 The topic of my talk is going to be looking at some of the, let's say, community practices, but also some of the direction my lab has taken in trying to understand what was previously beautifully described as the complexity of connectivity, except in this case focusing on techniques we have available to map this in the human brain. And in order to do that, building on insights that we have from macaque monkey track tracing, data sets available in the marmoset, monkey as well as classical approaches and studies in the neuroanatomy literature. But to kick this off, I've just read a wonderful biography of Alexander von Humboldt. And I was so compelled by this image that was put together. When Humboldt went off to South America, the state of the art of understanding the natural world, of understanding botany was taxonomy. It was to divide it up into hierarchies of divisions between different species. And after his trip, he had a vision or a way to synthesize this knowledge by embedding it in the geography of the terrain. And what was so innovative about this approach, where he's placing different plant species in these mountain scapes that you see here at different points in this world, was to recognize the axes of relevance for characterizing systematic differences across these different species so that as before the approach was to embed them in a hierarchy in reference to each other, he was able to see the way in which they fit within a common space and what aspects of that space were specifically relevant for understanding their distribution and consistency. I think they're absolutely stunning and I also see them in some ways as one of the frameworks for thinking about what we can bring to mapping the human and more generally the cerebral cortex. What are the axes of relevance? How do different features overlap? And what is a way or a coordinate space in which we can look at it that allows us to better simplify and understand these layers of features that emerge? So there are numerous scales for mapping. And I think this is beautifully illustrated by Betsal and Bassett's review from Nermich from a few years ago. There are multiple different scales and let's say characteristics or features which we can use for mapping the human brain in which are present in the literature. Two of the most common I would say or present are looking at different regions, areas or networks in a spatial sense and looking at graph topology which has already been addressed in the prior talk beautifully. So for those familiar and those not, functional networks or dividing the brain up into distinct long-distance patterns of let's say functionally homogenous systems is one of the major approaches looking at connectivity topology which is already addressed as another. The approach of looking at networks really emerged conceptually from cognitive psychology and its integration with the tools of neuroimaging, specifically this notion of localized function which in this case just gets reincarnated as functions that are spatially distributed across distant regions. But in terms of thinking about this as connectivity organization it's still quite limited. On the other side of the spectrum we have graph theory based approaches which are incarnated as exploring the topology of human brain connectivity and then yet reduced to these various principles of organization, small worldness, efficiency, et cetera. And while those have been quite valuable they also become completely divorced from the underlying anatomy of the cortex. And so one of the challenges to jump very quickly in the interest of time, one of the challenges in my lab and others as well has been precisely this point that was brought up before of how to bring anatomy, how to recognize the consistency of the spatially embedded structure in which all this connectivity is taking place and how to integrate that into our way of thinking about the organization of connectivity. So essentially going back to anatomy. And I'll go through this story very quickly to jump to the main point which is that there's a very rich tradition in classical neuroanatomy texts that over decades developed the notion of a hierarchy of cortical areas based on laminar structure which is translated to a hierarchy of cortical areas based on their connectivity with one another. And I think this comes together quite beautifully in review by Marcel Messelam in the late 90s who integrates across these different levels of neuroscientific research to propose a hierarchical model of global integration, a global hierarchy essentially with functional consequences and is able also to relate that to various clinical questions in the human as well. All of this that he proposes and this is of course schematized is at the time based on tract tracing work in the macaque monkey which he hypothesizes to be present in this structure in the human brain. And it's precisely this structure that we then aim to investigate using empirical data which is now available. So essentially what the model proposes is a sensory segregation which again I know there is evidence to the contrary on this point but this is trying to capture the major characteristics or the broad strokes of these patterns even though there are exceptions that may be quite relevant. But sensory segregation along the angle here and then as we move towards the center of these concentric circles along individual synaptic steps from area to area describes this as a hierarchical processing towards a center an integrative set of regions of trans-modal regions shown in red. To move very quickly here we the main technique that we use is functional connectivity resting state functional connectivity where we simply put people in the scanner have them look at a crosshair do their best to stay away, hold very still and we acquire anywhere from five, six minutes to an hour of data. And what's been notable about this approach and about the field in general is that we're able to map cortical cortical connectivity patterns with a high degree of spatial precision in the prior work that I'd be happy to discuss over the next couple of days. We've been able to demonstrate the convergence with what we observe from track tracing studies in the macaque monkey but I think some of the concerns that have been raised about diffusion-weighted imaging-based tractography are certainly present in resting state data. However, one of the reasons I've stuck to it as a methodology is that it allows for this high degree of spatial precision in mapping cortical cortical patterns of connectivity. What you're observing here are fluctuations as they would be acquired in the scanner. Red