WEBVTT 00:00:00.017 --> 00:00:06.137 Yeah, right. I guess we need some guidance on how to interpret it. Oh, yeah, we can do that. 00:00:06.797 --> 00:00:08.677 That's not a problem. Okay, good. 00:00:10.597 --> 00:00:14.717 This is the Convergent Science Network podcast. But you'll be interested in Nunga. 00:00:14.917 --> 00:00:18.997 Yeah. Leading researchers in the domain of neuroscience, brain theory, 00:00:19.137 --> 00:00:23.517 and technology are interviewed by Paul Vershoor and Tony Prescott. 00:00:24.217 --> 00:00:28.637 This is Paul Vershoor with the Convergent Science Network podcast, 00:00:28.637 --> 00:00:36.757 And I'm here with my colleague Tony Prescott in the BCBT workshop of 2014. 00:00:36.937 --> 00:00:44.677 And we're with our speaker, Henry Kennedy, who has been talking about architecture 00:00:44.677 --> 00:00:46.297 of a particular neocortex. 00:00:46.997 --> 00:00:51.697 And Henry, you sort of, you came in not sort of opening the door, 00:00:51.757 --> 00:00:54.597 you sort of knocked down the door by just saying, look, all the stuff you guys 00:00:54.597 --> 00:01:00.337 know and love is wrong. like small world networks you should take with a big grain of salt. 00:01:01.777 --> 00:01:03.457 Standard ideas about how we 00:01:03.457 --> 00:01:08.457 think about the connectivity of the human neocortex are maybe incorrect. 00:01:09.177 --> 00:01:12.897 But what's your position exactly towards these standard views? 00:01:14.297 --> 00:01:18.017 Not that the small world network is wrong, It's not. 00:01:18.197 --> 00:01:23.917 It's simply a proposition based on inadequate data. 00:01:24.617 --> 00:01:31.637 If you look at the complete data of inter-aerial networks, it just so happens 00:01:31.637 --> 00:01:32.857 that they're very high density. 00:01:33.757 --> 00:01:39.877 At that kind of density, the small world model is not appropriate. 00:01:40.197 --> 00:01:41.457 It doesn't tell you anything about it. 00:01:43.177 --> 00:01:47.437 Actually, Actually, what we're trying to say is that what does give you a much 00:01:47.437 --> 00:01:54.097 more meaningful explanation about what the architecture corresponds to is if 00:01:54.097 --> 00:01:59.517 you take into consideration the distance and the weight, the strength of connections. 00:01:59.957 --> 00:02:03.517 And this breaks away completely from the small world tradition because you're 00:02:03.517 --> 00:02:07.837 no longer dealing with a binary network. You're dealing with a directed weighted network. 00:02:08.217 --> 00:02:10.017 That was the point we were trying to make. 00:02:11.049 --> 00:02:16.369 So you're saying it's just too crude a view to really help you understand how a cortex is wired up. 00:02:16.629 --> 00:02:21.469 It's ignoring too much about what actually specifies the network. 00:02:21.589 --> 00:02:28.269 If you have a network which has a density of 70%, it's not what is connected 00:02:28.269 --> 00:02:33.409 to what that will tell you very much, it's how strongly which area is connected to which area. 00:02:33.609 --> 00:02:38.149 It's the strength of the connections, and it's a wide range of strengths that 00:02:38.149 --> 00:02:43.329 makes the cortical network work so very, very surprising in some ways. 00:02:43.409 --> 00:02:45.869 There's a big range of strengths of connections. 00:02:46.129 --> 00:02:51.629 They cross over five orders of magnitude, and it's taking those strengths into 00:02:51.629 --> 00:02:57.149 consideration that you'll have an insight into what is actually the fingerprint, 00:02:57.409 --> 00:03:00.469 the connectivity profile of a particular area. Right. 00:03:00.969 --> 00:03:07.089 So now you've been spending a lot of time doing very detailed studies using 00:03:07.229 --> 00:03:12.929 tracer injections in understanding the connectivity in particular of the cat neocortex. 00:03:13.249 --> 00:03:19.109 And I guess with emphasis on the visual areas, right? I started off in cat visual 00:03:19.109 --> 00:03:21.069 cortex many, many years ago, yes. 00:03:22.249 --> 00:03:26.109 More years than I care to remember, 20 years maybe since I've touched a cat, yeah. Right. 00:03:27.009 --> 00:03:33.729 But just to make the point that you're not basing your conclusions on let's 00:03:33.729 --> 00:03:37.229 say a functional description of the system or in terms of what you might want 00:03:37.229 --> 00:03:38.489 to call a functional connectivity, 00:03:38.809 --> 00:03:47.389 you base it really on a detailed study of, let's say, injected cells, right? 00:03:47.449 --> 00:03:50.269 So it's a very direct measure of connectivity. Okay. 00:03:50.849 --> 00:03:54.809 Are these... So then given that data set... 00:03:55.862 --> 00:04:01.042 How do you think of, what is the template you have in mind of cortical connectivity? 00:04:01.422 --> 00:04:05.242 How is it different from what a small world would tell you? How is it really 00:04:05.242 --> 00:04:06.802 wired up? What are the wiring rules? 00:04:08.742 --> 00:04:16.422 Well, there is a wiring rule, and that we felt was the backbone of the publications 00:04:16.422 --> 00:04:17.642 we've been making recently. 00:04:17.842 --> 00:04:20.382 There is a very simple, very straightforward 00:04:20.382 --> 00:04:24.222 wiring rule, and that is that there's a minimization of wire. 00:04:24.222 --> 00:04:29.042 And we're certainly not the first people to find it but what I think we have 00:04:29.042 --> 00:04:34.062 been able to put our finger on is a principle which explains that wire minimization 00:04:34.062 --> 00:04:38.802 which is such a strong constraining feature of the cortex. 00:04:39.082 --> 00:04:42.082 So the constraint basically that 00:04:42.082 --> 00:04:45.162 we've been able to demonstrate is that 00:04:45.162 --> 00:04:51.262 there's a weight-distance relationship so there's an exponential fall of strength 00:04:51.262 --> 00:04:58.642 of connection with distance And if you take that on board and you generate random 00:04:58.642 --> 00:05:02.182 networks based on that particular feature, 00:05:02.402 --> 00:05:07.122 using the space constants, the lambda value that we've observed, 00:05:07.602 --> 00:05:14.142 you generate networks which in many ways reflect the properties of the cortical 00:05:14.142 --> 00:05:17.882 network that you can look at down the microscope that you can actually measure. 00:05:18.082 --> 00:05:21.682 Just to clarify, we're talking about a primate, not a cat, aren't we? 00:05:22.362 --> 00:05:27.682 It's a non-human primate. But then, isn't this contradictory what you said earlier about weights? 00:05:27.942 --> 00:05:32.722 Because the data is roughly telling you that over a distance of, 00:05:32.762 --> 00:05:38.182 let's say, 60, 70 millimeters, you would have a roughly exponential kind of 00:05:38.182 --> 00:05:41.882 decay of the probability to connect and of the connection strength. 00:05:42.262 --> 00:05:45.402 Yes. But earlier you said the…. 00:05:46.545 --> 00:05:51.805 The small world view doesn't help you because you have maybe low probability 00:05:51.805 --> 00:05:54.045 connections, but their strength matters. 00:05:54.385 --> 00:05:59.785 Their strength actually can tip the balance, if you want, into a significant 00:05:59.785 --> 00:06:00.965 function relation or not. 00:06:01.045 --> 00:06:04.685 But now your data shows that the weight also drops off with distance. 00:06:05.045 --> 00:06:09.365 So would that then not undercut your earlier argument against the small world network? work? 00:06:10.365 --> 00:06:16.465 Well, the point about the criticism about the small world network is simply 00:06:16.465 --> 00:06:22.685 if you take the density, that is to say the number of connections that you have 00:06:22.685 --> 00:06:24.065 between different cortical areas, 00:06:24.345 --> 00:06:26.885 it turns out to be about 70%. 00:06:26.885 --> 00:06:32.165 So 70% of the connections that can exist actually do exist. That's a very, very high density. 