from the others. So in this paradigm, which still is quite dominant in the field, one spends relatively little time designing the experiment and analyzing the data, spent a lot of time collecting the data. In the past 20 years or so, people have been developing different multi-channel technologies, including multi-electrode arrays. I'm depicted here one of them, which is the Utah Array developed by Dick Norman at the University of Utah. It's a silicon-based array composed of 100 electrodes where each electrode is separated by its neighbors by 400 microns. And you get a sense of the scale, that's my fingertip. So you can see it's a relatively small, rigid array that's inserted into the brain. And one can then record from the tips, the electrode tips right here, which you can see the tips look a little different in color. Those are metallized, and those are the sites where one picks up electrical signals, extracellulary. So these are not electrodes that are penetrating cells, but in the intercellular space. One can pick up field potentials and then extract information. So with this technology and other similar technologies, the paradigm, oh, and by the way, with these, with this technology, this is the only gory picture I have. One can not only record from one area, so this is one of these arrays. This is a monkey's brain, by the way, exposed. One can implant one site, but one can also implant multiple multi-electrode arrays in different sites, and therefore understand not only the computations locally within a given cortical region, but interactions between different cortical regions. So with this technology, in this case, we're now recording from 300 different electrodes, and you can imagine how much data we have coming out of that. So this is resulting in a different paradigm where in principle, one can design an experiment, collect a set of data in one recording session, analyze the data, then design a new experiment, collect some data, analyze the data, and so forth. So in this case, one spends very little time collecting the data, spend a lot of time designing, which presents its own challenges, particularly training an animal to do a particular task requires a lot of time, and so that hasn't really been solved. In principle, one could go from one day to the next with a completely new experiment in the case of humans, but in the case of animals, it takes a lot more time. As far as analyzing data, of course, that also presents a problem, which I'll mention now for the rest of the talk. We now have multi-electrode recordings. One doesn't simply analyze each electrode separately, but one can also look at the interactions between different sites. So the challenges we face that might be of interest to you are first of all, visualizing the data, obviously, characterizing different kinds of patterns, spatial temporal patterns, linking these patterns to behavior, because that's ultimately my goal as a behavioral electrophysiologist, and then determining statistically significant interactions among neurons. So I'll give you some examples of each type of challenge, but before I do that, let me just briefly mention what's the data that we're getting from these electrode arrays. So if you look at the top panel right here, this is a typical trace, electrical trace will record from one of these electrodes. And by the way, the arrays that I'm recording from are recording from the frontal lobe, primarily the motor cortex and premotor areas of the macaque monkey. So here's a typical trace. We're recording at 30 kilohertz, usually tens of thousands of hertz, and most experimenters do, and this is continuously recorded. One can see interesting patterns, for example, these spikes, which are the action potentials from a given neuron. We can also see these slower fluctuations, which are often referred to as local field potentials. In fact, it's kind of a misnomer, this dichotomy between spikes and local fields. They're all local field potentials, but for some historical reason, spikes have their own designation, these little spikes in voltage. Those are distinct from the local field, which is typically signals that are below a certain frequency band, and that's typically below 300 hertz. So if you take this trace right here, we high-pass filter it, we can now accentuate these spikes, and then use different software to analyze to identify action potentials corresponding to particular neurons, so-called spike sorting problem. And then once we have the spikes, we can basically ignore the rest of the signal, basically the time. So essentially the problem becomes of just extracting these spikes when they occur, and the time to these occurrences is the relevant data. When it comes to the local field, or that is the data filtered below 300 hertz, one now records continuously, one has a continuous signal right here, and here's an example of very, very slow fluctuation below 10 hertz. Here's another fluctuation, which is particularly interesting to me, the so-called beta fluctuation between 10 and 40 hertz, and once these little wiggly lines, so-called oscillations, that are very prominent in motor cortex. If one looks at this local field, now this is the signal low-pass filtered below 300 hertz, and this is within a particular behavioral context, and in this case the monkey is performing a task where he's using his arm to move a cursor from a center position here to one of eight possible targets in the periphery. So he's making horizontal arm movements to move a cursor. And this trace is the local field, zero is when the target comes on, that is that yellow square where he has to move. The dotted line right here corresponds to when the go cue comes on, that's instructing the animal when to start moving, and then the solid line, vertical solid line, is when the movement begins. And what you see during this period of time when the monkey is not moving is this prominent beta oscillation. So you can see it in the raw local field potential trace, this sawtooth pattern right here. If you now band pass filter between 10 and 40 hertz, one can really see it quite prominently. So it's typically evident during preparation, it's also evident although attenuated during movement. Now what's interesting to us is with these solid, these rigid arrays, one can not only look at these temporal patterns, but also look at spatial patterns and spatial