 I come from Lyon. This is Lyon. This is a view from my kitchen window. So, yes, in the introduction you mentioned our work on tracing connections in the macaque. We've been doing this for a number of years using techniques which were put together in the 70s. So, talking about modernity, the problem that we confronted was that it was claimed that we had a very good understanding of the connectivity of the macaque brain because numerous groups across the world have been doing this for many years and publishing their results in excellent journals. But they had been looking at the connectivity of areas which they believed to be interacting together in a specific fashion. So, auditory areas were connecting to auditory areas, visual areas to visual areas, and there was a very poor understanding of the global connectivity. Our approach was very different because we discovered auditory projections into the visual cortex, into the primary visual cortex, back in 2005 or something like that. And this meant that the cardinal rule of what we believed to be a primary area was actually not holding because a primary area was considered to be an area which did not receive input from other primary areas. Of course, that's absolutely not true in rodents, as we all know. But it was thought to be true in the macaque. So we re-approached the problem and discovered that something like over a third of the connections which actually exist have been ignored. That's a conservative estimate, actually. Any given study was underestimating by a lot more. And a lot of the work that had been done had been done on a collated database looking at connections over many different publications. And the problem was that people were using different sorts of traces. They were using different sorts of definitions of what an area is. And so they were grossly underestimating the density, that is to say the percentage of connections that exist of the possible connections. It turns out that the macaque cortex has a density of 70%. Over two thirds of the possible connections actually exist. So most areas connect to most other areas. So if you want to understand the connectivity in terms of functional terms, if you want to cover an insight into function, actually knowing what area is connected to what area is trivial, what you need to know is the strength of that connection. And of course a lot more about the integration of the connection into the complexity of the area. So I think that these results in themselves lead to a reapprisal of the organizational principles of the cortex because the fashionable idea had been for many years that the cortex, the aerial interactions of cortical areas were organized according to a small world organization. In other words, you had an economy of connections, you had very few connections leading to a high degree of integration. With a high density, that becomes a trivial observation because you have no economy of connections, but you have a high degree of connectivity. So the title I gave to my talk here is specifics of the primate connectome because more recently we began to look at the connectome of the mouse. When we started this, the Anand Brain Institute had not yet published their publication in Nature in 2014. And so we used similar types of techniques in conjunction and collaboration with Andres Burkhauter and we're able to make a comparison between mouse and macaque cortex. What this shows you is that the mouse cortex has a density of 100%. All cortical areas project to all cortical areas. And the Anand Brain Institute's publication in 2014 made a number of mistakes. They were using computational methods to infer the connectivity, using very large injections which involve multiple areas. And their claim was that this was leading to a density of about 30%. Well, we published this, this is published, I don't want to go into the details of that here today because it would require a lot of neuroanatomy and I don't think that would be appropriate for this venue. But the point I'd like to make here is that if you're working on cortical processing in 2020, you're not going to work on the macaque monkey. It's not an ideal model by a long shot. You can't use genetic dissection of the processes that you're interested in. You can't use optogenetics in an easy fashion. So you're going to work on the mouse. In that case, the presumption is going to be that the mouse is an adequate model for what society wants to understand. Society is actually not very interested in the mouse cortex as such. The human society is very interested in the problems that confront a species which relies increasingly on its brain. Given that the brain falls apart and disintegrates with age. And so the challenge is to understand the basics of the primate cortex. Now, if you look globally at the neurobiology of the cortex, and I'm interested in the development of cortex for example, as well as its connectivity, the differences between mouse, rodent in general and primate are enormous. And my claim is that it's not possible to make a smooth transition in our understanding based on investigation uniquely of the mouse cortex. So the situation 20 years ago would have been that if you were interested in cortical processing, you worked on the macaque because it was, in those days, an ideal model. You had behaving macaque monkeys. You could record from single neurons while monkeys were doing complex cognitive tasks. And you could record the activity of individual neurons and correlate it to the task in question. That has now