Growing Living Rat Neurons To Play... DOOM?

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
this video is brought to you by Squarespace inside your head is one of the greatest computers in the universe which is surprising because I've seen your browser history and yes you do need to eat more fiber but generally the human mind is one of the most capable learning machines we know of composed of 86 billion neurons and another 90 odd billion glia and other cells with wires forming trillions of connections and while humans do have very impressive brains even small numbers of neurons are capable of learning complex behaviors a few hundred can control microscopic animals like tardigrades a few tens of thousands can control insects billions and you've got a mind capable of Reason logic math and art however there's a problem it's just that the brain is made out of meat what what does the thinking then the brain does the thinking the meat that means we can't use all that computing power unless it's attached to more meat so until our meat robot is finished we can't do much and no that's not a euphemism we're building a meat robot if you want to see us build the first pieces of it there's a link below but what if we could grow the neurons directly attached to a computer from the start then we might be able to use them as a tool to control things for us living neurons are a lot more capable of learning than the simulated ones we use for AI at least in theory so we're trying to do just that this weird looking device is a prototype neural multi-electrode array it's got a bunch of microscopic electrodes in it with tens of thousands of living brain cells growing on top and today we're going to be talking about how we made it and how to grow neurons but those electrodes aren't just decorative this array is designed to be connected to a computer and through that connection the neurons are being trained to play Doom well that probably sounds impossible it's actually anything but we're building on Decades of Neuroscience and previous examples of groups using living neurons to control things for example several years ago a group from the University of Florida trained a dish of rat neurons to fly a plane in a simulation and where a lot of the inspiration for this project the neurons had to control the pitch and role of the aircraft and keep it steady which it handled well even when turbulence or wind was introduced others have used living neurons to control little robots this one from Reading university in the UK uses ultrasonic distance sensors to tell the array of living neurons about the robot's environment and let it decide what to do about it which manifests naturally as active Hazard avoidance most recently a group got a batch of neurons to play Pong which may sound less impressive than the other examples but the way they handled their sensory input into the neurons was really clever and is where we're taking a lot of the inspiration in our control scheme though we're going to get to that in a second point is neurons are little prediction machines so if you design the stimulus you feed them correctly you can make them learn to control anything at least in theory I know Kung Fu now the reason we chose Doom is because it's a sufficiently complex task to be impressive but it can be deconstructed into a simple signal problem if you set it up properly Doom only looks 3D in reality it's much closer to this the character is basically just an arrow enemies are circles hazards are just shapes all on a 2d surface the game just renders it to look 3D but there's no real difference you can't actually tilt the camera the gun just shoots anything in the center of the screen regardless of elevation in order to play the game the array only needs to be able to do a few things move forwards or backwards turn left or right fire the gun change weapons and interact with buttons and objects as sensory input the neurons need to know if the character is pointing at an enemy approaching a hazard near something to interact with or pick up and if they're near a walled they also need to know if the clip of the gun is empty and their health status the other thing to know is that neural runs like some types of electrical noise and dislike others repetitive predictable noise like a simple tone or repeating pulses they like and will cause them to spike more chaotic unpredictable noise they dislike and will actively try and minimize let's look at one example and see how locating an enemy would work with only two electrodes there's a way to encode where an enemy is relative to the player I think the easiest way to understand this is to experience it so here's what I want you to do if you're watching this with headphones on already congratulations you don't need to do anything but if you're not pause the video and grab headphones don't worry we'll wait while you do one eternity later all right are you back okay good I'm going to put you into the perspective of the array listen to the tones as the player turns we may actually use a third electrode as an added stimulus for when an enemy is straight ahead but I think you can see how it's fairly easy to figure out where the target is just with two there's a natural point when the player is aimed directly at the enemy where the tones line up for multiple enemies we'll just program it to prioritize whichever enemy it's closest to start with but eventually a much larger array could include a lot more stimulus to track multiple enemies at once this isn't actually all that different from how some missiles work only the thing deciding what to do about the tones is living neurons you need to remember to feed though unlike a missile the array doesn't know where it is or for that matter where it isn't it's just listening to electrical Jazz and trying to guess the next note if we did the same basic thing for the other parameters the array will have a pretty decent amount of information to work with to be able to navigate around the way we know what the neurons want to do is