UNIVERSITY OF PENNSYLVANIA INSTITUTE FOR RESEARCH IN COGNITIVE SCIENCE 18th Annual Pinkel Endowed Lecture: Grid Cells and the Neural Map of Space February 25, 2016 Ubiqus/Nation-Wide Reporting & Convention Coverage 61 Broadway – Suite 1400 – New York, NY 10006 Phone: 212-346-6666 Fax: 888-412-3655 18 Annual Pinkel Endowed Lecture: Grid Cells and the Neural Map of Space [START RECORDING] DR. DAVID BRAINARD: Good afternoon. Welcome. I'm David Brainard. I'm the Director of the Institute for Research in Cognitive Science. And this is the 18th Annual Pinkel Endowed Lecture, which we run through the Institute. This year, we've teamed up with the Mahoney Institute, thanks to t he generosity of John Dani. And the Pinkel Lecture is part of our campus-wide Year of Cognition, so it's one in a series of lectures we've had this year. In a second, I'll turn things over to Mike Ka hana to introduce our speaker. But before I start, I wanted to tell you a little bit about the Pinkel Endowed Lecture series. This is endowed in honor of Benjamin and Anne Pinkel , through the generosity of their daughter Sheila Pinkel and on behalf of their estate. Benjamin and Anne Pinkel--it serves as a memorial to Sheila's parents. Benjamin received his Bachelor's degree from Penn Electrical Engineering in 1930 and was at that time and remained actively interested in the philosophy of the mind, published his own monograph on this, titled, "Consciousness, Matter, and Energy, the Emergence of Mind in Nature." And the Pinkel series is devoted to modern scientif ic explorations of that topic. And it's intended to advance the discussion and rigorous study of the sorts of questions which engaged Dr. Pinke l. With that, I think I'll turn things over to Mike Kahana to introduce our speaker. DR. MIKE KAHANA: Thank you, David. It is my distinct pleasure today to introduce Professor Edvard Moser, who is Director of the Kavli Institute for Syst ems Neuroscience in Trondheim. Professor Moser did his undergraduate and graduate work in psychology and neurobiology at the University of Oslo, completing his PhD in 1995. And then he conducted postdoctoral work at the University of Edinburgh and University College London. Professor Moser's work on the neural representation of space, which he will tell us about today, my own introduction to this work a number of years ago left me with the reaction, “Oh, my goodness. Is it actually possible that these cells really exist, gri d cells that encode location within an environment in a tessellated map? ” I thought, “It can't be. It can't be true.” And of course, after reading paper after paper after pa per, I saw, well, it must be true, but let's see if we can find the same kind of neurons in the human brain. And Josh Jacobs in my lab did discover grid cells in the human brain. And of course, by this point, we now know that the work of Edvard and May -Britt Moser has yielded an incredible, detailed picture of how the brain repre sents space across species. And that 1 work has changed the way we think about higher cognitive processes, not only in rodents, but even across more advanced species and in humans. Professor Moser is the editor of Current Opinion in Neurobiology and the recipient of numerous awards, including the Louisa Gross Horwitz Prize, the Karl Spencer Lashley Award, and the 2014 Nobel Prize in Physiology, shared with May -Britt Moser and John O'Keefe. And on that note, I introduce our speaker, Professor Edvard Moser. [Applause] DR. EDVARD I. MOSER: Can you hear me now? Is it fine? Thanks a lot, Mike, for that nice introduction. Thanks also for choosing me as this year's lecture speaker and also fo r filling up the lecture hall. I appreciate that. I want to start where Mike ended: namely, one of the really big advances in neuroscience the last few decades is that we are now able -- we are actually still beginning -- to understand the neural basis of higher cognitive functions, the most complex functions of the brain. And this is particularly rewarding for me and probably for many of you because I started out as a psychologist . And when I started in the 1980s, it was almost unthinkable that we would be able to talk about psychological functions in neural mechanistic terms. But the world has really changed. A lot has happened. And I will try to introduce to you one of the areas where there has been much progress, namely, the understanding of space, which has helped us get the ideas about how the cortex wo rks, not only for space, but maybe somewhat more generally. But understanding cortex is not entirely new. And there has been important groundwork. Is there a pointer, by the way? Yeah. So there has been important work that started almost 60 years ago. And you are all probably familiar with Vernon Mountcastle's studies and the discovery that a cortex has a columnar organization, and then a few years later the demonstration of such columns and also properties of individual cells in the visual cortex with David Hubel and Torsten Weisel. I just used these as illustrations of the fact that already more than 50 years ago, much very important work was done which started out along several decades of work, where we gradually began to understand how the b asic coding worked in the first stages of visual cortex. This is a diagram from Felleman and van Essen, which many of you probably are familiar with. It shows the various visual areas of the brain. And at the bottom here, nearest the retina and the L GN, you have V1, which is the area where very, very much of the mechanistic work on visual cortex has been done. So, there has been a lot of advance here , but what about the rest of the visual system , or the rest of the cortex from V1? There's a long way up, and there are many, many areas. And you can go all the way up to the entorhinal cortex and hippocampus, which 2 often are put at the top because they are indirectly connected to almost the entire cortex. But as you go up, our understanding of how these systems work gets more diluted, to put it mildly. There are a few exceptions, and I will focus on one of them. We have, for example, seen that in some of the higher visual areas, there are neurons that respond specifically to objects or to faces. So there are some correlates to things tha t happen in the outside world. But by and large, it's very hard to interpret activity in these areas and then introduce one exception, which is the hippocampus, and now also the entorhinal cortex at the very, very top of the hierarchy, where cells actually have responses that reflect properties of the outside world in a relatively simple manner. So this started out in 1971, when John O'Keefe and Jonathan Dostrovsky recorded from rats that were running in open boxes or in various mazes , that's what they used at that time. And when these rats were running around to collect small rewards, then they were recording from the hippocampus of their brain. And so with recording electrodes inserted into the hippocampus, and then they could watch the activity on an oscilloscope, and this was stored. And at the same time, when they did this, they could then actually try to find external correlates of the activity. So they had no clue what they would ac tually find. But the findings were very clear, and I'll show you this movie that illustrates the concept of place cells. So what they found is that many, many of the neurons in the hippocampus actually respond to the location of the rat. This is a rat running around in a box that is 1 by 1 meter. It's running around and collecting small chocolate pieces. And a cable has been connected to the rat's head, and then this cable is then in contact with small electrodes that are inserted into the hippocampus. These electrodes are very, very small, about 17 micrometers wide, so small that they