been, that is now no longer the case. And I think that the urgency now is to bring our understanding that we can gain from the reductionist model in the rodent and bring to bear that onto questions which need to be posed and to develop the techniques which will make it possible to investigate those questions in the macaque cortex. So that's the major frust of what I'm going to be talking about. I think the other problem that confronts a student of neurobiology of cortex is your, you read in the textbook that the cortex is responsible for higher cognitive functions and it sounds extremely noble. And then you have this increasing pace of artificial intelligence. And so what your iPhone can do, what your mobile telephone can do is enormous. And you get confronted by the fact that the brain maybe is not as special as you might presume. And the frontier between artificial intelligence, if you look at television programs about real humans, for example, you have this idea that actually robotics and artificial humans are going to be perhaps more real than real humans. It's not trivial. It's actually not trivial. You know, you have things like Blade Runner but it's completely dated because artificial intelligence is so compelling. It's so compelling in many ways. So this is a photograph which I got from Rodney Douglas and it shows you a main frame IBM computer of some years ago. And if I had a pointer, the trick is to say, well, what's really interesting in this room here? Does this work? Yes, it does. This produces an enormous amount of heat. This whole IBM machine is producing a lot of heat because it's doing the same thing over and over and over again. That's how it works. This man who's cleaning the machine has a brain that works on a couple of bananas and it produces practically no heat. It's a completely different phenomena. There's no comparison. It's worthwhile remembering that. And the mystery about how this works is, I believe, total. We don't really know how it works and that's what we're challenging. This is what these are the questions that we want to answer. These are the questions that have to be answered if we're going to bring society the solutions. So let's not get confused about artificial intelligence. It's useful, it's important, it's a tool. But it's not an explanation of what the human brain is doing by a long shot. And we have to be very firm about that because society who's funding us isn't. So we have two dilemmas here. The ideal model is the rodent brain. You can do many, many things in that and it's going to be an important frontier for how biological systems actually can achieve extraordinary performance. But I believe it's going to tell us very little about the mechanism's underlining human consciousness. It requires a transition via a model which resembles, to some extent, the human brain. I say to some extent because let's not forget that the McCack model is never going to be an absolute model of the human brain. This is making reference to the work of Frank Puller who with his student Charié, Cicille Charié, back in 2012 was working at RSA Gap II in Abition. So this is a gene which undergoes a specific inhibition in humans and it changes the nature of the spines in the human brain. And it doesn't exist in chimpanzees. And it actually renders those spines neotinic. So the spines in the adult human have characteristics which are usually found in immature animals. This is one example, this is not saying that the human brain is at the top of some extraordinary pyramid. It's just saying that, hey, hang on, you get very, very specific gene expressions in different species which might make it very difficult to transit from one to the other. So there's going to be a triangulation. If we're going to understand the human brain, we're not going to have a perfect model that we can simply transcribe into understanding the human brain. We're going to have to understand different species, different comparisons. So the work I'm going to be talking about today is very much to do with the connectivity of the brain and I will talk about the consequences of this very high density of the inter aerial connectivity. And I will argue that this has important consequences, for example, on the existence of small worlds. This has been published. But more recently, there's been a lot of interest in the notion that you have hubs in the brain. You have areas in the brain which have very high in and out degrees which are more interconnected to other cortical areas. The claim is that these so-called hubs will play an important role in routing information through the brain. This gives a vision of information being channeled from one place to another via different routes and the hubs would play an important role in that. Now there's another layer to this because if the hubs are more interconnected than statistically probable, you have a phenomenon called a rich club. And the rich club is by reference to human society, a lot of these network investigations are based on social networks. The rich club in a social network is a group of individuals that have a high degree of control across the society because they're involved in very, very different entities. So the hubs which are highly interconnected will give a rich club and that will exert a manifest control over the overall system. I find this extraordinary. The idea that information is routed through the cortex, through the brain, for anything, I think completely ignores what we understand about how nervous systems work. There aren't any messages