to pick a group of electrodes on the array and then arbitrarily set them as an action when the amount of neurons in that area producing electrical spikes increases Beyond some threshold we'll set we count that as the neuron's choice of action there will likely be a lot of tuning to get the levels of this stuff set properly and it'll also probably take time for the neurons to learn to respond and update their wiring but this same basic concept can be used to break down the entire game into a series of signals the neurons can learn to interpret and act on as a final touch to help train the neurons further and help encourage them to play the game every time they get a kill every stimulation electrode will play a pleasant tone if they get killed Say by walking into a hazard or losing Health every stimulator electrode will play an annoying tone for a moment so yes we are trying to Pavlov the neurons but jury's still out on whether or not the pleasant stimulus is the equivalent to a treat or an orgasm the little squish up to the neurons they'll need to figure out how to play the game and score points purely from the incoming signals but to them they'll have no idea that's what they're controlling they're just trying to anticipate the noise coming in and then adjust if they're wrong we've set it up in such a way to avoid situations where the array can find ways to not play the game now if we ever want to be able to test this we should probably get building so let's now look at where we stand on making this real this video is actually the second part in a series as we've been working on this for quite a while there are a few basic pieces to the project the first is of course the hardware this includes the array the incubator to keep the cells alive and the electronics for connecting the array to the computer the next part of the project is the neurons that we grow in the arrays these are living cells and require regular feeding and careful tending to keep alive and finally there's the software that will work as the bridge to interpret the signals that we collect from the growing neurons and feed signals into the array for the neurons to interpret and process as you can imagine that's a lot of things that need to work properly for it to actually function we realize that trying to do the whole project in one chunk is way too much to chew so we're breaking it into smaller parts the critical path has three parts growing neurons properly getting signals out and then putting signals in to control them but there are some side quests too and a final boss but we're going to talk about that near the end of the video today Beyond just an update we're going to be focusing on the first part growing neurons something that became very clear from Reading seemingly endless numbers of papers on this topic is that nobody actually knows the best way to grow them literally every company and academic lab does it differently and while there are some commonalities the only way we were going to figure out which particular witch's brew works best and how delicate the little bastards actually are was to start trying to grow them and see what works and I'm happy to report that we have now stress tested a protocol that seems like it works really consistently but we'll come back to that shortly because before we can grow the neurons we need to talk about what we're going to grow them in when we buy neurons they usually come Frozen on dry ice as a tube of about 1 million live Frozen cells last time we used stem cell derived human neurons but at least for the next few experiments we're going to be going with cortical rat neurons this is for the sake of cost availability and the fact that cortical neurons from a rat can learn well enough there I have absorbed all the knowledge in this computer database since neurons don't divide however many cells you put into your culture dish or well is the maximum number that will ever be in that dish and once they're growing you can't mess with them or move them to another dish so you need to plan your experiments out really carefully in advance there were a few questions that I wanted to answer with this set of experiments and since every time we buy and grow neurons is really expensive it's important to make sure that we answer as many questions with each attempt as possible and of course this is also where our wonderful patrons and other supporters come in to help feed our monstrosities and get those answers speaking of which I tend to buy my monster channel from an online supplier but what if you've got a great idea for a new blend with cinnamon spice and everything nice then you're going to need a website to sell it and for that look no further than the Fantastic sponsor of this video Squarespace Squarespace is the amazing all-in-one platform that makes setting up a website incredibly easy featuring Best in Class website templates that let you customize every detail and there is a huge collection of extensions to add even more functionality to your site blog about your brain dishes sell Monster muffins or exhibit experiments no matter the use there's a template that'll fit your needs so head to squarespace.com to start your free trial and when you're ready to launch your site go to squarespace.com the thought Emporium to save 10 off your first purchase of a website or domain now let's get back to science the question I wanted the answer to the most was the correct seeding density because there was a huge amount of variation in the literature and if the density is wrong that is if there's either too many or too few neurons they they won't grow and spread and be able to learn properly and we actually saw the effect of this in the experiments that you'll see in a second so we'll be growing a normal 12 well plate of neurons to figure out the correct density using the following amount of cells in each well that should leave enough neurons left for a few tests of the new arrays speaking of which there have been some big changes since last time for this array prototype