WEBVTT 00:00:00.017 --> 00:00:03.737 You look worried, Edward. Yeah, I'm not sure what's coming now. 00:00:05.257 --> 00:00:09.797 This is the Convergent Science Network podcast. So it's good cop, 00:00:09.877 --> 00:00:11.537 bad cop, and Paul's the bad cop. 00:00:11.877 --> 00:00:15.317 Leading researchers in the domain of neuroscience, brain theory, 00:00:15.477 --> 00:00:19.817 and technology are interviewed by Paul Verschure and Tony Prescott. 00:00:19.917 --> 00:00:22.437 After he's finished beating you up, I'll give you some easy questions. 00:00:24.477 --> 00:00:31.197 So, all right. So this is Paul Verschure, together with Tony Prescott for the 00:00:31.197 --> 00:00:36.577 Conversion Science Network podcast that we're recording at our BCBT summer school 00:00:36.577 --> 00:00:37.957 here in Barcelona, 2015. 00:00:39.097 --> 00:00:44.877 And our guest today is Edvard Moser, who gave a fantastic talk this morning 00:00:44.877 --> 00:00:47.577 about how the brain knows about space. 00:00:47.577 --> 00:00:54.637 And you started your talk emphasizing this whole challenge of combining psychology 00:00:54.637 --> 00:01:00.977 with physiology to find, if you want, this sort of physical mechanistic perspective on psychology. 00:01:01.297 --> 00:01:08.097 Is that really the motivation that drives this work? Yeah, in a general sense, it has always been so. 00:01:09.197 --> 00:01:16.717 When we started out as psychology students many years ago, 00:01:16.837 --> 00:01:27.677 it wasn't possible to say much about the physical substrate of psychology or behavior in any sense. 00:01:27.677 --> 00:01:31.137 But still that has been a major driving force. 00:01:31.337 --> 00:01:40.297 The fact that I ended up in work on space is kind of a coincidence. 00:01:40.377 --> 00:01:47.537 It's partly because I started out in the hippocampus and partly because it turned 00:01:47.537 --> 00:01:55.517 out that this part of the brain has cells that are so directly related to what's going on in the outside. 00:01:55.517 --> 00:02:02.157 It's actually an easy way into the cortex, but it could have been any function. 00:02:02.337 --> 00:02:06.497 So, I mean, whether it's space or if it's some other cognition, doesn't matter. 00:02:06.637 --> 00:02:16.417 I think what is the underlying drive to me is that it is informative about the workings of the cortex. 00:02:17.257 --> 00:02:23.477 So, in a more general sense, it helps us to begin understanding how the cortex 00:02:23.477 --> 00:02:27.037 computes, how functions might arise. 00:02:27.197 --> 00:02:32.677 Of course, only an early beginning, but it's an easier place to start than many other brain areas. 00:02:32.997 --> 00:02:36.557 But now when you made the decision to then go for hippocampus, 00:02:37.237 --> 00:02:41.637 was it in any way, let's say, inspired by the cognitive behaviorism of Tolman? 00:02:42.457 --> 00:02:49.237 To some extent it was. It was inspired by the fact that much of, 00:02:49.517 --> 00:02:57.457 this was in the 80s and early 90s, and then there was a huge interest in LTP, 00:02:57.657 --> 00:03:00.657 long-term potentiation, and its relation to memory. 00:03:00.657 --> 00:03:03.757 And of course memory was strongly linked to the hippocampus. 00:03:03.757 --> 00:03:10.457 So there was much interest at that time in finding a cellular mechanism of a 00:03:10.457 --> 00:03:12.637 behavior, which then in that case was memory. 00:03:12.917 --> 00:03:18.877 So that was the background for starting with the hippocampus. 00:03:18.917 --> 00:03:24.217 At that time there was not much work or interest yet in neural networks. 00:03:24.217 --> 00:03:30.177 That came or increased quite a lot during the 1990s, but nonetheless, 00:03:30.477 --> 00:03:36.777 the possibility for bridging two levels was perhaps more developed in hippocampus 00:03:36.777 --> 00:03:38.157 than most other parts of cortex. 00:03:38.797 --> 00:03:44.097 But in that sense, I was a bit surprised that you didn't mention Pavlov as a 00:03:44.097 --> 00:03:49.837 source of inspiration, because what always struck me in Pavlov was that he had to make that decision. 00:03:49.837 --> 00:03:52.797 Vision like here the dog has expectations so what 00:03:52.797 --> 00:03:56.997 do i do i'm going to speculate about it or i'm going to build a physiology of 00:03:56.997 --> 00:03:59.897 this of the psychic reflex as he called it right yeah 00:03:59.897 --> 00:04:06.577 no no that's pavlov was uh was really uh he really changed the field just in 00:04:06.577 --> 00:04:12.297 that sense that he dared to link the two levels so um i mean i could have mentioned 00:04:12.297 --> 00:04:17.897 pavlov to my history i lasted about four minutes So I couldn't mention everyone, 00:04:18.097 --> 00:04:26.397 but of course, Pavlov is a major part of the history of physiological psychology. 00:04:26.917 --> 00:04:35.777 So now your entry point into, let's say, this neural substrate of psychological 00:04:35.777 --> 00:04:38.177 function is hippocampus, right? 00:04:38.217 --> 00:04:41.697 So you started looking at place cells and place cell responses. 00:04:41.697 --> 00:04:45.357 So, however, that was not where you ended up. 00:04:45.477 --> 00:04:49.717 So, what was the trajectory there of, let's say, discovery that brought you 00:04:49.717 --> 00:04:51.097 to these extra hippocampal areas? 00:04:51.797 --> 00:04:57.217 So, it began with place cells. And around 1990, 00:04:57.617 --> 00:05:05.137 it was a common view that place cells were formed quite strongly by intrinsic 00:05:05.137 --> 00:05:07.237 processes in the hippocampus. 00:05:07.237 --> 00:05:13.277 And the reason was that at that time, no really specific spatial signal had 00:05:13.277 --> 00:05:17.197 been discovered in the entorhinal cortex outside the hippocampus. 00:05:17.737 --> 00:05:24.977 So it began with, as I mentioned in the lecture, a study where we disrupted 00:05:24.977 --> 00:05:29.517 the intrinsic hippocampal circuit and then saw that in the remaining part, 00:05:29.637 --> 00:05:30.877 there were still spatial signals. 00:05:30.917 --> 00:05:35.297 So it sort of forced us out to the next stage, which was the entorhinal cortex. Cortex. 00:05:35.577 --> 00:05:38.917 And then, uh, because we had, uh, 00:05:39.631 --> 00:05:45.471 and that collaborated strongly with a neuroanatomist, Menno Witter. 00:05:46.331 --> 00:05:49.911 Who has later moved to our institute in Trondheim. 00:05:50.011 --> 00:05:55.931 Then we had an expert on how the entorhinal cortex was organized and what would 00:05:55.931 --> 00:05:58.531 be the best way to target electrodes into it. 00:05:58.591 --> 00:06:05.151 So we dared to jump into that area, and then suddenly we found spatial cells, 00:06:05.151 --> 00:06:11.251 cells and later found also that they were actually quite strikingly hexagonally organized. 00:06:11.751 --> 00:06:17.191 But now, how many observations did it take for you to be convinced that these 00:06:17.191 --> 00:06:22.091 cells had these very specific properties that you found? Now that took a while. 00:06:22.331 --> 00:06:26.851 I mean it was gradual because we realized quite early that they had spatial 00:06:26.851 --> 00:06:30.531 fields like place cells in hippocampus. 00:06:30.611 --> 00:06:36.071 And we also noticed that there There were regular patterns, and we had, 00:06:36.151 --> 00:06:41.071 along with the first paper that we published in 2004, we noticed that it was 00:06:41.071 --> 00:06:44.831 extremely regular, much more than you would expect by chance. 00:06:44.851 --> 00:06:50.711 We showed that, but the data were not sufficient to really tell what kind of 00:06:50.711 --> 00:06:54.091 pattern it was, and that required bigger environments. 00:06:54.351 --> 00:06:57.031 And then the year after, we then tested them in larger environments, 00:06:57.031 --> 00:07:02.991 And then it was very clear, really. 00:07:03.131 --> 00:07:08.611 So, I mean, that didn't take a lot of work, 00:07:08.731 --> 00:07:14.991 but we also needed to rule out pretty obvious things like whether the grid pattern 00:07:14.991 --> 00:07:19.251 maybe was an artifact of some part of the electronics of the system or so, 00:07:19.271 --> 00:07:23.651 because it was so regular that then alarm clocks started to ring. 00:07:23.771 --> 00:07:26.051 But that was not the case. 00:07:26.711 --> 00:07:32.551 But now, I remember these first publications that came out in 2005 about on 00:07:32.551 --> 00:07:36.291 the grid cells, it still looked like a very risky proposition. 00:07:36.411 --> 00:07:41.371 It looked like iffy in some sense, like, oh, well, maybe they're over-interpreting this data. 00:07:41.591 --> 00:07:48.051 So you found these cells in an medial entorhinal cortex, supposedly an input 00:07:48.051 --> 00:07:51.751 station to the hippocampus, even though this has an interesting twist a bit later on. 00:07:53.571 --> 00:07:57.571 So, you say, okay, there's a grid-like response, they have a triangular kind 00:07:57.571 --> 00:08:03.651 of response field in the environment with a certain facing and orientation and spatial scale. 