WEBVTT 00:00:00.017 --> 00:00:05.937 So this is Paul Verschure with Olof Sporns. And he's part of our Olof's Also 00:00:05.937 --> 00:00:08.717 Wanted speakers at our BCBT summer school. 00:00:09.397 --> 00:00:15.717 And Olof has been introducing this notion of, if you want, the connectome of the human brain. 00:00:16.177 --> 00:00:20.997 So Olof, maybe you can say something about what this actually means. 00:00:20.997 --> 00:00:26.897 Well, it needs the set of structural connections between regions or individual 00:00:26.897 --> 00:00:30.357 neurons that exist in the human brain. 00:00:31.757 --> 00:00:36.477 Connectome is a structural description of the connectivity of the network that 00:00:36.477 --> 00:00:37.517 makes up the human brain. 00:00:37.657 --> 00:00:43.657 Now, I mentioned individual neurons, and that's going to be challenging to get to. 00:00:45.357 --> 00:00:50.837 All right, so all of you were at the Connectome and were telling me what it means. 00:00:50.997 --> 00:00:55.497 It's a description of the network of the human brain. 00:00:55.717 --> 00:00:59.657 And network means it's got nodes and connections. 00:00:59.977 --> 00:01:04.277 The nodes could be as small as individual neurons, or they could be a little 00:01:04.277 --> 00:01:09.897 larger and could be regions of the brain that we can distinguish from each other. 00:01:10.517 --> 00:01:15.597 And when we proposed that for the first time years ago, really it was motivated 00:01:15.597 --> 00:01:19.317 because we didn't have such a description for the human brain at all. 00:01:19.317 --> 00:01:24.117 And it seemed to me that we needed one, because everybody out there is doing 00:01:24.117 --> 00:01:28.177 functional brain recordings and dynamic measurements of brain activity, 00:01:28.417 --> 00:01:33.217 but we don't really know, in my view, where all this is coming from and how 00:01:33.217 --> 00:01:34.637 it's actually generated unless 00:01:34.637 --> 00:01:37.797 we have an understanding of what the connectivity looks like underneath. 00:01:38.577 --> 00:01:42.877 And it was a challenging goal to have because, you know, we don't have really 00:01:42.877 --> 00:01:47.797 good methods yet to map the connectome, certainly not at the level of individual neurons. 00:01:48.177 --> 00:01:50.637 That's really challenging because we've got so many of them. 00:01:50.977 --> 00:01:53.877 And we can't do it in humans in a non-invasive fashion. 00:01:54.177 --> 00:01:57.397 What we can do now is we can do diffusion imaging, 00:01:57.637 --> 00:02:04.977 which allows us to infer connectivity in live human beings without invasively 00:02:04.977 --> 00:02:08.197 going in there and destroying So that's good, 00:02:08.317 --> 00:02:11.777 and people are starting to do this, and it's beginning to give us a picture 00:02:11.777 --> 00:02:13.297 of what this network might look like. 00:02:14.037 --> 00:02:20.597 But in some sense, you could argue, well, this idea that connectivity is important 00:02:20.597 --> 00:02:23.217 to explained function is with us for some time. 00:02:23.357 --> 00:02:27.037 And also, anatomists have developed many different methods to do this. 00:02:27.077 --> 00:02:30.297 Maybe not at this massive scale. Absolutely. So…. 00:02:31.376 --> 00:02:35.736 I don't think it's a new idea in the sense, you know, I don't want to claim 00:02:35.736 --> 00:02:39.116 that, you know, this is the first time people have proposed connectivity is important. 00:02:39.276 --> 00:02:44.256 No, no. I mean, it goes back all the way to certainly Ramon y Cajal and perhaps earlier than that. 00:02:44.256 --> 00:02:48.856 And he, of course, was perhaps the greatest anatomist of all time and had these 00:02:48.856 --> 00:02:50.456 intuitive insights that are 00:02:50.456 --> 00:02:56.736 really fascinating to look at even now about how the brain was a network, 00:02:56.976 --> 00:02:59.816 even though he didn't – I'm not sure he used the word network. 00:02:59.936 --> 00:03:03.176 In fact, the opposing theory was a particular theory by Golgi, 00:03:03.296 --> 00:03:05.676 and he, in some sense, was more of a networker. 00:03:06.136 --> 00:03:10.656 Nevertheless, I mean, he – and also to some extent, Golgi gave us exquisite 00:03:10.656 --> 00:03:14.676 descriptions of how things were connected. And so that notion has very, 00:03:14.736 --> 00:03:16.396 very deep roots in the history of neuroscience. 00:03:16.736 --> 00:03:22.876 So I'm not saying that that was the first idea about it, but it is the case 00:03:22.876 --> 00:03:27.996 that until about a few years ago, we had a very sketchy understanding of, 00:03:28.076 --> 00:03:30.436 especially in the human brain, how things were connected. 00:03:30.576 --> 00:03:34.656 We had a much better understanding in primates, in non-human primates, 00:03:34.676 --> 00:03:41.216 in macaques and a few other species where people had done invasive anatomical studies with tracers. 00:03:41.216 --> 00:03:46.996 And so we had some picture of how that actually hung together as a network. 00:03:47.116 --> 00:03:50.376 And that was very informative, certainly for me and also, I think, 00:03:50.376 --> 00:03:53.516 for others to study those network diagrams. 00:03:54.036 --> 00:03:59.496 And so we felt like we wanted to push for, you know, doing something similar for the human brain. 