can pick up signals from individual cells. And we are going to listen to one single cell, and also see where this cell fires. [Recorded sounds play] So whenever you hear the sound, the cell is firing. And you can see this also in the red dots. So firing happens all the time when the rat is up in this area. And when the rat is in other areas, then there's no activity. So this is a typical place cell. It fires only at one place and location of the environment. Otherwise, it's quite silent. So this can also be shown in a color -coded manner, like here. The blue is very low activity -- no activity, actually. And the dark red is high activity. So this is the place field of the cell. It's the area where it fires. So what O'Keefe and his colleagues found during the subsequent years was that different cells had different preferred firing locations. And this led them, O'Keefe and Lynn Nadel in 1978, to suggest that place cells were actually the basis of an internal map or a cognitive map or a Tolmanian map, because the concepts were suggested by 3 Edward Tolman some decades earlier . In the activity of many place cells, the location of the animal was always represented. It was a dynamic map that always told the rest of the brain where the animal actually was. So this was 1978. And then many years passed, and some questions were resolved. But some of the really difficult questions were not resolved. And one of them is, “How does this place cell signal arise?” Because, you remember, the hippocampus is really at the top here. It's very, very far from any of the sensory receptors, whether they are visual or from other senses. So how is it possible for such a sharp signal to be generated? Because there are no space sensors on our body, so somehow this sharp signal, this location signal, must be constructed in the brain. And what areas and what mechanisms are involved in this construction process? So this is something that caught our attention, and which we started out with in our lab. And in 2004, after some initial work, where we found that it could not be in the hippocampus itself because you could disrupt the hippocampal circuit and still retain place fields in the hippocampus. And we started to go out of the hippocampus. And the obvious place to go was the entorhinal cortex because the entorhinal cortex leaves the hippocampus with the rest of the cortex. So what we did in 2004 was to implant similar electrodes into the entorhinal cortex, in the remedial part of the entorhinal cortex. This is a rat brain seen from behind, the brain stem here, cortex here. The colored area is the entorhinal cortex. And the upper half here is the medial entorhinal cortex. And what we did then was to insert electrodes particularly into the dorsal part because the dorsal part is the part of the entorhinal cortex that has the strongest connections to the dorsal part of the hippocampus, which is where almost all place cells have been recorded before. So with the obvious place to go, no one had ever touched this region because -- or actually, I don't know why that hasn't happened -- but most of the recordings in the entorhinal cortex have been in much more ventral areas, which are not connected to the same parts of the hippocampus. Staring out the recorder, we observed the following. So I'll show you another movie. And now, there's no sound. But you can see similar dots, white dots, that illustrate where the cell is firing. So a rat is, again, walking around collecting small chocolate crumbles. But now, you can see that the cell is not really firing at one particular location. It's firing at many locations. So at first, the signal doesn't really seem very special because it fires almost everywhere. But as we let the rat run around a little bit, you will notice that there are actually multiple fields, so the firing is not random. The cell fires at some locations, and between those locations, it is completely sile nt, so multiple firing fields. And the other thing is that those fields seem to be distributed more regularly than you would expect by chance. So the distance between these firing areas is quite regular. This puzzled us, 4 and we didn't in 2004 really understand what the pattern was. So what we did was to extend the space so that we could see the pattern better. And here is what you see. This is a later recording, but from a box that is much, much larger, 220 by 220 centimeters, compared to the 1 meter that we had used so far. And what you see here in gray is the path of the animal. So it's been really running around, visiting every possible place of the box. And in black, you see the dots that correspond to the firing locations of the cell. What you can see is that, again, the cell has many firing locations. But now, you can also see the pattern because it is really a triangular or hexagonal pattern that repeats itself over the entire space. This, you can also see here. This is from the original study. So here, I put red lines on top just to illustrate how this pattern is actually a repetition of equilateral triangles or, if you want, hexagons over the entire space. And because this formed a grid-like pattern, we call them grid cells. There was a grid that covered the entire environment. And the regular structure also introduced the idea that this might be a possible metric for the brain's neural representation of space , because it contains informatio n, both about distances -- the distance between the peaks -- and directions, namely the orientation of the grid axis. So this was in 2005. One of the things we asked when we had seen these grid cells is , “What is the variation between different grid cel ls? What are the properties?” And one of the differences between different grid cells is that they do, of course, not all fire in the same locations. So some grid cells fire in these locations. Other grid cells have peaks that are in between those. And that's illustrated here for three different grid cells that were recorded simultaneously. One is shown in blue, one in green, and one in red. And you can see that they are shifted relative to each other. So they have different x-y locations, or as we call it, they have different phase, different grid phase, different x-y locations. These are three simultaneously recorded cells, so here is the illustration of the recording electrodes, the tetrodes, and below, an illustration of how these cells could be arranged, so the blue one, the red one, and the green one. And there seems to be no clear relationship in how these cells were arranged because in any particular location, there could be grid cells o f all phases. So in other words, the entire space was represented at any possible location in the entorhinal cortex. So another property that varied between cells is the scale of the grid. And that is shown here. So this is a slice through a rat br ain, the saggital section. This is the hippocampus. And this colored area is the entorhinal cortex. This is the top. This is the bottom. So top is dorsal, bottom is ventral. And this shows how grid cells change from the dorsal to the ventral part. So as you begin at the top, then grid cells have a high frequency or a small scale. And as you go down, the scale increases and gets very, very large so that while the distance between 5 the peaks here is something like 30 centimeters, as you get down here, it may be several meters and get very difficult to measure, of course, because the environments aren't large enough. So unlike the phase of the grid, the scale is actually topographically organized in a sense, such that the scale increases as you go from dorsal to ventral. So for quite a while, we thought that this was a continuous change in the scale of the grid. But we couldn't really tell because we never had nearly enough cells from the