00:08:04.251 --> 00:08:10.171 So, but how much resistance did you then receive in actually getting that published? Yeah. 00:08:10.842 --> 00:08:16.802 Well, not a lot of resistance. Actually, people tended to believe it right away. 00:08:17.022 --> 00:08:24.202 So I think people were amazed, but there was very little skepticism. 00:08:24.422 --> 00:08:29.082 And I think the data were quite clear. I mean, you could see it in individual 00:08:29.082 --> 00:08:32.042 cells. It didn't really depend on any sophisticated analysis. 00:08:32.382 --> 00:08:39.882 So it was very hard to actually think of alternative ways that you could get this data. 00:08:40.842 --> 00:08:45.742 You mentioned that you had the help of somebody who knew entorhinal cortex. 00:08:46.122 --> 00:08:50.462 So is that why you found these? Because people had looked before and hadn't 00:08:50.462 --> 00:08:52.662 found anything that was a signature for space. 00:08:53.062 --> 00:09:01.102 Yeah, and also what we did was that we started recording in a more dorsal, 00:09:01.122 --> 00:09:08.482 more superficial part of the entorhinal cortex where no one had recorded before. 00:09:08.482 --> 00:09:14.942 And in the collaboration with this neuroanatomist, 00:09:14.962 --> 00:09:22.622 it was possible to target the electrodes precisely to an area that had the maximal 00:09:22.622 --> 00:09:27.422 connectivity with the play cells that had been recorded in the hippocampus. 00:09:28.182 --> 00:09:33.822 What had been done in the earlier studies was that people recorded in areas 00:09:33.822 --> 00:09:36.482 that were too ventral, too deep into the brain. 00:09:36.482 --> 00:09:41.882 And they were probably also grid cells, but because the scale is so different, 00:09:42.102 --> 00:09:46.202 the distance between the fields is different and the fields are so much bigger, 00:09:46.342 --> 00:09:50.782 then when they recorded in standard size boxes, they didn't see the periodicity 00:09:50.782 --> 00:09:52.522 simply because they didn't have enough fields. 00:09:54.881 --> 00:10:02.181 So, okay, you discover these cells, people buy it, and they're convinced because 00:10:02.181 --> 00:10:03.341 there's no alternative explanation. 00:10:03.821 --> 00:10:07.561 Yes. But now, what do you see as their key properties? 00:10:07.781 --> 00:10:12.021 And how do you see these key properties organized in this piece of brain? 00:10:14.621 --> 00:10:20.861 Well, it has several key properties. It's quite actually organized in the sense 00:10:20.861 --> 00:10:25.881 that they vary along several dimensions. They have different phase or XY firing 00:10:25.881 --> 00:10:28.581 locations, they vary in scale, they vary in orientation. 00:10:29.381 --> 00:10:31.961 So one of the key properties, it 00:10:31.961 --> 00:10:35.681 has turned out later, is that they are organized in what we call modules. 00:10:35.981 --> 00:10:39.561 So clusters of cells with very similar firing properties. 00:10:40.641 --> 00:10:44.221 So there are at least four or five of them, maybe as many as ten, 00:10:44.321 --> 00:10:52.321 of cells that within each module, the grid cells behave in a very rigid way. 00:10:52.321 --> 00:10:58.061 So that two cells that, for example, have similar phase or similar firing locations 00:10:58.061 --> 00:11:01.041 in one environment will also have it in another. 00:11:01.321 --> 00:11:07.661 If they have a special orientation difference, then they would have it in a different one too. 00:11:07.801 --> 00:11:13.541 So the whole map is very, that's another very salient property of the network, 00:11:13.641 --> 00:11:14.961 that it's extremely rigid. 00:11:14.961 --> 00:11:19.501 So you can almost take the map from one environment or from one grid module 00:11:19.501 --> 00:11:24.061 and apply it onto another environment and you will see the same relationship. 00:11:24.381 --> 00:11:29.541 And that's probably a property you would expect of any system that serves at 00:11:29.541 --> 00:11:35.941 least partly as a metric for space because you don't want to reinvent that mechanism 00:11:35.941 --> 00:11:42.481 for every representation of maybe several thousand environments that you have stored in the brain. 00:11:43.801 --> 00:11:46.861 So that at least two very important 00:11:46.861 --> 00:11:50.741 properties of the network but 00:11:50.741 --> 00:11:53.861 now these cells don't emerge by magic right 00:11:53.861 --> 00:11:59.341 they themselves are also dependent on on external inputs yeah of course they 00:11:59.341 --> 00:12:04.941 are embedded in a wider network and and first of all although we believe that 00:12:04.941 --> 00:12:11.561 the clues to the the hexagonal pattern lies in the cortex or maybe even in the 00:12:11.561 --> 00:12:12.781 entorhinal cortex itself. 00:12:12.861 --> 00:12:15.561 It cannot arise in isolation. 00:12:15.941 --> 00:12:18.801 So it depends, for example, it must depend on, 00:12:19.485 --> 00:12:24.885 on both speed and direction inputs that are likely to come from outside because 00:12:24.885 --> 00:12:31.445 there's no other way that you can actually create a dynamic map that reflects 00:12:31.445 --> 00:12:35.305 the distances that an animal moves in the environment. 00:12:36.045 --> 00:12:41.065 So fundamental inputs are information about speed and direction, 00:12:41.285 --> 00:12:43.565 instantaneous speed and instantaneous direction. 00:12:43.565 --> 00:12:52.625 In addition to that, so this is what is required to generate a map that is based on self-motion. 00:12:52.725 --> 00:12:57.065 But in addition, you need to calibrate that map all the time against other sensors, 00:12:57.165 --> 00:12:58.285 for example, visual inputs. 00:12:58.605 --> 00:13:02.585 So there is probably a continuous correction process going on all the time as 00:13:02.585 --> 00:13:04.985 well. What do you mean with correction in this case? 00:13:05.365 --> 00:13:10.605 Yeah, so path integration is a clue here. So the grid cells, 00:13:10.785 --> 00:13:19.585 if you let an animal walk in an open environment, it may have a grid pattern. 00:13:21.405 --> 00:13:27.165 So if there are lots of visual inputs available, then the grid pattern will 00:13:27.165 --> 00:13:32.665 use those visual inputs for the cell to fire at the same locations all the time. 00:13:32.685 --> 00:13:36.625 So that's a stable grid pattern. If that input is not available, 00:13:36.865 --> 00:13:42.525 it will start to drift over time, because even if it uses the animal's own motion 00:13:42.525 --> 00:13:45.645 to generate an approximate firing map, 00:13:45.865 --> 00:13:51.145 then there are occasional errors and they add on to each other unless you have 00:13:51.145 --> 00:13:54.045 other inputs that tell you that now you're drifting off. 00:13:54.045 --> 00:13:59.185 And what I believe happens in real life is that when lots of other cues are 00:13:59.185 --> 00:14:04.145 available, like visual cues, visual information, then this is used to sort of 00:14:04.145 --> 00:14:08.585 get the motion-based map on track again. Mm-hmm. 00:14:09.204 --> 00:14:13.764 Yeah, so this would also allude to, let's say, these experiments you've done, 00:14:13.804 --> 00:14:16.944 even though you didn't discuss them here, where you, for instance, 00:14:17.064 --> 00:14:18.884 morph the environment slowly, right? 00:14:18.884 --> 00:14:24.104 Where you can really sort of try to show how this kind of integration of other 00:14:24.104 --> 00:14:30.364 sensory states, like visual information, with space might occur in this whole 00:14:30.364 --> 00:14:32.764 loop that then starts with the entorhinal cortex. 00:14:33.184 --> 00:14:39.464 Yeah. So, but then do you still see that there is a special role for grid cells 00:14:39.464 --> 00:14:40.584 in that integration process? 00:14:40.764 --> 00:14:43.564 Because it means, okay, here I have sensory states in the world. 00:14:43.584 --> 00:14:45.644 They come in over my lateral interanal cortex. 00:14:46.164 --> 00:14:50.904 I have heading direction and velocity driving my grid cells, tells me about space. 00:14:51.144 --> 00:14:54.484 This information gets further processed in the hippocampal loop. 00:14:54.844 --> 00:14:59.404 But then do you see these two sources of information as being equally weighted 00:14:59.404 --> 00:15:05.164 in that integration? or do you see the grid cells as having a higher priority in that process? 00:15:06.944 --> 00:15:10.964 Well, I wouldn't say that one has a higher priority than the other. 00:15:11.104 --> 00:15:17.004 I mean, the grid pattern is probably intrinsically generated and as such is 00:15:17.004 --> 00:15:23.204 quite fundamental, but it needs that other input to be aligned to the environment. 00:15:24.304 --> 00:15:29.364 And of course, both of them are equally important. And I should also add that 00:15:29.364 --> 00:15:32.804 grid cells may help to create a spatial reference frame, 00:15:32.964 --> 00:15:36.964 but what goes into the hippocampus is equally much dominated by, 00:15:37.084 --> 00:15:39.784 for example, inputs from the lateral entorhinal cortex, 00:15:40.044 --> 00:15:45.964 which we understand much less, but which may be informative about all the other 00:15:45.964 --> 00:15:51.364 types of changes that occurs in an environment like the experiences that an 00:15:51.364 --> 00:15:54.224 animal has while it's walking around in the space. 