going from one place to another. That does not exist. This analog, this metaphor for the brain, like the worldwide web, I've heard people stand up and say, well, you know, the brain's like a worldwide web. There you do have a piece of packets of information that go from one place to another and indeed you have hubs which will route this information from one place to another. But brains don't work like that. Nervous systems don't work like that. Axons, cells and these cells integrate the activity coming along the axon and they fire. And so the horizon for any cell is the cell that came into it and the cell it's projecting to. You integrate across maybe 5,000 inputs coming into a large pyramidal cell. And what is then signaled by that cell is a completely different entity than what came in on any of those synapses. So I don't believe that routing of information is going to be a useful concept. And if routing information is not a useful concept, I don't believe it makes much sense to think about hubs and rich clubs. But let's come back to the density of the cortex. I told you that the density of the mechat cortex using track tracing, and I'm going to claim that that is the ground truth, is 70%. So they can't be a very large range of inputs and outputs. No cortical area is going to be hugely more connected than any other. And in fact, that's what we show. There is, we have some information, we have some data and I will just refer to it very briefly concerning the claustrum which suggests that the claustrum may be fulfilling a role that people have been trying to ascribe to hubs because the claustrum projects to every single cortical area that we've looked at. And the input to every cortical area is the strongest sub-cortical input that you can find. I say sub-cortical with some embarrassment because actually the claustrum is not sub-cortical. It's a layer of cells which is wedged underneath the insular cortex, which is trapped in the white matter and is generated by a germinal zone which actually generates the insular cortex. So the best way to think about the claustrum is the lost layer of the insular cortex. So it's actually not sub-cortical, it's cortical. And you can best think of it as a satellite. So what's extraordinary about the claustrum is it projects equally to every single cortical area of the brain. And people when they talk about the claustrum often refer back to the publication of Koch, Christoph Koch and Francis Crick where they suggested it had an important role in consciousness. So this is the sort of image that people like to show when they want to refer to modernity to show that modern techniques are allowing us to make enormous progress in understanding the structure of the brain. It's a tractography based on diffusion MRI of the human brain. And it's been believed for many years that this will enable you to understand, to put into evidence the full connectivity of the human brain, of an individual brain. The sort of work I've been talking about looking at the macaque cortex using tract races and I'll come back to a little bit more detail in that uses many, many individuals because you only have three, four, five, maybe six tracers in any case. And so you have to do six or seven injections of tracers and look at the distribution of retrogradedly labeled neurons and you end up with a kind of composite. Now if this was to do what it's claimed to be doing you would have in one instance the full connectivity of any given individual. And this is very much the basis of the human connectome project because you can then compare the connectivity of a genius and an idiot, of a depressed man and a schizophrenic and build up all sorts of understandings of human pathology, aging and what have you based on this technique. The problem is that we have a very poor understanding of exactly what the technique is telling us. So a few years ago we collaborated with a number of teams, Tim Barons in Oxford and David Van Essen in St. Louis. And what we were basically doing is comparing across individuals the connectivity that you can obtain using tract tracing with the connectivity that you can obtain using tractography from the best sort of diffusion that we could get back in those days. This was about five years ago. And the long shot is it's very, very, very poor. It's extremely poor. Basically you can reveal very short distance connections and it's not clear whether the phenomenon that you're recording looks similar to the connectivity whether it has anything to do with transport or whether it has anything to do with the diffusion of water molecules because the water molecules are diffusing in and around the axons. So we're not doing tractography. We are further investigating tractography and it turns out that it's highly variable across cortical areas in an individual brain. So if you compare the tract tracing and the tractography some areas give you quite a reasonable estimation with tractography of what the tract tracing shows you and other areas don't. And I believe that this is very, very much to do with the configuration of the cortex. So in my book, tract tracing is not really very useful, not very convincing, not very powerful. But just in case you don't want to believe me, this is an early result that we find it's no longer in preparation. If you put, this is a summary diagram of what could be a way of thinking about things. If you do an injection of tracer in area V1 in the central representation, generations of anatomists would tell you that in V2 you will expect to find