 Good morning everyone. Today I would like to present to you the Marmoset Brain Connectivity Atlas and demonstrate to you how you can use this open access resource as a platform for discoveries. In the two previous presentations we learned how we can investigate various kinds of connectivity in a human as well as non-human primates and rodents and let me quickly recapitulate the main points of the two previous presentations. So in essence it is important to study the comprehensive wiring diagram of the brain because it is critical, it's fundamental for understanding how the brain processes information. And we do it by creating models of various kinds, theoretical models, computational models, conceptual models, any kind of models in fact. And these models in order to be useful, in order to provide reasonable insight, have to be fueled with a reliable data, in this case data on connectivity. And as we've seen, especially when it comes to human, we cannot always measure everything we would like to. For instance we cannot perform any invasive experiments and therefore we fall back to animal models such as mouse or non-human primates and among them the common marmosets. Common marmosets are new world primates and they are relatively small, so are their brains. As you can see on this comparison to the surface of the marmoset cortex is approximately 12 times smaller in comparison with macaque. Nonetheless the marmoset brain preserves all the defining features of the primate brain. And marmosets have become increasingly popular among recent years and they turned out to be extremely useful animal model. They're the first non-human primates for which genome was sequenced and stable transgenic lines have been obtained. Marmosets as you can see on this movie are very social animals and they exhibit amazing repertoire of hearing and vocal communication behavior and which is a subject of numerous studies. Finally they are a convenient model for investigating mesoscale connectivity and as we will learn tomorrow from Alex, for instance they constitute a fundamental part of the Japanese program on brain research. But as we heard from David Kennedy investigating connectivity in primates is extremely challenging. It's very difficult and this could be noticed if we compare the existing body, the existing resources and mouse connectivity when the injections are counted in hundreds if not thousands and they cover the entire brain and sometimes they go even outside the brain. And the data is provided in a very sort of modern way. Let's say that this is specially co-registered and it's all available online. For primates we simply do not have this throughput. It's much more challenging. And so perhaps we can place marmosets somewhere in between to fit this gap. Let's see. Regardless of the species that we use, if we think of an ideal toolkit, ideal resource for investigating connectivity, what features should it have? How should it look like? And there's a consensus that of course from the technical standpoint it should obey all the standards and best practices including the fair principle. But from the perspective of the end user it primarily it should provide access to the quantitative results and connectivity as well as it should allow us to access the primary data, the primary unprocessed experimental data. And this is critical from the perspective of re-analysis of the data in the future in perhaps with new tools and methods as well as it will allow us to reinterpret the data in the light of evolving knowledge, novel hypothesis. It also should allow for broad range of analysis. For instance, we would definitely like to do all the good old graph-based network analysis but some of us tend to choose alternative form. They would like to abstract for any kind of persuasion and perform purely specially based analysis. Of course it is critical to be able to use the results in a translational and comparative context and therefore there should be sufficient level of compatibility and interoperability with other tools, methods, modalities or projects that investigate connectivity. And as you can see these requirements, they pose a lot of challenges. The most critical I would say it's an organizational one but we have to address all of them. So how we address this in our project? Let me start by explaining the experimental paradigm behind our project. It essentially relies on injections of monosynaptic retrograde fluorescent tracers into the marmoset cortex. The tracer is injected. It is picked up by axonal terminals. It's transported retrogradally across the entire brain where it labels bodies of individual neurons. And then the brain is sectioned. The sections are some of them are stained, some of them are investigated under fluorescent microscope where locations of individual cell bodies are identified. This provides us information on the direction of a connection because we rely on retrograde connectivity, retrograde axonal transport. This also facilitates subsequent quantification of the results because we have our well defined quantum of connectivity which is single labeled neural body. And so far the body of data that we accumulated comprises approximately 300 injections. And as you can imagine processing this amount of data manually is next to impossible due to extremely laborious nature of such process. Therefore, we have to create some sort of computational solution to facilitate and streamline this process. And the infrastructure behind the project is quite complex. However, if I were to pinpoint the single critical step that would be mapping the experimental data set into the sur-taxic space of the reference brain atlas. And this process consists of a few steps. In the first one, we take the unaligned stack, we reconstruct it into what resembles a single hemisphere of the marmoset brain, which is then further refined by the means of the formable mappings, which result in a reconstruction that is much more smooth in which individual anatomical features are much more pronounced. And this 3D image is then mapped into the sur-taxic atlas with full 3D to 3D mappings. A unique feature of our pipeline is the use of so-called label maps, which are indicated here with the patches of various colors. These are essentially outlines of individual cortical areas drawn manually by new anatomists. And while you may think that including such a step in a process that's supposed