00:03:59.736 --> 00:04:02.936 And really, it's been proposed before. I mean, when you go back in literature, 00:04:03.136 --> 00:04:08.596 you find people saying this, you know, going back certainly 10, 15, 20 years. 00:04:09.156 --> 00:04:11.396 So there's no claim here this is in any way original. In fact, 00:04:11.476 --> 00:04:16.756 the only original part of it is, in some sense, that we were aiming to derive 00:04:16.756 --> 00:04:22.416 this description and then make it a basis for understanding in a dynamic computational sense, 00:04:22.676 --> 00:04:27.436 how do we get from the structural network to the dynamic, the changing dynamic 00:04:27.436 --> 00:04:33.916 picture of activations, coactivations, all that rich repertoire of dynamic states 00:04:33.916 --> 00:04:35.376 that we have available to us? 00:04:35.376 --> 00:04:41.636 How do we get from this anatomical network to this rich dynamics that we're 00:04:41.636 --> 00:04:44.336 generating every instant in our lives? 00:04:44.616 --> 00:04:50.096 But another contributing factor, it seems that there are also new quantitative 00:04:50.096 --> 00:04:52.296 methods to look at these kinds of networks. 00:04:52.356 --> 00:04:56.056 Yeah, that's something that I think we have only had for a few years now. 00:04:56.834 --> 00:04:59.934 Uh the fact that we can go in and and take 00:04:59.934 --> 00:05:03.054 a network and actually say something about it not just describing it 00:05:03.054 --> 00:05:05.814 you know this is connected that and it's not connected so not not not just 00:05:05.814 --> 00:05:08.754 give a narrative description of what's connected to what but also say something 00:05:08.754 --> 00:05:14.354 about the organization of it and it's driven actually by theoretical developments 00:05:14.354 --> 00:05:17.634 that are in other that have occurred in other fields than neuroscience you know 00:05:17.634 --> 00:05:21.894 people have been studying social networks for a long time going back all the 00:05:21.894 --> 00:05:23.174 way to the 1930s, really, 00:05:23.314 --> 00:05:27.134 and then blossoming really in the 60s and 70s. 00:05:27.414 --> 00:05:33.854 And people coming from the physics community have increasingly applied network 00:05:33.854 --> 00:05:38.754 metrics and graph theory measures to complex networks ranging all the way from 00:05:38.754 --> 00:05:44.834 the web to transportation and power grid and collaboration networks, citation networks. 00:05:45.154 --> 00:05:48.714 And it's been, I think, on the whole, very informative about, 00:05:48.914 --> 00:05:52.214 it sort of allows you to unpack something about the structure of these networks 00:05:52.214 --> 00:05:55.374 and also compare them and say, how is this differently organized from that? 00:05:55.894 --> 00:06:00.194 How these networks come to be, it's driven by human activity, 00:06:00.414 --> 00:06:06.554 it's driven by technology, it's driven by social developments in social systems. 00:06:06.754 --> 00:06:10.014 And in the brain, it's driven by development and perhaps evolution. 00:06:10.574 --> 00:06:14.054 And so these are interesting ways by which we can explore 00:06:14.054 --> 00:06:16.794 networks and they haven't been around for very long and so that's i 00:06:16.794 --> 00:06:19.474 think maybe another new ingredient here and with those 00:06:19.474 --> 00:06:22.294 it seems to suggest or in some sense you take 00:06:22.294 --> 00:06:25.334 a more an anti-reductionist stance in 00:06:25.334 --> 00:06:28.254 this perspective right because sometimes you're saying also if you're going 00:06:28.254 --> 00:06:32.094 to say about the social social networks um social networks are more than their 00:06:32.094 --> 00:06:36.334 parts yeah and in some sense you make a similar statement then on brain organization 00:06:36.334 --> 00:06:42.894 i personally do yes i think that when it comes to I like to use the economy as a metaphor. 00:06:43.994 --> 00:06:50.794 In a trivial sense, the economy at a large scale is the sum total of all the 00:06:50.794 --> 00:06:53.754 individual transactions that go on between people. 00:06:54.714 --> 00:06:59.354 If I purchase an item or take out an insurance policy or a mortgage, 00:06:59.574 --> 00:07:04.354 those are interactions that involve me as a single agent and perhaps a few other 00:07:04.354 --> 00:07:06.994 agents that are representing banks or what have you. 00:07:07.914 --> 00:07:13.274 But the entirety of the economy, you know, the global state that it's in, 00:07:13.454 --> 00:07:16.574 perhaps, you know, whether economies globally are growing or contracting, 00:07:16.814 --> 00:07:22.874 whether they are doing well or not so well, it's hard to reduce that to me as 00:07:22.874 --> 00:07:26.774 a, you know, it's not really, doesn't exist at my level. I contribute to it. 00:07:27.134 --> 00:07:29.354 I'm an atom in the economy. 00:07:29.914 --> 00:07:33.854 But what the economy is cannot be reduced to my actions, really. 00:07:34.446 --> 00:07:39.886 I think the same thing it is in the brain. Neurons are obviously the building blocks of what goes on. 00:07:39.986 --> 00:07:45.286 They each contribute in various ways to the overall pattern of activity that's there. 00:07:45.626 --> 00:07:51.926 But the pattern as a whole is not just the sum total of what goes on in individual 00:07:51.926 --> 00:07:53.966 neurons. There's more to it. There's organization. 00:07:54.526 --> 00:08:00.726 There's emergence, if you wish. It's a very difficult term to use, 00:08:00.786 --> 00:08:02.186 perhaps, but I'll use it anyhow. 