same animal. And that changed in 2012, when we were able to r ecord many dozens of cells, grid cells from the same animal, up to 186 cells in the best case, from grid cells from the same animal. And then we could see that actually, it wasn't continuous. So this is illustrated here. So here, each dot here is one cell. And on the x-axis, you have the position along the dorsal-ventral axis, with dorsal here, the ventral here. And on the y-axis, you have the scale of the grid, or the distance between the fields. And what you can see is that, as you start from dorsal and go to what's ventral, yes, the scale increases, as we knew. But it's not a continuous increase. It's actually a stepwise increase. And the change can be described by only four steps in this case. And actually, when we recorded in the upper half, we never got more than four, perhaps five in one case. So it's a very low number of modules. And the first module, M1, is shifted very much to the dorsal part. So in the most dorsal part, you only have the first module. Then you recruit the second and the third and the fourth, but you still retain some cells from the lower modules, even if you go up to the higher ones. So one of the things we asked when we saw this was, “What is the relationship between these modules? What is the scale relationship?” What we did was simply to measure the scale for M1, 2, 3, and 4 and then the scale ratio, so M2 divided by M1, M3 by M2, and M4 by M3. And what we found was that this ratio is actually more or less constant. So the factor that you multiply the gri d size with, as you go from one module to the next, is the same whether you go from M1 to M2, M2 to M3, or M3 to M4. And that factor is approximately 1.42 or 1.4, same as square root of 2. So this is clearly a geometric progression, in the sense that you multiply by the same factor for each step here. So why is it like that? Of course, we don't have any experimental evidence for that. But several groups, both Vijay Balasubramanian’s group here and Andreas Herz’ group in Munich, have suggested that this is a way that actually causes an optimal representation of space with a minimal number of neurons. So that is possibly a good reason for why it could be organized in this way. These are grid cells. And then finally, in this introduction, I also want to say that there are other types of cells. Head direction cells, these cells were actually discovered already in 1985 by Jim Ranck at Brooklyn. And those cells were discovered in the 6 presubiculum, which is a neighboring structure. In 2006, we also found them in the entorhinal cortex. And I'll illustrate these cells here. So again, this is a sagittal brain section. Recording is in the medial entorhinal cortex here or here, two different cells. You can see that they have no preferred firing location, so not really any structure especi ally. But to the right, you see directional plots. So this is the firing rate as a function o f direction of the rat's head. And you can see that this cell, on two different occasions, fires only when the rat is looking to the left in the box. And this cell fires only when the rat is looking to the left and up in the box. So direction is represented in these cells. And in 2008, then we found out that there is yet another type of cell in the same system. These cells we call border cells because they fire specifically along the borders of the local environment. So this is, again, a box where a rat walks around. And blue means low firing rate. Red means high firing rate. And you can see that this cell fires only along the right wall here. And that happens even if you stretch the box in the x direction or even in the y direction. It just keeps firing along this wall. And we take the cell into another room; it fires now along the left wall instead. And if you insert a wall, it even chooses to fi re along that wall, too, on the corresponding side. So these cells, the border cells, they really only fire along walls or borders. If the rat is walking on the table, it will fire along the edges instead. But the cells really indicate where the local environment stops. Border cells are distinct. So a border cell can never become a grid cell, can never become a head direction cell, or vice versa, although border cells can be directional modulators often. There is a system in the entorhinal cortex, i n the medial entorhinal cortex, that consists of several types of specialized cells, grid cells, head direction cells, border cells. And so each cell type has a quite specific function. These cells were all discovered in rats. And then we found them also in mice. So this shows the phylogenetic tree for mammals. And one interesting observation was made a few years later by Nachum Ulanovsky's group at the Weizmann Institute, where they recorded or found grid cells in bats, in Egyptian fruit bats. And that was important because bats are actually on a different branch of the mammalian phylogenetic tree, suggesting that if they exist also on this branch, and with so similar properties -- it has mostly similar properties still, although they don't look so nice in the bats. But that suggests, then, that the grid cells must have emerged quite early in mammalian evolution. And consistent with that, then, they were later found in somewhat different properties in monkeys, in Elizabeth Buffler's group, and finally Mike Kahana with Josh Jacobs in the human brain in a virtual reality environment . So, they probably exist in ourselves as well. 7 So now, I've talked about time averaged maps. And I've indicated that there are different types of cells that fire at d ifferent locations or different directions if you average over time. But these maps, they are dynamic. They have to be updated whenever you move. So if you go from one place, like where I'm standing now, and if I move just a meter to the right, then a d ifferent collection of grid cells will take over. So this updating process happens all the time, so that the active map of the brain is actually one that corresponds to the location where you actually are. So how does this process happen? It is quite obvious that for this to happen, there has to be some sort of information that reaches the entorhinal cortex that tells the cells about how fast and what direction the animal is moving. So direction is fine. That could be signaled by head direction cells. But speed is more difficult to understand, from what we knew until very recently. So the question, then, is whether there is any signal that tells this part of the brain about how fast the animal is actually moving. So that movement in the outer world would correspond to, or be matched by, movement in the internal networks between cells in the brain. So we then started out to search for speed cells. This is many years ago, seven years ago, so it took a while to get it done. What we did was that we built a cart, called it a Flintstone car. And the reason is that the rat, which was running on this cart and is going back and forth on a linear track , a 4-meter-long linear track. And this cart was driven by the computer, so the computer decided the speed. But the rat had to run inside here, just like Flintstone walked or ran in his car. And the advantage was that the computer could then completely determine the speed, so we could run it at 7, 14, 21, 28, or whatever centimeters per second or what ever speed we wanted and then could see if there were any cells that actually responded to this speed. And that turned out to be the case. So this is illustrated here. So position on the track is on the x-axis. And then in gray, you see the speed of the rat, so increasing in this case. And then in green, you see the firing rate of one particular cell, which is increasing correspondingly. Same you see here. And here in the