 Okay, my name is Daniel. I'm a PhD student in the Department of Neuroscience here at Karolinsk Institutes And I'm originally trained as a clinical psychologist, but these days I do more like molecular biology So just to get a feeling it would also maybe be good to sort of get like Do people here have any sort of like background in molecular biology or like no idea about the general biology Yeah, okay. What what kind of like background? Okay, okay Yeah, but then I'm going to go over those parts a little bit more in detail and like not not dwell on the details the drug like Show what can be done with molecular biology these days So this is all in mice and I'm going to talk essentially about what I refer to as a never anatomical information system and The idea behind never anatomical information system is to be able to create essentially maps of the brain And when I talk about maps of the brain is digital maps So there's a perfect analogy between a never anatomical information system and Geographical information system that for example Google Maps use and other mapping surfaces So the idea is that like individual researchers can create their own whole scale Whole brain representations of their data So you do not have to rely on Allen Institute anymore. You can essentially generate your own whole brain data With a brain Atlas from a single like laptop computer like this and a microscope It works on a cell phone it works essentially there's no requirements. So The software suit is actually composed of many two parts. So there is a Back-end application and desktop application, which is essentially image processing and image analysis That's mostly coded in C++ and then The user integrates with it through an R package called the whole brain simply So you have all the plotting function now On top of that, it's a front-end server as well, which is called the whole open brain map of work Which is also the whole never anatomical information system in terms of like brain at us that is open source and rely some gets to be updated So that's basically two Different things so more into detail when we talk about Geographical information system. So if you come from an fMRI by background This might not really make sense to you because essentially an fMRI there is only one kind of like Computer graphics and that's raster graphics You're representing an object, but you can also do a vector representation. So if you look at like Google Maps That's a vector representation for most kind of informatic system when you have read a high resolution spatial resolution That's what you want in them when you do your computation and when you do your analysis The raster information. It's nice like a satellite image You can see certain things and you can use it in qualitative ways, but it's not the end product of complication So what we essentially want to be able to do in the brain is having these kind of information system And how are we going to develop systems where we can actually get to this point? So you see red souce vague That's where my lab is and the bagpipers. I don't know if people are familiar here with it It's a beer place where I think it's like 80% postdocs from Kerensky Institute. There are inhabitants there Through the Google Maps interface I can query about pathways through from this point to this point I get a lot of information I see that for example There is some traffic jam here and I get an estimate of the time if I take it by car if I walk and I can See different pathways. We don't really have anything sort of the equipment when it comes to the brain Look at the Alan brain at last. It's essentially a huge qualitative notebook There the quantity data is so low resolution and for data integrity But it cannot really be used in this kind of like various other way and the resolution and the data system are not really in place there So what do we need in order to get this kind of like? information system Yeah, just as a point like rastrographic is the norm in neuroscience these days and even like the labs that really tried to push it In terms of like freeman and colleagues at the Amelia farm. This is from the recent nature or no It's a whole brain analysis of calcium imaging in the zebrafish embryo So you can basically see here that you can map different behaviors towards different pixels But you also see that the pixel Analysis creates a lot of artifacts and you have actually a cellular resolution in the image, but you're not using that information So this just creates parametric maps and This is like the norm in neuroscience and if you look at all the C++ Different like libraries. They are geared towards this kind of analysis based on fmri and it's Different coding regions that started way back in the 90s and it has been with us Yeah, so why this this doesn't really work for the new kind of data that were that is generated because we're starting doing Imaging on the subcellular level and we do it through across the whole brain So if we compare for example here in neuroscience like a standard data set could be like this and here was there a midiolateral Ventral this is kind of like a pixel dimensions. We're talking about When we're doing mouse brain images, we're talking about like sliced tissue Hundred sections if you do clarity and those kind of techniques. We're talking about 2,000 Sections optical section and we're talking about 20,000 times 13,000 pixels and that kind of