00:06:32.505 --> 00:06:36.725 And at that density, you have a small world. So you don't have to measure the 00:06:36.725 --> 00:06:39.125 clustering index, path length, or what have you. 00:06:39.465 --> 00:06:44.445 Everything is virtually connected to everything else. Every area is one and a half hops away. 00:06:44.865 --> 00:06:48.685 Now, what you're referring to now, the fact is that when you look over very 00:06:48.685 --> 00:06:52.825 long distance, yes, very few areas are connected, and that's interesting because 00:06:52.825 --> 00:06:56.145 it means that you have a sort of binary specificity over those distances. 00:06:56.705 --> 00:07:01.705 Presence of a connection in itself is going to be significant in terms of trying 00:07:01.705 --> 00:07:06.825 to understand the biology. Our point is that globally, overall, 00:07:07.185 --> 00:07:09.225 it's going to be the strength of connections which matter. 00:07:09.465 --> 00:07:11.925 But there are exceptions over these long distances. 00:07:12.165 --> 00:07:14.005 So for example, if you take... 00:07:15.262 --> 00:07:20.022 The frontal eye field projection to area v4 that connection is actually rather 00:07:20.022 --> 00:07:23.802 strong it's stronger than what you would predict so it's an it's an outlier 00:07:23.802 --> 00:07:28.982 and i think that that's uh something which comes out of uh our investigations 00:07:28.982 --> 00:07:31.642 which i think need to be considered in more detail, 00:07:32.722 --> 00:07:35.762 you're really doing the analysis two spatial scales here 00:07:35.762 --> 00:07:39.702 so you're talking about connectivity between cortical 00:07:39.702 --> 00:07:42.502 areas yes of what which how many are 00:07:42.502 --> 00:07:45.322 we talking there sort of well we're working with an atlas 00:07:45.322 --> 00:07:49.142 of 91 areas in the in the macaque monkey right and 00:07:49.142 --> 00:07:52.202 so and then at another spatial scale you're talking about 00:07:52.202 --> 00:07:58.142 connectivity within an area which you're saying is 80 to 90 percent of the connections 00:07:58.142 --> 00:08:03.122 are within cortical areas yes yeah but and then spilling out into surrounding 00:08:03.122 --> 00:08:09.102 areas with the remaining 10 to 20 percent of connections yeah yeah so um but 00:08:09.102 --> 00:08:10.662 i think when you're analyzing the data, 00:08:10.822 --> 00:08:15.602 are you analyzing it in different ways when you're doing these two analyses or is? 00:08:15.762 --> 00:08:21.202 We haven't, the connectivity within the cortical area, this is a local connectivity, if you will. 00:08:21.922 --> 00:08:28.322 So if you take a point in a cortical area, 80% of the projections to that point 00:08:28.322 --> 00:08:30.622 come from within two millimeters actually. 00:08:30.862 --> 00:08:35.702 That's the local connectivity and we're not doing, we're not looking at that in any great detail. 00:08:36.702 --> 00:08:40.442 Actually, it should be looked at because there's been very little work on that. 00:08:40.542 --> 00:08:44.862 So you have the canonical model, which is developed by Kevin Martin and Rodney 00:08:44.862 --> 00:08:46.922 Douglas, that was for area B1 of the cat. 00:08:47.742 --> 00:08:51.342 It's never been done for other cortical areas. We don't know what that looks 00:08:51.342 --> 00:08:53.102 like. We're not doing that. 00:08:53.522 --> 00:08:56.982 So when you say most of the connections are within an area, they're actually 00:08:56.982 --> 00:09:01.462 within a very small part of that area. Absolutely, yeah. 00:09:02.182 --> 00:09:06.442 Well, I think that the range you were talking about in the data you presented 00:09:06.442 --> 00:09:11.002 another day was, like I said earlier, up to 80 millimeters or something, right? Not beyond that. 00:09:11.642 --> 00:09:14.902 For the scaling loss that you showed. Yeah, for the macaque monkey, 00:09:15.142 --> 00:09:19.282 it's about 7 centimeters is about the size of the brain, yeah. 00:09:20.302 --> 00:09:22.642 Okay. Of course. 00:09:24.602 --> 00:09:26.042 So, what is then the, 00:09:27.442 --> 00:09:32.322 So if we now look at these rules of connectivity that you presented, 00:09:33.462 --> 00:09:37.062 that seems to suggest that indeed, like you said earlier, everything is connected 00:09:37.062 --> 00:09:40.202 to everything, directly or indirectly. 00:09:41.102 --> 00:09:45.922 So that would seem a little bit unspecific to talk about, let's say, 00:09:46.542 --> 00:09:48.762 an architecture that actually has a certain function. 00:09:48.962 --> 00:09:51.142 And also when you look at the physiological properties of it, 00:09:51.242 --> 00:09:54.262 it doesn't necessarily look as some sort of uniform structure. 00:09:54.502 --> 00:09:59.502 It just seems much more fractionated in its dynamics. So, how do you match these two? 00:09:59.922 --> 00:10:06.382 So, is there an element of fractionation and specialization that we just have not seen in this data? 00:10:06.682 --> 00:10:09.702 Well, I think you were describing it, though, weren't you, with the Dossler and Banchel stream? 00:10:10.102 --> 00:10:13.502 There was some fractionation. Well, no, this is what I want to get to, right? 00:10:13.602 --> 00:10:19.462 So, how do we get, just at face value of this Markov kind of data, 00:10:19.582 --> 00:10:23.742 the scaling laws, you could say, okay, well, this looks pretty uniform wherever I go. 00:10:23.842 --> 00:10:26.922 It's sort of connected in a similar way. way, similar kinds of weight distributions, 00:10:27.142 --> 00:10:28.182 similar kind of length distributions. 00:10:28.802 --> 00:10:33.042 But how does it give you functional specialization of areas and functionally 00:10:33.042 --> 00:10:34.942 organized forms of dynamics? 00:10:35.182 --> 00:10:40.282 That's the question, though. Okay. So the short answer to that is the binary 00:10:40.282 --> 00:10:42.922 specificity is low, as you pointed out. 00:10:43.802 --> 00:10:47.982 Everything is not connected to everything, but there is, at least within. 00:10:49.588 --> 00:10:55.408 Within, say, 15 millimeters of an area, there is a very, very high connectivity, 00:10:55.528 --> 00:10:59.948 so you are virtually approaching 80%, maybe 90%. 00:11:00.768 --> 00:11:03.708 So it's going to be the strength of connections which are important. 00:11:04.068 --> 00:11:06.208 So we've looked at the global properties of these networks. 00:11:06.448 --> 00:11:11.488 So you can make an efficiency measure, which is a sort of conductance where 00:11:11.488 --> 00:11:15.608 you're treating the strength of connection as an inverse resistance. 00:11:15.608 --> 00:11:21.948 When you do that, you look at local efficiency and global efficiency, and then you can look, 00:11:22.608 --> 00:11:29.408 how the global efficiency, the local efficiency, is affected by fresh-holding, 00:11:29.428 --> 00:11:34.748 removing the weakest connections until you have just the backbone of very strong connections, 00:11:35.008 --> 00:11:41.088 and then look how your efficiency changes with removal of connections, 00:11:41.448 --> 00:11:43.568 and show that the, 00:11:45.148 --> 00:11:50.388 network which you've produced using the exponential distance rule actually very, 00:11:50.508 --> 00:11:56.828 very faithfully mimics the efficiency changes you can observe as you remove connections. 00:11:58.908 --> 00:12:05.648 The point I want to make is that this distance rule, the exponential distance 00:12:05.648 --> 00:12:09.368 rule, actually gives you a handle on looking at global properties, 00:12:09.548 --> 00:12:15.668 not just the motives and the click, which I was talking about earlier on. 00:12:15.788 --> 00:12:19.128 The click actually is coming back to the core of the previous speaker, 00:12:19.288 --> 00:12:21.948 so it's not without its own significance. 00:12:21.948 --> 00:12:29.048 The strength of connection does give you a very strong degree of specificity, 00:12:29.088 --> 00:12:32.828 but again, it requires looking at the strength of connections. 