middle, the speed is changed. That's two meters. And you can see that the firing rate of that cell also decreases correspondingly. And similar changes here, with the raster plots, and here. And this is also summarized for five different cells here at four different speeds. And the firing rate is shown on the y -axis, different cells in different colors. And you can see that they all increase their speed, increase their firing rate along with the speed that the rat is running in the car. So if such cells existed in this Flintstone car, then they should also exist in the open field where we had done all the other studies. So we went back to those studies and analyzed some 2,000 cells or so. And indeed, there were also speed cells. Twelve different cells are shown here. 8 So first of all, you can see that these cells from the color -coded maps here, you can see that they have no particular firing preference. They fire almost anywhere in the box. So it's very evenly distributed. They spike all over the box. But to the right here, you can see firing rate as a function of speed. And you can see, then, that the firing rate for these cells increases quite linearly as the speed increases. And quantifying this, then, with about some 2,000 cells, 358 passed the quite strict threshold for a definition of speed cells. So this is using a shuffling procedure and taking only the cells that pass the 99th percentile of the shuffle data. And then 315, or 15%, pass the threshold where we expect only 1% by chance. So certainly, many cells in the medial entorhinal cortex do respond to speed. And other analysis also showed that these cells, they were distinct because they w ere neither grid cells nor head direction cells nor border cells. They were mostly their own category of neurons. Some such cells are shown here. What you see here is an extract of two minutes of recording. And the speed of the animal is shown in gray . And the seven different cells are shown with different colors here. And you can see for each of the cells how well the firing rate, which is shown on the y -axis, actually matches the speed of the animal. So just look here, for example, at the yellow example, where the cell fires -- almost exactly follows the speed of the rat. So, then, it's not surprising that you can actually use these inputs o r feed the relationship between speed and firing rate into a computer, do that for maybe one part, the first half of a recording session, and then let the computer, based on that, guess what is the speed of the animal, knowing the previous firing rates /speed relationship of these cells. And the guess of the computer is then shown here in blue. And then the actual speed on a later part of the trial is shown in gray. And you can see that you can actually predict the speed of the animal very well, even with a very low number of cells. The relationship to number of cells is shown here, and then the correlation with actual speed is shown here. And you can see that already, with five, six cells, you can do very, very well. So that means that you don't really need many of these cells. And because we know that most likely, many of them or most of them may be interneurons, they then project to many, many cells in the local circuit so that this information may be distributed very widely within the local circuit by maybe 10, 15% of the cells. So most likely, this information is available to the entire network. So then there is such a speed signal. But then they wonder, “Where does it come from?” because speed is certainly not something that originates in the entorhinal cortex. So what could be the possible origins of such a signal? And what we thought was that perhaps we have to go down to the deeper parts of the brain. And one area that might be interesting is mesencephalon, so the mesencephalic locomotor region is a very broadly defined, vaguely defined area of the brain, which is 9 down here in the upper part of the brain stem , and the cerebellum here. And this is the sagittal section. And this square box illustrates approximately where the area is. This is a magnification. And here is the location of recording electrodes. We also had optical fibers there so that we could actually stimulate in this area. And I will show you another movie that illustrates what happens if you stimulate in that area. So now, you see a rat here. It's a bit difficult to see. And in the upper right corner, you'll see a blue light. So when this blue light is on, stimulation is on. And you will then see what happens with the behavior of the animal. So the animal is just sitting there. Now comes light on. And you can see the animal starts running. Light off, it stops. And then light on, it starts running again, and so on. So in this area, if you stimulate the cells there, it's very, very common that the rats just instantaneously start to run. So this could be a possible origin on the speed signal. And indeed, if you look back in the literature, there is, for example, a paper from the Stryker-Niell lab from 2014, which was -- the study was actually performed for a different purpose. But it's very clear from one of the figures that some of the cells in this area may actually have firing rates, shown in black here, that are related to the instantaneous speed of the animal. So this is a mouse. And time again here, and you can see that the speed, shown in green, g oes up and down. And then the firing rate of one particular cell goes up and down accordingly, very much following the speed of the mouse. So there may be speed cells in this area. And to investigate that, we recorded many, many cells in the same area and focused specifically on the pedunculo-pontine tegmental nucleus within this mesencephalic locomotor region. So this area is then shown again here. This is an acetylcholine esterase stain so that you can identify the region, because it has a high dens ity of cholinergic cells. And then with the tetrodes in this area, there are many cells that responded like this. So here, you have time on the x -axis, firing rate on the y-axis. And then in gray, you see the speed of the animal. This is a rat. And in red, you see the firing rate of one particular cell. And you can see that the firing rate very much follows the speed that the animal is running, very much like the speed cells in the entorhinal cortex. And in the population of cells, if you plot the correlation between speed and firing rate on the x-axis, and then the number of cells with that correlation on the y-axis, then you see that this is not distributed evenly around zero, as you would expect if there was no relationship, but shifted to the right, where some cells have really high correlations with the speed of the animal. So there are such cells. But how could this speed signal reach the entorhinal cortex, if it does, because there's no direct connection from the PPN to the medial entorhinal cortex. So there are several routes. But perhaps one of the most important ones might be the one that goes through the medial septum, in particular the diagonal band, because the diagonal band, which is a kind of continuity of the medial septum into the 10 lower basal forebrain, has strong connections to the median entorhinal cortex, and particularly the vertical limb of the diagonal band. So there could be possibly a connection from the pedunculo -pontine tegmental nucleus to the diagonal band, and from there on up to the medial entorhinal cortex. So to test this hypothesis, we injected fast blue, a retrograde tracer, into the medial entorhinal cortex, and asked whether cells were labeled in the diagonal band. And at the same time, we injected an anterograde tracer, BDA, into the PPN, and then aske d if fiber terminals could be seen in the same region. And indeed, there was overlap. So this is