dimension So what kind of vector graphics are there in neuroscience these days? Well, it's actually very little. It was an attempt by By currency enga To essentially create a lot of scaling variant atlases with vector graphics But as you see the resolution is really poor because all they did is taking a pixel atlas and convert it into vector graphics I mean you starting by very low resolution and you end up with even poor resolution when you do that approach You have to rebuild the whole atlas and start with a vector Graphic revolution to start with And the basic argument is also like this, but this they don't have to tell you in this kind of crowd-setting It's about like having transformations that are in variants So for example in raster graphics, I can do a transformation I'm gonna transform it back and there is a lot of information that's completely lost Those kind of things never happen in vector graphics So what we essentially want to do is build this kind of system Never anatomical information system where we have the object We want to represent in this case a mouse brain and we slice it, right? So we put it on a shock board and then we slice it in very very thin micrometer slices and we image the slices and then we want to map it back into a vector Representation and do it through a lot of sections. We also want to map features in the game So with analogy of like the Google Maps that you have different regions Like cell bodies, you know exactly their location and all the information can be clear So how do we create a Staley-Marion representation so their current effort the the initial effort would for example Alan reference Atlas was originally drawn on missile plates like By hand and that was actually a vector representation, but it was just a hundred sections or so Later on when the computer scientists came in they started doing like basically fMRI research They just melded everything into like an area representation. So you have this like 25 micrometers No, nice a topic resolution. So each voxel you can query What region does this voxel belong to it belongs to with this identity number, which is an unsigned integer 32 bit That's a very inconvenient format for streaming on a web It's a very inconvenient format to work with in terms of gits. It doesn't really work It works for a lot of things if you take it and push it through an fMRI pipeline So currently what Alan is trying to do is like they're trying to redraw that that's with polygons That's also a bad idea because you will get good resolution But in the future we want subcellular resolution So you're going to start seeing these jagged lines and you constantly have to read or the Atlas a better representation is to do it with these lines So these points you can have in different formats I mean one very common format on the web based on the XML based format with this The good part with having representation in this kind of format is you can treat it with it So if someone there is a discussion about me dissenting a definition of an anatomical region You can come to an agreement by using this then I mean business as usual The important thing here you don't have an array of data that is like on the order of several hundred megabytes binary data here you have like something that's readable And So basically this is the regional and reference that's like when you download it and try to do a certain computations on it you have like this I Forced isotropic resolution because the cutting is in 100 micrometer section But you have essentially an array of data where each of these voxels have an Unique identity number depending on what region in the brain it is What you're not trying to do with Alan is Essentially creating polygon data where they draw polygons on top of this and the final reading is in terms of polygon measures What we instead try to do is actually creating nerve surfaces, which is essentially 3d representation of Beespline it's a non-uniform rational beesplines. It's what's using cat softwares For example most likely the shares here the table all these things are done with nerve surfaces at some point They also makes computations really really simple when terms of for example computing intersections with a plane so you can like redefine that plus depending on how you want to cut the brain And as I already said since you have the format in some sort of like XML Jason kind of format You can then do an aftermath of social construction So you can have like different people don't agree on a certain region I mean certain regions were defined by some old guy 100 years ago By looking for a microscope things evolved and maybe want to have a social construction how we agree on the anatomical boundaries So the boundaries are also highly questionable. So this can easily be done with with git So there's also this part where maybe nor anatomy shouldn't really be defined by social construction I mean with the current molecular tools such as CRISPR engineering You can actually do lineage tracing and trace the development from a single cell to like the whole Organize throughout development, and I think in the future That's how we're going to define your anatomy, and that's how we want to do But importantly you can apply the same kind of git framework to the Never-anatomical boundaries when they come when they happen and do margin pull request and Get at that sort of like that So what can we do with this kind of