00:15:54.224 --> 00:15:59.144 But now if we just look at, let's say, cell numbers, which of these two divisions 00:15:59.144 --> 00:16:03.644 would be, let's say, dominant from just a perspective of cell volume? 00:16:05.010 --> 00:16:08.730 You mean medial versus lateral entorhinal cortex? 00:16:09.010 --> 00:16:13.450 I would think that they are both important. I mean, the number of cells is about the same. 00:16:14.590 --> 00:16:19.050 It's just that we don't understand the lateral input much at all. Right. 00:16:19.850 --> 00:16:25.690 You mentioned that the grid cells occur in these modules, and the modules are 00:16:25.690 --> 00:16:27.550 at different spatial scales. 00:16:28.250 --> 00:16:34.050 Does that mean that across the whole set of modules, you're imagining that the 00:16:34.050 --> 00:16:39.250 animal has access to a unique code for its location in space. 00:16:40.530 --> 00:16:44.930 Will it operate in that way? Because obviously within any single module, 00:16:45.390 --> 00:16:49.630 you know you're on the grid, but you don't know exactly where you are on the grid. 00:16:49.890 --> 00:16:53.950 But maybe relative to the other modules you can build up a more unique… Exactly. 00:16:54.230 --> 00:16:57.630 You need to use the modules in combination to get a unique code, 00:16:57.710 --> 00:17:00.450 because otherwise there are multiple solutions. Yeah. 00:17:00.570 --> 00:17:06.930 So that can either happen within the entorhinal cortex itself or just as likely 00:17:06.930 --> 00:17:12.670 it may use the hippocampus where the modules are likely to converge onto play cells. 00:17:12.850 --> 00:17:16.610 So how that occurs is not known, 00:17:16.750 --> 00:17:22.730 but I would think that it would be very smart of the brain to actually compare 00:17:22.730 --> 00:17:29.150 the activity of different modules at any given time and not just let them drift on their own. 00:17:29.150 --> 00:17:34.330 And does the connectivity into the hippocampus suggest that, 00:17:34.390 --> 00:17:38.950 for instance, place cells have that ability to read out from the multiple modules? 00:17:39.430 --> 00:17:43.250 Well, that's what we are investigating now. It's not a very simple environment 00:17:43.250 --> 00:17:48.530 because you need actually to determine what are the inputs to an individual place cell. 00:17:49.590 --> 00:17:56.290 But now with rabies virus tracing, you can actually find the functional inputs 00:17:56.290 --> 00:17:58.530 onto single place cells. 00:17:58.530 --> 00:18:04.070 And the aim then is to determine whether grid cells from different modules converge 00:18:04.070 --> 00:18:08.370 or whether they actually somehow stay separate in hippocampal cells. 00:18:09.457 --> 00:18:14.997 So in this example, we are following, let's say, a causal chain that would go 00:18:14.997 --> 00:18:18.877 like, well, we have entorhinal cortex, these inputs go into hippocampus, 00:18:18.937 --> 00:18:24.057 grid cells are a part of that, 50-50 with lateral entorhinal cortex telling you other stuff. 00:18:24.997 --> 00:18:30.117 But this provides some key information for a place cell to give a response to a location. 00:18:30.117 --> 00:18:35.017 And this if you want we could call this sort of the standard interpretation for quite a while, 00:18:35.877 --> 00:18:38.857 but that that interpretation is now under some 00:18:38.857 --> 00:18:44.017 under some challenge right because apparently also without grid cells or without 00:18:44.017 --> 00:18:48.197 this connection or the grid cells might not even target those cells directly 00:18:48.197 --> 00:18:51.817 them or you can even get rid of them so what's the situation there in your perspective 00:18:51.817 --> 00:18:56.837 are the grid cells really key in driving a play cell response or they're extra Yeah, 00:18:56.917 --> 00:19:01.397 I'm pretty sure that the grid cells are still key in driving place cells under normal circumstances, 00:19:01.457 --> 00:19:06.537 but the place cells appear to be responsive to a lot of inputs and can probably 00:19:06.537 --> 00:19:13.357 make some sort of spatial signal even out of other cell types that have a spatial bias, 00:19:13.497 --> 00:19:18.057 like cells in the lateral entorhinal cortex that are also weakly spatial, 00:19:18.977 --> 00:19:23.797 but still have enough information that at least if you average over many cells, 00:19:23.917 --> 00:19:26.157 you can tell quite precisely where you are. 00:19:26.377 --> 00:19:29.297 And place cells seem to be able to use that. 00:19:29.457 --> 00:19:32.697 It doesn't mean that they don't normally rely on grid cells. 00:19:32.837 --> 00:19:36.537 I would still think that grid cells and border cells are the major inputs. 00:19:37.778 --> 00:19:42.618 But play cells are able to do the best out of very little. 00:19:42.778 --> 00:19:50.858 Though when play cells, in cases where the medial entorhinal cortex is lesioned 00:19:50.858 --> 00:19:55.138 or grid cells are somehow otherwise inactivated, 00:19:55.278 --> 00:19:59.358 the play cells are usually not very normal. 00:19:59.858 --> 00:20:04.998 I mean, they're very unstable, for example, and they are also apparently not 00:20:04.998 --> 00:20:10.478 able to switch between environments, so changing maps is also not easy. 00:20:10.738 --> 00:20:12.898 So it's not a normal network, but 00:20:12.898 --> 00:20:17.378 the threshold for hippocampal cells to become a play cell is quite low. 00:20:17.718 --> 00:20:22.178 But that means that you see actually two complementary models of play cell formation, 00:20:22.458 --> 00:20:26.738 because one could be also just this older idea from O'Keefe and Burgess and 00:20:26.738 --> 00:20:31.258 others, that you just have weakly tuned spatial responses coming in from the 00:20:31.258 --> 00:20:32.318 lateral and thoracic cortex. 00:20:32.378 --> 00:20:36.438 And by just integrating over those, you can then get a space-specific response. 00:20:36.978 --> 00:20:40.538 And the complement would be the grid cells that will give you a redundant response. 00:20:40.838 --> 00:20:43.858 But by averaging over many of them, again, you get specificity. 00:20:43.858 --> 00:20:47.338 So then these two modes would be operating in parallel. 00:20:47.598 --> 00:20:50.158 Yeah, it's two ways. And they are not mutually exclusive. 00:20:50.518 --> 00:20:55.018 But I think what it shows is that it helps 00:20:55.018 --> 00:20:58.038 to have these spatial inputs so you can probably use several of 00:20:58.038 --> 00:21:01.458 them and then there must be some intrinsic hippocampal 00:21:01.458 --> 00:21:06.178 processing that we still don't quite understand may involve neuronal plasticity 00:21:06.178 --> 00:21:13.238 may also involve other circuit mechanism that then that help shape a play cell 00:21:13.238 --> 00:21:19.878 out of maybe not so precise spatial activity right so the other thing you can do is 00:21:20.118 --> 00:21:24.438 you can combine if you like the sort of metric properties of your grid cell map 00:21:24.618 --> 00:21:28.038 with the sort of more topographic relationships that 00:21:28.038 --> 00:21:31.278 you would get from sort of cues in the environment yes 00:21:31.278 --> 00:21:36.258 and so you could build up an idea about what's adjacent to what based purely 00:21:36.258 --> 00:21:40.018 on those features and that could give you activity in your play cells independent 00:21:40.018 --> 00:21:46.418 of your metric map yeah i believe that both both mechanisms are likely to be 00:21:46.418 --> 00:21:48.198 used so it is somewhat redundant. 00:21:50.478 --> 00:21:53.958 That's also probably explains why at 00:21:53.958 --> 00:21:59.998 very early ages when grid cells are still not highly periodic and in an immature 00:21:59.998 --> 00:22:05.418 state you can still get nice place cells because for example you have the border 00:22:05.418 --> 00:22:11.678 cell inputs are already intact from the first day and there's also evidence from. 00:22:12.716 --> 00:22:18.056 From groups at UCL in London, which suggests that at that early stage, 00:22:18.316 --> 00:22:22.896 the place cells are more precise near the borders of the environment and then in the middle, 00:22:23.016 --> 00:22:27.576 which is consistent with grid cells having a role where they sort of map the 00:22:27.576 --> 00:22:31.056 entire environment and the metrics of the environment where border cells are 00:22:31.056 --> 00:22:38.876 maybe more responsive to the specific landmarks and especially in geometric references. 00:22:39.336 --> 00:22:44.756 I think also we touched on the talk on the effect of environmental context and 00:22:44.756 --> 00:22:47.796 things like the presence of daylight. 00:22:48.296 --> 00:22:53.276 So if you're moving in darkness, you may be more reliant on this grid cell map. Absolutely. 00:22:53.556 --> 00:22:57.076 If it's a nice sunny day and you've got access to lots of visual cues, 00:22:57.316 --> 00:23:00.236 maybe you don't need that information so much. No, no, that's true. 00:23:00.376 --> 00:23:05.296 So, I mean, in most cases you have much more cues than you actually need. 00:23:05.296 --> 00:23:09.216 So it's quite hard to perturb the system. 