 Thank you so much Jota, and now we'll do Samantha Estudis from University of Toronto. And she'll be telling us about investigating then right self avoidance using computational analysis of time-lapse 3D images. So Samantha, is it okay for my desk? Sure. So, I'm Samantha Inastavis. I'm a fourth year PhD student in Julia Lefebvre's lab at the University of Toronto. And I'm going to talk today about one of my projects that's using computational methods of analysis to applying that to time-lapse images to try to understand how dendrites are able to develop their specific shapes. So for context, because I think people who do models sometimes forget, what I'm showing here is just one subtype of neuron that's been sparsely labeled. So it gives you an idea of how incredibly dense these neural tissues are. And within these neural tissues, multiple different subtypes need to share this space and develop the specific morphology they need for their specific function. So what I'm showing you here are six different neural subtypes, and you can appreciate how terribly variable their morphologies are. And what's really unknown in development is what are the cellular mechanisms that occur in order to result in these morphologies. And if you take the starburst, and these morphologies are known to be really specific to, implicated in the function of these neurons. So if you take the starburst-american cell, which is the model system that I'm working with, it's a relatively planar cell within the retina. Past work in our lab has shown that with genetic methods, if you perturb this morphology, it inhibits the ability of the cell for specific function within the circuit, which is to tune the direction selectivity within motion detection, so inability to detect motion. But despite this mature cell having this beautifully radial morphology with minimal overlaps, known as dendrite self-avoidance, early in development, you can see this time course of development approximately from post-anil day one to six. You can see that this is not the case. In development, their dendrites are very complex, and they create this really dense web that has these self-contacts. So this is an example of a P3 starburst-american cell. So in their adult morphologies, they have these really radial morphologies, but in this context, you can see it's this really dense web, and the take-home message here is that this web is forming self-contacts. There are contacts occurring within these dendrites, within these self-dendrites. So the question I'm really trying to understand is how do you get from this morphology to the mature morphology, and what are these cellular mechanisms? So to try to understand this problem, I first went in vitro. So here you're looking at a cultured neuron. So this neuron is taken out of that environmental context, and it's really given its space to develop this morphology. So it's a single cell neuron, and what you notice are growth cones at the distal tip. So classical work knows that in axon guidance, these growth cones emerge. In this example, it's dendritic growth cones. But moreover, there's these dendrite bridges. So outlined in green, you can see when I'm defining as a short orthogonal projection, which is a dendritic bridge, and this is the point of self-contact. And it forms this really ladder-like or web-like structure. So this is in vitro. So then I looked in vivo to see do these morphologies come out in vivo. So again, this is in that very dense neural tissue, and you can see outlined in green and pink the growth cones at the distal tip, as well as the dendritic bridges. So this allowed me to develop my model of dendrite self-avoidance or dendrite development, where you have growth cones that have multiple bifurcations at acute angles in contrast to dendritic bridges that project at orthogonal angles, and as well the growth cones are at the distal tips, whereas the bridges are more proximal. But all of the static analysis is limited. So what it can't tell you are the temporal dynamics. So how are these morphologies changing throughout development, and if you track these morphologies throughout development, what is their role in leading to that final morphology? So to answer this question, I'm using time-lapse imaging so I can follow these morphological events in real-time. And what I'm going to show you is one experiment, an example of one experiment. So this is a P4 starburst amocrine cell, and I'll just play it through. So I hope you can appreciate here how much remodeling is occurring within the neuron. And although there are points that are being extensively remodeled, there are points that are stabilized, that I hypothesize will become the primary branches that are stabilized. So just to show you some frames within that video, there was a growth cone at a distal tip, and the time course of these growth cones is approximately an hour. And this is in contrast to the bridges, where they have much faster temporal dynamics. So in dark green are the invariant branches that I hypothesize will be stabilized. And in light green are the very transient and rapid dynamics of these dendrite bridges. And this video will loop one time. So you can see this is a 15-minute time step, and although the green branches aren't extensively remodeled, the dendrite bridges are extensively remodeled. So everything I've shown you up until here has been very qualitative. So this leads me to my hypothesis adding to my model of dendrite development, the time courses, and as well the role and development where growth cones are responsible for radial outgrowth, whereas the bridges are responsible for dendritic spacing. So the question here is, everything I've been told you has been qualitative features, and typically you scroll through stacks and a graduate student pulls out features and tries to quantify them at relatively low ends. So how can you use computational methods analysis to extract quantitative features from these time-lapse volumes? So typically if you're trying to extract morphological features, you start with the neuron trace. So what I'm showing you here is that same experiment, and then this neuron trace. And neuron traces are the most widely used open source file format for these neuron traces is the SWC file. But one main caveat in doing specific types of analysis is that SWC files are a ball and stick