00:08:02.186 --> 00:08:09.666 And so I think the network perspective also takes you away from trying to reduce things to, 00:08:09.706 --> 00:08:15.686 you know, reducing memory or attention to properties of single neurons and rather 00:08:15.686 --> 00:08:17.966 view them as properties of a system. 00:08:18.566 --> 00:08:22.046 Right. But there are also some new terminologies coming in, right? 00:08:22.086 --> 00:08:25.366 Because now, for instance, you would talk about hubs, where you would say, 00:08:25.446 --> 00:08:32.126 well, these would be areas that play a special role in these larger dynamical systems. 00:08:33.106 --> 00:08:38.186 So how should we think about this notion of a hub? Yeah, that's a good question. 00:08:38.246 --> 00:08:42.966 In fact, it gets us back, in some sense, it inverts my statement I just made. 00:08:43.626 --> 00:08:48.526 Network approaches also allow you to analytically decompose a network again 00:08:48.526 --> 00:08:52.106 into elements, and then ask questions about these elements. 00:08:52.246 --> 00:08:56.686 How are they embedded in the network? What is the contribution they make to the network? 00:08:56.906 --> 00:09:00.326 What happens if they are deleted, if they are damaged? What are the consequences 00:09:00.326 --> 00:09:05.546 for the network as a whole? And in other disciplines, people have looked at 00:09:05.546 --> 00:09:08.306 these questions. For instance, in, 00:09:09.168 --> 00:09:13.088 protein networks, interaction networks, genomic networks. The question is, 00:09:13.108 --> 00:09:16.508 you know, if I have a protein here and this protein is deleted from the cell, 00:09:16.588 --> 00:09:21.028 it disappears, what are the consequences for the integrity of the other, 00:09:21.148 --> 00:09:23.568 of the remaining interactions in the cell? 00:09:23.688 --> 00:09:27.668 Do they all become disrupted as well? This is completely inconsequential. 00:09:28.568 --> 00:09:30.548 Similar questions can be asked in the brain. 00:09:31.848 --> 00:09:35.228 They're beginning to be asked. I think we have no good answers yet. 00:09:35.408 --> 00:09:38.288 I think it's a good question. What does it mean to be a hub, 00:09:38.308 --> 00:09:40.468 okay, in the brain? Is that a meaningful question? 00:09:40.748 --> 00:09:44.208 I believe it is, but we haven't got a lot of empirical evidence yet. 00:09:44.688 --> 00:09:48.168 What happens when we delete certain regions of the brain when there's a lesion 00:09:48.168 --> 00:09:53.568 happening in a primary sensory area versus an area that's perhaps involved in, 00:09:53.588 --> 00:09:55.088 quote, high-level processing? 00:09:55.688 --> 00:10:01.108 Can we, does graph theory, does network theory give you some handle on understanding 00:10:01.108 --> 00:10:05.968 what the consequences are of that? that, make predictions that you can empirically test. 00:10:06.108 --> 00:10:08.308 I think these are open questions that are beginning to be asked. 00:10:08.588 --> 00:10:15.528 But in a recent paper, if you want, your co-authors did put your finger on a 00:10:15.528 --> 00:10:19.108 potential hub, which was actually also a neglected piece of the brain. 00:10:19.588 --> 00:10:25.268 Yeah, we discovered really, in a sense that we didn't know beforehand anything 00:10:25.268 --> 00:10:28.548 about it, we discovered that there was a region in the medial parietal cortex 00:10:28.548 --> 00:10:30.268 that was really quite prominent. 00:10:30.408 --> 00:10:34.508 It stuck out when we looked at the network. It stuck out in terms of the number 00:10:34.508 --> 00:10:36.228 of connections that it maintained, 00:10:36.668 --> 00:10:42.328 the number of connections that were attached to it, the centrality of it, 00:10:42.388 --> 00:10:47.288 which means how central it is relative to the different paths that go through the system. 00:10:47.568 --> 00:10:52.868 And only at that point did we then read the literature and actually discovered 00:10:52.868 --> 00:10:58.408 in the literature that people had been describing this piece of brain before in the context. 00:10:59.806 --> 00:11:04.726 The default mode network, where it is a central component, but also brain activation 00:11:04.726 --> 00:11:08.306 studies that have to do with self and with memory retrieval, 00:11:08.406 --> 00:11:09.566 with the imagined in the future, 00:11:09.966 --> 00:11:14.826 lesion studies, where if you do have damage in that area, it's very disruptive. 00:11:15.246 --> 00:11:20.206 And so it had already been noted as an extraordinary brain region before, in some sense. 00:11:20.366 --> 00:11:28.566 It came up in the network study as being also rather prominent and that perhaps provides predictions. 00:11:28.886 --> 00:11:32.526 One of the things that people have noted about this is that it's very highly 00:11:32.526 --> 00:11:35.086 activated when you are at rest. 00:11:35.246 --> 00:11:37.726 This is the part of the brain that uses up the most energy. 00:11:38.746 --> 00:11:44.306 Now is that a cause or is that a consequence of it being a hub region? 00:11:44.726 --> 00:11:49.546 Perhaps hub regions are just busier. Perhaps they are more engaged in the ups 00:11:49.546 --> 00:11:52.286 and downs of what goes on in the brain and therefore for a brown more energy. 00:11:52.966 --> 00:11:56.846 So that's a prediction one could test, perhaps in an empirical setting. 