00:23:09.416 --> 00:23:13.936 But in darkness, self-motion is more important. 00:23:14.076 --> 00:23:20.776 Although you can still do quite well even with just tactile cues. 00:23:20.896 --> 00:23:22.976 You bump into the corners and so 00:23:22.976 --> 00:23:28.156 on, so that you can at least occasionally reset and get the map to work. 00:23:28.596 --> 00:23:32.736 But of course, as you get out in the open space, there's no other reference 00:23:32.736 --> 00:23:36.496 than your own motion. And for that, grid cells are probably quite important. 00:23:37.016 --> 00:23:42.156 But now the situation has gone, actually has become more complicated because 00:23:42.156 --> 00:23:48.076 now you have also identified many other cell types in this little bit of brain, entorhinal cortex. 00:23:48.476 --> 00:23:53.056 Like we're now with border cells, with speed cells, we have head direction cells, right? 00:23:53.136 --> 00:24:00.456 So what's the relative proportion of these different types of cells in entorhinal cortex? 00:24:00.896 --> 00:24:06.116 Yeah, in our experience, still the most abundant cell is definitely the grid cell. 00:24:06.216 --> 00:24:10.596 At least if you search in the superficial layers in layer two and also to some 00:24:10.596 --> 00:24:17.976 extent in layer three, maybe even almost half of the cells in layer two seem to be grid cells. 00:24:18.136 --> 00:24:22.556 It's a bit hard to tell because as you get, at least if you get further deeper, 00:24:22.736 --> 00:24:27.816 then you get cells with larger scales, not always so easy to tell if they are grid cells. 00:24:28.356 --> 00:24:32.296 But in addition then, we also have, as you said, the head action cells there. 00:24:32.990 --> 00:24:38.210 In layer 3 and 5, they are very abundant, maybe the most abundant cell type. 00:24:38.870 --> 00:24:41.950 And then you have border cells, seem to be around 10%. 00:24:42.510 --> 00:24:46.850 They are in all layers, but especially 10% also in layer 2. 00:24:47.310 --> 00:24:51.530 And speed cells are maybe about 15% across all layers. 00:24:53.450 --> 00:24:57.530 But for the speed cells, for example, even if they are only 15%, 00:24:57.530 --> 00:25:01.090 many of them seem to be interneurons. They have interneuron-like properties, 00:25:01.250 --> 00:25:04.010 at least, which means that they connect to a large number of cells. 00:25:04.130 --> 00:25:08.990 So their influence, even if it's only 15%, is probably extremely important. 00:25:09.190 --> 00:25:12.250 They probably influence every single grid cell. 00:25:12.510 --> 00:25:16.450 But you interpret, so as you say now, right, you see those different cell types 00:25:16.450 --> 00:25:20.090 very much as forming local circuits that help the grid cells to stay on track? 00:25:20.430 --> 00:25:23.570 Or you see them as a forward pathway into the hippocampus? 00:25:23.570 --> 00:25:29.330 No, well, both, of course, but I do think they are important for the local circuit 00:25:29.330 --> 00:25:37.910 because the grid cells need the speed and the direction cells to stay updated. 00:25:38.930 --> 00:25:43.790 But, of course, the result of all of this is then fed into the hippocampus. 00:25:44.510 --> 00:25:49.710 So, I mean, they're not mutually exclusive. But so now, also in your talk, 00:25:49.790 --> 00:25:54.170 you then gave us an interpretation of how these different cells might work together. 00:25:54.310 --> 00:25:58.210 In particular, you were talking about how the border cells might actually be 00:25:58.210 --> 00:26:02.710 interacting with grid cells to sort of help them in aligning to an environment. 00:26:02.990 --> 00:26:05.030 So how does that work out? 00:26:05.830 --> 00:26:11.990 Yeah, I mean, what the role is of each cell type is of course still a matter 00:26:11.990 --> 00:26:15.290 of speculation because we don't have the tools to manipulate it. 00:26:15.290 --> 00:26:20.030 But what we see with the grid cells is that they are heavily influenced by the 00:26:20.030 --> 00:26:22.290 borders of the environment. 00:26:22.430 --> 00:26:26.090 So as you get close to the borders, then the grid cells tend to get distorted. 00:26:27.363 --> 00:26:30.863 So they are not perfectly hexagonal any longer. 00:26:31.023 --> 00:26:37.383 And that can be explained by forces that operate along the walls and then both 00:26:37.383 --> 00:26:41.483 deform and rotate the grid patterns in certain ways. 00:26:41.783 --> 00:26:48.463 So most likely this is mediated through border cells because they have activity 00:26:48.463 --> 00:26:52.483 that corresponds to the orientation of the walls 00:26:52.583 --> 00:26:57.423 and also activity that decreases as you get away from the walls. 00:26:57.623 --> 00:27:03.163 But how this is implemented exactly in the network is still very much an open 00:27:03.163 --> 00:27:04.763 issue. I don't know how that happens. 00:27:05.083 --> 00:27:10.283 Because there are different ways to… Maybe we could argue that the notion border 00:27:10.283 --> 00:27:15.283 cell might be a bit too restricted interpretation of what they do. 00:27:15.463 --> 00:27:19.243 You could also argue that maybe these are just cells that go for let's say salient 00:27:19.243 --> 00:27:22.903 aspects of the environment that can be exploited as anchors. 00:27:23.243 --> 00:27:27.703 So it's more like a salience or a landmark cell. 00:27:27.943 --> 00:27:34.663 It could be. Now we have put individual more point-like landmarks into the environment 00:27:34.663 --> 00:27:42.403 in the past and usually they have quite limited influence on the cells that we have recorded, but. 00:27:44.243 --> 00:27:47.463 At least in the medial entorhinal cortex. In the lateral entorhinal cortex, 00:27:47.563 --> 00:27:53.383 there are cells that actually do respond to particular objects and their location. 00:27:53.383 --> 00:27:57.023 So they fire around those objects, even if you then remove them afterwards. 00:27:58.063 --> 00:28:02.643 But those are not really part of the medial entorhinal network. 00:28:03.183 --> 00:28:06.123 But your interpretation would really be like, I have my grid cells. 00:28:06.543 --> 00:28:11.583 They give me, let's say, an initial description of a space in which I can operate. 00:28:11.583 --> 00:28:15.043 This space is seen in a planar perspective it's 00:28:15.043 --> 00:28:18.503 not three-dimensional it's a two-dimensional plane and my 00:28:18.503 --> 00:28:21.863 border cells would really be telling me where the where the 00:28:21.863 --> 00:28:27.283 end is of that of that flat world in which i exist you would agree with that 00:28:27.283 --> 00:28:31.863 uh yes but not only where the end is i would rather say where there are signal 00:28:31.863 --> 00:28:38.063 significant or salient uh reference directions so if you put a wall into into 00:28:38.063 --> 00:28:40.983 the middle of an environment, 00:28:41.143 --> 00:28:45.063 you will still get the border cells, some of the border cells to fire along that wall too. 00:28:45.223 --> 00:28:50.303 So it doesn't mean it's the end, but they're very significant for anchoring the grid. 00:28:50.423 --> 00:28:55.603 And for anchoring, it's of course most effective actually to use straight lines wherever they are. 00:28:56.043 --> 00:28:59.203 But now you could also argue that maybe the border cells are just responding 00:28:59.203 --> 00:29:02.583 to the dynamics of the grid cells because if I'm running over, 00:29:02.623 --> 00:29:05.743 let's say I'm running over this table and there are edges, that also means there 00:29:05.743 --> 00:29:07.063 are certain positions in space. 00:29:07.063 --> 00:29:11.723 In other words, there are certain attractor states of my grid cells I will never reach. 00:29:13.103 --> 00:29:17.863 And these might be giving you transient responses in the population of grid 00:29:17.863 --> 00:29:22.463 cells that you can pick up and then say, aha, this is an important transient in my dynamic. 00:29:22.783 --> 00:29:24.843 So that's the grid cell driving the border cell. 00:29:25.303 --> 00:29:28.223 Would you buy that interpretation or there's something missing in that? 00:29:29.203 --> 00:29:34.763 No, I think it goes both ways. So I think the border cells influence the grid 00:29:34.763 --> 00:29:38.103 cells, but the grid cells will then also So again, influence the border cells. 00:29:38.203 --> 00:29:44.403 I think they are probably bidirectionally connected and they work together all the time. 00:29:44.663 --> 00:29:50.623 Okay. So now we have our map of our grid cells. 00:29:51.023 --> 00:29:57.923 And this is tightly coupled to hippocampus, which sort of now loops back the cortex onto itself. 00:29:59.703 --> 00:30:02.443 So what is the hippocampus doing with this information? 00:30:04.327 --> 00:30:09.827 Well, one striking difference between replace cells in the hippocampus and the 00:30:09.827 --> 00:30:13.467 grid cells, or all of the cells actually in medial entorhinal cortex, 00:30:13.807 --> 00:30:21.247 is that in the hippocampus, replace cells form individual maps for every single environment. 