00:12:33.028 --> 00:12:38.828 I think one of the reasons why this has been perhaps ignored in the literature is, first of all. 00:12:40.226 --> 00:12:43.026 Um you can't see this kind of range of strength of 00:12:43.026 --> 00:12:47.326 connections with diffusion mri you can see 00:12:47.326 --> 00:12:50.226 it with track tracing if you're going to do this with track tracing then 00:12:50.226 --> 00:12:54.986 you've got to count neurons or synapses and that's a lot of work so if you want 00:12:54.986 --> 00:12:59.886 to have that kind of data then you you can't just do what a lot of us have been 00:12:59.886 --> 00:13:04.246 doing for many many years and myself included which is saying strong weak and 00:13:04.246 --> 00:13:07.586 medium strong weak and medium isn't going to tell you about 00:13:07.726 --> 00:13:11.366 the range of strengths of connections where the specificity is coming in. 00:13:11.446 --> 00:13:16.546 It requires a much more thorough approach than that. 00:13:16.646 --> 00:13:20.106 And what that shows you is that you have these five orders of magnitude. 00:13:20.426 --> 00:13:25.506 That means that the counts that you're making run into very, very high numbers. 00:13:26.446 --> 00:13:29.406 And that's an important point. 00:13:29.706 --> 00:13:34.266 But the other thing that I would like to understand better, So if I take now 00:13:34.266 --> 00:13:39.326 these scaling laws for weights and for the probability to connect, 00:13:39.446 --> 00:13:41.206 these are probabilistic. They're probability distributions. 00:13:41.826 --> 00:13:46.266 So I just take a lattice of units and I'm going to wire them up following these 00:13:46.266 --> 00:13:51.546 two probability distributions of lateral conductivity and the strength of these 00:13:51.546 --> 00:13:52.646 connections. directions, right? 00:13:53.066 --> 00:13:57.886 Then you would expect if I start to sort of chip away those guys and I removed 00:13:57.886 --> 00:14:02.946 the weights with the smallest, the lowest value, the lowest strength, 00:14:03.106 --> 00:14:05.126 that I would get also the random patterns. 00:14:05.526 --> 00:14:10.046 Because these two scaling laws or wiring laws themselves don't give me any kind 00:14:10.046 --> 00:14:11.366 of symmetry breaking, right? 00:14:11.446 --> 00:14:18.646 So if I do this a million times, I would have a million different kinds of topologies or not. 00:14:18.646 --> 00:14:25.886 No, you don't. What you find when you take the data, the interaerial connectivity 00:14:25.886 --> 00:14:31.006 database that we have, which is 29 areas and some hundreds of connections, 00:14:31.366 --> 00:14:36.286 and you remove the weakest connections, you get a backbone. 00:14:36.506 --> 00:14:38.386 You find the backbone of the cortex, 00:14:38.646 --> 00:14:42.346 and that gives you those features and characteristics of that backbone. 00:14:42.646 --> 00:14:46.366 When you do the same thing to your random networks that you generated using 00:14:46.366 --> 00:14:53.206 the exponential distance rule that we have, what we're saying is that is a probability, 00:14:53.466 --> 00:14:57.646 so you pick more frequently the very strong short-distance connections, what have you. 00:14:58.006 --> 00:15:03.646 When you remove the weak connections and you end up eventually with your backbone, 00:15:03.846 --> 00:15:07.066 you find that it has many of the features of the backbone in the data that you've 00:15:07.066 --> 00:15:08.766 observed. Which features do you recover best? 00:15:10.105 --> 00:15:15.365 Well, I mentioned the motives. That's pretty good. The correspondence between 00:15:15.365 --> 00:15:16.865 the motives is actually excellent. 00:15:17.305 --> 00:15:22.385 So you have 16 different motives of, say, three nodes. 00:15:23.045 --> 00:15:29.505 Of those 16 different motives, your distribution, which is captured by the exponential 00:15:29.505 --> 00:15:35.985 distance rule, is actually excellent. Then I referred to this question of hubs. 00:15:36.145 --> 00:15:42.545 Now, people have been finding hubs, let's just say, areas with high degree distribution 00:15:42.545 --> 00:15:46.165 to them, a large number of connections with other cortical areas. 00:15:47.785 --> 00:15:54.565 And the idea of the cortical core is that the hubs form more connections between 00:15:54.565 --> 00:15:57.225 themselves than you would predict statistically. 00:15:57.225 --> 00:16:00.125 Statistically so that's actually uh what is 00:16:00.125 --> 00:16:03.305 referred to as a rich club the rich club analysis was introduced 00:16:03.305 --> 00:16:06.485 in in network science uh by coriza 00:16:06.485 --> 00:16:09.165 and other people about seven or eight 00:16:09.165 --> 00:16:11.965 years ago now you can't do that 00:16:11.965 --> 00:16:15.145 kind of analysis on a high density network it 00:16:15.145 --> 00:16:18.045 the normalization doesn't work so what 00:16:18.045 --> 00:16:20.865 we've done is look at the clicks so the clicks are 00:16:20.865 --> 00:16:24.005 sets of areas which are 100 connected 00:16:24.005 --> 00:16:26.705 connected amongst themselves and we find a very very 00:16:26.705 --> 00:16:29.365 large number of clicks and when you look at 00:16:29.365 --> 00:16:33.465 the probability of finding that by chance it's extremely low when we do this 00:16:33.465 --> 00:16:40.265 same analysis of networks which we've constructed using this EDR we find the 00:16:40.265 --> 00:16:45.545 same number of clicks with a very very similar frequency and the correspondence 00:16:45.545 --> 00:16:47.205 is really very remarkable, 00:16:48.085 --> 00:16:53.005 you're not imposing any other constraints you just follow this exponential distance 00:16:53.005 --> 00:16:55.085 rule and that's it Absolutely. Okay. 00:16:56.115 --> 00:17:01.335 So some fairly simple geometric rules are giving us a lot of information about 00:17:01.335 --> 00:17:02.975 connectivity within the brain. 00:17:03.415 --> 00:17:08.775 Do you then draw inferences about the developmental processes that are building brains? 00:17:09.315 --> 00:17:12.995 And because this has been a week where we've looked at evolution and development 00:17:12.995 --> 00:17:17.295 and their impact on the way the adult brain is. 00:17:17.295 --> 00:17:23.695 And your data, I guess, gives hope to people that want there to be some useful 00:17:23.695 --> 00:17:25.195 developmental rules that we can 00:17:25.195 --> 00:17:30.955 apply, perhaps, to build a brain on which experience then has some impact. 00:17:31.315 --> 00:17:36.775 Right. I think that's a very important issue, which we haven't looked at, 00:17:36.835 --> 00:17:42.975 despite the fact that I also have another side to myself, which is to do with cortical development. 00:17:43.415 --> 00:17:49.175 I mentioned in my presentation that there's this log-normal distribution of weights. 00:17:51.435 --> 00:17:56.375 A number of studies have looked at distribution of synaptic weights at the single-cell 00:17:56.375 --> 00:17:58.655 level and also find a log-normal distribution. 00:17:59.315 --> 00:18:07.115 There's a recent paper review in Nature Neuroscience looking at log-normal distributions 00:18:07.115 --> 00:18:09.975 in frequencies of firing and another sort of phenomena. 00:18:10.295 --> 00:18:13.955 It seems to be a sort of signature you're finding a lot. 00:18:15.375 --> 00:18:20.135 The log-normal distribution in itself suggests that there could be a very simple 00:18:20.135 --> 00:18:24.375 algorithm which would be controlling outgrowth of axons. 00:18:25.195 --> 00:18:29.715 For many years, I was interested in the formation of connections, 00:18:30.015 --> 00:18:32.315 and that was in cat and monkey cortex. 00:18:32.615 --> 00:18:39.855 In those days, the predominant theme in cortical development was this notion of exuberance. 00:18:39.975 --> 00:18:45.075 You had an overproduction of connections and the selective pattern, 00:18:45.215 --> 00:18:50.455 the precise pattern which is characteristic of the adult, emerges through a pruning process. 00:18:52.035 --> 00:18:57.115 This pruning process is actually necessary because there's not enough information 00:18:57.115 --> 00:18:59.955 in the genome to set up the correct connectivity. 00:19:00.775 --> 00:19:06.435 We challenged that, And every time we challenged it, it turned out not to be 00:19:06.435 --> 00:19:08.275 the case. So what did you find? 00:19:08.655 --> 00:19:13.875 Well, in the case of inter-hemispheric connections, that was the model which 00:19:13.875 --> 00:19:21.635 was hugely used between 1970 and 1985 by Giorgio Innocenti, 00:19:21.955 --> 00:19:25.935 Doug Frost, and many, many people were looking at Herb Kalacki. 