is high blue is low green is someone in the middle this is just percent bold signal change and it's these very slow fluctuations that we're able to use to map patterns of spatiotemporal dependency. The approach that we use to go back to Messlem's model and to characterize these kind of global patterns of connectivity or spectra of connectivity just to give an illustration of this idea is to take the connectivity graph that you see on the left here where let's just say that the position of nodes reflects the similarity of their connectivity patterns or their degree of connectivity. We can also reflect that as a connectivity matrix or an affinity matrix and then using a technique called the fusion map embedding which is essentially a nonlinear decomposition technique that allows us to capture the maximum variance along several dimensions in this connectivity matrix. We can then map out these components of connectivity or these dimensions of variance in connectivity patterns. So this is simply illustrating a reordered matrix based on the principal component or the component capturing the maximum variance but we can map out then orthogonal dimensions as well. If we apply this to functional connectivity data in this case using the human connectome project data what we observe is a structure quite similar to what is present in the theory proposed by Messlem in the late 90s where we have segregation between different sensory modalities visual sensor motor auditory in blue and green and then integration towards this set of regions these transmodal regions presented in red here towards the center of that space. Now just focusing in on that dimension gradient one what we find is that at the peak of that we spread out the colors to see where the peak lies it's a set of regions that have been described in the human as the default mode network and so what this appears to capture is a kind of spectrum of connectivity networks that's been considered a canonical set of resting state networks or intrinsic networks in the human that fall along this dimension such that we have unimodal networks on one end in blue and purple visual and motor set of attention networks and then moving towards these higher order memory related networks such as the front bridle and the default mode network and allows us to shift from thinking about these as discreet entities towards interrelated interrelated connectivity patterns that fall along the spectrum again just in the interest of time I'm going to present this very briefly but I believe the following talks will go into this in a bit more detail we've also been able to observe this in macaque monkey data as well as marmoset data which raises a question of the degree to which this may be a preserved or conserved to access a phylogenetic expansion in connectivity patterns it also allows us a framework for describing these network structures across species because we no longer have to enter in explicitly to the various functional definitions of for instance the default mode network to debate whether or not it exists in these other species as it does in the human we can look at these dimensions of organization and use that as a reference frame to then characterize whether or not these different patterns or what are the correspondences across these different species in these patterns of connectivity and in these functional systems so it allows us a different reference frame which I think is quite valuable when we're moving into species where it becomes more progressively more difficult to characterize precisely what the functions are as this organization emerge and here I'm going to move to the question of cortical geometry so stepping away from connectivity focusing directly on the structure of the cortex how does this emerge and this was actually one of the starting points for us for this project Buckner and Kreenin proposed a hypothesis or a theory a few years ago in a wonderful review that started with the basic question of why is it or how is it that we see the emergence of such a large degree of association cortex in the human brain how does that occur what they propose is the tethering hypothesis and the basic idea of this hypothesis is that in the ancient Malian cortex which is quite a bit smaller than the human we have different portions of sensory cortex that are quite proximate to each other they're determined by molecular gradients during cortical entogenesis and there is a a hierarchal structure between them and that that's that that determines the position of them but then we have dramatic cortical expansion in the human brain we still have those gradients that determine the position of primary cortex we have inputs from the thalamus that establish their functional specialization however there is all of this other cortex surrounding those portions that were functionally determined which are left to essentially interconnect by various other mechanisms they're untethered from the gradients to determine the specialization of primary cortex what this suggested to us when we read it is that if this is the case there should be a highly precise spatial mapping between the position of association cortex and the location of primary areas so in this case we aim to test that by using the distance along the cortical surface as a way to kind of characterize this expansion is a relationship between regions and so what we're going to do now is place seed regions in those white points that you see on the cortex those are located at the local maxima of the principal gradient and just measure every other point the cortex how it's shortest distance to any one of those points so every other point in the cortex how close it is to any of those positions just using the geometry of the cortex and here's what we observe so going from red now to blue blue is further away the gray lines that you see are the equidistant positions between these points and what emerged what was quite surprising to us when we first saw this was the convergence with the morphological landmarks of primary sensory motor areas so that the central sulcus located here is equidistant from the adjacent landmark or points of transmodal cortex or default mode nodes if you're familiar with that literature likewise the calcurine sulcus falls at an equidistant position between these peaks of transmodal cortex or at the other end of the spectrum of this connectivity spectrum transverse sulcus as well so auditory cortex so there appears to be a spatial correspondence between these systems that are on one hand oriented towards incoming information from the environment primary sensory areas and at the other end of