00:11:57.426 --> 00:12:02.986 But now the kind of network data on which you base this interpretation is in 00:12:02.986 --> 00:12:05.906 some sense fairly abstract, right? 00:12:06.006 --> 00:12:10.206 It's not like a direct anatomical measure of connections. 00:12:10.626 --> 00:12:15.446 It's a derived measure. Like in this case, you do this tensor diffusion imaging. Yeah. 00:12:17.126 --> 00:12:22.386 So what are the problems that you're facing using these really abstract methods 00:12:22.386 --> 00:12:25.046 in inferring connectivity? 00:12:25.466 --> 00:12:31.106 I think that's another good question. I mean, I like to point this out to people all the time. 00:12:32.186 --> 00:12:36.086 The graph description or network description is an abstract model. 00:12:36.346 --> 00:12:40.206 It is not the real brain. There are many assumptions that go into constructing 00:12:40.206 --> 00:12:45.746 it, starting with defining what your nodes are and then finding some way of measuring the edges. 00:12:46.066 --> 00:12:51.466 What is that really that we get from diffusion imaging? Is that the number of 00:12:51.466 --> 00:12:53.486 axons that go through a particular volume? 00:12:53.726 --> 00:12:58.126 Is that a measure of their myelination status, how myelinated they are? 00:12:58.386 --> 00:13:02.646 Is it a measure of their thickness or their integrity? 00:13:03.386 --> 00:13:05.826 So these are good questions. For example, 00:13:05.826 --> 00:13:09.366 we have no access right now to physiological efficacy or strength. 00:13:09.806 --> 00:13:14.486 We could be finding axons that actually are physiologically very weak. 00:13:14.826 --> 00:13:19.026 We could find axons that are physiologically very strong, and we would never 00:13:19.026 --> 00:13:22.686 be able to tell the difference right now using just this one technology. 00:13:23.106 --> 00:13:28.646 So that is clearly a limitation of that particular technique. 00:13:28.926 --> 00:13:34.466 My sense is, though, that as we move along in time, we're going to get deeper insights into this. 00:13:34.566 --> 00:13:37.546 Other methods will come along. Perhaps this method will be more refined. 00:13:38.986 --> 00:13:44.186 There's also the combination of doing diffusion imaging and perhaps a functional 00:13:44.186 --> 00:13:48.866 MRI or perhaps EEG or MEG, combining different modalities, you know, 00:13:48.866 --> 00:13:50.846 getting a better handle on what goes on dynamically, 00:13:51.146 --> 00:13:54.566 because that can give you a clue as to how strong these connections really are. 00:13:54.746 --> 00:13:57.566 So it's, it doesn't really, um. 00:13:58.699 --> 00:14:03.879 While all these limitations are certainly in place, and I always like to point 00:14:03.879 --> 00:14:10.439 out that there are limitations, to me it doesn't detract from the value of the approach itself. 00:14:10.559 --> 00:14:14.579 Because the approach is ultimately motivated by a theoretical argument, 00:14:14.859 --> 00:14:16.459 namely networks are really important. 00:14:16.679 --> 00:14:20.159 They're not everything, but they're really important for understanding the brain. 00:14:21.159 --> 00:14:26.519 Technologies will always be turning over. I don't think in five or ten years, 00:14:26.719 --> 00:14:31.099 who knows, we will be doing this much fMRI anymore. We might be doing something different. 00:14:31.759 --> 00:14:36.479 But the questions will remain the same as they have since our style, right? Sure. 00:14:36.899 --> 00:14:41.599 But now there's a problem, right? Because in some sense, the methods you use 00:14:41.599 --> 00:14:46.959 to quantify a network structure are neutral to their interpretation. 00:14:46.959 --> 00:14:51.139 And in some sense, there are multiple ways in which you can generate complex 00:14:51.139 --> 00:14:55.039 data sets in which you can now discover network-like organizations. 00:14:55.099 --> 00:14:59.159 But maybe 95% of these are spurious networks. 00:14:59.319 --> 00:15:02.339 Like, for instance, the case you just described. Indeed, we might have myelinated 00:15:02.339 --> 00:15:06.759 fibers going somewhere with synapses that are not effective at all. 00:15:07.079 --> 00:15:12.239 So aren't you, in some sense, adding more noise to the debate? 00:15:12.239 --> 00:15:16.839 Or how do we prevent adding more noise to the debate that is already there? 00:15:17.079 --> 00:15:21.999 I think the only way to really come to a consistent picture in the end is to 00:15:21.999 --> 00:15:27.679 really do multiple different studies, come to some sort of agreement between 00:15:27.679 --> 00:15:30.759 investigators who use slightly different methodology. 00:15:31.139 --> 00:15:34.799 It would be wonderful if we had a totally different way of measuring brain, 00:15:34.919 --> 00:15:37.959 a connectivity anatomy, in a way that's not invasive. 00:15:38.239 --> 00:15:42.139 And it might happen one day. I don't know what it will be, but I'm sure it'll happen. 00:15:43.019 --> 00:15:45.099 And then this will be applied again, 00:15:45.279 --> 00:15:48.319 and we have a way of going back and distinguishing signal from noise. 00:15:48.799 --> 00:15:52.679 Remember, there's always classical anatomy. We always can cut up brains and 00:15:52.679 --> 00:15:55.879 look at them histologically, and that's by many considered to be the ground truth. 