temporal patterns. So what I've shown you here is the array at different time points and I've color coded the voltage. This is the voltage in the local field potential, band pass filter between 10 and 40 hertz. So I'm really accentuating this beta oscillation which is a very prominent oscillation in motor cortex. And I've color coded it to sort of really show you what's going on in terms of the spatial variations in the beta. And what you can see in this example over time, this beta is actually showing some sort of progression from the top left to the bottom right as a propagating wave of activity. And I wanted to show you a movie right here if I can get out of here for a second. Let's see. Okay, this is a movie showing you this wave of propagation. What's plotted here now is each square is one electrode, just the local field, raw local field potential, band pass filtered in the beta frequency range. And you can just see it by eye the patterns that we're seeing. What's plotted on the top right is a polar plot that's showing you the instantaneous direction of propagation of the wave during a particular trial of behavior. What the monkey's doing in this case is he's preparing and then moving his hand to reach out and grasp an object. Okay, so what's plotted on the lower right is the instantaneous velocity of his hand, of his wrist. So right now, and the blue circle is just showing you where in time he is along this velocity profile. So right now he's just sitting there waiting to reach to grasp an object. And at some point he'll begin the movement and he'll increase his velocity, he'll reach a peak and then he'll decelerate as he approaches the object. And the only point I wanted to make here is we see very interesting patterns of propagation. They typically propagate along this axis right here. So they're propagating either the top right or the lower left. And there's something very interesting that happens right around movement onset which you probably missed. There seems to be a bias such that the waves tend to propagate along one direction, right around movement onset. That's, and that direction corresponds to a direction from the front of the brain to the back of the brain. And I'll make that clear in a moment when I get back to my presentation. Okay. So I showed you an example of one kind of pattern that we see which is this planar propagating activity. We see other kinds of patterns although they're much less frequent. For example, right here, we see an example of a circular wave where it tends to propagate around a circle along the array. But the most prominent pattern is this planar linear propagation. So we wanted to, we're still trying to understand what its significance is. But I'll spend a little bit more time describing how we characterize it. So one way we characterize it is by focusing only on its phase. So we take this voltage signal and we just, using the Hilbert transformer, we're just gonna focus only on the phase and ignore the amplitude of the signal. And from the gradient of the phase across the array, one can then estimate the instantaneous direction at every point along the array. And then using this, the quantity called the phase gradient of directionality, PGD, what this tells us, we get a measurement, a number, a scalar number between zero and one, where one indicates that all the phase gradients are oriented along the same direction. Suggest a planar wave across the whole array. Whereas a PGD of zero or close to zero would indicate that each local phase gradient is oriented in a random direction. Therefore, no consistent propagation direction. So by using a threshold of PGD, one can then extract only waves that consist of planar activity. And that's what we've been focused on right now. And if one uses a PGD threshold of 0.5, which is relatively high in our data, one sees that planar wave activity consists of a wave occurring in one of two directions. So this is a histogram showing the direction, the distribution of directions along one of two directions. A strong peak in one direction and another peak in exactly 180 degrees in the opposite direction. And if you plot those modes in that bimodal distribution, one plots out an axis along the brain. This is the central sulcus right here. This is the front of the brain, the arcuit sulcus. So we're look this way is the front, this way is the back. This is somatic sensory cortex, motor cortex, and this is premotor cortex. Each plot here is from one different monkey and one can see the axis is typically oriented orthogonal to the central sulcus. Or in other words, propagating along the rostral caudal axis. In premotor cortex, we see a different pattern where it typically propagating along a bimodal axis in the medial lateral direction. So that's one way we characterize these waves, but any other help we could get from the community to quantify these patterns would be very useful. Now, the second challenge is trying to link this, these patterns to behavior. And one approach one can use is information theory where one conditions on each behavior. So for example, in one task, we had the animal reaching out and grabbing different objects, okay, of different shapes and different sizes. And so we can condition the data on each kind of object that monkey was gonna reach for and look at the distribution of waves, wave directions and use information theory to say, well, is there differences in propagation direction for each of the object types? And what's shown here is that right around movement onset, which is zero right here, one sees a transient increase in information, that is, there are differences in wave propagation directions that characterize what kind of object the monkey's gonna reach for. One sees a second peak here, this is at the end of movement, after the object is being touched. Okay, then finally, I wanted to talk about determining statistical interactions between single neurons. And now instead of looking at local fields here, now we're gonna be looking at spikes, individual spiking patterns among a collection of neurons. And one approach we've taken in the context of these waves is by looking at each of the neurons spiking activity. So imagine we have a collection of neurons, X1 to XN, this is the firing rate of a neuron measured, of neuron one at time T. And what we do is we develop, we're trying to come up with a