Finally I want to discuss with you today, recent breakthroughs, that I think, I feel, that became very, very, very important, recently, to enable us really, eventually. I think it's like a tour de force, really, in an attempt to understand neurons as computing device beyond ideas, because all that I showed you before were theoretical ideas. I did not really know how a particular neuron computes the direction of motion in the visual cortex, these were ideas. Now what do I need to know in order to make these ideas concrete, and really solve the issue, how does this neuron compute direction of the motion, for example. So, I want to highlight basically two really experimental breakthrough that really brings us close to be able to answer this kind of questions experimentally. One work come from Arthur Konnerth's lab in Munich. Whereby he developed the optical system, the two photon microscope system that I mentioned already before, to the level that enables us to ask really, at the level of single neuron, questions like how does the neuron compute direction of motion. So, here is the very first experiment. So, here is a mouse looking at the world. Here is the world and the world with these lines on the screen are moving in this direction or in this direction and so on, and he records from a particular cell. So, this is the cell he was recording from in layer two, three of the cortex, the visual cortex of the mouse, V1, records from this cell with an electrode, showing the different directions to the living mouse. So the line was moving this direction, recording from the cell body. Then you see dot, dot, dot, dot, dot, the cell fires. In this direction of moving bars, much less spikes, in this direction, this direction, from here to here, more spikes, less spikes and so on. You see that this is a directional selective neuron, because in some directions, particular this one, from bottom to up, top. It fires a load this cell, it fires a load less, no, no, less and so on. Direction and selectivity, okay? So, if I sum what I just said, I said that this particular cell is oriented, it responds to this direction, you can see that this polar plot, there are a lot of spikes, a lot of spikes. When the movement is in this direction, which is this case. And there are also spikes when the movement is in this direction, which is this case. Okay, so this is, the computation, that the cell performs, is computing the direction. In this case the cell likes this direction and this direction, but, but how does it do it? What do I need to know? How does it do it? So, Arthur Konnerth, extended these techniques of two photon microscope to be able, not only to look at the output of this particular cell, but also at the orientation selectivity of the inputs of this synaptic inputs to this particular cell. So, this is a summary from his paper in nature, from three years ago. And here is the result, the result is that if you zoom in into this input synapse, into this EPSP, you see that this particular input synapse is sensitive to this direction. Meaning that this synapse is active when there is line moving or bar moving in this direction. This other synapse, let's say synapse number four, here, likes these two directions. You can see it here, synapse number four, the EPSP using light vacuum sensitive dyes locally, locally here, or here in this case synapse number four you can see that synapse number four likes to respond to this direction and also here to this direction. This is synapse number four that likes to give, to respond to this orientation, and to this orientation, and so on. So, we are now at the really miraculous time, where not only you can record the output of the cell which is a summation of all its inputs. But also, you can record from the input sensitivity itself, what direction this synapse, this synapse is sensitive to, or this, or this. So, for example, you see in this case something we did not know, really, that this particular cell, the layer two, three pyramidal cell in the mouse visual cortex V1. All orientations are presented on the dendrites. So, the dendrites experience as an input, all the orientation, all the synapses. The different synapses code for different orientation onto the dendritic tree, but the output is oriented to a particular orientation, like the input that is sensitive to particular orientation. Also the output of this cell is sensitive to particular orientation. How does this dendritic tree transform the sensitivity of dendritic tree, so to speak? To many orientations, to making an output representing a particular orientation. Unsolved as yet, we don't know, but some computation is performed on the dendrites, for example, it could be, we don't know. That most of the synapses, it has many, many synapses here, not only 20, but, but 5,000. Maybe if you just count all the orientations selectivity of the input synapses, you found that most of them are 90, 90 degrees oriented, the inputs. So, the cell just is a follow-up, follower of the majority, just majority. So democratic neuron, just counts the majority of orientations, and decides that the majority dictates my orientation. Maybe there is another computation being performed with the dendrite. This will be