00:30:21.447 --> 00:30:25.787 So for every environment where a rat is tested, 00:30:26.027 --> 00:30:30.447 it seems to generate orthogonal maps almost, 00:30:30.527 --> 00:30:33.827 maps that are completely independent, which fits very 00:30:33.827 --> 00:30:36.787 well with what you would expect from a memory 00:30:36.787 --> 00:30:40.287 perspective on the hippocampus that you actually form discrete 00:30:40.287 --> 00:30:44.067 representations for different experiences 00:30:44.067 --> 00:30:47.647 in the animal's life 00:30:47.647 --> 00:30:50.487 which is very much in contrast to what we have seen in 00:30:50.487 --> 00:30:54.027 entorhinal cortex where the firing relationships 00:30:54.027 --> 00:30:56.807 are sort of preserved from one environment to the 00:30:56.807 --> 00:30:59.587 other and so that that is 00:30:59.587 --> 00:31:03.627 an important feature of hippocampal activity how 00:31:03.627 --> 00:31:09.367 that is transformation is generated that's a very important task that we still 00:31:09.367 --> 00:31:15.167 have no data for it can happen perhaps by combining activity from different 00:31:15.167 --> 00:31:20.427 modules because by differentially combining activity 00:31:20.707 --> 00:31:24.527 from modules, you can get a large number of activity patterns. 00:31:25.947 --> 00:31:31.607 But that's still quite uncertain. But of course you could also argue that entorhinal 00:31:31.607 --> 00:31:35.487 cortex is maybe not a main source of information for hippocampus, 00:31:35.487 --> 00:31:38.347 but it's actually a main source of information for the rest of the cortex. 00:31:39.670 --> 00:31:44.650 So how do you see that exchange? Well, the exchange between entorhinal cortex 00:31:44.650 --> 00:31:48.130 and rest of cortex is not very well understood at all. 00:31:48.850 --> 00:31:56.590 Of course, just based on pure connectivity, we know that entorhinal cortex is not an island. 00:31:56.710 --> 00:32:00.450 It really interacts with almost the entire rest of the cortex. 00:32:00.830 --> 00:32:07.290 And we also know that navigation is not only a hippocampal entorhinal phenomenon. 00:32:07.290 --> 00:32:10.650 The navigation maybe the creation 00:32:10.650 --> 00:32:14.110 of an internal map involves those two structures particularly 00:32:14.110 --> 00:32:17.030 but the internal map needs to be 00:32:17.030 --> 00:32:23.090 used for the animal to get from a to b and for navigation in a broader sense 00:32:23.090 --> 00:32:28.470 beyond forming maps you actually need the entire brain and then that involves 00:32:28.470 --> 00:32:35.330 for example prefrontal cortex which is important for planning how you get from one place to the other. 00:32:35.490 --> 00:32:38.870 So it's very important to remember 00:32:38.870 --> 00:32:44.490 that the entorhinal hippocampal network is part of a wider system. 00:32:44.730 --> 00:32:50.090 And I think in the future, we'll probably try to understand these other cortical 00:32:50.090 --> 00:32:52.850 regions too, although it's a quite challenging task. 00:32:53.090 --> 00:33:00.650 Of course. But now the other amazing result you presented today was in some sense taking 00:33:00.750 --> 00:33:04.390 away all doubt anyone might have about the metric properties of these grid cells 00:33:04.390 --> 00:33:09.990 because you show how amazingly precisely they are aligned with the space in 00:33:09.990 --> 00:33:10.870 which the animal operates. 00:33:11.330 --> 00:33:14.210 So what are the basic observations there? 00:33:16.050 --> 00:33:19.650 With regard to the metric, I mean the metric 00:33:19.650 --> 00:33:23.350 of… well first 00:33:23.350 --> 00:33:26.730 of all when you started out i mean the the grid 00:33:26.730 --> 00:33:29.430 cells um appear to be 00:33:29.430 --> 00:33:32.370 i mean we're struck by their enormous regularity and 00:33:32.370 --> 00:33:37.190 the fact that they form a perfect grid that with 60 degree angles that repeats 00:33:37.190 --> 00:33:43.150 itself all over the space but then as we start looking closer and especially 00:33:43.150 --> 00:33:48.930 in large environments it's easy to observe that the grid is actually slightly 00:33:48.930 --> 00:33:50.710 deformed especially near, 00:33:51.230 --> 00:33:56.050 the walls or the ends of the environment where you see that borders, 00:33:56.310 --> 00:34:01.130 walls have strong influences and sort of deform the grid and of course that 00:34:01.130 --> 00:34:07.990 then raises the question whether does this have any consequence for the use 00:34:07.990 --> 00:34:12.370 of the grid to infer directions and distances. 00:34:12.370 --> 00:34:18.390 Instances, I would still say that, by and large, you can infer position and 00:34:18.390 --> 00:34:22.810 direction from the grid even as it is, even with these slight distortions. 00:34:22.830 --> 00:34:27.010 But of course it would be interesting to see if these distortions that are present 00:34:27.010 --> 00:34:31.650 in the map would also transfer to behavior, so maybe our judgment of position 00:34:31.650 --> 00:34:34.890 might be slightly distorted also. 00:34:35.130 --> 00:34:38.630 That's something that would be interesting to have tested at some point. 00:34:39.472 --> 00:34:44.992 But you also showed that the grids align with the cardinal axis of the environment. Yes. Right? Almost. 00:34:45.452 --> 00:34:51.072 Oh, and this is the weird thing about it, but it was sort of perfectly aligned with a tiny offset. 00:34:51.392 --> 00:34:54.092 And they are an offset of seven and a half degrees on average. 00:34:54.212 --> 00:34:58.652 So that's the axis of the, the grid has three axes. 00:34:58.832 --> 00:35:04.212 So the axis that is closest to one of the walls is usually offset by on average 00:35:04.212 --> 00:35:05.392 seven and a half degrees. 00:35:05.392 --> 00:35:11.072 And that we interpret then as a result of what we call a shearing process, 00:35:11.652 --> 00:35:17.092 which probably begins on day one when the animal experiences the environment 00:35:17.092 --> 00:35:22.952 that is forces along walls that when distort the grid pattern and as part of 00:35:22.952 --> 00:35:26.772 the distortion process also gets it to rotate one, 00:35:26.852 --> 00:35:28.772 especially one of its axis. 00:35:29.172 --> 00:35:32.852 Right. But also what was really interesting is that on the one you see that 00:35:32.852 --> 00:35:37.392 one of the coronal axis is taken as the anchor point, if you want, 00:35:37.472 --> 00:35:40.052 and the orthogonal one is ignored. No? 00:35:40.452 --> 00:35:44.572 In that particular experiment, yes. Ah, okay. This is not always the case. 00:35:44.652 --> 00:35:46.852 Not always the case. So, quite, it depends. 00:35:47.532 --> 00:35:52.192 And in that particular experiment that I referred to, where it always chooses 00:35:52.192 --> 00:35:57.392 one of the axes, it's important to remember that the rats were introduced to 00:35:57.392 --> 00:35:59.672 the environment in a very, very consistent way. 00:35:59.812 --> 00:36:07.772 So, all of the rats were placed in exactly the same corner with all or much 00:36:07.772 --> 00:36:13.012 of the focus to cues in that along one particular wall of the box. 00:36:13.192 --> 00:36:21.012 Perhaps this may have shaped the animal as it's in its initial formation of 00:36:21.012 --> 00:36:26.672 the grid cell map that they may have used cues along one wall more than another. 00:36:26.852 --> 00:36:29.972 That's our hypothesis. This is, of course, something that we will need to test. 00:36:30.172 --> 00:36:35.552 But the effect of, I would assume that the early environment, 00:36:35.972 --> 00:36:40.352 I mean, the environment as it is on the very first experience is very important 00:36:40.352 --> 00:36:42.992 for how the grid actually is anchored. 00:36:43.672 --> 00:36:46.932 Right. And also there we should not forget that these animals are, 00:36:46.992 --> 00:36:50.812 of course, come pre-equipped with a lot of, let's say, stereotype behavior. 00:36:51.112 --> 00:36:54.292 So you put them in that corner and they will all start to do tic-mo-taxis. 00:36:54.292 --> 00:36:57.252 It's always run around the walls, either clockwise or counterclockwise. 00:36:58.052 --> 00:37:02.712 So that means this early experience is shaped in a very stereotyped way for all of them. 00:37:03.712 --> 00:37:08.812 So, and this is a bit surprising, you have this sort of more mechanical metaphor 00:37:08.812 --> 00:37:12.592 to interpret the distortion of the grid in terms of shearing. 00:37:13.512 --> 00:37:17.372 But the alternative would be a more behavioral interpretation where you say, 00:37:17.432 --> 00:37:21.832 well, look, if I'm throwing the environment or with all my friends always in 00:37:21.832 --> 00:37:25.672 the same corner, And we all will have a stereotype response of Tecmo Texas running 00:37:25.672 --> 00:37:27.832 along their walls for quite a while. 00:37:28.670 --> 00:37:33.570 This will give me quite a bias in my input sampling that might then lead to 00:37:33.570 --> 00:37:36.110 that distortion of my grid cell response. 00:37:36.530 --> 00:37:40.110 Yeah, I think those explanations are not mutually exclusive. 