00:19:25.935 --> 00:19:32.935 Monkey, there you could observe a widespread connectivity in the very young 00:19:32.935 --> 00:19:36.735 animal which would decrease as the animal would mature. 00:19:36.935 --> 00:19:44.315 We challenged that by looking at the collosal connectivity of the prenatal monkey in the visual system. 00:19:44.715 --> 00:19:51.355 The characteristic of the monkey is that the area V1 is a collosal in the adult. 00:19:51.535 --> 00:19:54.015 Then you can really ask the question, is there. 00:19:54.504 --> 00:19:58.504 Is it a collosal during development? Because it shouldn't be if this exuberancy 00:19:58.504 --> 00:20:00.964 rule is going to be general. And it's not. 00:20:01.264 --> 00:20:06.124 We were able to show that there's never a collosal connection going into area V1 of the monkey. 00:20:06.464 --> 00:20:10.544 So I think it's very much a problem. You've got to be able to distinguish with 00:20:10.544 --> 00:20:18.344 the growth of the brain and the expansion and the increasing and therefore the decreasing density. 00:20:18.344 --> 00:20:21.024 Density simply because of the expansion you're going 00:20:21.024 --> 00:20:23.944 to be able to sort that out from the actual 00:20:23.944 --> 00:20:26.764 creation of the specificity so that was 00:20:26.764 --> 00:20:30.944 in inter inter hemispheric connections but we then did the same thing looking 00:20:30.944 --> 00:20:38.144 at the connections between v2 and v4 which are in uh you have these bands where 00:20:38.144 --> 00:20:42.724 you have projections to v4 and projections to mt and we were able to show that 00:20:42.724 --> 00:20:45.524 they're correctly aligned from the word go. 00:20:45.704 --> 00:20:52.204 There's an overall decrease in density, but it's not the decrease in density 00:20:52.204 --> 00:20:53.604 that creates the pattern. 00:20:54.164 --> 00:20:59.664 But then this old-fashioned idea, we can call it now, of exuberance and pruning 00:20:59.664 --> 00:21:05.604 would still hold, but for a smaller set of, let's say, neurons and connections? 00:21:05.944 --> 00:21:09.544 Or you think it's really the wrong way to think about how the system gets wired 00:21:09.544 --> 00:21:12.024 up? I think it might be an oversimplification. 00:21:12.244 --> 00:21:15.884 I think the idea that you You don't have enough genes to specify connections 00:21:15.884 --> 00:21:18.224 is not really very interesting. 00:21:18.344 --> 00:21:23.184 I think the question, the confusion which was made was not, is there a decrease 00:21:23.184 --> 00:21:26.144 in density of connections with growth? There is. 00:21:26.824 --> 00:21:31.324 The question which was posed was, 00:21:31.444 --> 00:21:36.344 is it that decrease in connections which creates the specific pattern? 00:21:36.624 --> 00:21:40.704 And that, I think, has never really been satisfactorily shown to be the case. 00:21:40.704 --> 00:21:45.024 I think it's not, I'm not saying it plays absolutely no role at all, 00:21:45.124 --> 00:21:51.544 but I think there's much more, there's much more guided growth and target specification. 00:21:52.104 --> 00:21:57.244 But couldn't you argue that if you apply the exponential distance rule to a developing brain, 00:21:57.564 --> 00:22:00.824 you should see something that might look like pruning by definition, 00:22:01.044 --> 00:22:07.184 because you're imposing a distance relationship of your connectivity in an expanding volume? 00:22:07.804 --> 00:22:11.484 Well, I think that would be an interesting thing to look at. 00:22:11.784 --> 00:22:17.704 Simply look at the brain and see how does this exponential distance rule look 00:22:17.704 --> 00:22:19.724 in a very immature animal. 00:22:19.944 --> 00:22:24.944 The problem is that numbers get very big, and you really need much more automated 00:22:24.944 --> 00:22:28.024 techniques in counting, which are coming through actually. 00:22:28.544 --> 00:22:34.084 So I think that sort of thing is something that will be addressable. 00:22:35.024 --> 00:22:41.104 So one of the other things that you discussed today was how the primate brain 00:22:41.104 --> 00:22:48.124 seems to minimize the wiring length as a consequence of applying these rules but and. 00:22:49.851 --> 00:22:55.611 Bearing in mind this point about changing brain size, we also discussed some 00:22:55.611 --> 00:23:02.351 data about animals with smaller brains like mice, and that the same rules don't necessarily apply. 00:23:02.691 --> 00:23:08.291 So are there some scaling issues here for brains that perhaps would help us 00:23:08.291 --> 00:23:09.771 unravel some of these issues? 00:23:11.231 --> 00:23:14.691 Absolutely, yes. This is something we're very interested in looking at. 00:23:14.691 --> 00:23:20.131 With Zoltan Tarakai and his postdoc who's now back in Romania. 00:23:20.471 --> 00:23:25.051 We have funding with John Cass to look exactly at that. 00:23:26.031 --> 00:23:31.731 We're looking at the mouse and the microcebus, which is a small primate, 00:23:32.251 --> 00:23:37.791 and we want to look at the baboon, which is about as big as we can go in primates. That's not a huge deal. 00:23:38.051 --> 00:23:44.711 The idea is to see how this exponential distance rule plays out in changes in brain size. 00:23:47.611 --> 00:23:53.111 We're still working with our own database, which we're putting together with Andreas Burkhalter. 00:23:53.711 --> 00:23:58.351 To keep us going on this, we've been looking at the Allen Brain Institute database, 00:23:58.651 --> 00:24:00.611 and it's actually really rather interesting. 00:24:01.437 --> 00:24:06.097 So this is very, very preliminary, and with Zoltan Torukai and Ken Koblok and 00:24:06.097 --> 00:24:08.757 our colleagues, we're still analyzing this data. 00:24:08.897 --> 00:24:16.637 But the glaring thing is in the mouse, there's a huge amount of wire economy 00:24:16.637 --> 00:24:17.937 that you can impose on it. 00:24:18.037 --> 00:24:23.757 So the areas in the brain are not localized optimally as they are in the primate. 00:24:23.757 --> 00:24:26.757 And and that makes it look straight 00:24:26.757 --> 00:24:29.697 away very strange because you have areas which are 00:24:29.697 --> 00:24:33.397 not connected which are actually quite near to another area and they're not 00:24:33.397 --> 00:24:37.477 connected to it and that's what you you absolutely don't see in in in the monkey 00:24:37.477 --> 00:24:44.277 and um the the thing that uh we're very interested with tori kai and and uh 00:24:44.277 --> 00:24:49.397 the the the group in general is to see how folding comes into to this. 00:24:49.957 --> 00:24:55.097 So the comparison between mouse and monkey is a bit complicated because we have 00:24:55.097 --> 00:24:58.457 a change in brain size, but we also have a change in folding. 00:24:58.797 --> 00:25:02.197 So what we've been doing with David Van Essen is we've been, 00:25:02.317 --> 00:25:07.217 instead of taking the distances which we've been using up until now, 00:25:07.337 --> 00:25:09.377 which have been white matter distances between areas, 00:25:09.597 --> 00:25:15.037 approximating the trajectory of the axon, we've been taking surface distances. 00:25:15.597 --> 00:25:20.437 And when you do that, you're sort of unfolding the brain as it were right and 00:25:20.437 --> 00:25:25.017 and when you do that you get a very very different uh distribution of distances. 00:25:25.737 --> 00:25:28.917 In the monkey so this is unfolding the monkey brain this 00:25:28.917 --> 00:25:32.237 is flattening the cortical sheet virtually yeah and 00:25:32.237 --> 00:25:35.117 then when you look at your distance distribution you get something 00:25:35.117 --> 00:25:39.837 much flatter it's it's not at all like a gaussian pointed a very pointed gaussian 00:25:39.837 --> 00:25:44.677 and it looks just like the mouse it looks just like the mouse and so we're now 00:25:44.677 --> 00:25:50.697 playing with that and seeing how this distance distribution interacts with this 00:25:50.697 --> 00:25:54.037 exponential distance rule to set up the specificity that we're seeing. 