solved, I'm sure, within a year or two, that's for sure. That's one Interesting and important new technology. That enables us not only to record the output of a cell, in VIVO, in the behaving, in the full in VIVO situation. But also the input to this cell and some computation is being performed by the dendrites generating a particular direction of selective output. The other system I want to discuss is direction selectivity in retina ganglion cells. Very fast about the retina. So, the retina is built from several layers. The most important ones are marked here, so the receptor layers is the layer of neurons receiving light. So light comes in, absorbed, the photons are absorbed by the receptors, and then there are several stages. The bipolar cells, which is the next stage receiving the input from the receptors. And then in next stage the amacrine cells here, and then the next stage the ganglion cells, and this ganglion cells have this axons the optic nerve. So, the output of the retina is here, the input to the retina light is here, and there is a network in between, and this network eventually computes something. Because already in the retina, if you'll recall from some, some of the ganglion cells, you'll see that these cells are directional selective. Here is an example. So, this is a recording from a ganglion cells in the retina, and you, you see that for some direction of motion, there are spikes. For some other direction of motion, 80 degrees, 135 degrees, there are no spikes. So, this is a, a tuned cell that tends to fire only in the retina already. Only when there are this direction of motion 225 degrees or 270, and no others. So, I'm saying that already the retina, before the cortex, before [UNKNOWN], before the cortex. There are already cells at that directional selective, even when you cut it from that and take the retina out the, computation is implemented at the level of this local network at the Retina. How, how? So, there were many ideas along the years, one of the most fundamental one is what is called the Reichard detector. The idea is the theoretical idea comes with many ingredients, but the general idea that there's is some symmetry in the system. Whereby in one direction there is more inhibition coming at some location than in the other direction. So, some asymmetry within the network causes the one direction to, to involve less inhibition compared to the other direction somehow. And this may explain the fact that, in some direction, in the null direction, you will not get an output in the null, in the, in the preferred direction you will get an output. And particularly, for the, for the retina, this was a possibility suggested. And here is just an example. So, if you move a light in this direction, these are the receptors. The photons are being absorbed, and if there is some kind of a symmetry, so this is the ganglion cell. And if there is along the way, a symmetry in terms of location of inhibition, compared to excitation, then this system will generate directions selectivity. Because, if lights move in this direction, this receptor will activate this inhibitor cell, amicrane cell in the retina, and this inhibitor cell will be activated first, just because you move here. And this cell is activated, bipolar cells activate this amicrane inhibitor cell. This inhibition Impinges on the ganglion cell, so you inhibit this cell, and then when next you come here, this excitatory path is all the way inhibited, inhibited by this cell that was previously activated, yes, and then there is no output. But if you move the other direction, you first activate excitation, and then later on, you activate the inhibition. This inhibition comes too late, if there is this asymmetry in first excitation. And only later inhibition, this inhibition will come too late, and there will be an output. This was an idea, it was an idea. That there is some asymetry in location of inhibition versus excitation. And in one direction first inhibition comes in and v to the excitation that comes later in the other direction. First excitation comes in, and when the inhibition is active it's too late. It was an idea. So how do you solve this idea? Very good idea, but what do you need to do? Basically, you need to see whether they are not [UNKNOWN], I just mentioned, is it possible anatomy really exist in the system, and that's where the connectomics comes in. So, we already mentioned in the first lesson. That there are these new techniques, and there is a group of people using this very sophisticated techniques to cut and reconstruct fully the connectomics. Kevan Martin from Zurich, was one of the first to try to connect anatomy at the level of synapses and neurons to a particular computation. Winfried Denk who developed the two photon microscope, and now doing all this connectomic work. Jeff Lichtman from Harvard, Sebastian Seung for MIT, Mitya Chklovskii from Janelia Farm, Clay Reid now in Allen Institute. All these people are trying to reconstruct at the level of synapses, because I need to know where this inhibitor of synapses sit. Where this exciter of synapse sit, when are they activated? So, I need both an, anatomy and electrophysiology to connect the two, in order to understand