representation of your neuron tree. But the main assumption here is that each individual node within the tree cannot have more than one parent. So because of this assumption you cannot directly encode closed loop structures. So how do you get around this problem? So the challenge here is you need to create faithful neuron traces and when you have very complex cells that are early in development that are near diffraction limited, how are you able to create and validate these neuron traces? And then once you have these nodes of really complex neurons, upwards of a thousand nodes, how are you able to link these nodes through time in order to follow their development going into that fourth dimension? And these morphologies are known to contain closed loop structures. So if your file type can't even encode the morphology you're looking for, how can you answer your questions? So the solution that I'm proposing and I'm working on currently is an analysis pipeline that's using various open source and what will soon become an open source program, the dynamo program that's still in development, to address these problems. So using Vaw3D to create automatic traces of every time point and then 3D registration with Dynamo and importing the output of Dynamo into Python so you can extract your geometrically defined features. So the acute angles or the orthogonal projections. So the pipeline starts with this is created with Vaw3D. This is a three dimensional visualization of one of my experiments. And you can take this as just one time point and input that into Vaw3D. And this is an example of Vaw3D has many features. But one of the dropdowns it has is all of these neuron tracing algorithms and a lot of them are developed by third parties. It's not all of the Vaw3D team. But this trace that I'm going to show you which was the output of the Vaw3D tracing was created by Vaw3D's own algorithm here. So you have a neuron trace and it's an SWC file that can't contain closed loop structures. And as well all of these nodes are not linked through time. So this is my way of visualizing that you have all of these nodes which are supposed to be the same neuron developing through time. So how do you link all of these nodes? And with my neurons specifically there's about 800 to 1200 nodes and all of those nodes are connected by segments. So with Dynamo. Dynamo is a program that came out a MATLAB version about 10 years ago I think but new efforts have been going into converting their MATLAB version into a Python open source version. And I'm currently working with a student in Kurt Haas' lab at UBC to get the specific feature of Dynamo although it does tracing as well. They're auto registration. So the ability to follow these nodes and track the development of these morphologies through time. So then once I have my nodes linked through time then I can export this and what I'm showing you here is what all of these 2D projection of these nodes look like in Python. And then in Python I can I have my geometrically defined features and then I can extract them and in that way analyze the whole arbor at once. This is very much ongoing. It will be the next 2 years with my PhD. So what I've told you is that static analysis can't give you the whole picture. Timelapse imaging can reveal dynamic features that are relevant in development and one of the things that it revealed is that these closed loop structures are relevant especially during development but the data structures that are open source and often used don't have the ability to accurately encode these morphologies possibly obscuring these biological phenomena and biasing our interpretation. So what I'm really looking for is an open source file format that's able to accurately represent these morphologies these early in development morphologies i.e. the closed loops and my current workaround which isn't a direct solution is to build this pipeline starting with ball 3D dynamo for the auto registration and then Python. And I just want to say that I'm giving a poster so in this brief talk about my model of dendrite self-avoidance in development our lab also works with a set of molecules exploring this really interesting idea of neuron identity so if you want to come check out my poster I can give you sort of a broader depth of this model of self-avoidance everything from the molecular up to circuit so 2 or 3 cell circuit level Thank you How automated is the process of neuron tracing? Like if you wanted to trace a single neuron ever development about how much time would that take manually? Oh manually Or I mean, sorry, with the pipeline how much time would it still be required manually? For an experiment with 10 to 30 time points probably 10 seconds per time point How much time, totally manually? Manually? Yeah Like 3 undergrads a week probably Thanks Yeah, typically these are not traces like you have to get at your ground truth so that's one of the things how do you validate automatic traces and typically it starts with a manual reconstruction which tends to be a team of summer undergraduate students Which file formats didn't work? I imagine it's been a while since I looked at some of the 3D branching ones but I'd imagine that you know explicitly support closed loop that you could just interpret 2 points in a branch having the same coordinates as being closed Yeah, so that's eventually, yeah, so that's definitely what I will do in Python Yeah, but it's definitely a problem within encoding these structures and for me I'm specifically looking for closed loop structures because I previously developed my hypothesis of branching where you're doing an automatic trace and you're looking at something else and it will automatically miss these contact points So you mentioned that the cross links space the dendrites as they go out so is there some relationship between the number of cross links and where they are between the dendrites because if they're producing some sort of tension you couldn't have sort of 2 on one side and 1 on the other otherwise everything would get messed up So my hypothesis, that's something that I want to investigate in my analysis is where do these bridges occur on the dendrite as well as relative to other bridges I see them proximally closer to the SOMA and less so at the distal edges but that's definitely something that I'm interested in figuring out