we tried to be clever by ordering the new array design on clear pet film the thinking was that if we just adhere the clear pet to a glass slide we get a see-through set of electrodes and it's cheap and fast to make this didn't work as planned but we'll get to that this design was created by our team member Jonah who you might have seen pop up in some recent videos the array features 46 electrodes with 120 Micron tips as well as two reference electrodes on the outside and it uses simple ribbon connectors to connect to the electrodes but clear pet doesn't work the way normal pcbs do and it's where we ran into trouble the way they protect the traces is by sandwiching the metal between a second piece of pet and for electrical connections they just cut holes in the film but the electrodes that we ordered were so small they couldn't cut holes and glued the top layer on properly so the company just decided to cut one big hole and send us that all told not thrilled this means huge stretches of the electrodes are exposed so any signals we get couldn't be localized anymore and if the electrodes are too big you just can't detect anything because all the electrical noise from all the different neurons gets averaged out over such a large area we initially attempted to fix this by covering the electrodes in some of the biosafe epoxy and then trying to use our new fiber laser to expose the electrode tips but we can never get this to work so for today we're just going to use the arrays coded in epoxy and next time we're going to use traditional PCB materials at the cost of being able to see the neurons otherwise we'll just get commercially made arrays which I'm not opposed to controlling neurons is the part of the project I'm actually interested in not really making neuron arrays that said if you you work for a company that makes electrode arrays or neural recording equipment and want to send us stuff feel free to reach out we would be happy to use your system for the video to finish off the arrays we epoxied the pet film onto glass microscope slides then once it was cured did the same thing as last time and used the tops of Falcon tubes glued on with biosafe epoxy but this time I cut an extra hole in the lid and glued on little laser-cut discs of acrylic when doing cell culture like this we use a special inverted microscope that looks up through the bottom of the dish but this means that the light has to come down through the lid in order to illuminate it so these little windows will let light in while letting us keep the lids closed that said these Lids still not a perfect solution so are something that are very likely to be replaced in the next version but that's pretty much all the prep so we can actually start growing stuff now as always all of our tissue culture work will be performed in our flow Hood with everything being sterilized and sprayed down with alcohol before bringing it into the hood before we take the neurons off the ice we need to prepare their food which is a special growth media the particular potion that I picked was called neurobasal medium plus a B27 supplement it comes from the manufacturer as a two-part mix that you combine when you're ready to use it to this I also added some pen strep which is a mixture of antibiotics to help keep the neurons healthy and sterile this is especially important because we're using homemade stuff that can't be autoclaved speaking of which we learned from last time that using heat to sterilize the arrays ends very badly so this time we went with chemical sterilization we started by water testing each of the arrays to catch any that leak before we add any neurons to them once we found a few that held liquid they were rinsed out and then sprayed down inside and out with 99 ethanol then we filled the arrays and their lids with a mixture of alcohol and three percent hydrogen peroxide if I can't use actual fire to clean these chemical fire will have to do after a few minutes of soak time the liquid was carefully drained off and replaced with several rinses of sterile water all of our other culture dishes usually come pre-sterilized so don't normally need this treatment however we're not going to be using the dishes on their own into the bottom of each well we're going to be placing a glass coverslip these thin discs of glass are extremely fragile and can also even be tricky to see the reason we're going through this effort is because of our microscope situation we have two microscopes one inverted microscope that can see the cells while they're growing in white light and looks up through the bottom the other is our fluorescent microscope and it can take these pretty fluorescent images but it looks down onto the sample I can't dunk the microscope objective into the well to see the neurons so this way once the neurons are grown we can just fish out the coverslip and flip it onto a microscope slide that way we can take nice images without needing a new microscope but the neurons are contaminated and definitely ruined by doing this which is why we need duplicates the cover slips were first cleaned in alcohol and then placed into the wells then they were given the same alcohol and peroxide wash followed by a bunch of saline rinses while there are some cell lines that could stick to Bare glass neurons most definitely are not one of them so the last thing that we have to do before we add cells is coat the wells with a polymer called polyd lysine thankfully this is super easy to do I have a pre-made solution of it that I just add about a milliliter to each well and to the arrays think of this stuff as Nano netting for the cells to grab onto we let that sit for at least an hour then it's rinsed off with a couple rounds of sterile water this is drained off and then replaced with culture media we add one milliliter to each array and two milliliters to each well of the 12 well plates these were then all placed into our meat cabator to equalize if this is your first time seeing the meatcubator its proper name is a humidified CO2 incubator it