to be automatic, it doesn't make really a lot of sense. But on the contrary, it does. It actually is very important because it greatly increases the accuracy of the registration process by forcing the algorithm to overlap corresponding label maps. Sometimes it actually even decreases the time that it takes to compute such mapping. But fundamentally, it's a critical part of our quality assurance procedure, in which we make sure that whatever mapping we calculate, it is biologically relevant that we do not generate any spurious results. Yes, so once we have a recipe for traversing between our experimental dataset and sur-taxic coordinates, we can use it to map locations of individual neurons into the sur-taxic space. But we can do the contrary. We can take the Atlas segmentation and map it back onto our experimental dataset. This gives us information on sur-taxic location of each labeled neuron, as well as allows us to easily determine the number of neurons in each cortical area. And we apply this procedure to over 50 marmoset brains, which translates into almost 150 injections, encompassing a little bit less than the half of the all cortical structures that are defined in the Atlas. And for each of the inter-areal connection that we found, we determined its direction and strength using the fraction of extrinsic-labeled neurons, which is a measure that is commonly used in track tracing to quantify the results of track tracing studies in mouse and macaque. But also to add another dimension to our data and to allow analysis that concern feed-forward and feedback-nature of the connections. We also quantified the fraction of supra-granular neurons in each of these inter-areal connections that we found. This altogether provided us with a comprehensive quantitative data set on retrograde cortical-cortical connectivity, which at this specific moment of time might be the most comprehensive, which is probably going to change in the future. But at this specific moment, that might be the case. But this is still not enough. As I mentioned at the beginning of my presentation, just merely getting the data is not enough. An ideal resource should go well beyond this point. First, by providing researchers with a means of accessing and exploring the data, and secondly by adding additional layers of data to better contextualize the results of the connectivity studies. And to address the first issue, we developed the Marmosetbrain.org portal that you can access anytime. And it has this nice and sleek interface that allows you to pick any of the 143 injections that we upload that and investigate the underlying high-resolution histology, check out the locations of the specific individual labelled neurons, as well as take a look at the overall patterns of connectivity using the flat map view. This interface provides also an access to the connectivity matrix, as you can see here. You can explore it down to the smallest detail and you can customize and adjust the view just to get better view of what you're interested in. There's alternative view that we call the graph view and the purpose of this one is to allow you to better appreciate the spatial distribution of the projections. Each case comes with extensive set of metadata, including comments, sometimes very personal comments from the neuronatomists who actually conducted the surgeries and helped in processing the data. There are a few other widgets like this one where you can grab the brain, spin it around and simply enjoy the view, I believe. So I highly encourage you to go online and check out this portal. Now let's move a little bit behind the scenes. This is the structure behind the portal. I don't want to go through this because it's complex and necessarily complex. But the main point here is that we provide two streams of accessing the data. One through this web browser interface that I just showed you and the second one it's a programmatic way of accessing the data through dedicated application programming interface and therefore you can employ computers to go to the portal, fetch the data and run any calculations you're interested in. A few numbers, as I mentioned, there are like 53 cases. This translates into over seven and a half thousand of individual sections, which in total is like 2.7 terapixels of imaging data. But for the purpose of serving it online we compress them down and it's only 85 gigabytes more or less to put up online. Recently we started monitoring who is visiting out, who is using our portal because that feels interesting, right? And it seems that we have a pretty decent fan base. You may say that that's not really a lot and well just think how many researchers would like to access a data set on structural connectivity in Marmoset on everyday basis. I think there are a few but not that many. However, however, but in terms of how long the users are staying, most of them are bouncing off but this is typical for any website. But we have a group of hardcore users who seem to be really enjoy browsing our website and stay over 10 minutes regularly. In terms of the countries of origin, it's mainly US followed by Japan, then there is Australia, however the internal traffic is excluded. So it's mostly it's mostly Sydney. This is followed by Canada and France and then the rest of the world. Right. As we as we see on the during the first lecture it's very important to be able to study the connectivity versus topographical organization of the cortical areas and we would like to enable these the same kind of analysis in Marmoset. Therefore, we introduced a computational model that allows us to incorporate these factors. It stems from an assumption that an axonal bundle connecting two points in the brain would prefer to choose the shortest or the cheapest way that should as a path and we could formalize it in the in the following way. Let's assume that it's cheap to traverse or easy to traverse through white matter. It's very expensive to go through grey matter and going outside the brain it's impossible. It's no no. And let's impose these conditions and use fast marching method to solve this, numerically solve this problem and the result is going to be the geodesic path and we associate the distance of this geodesic path with a distance. And this formalism allows us to introduce the concept of inter aerial distances which essentially it's a measure that tells us how far apart from each other