00:15:56.539 --> 00:16:00.619 I think we will see new methods in that realm pretty soon for humans, 00:16:00.839 --> 00:16:04.939 for human brain, postmodern brain, to go back and really slice them very thinly, 00:16:04.939 --> 00:16:08.519 to try to reconstruct using microscopy 00:16:08.519 --> 00:16:11.539 the direction of fiber pathways at least 00:16:11.539 --> 00:16:14.779 and perhaps something about their density that'd be a totally different way 00:16:14.779 --> 00:16:21.019 of looking at it i would hope that we can then come to an overlap between these 00:16:21.019 --> 00:16:26.119 two approaches but what is interesting is that you don't seem to refer at all 00:16:26.119 --> 00:16:29.079 to any kind of animal model why is that. 00:16:30.994 --> 00:16:38.934 Well, I like animals and I like animal models even more, but a lot of work has 00:16:38.934 --> 00:16:42.554 been done in mice, obviously, for the genetics, and we have a lot of data there. 00:16:42.714 --> 00:16:46.834 It'd be very important to get some good connectivity data in the mouse. 00:16:47.034 --> 00:16:48.834 I think people are working on that. 00:16:49.234 --> 00:16:54.374 The macaque has been for a long time a preferred model for primates. 00:16:55.014 --> 00:17:01.294 So that's all good. I think if we had even a much wider range of animal species, 00:17:01.714 --> 00:17:05.654 we could get into comparative studies and that would be very exciting to figure 00:17:05.654 --> 00:17:09.694 out, you know, so how is the, what are the architectural principles that vary 00:17:09.694 --> 00:17:13.494 across different animal species, perhaps within the mammalian, 00:17:13.514 --> 00:17:16.534 the animal kingdom, perhaps even including invertebrates, you know? 00:17:16.694 --> 00:17:20.154 How are these brains different in terms of the architectural principles? 00:17:20.314 --> 00:17:25.114 So that's all great. I personally am interested in humans, and I think humans 00:17:25.114 --> 00:17:31.574 deserve that interest because we are interested also in figuring out what goes 00:17:31.574 --> 00:17:34.734 wrong in brain diseases that are very common. 00:17:34.734 --> 00:17:37.734 And it turns out that my hunch is 00:17:37.734 --> 00:17:42.054 that a lot of the conditions that we're struggling with in human beings, 00:17:42.214 --> 00:17:48.174 from Alzheimer's to schizophrenia to other forms of brain damage and so forth, 00:17:48.234 --> 00:17:50.774 have ultimately something to do with connectivity. 00:17:51.534 --> 00:17:57.334 Disturbances of connectivity that manifest themselves in disturbances of mental processing. 00:17:57.674 --> 00:18:02.854 And that's one of the things that I personally am interested in, 00:18:02.854 --> 00:18:06.714 you know, understanding that better. I think it's very difficult to get to animal 00:18:06.714 --> 00:18:08.714 models of any of these conditions. 00:18:08.854 --> 00:18:11.334 People have tried very hard, and it's not easy going. 00:18:11.894 --> 00:18:15.654 If we study humans directly, I think we might have a better way to go. 00:18:16.094 --> 00:18:23.734 But I could imagine that some sort of control experiments might be deployed, 00:18:24.034 --> 00:18:27.094 where we actually look at calibration with specific animal models, 00:18:27.154 --> 00:18:33.714 where at least we have some sort of understanding of the physical connectivity and their activity. 00:18:34.334 --> 00:18:37.994 So that's really, I totally agree with that. I mean, people have actually, 00:18:38.214 --> 00:18:40.574 myself included, have tried to do that with macaque. 00:18:41.190 --> 00:18:46.410 Where we have vesicle track tracing and still being done by some laboratories. 00:18:47.270 --> 00:18:51.310 And we can also do the fusion imaging on the macaque brain. 00:18:51.370 --> 00:18:58.850 We can take a post-mortem brain and put it in a scanner and actually scan at 00:18:58.850 --> 00:19:03.910 very high resolution because post-mortem, you don't have to worry about keeping 00:19:03.910 --> 00:19:06.230 it in there for a long time. It doesn't move and so forth. 00:19:07.150 --> 00:19:11.910 No, she waits long enough. She waits long enough, it moves. And it turns out 00:19:11.910 --> 00:19:15.590 that in the... Now, that tissue is very hard to come by. 00:19:15.690 --> 00:19:24.550 It's very precious, obviously, and you don't want to be working with that on any kind of large scale. 00:19:24.650 --> 00:19:29.050 So there's only very few brains that become available more or less by accident. 00:19:29.350 --> 00:19:34.310 And in the one case where I myself have looked at a macaque hemisphere, 00:19:34.710 --> 00:19:40.250 diffusion imaged, and then compared that to a classical Pocomac database description 00:19:40.250 --> 00:19:41.930 coming from track tracing. 00:19:42.410 --> 00:19:48.470 I forget the exact numbers now, but overall about 5% of the fibers that were 00:19:48.470 --> 00:19:55.270 found in the diffusion imaging were in places where we know fibers not to exist in the track tracing. 00:19:55.430 --> 00:19:59.210 So we had relatively few false positives. 00:19:59.210 --> 00:20:06.150 However, we also know that diffusion imaging is notorious for struggling with 00:20:06.150 --> 00:20:09.310 crossing fiber pathways, with inter-hemispheric connections, 00:20:09.710 --> 00:20:10.910 with very long connections. 