from the diagonal band. And it's a bit hard to see, but what you can see here is in blue. Those are cells that are labeled because of the fast blue injection into the medial entorhinal cortex. And at the same time, you can see these fibers in between here, which originate from the anterograde tracer that was injected in the PPN. So they overlap. Of course, that doesn't prove that there is a synaptic contact. But it's not unlikely. So if this is an intermediate step between the PPN and the medial entorhinal cortex, there should also be speed cells. And indeed, that's the case. So recording in the diagonal band then, again, you can see that speed is shown in gray here. And then in green, you see the firing rate of one particular neuron. And you can see how well that neuron actually follows speed of the animal. So certainly, speed cells have firing rates corresponding to speed in a pretty linear manner. And again, you have right-shifted distribution, showing that there are many cells with very high correlations between speed and firing rate in the diagonal band. But then we wonder, if you go back to this, if there are actually functional contacts between the input from PPN and the cells that project to the MEC. We're still working on that, so I can't give you a final answer. But what I can show here is data -- it's actually just less than a week old -- where Miguel Carvalho and Nouk Tanke in our lab recorded from the diagonal band in response to stimulation in the PPN. So they injected AV virus into the PPN. The AV virus carried channelrhodopsin, and then light stimulation into the PPN resulted in responses in the cells in the diagonal band. So this shows time from light onset here. Light was on from 0 to 10 milliseconds. And then trial number here, and you can see that on each trial, the cell responded approximately 30 milliseconds after the onset of the light stimulus. And this suggests that it is a multisynaptic process, probably two synapses or so, because latency to discharge one neuron in a nonsynaptic manner is about 9 milliseconds with this channelrhodopsin. And if you have one synapse, then it is around 20 milliseconds, and so on. The point is that there is a functional connection onto a speed cell, suggesting that actually the diagonal band could mediate responses from the brain stem and up to the entorhinal cortex to provide the entorhinal cortex with a speed signal. In other words, locomotion signals come down from the spinal cord and are relayed up to the 11 entorhinal cortex to update the map with the amount of movement, how fast the animal is moving at any given time. I want to spend a little bit of time on the types of environments where the animals are walking because many of you may have thought that these small boxes where the rats are walking, and almost all of the experiments are done aren't ver y natural because rats are not really walking in boxes in their natural lives. So what would it look like under more natural circumstances? And one of the things that characterize s natural environments is that they are not only complex, but they are compartmentalized, so they're divided up into small sub environments. So we try to simulate that in the maze. In 2009, we trained animals first to walk in a square environment like I've shown you many times now. And then we divided the environment into many others so that the rats start in one end and then went back up and down, up and down, up and down, until it came out here in hairpin-like manner, so whenever we call it a hairpin maze. And then finally, it went back to the open environment again. And the question was when you divide the environment up into these alleys, “Is the grid pattern retained?” Is there a master grid that covers the entire environment? Or is it somehow changed?” And it turned out that it is somehow changed. So this is illustrated here. These are two different cells. First at the beginning and at the end, when the environment is open, you can see that there is a grid pattern. But when you break up the environment into these alleys, you can see there is no overall ov erlaid grid pattern anymore. Instead, the cells are firing at certain distances from the turning points. And those distances are similar on every single alley. So these ask if the grid is actually broken up and restarted whenever the animal makes a turn into a new alley. This suggested, then, that the grid is actually fragmented or consists of smaller parts that, in a way, are fused together. This is not only the case in this type of maze , because if you go back and look at grid patterns in large open environments, it actually also suggests that the grid is fragmented. So what you see is one typical grid cell here. And if you look more closely, if you draw lines between neighboring point s of the grid, you can actually see that the grid breaks somewhere along the diagonal in the middle here. You can see there's a turn in the middle here, which then suggests that these two parts of the environment are probably aligned with different cues o n the outside. So maybe this bottom part here is aligned with cues here, and this upper part is aligned with cues up here. And if you look more at the grid pattern from one o f the example cells I have shown, what you actually see is that even within thi s grid, which we thought was so regular, is actually differences in the orientation of the grid in the upper and the lower parts. So 12 this suggests that the gird isn't actually as we thought after all. And this is even more illustrated here. So this is an analysis of the average pattern of many, many grid cells that belong to the same module and were recorded simultaneously. And based on this analysis, this population analysis, it was possible to infer the deformations of the grid pattern. So you can see that up in the corners, both here, here, and here, a little bit here, the grid is actually quite distorted compared to what it is in the central parts, where the grid has a more circular structure. And this turns out to be very common. And what it illustrates is that grid patterns are actually very much influenced by the borders, maybe through the border cells. That, we don't know. But geometric borders -- walls in this case, and particularly corners -have a strong influence on the actual shape of the grid pattern. That, then, raises the question, “If the grid pattern is fragmented, like we saw in the hairpin maze -- or like we saw in these environments too, that they're actually somewhat different in different parts of the environment -- how are these parts, then, merged together?” What we did in one series of experiments was to try to train rats to walk in two different compartments that were divided by a wall . So compartment A and B, which each were rectangular, and then you get grid patter ns in each of them. And then after a long training, the wall was removed. And we then observed a grid pattern like this. And the question, then, was, “How can this grid pattern be related to the original ones?” So is it so that the original patterns are fused together into a single grid? Or is the new grid pattern perhaps an extension of one of the original ones, either the A or the B map? Or is a completely new map formed in a merged environment? It turned out that it is most likely to be some s ort of fusion because in this analysis, we took the merged box environment and correlated it with the original environments A and B and then color-coded this correlation, where we used a sliding box approach so that red or yellow is high correlation, and b lue is low correlation. This is the merged box on top of the A and B boxes. And what you can see is that this correlation is highest along the peripheral walls to the left and to the right – or, in other words, here on the left side