00:25:54.317 --> 00:25:56.537 So what kind of rule would hold in the mouse brain? 00:25:57.512 --> 00:26:01.952 That's a key question. Certainly, the exponential distance rule is part of it. 00:26:03.292 --> 00:26:05.632 But the jury is out completely. 00:26:06.052 --> 00:26:15.192 I mean, Zoltan Tarakay and Maria Havaz is doing simulations on that, and we don't quite know. 00:26:16.212 --> 00:26:21.392 But you seem to suggest this morning that the probability to find long-range 00:26:21.392 --> 00:26:25.732 connections is higher in the mouse brain than in the macaque brain, 00:26:25.892 --> 00:26:26.932 or did I misunderstand that? 00:26:27.512 --> 00:26:33.532 No, that's correct. So when you look in the macaque, you have an exponential 00:26:33.532 --> 00:26:36.832 lambda value with a very sharp drop. 00:26:37.532 --> 00:26:44.032 When you look in the mouse, the decline in strength with distance is much more shallower. 00:26:44.132 --> 00:26:49.812 And that's certainly part of the exponential distance rule being altogether 00:26:49.812 --> 00:26:55.652 specifying much less specificity in your mouse. 00:26:55.652 --> 00:27:00.532 So you we talked about the cortical corn in the macaque you have this very large 00:27:00.532 --> 00:27:04.132 number of clicks Which is extremely improbable in the mouse. 00:27:04.272 --> 00:27:06.912 You have a much smaller number and don't forget what? 00:27:07.652 --> 00:27:10.232 Although the overall number of areas are the same in two species. 00:27:10.532 --> 00:27:15.272 We're actually making these networks on same number of areas using the same 00:27:15.272 --> 00:27:21.792 column Comparable number of areas so The cortical core is much less defined. 00:27:22.032 --> 00:27:27.332 So what I think this is pointing to is Whereas in a world where one might be 00:27:27.332 --> 00:27:33.532 tempted to say, well, the mouse can be a very interesting model for the brain, 00:27:33.712 --> 00:27:36.612 it might be because you can have knockouts, you can have knock-in, 00:27:36.732 --> 00:27:41.012 you can use optogenetics very easily and what have you. 00:27:41.332 --> 00:27:47.852 But simply from these kind of large-scale properties, it appears to be very, very, very different. 00:27:49.752 --> 00:27:55.612 But in the areas you're looking at, the cell densities are comparable to primate? No, they're not. 00:27:55.992 --> 00:28:03.032 So the scaling rules, and John Kaus could talk about this, he's with his colleague 00:28:03.032 --> 00:28:06.432 Herculeano Huzel, they've done a lot of work on that. 00:28:06.532 --> 00:28:09.212 And the scaling rules in primates and rodents are quite different. 00:28:09.932 --> 00:28:12.652 So basically, I mean, the way I've understood it, in fact, 00:28:12.672 --> 00:28:20.172 I want to talk to this about John at this meeting, is that the um your um as 00:28:20.172 --> 00:28:25.652 the brain changes in size you basically you can change the density of cells 00:28:25.652 --> 00:28:32.352 in in rodents and my feeling is that this might lead to a capacity for miniaturization. 00:28:33.710 --> 00:28:40.950 The microcebus is about a centimeter and something, and it's the smallest primate brain that exists. 00:28:41.670 --> 00:28:46.090 And as the lady this morning was pointing out, there's a huge radiation in primates, 00:28:46.130 --> 00:28:47.010 and there's a huge adaptation. 00:28:47.950 --> 00:28:53.370 The mouse, the rodent brain, can go down to tiny little things, 00:28:53.590 --> 00:28:59.350 I mean, they have moles and voles and what have you that have… This means, 00:28:59.430 --> 00:29:04.910 Henry, that possibly the scaling law still holds, but it is modulated by cell density. 00:29:05.050 --> 00:29:07.370 It scales in turn with cell density. 00:29:07.670 --> 00:29:11.230 So the cells are packed more tightly, and that gives you an exponential decay. 00:29:11.770 --> 00:29:15.450 But if the cells are packed more loosely, it gets stretched out. 00:29:15.570 --> 00:29:16.450 Would that be reasonable? 00:29:16.770 --> 00:29:20.270 I think that's what Herculeano Huzel's results are saying. 00:29:20.350 --> 00:29:25.670 I think she's saying that the density has a much bigger variation in rodents 00:29:25.670 --> 00:29:28.870 so that you can make smaller and smaller brains by making smaller and smaller cells. 00:29:29.190 --> 00:29:33.770 But the other thing about the mouse example that might be problematic is that 00:29:33.770 --> 00:29:38.510 for the primate brain, you were saying, look, these laws we have on connectivity, 00:29:39.370 --> 00:29:43.730 tells you something about the optimal wiring of a brain because you see that 00:29:43.730 --> 00:29:46.570 regions that are connected are placed together and so on, right? 00:29:46.650 --> 00:29:50.490 And you make this distinction, or the example of the ventral dorsal visual stream. 00:29:51.670 --> 00:29:54.530 However, now for the mouse case, you say that you find 00:29:54.530 --> 00:29:57.370 regions there that are adjacent but not connected not adjacent 00:29:57.370 --> 00:30:00.610 but nearby yes no but okay still right so 00:30:00.610 --> 00:30:05.150 they're nearby but not connected so that seems to be a violation of the principle 00:30:05.150 --> 00:30:08.750 that you identify so that means a mouse brain in its development is really setting 00:30:08.750 --> 00:30:12.870 up fixed borders which i know we're not going to wire these guys up yes while 00:30:12.870 --> 00:30:18.230 you apparently are not doing that in the primate brain is that correct exactly yes that that's what, 00:30:18.690 --> 00:30:22.390 I found so very, very surprising. I didn't expect that at all. Yeah. 00:30:22.630 --> 00:30:26.170 And if you, if you optimize your, your mouse brain. 00:30:27.283 --> 00:30:32.683 So you place the areas optimally so you don't have these absent connections, as it were. 00:30:33.023 --> 00:30:35.803 And then you look at your lambda value. It looks much more like a monkey. 00:30:36.443 --> 00:30:40.883 So you can convert it into a monkey. Now, how we got to that state, I don't know. 00:30:40.963 --> 00:30:49.383 But I'm wondering if the ancestral primate was actually, I think, probably quite large. 00:30:50.003 --> 00:30:53.863 And so today's rodents might have undergone a miniaturization. 00:30:53.863 --> 00:30:57.123 And is that the cost that you're paying for miniaturization, 00:30:57.283 --> 00:31:02.063 or is it another kind of adaptation that we're not putting our finger on? 00:31:02.763 --> 00:31:06.523 So about optimality, what I would like to understand is how you define that. 00:31:06.663 --> 00:31:11.423 Because if you talk about optimally wiring a brain, what does that really mean? 00:31:11.563 --> 00:31:16.663 Does it mean that you have a constant number of wires crossing certain distances 00:31:16.663 --> 00:31:19.103 independent of where you are on this cortical sheet? 00:31:19.103 --> 00:31:23.543 Does it mean that you want to optimally transfer signals between areas that 00:31:23.543 --> 00:31:27.943 you want to use as a minimum number of intermediate steps to get from A to B? 00:31:28.023 --> 00:31:29.183 What's optimality here? 00:31:30.183 --> 00:31:35.183 Well, I've been using the term maybe rather loosely. What we're talking about is optimal placement. 00:31:35.383 --> 00:31:40.563 Can you replace the areas in a monkey brain in such a way that you would be 00:31:40.563 --> 00:31:44.643 using less wire given the strength values that we observe? 00:31:44.863 --> 00:31:47.023 And the answer to that is no, you can't. 00:31:48.023 --> 00:31:50.863 There's no way you can do that these 00:31:50.863 --> 00:31:53.623 are simulations that take days and days and days 00:31:53.623 --> 00:31:57.563 to do but there's there's no way of doing that either to the binary or 00:31:57.563 --> 00:32:01.723 that maybe there's one or two areas you can flip positions but the weight is 00:32:01.723 --> 00:32:07.663 absolutely not in the mouse you can get 15 percent reduction in wire both for 00:32:07.663 --> 00:32:11.783 the binary and for the and for the weighted network so we're talking about optimal 00:32:11.783 --> 00:32:17.423 placements the the second part of the your question touched on this thing that I was talking about, 00:32:17.523 --> 00:32:20.743 that if you take this exponential distance rule to heart. 