00:20:11.230 --> 00:20:18.090 And I think we are dropping out connections there at a rate that we need to improve on. 00:20:18.310 --> 00:20:24.270 So I think where we find fibers, I think we have them, but we also are missing 00:20:24.270 --> 00:20:28.210 some, and that's something that we need to work on further. But now if you find 00:20:28.210 --> 00:20:32.070 fibers, how can you say something about their causal organization? 00:20:33.151 --> 00:20:36.211 Right, what the input-output structure really is. 00:20:36.431 --> 00:20:41.471 Yeah, and that's, again, classical track tracing allows you to say much, 00:20:41.491 --> 00:20:44.271 much more about that, because you have directionality. 00:20:44.291 --> 00:20:48.111 You actually know where you've injected, and then you know how the tracer is transported. 00:20:48.991 --> 00:20:51.831 And that tells you something about the direction of the pathway itself. 00:20:52.251 --> 00:20:58.431 And not every pathway is bidirectional. There are definitely pathways that go 00:20:58.431 --> 00:21:01.771 only in one direction. and they could be extraordinarily important by setting 00:21:01.771 --> 00:21:03.171 up the dynamics, it turns out. 00:21:03.611 --> 00:21:08.591 People have been studying this. Now, we have no information on that coming from 00:21:08.591 --> 00:21:11.531 these non-invasive diffusion imaging methods at the moment. 00:21:12.391 --> 00:21:16.951 I think there are some people out there who are working on this, 00:21:17.051 --> 00:21:21.151 but I have not seen a single study where people have been able to determine directionality. 00:21:21.251 --> 00:21:23.371 It's a big gap in our knowledge. 00:21:24.511 --> 00:21:30.291 Okay. But there may be also another interest of yours could come in, right? 00:21:30.311 --> 00:21:34.611 Because certainly, I mean, you're around in this field now for some time and 00:21:34.611 --> 00:21:38.351 also with a strong theoretical interest and having worked and published on a 00:21:38.351 --> 00:21:39.711 number of models of the brain. 00:21:40.331 --> 00:21:43.971 But in some sense now you seem to be more in a descriptive mode you're going 00:21:43.971 --> 00:21:47.951 to describe these complex networks it's difficult to get access to causality 00:21:47.951 --> 00:21:52.411 so how do you think are you going to bring this back into a more theoretical realm, 00:21:53.910 --> 00:21:57.130 Putting together, let's say, a more broad picture on brain function. 00:21:57.270 --> 00:22:01.370 How are you going to achieve that? I think I personally do want to do that, yes. 00:22:03.090 --> 00:22:07.610 It's true right now. I mean, making a map of the connections of the human brain, 00:22:07.690 --> 00:22:11.870 let's say, is primarily a descriptive task. You're not testing hypotheses. 00:22:12.230 --> 00:22:17.930 You're not really interrogating the architecture. You are trying to make a map. 00:22:18.310 --> 00:22:23.330 Now, I happen to believe that having a good map really helps you if you want to get places, right? 00:22:23.330 --> 00:22:28.270 Because, you know, now you know something about how things are organized, 00:22:28.270 --> 00:22:33.890 and then perhaps you can ask good questions that are open to experimentation 00:22:33.890 --> 00:22:37.510 or open to modeling, which is not just descriptive, 00:22:37.730 --> 00:22:40.210 but also serves to test hypotheses. 00:22:40.210 --> 00:22:44.570 So, one of the questions that I have in my mind is we've been describing connections 00:22:44.570 --> 00:22:45.990 and we've been making maps. 00:22:46.170 --> 00:22:49.350 I want to bring this back to human cognition, right? 00:22:49.690 --> 00:22:54.310 We haven't talked about that yet, but this network in there is doing something, right? 00:22:54.430 --> 00:22:57.670 It's doing something right now. It's making me speak and making you listen and 00:22:57.670 --> 00:22:59.630 ask the next question and this is all happening, right? 00:22:59.630 --> 00:23:05.610 So how do we get from a description, these fibers go here and these fibers go 00:23:05.610 --> 00:23:11.650 there, to an account what is going on in the architecture as we're engaged in 00:23:11.650 --> 00:23:14.050 this interaction, dynamically engaged? 00:23:14.410 --> 00:23:19.210 That cannot be reduced to the anatomy. The anatomy is like a skeleton that maybe 00:23:19.210 --> 00:23:22.710 enables all this to happen, but it isn't going to make it do by itself. 00:23:22.710 --> 00:23:29.230 And so my hope is that when we understand the map a little bit better, 00:23:29.270 --> 00:23:34.330 maybe we also see some relationships that this map has to broad categories of 00:23:34.330 --> 00:23:36.010 human mental processing. 00:23:36.250 --> 00:23:41.610 You know, things like memory and attention, planning, movement. 00:23:42.450 --> 00:23:44.950 But in a correlational sense, or... 00:23:46.115 --> 00:23:49.635 Do you see yourself build models, large-scale models? 00:23:49.855 --> 00:23:56.075 Hopefully more than correlation, because if you can make models that are structurally 00:23:56.075 --> 00:24:02.075 based and that get you into the realm of neurodynamics, but also perhaps cognitive processing, 00:24:02.295 --> 00:24:08.655 engagement with the external world, in a modeling context, you can now manipulate the model, right? 00:24:08.655 --> 00:24:14.035 It's a little bit like mutating a protein and seeing how the mutation affects 00:24:14.035 --> 00:24:16.095 what it actually does in the cell. 00:24:16.795 --> 00:24:21.495 You can manipulate it. That's a perturbation of the system. 