the spectrum furthest ways you can possibly get along the cortical surface are the regions that are involved in the most higher order integration or the center of that connectivity space and it suggests to me I think I'd be very interested to discuss this further it appears in this case if we interpret the geometry the intrinsic geometry of the cortex as reflecting in aspect of intrinsic connectivity it suggests that this may be a critical organizing mechanism for what also emerges as long-distance connectivity and I should I want to really emphasize at this point that we're talking about one dimension of a extremely high-dimensional space of connectivity the principal gradient is capturing one but it's the axis that represents the maximum variance in that space and so here this raises the question to what degree that might be accounted for by aspects of underlying local connectivity away from primary cortex or away from transmodal cortex but a convergence between these two aspects of connectivity okay in the about 10 minutes or so great so what I've shown so far is a relationship between geometry and large-scale connectivity and this convergence in a similar space what I'd like to show now is some lines of work from my lab as well as from Boris Bernhardt a collaborator who's doing beautiful work at McGill that try to relate this now to aspects of laminar structure to the degree that we can in the human brain and there's been observations to go back decades now that there's a convergence between laminar structure and laminar differentiation as illustrated here in the frontal cortex and connectivity patterns such that like preferentially not exclusively as was discussed but preferentially connects to like areas that have similar degrees of cytokinetic differentiation preferentially interconnect to one with one another Alex Gulas who was in my lab a few years back and is now in Hamburg had a wonderful idea of could we explore this in the human brain and how might we be able to do that and so with Julia Hudenberg we looked at ways to be able to characterize degrees of cytokinetic differentiation or various aspects of the laminar pattern in the cortex and to relate that to human brain connectivity and now this is showing a whole brain distribution of cord intercortical myelin as an illustration there are ways of being able to characterize this using t1 maps which has been validated in at the Max Planck and Leipzig a few years back in another measure using t1 t2 ratio which is something I'll present further in a couple of slides this has also been used by the human connectome project as a way of providing additional information about cortical structure in order to differentiate different areas is very briefly Julia acquired high-resolution data at 7 Tesla across 8 subjects and was able to demonstrate a similarity in the distribution of values from t1 maps reflecting intercortical myelin with the principal gradient from these same subjects as illustrated here and the maps don't correspond perfectly but they illustrate a general pattern of consistency between what we're observing purely in microstructure or in a proxy for microstructure and large-scale connectivity this work has been extended further taking advantage of the remarkable resource of the Big Brain project that has been made available on Loris if you're interested in downloading it and this is a histological preparation at extremely high resolution that enables us to now relate using a consistent space to be able to map between connectivity and the non-invasive measures of connectivity have with aspects of cytore architecture and histology based information about the human brain Conrad Wagstil has done some analytic development to enable for the automated mapping of laminar profiles in the Big Brain data set and this is an example of that where you can now move across different regions of the cortex and characterize variation in the neuronal density over the PL to white surface so this is just illustrating this across different Von E. Connell regions and I'm going through this quickly because of the next step in which Casey Paquala from Boris Bernhardt's lab then took these different maps of cortical profiles map the similarity between them creating a matrix of similarity between the laminar profiles from the Big Brain data described as microstructural profile covariance sorted this in a similar way that I described before using connectivity data and then there we go was able to present the dimension of maximum variance in variation of these profile patterns across the brain and as you can see here it reflects these levels of cytorectonic differentiation that I briefly described before in many ways this provides a kind of validation of the approach for looking at variance in these patterns but it also allows us now to move back to the T1 T2 weighted ratio which is something that we can characterize in vivo in the human brain and to be able to map connectivity and these MR values using this microstructure profile covariance approach so in this case just to go through it again rather quickly in the interest of time both on a group and individual level Boris and his lab were able to beautifully map a relationship between the principal gradient and connectivity and covariance in these laminar profiles in vivo demonstrating a convergence between aspects of cortical structure and long-distance connectivity again I think one of the exciting questions again is to go back and understand to what degree this relates to features of cortical geometry on the individual level I'll just very quickly I think I just have a couple more slides to take the next step in asking the question of the degree to which this space helps us to understand distribution of functions this is an illustration from a review by Rakhir Mahra's Sajjababdi and colleagues in which they demonstrate this principle that if you look at the distribution of specific state in this case it was a contrast between math and story from the human connectome project task data set if you just use the anatomy to reflect that this kind of gets back to that that mountain picture that I showed at the beginning if you just use the anatomy to try and understand that spatial distribution they're quite intermixed it's difficult to differentiate them there's not a lot of structure that is immediately recognizable allows one to differentiate those patterns by projecting the data into a space defined by these patterns of connectivity in which we have the different dimensions reflecting progressively less variance in connectivity patterns it allows us to very clearly demarcate a boundary between