maintains a temperature of 37 degrees to mimic the inside of a body and fills its inner atmosphere with five percent CO2 this diffuses into the liquid and helps hold the pH of the growth media constant which is really important as any drift could kill the neurons cells are actually much more sensitive to pH changes than they are to temperature by leaving the plates and arrays in the meatcubator it gives the liquid time to warm up and equalize so it's perfect when we go to add the neurons once they were warmed and ready to go we quickly thawed the neurons in a hot water bath until there was only a tiny Ice Crystal remaining then they were sprayed down and the dishes and neurons were brought back into the hood the nice thing is that there's one million neurons in one milliliter of liquid so I can just put one microliter of cell suspension for every 1000 neurons I want to add and then everything was closed up labeled and then quickly transferred back to the meatcubator to grow rather than screwing the array's lids down they were just left loose to try and let the air in on the first day the cells just look like little round balls floating around and at this point they haven't yet attached to the glass after only a day that had changed all the little balls had now stuck to the ball bottom and many were already beginning to spread out and send out very clear dendrites but take note at how spread out they are because as I learned neurons move a lot I guess we think of neurons as just being in their little spot chilling sending signals but both during development and I suspect during your life they move around a fair bit when the brain is first growing and developing neurons need to crawl around so that they can get to the correct places in the brain there are dozens if not hundreds of types of brain cell so they often need to relocate from the place they form to where they're needed this is one of the behaviors I only learned about after seeing it for myself and it honestly really caught me off guard at this point I was still treating the neurons like Delicate China so after the first 24 hours when I needed to do the First Media change I pre-warmed some media in a six well plate in the meatcubator before use I think this is probably Overkill and just pre-warming the media is probably sufficient which is all I did for the rest of the culture period neurons don't actually like it when you do a full media change because they release a lot of stuff into the growth media that they use to function so every three days when I need to change the media I would just remove half and replace that half with fresh pre-warmed media each day we checked on the neurons they grew and crawled more than the day before and I was pleasantly surprised to see both how quickly and how easily the dendrites form and become visible also a trend began to develop really quickly in how the cells were growing at high densities they would form all of these little balls of cells that just grew larger as time went on and they send out thick cords composed of hundreds or thousands of dendrites in a bundle at the lowest density the cells stayed pretty nicely spaced out and slowly growing more and more dendrites the medium density was somewhere in the middle some clumps but even after a couple of days it was obvious a dense network was forming on day four things get exciting and we get to our first stained samples we'll be using a stain called calcine am this stain is really cool because not only will it make the neurons much easier to see it'll only stain healthy living cells and it's super easy to use all we do is replace the growth media with a special kind of saline that we use for tissue culture called hbss then we add a few microliters of stain on top it starts as a clear non-fluorescent liquid but in healthy cells the enzymes in the cells convert it to a fluorescent form which gets trapped inside the cell I let the cells process the dye for 30 minutes in the meatcubator before carefully fishing out each of the delicate cover slips and through the magic of editing it looks as if I got it on the first try I don't know about you but I would say those look like happy neurons at the lowest density the cells are still pretty spread out and are making connections but you can still see isolated cells but also wow does the stain make things easier to see I totally understand now why that's usually what you see in papers compare the medium density in white light where you can see some dendrites to it with some stain it goes from looking like shattered snot to a galactic web not only can you see the incredibly dense Network that has already formed you can even see what looks like synapses and the incredible green color means they are all very healthy at the highest density it's more of the same but you can get a better idea of what the blobs look like they're made of a shell of neurons with a core of Tangled dendrites but also look at how many connections they've made so I'd call that a huge success we can clearly grow and maintain neurons and keep them healthy now let's Jump Ahead a few days to day 10. at this point things were getting a little weirder the cells were still healthy but we were starting to see some really big chungus looking blobs forming at the highest density so I think it's safe to say that the highest density is too high for our experiments as these blobs won't sit on the array properly but they do look cool at medium density things looked better though still a little bit clumpy but look at how many connections have formed in some areas it's almost a solid mat of Twisted dendrites I hope this gives you an appreciation for just how complex your brain is this is what a few thousand neurons do imagine how many little wires are in your head the lowest density at this point actually looked the best with a nice Network forming but very little clumping and that Trend continued onto the final day where they had formed a beautiful and dense network with very little clumping so I think the verdict is that