00:24:21.735 --> 00:24:26.775 Something that you can now do to ask questions about, if I delete this pathway 00:24:26.775 --> 00:24:30.715 over here or if I make it stronger, what is the effect of that? 00:24:30.715 --> 00:24:35.255 That, not just on that pathway, because you've deleted it, but also on the rest of the brain. 00:24:35.395 --> 00:24:40.035 How does it reorganize? How does it reshape itself to deal with that disturbance 00:24:40.035 --> 00:24:41.195 that you've just included? 00:24:41.395 --> 00:24:45.295 But is it going to be like a brain in a vet model, or it's going to be a model 00:24:45.295 --> 00:24:47.855 that will be instantiated in some form in the real world? 00:24:47.915 --> 00:24:50.775 Well, I sure hope it has to be embodied. I mean. 00:24:55.198 --> 00:25:01.138 To be more conservative might be happening more in a closed setting where you're 00:25:01.138 --> 00:25:03.298 just having a big network and 00:25:03.298 --> 00:25:06.978 you are studying dynamics within it and it's actually not doing anything. 00:25:07.298 --> 00:25:11.998 That's what we've been doing in the past couple of years and that can be extraordinarily 00:25:11.998 --> 00:25:16.018 informative about what the capacities of that network are, you know, 00:25:16.038 --> 00:25:18.678 what kind of dynamics can it actually generate by itself. 00:25:18.858 --> 00:25:22.818 But it doesn't tell you how it reacts to a perturbation coming from the outside, 00:25:22.858 --> 00:25:29.138 a stimulus. And even that's not enough, because you also have to allow the brain to act on the world. 00:25:29.298 --> 00:25:34.198 Otherwise, if I took your brain out, forgive me for suggesting that, 00:25:34.298 --> 00:25:36.918 put it into a vat, I don't think you would be you anymore. 00:25:37.278 --> 00:25:44.018 And not just in your physical appearance, but also I think your cognition would shrink dramatically. 00:25:44.418 --> 00:25:46.398 You wouldn't be able to do anything. I might even be more boring. 00:25:47.278 --> 00:25:50.278 No i mean you know we are we are always we 00:25:50.278 --> 00:25:53.998 always embodied every moment of our existence we are connected 00:25:53.998 --> 00:25:56.718 to our body our body allows us to act on the 00:25:56.718 --> 00:26:00.478 world it's the only chance we have to do anything um that 00:26:00.478 --> 00:26:03.738 has got to be important i know it is important and mapping 00:26:03.738 --> 00:26:09.558 the brain brain's connectivity should not distract us from realizing that exactly 00:26:09.558 --> 00:26:16.798 so to switch a little bit i I also understand that this initial more intuitive 00:26:16.798 --> 00:26:21.638 idea about the connectome has really been picked up at a much larger scale in the US. 00:26:21.858 --> 00:26:26.538 And now NIH is supporting a number of projects in this area, 00:26:26.758 --> 00:26:31.078 of which also you and your collaborators have got one. 00:26:31.558 --> 00:26:40.698 So what's the objective of this project? It's really to get a ground-level map of the adult, 00:26:40.938 --> 00:26:48.598 normal, that is clinically normal human brain over a wide, over a large population 00:26:48.598 --> 00:26:50.718 of normal, healthy adult brains. 00:26:51.300 --> 00:26:55.700 And the reason for doing it that way, I think, has to do with the fact that 00:26:55.700 --> 00:26:56.900 brains are quite individual. 00:26:57.340 --> 00:27:01.680 So if we just took one brain and we got a map for that brain, that'd be great. 00:27:01.820 --> 00:27:06.140 But you know, we wouldn't know, well, we would know maybe with 80% certainty 00:27:06.140 --> 00:27:10.480 how your brain looks like, but that's not really quite enough. 00:27:11.220 --> 00:27:13.900 There's a lot of variation in individual brains, even normal, 00:27:14.040 --> 00:27:18.540 healthy adult human brains vary tremendously in the way they are physically 00:27:18.540 --> 00:27:22.560 organized. Which is, by the way, a real interesting fact in itself. 00:27:22.820 --> 00:27:30.200 How can we be so alike cognitively and yet have such greatly varying architecture 00:27:30.200 --> 00:27:31.100 that we're working with? 00:27:31.100 --> 00:27:36.200 So part of that, the goal of what NIH wants to do, I think, is to establish 00:27:36.200 --> 00:27:43.420 a foundation in healthy adults and also correlate that with behavioral and genetic data. 00:27:44.700 --> 00:27:50.340 And then I think the next step for others will be to look and compare how these 00:27:50.340 --> 00:27:57.420 normal brains compare with brains from people with certain conditions. 00:27:57.420 --> 00:27:59.820 Another interesting question is development. 00:28:00.140 --> 00:28:04.360 I happen to have done a little bit on that just recently, working with colleagues 00:28:04.360 --> 00:28:07.780 at MGH and Patrick Hagman at EPFL. 00:28:08.000 --> 00:28:12.020 And it turns out that the brain changes tremendously in terms of its connectivity 00:28:12.020 --> 00:28:17.300 structure between ages 2 and 18, which is relatively late development in most 00:28:17.300 --> 00:28:20.140 people's books. At age 2, we are already pretty autonomous. 00:28:20.720 --> 00:28:25.860 And yet you see continuing changes to network changes, connections change, 00:28:26.100 --> 00:28:30.740 there's different network metrics that systematically change across development. 