and here on the rig ht side -- but very low in the center, where there used to be one. So that suggests that actually, the new map is the old map along the walls. It's retained here. But as the two original maps meet in the center, then the map gets much more different from what it used to be when there was a wall. So what is happening here at the fusion point? Well, some clues are provided by this. So what this analysis shows for the first half of the experiment, where there is a wall, 13 and for the second half, where t here is no wall, it shows all the fields of all grid cells, shown in either blue or red. So blue is grid fields that have low variation in the distance to the neighboring fields. It means that the neighboring fields, the six fields around, are all approximately at a similar distance, so that means very periodic. Red means high variation. That means that some fields are close, and some are much further away. This is what you would expect when there is a wall because the two grids on the two sides are not in phase. They are not aligned. But when you remove the wall, what you can see is that the red dots disappear here in the center, and you actually get similar distances to neighboring fields also in the center. This is also shown here. So this is just an average of all y positions. So you can see that before, when the wall is there, there is a high variance at the center, as expected. But when you remove the wall, this high variance disappears. And as you can see also here, it's very similar throughout. And that suggests that there's actually a formation of hybrid grid patterns, so that the distances are equal to neighboring fields also in the center. So even if the original maps are retained on the sides here, they somehow make a nice grid also in the center. And the processes underlying this, we still don't quite understand. But it's very clear that fragments of different grids actually meet in the center and then try to form a kind of compromise grid that has a smooth transit ion from one end to the other. All right, now I'll spend the last five minutes -- maybe seven, eight minutes -- to talk about development because also in this field, one may pose nature/nurture questions. So the spatial system here, is it “innate” in quotation marks, because you can never tell entirely, of course? Or is it also influenced by early experience? When it comes to place cells, shown here, and border cells, shown here, and head direction cells, shown here, it turns out that those have adult -like properties already on the first day when an animal can be tested outside its own nest. So this is that P17, when the rats are 17 days old. That's when they start walking outside and are willing to walk in an open environment. And already on the first exposure, then they have both nice border fields, place fields. And head direction cells can even be seen before, even before they actually open their eyes. But it's different with grid cells because there is some periodic pattern from the beginning. But it doesn't really form the nice, periodic fields of an adult animal until the animal is about four weeks old, at around P28, plus/minus. So this suggests that grid cells are a bit later to develop. They really require the entire circuit. And one may wonder whether that also leaves a window for experience to shape the network when they are so late to develop. To address this question, what we did was to train animals in three types of environments. So some animals were raised all their life, or at least the first two or three months, in a sphere. And the reason for putting them in a sphere is that there 14 are no vertical boundaries at all. So there's nothing to refer the grid patterns to, so they couldn't really anchor or calibrate the grids to the external environment when they were raised here. Some animals were raised in a cube environment. Both of them are opaque, by the way. But these have references, so they can use the corners and the walls to anchor their grids. And then some were raised in normal or spatially-enriched environments. And it turns out that those that were raised in the sphere, they don't really have any clear grid cells when you, for the first time, put the animal into an open environment at adult age. So this shows that most grid-like cells that we could find, the upper 20% of distribution. And the red shows the statistical threshold for passing as a grid cell. And the two cells just past -- you can't really see much of a grid pattern here because they're really, really bad. This is different in the cube -raised groups, where you have clear grid cells, and also in the enriched group, where you have clear grid cells. And the number of grid cells, compared to a statistical threshold here, and chance level at 1%, is shown here. So in the sphere group, it's just above and not really significantly different. And in the two other groups, it's a high number of grid cells. And this also is on the second day of testing just the same. But if you then continue, in the end, the number of grid cells increases. And after four, five, six days, then you start to get them also in the sphere group. So it is temporary. They finally recover. Perhaps this suggests that experience isn't really necessary for grid patterns to form. But you may need a grid pattern to learn how to anchor it to the geometric references of the outer world. So experience is important, perhaps most important for actually aligning the grid with the rest of the environment, which you can't do if you're raised in a sphere. But this can be learned, even at adult age. And finally, for the very, very last couple of minutes, I want to say a little bit about development of the entorhinal circuit. So when are these cells actually born, and how do they mature? So when they're born, that can be addressed in several ways. But one common way to address this is to inject BrdU, which you can inject I V into the mother. And this will also be taken up by the embryo, and BrdU will label cells that are in the S phase of the cell division cycle. So you can label cells within a temporal window, maybe plus/minus six hours or so. And then if you do these injections on different embryonic days, you can actually find where are the cells that are born at differe nt ages. So this is when cells are born. But then maturation of the cells is, of course, much slower. And one way to measure that is to measure double cortin levels, which 15 characterize cells up until a late stage. When they reach maturity and finish branching, and so on, then doublecortin disappears. One could count the number of cells that are doublecortin negative and define them as mature. And then you can ask, are these Reelin positive, Calbindin positive, and so on. And parvalbumin, this is for excitatory cells. For inhibitory cells, we will also see that the level of parvalbumin will increase as the cells mature. So how is this development? This is shown here. If you focus first on the left column, this shows the development across time, embryonic days shown on the x -axis. It's a bit difficult to read. Embryonic days measured is shown on the x-axis. And then on the yaxis, you have the fraction of cells that are born. And you can see that for parvalbumin-positive or Calbindin-positive cells, that means the interneurons and the granule cells, then different colors. They indicate different dorsal-ventral levels of the medial entorhinal cortex. And dorsal-ventral level doesn't matter because cells are born gradually from around E12 to E16. But they are evenly distributed across dorsalventral levels. This is not the case for the stellate cell s, those that are positive for Reelin. Because then the cells for the firstborn are those that are at the most dorsal le vel. Those are shown in blue. And