whether the computation that I suggested is really implemented, can be implemented by this particular system, yes or no. And really recently, we have a, a breakthrough, in trying now to reconstruct both the orientation selectivity of a particular cell. And then reconstructing the anatomy surrounding the cell, the environment of the cell at the level of synapses using this connectomic approach. I'm not going to go into the details, now just to explain to you that eventually this work, and this is coming from Winfried Denk from Heidelberg showing in his theme showing that indeed. If you look at the ganglion cell, this is the green cell, this is directional selective cell that was really recorded from. And also, you look at the presynaptic cell, the amacrine cells making inhibitory synopsis onto the ganglion cell. So, this is the Amarcrine using inhibitory cell, and you can now count the inhibitory synapses. Here is one inhibitory synapse between this, this presynaptic cell to the ganglion cell. Another one, another one, another one. Now you can count all the synapses. You can see there is symmetry immediately. You can see that to the right of the green cell you get more inhibition than to the left. If you just count how many synaptic inputs that are inhibitory inputs, you will see there are more red inputs here and less red inputs there, okay? This will be the source of direction selectivity, because now you understand. That this point of light is moving in this direction, from left to right, it will first activate less inhibition onto this cell, then excitation will come to the cell and will tend to activate it. If the input is coming from right to left, you will activate more inhibition, then the excitation will be vetoed by this inhibition. So, this could be, it's not completely resolved yet, but this could be a source from this direction selectivity, but now grounded in real experimental work. Both characterization of this direction of selectivity, physiologically but also anatomically charting, drawing the environment of synapses. That this cell is receiving, this is only the inhibitory cells. So, this is a breakthrough, because now we can connect the anatomy to function, from structure to function. Let me summarize by a personal note, this lecture. I want to say that the brain, the first from other physical system in many ways. But then one way in which the elementary units are, in other system, in most physical systems, the elementary units are simple and uniform. And you cannot really count the effect of one atom, or one molecule, on the overall physical system, but the brain is composed of nerve cells of neurons. Which are inherently complex as I showed you, and they are dynamic and they are plastic. And I think this eventually the formation of a network built in this case from complicated units makes this physical system differently than the classical physical system, where you can take an average of many many elements, each one is simple. Here there are many many elements, but each one is complex and each one may implement specific computation. So, as in with humans, I want to say metaphorically that the individual neuron seems to enrich the dynamic of the society. So each one of us are, is complex. And this complexity together generates a complex environment. You cannot ignore the complexity of the neuron, if you want to understand society. Okay, it enables us ,the neuron, the synapses enables us to learn, to change and eventually, successfully function in this complicated, unexplained and unpredictable environment. So, what we really need to develop, and it's still missing in the grand scale. We need to develop conceptual and theoretical tools for connecting the low level, so to speak, the neuron, the synapse. The low level of computation and learning capabilities performed by single neurons and their synapses, to the global computational function of the nervous system. This is the central role of the field called computation neuroscience is to go from the computation performed by synapses neurons, as I just mentioned today. To the global computational capabilities of a large computing network of neurons in the brain. And that's really the, the, the, the mission of computation neuroscience which is relatively new field, maybe 20, 25 years old. And, and finally, as I just try to show you in the last example, only very tight interaction between theoretical ideas, modeling and simulations. And experiments now with connectomics and other methods to record the activity of many neurons. Only the interaction between theory and modern experimental tools and experiments will provide the very, very missing still. Breakthrough in understanding how does neuron utilize, how does the nervous system utilize the neurons and the synapses to eventually enable me, for example, to stand and speak here. And for you now to listen and understand, move, go to the fridge, take a drink, go outside, and do all other things that you do computationally using neurons, synapses, and the network that they are build for. So, thank you very much, this is the end of lesson six. In the seventh lesson, we shall discuss the blue brain project and the human brain project. So, let's meet next week.