these functional states as shown here and the challenge becomes to not simply conduct this type of analysis in a data-driven way but to have a clear neurobiological interpretation of what these dimensions reflect and why it is that we're using these dimensions to capture various aspects of cortical organization it's interesting that we see such convergence across these approaches and that suggests that this is a legitimate set of axes for characterizing the functional domain but we want to understand the mechanisms by which they arise and by which they determine the distribution of specialized function and this is simply a meta-analysis I think I got a couple of minutes left so this is a meta-analysis in which we looked at the distribution of functions along the principal gradient and as you might expect from looking at the distribution of these patterns at one end we do see the kind of motor sensory processing followed by domain general and then all of these memory-driven higher-order integration abstract processes at the opposite end and again this is simply a caricature of what we're observing there but it does suggest this progression that was initially described by Messlem when he proposed this theory of integration from sensory input towards output or a sensory-fugal pattern of organization this also enables us to ask questions about where the this large-scale global cortical hierarchy might be disturbed in various psychiatric disorders in this case in autism spectrum disorder this is again work from Boris Bernhardt who's been taking this line of research in numerous really exciting directions in this case what Boris's group observed is a contraction along the principal gradient so that in this spectrum of disorders that are in part characterized by sensory processing differences in sensory processing there appears to be differences in the way that that information is integrated towards the default mode end of the spectrum and rather than trying to look specifically at one end versus the other this allows to characterize it as a continuous space so all in all where this is going is towards establishing a dimension based in connectivity or otherwise of a coordinate space for the cortex and I've already described this a bit but we have these several observations that appear to be consistent in how these features are distributed the notion of there being a dimension in the cortex that begins with primary and goes towards higher-order areas has been around for a very long time there's nothing new here but we have techniques and data available now to be able to actually map out the nuances of that space and to understand where they both converge and where they diverge and why that might be the case and what additional properties might be conferred by that divergence and we recently can propose that distance might constitute some form or is a starting point for thinking about a cortical coordinate space and I know this is an idea that has been discussed I think what at least one of the points I'd like to enter into the discussion with this is that the a two-dimensional space may not while while it has great utility doesn't appear to be the way the cortex itself is actually functionally organized and so we may need to consider higher-dimensional spaces that are consistent across species in varying ways and what the most meaningful aspects of that geometry are so here we've proposed the distance from the transmodal peaks as well as relative distance from various sensory modalities but I think that there's still a lot of work to be done here and this was merely a proposal for basic structure it gets back to that question of what is the what are the axes that are meaningful when we're thinking about these spaces and what's going to best reflect the various features in a as low-dimensional way as possible so with that of course like to thank my old lab in Leipzig funders all of the data and code is available there and of course Boris Bernhardt's lab at MNI for much of the work that presented as well and thank you and happy to discuss thank you we have time for a few questions and then maybe Piotr can rig up questions hello um wonderful talk thank you I think could could you elaborate on maybe how to start communicating to our colleagues more effectively that the fact that despite the fact that we're embedding things in these spaces like as you just said two-dimensional that maybe biology doesn't know anything about space to start with and so looking how do we how do we then go look for those higher-dimensional spaces it sounds like a very interesting question but yeah so so what what is what is the next step in looking for those higher-dimensional spaces oh um you mean going towards higher-dimensional spaces I mean yeah okay great I think I'm with you um and please tell me if I didn't follow exactly but it's very tempting to be driven by the by the algorithms here and of course there are all sorts of constraints in these and we could we applied diffusion map embedding we've since tested and of course looked at various other approaches and it seems PCA principal component analysis is perfectly reasonable for capturing this as well as long as um but how to not simply be driven by by the algorithms for pulling out these higher dimensions and how to ensure that each dimension that we're including in this kind of a model or approach is actually meaningful yeah we've been taking it quite slowly and carefully for that reason so I think that that is a challenge we're trying to return to the anatomy as quickly as possible from these decomposition techniques I think it's important to use these data driven techniques because otherwise we're simply pointing at different features um rather than letting the data kind of speak to us but then to return back to the anatomy to try and understand how that might be capturing um these features that are being observed in the data um I should mention I think folks here are probably more familiar than I am that these techniques using PCA components in neuroimaging has been around since the mid 90s since the beginning of the field it just wasn't really adopted widely in the way that resting state networks independent component analysis and those types and I think in part that has to do there wasn't a good framework that was introduced alongside it for how it should be interpreted within the field in disgust and so one of the challenges has been to be very careful and slow about about um motivating this through through both the cognitive neuroscience and the anatomy okay thank you very much I uh I look forward to continuing the continued discussion afterwards