00:28:31.760 --> 00:28:36.840 And these are just some of the questions that can be asked and can be productively 00:28:36.840 --> 00:28:39.920 hopefully addressed using these kinds of connectome data. 00:28:40.660 --> 00:28:44.020 So to close off, two questions. 00:28:44.340 --> 00:28:49.540 So you're a man with some experience in this field and outside this field. 00:28:51.000 --> 00:28:58.460 So what's the Olaf Sporn's law that we should remember so what's Olaf's law 00:28:58.460 --> 00:29:05.740 about what science life in general exactly I don't know I would say, 00:29:06.480 --> 00:29:10.280 I can say one thing as someone who's working connectivity as you've pointed 00:29:10.280 --> 00:29:14.740 out it made me feel old actually I've been doing this for a while, 00:29:15.640 --> 00:29:18.480 10 years ago nobody wanted to hear about this really. 00:29:18.800 --> 00:29:21.980 There were very few people who were doing this and were interested in it. 00:29:22.500 --> 00:29:27.940 And neuroscientists, quite frankly, didn't understand why we needed any theoretical 00:29:27.940 --> 00:29:31.880 analysis or modeling of this kind that we've just discussed. 00:29:32.680 --> 00:29:39.340 I think with the arrival of these new methodologies that spit out enormous amounts 00:29:39.340 --> 00:29:44.860 of data very quickly, there's a growing realization we have to have some insights here. 00:29:44.940 --> 00:29:49.280 It's not just all about generating masses of data. we have to have some guiding 00:29:49.280 --> 00:29:50.420 principles that actually. 00:29:51.490 --> 00:29:55.030 Allow us to make sense of it and uh i'm 00:29:55.030 --> 00:29:58.110 not sure that it's necessarily uh the spawns law 00:29:58.110 --> 00:30:00.970 but i i would say you know um we do need 00:30:00.970 --> 00:30:03.870 that and i'm very happy to see that we're beginning to 00:30:03.870 --> 00:30:06.710 make progress in that direction so spawns law would 00:30:06.710 --> 00:30:11.790 be something like less data more ideas no no let's say i think more more data 00:30:11.790 --> 00:30:16.830 and more ideas okay data is good generous data now data data is good it's important 00:30:16.830 --> 00:30:20.150 you know we if we didn't have this data that we're having now i think you know 00:30:20.150 --> 00:30:23.970 it was sort of what we were talking about years ago, we didn't have any data like that. 00:30:24.110 --> 00:30:30.230 Now we have some, but we also need simultaneously in neuroscience to be open 00:30:30.230 --> 00:30:35.330 to theoretical ideas, which ultimately allow us to ask new questions and understand 00:30:35.330 --> 00:30:37.250 the material that we're dealing with. 00:30:37.350 --> 00:30:40.230 It's not possible to do science without theory, okay? 00:30:40.350 --> 00:30:43.990 Some people seem to think if we just have more data in the end, 00:30:43.990 --> 00:30:46.550 it'll just sort of fall out, you know? That's not going to happen. 00:30:47.750 --> 00:30:53.470 All other sciences are involving theoretical ideas always, and we should do 00:30:53.470 --> 00:30:54.210 the same in neuroscience. 00:30:54.530 --> 00:30:58.190 Not to say that most of them aren't going to be false. That's fine. 00:30:58.850 --> 00:31:06.170 Progress is made by rejecting ideas that are wrong, and we need to be open about that. 00:31:06.690 --> 00:31:10.370 But we can't make progress without theoretical insights. 00:31:10.630 --> 00:31:14.690 I don't mean just computation. I really mean theory that is having some understanding 00:31:14.690 --> 00:31:16.610 of what really goes on at the fundamental level. 00:31:18.010 --> 00:31:20.970 So my last question would then be what's what's 00:31:20.970 --> 00:31:24.070 the one prediction you really want to stick your neck out for today so 00:31:24.070 --> 00:31:27.330 i can come back to you five years from now say look olaf this was 00:31:27.330 --> 00:31:32.490 your prediction was wrong now you owe me a beer what what's that one prediction 00:31:32.490 --> 00:31:38.970 oh that's a difficult one because uh when you're you're gonna do that i know 00:31:38.970 --> 00:31:41.850 that you're gonna be there in five years and you're gonna ask me about this 00:31:41.850 --> 00:31:46.790 and then i'll remind you every time i meet you i don't know if I have a single prediction, 00:31:46.970 --> 00:31:55.790 but I think that if the model that is currently being constructed, 00:31:56.150 --> 00:32:00.330 that is the brain being a complex network with modules that are interconnected 00:32:00.330 --> 00:32:05.670 through hub regions and perhaps engage in complex dynamics of the kind that 00:32:05.670 --> 00:32:07.170 you find in critical systems, 00:32:07.410 --> 00:32:12.890 if that were to be discarded five years from now, I'd be very unhappy because 00:32:12.890 --> 00:32:15.890 because I wouldn't have a second option to go to. 00:32:17.010 --> 00:32:20.310 That's not a single prediction, but the model that's sort of crystallizing right 00:32:20.310 --> 00:32:24.910 now about the brain as a multistale hierarchical network with complex dynamics, 00:32:25.450 --> 00:32:29.630 where the complexity really matters for how it operates, that I think I would 00:32:29.630 --> 00:32:33.450 hope to see confirmed in five years or in 10 years. 00:32:33.590 --> 00:32:39.270 I don't know. But if that goes out the window, then I'm back to not knowing how it works. 00:32:39.790 --> 00:32:42.270 Very good. Olaf Sporns, thank you very much. Thank you.