statistical approach to determine functional connections between neurons. And using an approach from economics called Granger Causality, one can try to infer, if we could predict the response of neuron one based on the response of neuron N at some time point in the past. If this past activity can predict the future activity of neuron one, we would then determine through some statistical test that this causes in quotes, causes or predicts the response of neuron one in the future. And actually I'm gonna skip this, I'm gonna just end with this figure. So what we've applied this approach, this approach allows us to come up with directed graphs of connectivity. And what's plotted here is such a directed graph. This is data from one monkey, a second monkey, and a third monkey. This is a different task now I'm looking at. I'm not looking at reaching the grab objects, but rather reaching in a 2D plane, just like the first task I told you about. And I'm looking at two different time windows. One time window is before the stimulus comes on to tell the animal where to go. And this window is after the stimulus comes on. Now that monkey is given the stimulus, is preparing and executing the movement. And one sees a variation in this connected graph as a function of time before and after the stimulus. That's the first thing to note. The second thing to note is if you want now plots in a polar plot, the distribution of directed connections. And now along space, along the cortical surface, one sees that this distribution of directed connections is not isotropic. And in fact, it shows an orientation, which is actually quite consistent with the wave propagation axis I mentioned before. So the wave propagation axis is shown by this CW, RW axis right here. And you can see this distribution of connections is oriented along that axis. And that seems to be true for these other two monkeys as well. Finally, if one looks on this panel right here, one can break up these connections based on the sign of the functional connection, whether it's excitatory, that is positive or negative. If one looks at the excitatory connections, they seem to be oriented along the axis of propagation. The inhibitory or the negatively weighted connections seem to be somewhat oriented or orthogonal to that axis. So we're trying to make sense out of this, both at the local field potential level, as well as the spiking level, as to what this wave, is this simply an epiphenomenon, such as Barry Richmond, who's in the audience would argue this is simply an epiphenomenon. I actually think it has some more interesting implications for behavior. So I'll finish there and take some questions. Thank you. You could tell which way they're going. But it also seemed that their scale was larger, much larger, their spatial scale, much larger than their window. And so I'm curious whether to investigate them, would it be also fruitful to go to a larger scale kind of recording more parts of the drain, maybe less spatial definition like pasta pieces or other things? Right, so we're doing that right now. In fact, we're using so-called e-cog grids. These are electrocortical grams. These are arrays of electrodes that are actually sitting on the brain, not in the brain. And their spacing is much broader. And we implanted one monkey that encompasses a much larger area. And we're beginning to analyze that data. So there's another scale, not another scale, but another dimension that I've ignored, which is also interesting, which is the third dimension, which is depth along the depth of the cortex, which is something I also am interested in. And their technology is now being developed such as by NeuroNexus, where you now have three-dimensional arrays that are recording not only in 2D space, but also along the shank of the electrode in the third dimension. Yes, right, well, that's the obvious question. And we haven't, historically in that area, people have documented a topography, and namely a somatotopy. So, well, it's complicated. And some people argue that unlike, it's clear in sensory cortices, there's a very strict topography, or a stricter topography. In motor cortex, there's a controversy. But yes, some people argue there is a somatotopic organization. And we're trying to understand the relationship between this anisotropy of wave propagation and the topography. But we could talk about it in more detail. It's not trivial. Maybe to comment on that, because there may be the question whether this is so variable. Did the monkeys do all the same task? And why did this pattern look so different? So, the spatial anisotropy that wave axis actually is quite consistent across behavioral tasks. So, I mean, the details might vary as to the dynamics of, you know, I show you the dynamics over, as a function of time, that might vary with tasks. But the overall spatial anisotropy that rostropodal axis is quite consistent over, we've now tested at least three different tasks. And by the way, we've also tested it in a human with this Utah array in motor cortex through sort of an accident of a part of a clinical trial. We were able to show a similar kind of axis in human motor cortex. Hi, so this looks like wonderful data and it looks like the electrophysiology is finally moving to high throughput. This is fantastic. Have you considered what to do with your data if you want community contributions? Have you made any of this data available to the community yet through a platform like Carmen or any of the others? I'm open to it. I'm open to sharing the data. I've been, whenever anybody's contacting me for the most part, I don't think I've ever rejected anybody, I'm collaborating quite extensively with different people who ever want to look at the data. In fact, I'm collaborating with Rob Cass here who was interested in this data and from a statistical point of view and I've been sharing it. So yeah, I'm very welcome to sharing it. But I haven't provided the data to a platform that's well-docketed. By the way, I mean, this electrophysiological data is not, you can't just take the data and run with it without really understanding what's going on. And it's not, I haven't annotated the data and provided a context. And you need to know that context to really understand it. Yeah, coming to my lab and yeah. Okay, I think we will come back to this question and let's speak again and let's move on. Thank you.