00:32:21.756 --> 00:32:25.936 And you look at the visual cortex, for example, and you look at the central 00:32:25.936 --> 00:32:30.296 visual one and peripheral area visual one, you'll notice that they're in very 00:32:30.296 --> 00:32:34.196 different neighborhoods in terms of the areas which are surrounding them. 00:32:34.736 --> 00:32:40.196 That would suggest that the central area V1 should be more strongly connected 00:32:40.196 --> 00:32:44.976 with a very different population of areas than the peripheral V1 and the same for V2. 00:32:45.176 --> 00:32:49.716 And we've looked at that. And the answer is, yes, the connectivity is very different. 00:32:49.716 --> 00:32:57.016 So central V1 is in front of the temporal areas, so you find that central V1 00:32:57.016 --> 00:33:00.576 and central V2 look like ventral stream areas. 00:33:00.876 --> 00:33:05.336 And if you look at the peripheral representation, they're in front of the parietal 00:33:05.336 --> 00:33:08.136 cortex, and you look at the strength of the connection and you do a summation 00:33:08.136 --> 00:33:12.356 of strength, then these areas, or these parts of these areas, 00:33:12.456 --> 00:33:14.936 appear to be dorsal stream areas and quite different. 00:33:14.936 --> 00:33:21.776 So from that, I'm wondering if we have an optimization of the position of areas, 00:33:21.916 --> 00:33:24.936 but we also have an optimization of the shape of areas. 00:33:25.276 --> 00:33:29.336 And if you look at the flat maps of David Van Essen and others, 00:33:29.556 --> 00:33:31.516 they have very, very distinctive shapes. 00:33:31.976 --> 00:33:34.736 I mean, you remember the motor cortex and the somatosensory cortex, 00:33:34.956 --> 00:33:40.576 these long, thin strips of cortex where you put the whole homologous in this peculiar sort of strip. 00:33:40.696 --> 00:33:44.736 You could imagine something very different. Would you accept the hypothesis 00:33:44.736 --> 00:33:47.696 that what you're optimizing is a transduction delay between these areas? 00:33:49.256 --> 00:33:55.776 Well, yeah. I think that was part of the motivation that we had to look at the 00:33:55.776 --> 00:33:57.356 distance through the white matter. 00:33:57.636 --> 00:34:02.036 So we felt that we were looking at the… And I think in the first instance, 00:34:02.276 --> 00:34:05.896 we measured the white matter distances and the surface distance. 00:34:06.416 --> 00:34:11.416 And you could imagine that these distance relationships we've been reporting, 00:34:12.156 --> 00:34:19.636 could simply be reflecting change of the cortical property, in which case the 00:34:19.636 --> 00:34:21.196 surface distance would be very good. 00:34:21.996 --> 00:34:26.416 Or it's something to do with transduction signals, in which case your trajectory 00:34:26.416 --> 00:34:28.196 through the white matter should be very good. 00:34:28.676 --> 00:34:32.636 And so I was expecting to have a sort of black and white answer comparing those 00:34:32.636 --> 00:34:35.756 two. And that didn't happen. What did happen? 00:34:37.376 --> 00:34:43.696 Well, not much actually. We stuck to the white matter distances because that's 00:34:43.696 --> 00:34:44.916 what we'd started off with. 00:34:45.256 --> 00:34:50.236 But then when we were looking, when we were making this monkey-mouse comparison, 00:34:50.776 --> 00:34:54.356 that's where we asked ourselves the question, what about if we unfold the monkey 00:34:54.356 --> 00:34:55.896 cortex and how will that behave? 00:34:56.376 --> 00:35:01.176 And we're still looking at that. But it certainly changes the distribution of distances. 00:35:02.536 --> 00:35:08.756 When you say that you want to do this minimal wiring test and see if you can 00:35:08.756 --> 00:35:09.856 rearrange cortical areas, 00:35:10.596 --> 00:35:15.236 to reduce the wiring, how do you deal with the fact that the pieces of the jigsaw 00:35:15.236 --> 00:35:18.216 are all very different shapes and there's really only one way it fits together? 00:35:18.456 --> 00:35:21.216 So if you rearrange it, how do you compensate for, 00:35:22.197 --> 00:35:26.917 Oh, well, so what we're rearranging, we're not tackling shape here, Tony. 00:35:27.757 --> 00:35:32.457 Yeah, and I've perceived you're not. Yeah, so we're measuring distances between 00:35:32.457 --> 00:35:35.277 the barycenters of the areas. 00:35:35.497 --> 00:35:39.357 And so we're switching barycenters around in space. 00:35:39.617 --> 00:35:43.637 We're not trying to shift these peculiar shapes and make them all fit in. 00:35:43.717 --> 00:35:44.977 It's not a jigsaw puzzle thing. 00:35:45.397 --> 00:35:51.937 So we're moving the barycenters. and we're considering the distance from one 00:35:51.937 --> 00:35:55.297 barycenter to another is the distance between one cortical area and another. 00:35:55.597 --> 00:35:59.717 But as you say, the shape of these areas can be quite extreme to long thin strips. 00:36:00.097 --> 00:36:02.177 Yeah. Which then does have an impact. 00:36:03.457 --> 00:36:09.557 Yeah, that introduces its own problem about what is a barycenter of a long oblong shape. 00:36:09.857 --> 00:36:17.877 Yeah, but that's where we are. So then if you had to choose one objective that 00:36:17.877 --> 00:36:24.557 these wires are optimizing, it would be just to minimize the physical wires between areas. 00:36:24.657 --> 00:36:29.577 That would be your bet today. day um i 00:36:29.577 --> 00:36:33.017 as i say i'm i'm not sure if you look at if 00:36:33.017 --> 00:36:38.617 you look at area v1 and v2 in the monkey in the macaque it's the the two biggest 00:36:38.617 --> 00:36:45.217 areas uh in in the macaque brain in fact they're very very large if you relatively 00:36:45.217 --> 00:36:50.637 speaking compared to any brain they're huge and in the central representation representation, 00:36:50.817 --> 00:36:56.497 they're folded in such a way that V1 and V2 lie opposite each other, 00:36:56.617 --> 00:36:58.837 separated by one millimeter of white matter. 00:36:59.377 --> 00:37:03.557 So it's the most extraordinary engineering to make sure that you've really minimized 00:37:03.557 --> 00:37:07.337 the distance between the two biggest areas of the macaque brain. 00:37:08.357 --> 00:37:13.477 That tends to make one feel, either it's because, well, you know, 00:37:13.477 --> 00:37:19.637 if you had as much white matter in your brain as a mouse, it would be the size of a bathtub. 00:37:20.317 --> 00:37:22.697 Which would make getting out of the door rather awkward. 00:37:23.417 --> 00:37:27.597 We know that there's a reduction in white matter, there's a reduction in total 00:37:27.597 --> 00:37:29.457 connectivity as brains get bigger. 00:37:31.477 --> 00:37:36.897 It could be that that piece of engineering is dealing with that problem. 00:37:37.877 --> 00:37:42.497 Alternatively, it's dealing with the transduction problem. You really need that 00:37:42.497 --> 00:37:46.217 kind of very, very short distance for V1 and V2 to do its job. 00:37:46.217 --> 00:37:50.997 And so, in fact, that could maybe explain some of these scaling distances is 00:37:50.997 --> 00:37:54.657 that the amount of wire you have in the brain really becomes. 00:37:56.064 --> 00:37:59.444 Under pressure as you get bigger brains so you really 00:37:59.444 --> 00:38:02.444 have to optimize for wire length in a way that in a small brain 00:38:02.444 --> 00:38:06.664 it's not so important absolutely yes but now another element of this is that 00:38:06.664 --> 00:38:12.224 we could argue well these you measure this in adult in adult sub monkeys right 00:38:12.224 --> 00:38:19.464 so this is a cortex that has been learning and changing its properties due to plasticity rules, 00:38:20.244 --> 00:38:23.424 plasticity rules are activity dependent so what you're looking at is really 00:38:23.424 --> 00:38:25.284 the history of activity in this system. 