gradually, the cells then go from dorsal towards ventral, which are shown in red. So there's a gradient of cells that begin at the dorsal and end at ventral levels. This is also reflected in the maturation of the cells, measured with the doublecortin levels. So first mature are the ones at the dorsal, and last mature are the ones at the ventral level. It's different for the pyramidal cells and the interneurons, because even if they are born, there is no dorsal-ventral gradients in where they're born, there is actually maturation gradient that begins dorsally and goes towards ventral ly for each of these. So what could be underlying this? I skip this slide, but what we did was to find out what caused this pattern. We blocked all excitatory activity in the system by using a DREADD approach. So we injected on P1, just after the mice were born, we injected into the medial entorhinal cortex a combination of viruses that caused HM4D expression in the excitatory cells in the entorhinal circuit. And then, we implanted osmotic mini -pumps at P14, and then from P14 to P20, CNO was injected subcutaneously. And this was then enough to actually silence the cells via this DREADD receptor, silence the excitatory cells. And what happens if you silence the excitatory cells? What you see here is for the Reelin-positive cells, for the Calbindin-positive cells and for the parvalbumin -positive cells. And then you see the fraction of mature neurons. So you can see that, when you block excitatory activity, there's no effect in the stellate cells. So what you should see here is that the green ones, which are the blockade of entorhinal cor tex, or the two green ones here, they are not different from the blue ones, which are t he controls. 16 But when it comes to the Calbindin and the parvalbumin cells, then you can see that the number of mature cells is zero. They are completely blocked. So the cells don't mature as long as you don't have excitatory activity in the circuit. I'll skip this. And then if you block hippocampal excitatory activity, there is no effect. So it's really the local entorhinal blockade that causes the maturation to stop in these two cell types. And what we are now doing is that we are trying to in activate selectively the stellate cells so that we most likely will find that activity in the stellate cells is actually necessary for maturation of the pyramidal cells that express Calbindin and the fastbinding interneurons and for the circuit to develop . So in other words, the stellate cells mature regardless of activity, just follow their own program. But their activity may be necessary actually for the rest of the circuit to finish. So with that, I want to first of all say that there are a lot of people that have been involved in this research. And first in all work, of course, May-Britt Moser. But then for the grid cells, the early work on grid cells, Marianne Fyhn, Torkel Hafting, and Sturla Molden were very important. And also the later work modules, Tor Stensola and also several external collaborators like Alessandro Treves, Bruce McNaughton, Carol Barnes, and especially Menno Witter, who helped with all the anatomy. And speed cells -- I talked about those -- that was the work of Emilio Kropff, the early work and the discovery of the speed cells in entorhinal cortex. And the more recent work with tracing them into the brain stem was Miguel Carvalho and Nouk Tanke, and again, a collaboration with Menno Witter. And then the fusion of the grid maps, Tanja Wernle and Maria M orreaunet particularly, and the earlier work by Dori Derdikman. And then the developmental studies, especially Ingvild Kruge. And then the last part on the role of activity in stellate cells for development of the system, that was especially Flavio Donato. So with that, I want to just leave the conclusions here. I'm not going to repeat my own lecture. But I think it's time to stop. And I thank you all for the attention. [Applause] DR. BRAINARD: We have time for some questions and discussion. And there are microphones that will kind of be passed around – [Pause] AUDIENCE MEMBER 1: That was certainly a pleasure. Are your experiments in this field, when the rat was raised in a sphere, and initially, there were no grid cells , I noticed that within the hexagonal pattern, there was a pattern. And I wonder whether it means anything to you. 17 DR. MOSER: Well, that's a good point because it is -- let's see, here, there is some sort of pattern. The cells have firing fields. But what it lacks is a hexagonal pattern. It's also not very stable. So to some extent, there might be some grid pattern, especially along the walls, but especially in the middle, it's very disorganized. And perhaps this is what you would expect if the cells are not able to refer to the external walls or boundaries actually to find out where should they fire and where should they not fire. So this is a learning process. They have to learn actually where to fire consistently in relation to the outer environment. But I do think that the grid pattern itself is not disrupted. It's just that it's sort of floating around because I can't put it in the right place. And that's probably why you see that there is some sort of pattern. It's just not hexagonal. DR. BRAINARD: Russell? DR. RUSSELL EPSTEIN: I'd like to ask a psychological question. And it's about humans. So this seems like an exquisite system for the animal to keep track of whe re it is and keep track of how many steps it made and to path integrate. And we seem to have these cells in humans. But humans are terrible path integrators. So what is the system doing in our brains? DR. MOSER: Well, I think it's more than just space . So first of all, what these cells are doing -they are doing path integration. I actually think they are doing even in humans. But they operate at a local scale. But what they also do, both in humans and in rats, is that they are constantly updat ed by other types of input, and especially visual inputs, I would say, in humans. So there is a constant interaction between the path integration input and especially visual input that sort of recalibrates all the time. But I'm not sure I would agree that -- if that's what you're saying -- that humans don't use path integration input. I mean, not very good at it, and it causes a lot of error. But I think if you only had the visual inputs, for example, and we would have to estimate position by triangulation in some way, that would also cause a lot of error. So I think what is good is actually the interaction between locomotor and visual and optic flow signals, many, many sources. But then what is it doing in humans? I definitely think it's doing other things, too, both in humans and in rats. And one idea we have that we are currently testing is that because you have different modules of grid cells, they may actually be involved in memory because one of the properties of hippocampal memory is that it has such a high capacity. It's very different from entorhinal cortex, where you have more or less one pattern that is expressed all the time. But if you have a small number of modules, and you can combine them in different ways because all of them have different x -y positions, it is almost like a combination lock, where you have different solutions for each of them, and you can get a very, very 18 large number of combinations by just using the combinations of these grid cell modules onto the hippocampus. There could be a way that actually you can create thousands of different maps in the entorhinal cortex based on just very few in the entorhinal cortex. And of course, that's good for space, but it's also definitely good for memory -- but still a hypothesis. DR. BRAINARD: In the front here? AUDIENCE MEMBER 2: I have a question about the first two videos, space