00:37:40.330 --> 00:37:44.450 I think you can get what we observe or mechanically can describe as a shearing 00:37:44.450 --> 00:37:52.610 process through behavioral filtering in a way where the animals actually focus 00:37:52.610 --> 00:37:57.790 on certain cues and perhaps have much more attention on those than others. 00:37:57.790 --> 00:38:01.430 So I think that is a possible implementation. 00:38:02.290 --> 00:38:07.050 Wait, we don't need to consider attention because imagine I just do ticmo taxes. 00:38:07.330 --> 00:38:11.490 I just run around and whether I'm paying attention or not, I will drive my grid 00:38:11.490 --> 00:38:13.630 cells in a very specific order. 00:38:13.950 --> 00:38:17.810 And given that this is a hyperplastic system, that then gives you the bias. 00:38:18.130 --> 00:38:22.570 That's possible. I mean, you can even do it probably at least in principle without 00:38:22.570 --> 00:38:25.890 attention simply by activating cells. I agree with that. Exactly, yeah. 00:38:27.330 --> 00:38:36.390 So, but now the interpretation of this realignment of the grids or their formation, 00:38:37.810 --> 00:38:41.390 to what extent do we really have to think about the box in which the animal 00:38:41.390 --> 00:38:45.310 is or the animal will also consider, also going back to Tolman, 00:38:45.390 --> 00:38:48.250 right, the global cues that are out there in the environment. 00:38:48.610 --> 00:38:53.950 So, are these controlled for? Is this all local sensory information taken into 00:38:53.950 --> 00:38:55.650 account? Is it global sources? 00:38:56.070 --> 00:39:01.050 No, of course it uses all kinds of cues. But yet it is, if you test an animal 00:39:01.050 --> 00:39:06.550 in a box like the ones we do, the strong, the local cues and especially the 00:39:06.550 --> 00:39:08.090 walls of the environment are, 00:39:08.870 --> 00:39:12.510 have a strong, very strong influence on most of the grid cells. 00:39:14.170 --> 00:39:18.810 At the same time, we also see that they sometimes respond to cues that are further 00:39:18.810 --> 00:39:21.970 away and that may even differ across grid cells. 00:39:21.970 --> 00:39:26.490 Maybe some respond, some modules maybe, depending on scale, may respond more 00:39:26.490 --> 00:39:29.790 to distal and others more to proximal cues. Still not settled. 00:39:31.330 --> 00:39:38.490 But in some experiments, we simply try to close out all the distal cues by simply 00:39:38.490 --> 00:39:40.070 pulling curtains around. 00:39:40.270 --> 00:39:44.190 And you get essentially the same results. So the local cues are very important. 00:39:44.310 --> 00:39:46.670 But of course, they all matter. Yeah. 00:39:47.476 --> 00:39:50.456 But now you use the concept of anchoring also for this, right? 00:39:50.476 --> 00:39:55.036 That you have to anchor that grid cell map in some properties of the environment 00:39:55.036 --> 00:39:56.856 or I guess some intrinsic signal. 00:39:57.816 --> 00:40:03.956 So if you would have to define this anchoring as a sort of a neural mechanistic 00:40:03.956 --> 00:40:08.436 sense, how would you realize that kind of anchoring? 00:40:09.016 --> 00:40:14.836 Well, the way I often think about it is that it's synaptic plasticity. 00:40:14.836 --> 00:40:19.856 It's a kind of learning process where you associate the firing pattern of the grid, 00:40:20.076 --> 00:40:24.996 a free-floating grid in a way, with a certain environmental reference that is 00:40:24.996 --> 00:40:28.556 then somehow mediated to the entorhinal cortex via, for example, 00:40:28.556 --> 00:40:30.676 visual inputs. It could also be other ways. 00:40:30.856 --> 00:40:37.196 And that there is plasticity taking place, usually, I would assume, 00:40:37.276 --> 00:40:39.016 quite early on, on the first trial. 00:40:39.156 --> 00:40:44.776 And then that sets the grid in a certain way, and then it remains that way for the rest of the time. 00:40:44.776 --> 00:40:49.496 But that would mean that you're saying there's some external signal that says 00:40:49.496 --> 00:40:55.376 now this is important, anchored to this, or you see it more as a continuous 00:40:55.376 --> 00:40:59.396 sampling that converges into some attractor state that is this anchor? 00:41:02.036 --> 00:41:07.416 Well, I think it will use whatever sensory input is present right from the beginning. 00:41:07.596 --> 00:41:10.196 I think this happens almost instantaneously. 00:41:10.656 --> 00:41:18.516 Exactly how it happens, that I don't know. I think whether that involves an 00:41:18.516 --> 00:41:21.856 attractor process as such is maybe not necessary, really. 00:41:21.996 --> 00:41:26.196 You just need to, even if you have an attractor that shapes a grid pattern, 00:41:26.256 --> 00:41:28.416 you just need to associate with 00:41:28.416 --> 00:41:31.696 certain inputs so that you can start out there next time you come back. 00:41:31.876 --> 00:41:35.916 And then in that, let's say it is an attractor state, and then you get the starting 00:41:35.916 --> 00:41:38.716 point for the next trial when you come back. 00:41:38.856 --> 00:41:42.696 And then in that sense, re-experience activity. But have you in that sense considered 00:41:42.696 --> 00:41:46.976 a role for neuromodulators because here I am, I'm your rat, you put me in this 00:41:46.976 --> 00:41:50.376 environment, I've never been here before, lots of novelty, lots of stress, 00:41:50.736 --> 00:41:55.016 acetylcholine is coming out, dopamine is being released, very strong learning 00:41:55.016 --> 00:41:59.136 signals that can serve then to sort of define an initial anchor. 00:41:59.396 --> 00:42:06.516 Exactly, yeah. So that happens very fast and that certainly helps to stabilize 00:42:06.516 --> 00:42:08.476 the map right away from the beginning. 00:42:08.476 --> 00:42:11.816 So your prediction would be if you would, let's say, use antagonists to these 00:42:11.816 --> 00:42:15.636 kind of neuromodulators, anchoring would be, let's say, compromised. 00:42:16.076 --> 00:42:18.736 Yeah, I wouldn't be surprised if that happens. 00:42:19.506 --> 00:42:26.146 Okay. So now that we know a bit more about the grid cells, you have then also 00:42:26.146 --> 00:42:29.406 looked in detail really at very 00:42:29.406 --> 00:42:34.266 much a psychologist's question of is this nature, is it nurture, right? 00:42:34.286 --> 00:42:38.566 Is it an innate system, it's an innate automator in some way, 00:42:38.686 --> 00:42:40.646 or is it really dependent on experience? 00:42:41.366 --> 00:42:45.966 So what have you learned from those experiments? Well, it's still ongoing work, 00:42:46.106 --> 00:42:51.006 but the initial experiments that we did a couple of years ago have shown basically 00:42:51.006 --> 00:42:53.626 that for several of these cell types, 00:42:53.866 --> 00:42:58.846 they express adult-like properties almost from the beginning, 00:42:59.006 --> 00:43:00.506 at least when we can measure them. 00:43:00.506 --> 00:43:09.486 So place cells look more or less adult-like from the first day when we can actually record activity. 00:43:09.666 --> 00:43:14.386 When animals start walking out of the nest and cover at least small environmental 00:43:14.386 --> 00:43:17.966 spaces, you can already measure place cells. 00:43:18.486 --> 00:43:24.906 Head direction cells have directional preferences even before the rat pups leave 00:43:24.906 --> 00:43:28.066 the nest because we can record them even before the eyes open. 00:43:28.066 --> 00:43:31.406 So a directional tuning is present very 00:43:31.406 --> 00:43:34.306 very early on and border cells also like place 00:43:34.306 --> 00:43:37.266 places are very early there grid cells though 00:43:37.266 --> 00:43:40.186 are a bit slower so they take another week or 00:43:40.186 --> 00:43:43.466 two before they get adult like property so 00:43:43.466 --> 00:43:46.806 in this early stage during the 00:43:46.806 --> 00:43:50.326 first week or two after the animals start walking around then 00:43:50.326 --> 00:43:56.866 they they have periodic firing patterns but it's not very regular And that probably 00:43:56.866 --> 00:44:05.626 leaves a window for experience to actually influence the network and perhaps 00:44:05.626 --> 00:44:09.306 help with the anchoring process. 00:44:09.726 --> 00:44:16.306 So we did then, I presented experiments that are still going on where we have 00:44:16.306 --> 00:44:22.686 raised rats in environments where the animals are really deprived of borders. 00:44:22.686 --> 00:44:27.186 So one group was then raised in a spherical environment until they were adult. 00:44:27.666 --> 00:44:34.226 And in the absence of environmental borders, we then tested these animals when 00:44:34.226 --> 00:44:37.746 they became adult, whether they actually could form normal grid cells. 00:44:37.966 --> 00:44:41.006 And it turns out that that probably takes a longer time, 00:44:41.086 --> 00:44:45.686 or it maybe even may not happen, at least if the environment is large enough, 00:44:45.686 --> 00:44:52.046 of, compared to what happens if you raise the animals in environments that do 00:44:52.046 --> 00:44:55.586 have borders and cubes which otherwise are very similar. 00:44:55.806 --> 00:45:00.806 So is it possible that the difference between the spherical environment and 00:45:00.806 --> 00:45:04.746 the rectangular environment is that this mechanism whereby. 