00:38:25.504 --> 00:38:30.604 Now, the history of its activity is strongly constrained by subcortical systems. 00:38:30.964 --> 00:38:33.944 Like in the case of cortex, you'll depend on your thalamus. And now, 00:38:34.064 --> 00:38:36.264 thalamic projections are not random. 00:38:36.804 --> 00:38:42.104 Thalamic projections actually also define very specific envelopes of interaction 00:38:42.104 --> 00:38:42.984 with cortex. Absolutely. 00:38:43.144 --> 00:38:47.644 So I could argue what you're actually measuring is an echo of the combination 00:38:47.644 --> 00:38:53.644 of the envelope of thalamic projections into this system modulated by the local 00:38:53.644 --> 00:38:58.464 plasticity rules that make these guys wire up together. Would you accept that interpretation? 00:38:59.024 --> 00:39:02.704 I'd go a long further, much further than that. The work we've been doing with 00:39:02.704 --> 00:39:08.124 Colette de Hay over the last 10 years shows that the thalamic fibers release 00:39:08.124 --> 00:39:12.464 a mitogenic factor which governs the proliferation in the germinal zones. 00:39:12.844 --> 00:39:17.984 So the size of your areas, the size of the cortex is largely determined by the 00:39:17.984 --> 00:39:20.904 interaction between the thalamic fibers and these germinal zones. 00:39:21.524 --> 00:39:28.164 That whole relationship between the thalamus and the cortex setting up, 00:39:28.224 --> 00:39:32.324 for many years people thought that the thalamus was playing an important role 00:39:32.324 --> 00:39:35.184 in the specification post-mitotic. 00:39:35.444 --> 00:39:38.364 You had the barrel cortex and the effect of 00:39:38.364 --> 00:39:41.584 plucking whiskers and seeing the representation the 00:39:41.584 --> 00:39:44.384 barrels changing in some fashion on 00:39:44.384 --> 00:39:47.444 in on the cortical surface so what we've been arguing 00:39:47.444 --> 00:39:50.684 for for a long time now is that the the thalamic 00:39:50.684 --> 00:39:55.664 fibers are actually their primary target and particularly in primates is it's 00:39:55.664 --> 00:40:02.004 much more accentuated is not the cortical plate it's the germinal zone and and 00:40:02.004 --> 00:40:06.684 um so that the thalamic fibers get into the cortex in the primate very very 00:40:06.684 --> 00:40:08.664 early, before there is any cortical plate. 00:40:08.844 --> 00:40:12.924 In fact, before there is any release of mitototic neurons into the cortex. 00:40:13.264 --> 00:40:19.684 And what they're doing is actually, I think, hugely to do with shaping the proliferation 00:40:19.684 --> 00:40:21.364 and possibly the specification. 00:40:21.524 --> 00:40:25.304 But would it also mean that if we look at these two types of thalamic projections, 00:40:25.664 --> 00:40:30.884 like powerful magnocellular, where the powerful cellular seems to be not very 00:40:30.884 --> 00:40:35.524 plastic and the magnocellular is, shouldn't there then be a correlation between 00:40:35.524 --> 00:40:38.564 parvocellular projections and your scaling law? 00:40:41.300 --> 00:40:44.300 Um well for me the scaling law 00:40:44.300 --> 00:40:47.160 is is uh a little different 00:40:47.160 --> 00:40:51.100 from that it's it's to do with how you how 00:40:51.100 --> 00:40:54.560 white matter gray matter changes over a 00:40:54.560 --> 00:40:59.980 range of brain sizes and so what people have been able to show is as brains 00:40:59.980 --> 00:41:04.580 get bigger the volume of white matter doesn't increase at the same rate as the 00:41:04.580 --> 00:41:09.040 volume of gray matter and so this is where this this this notion that was It 00:41:09.040 --> 00:41:12.940 was introduced by Ringo some 20 years ago that as brains get bigger, 00:41:13.040 --> 00:41:16.240 there's a huge pressure to economize numbers of connections, 00:41:16.500 --> 00:41:20.600 which if you think about it, is really rather extraordinary because brains are 00:41:20.600 --> 00:41:21.820 all to do with connections. 00:41:22.100 --> 00:41:25.440 And you've got to actually reduce the number of connections and reduce the number 00:41:25.440 --> 00:41:26.360 of long-distance connections. 00:41:29.520 --> 00:41:33.340 But the question is, how do you do that? How do you control the developmental 00:41:33.340 --> 00:41:38.760 program to achieve this, to economize on these connections? So then the question 00:41:38.760 --> 00:41:43.420 is, is the thalamus then the key to understand the genesis of your scaling law? 00:41:44.260 --> 00:41:47.500 Well, I don't know. I think if I understand Herculeanus' work, 00:41:47.700 --> 00:41:50.540 Herculeanus' work correctly, 00:41:50.860 --> 00:41:57.340 as your rodent brains get bigger, the size of the neurons get larger, 00:41:57.480 --> 00:42:01.160 so you have this change in density which you don't have in primates. 00:42:01.220 --> 00:42:07.100 So that seems to me to be a very different sort of algorithm for building the 00:42:07.100 --> 00:42:10.220 brain. If you say, well, okay, we're going to have primates, 00:42:10.220 --> 00:42:14.200 and primates are going to have basically a small variation of cell size. 00:42:14.940 --> 00:42:19.200 We want to go from a small brain to a big brain, so we're going to economize on connections. 00:42:19.640 --> 00:42:22.200 And then you do the same thing for rodents. You say, well, we're going to have 00:42:22.200 --> 00:42:23.780 to do the same economy on connections. 00:42:24.620 --> 00:42:30.040 That's a kind of given. But here we can change the neuron size a bit more. 00:42:31.260 --> 00:42:35.240 So you seem to run into a bit of a problem with creating very big brains. 00:42:35.380 --> 00:42:41.760 You have these South American capybaras. You have these big South American rodents. 00:42:42.700 --> 00:42:46.820 But then I'm wondering if the adaptation is much more adapted to making small brains. 00:42:46.900 --> 00:42:50.980 So that is the sort of rules I see for the scaling. 00:42:51.240 --> 00:42:55.220 I haven't been thinking about how the… Because what I'm after, 00:42:55.300 --> 00:43:00.060 what I try to figure out is… Now what we're trying to do this week in the school 00:43:00.060 --> 00:43:01.980 is this relationship between genetics and development. 00:43:02.580 --> 00:43:05.900 And as we discussed earlier what you measure with the scaling law is like an 00:43:05.900 --> 00:43:08.220 echo of these two processes working together. Right. 00:43:09.082 --> 00:43:13.282 So then I was trying to push you a bit and try to come to some understanding, 00:43:13.402 --> 00:43:18.962 okay, what are the causal factors that give rise to these scaling laws that you measured? 00:43:19.222 --> 00:43:25.002 Right. Okay. Well, we're going to look at a particular case which would probably come back to that. 00:43:25.082 --> 00:43:29.162 You have this situation of microcephaly where you have very, very small brains. 00:43:29.362 --> 00:43:34.242 And there's a lot of interest in understanding that because expansion of the 00:43:34.242 --> 00:43:39.662 brain is very much a hallmark of human evolution. And so microcephaly raises 00:43:39.662 --> 00:43:43.482 something of a question about our evolutionary origins. 00:43:44.222 --> 00:43:50.762 People with that condition actually show remarkable levels of cognitive capacity. 00:43:51.262 --> 00:43:55.222 And so we're now going to start working with a mouse which is a model for this, 00:43:55.382 --> 00:44:00.962 and so it will explore what is the genetics regulating brain size and how that 00:44:00.962 --> 00:44:03.622 will touch on this exponential distance rule, 00:44:03.702 --> 00:44:10.522 and also on the the whole relationship between brain size and the explanation of distance rules. 