the correct seating density is probably around 15 000 cells per square centimeter just a little bit higher than the lowest density now let's talk about the arrays how they'd fare through all this well there's good news and bad news the good news is the neurons adhered properly and did grow at least a little bit bad news is threefold first we can't stain them because we'd have to break the arrays to do that so we can't use our nice indicator to see if they're healthy and second I think that there wasn't enough airflow because the lids thread on even when loosened I think they have a tendency to seal and prevent oxygen from getting into the cultures so the neurons eventually asphyxiated finally while the little windows on top work to let light in it was still pretty restrictive of the area we could look around so for next time we're probably going to make custom vacuformed Lids that will both let in more light and more air but otherwise I'd say this has been a big success we can definitely now grow neurons and keep them alive in culture which means we can check off the first step of our Tech Tree so where do we go from here well let's wrap this video up by talking briefly about what's coming up in the series if we look at the main quest line you can see that next up is getting signals out of the arrays this means we'll need to finalize the array design or buy a commercially made one and finish our Electronics we already started building and testing that side of this project and currently plan on using this amazing little chip which is a purpose-built neuron amplifier it's made by a company called intan Technologies and they have chips that either only do recordings or chips that do both recordings and stimulation so we're currently looking at buying the head stages from them which include the chip and then building our own back end for it that said if you're from antenna and you're watching this and want to send us one of your lovely complete systems we would be happy to show it off and use it beyond that we have a set of side quests which I mentioned near the beginning of the video one issue is that neurons are really expensive but also they don't divide so you have to buy them every single time you need them would it be much more convenient is if we could just grow them ourselves and while blending a pet store rat might work there's a much more elegant way to get neurons by far the most interesting way is to start with skin cells and then using a combination of DNA and small molecules you can directly convert those skin cells into neurons by reprogramming them there are pre-made kits that just let you do this which you can just buy but if you're already making new neurons it opens up another interesting possibility neurons grown in a sheet like we've just shown aren't exactly a perfect model of brain physiology while they still can learn informed connections they don't have any of the intricate structure you get in a normal brain but there's a technique which grows tissue that is much more like a real brain and it's called making organoids basically you purposely grow balls of tissue just like the big chungus we made by accident but because you're starting from a stem cell-like State the cells start to organize themselves much more properly into tissue that looks more like real brain behaves more like it too and even produces the right cell types so future iterations may use organoids instead of sheets of neurons and there's also ways that we can improve the array itself if we switch to organoids using microfluidics we should be able to upgrade the array to only use a fraction of the amount of growth media and keep the cells much healthier and by mixing vascular cells into the organoid as it's forming you can make little veins form to keep the core of the organoid oxygenated with a few electrodes we'd have the ultimate micro brain processor and since we can make them starting from skin cells there's nothing to say that we couldn't take a skin sample from say everybody at the lab and grow our own little mini brains then we can have an off this Quake tournament played by our own lab-grown brain dishes and see whose cells learn better which I think is about as sci-fi as this project could really get boy that escalated quickly I realize that's all probably a lot to take in but if you're as excited for this journey as we are then be sure to subscribe and ring that Bell so you don't miss any of it and this time I don't think you're gonna have to wait nearly as long for the next update part of the reason for the huge gap was that it just took a really long time to build up enough of the infrastructure to work on this project but now we have an amazing lab as well as the time and with your support the money to give this project the attention it deserves on that note one way you can do that is by picking up one of our amazing new t-shirts that we just added to our store we've got a bunch of designs that let you show your inner mad scientist on the outside and our new neuron array design is now probably my all-time fave of our lineup so grab something awesome to spruce up your style and join our amazing patrons channel members and supporters on Kofi in helping to feed our monstrosities let's build the cyberpunk dystopia that we want instead of the one that we have but that's where we'll leave it and we'll see you next time
Info
Channel: The Thought Emporium
Views: 3,284,987
Rating: undefined out of 5
Keywords: biology, experiment, science, neuron, neural, doom, video game, neural array, neuroscience, biotech, growing, growing neurons, tissue culture, biochem, squarespace, science fiction, sci fi, til, brain, learning, ai, neural network, computer science, microbe, microscope, fluorescent, glow, interesting, electronics, incubator, cortical, cortex, stem cell, cells, dna, classic, meat, food, robot, robotics, learning model
Id: bEXefdbQDjw
Channel Id: undefined
Length: 27min 10sec (1630 seconds)
Published: Mon Jul 10 2023
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.