framework? So here is an example section of a coronal section of the brain. This is around like Middle of the brain. So here you see the stratum into hemisphere for you. How many people are familiar with mouse brain? No, it's just like yeah, you recognize this image where it is in brain Okay, this is straight them here. You see the cortex here We have genetically labeled a population in the stratum. This is a D1 positive neurons They have the D1 dopamine one receptor Express and we label them with GFP But we also have engineered a virus that basically jumps monosynaptically and also labels the monosynaptic Direct input from cortex. So we're looking at the corticostreatal pathway here This is how we can get the connectivity tracing throughout the brain genetically So through the framework we can then fit Atlas and also segment out all the labeled features This is roughly 1000 to 2000 or somewhere there But you also see in the image that we have labeling of axons For example here, we see the contralateral input to the this side of the straighten So we can basically trace where are the fibers crossing the corpus callosum at what they and this we do by Segmenting the processes as well. So we have a representation that more looks like a Google Maps kind of framework Automatically we get the quantification of like how many cell count if the latter all of this can be queried through our package You can plot it here. I have like how many fibers you have in the different regions We see that most of the fibers are not unsurprising An ipsilateral side to the injection and they are in the codotifatoma the straight and then the second largest corpus callosum Most cells here are in the stratum. That's the injection side And then we see the most abundant input is from layer five in motor clinics So for example when we are mapped it back to Atlas, which is the reference Atlas This is how the data can look from a single section and this is then how it looks on the web So this is the intermediate step on the computation to get the fiber processes And then you can represent it like the highway you saw in the Google Maps You also then have the cell bodies and everything and everything can be queried in a uniform reference Atlas for multiple different So when we then do it across the brain Everyone gets that approach. This is the kind of black data we start with like essentially taking sections and then Transforming them to the reference Atlas and building it up a single brain in 3d So these are all 80,000 neurons that sense input to 2,000 neurons in graphonucleus These are certain nugget neurons. So these are all the monosynaptic neurons that directly control graphonucleus and the production That's how that can look So if you then so this is how the Atlas looks The reference Atlas, so this is the main starting page of openbrainmap.org So you can see here immediately that we can like zoom in and have a really high resolution detail We have a small Small inset here where we can quickly move across the image and have both resolutions levels Simultaneously so we can get a good feeling of our image and move around so for example here We are in hypothalamus and when you are at a certain resolution I would also see that the boundaries are defined And you can see that this is the boundary to MPO the LPO without you even like having to like Look at the centroid of that So that's how it looks You can then query that last for example, I can put markers here We can both draw if you want to for example create new brain regions You don't agree on this kind of definition of straight and you think this is a good definition of straighten You can draw it then someone else can go in and essentially say like no you're wrong I think actually we should define it more like this and change it Here the markers Basically the markers here is that I put marks a single location in the brain I both get pixel coordinates They are useless because microscopes you can do imaging and different pixel resolutions all of it This is completely useless unless you're really interested in developing image processing algorithm More important you get still tactic coordinates and these are the coordinates that I use when I do Injections in the brain when I do surgery put in probes This tells me a millimeter where I'm at the brain It's in from anterior posterior Meteor lateral and then dorsal ventral so I know exactly where to go in the brain through this coordinate system I can then query what are the most abundant genes in this region compared to the rest of the brain And I get a list of two thousand most abundant genes sorted according on the server And now we're going to add all the like inputs and outputs as well. So you can query from one location to the other For example, I can take up simultaneous in the Regions like that and see the difference You also have a different set of tools for example if you want to quantify your injection site You see in real time the radius is like, yeah, 500 micrometer. I don't know radius there So all the measurements are in real time you can draw Can your surges for example go distances? All of this can be done saved and download through this button You get it as text file which you can import into our directly in your console And you get the same sort of like your metric You can search that class and you have the same sort of like hierarchy here and search through region And also this