00:44:10.682 --> 00:44:16.002 So these are questions which we will address in mouse. Okay. 00:44:17.302 --> 00:44:21.962 Tony, you have any more? Okay, so Henry, now looking at this, 00:44:22.102 --> 00:44:26.502 your tour through, let's say, the anatomy of the brain that's going on for quite 00:44:26.502 --> 00:44:29.982 a while, and which also you have made amazing progress, 00:44:31.142 --> 00:44:35.742 if we want to follow in that trajectory, what's Henry's law that that we should 00:44:35.742 --> 00:44:43.442 adhere to um i think there's a number of cases without saying any names, 00:44:44.122 --> 00:44:46.902 where people you do you remember the decade of 00:44:46.902 --> 00:44:49.802 the brain very well so the decade 00:44:49.802 --> 00:44:56.062 of the brain was spurred by a book um i won't say the name of the book which 00:44:56.062 --> 00:45:02.522 said well we've got people endlessly producing data and you go to the sfn meeting 00:45:02.522 --> 00:45:06.802 and it's so depressing because you've got We've got miles and miles of posters 00:45:06.802 --> 00:45:08.762 of people showing endless data. 00:45:09.002 --> 00:45:14.142 What we now need is somebody to come along and give us a model of this, 00:45:14.742 --> 00:45:19.862 much in the way that Watson and Crick were able to do with the double helix. 00:45:20.082 --> 00:45:24.842 And we'll have a kind of Eureka moment, and we'll suddenly understand everything, 00:45:25.062 --> 00:45:28.502 and we will have understood the brain. Basically, we would have done it, gone there. 00:45:28.882 --> 00:45:31.702 It would be a finished story, and we can pass on to something else. 00:45:31.702 --> 00:45:35.342 I find this absolutely ludicrous because um. 00:45:36.863 --> 00:45:40.923 Based on an idea and there's a there's a present a european project to understand 00:45:40.923 --> 00:45:46.003 the brain and it has this idea built into it that we've got a lot of data and 00:45:46.003 --> 00:45:47.283 we can go back and accumulate, 00:45:47.983 --> 00:45:53.383 over 150 years of journal comparative neurology and and skim through and and 00:45:53.383 --> 00:45:57.783 and take out all this data and pile it up together and it will add up to some 00:45:57.783 --> 00:46:02.023 total explanation and i think this is dangerous. 00:46:02.183 --> 00:46:07.443 I think it underestimates the challenge to understand intelligence, 00:46:07.643 --> 00:46:08.603 biological intelligence. 00:46:09.143 --> 00:46:14.263 I think it underestimates the challenge of relating that to neurological principles. 00:46:14.703 --> 00:46:22.303 It forgets the fact that any experimental result is done in a certain intellectual 00:46:22.303 --> 00:46:27.383 framework, and the interpretation of those results is not extendable. 00:46:27.463 --> 00:46:30.503 You can't extract these this information willy-nilly and 00:46:30.503 --> 00:46:33.623 apply it across the board so i think 00:46:33.623 --> 00:46:37.903 that either governments are going to decide that understanding how brains work 00:46:37.903 --> 00:46:43.063 is worthwhile and they will provide money to do basic research which means not 00:46:43.063 --> 00:46:46.863 endless science papers and nature papers and what have you but actually pay 00:46:46.863 --> 00:46:51.763 people to how many synapses do you have on your pyramidal cell what What is the, 00:46:51.763 --> 00:46:59.063 you know, the connectivity of your average whatever and pay people to do that kind of work. 00:46:59.143 --> 00:47:02.323 And so I think that there's a need for empirical data. 00:47:03.303 --> 00:47:07.963 I think that the power of simulation is fantastic and has to go along hand in hand. 00:47:08.083 --> 00:47:12.363 But if you don't actually have the investigations, if everything's got to be 00:47:12.363 --> 00:47:14.803 a breakthrough now and then, you know, immediately, 00:47:15.043 --> 00:47:19.783 if you're only going to have high profile kind of projects, then you're going 00:47:19.783 --> 00:47:23.463 to find that you're actually playing around with inadequate data. 00:47:23.623 --> 00:47:28.243 And I think that this small world thing, it's not that there isn't a small world complex network. 00:47:28.243 --> 00:47:30.963 Work there is what i want the point i want to 00:47:30.963 --> 00:47:33.763 make is it doesn't exist at the aerial level and there's 00:47:33.763 --> 00:47:37.103 been over i don't know how many dozens of publications claiming 00:47:37.103 --> 00:47:41.183 that that is the case and they've been using inappropriate data somebody's got 00:47:41.183 --> 00:47:47.243 custard on their face and i think that the um the willingness to challenge big 00:47:47.243 --> 00:47:51.523 questions i think there needs to be much more interaction between people doing 00:47:51.523 --> 00:47:55.503 um the experimental work and the people doing the simulation and i I think these 00:47:55.503 --> 00:47:56.663 have got to be hand in hand. 00:47:57.143 --> 00:48:03.383 And this is exactly what we've been trying to do with Zoltan Torekai and now 00:48:03.383 --> 00:48:05.843 with Chao Jingwang and others. 00:48:06.043 --> 00:48:10.523 And I think it's tremendously exciting because you can see your anatomy in a 00:48:10.523 --> 00:48:12.903 much larger framework, in a much bigger context. 00:48:13.243 --> 00:48:16.663 And I think this will ultimately lead to real breakthroughs. 00:48:16.663 --> 00:48:24.123 I think sort of collapsing things and overselling and saying, 00:48:24.223 --> 00:48:29.043 well, we're going to have a decayed with the brain and okay, we need big science. 00:48:29.203 --> 00:48:35.503 That's for certain. But we also need reason science and we need empirical data. 00:48:36.663 --> 00:48:39.763 And now, four years from now, Tony's going to visit you in Lyon. 00:48:39.903 --> 00:48:44.043 Yes. Have some pâté with you. No, no, andouille. Okay, even better. 00:48:44.843 --> 00:48:48.343 But he's also going to confront you with a prediction you're going to make today. 00:48:48.503 --> 00:48:53.503 So what's the one specific prediction you would share with us today that Tony's 00:48:53.503 --> 00:48:54.883 going to check out four years from now. 00:48:58.063 --> 00:49:01.363 Well, I think that we're going to find... 00:49:02.655 --> 00:49:09.835 That there's a very large range of solutions that biology has brought to the brain. 00:49:10.695 --> 00:49:16.655 The dream that you can have one brain and extrapolate from that and understand 00:49:16.655 --> 00:49:22.795 the brain principle goes right against all our understanding of zoology and how it works. 00:49:23.495 --> 00:49:30.095 So I think that when Tony comes to Lyon, we have a pot of white wine and an andouillette. 00:49:30.095 --> 00:49:34.335 – andriets, you have to have very, very white wine, very dry white wine, 00:49:34.415 --> 00:49:40.635 and in very large quantities – is that we'll know something about different 00:49:40.635 --> 00:49:43.115 sizes of brains and how folding, 00:49:43.715 --> 00:49:45.175 interacts with these things. 00:49:45.375 --> 00:49:49.755 And I think that the idea that you can use the mouse brain as a model brain 00:49:49.755 --> 00:49:54.235 for all brains will be seen to be completely fallacious. 00:49:54.515 --> 00:49:58.315 I hope between now and then somebody else will have done the local circuitry, 00:49:58.315 --> 00:50:00.315 because because this is where the machine really is. 00:50:00.395 --> 00:50:04.015 This is where the machine lies across different brain regions. 00:50:04.215 --> 00:50:07.175 At the moment, we have no idea about the local circuitry of area 46. 00:50:07.955 --> 00:50:13.535 We know the local circuitry of V1 in the cat, and we're extrapolating that across 00:50:13.535 --> 00:50:15.755 all species, all brains, and what have you. 00:50:15.835 --> 00:50:21.655 So I think that we'll have a much more deeper understanding of the variability 00:50:21.655 --> 00:50:24.855 of brains and the solutions they bring. 00:50:25.235 --> 00:50:27.935 Great. Henry Kennedy, thank you very much for this conversation. 00:50:31.095 --> 00:50:35.495 Thank you very much. I hope it was not too boring for you. The CSN podcast was 00:50:35.495 --> 00:50:40.155 produced by the Convergent Science Network of Biometrics and Biohybrid Systems, 00:50:40.615 --> 00:50:45.495 a project funded by the European 7th Research Framework Program. 00:50:47.215 --> 00:50:52.375 For more interviews, recorded lectures, or upcoming conferences in the field 00:50:52.375 --> 00:50:58.615 of biometrics and biohybrid systems, go to csnnetwork.eu. 00:50:58.640 --> 00:51:06.640 Music. 00:50:58.975 --> 00:51:00.795 And thank you for listening.