two cortical cortical areas are. But also we can go deeper and start investigating the distribution of distances for individual labeled cells. What do I mean by that? So here we have an injection that labeled approximately 60,000 neurons and we can go down and compute the distance from each labeled neuron to its injections to the injection site. And this opens new avenues of exploring the various properties of the connectivity patterns versus the distance. But I think it also enables us to study the intrinsic connectivity which do not take into account when investigating, when using the area based connectivity. All together this formalism allows us to gives us a framework for analyzing the connectivity data in the context of, in the spatial context. Now another, another layer of data is the is the neuronal density and we have measured this neuronal density in Marmoset Cortex for all 116 areas in the study, in the recently published study. In essence, except for providing an average neuronal density for each area, we also measured the, the, the neuronal density across cortical profile. Here you have two, two examples of such, such profiles. Yet again this data is available online, it can be used and it's there. A sort of byproduct of this study was a, was a measure of a cortical thickness for specific individual, individuals in which we measured the, the neuronal density. So this is essentially another, another layer of data. Now, now I'd like to show you an example analysis that we can run with the track tracing data from Marmoset that we, that we, that we got. And the tethering hypothesis and the concept of connectivity grading was explained in detail by Daniel just, just few minutes ago, but so I will do not go over this again. And, but one of the implications that, that probably requires highlighting is that, is that due to the cortical expansion and some, the, the connections in the, in the primary areas should be, or, or are expected to be rather compact and short-renched in contrast to the connectivity in, in trans-model or, or as association areas in prefrontal cortex, parietal and temporal cortex. So the question is whether this, this can be actually observed and whether this is phylogenetically conserved. So the first, one of the first studies that explicitly addressed this issue quantitatively used human resting state fMRI and indeed what they found is that the, the connectivity of the primary areas is compact and short-wrench but the further you go from the, from the primary areas, the connectivity is more widespread and more long-wrenched. But this is human resting state fMRI. The follow-up question is whether we can observe it in, in macaque. And indeed a study was conducted, a follow-up study was conducted, and a similar phenomenon was observed. So another question, can, are there any structural underpinnings of this, of this, of this functional observation? And with a, with a, with a available body of data on, on, on macaque tract tracing data from, from, from Prusar Kennedy Lab, the observation confirmed what was, what, what was found, what was known from the functional imaging. And you know where this is going? We took our marmoset tract tracing data and, and conducted very similar analysis. We calculated what's the average, the length of a connection for, for each injection and also for each area. And indeed as you can see when, when we take a look at the primary areas, these are the one outlined in, in, in black. Similar, similar tendency, similar trend emerges and this observation can be quantified and it's quite robust, and it's quite robust. Now there, there were a few regions of the brain which, which caught our attention and we want to conduct more detailed analysis, but then what happened is that we had a, a model user, which is Daniel Margulis and Randy Brugner, model, model user of our website, who went to the website and downloaded the data and use it to make discoveries, totally in independently from us. And what they found is that by arranging the, the areas from primary to trans model, they identified that there is a, there is a peak trans model network, which is highly interconnected. You can see this by, by dark brownish shape. And they concluded that, that marmoset seems to have what they called apex trans model network, which fulfills a condition for a structural homologue of a, to the human default mode network. What does it mean is that, is that, this requires of course further studies, but from the perspective of the resource, we can say that the data, the track tracing data that we published enabled and identification of such, of such homologue. And I think that's the, that's the, that's a great example of how this data can be utilized. Of course, this is, this is just one example of, of, of analysis. I think that the key, key word here is flexibility. I would try to provide, we try to enable the broad range of analysis in two ways by providing multiple layers of data, as well as making it easy to access the data and make a use of it. And the data has been around for 18 months or so and during this time there are several studies that were conducted by, by exclusively, exclusively based on our data or partially based on, on our data. By the way, we also found monosematic connections between auditory cortex and, and, and visual cortex. So also the, the disconfirmed observation that was making in, in macaque monkey. Right. I will, in the last slide I would like to emphasize that this is a highly collaborative project, primarily between, between the Nansky Institute and the Monash University, specifically the laboratory of Professor Marshall Rosa and this is essential where all the experimental work is conducted and in Poland just across the fence at the Nansky Institute all the computational and analytical part is taking place. And of course we collaborate with many other groups across the world including the, the Japanese brain, brain minds program and this project was possible due to multiple sources of founding and one of them is being the INCF seed founding which may, feels me obliged to stand here in front of you and just let you know what, what do you guys got in, in return. And I think that, that would be almost the last slide. Yes, so I invite you to visit my demonstration today and, and discuss either the connectivity or perhaps some other, other aspects of, of a Marmoset brain anatomy. Thank you.