00:45:05.715 --> 00:45:11.635 The borders or the visual cues for the borders correct for the slip and the path integration. 00:45:12.255 --> 00:45:17.555 Yeah, I think that is a major thing that the network has to learn, 00:45:17.755 --> 00:45:27.095 both how to anchor to associate with the environmental directional cues in the 00:45:27.095 --> 00:45:29.275 environment right from the beginning, 00:45:29.415 --> 00:45:35.075 but also to use that as the animal is walking around to correct at the time 00:45:35.075 --> 00:45:37.435 when the grid pattern is beginning to drift. 00:45:37.755 --> 00:45:43.055 And if that hasn't taken place in early development, that may compromise the 00:45:43.055 --> 00:45:45.455 grid pattern at the more adult stage. 00:45:45.775 --> 00:45:49.435 And something else, obviously, that's happening in those first few weeks of 00:45:49.435 --> 00:45:53.275 life is the animal is changing enormously in size and morphology. 00:45:53.495 --> 00:45:58.555 So if the mechanisms that are contributing to path integration are to do with 00:45:58.555 --> 00:46:05.335 gait and body length and speed, then you can't fix those in place at day 11 00:46:05.335 --> 00:46:06.635 when you first wander out the nest. 00:46:06.715 --> 00:46:12.055 You have to be able to calibrate as you grow in size and speed, 00:46:12.295 --> 00:46:15.795 and it may continue to calibrate I guess throughout life. 00:46:16.195 --> 00:46:20.935 Yeah, that's possible. I mean, I wouldn't say that it even stops at P30. 00:46:21.675 --> 00:46:24.875 So this is probably a process that's going on for quite a while, 00:46:24.995 --> 00:46:31.495 and to the extent that that just length of footsteps are used by the animal 00:46:31.495 --> 00:46:32.775 actually to calibrate position. 00:46:33.035 --> 00:46:36.995 This is something that has to go on for quite a while. But of course they don't 00:46:36.995 --> 00:46:40.995 only use that, they can use other sources of speed information too. 00:46:41.706 --> 00:46:46.406 But this experiment might really help us to better the causal chain of this 00:46:46.406 --> 00:46:50.026 system because actually the grid cells appear to last, right? 00:46:50.146 --> 00:46:53.346 So in some sense, it's telling us something really important about the scaffolding 00:46:53.346 --> 00:46:54.926 of this learning process. 00:46:55.066 --> 00:46:57.066 And apparently, you really start with speed, head direction, 00:46:57.306 --> 00:46:59.366 border cells, even place cells. 00:46:59.786 --> 00:47:05.306 And only once that scaffold is sort of put together can grid cells be formed. 00:47:05.586 --> 00:47:09.306 Is it also how you think about it? The grid cells really floating on that information? 00:47:10.206 --> 00:47:14.366 Information? Yeah, I mean, obviously, as you say, the grid cells, 00:47:14.606 --> 00:47:21.606 they do still provide some metric information even right from the beginning. 00:47:21.826 --> 00:47:25.306 So even if you don't really see that very well in individual cells, 00:47:25.546 --> 00:47:29.506 the population would still provide the hippocampus at least with place information. 00:47:29.906 --> 00:47:34.346 But it certainly allots us to the possibility that you, 00:47:34.346 --> 00:47:37.326 for the other cell types like for place cells 00:47:37.326 --> 00:47:41.286 you don't really need a lot of grid activity for 00:47:41.286 --> 00:47:44.146 or spatial activity for a spatial signal to 00:47:44.146 --> 00:47:48.846 be created but i would think that at that early stage there's still something 00:47:48.846 --> 00:47:53.346 that is missing and then that which is later contributed by the grid cells which 00:47:53.346 --> 00:47:57.666 is the precise metrics ability to actually calculate exactly where you are and 00:47:57.666 --> 00:48:03.766 especially when you're far away from visual cues and perhaps this is something that is not 00:48:03.946 --> 00:48:06.966 present until the animal is a bit older. 00:48:07.386 --> 00:48:12.026 But then do you interpret this as a stage-wise development where let's say first 00:48:12.026 --> 00:48:15.226 I get my place cells, border cells, speech cells, et cetera, 00:48:15.366 --> 00:48:19.586 I build my grid cells and now I liberate my hippocampus to start to dedicate 00:48:19.586 --> 00:48:20.886 itself to different tasks. 00:48:21.166 --> 00:48:23.626 So I move now to a different phase of operation. 00:48:24.406 --> 00:48:29.186 Or do you believe that the system will be operating in the same state as it 00:48:29.186 --> 00:48:30.246 was in when you were born? 00:48:31.327 --> 00:48:34.347 Well, I don't think that you liberate the hippocampus to do other things, 00:48:34.467 --> 00:48:37.147 but I mean, hippocampus may take on other tasks. 00:48:37.267 --> 00:48:44.187 That's very well possible, but the entorhinal network is not really mature until 00:48:44.187 --> 00:48:47.707 very late, actually, because it continues to develop. 00:48:47.807 --> 00:48:52.007 The connectivity, especially via the interneurons, is not mature until you're 00:48:52.007 --> 00:48:55.847 about, if you're a rat, until you're about almost four weeks old. 00:48:55.927 --> 00:48:59.727 So that's very late. and i think that 00:48:59.727 --> 00:49:03.047 is important because it really allows for experience 00:49:03.047 --> 00:49:05.967 to shape the animal and maybe experience is 00:49:05.967 --> 00:49:10.687 important because it's not only about creating a single grid pattern you actually 00:49:10.687 --> 00:49:15.167 need to probably link together grid patterns for small patches of space and 00:49:15.167 --> 00:49:23.047 and that may require experience with spatial environment right okay but on the 00:49:23.047 --> 00:49:24.647 other hand And of course, 00:49:24.667 --> 00:49:27.587 it also means that more frontal areas in the brain are also still developing. 00:49:27.707 --> 00:49:32.587 And the rhino cortex also has to deal with that as an interface to the hippocampus. 00:49:32.587 --> 00:49:35.907 This might be another constraint that we have to take into account. 00:49:36.227 --> 00:49:40.827 And especially when it comes to planning how to use the maps and linking them 00:49:40.827 --> 00:49:43.007 to action, that may still not be mature. 00:49:43.107 --> 00:49:47.447 That hasn't been studied, but it's very well possible that it takes even longer time. Right. 00:49:47.867 --> 00:49:51.307 Now, you're also talking about this link to other brain structures. 00:49:51.307 --> 00:49:56.067 You shortly mentioned this link to prefrontal cortex that you started to look 00:49:56.067 --> 00:49:59.347 at more recently via the thalamus. 00:49:59.847 --> 00:50:04.627 So then how do you see that interface between hippocampus and prefrontal? 00:50:04.947 --> 00:50:08.647 You said already earlier, prefrontal will be the planning, your executive control 00:50:08.647 --> 00:50:15.127 system, but it is extracting specific information from the hippocampus or hippocampus 00:50:15.127 --> 00:50:17.207 extracting specific information from prefrontal cortex. 00:50:17.427 --> 00:50:20.947 So how do you see that interaction? direction well um it goes 00:50:20.947 --> 00:50:23.987 both ways for sure because i talked uh briefly 00:50:23.987 --> 00:50:28.907 only about the connections from prefrontal via reunions to the hippocampus and 00:50:28.907 --> 00:50:36.267 uh i believe that there is that is a route for prefrontal to influence uh hippocampus 00:50:36.267 --> 00:50:40.747 but there are reverse connections to at least via the subiculum and from more 00:50:40.747 --> 00:50:43.547 ventral parts of hippocampus was directly back to the entorhinal cortex, 00:50:43.967 --> 00:50:53.007 so that prefrontal circuits are updated probably by internal maps of the hippocampus. 00:50:53.547 --> 00:50:59.407 Like all other, or many other systems of the brain, it's bidirectional and then 00:50:59.407 --> 00:51:04.247 it gets so much more difficult to understand than if it's just a linear process, 00:51:04.407 --> 00:51:06.607 but that's the way the brain works. Absolutely, yeah. 00:51:08.533 --> 00:51:14.133 So in some sense, early on, the study of hippocampus was very much dominated 00:51:14.133 --> 00:51:19.133 by a neuropsychological perspective, where people thought strongly about episodic memory, 00:51:19.353 --> 00:51:24.433 how this was linked to certain deficits that would occur in humans after lesions to the brain. 00:51:26.053 --> 00:51:29.113 Mollison is a famous example of that. 00:51:29.113 --> 00:51:36.453 And in some sense, now the debate has shifted very much towards a perspective 00:51:36.453 --> 00:51:41.113 of space, more pragmatic, autometers, spatial representation, 00:51:41.653 --> 00:51:43.233 surviving in a box, right? 00:51:43.233 --> 00:51:50.433 But in some sense, I would imagine also as a psychologist trying to link psyche with the flesh, 00:51:50.673 --> 00:51:56.673 you want to claw your way back to this more high-level interpretation of, 00:51:56.773 --> 00:52:02.173 let's say, episodic memory in the context of human experience. 00:52:02.173 --> 00:52:09.393 Well, I still believe that the hippocampus is absolutely critical for certain memory functions. 