dense connections in the central representation of V2. In fact, tract tracing was basically used for many years to explore the topological relationships between sensory areas. If you do in the same experiment, if you put a tractographic seed in area, in the central representation of V1, you do find dense streamlines coming into the central representation of V2, but you also find dense seedlines coming into the far periphery, upper field and lower field. If you had a result like this with tract tracing and you sent it into the journal of comparative neurology, you wouldn't be able to publish it. That would be the end of it. You wouldn't want to publish it. It wouldn't be considered to be at all useful. So, I'm not quite sure what time did I start? I have 15 minutes to go. Right. Okay, that's gonna be a difficult one. So, basically, I mentioned that back in 2012 we made a series of publications where we looked at the tract tracing using tracers where we inject a tracer. It's retro-gagely transport. It's an active process. It's got nothing to do. It's not a passive diffusion. And we were able to show a large percentage of connections which had previously not been discovered. So, this is an illustration of those findings. This is a log scale. These are the newfound projections, projections which had not been reported in any journal that we could lay our hands on. And these are the so-called known connections. So, the very strongest connections, going over the very shortest distances, were largely known. But very, very quickly, you get a large increase in unknown connections even over very moderate distances over connections which have been well and truly studied. And there's a relationship between distance and strength. So, our claim is that this relationship between strength for connection and distance is actually a key to understanding global organization of the cortex. And basically what we've been able to show, I'm going to jump over the issue of the small world networks not being explained in a high density graph because I think it's relatively trivial and it's well and truly published. And come instead to the distance-weight relationships that we can observe. So, the distance relationships gives you the fact that over short distance connections are very strong and over long distances, they're very much weaker. And that in itself gives you a very obvious explanation of why wire minimization is such a key feature in the organization of the cortex. It also tells you something that you know. We know that the different cortical areas which are heavily integrated like the visual areas, auditory areas, the motor areas are grouped together so that you have a group of areas. And the strength of the connection between those areas is going to be much stronger because the distances are much shorter. When we look at the distance relationships, they're very consistent and they allowed us to develop a model where we were able to build artificial networks using the same characteristics that we observe in the cortical network and ask ourselves, does this generate features in the large-scale networks that we can form in this session which are reminiscent of what we've observed in the macaque cortex? And the answer is that we do. The answer is what we do. And one way of doing this is to look at these free motifs. So you take all groups of free in the global network and you look at the frequency of the difference of the different motifs that you get here. So you go from a motif, for example, the zero motif where you have no connections between set of free nodes to the fully connected triangles. And then the frequency of this is going to give you a characteristic distribution. And the claim has been made and I think there's many ways to understand that that this frequency of the motifs is going to be an important feature in understanding the function of the organizational principles that we're describing. And if you look at this motif here, for example, the motif number eight, where A goes to B and B goes to C and C goes to A, you find abnormally unrepresented in the cortical network. And then we look, we can compare this to the observed frequency and we show that there's a very, very high frequency of the motifs that we've observed in the cortex that you generate in this fashion. So we believe that the strength distance relationship is a cardinal feature and that it's determining many of the other features that we can observe, like the core periphery distribution, which we think could have something to do with the global workspace, which has been highly studied by Stan Hasdahan. So exit small worlds, I didn't go through the details of that, but I think it's easy for you to understand that in a fully connected graph, there is no explanation of clustering as being a generative feature of the distance between networks, between areas. And this is to pay reference to the notion of hubs. And what I want to show you is this, this is not published, we're submitting it this month. This is the different cortical areas with tracer injections in these areas. This is a log range here. So you go from, this is a range of strengths of inputs and the red dots are the strength of the clouser. So you can see from this that the clouser is consistently across all the cortical areas, a very, very, very strong input. And there's no influence of distance. What I've been telling you is that in the cortex, cortical areas projecting to each other, as the distance increases, there's a decrease in the strength of connection. There