cells. What determines the initial state of the grid that's set up? Is it the first glimpse, or is it light source, or is it something related to magnetic pulse? And then what if you make a subject dizzy before -- how does the grid emerge from the dizziness? What happens if the subject is blinded or deafened? Because I've heard that these same cells are implicated in auditory location also. DR. MOSER: Yes, so how does grid pattern emerge? One of the experiments we did -- one of the first experiments, actually -- was a pretty obvious experiment: test the grid cells in darkness, so you remove all visual inputs. And consistent then with a role for locomotion, they still fire. If you put the rat into an environment with light and then turn off the light, when it goes back, it will still fire in the same firing locations. And this question is then what happens if you put them into a new environment , the very, very first time they walk there. Then we can also see that as long as the environment isn't very large -- so you have multiple grids that fuse together -- the grid pattern is expressed from the very beginning, which is then consistent with a locomotor role for this to be instructed. So it means that I think as long as you can trace your position back to where you started, and you can sort of use the surroundings at the same time, you can actually have the grid pattern express itself right from the beginning. And I think that happens under all conditions. If you disorient the rat, of course, the grid pattern would perhaps still form. But it will be misaligned with what it usually used to be. DR. BRAINARD: Over here, in green. Yes, you. AUDIENCE MEMBER 3: I can just talk loud. Could anything be learned in the subjects where you arrested the development of certain neurons from their behavior rather than from what the cells did? DR. MOSER: Sure. I mean, whether the arrest of development has not only physiologic al but also behavioral consequences. That's an obvious thing to look at, so that would be a next step. So what you have shown now is that for maturation of the circuit, you need this activity in certain subsets of the circuit. And then you want to see how does this affect the development of the different types of cells, the border cells, the grid cells, and so 19 on. And finally, one of course wants to know how this affects the navigation behavior of the animals. And to do that, of course, you can't only use these open envir onments. You need to use specialized mazes. So that's another level of complexity, but certainly something that we'll do down the road. DR. BRAINARD: Down in the front there. AUDIENCE MEMBER 4: You mentioned orientation cells and speed cells. Do you ever see velocity cells? In other words, a combination of speed and velocity? And in bats, is it a 3-D grid or a 2-D grid? DR. MOSER: Yeah, those are two different questions. But then with the combination of speed and direction -- that's what we expected to find, actually. We didn't expect to have it in two different types of cells. That is still somewhat surprising. But that is what the results show. So speed is encoded pretty much independently of direction. And then you have direction in the head direction cells. But this is then integrated when they meet, probably in common with common inputs. When it comes to the other question, in bats -- it was the 3-D question, was it? Yeah, so the bat studies that I referred to, they were done on flat surfaces. The bats actually were crawling. So then, you could observe a two -dimensional grid pattern. But of course, bats fly, and that gives the opportunity to actually ask whether grid patterns are three-dimensional. So the group in Israel is doing that. What they have published so far is that head direction cells are actually three -dimensional. So if they are flying, or even if they're just walking, then they also respond to not only horizontal direction, but also in the vertical dimension. And since that happens for the head direction cells, it's certainly not unlikely. And it's also for place cells; they are spherical. Then it's not unlikely that a three-dimensional grid pattern is also there. So some preliminary data suggests that that may be the case, but the details are still to be seen. So it hasn't been published. DR. BRAINARD: Over there. AUDIENCE MEMBER 5: So I wondered a little bit about the generalizability of the results to other species— DR. BRAINARD: [Interposing] Use the microphone. It's coming over. Thanks. AUDIENCE MEMBER 5: --in particular, to humans, and about the discrepancies between the human and the rodent literature with regards to the hippocampus. If I think, for example, about patient HM, who had, of course, a bilateral hippocampectomy, including entorhinal cortex on each side. Patient HM has been through several neuropsychological testing, and never was it reported that he had any problem, let's say, orienting itself in the room or finding out where his head is facing in space. He had, of course, severe episodic memory problems . But in terms of navigation and in 20 terms of orientation, I don't remember anything being reported in that regard. Do you have any comments about that? DR. MOSER: Well, it depends on how you define navigation. So if you define navigation in the sense that very simple path integration, where you just walk a certain direction and then make a turn and then are able to come back to the same location. This was measured by - - and his group in other patients. And lesions that involve very much of the hippocampus and entorhinal cortex don't really seem to have much effect on that, which is not very surprising because I think this is such a simple computation that you can probably do it with many different brain systems. When it comes to more complex navigation, actually finding the way in more complex environments, this you may know better, and Mike may know better. But actually, I thought that -- at least if you read the initial reports -- that he was, for example, not able to find his way back to the bathroom and things like that. So I think it's more or less consistent. But one has to remember that space and navigation involve very large areas of the brain, that you can do it in many different ways that can compensate for each other if the tasks are too simple. Otherwise, I should also mention that the grid cells that are measured, at least in monkeys, are somewhat different than the grid cells in rodents because they respond apparently to eye movement. And perhaps they also respond to locomotor movement. That, we don't know yet. But eye movement, so far, hasn't been studied in rodents. And it's unclear whether there actually is any relationship -- probably not in the rodents. DR. BRAINARD: We’ll take one last question there. AUDIENCE MEMBER 6: Hi. I have a question about the development environments as well. I was really surprised that, for the enr iched environment, you didn't have more place cells. And it actually looks in your other slide that the number of place cells went down over the course of development for enriched environments. So is this a phenomenon that you see over - - or what's up with the complex environments? DR. MOSER: First of all, that's a good observation because we wouldn't expect it to go down. The reason is that the number of cells on this last day is very, very low. So I think you just have to wait and see until we have enough data for all of the groups. What I would expect is that these two should -- at least the enriched group -- shouldn't be below the cube group. That would simply not make sense. So I think when we get enough cells in the enriched group, which will happen in a few months, then I would expect those two to be similar. DR. BRAINARD: Let’s thank Dr. Moser again. [Applause] [END RECORDING] 21