kind of information. So this is where we want to similar to like Google Maps Unlike Google Maps, which is like a single pane the brain is multiple pain So we can move throughout the brain for example, if you know a surgery, you know I'm teriopostero medialatron dorsal ventral and this in millimeters. So let's move maybe minus two Then one millimeter to the right and then let's move maybe two millimeter to down in the brain And then it places a market exactly at that coordinate and we're taking here now to the Dental gyros the cell layer of Dental gyros in the hippocampus So you have the whole brain at last like this rather than multiple windows and a lot of light to differ in So The segmentation are words are based on wavelength transforms And the reason for that is because we want to define biological features Not by how they're labeled or anything like that, but rather the space to occupy So here you see for example an original image tile with this cortical Cells label you see that the dendrites here and by filtering the image in these kind of steps we can segment out the dendrites and the cell bodies in different way that's a platform What this enables us in which one now going to get both the processes axons dendrites and Cell bodies and this is just to show the signal a different key So the signal to noise ratio for neurons the cell bodies It's high as the D3 here the scale and this also has to correspond to about 10 micrometer Which is the size of our cell body a thickness of an apical dendrite is normally like two micrometer And that's also what we see that the peak of the single intensity for processes are around there And also what we see is that the GFP is most localized to the soma So we see this as really really high intensity signal compared to the processes It works in different kind of microscopes epipressants confocal microscopy and light sheet microscopy So light sheet microscopy enables us to do all of this in 3d as well So we can go through an optical stack and then segment out everything in 3d registration algorithms Since we work with thin plate-based lines And we get the contour automatically. This is actually the idea is very similar to snapchat actually We don't want to do this sort of like information Pixel by pixel shaking the information intensity and then optimizing that that mostly doesn't never work in this kind of data Because your section in the tissue every brain section is going to look very individual with so it's better actually to get a set of correspondence points between your atlas Here and your regional side section and this is equivalent to essentially what Here we generate the correspondence points is essentially what snapchat is doing but Snapshot you have several years of research into facial features to detect them But you detect these kind of facial features and then you can get a millisecond registration and run it on a cell phone If you get equivalent for these kind of like really large-scale tiff images We can have millisecond registration on like 500 megabyte tiff images and have 2,000 of them You can do a registration of a brain within a couple of sections in order to get the data We need a lot of brain images in this kind of format on our server roughly 2,000 images per chronal plane That's the basic idea here you can see how the framework looks on a single on your raw data So you can see what kind of resolution we're talking about that this kind of data have you see that we see single cell bodies You also see that this will work on your cell phone. There is no like so down I can both see all the region bounders that are fitted here. You see perfectly that this is a five, right? In Cortex and then I also can get my cell bodies here. So these are the 2,000 cell bodies Each cell body here. I can then click on anything I get the identity where it is and also since we're doing our stuff like that in your data frame What role we are in so this is how you look like at your output how it's working You can move around and sort of like look how accurate it is and you have the same set of tools here More importantly since we are in both the pixel image System and the still tactic system. I can actually plan my soldiers within this Tissue section. So if you look at the mouse when I'm moving around I see millimeter where I'm at in the still tactic frame seems in this last of war It's an image. What can we do with this? We can trace connectivity. I'm just going to do this very quickly We can trace connectivity throughout the brain with genetically engineer viruses. So rabies virus Everyone are familiar with the idea of a rabies virus. The rabies virus is a single-stranded RNA virus It's very aggressive in a sense that like if you get bitten by a rabies-impacted dog will travel Exclusively in the retrograde direction. So you will not travel anterior grade just retrograde And it will travel synapse by synapse up to the brain and then jump across synapses like this in a retrograde fashion The rabies virus is evolutionary very very primitive in the sense that you only have five genes and it was discovered a couple of Years ago that if you know knock down one of the genes here the g which is glycoprotein gene Which is the glycoprotein that coats the the matrix envelope If you do that the rabies virus cannot jump synapses it will Get into one neuron and then it cannot jump across synapses So the idea is what you actually do then