isn't in the clouser. The clouser does not obey the rule of the decline of strength of connection that I was talking to that we refer to as the exponential distance rule. This means that the clouser enjoys a very specific privilege in that it has this very strong connectivity across the cortex. One consequence of this distance rule is going to be that if you look at the cortex in terms of weights and strengths of connections, you would make a very simple prediction which would be that there should be a spatial organization which will be determined by proximity and that this will in turn be supported by the folding of the brain. So if you look at a flat map of the cortex that David Vanessa has been working on for many years, you can see that if you look at the foveal representation, you have a location of the foveal in terms of other cortical areas, which is quite specific. The foveal finds itself sitting in front of the ventral stream, which is the pathways which lead down into hypercampus and we believe is important for object recognition. If you look at the far periphery representation, the far periphery representation of area V1 and V2 is in front of the parietal stream and so the strength of connections would be expected to be much greater between these parietal dorsal stream areas and the far peripheral representation. And so this illustrates the fact that the exponential distance rule, the weight distance relationship of areas is going to, in addition to determining the motifs distributions, is going to be characteristic, characterizing the spatial layout of the cortex. I'm not really gonna have time to talk about hierarchy, I suppose I've got about five minutes. Seven. The key feature for understanding the role of connections and the way areas integrate is to understand that what we call cortical hierarchy is the integration of long distance connections into the local circuit. The distance, the integration of long distance connections into the local circuit is a key feature. Now, what you need to remember is that the local circuit, the local circuit constitutes something like 80% of connections in the cortex. The remaining 20% mostly come from the adjacent cortical area. So the long distance connections that we favor represent individually one or two percent. And the mystery of how this whole thing works is how those connections are integrated functionally into the local network. This is what we mean when we call cortical hierarchy because there's a basic pattern in that integration and we know this from the work of David, Vanessa and others whereby we see that super granular layer connections are projecting in a feed forward direction and I haven't got the time to go into explaining exactly what I mean by that. And infra granular layer neurons are projecting in a feed back direction and it turns out that this feed forward direction pathway targets preferentially layer four and the feedback pathway avoids integration into layer four. So the origin and the destination of the connections the long distance connections is extremely specific and I believe that that is going to have a huge impact on how the signals are integrated over these long distances. So the whole mystery about what the cortex is is to grapple with the integration of long distance connections into the local circuit. And of course the local circuit is composed of excitatory cells and inhibitory cells with very, very few, very few inhibitory cells projecting over any distance at all. The inhibitory cell is a local cell and of course the canonical circuit says that this, the interactions between the different cortical layers is highly specific. Now for those of you who might believe that my ranting and raving about the tractography was over the top, remember tractography will not ever at least in the way we understand it at the moment give any indication of the origin and termination of the long distance pathways. And that is the information which is key to understanding the integration of those pathways and I believe is of huge importance. So this is another diagram which is highly used. It's used often to say that we know a lot about the brain. I think it's actually, it tells us how much, how little we know about the brain but this is the notion of feed forward goes this way, the projections which go in a feedback come this way, this is area V1 down here, this is area 46 up here and so we know what the feed forward sweep does. It determines the receptive field properties. The mystery is what the back projection does and this is I think where we need to exploit what we can with the mouse and try and see how we can answer those questions in the macaque. I think that understanding the feedback pathway is going to be the challenge. I think that there's, I think the work of Karl Frisven and others is extremely, is very provocative but what I think that they're ignoring is that the feedback pathway is multiple and I think we're going to find that these different feedback pathways are going to play very important roles in understanding the range of activities, the range of roles of the feedback pathway. So I think before I forget I'd like to mention the work of people who've actually contributed to all of this. So a lot of my work at the moment is being done in China where I find that I can get the sort of support that we need to be able to do this track tracing. This is greatly enhanced by my collaboration with Sean here and his collaborators. Mooming Poo who heads the ION Institute in Shanghai where a lot of my work