00:52:10.493 --> 00:52:17.813 But what has happened during the last years is that we have understood much 00:52:17.813 --> 00:52:23.113 more out of one component of the system, which is the spatial framework, 00:52:23.393 --> 00:52:29.933 which I believe is a fundamental reference for memories to be created. 00:52:29.933 --> 00:52:37.173 So space is a fundamental element of all episodic memories, but it's not all. 00:52:37.333 --> 00:52:41.293 On top of that, we also have the experiences of what actually happens at those 00:52:41.293 --> 00:52:44.393 spaces, and hippocampus is critical for that. 00:52:44.473 --> 00:52:48.173 I think for that to be understood better, we need to know more about the lateral 00:52:48.173 --> 00:52:56.333 cortex, entorhinal cortex input, which is 50% of the input to the hippocampus. 00:52:56.333 --> 00:53:00.493 So, but I don't think these are mutually exclusive. 00:53:00.793 --> 00:53:06.793 I mean, hippocampus space is a very important element of what is encoded in the hippocampus. 00:53:07.333 --> 00:53:12.193 But unlike the entorhinal cortex, the hippocampal place signal or space signal 00:53:12.193 --> 00:53:17.553 is part of a representation that is unique for every single environment. 00:53:17.653 --> 00:53:20.913 So it's much more linked to individual memories. 00:53:20.913 --> 00:53:26.673 But now, I could also radicalize the view and say, yeah, but if I can just hijack 00:53:26.673 --> 00:53:30.753 my velocity signal, which in theory I could because it also travels over the thalamus, 00:53:30.893 --> 00:53:35.393 then I could basically impose any kind of metric onto the hippocampus. 00:53:35.433 --> 00:53:36.793 It doesn't know. It's not a speed. 00:53:36.873 --> 00:53:39.173 It doesn't represent space. It represents something else. 00:53:39.733 --> 00:53:42.553 But so do you see that grid cells 00:53:42.553 --> 00:53:45.633 might have that flexibility to also represent other kinds of metrics? 00:53:46.113 --> 00:53:48.293 Or do you really see it anchored to space? 00:53:50.233 --> 00:53:54.273 Well, that's hard to say. I mean, it certainly has the potential for doing it. 00:53:54.313 --> 00:53:59.753 But as far as we have observed, it's really related to space in the rat at least. 00:54:00.576 --> 00:54:06.756 So hippocampus is more different. I would think that the strict relationships, 00:54:06.756 --> 00:54:13.396 for example, between the animal's speed and the subtle changes in movement of 00:54:13.396 --> 00:54:15.856 grid fields in the entorhinal cortex, 00:54:16.176 --> 00:54:19.616 all these relationships are absent in the hippocampus. 00:54:19.616 --> 00:54:22.896 It's just that the signal, the influence of speed is much more indirect, 00:54:23.216 --> 00:54:26.176 even if it can be detected there. So I think it's secondary. 00:54:27.296 --> 00:54:33.356 But then, for instance, one view, not certainly that the view on grid cells 00:54:33.356 --> 00:54:35.416 has complexified in terms of 00:54:35.416 --> 00:54:39.216 we have moved away from a linear feed-forward interaction to hippocampus. 00:54:39.756 --> 00:54:44.116 Some people are suggesting that we can also then use that system for mind travel. 00:54:44.356 --> 00:54:48.616 This might explain how we can then use sweeps in hippocampus to sort of look 00:54:48.616 --> 00:54:51.016 ahead and do mind travel. 00:54:51.016 --> 00:54:54.056 Travel so do you also see 00:54:54.056 --> 00:54:58.396 that as a as a possible secondary function 00:54:58.396 --> 00:55:01.456 of this system or do you see that this is physiologically not 00:55:01.456 --> 00:55:04.456 fully supported no i mean at least in 00:55:04.456 --> 00:55:07.696 humans or primates i wouldn't be surprised if if 00:55:07.696 --> 00:55:10.836 mind travel is a central function of 00:55:10.836 --> 00:55:13.756 of grid cells perhaps even in the 00:55:13.756 --> 00:55:16.916 rat because i mean you 00:55:16.916 --> 00:55:19.916 can recreate activity based on memories 00:55:19.916 --> 00:55:23.476 so you can sort of re reactivate patterns 00:55:23.476 --> 00:55:28.736 that happen in real space and to some extent you see that even in in rodents 00:55:28.736 --> 00:55:33.836 in sleep that you that at least in hippocampus patterns of activity are replayed 00:55:33.836 --> 00:55:40.576 from that happen patterns that happened in awake experience are replayed in sleep. 00:55:40.776 --> 00:55:44.756 And probably, because this is expressed across many brain years, 00:55:44.816 --> 00:55:47.716 it's very likely to happen in the grid cell system too. 00:55:47.876 --> 00:55:54.376 So who knows? I mean, you can probably at least conceivably replay trajectories 00:55:54.376 --> 00:55:56.996 also in the entorhinal cortex. 00:55:57.176 --> 00:56:02.056 And if you can do that in sleep, then, I mean, that would probably also apply 00:56:02.056 --> 00:56:09.136 in a certain drowsy awake state. you can probably also do the same I would think. 00:56:09.396 --> 00:56:11.516 Right. So now um. 00:56:12.828 --> 00:56:19.748 You're sort of halfway your career. You have plenty of years in front of you to change the universe. 00:56:20.748 --> 00:56:23.668 But a lot of experience in studying the brain. 00:56:24.308 --> 00:56:28.788 With, of course, magnificent success. We also have to mention the Nobel Prize 00:56:28.788 --> 00:56:33.848 you shared with your wife and with John O'Keefe last year. 00:56:34.548 --> 00:56:40.188 So in that sense, you are in a unique position to also view the process of doing our science. 00:56:40.188 --> 00:56:46.608 The process of understanding the way in which we can make progress in coupling 00:56:46.608 --> 00:56:49.608 mind and brain in this case, bringing psyche and the flesh together. 00:56:49.808 --> 00:56:55.108 So in that sense, if we would follow your example, what is Edwards' law that 00:56:55.108 --> 00:56:57.808 we should follow in the study of mind and brain? 00:56:59.648 --> 00:57:13.528 No, that I don't know. I think one law is at least just to be brave and try to, 00:57:13.648 --> 00:57:21.648 if I would give any advice to young students, is that take one step longer than you usually think. 00:57:21.788 --> 00:57:24.168 So try to do what you think is impossible. 00:57:24.588 --> 00:57:29.228 And that's what we have tried through the years. Yes, we have used the opportunities 00:57:29.228 --> 00:57:35.228 and done experiments that perhaps we wouldn't have done if we had limited funding or so. 00:57:36.748 --> 00:57:42.428 And that has paid off. It's high risk, high gain, it's often called. 00:57:43.088 --> 00:57:45.228 So, Edward Sloat, be brave. Yes. 00:57:46.228 --> 00:57:51.088 So, then, look, Tony has plenty of money, so he likes to buy me tickets to fly 00:57:51.088 --> 00:57:54.548 all over the place. So five years from now, we're going to visit you in Trondheim 00:57:54.548 --> 00:57:58.788 because we're going to confront you with the outcome of a prediction you're going to make now. 00:57:58.868 --> 00:58:04.668 So what prediction do you want to make today that you will show to us you have 00:58:04.668 --> 00:58:10.468 validated five years from now, which will be a major next insight in how the brain operates? Yes. 00:58:11.874 --> 00:58:18.134 Well, I mean, in how the brain operates, I think I want to keep it to the circuits 00:58:18.134 --> 00:58:20.154 that I've been usually studying. 00:58:20.614 --> 00:58:26.114 But I think one of the questions that we really work on is, I think if we understand 00:58:26.114 --> 00:58:30.634 how the grid cell pattern is generated, that will tell us a lot, 00:58:30.654 --> 00:58:35.094 not only about grid cells, but actually how patterns are formed in general in the brain. 00:58:35.094 --> 00:58:40.654 So one of the predictions we're trying to test is whether the grid cell network 00:58:40.654 --> 00:58:45.254 actually has the connectivity that is required for such patterns to occur. 00:58:45.534 --> 00:58:49.714 For example, if cells with similar properties are linked and if cells with, 00:58:49.774 --> 00:58:54.654 also if they are linked in the way that you would predict if this network is 00:58:54.654 --> 00:59:00.934 going to use inputs from the environment to update the internal map. 00:59:00.934 --> 00:59:04.254 So that's not a very specific. 00:59:05.174 --> 00:59:14.494 No, no. Well, I mean, the specific prediction is that cells that fire at similar 00:59:14.494 --> 00:59:18.474 locations are preferentially connected. So that's one single. 00:59:18.774 --> 00:59:21.794 If you want to be specific, then you get much more details. 00:59:22.314 --> 00:59:26.874 But it's easy to test. All right, Edwin Mosers, thank you very much for this conversation. 00:59:27.874 --> 00:59:33.234 Okay, thank you. The CSN Podcast was produced by the Convergent Science Network 00:59:33.234 --> 00:59:41.274 of Biometrics and Biohybrid Systems, a project funded by the European 7th Research Framework Program. 00:59:42.874 --> 00:59:48.174 For more interviews, recorded lectures, or upcoming conferences in the field 00:59:48.174 --> 00:59:54.414 of biometrics and biohybrid systems, go to csnnetwork.eu. 00:59:54.960 --> 01:00:09.652 Music.