is you knock out or you remove the Glycoprotein instead you insert a GFP a fluorescent protein in there It's yellow fish fluorescent proteins so we can visualize it and now we have a virus that can enter a cell expressed GFP, but it cannot jump synapses. So how can we then Get the proper back again of the jumping synapses? Well, we express with a different set of viruses AV viruses Which are single strident DNA viruses we express in a Cree-dependent manner meaning we express it in individual cell types like dopamine neurons only or this kind of neurons We express the glycoprotein and then we will actually get a rabies virus that have it The glycoprotein in trans so it will sort of like coat itself with the glycoprotein We were able to jump once in apps But now we will find itself in a cell that doesn't have the glycoprotein anymore and it will stop there So that's how you get a genetic system where you can label Monosynaptic inputs and it's in in one direction retrograde And you can do it in a cell type specific manner So this is how you get this kind of like connectivity data Which we need in order to create this kind of queer database it where I can ask the same kind of question of like How do I get to bagpipers from here? I want to know if I'm in a particular brain region. What is the fastest way to? Synapse by synapse to go from here to here That's how you get that kind of data This is just like how it enters the Daxon terminal and then travels retrograde layer up into the cell body and expresses there These are the helper viruses where we label the glycoprotein as well But then also label with the fluorescent marker the Primary population so we can then actually get what are post-synaptic and what are presynaptic cells and count them So we did this in different cre lines and here you see essentially Several brains reconstructed and here we see all the 30. What is it? 40 50,000 cells that receive in that gives input to about 2000 neurons in the dorsus rate them and these are D1 neurons We see D2 neurons Then we can also trace connectivity in a step up like in motor cortex. What the motor cortex gives input to motor cortex? Well, we do a cancunus creed which is roughly excitatory neurons and then gabarra itself you can see the difference and Colonary into noras in straight them and through this we can build this kind of like Network of how the cells are wired in the brain So that's like the step one the step two is then looking at function of the brain regions We're here. We're giving a mouse. I'm just quickly show this video This is a very very good task. So Then it's an IP injection of either a sedan vehicle or Cookin so you have an extremely strong behavioral effect the mouse on the right that's okay So you have this enormous strong Locomotoric output and then we can look at activity markers like seafoas and count them in the brain like this and count Intensive of the protein levels as well like this and Then we can actually look so this is the mouse that you saw in the video with the cocaine and here is without the cocaine the whole brain activity Representation we can see for example in different brain regions how the states go up and down and then this is across animals We think can and correlated with the velocity the animal was moving at We cannot figure out like what kind of coding do we have in different brain regions for example here We can figure out the decoding scheme for orbital cortex in terms of which send inputs to the dorsus straight on is that it Works as an input game in terms of activity pattern Next what we're going to do is verify of course that this anatomical whole brain tracing works and the way we do that is essentially through This nice assay where the animal moves around in a virtual Environment what you have here is the equivalent of the old Mouse that weren't optical so it's this ball and then you have sensor on it So you get all the movement with millisecond precision when like slightest movement when the animal is moving with the good part about that is of course that We can record in the brain And we can then also catarise the animal and do Direct Infusions into the blood vein of cocaine and not getting this delayed where it goes through the liver So here you can see the cocaine injection and here you see the cumulative distance ran on this ball You see me laughter 20 second you get the effect of cocaine unlock a motion and Simultaneously in this assay you can of course give rewards and stuff like that to figure out What is the coding scheme for different neurons in terms of reward properties and and local motor effect That's how it works. Other things you can do is cancel types It also works on RNA it's got here is an RNA molecule and you can then classify cells according to where they are But these are the kind of like brain maps we're starting to build. So, yeah So questions Yeah, it should also end with if you're about to finish your PhD or anything else We started up a company now and we got funding for it So we will this summer start hiring engineers preferably we're looking for One developer with knowledge of C++ the CPU programming Open CV whips these kind of things is nice And you're working on the back end then and then we need a JavaScript and go lang mongo DB these kind of things So if you feel that this is you or you know someone and are interested in like developing this Google Maps for the brain One man who may