is increasingly taking place is not in this. My work with Zoltan Torakai has been a huge importance. Zoltan has been, has led me to understand many things about graph theory that we're finding to be very important. David Van Essen has played an important role. All our work in the mouse would not have been possible to even begin without Andres Berghouter. And I collaborate very closely with Ken Knoblock who's a great guy to work with and has a lot of insight. And these are students which are heading different projects which I've only very briefly mentioned so I won't go into further detail. I'd be very happy to take questions. Can I extend what the cortex is doing in terms of some kind of information processing? Or you've mentioned, okay, there is just 2% of this long distance connections but yet you are questioning the small world and the rich club which seems to be connected very densely still and have just a few connections outside so those two things are kind of... Well, we published in Science in 2013 an explanation why the small world is inadequate. We have also publications on the rich club so I haven't had time to go to this in detail. I've been able to mention that the small world is based on the idea that you have very sparse connectivity. It comes from understanding of the social network where the density, that is to say the percentage of connections that exist between individuals is 0.00 something. In the brain, the connectivity between causal areas is in the range of 70% end of story. I don't have anything to say beyond what we did in Science. It's published, it's there. Does that mean there is not a small world principle for interconnections between single units? There could well be. You could test that probability using viruses. I would love to do it. I'm not excluding that. It's very probable and I think recent results in the mouse say it's almost certainly the case because there's instances of high clustering which is a key feature which are very, very, very high in mouse. So for the rich, for the small world, I think the story is done. The same argument. If you have a very high density, you're not going to have any hubs. A hub is an area which has a high indegree, has a high degree of connections to other areas. And that means it's higher than some areas and some areas will have very few connections. This is indeed the situation in the worldwide web. It's absolutely not the situation in the cortex. We've published 70% density. We've doubled, in Shanghai, we've been able to double our number of areas that we've investigated and we find exactly the same statistics. So there are no hubs in terms of ranges of inputs and outputs. There's no area, there's no single area with very, very high degrees of connectivity, except the classroom. And interestingly, the classroom is a satellite of the cortex. So there is a massive hub in the brains of mice and men and monkeys. The classroom is highly conserved across species. You're perfectly right. I didn't have time to go into the really interesting thing of what we call hierarchy. Cortical hierarchy is this laminar constraint on the nature of long distance connections. It's been well shown across species that feed-forward connections target the middle layer and the feedback connections avoid the middle layer. I think that the mystery, the intact mystery, is how that integration occurs over the longer connections because the power, the strength of those connections are vanishingly weak. My detractors say, Henry, why don't you just threshold all this? Get rid of the weak connections because, hey, you know, you don't know how they work. Well, I think that that would be a mistake and I think, yes, I'm sorry for being confusing, but I think that the challenge is to come back to these weak connections and to try and imagine experiments and species other than mouse, which will tell us what they're doing. Thank you very much, Henry, for your very revealing talk. I'm here. You talked about mouse and the fact that you can obtain a lot of different types of information from the mouse and recently there have been a lot of reconstructed neurons being released in the mouse-light project and some smaller labs also are doing that. And what they show, these fully reconstructed axons, is that more often than not, long-range projecting neurons connect not one region to another region but to multiple other regions. So I was wondering if there is any way to assess this type of information in macaques. Absolutely. But you're referring to the work of Fogel in London who's been using barcoding to do that. We did these experiments back in the 70s and showed that the high degrees of bifurcation is a characteristic feature of the macaque. Yes, I think it's a hugely interesting idea. The point is that a single area projects with one neuron to three, four, five different areas. This is by a bifurcation. It doesn't mean it's identical information, by the way. Georgie and Achenty and others have shown that. It's not necessarily identical information going across those collaterals. But yes, that's very interesting. We have a lot of data on that, which is published. Using, so we use pairwise injections of different types of traces and look at double labeling. And that enables us to map out these incidents of bifurcation and it's as much a characteristic of the macaque, I believe, as it is as a mouse. And it's beautiful work that they've been doing recently. Yeah, that's an important additional mystery. Thank you, Henry. That was a very interesting presentation.