Nectome and aldehyde-stabilized cryopreservation

https://twitter.com/kanzure/status/981363642598920192

video: https://www.youtube.com/watch?v=Y524FbBaVHM

Robert McIntyre, Nectome

Introduction

Well, thank you for the kind words. It's a pleasure to be here with such a great audience. I would like to start off by gauging the neuroscience knowledge in the audience. How many of you here can read this image like a book, where you look at this and this all makes complete sense to you? All the pixels in there. Raise your hand, don't be shy, if you can read this. Okay, great. Okay, great. Well, my promise to you is that by the end of this talk you will be able to read this image. If you get nothing else out of this, you will enlightenment for reading neural tissue slides.

I'm going to talk today about aldehyde-stabilized cryopreservation. It's a brain preservation technology that I developed a few years ago called aldehyde-stabilized cryopreservation. This is the world's first nanoscale whole-brain preservation technique. What that means is that it's able to preserve the synapses of a brain such that they look like the highest grade connectomics brains that are generally studied in neuroscience.

Background

A little bit of background on myself. I studied at MIT under Minsky, Sussman and Winston. I studied neuroscience as well. I left MIT for a year and worked on this thesis for AI where little robot bodies would move around in a simulated world and they had skin and they had tendons and joints, they could feel themselves and move around, they could recognize each other and it worked okay. While I was doing this master thesis, I saw this science prize offered by the Brain Preservation Foundation called the Brain Preservation Prize. It was to be awarded to the first team that could preserve the connectome of a higher mammalian brain. The connectome would have to include all of the individual synapses in the brain that store memories. Your brains, each of htem, right now, has about 86 billion neurons in them and something like 1 quadrillion synapses. This kind of scales down to the size of the brain, but the goal is to preserve it all. I thought this was a cool scientific goal. I started volunteering for the Brain Preservation Foundation and there were a few competitors at the time. In the process of trying to explain what they were doing, I figured I could win this prize. I left MIT and worked for about 9 months to make what I thought would work a reality. We won the prize, and now I have started a company called Nectome which is devoted to extending this technique, scaling it up, pushing the limits of it, and having the ultimate goal of preserving memories stored in the brain in addition to the structure that we already have.

Aldehyde-stabilized cryopreservation (ASC)

http://aurellem.org/ASC/html/ASC-introduction.html

To very briefly explain this, what aldehyde-stabilized cryopreservation does is that it's combining two different chemicals to preserve the brain. The first is a chemical called glutaraldehyde. It's delivered to the brain. It does a process called fixation, which glues biomolecules together very quickly and stabilizes the structures of the brain. The innovation is the second part, which is adding cryoprotectants to the brain slowly over the time. This is the same kind of technique you would use for antifreeze for your car. It prevents ice formation. The fixation gives you super short term stability. If you just fix a brain, then it will last for several years without much degradation but it's still an aqueous water environment which is still causing molecules to move around. You need to further immobilize the structures with vitrification which is the process of freezing a brain without ice. Imagine honey you put in a freezer and it becomes more vicous over time, it doesn't flow anymore, it's more of a solid glass. The combination of these two techniques is aldehyde-stabilized cryopreservation and they are based off of two of the competitors. One of the competitors was doing fixation, and then trying to do a bunch of other steps. One of the competitors didn't have fixation, but they had vitrification. Put them together, you get ASC.

Also you need to add azide and SDS.

What's the point of all of this? Where is this headed? The goal here is-- how many of you have a computer science background? Okay, a lot. Have you ever heard of the Visual6502 project? It's really cool. It's a fun little project. Some guys took what's considered the first computer chip, the 6502, and they took microscope images of the individual transistors that are on this chip. Here's this dead chip, it has some connections burned out on it, but the overall circuitry is still good. If you're careful, you can map out where the transistors are and what's the connections are, and then you can create a logical description of this circuit, simulate it on the computer, and the simulation will do 6502 things. You can hook it up to memory, play games, you can play software that was written for 6502 except do it on your simulated version of the 6502.

We want to do the same thing for a human brain. That is a much larger-scale challenge. It's not really a difference of kind, merely a difference of scale. You preserve a brain, you convert it to a form that can be imaged, you take a tremendous number of images at the equivalent of the scale required for imaging transistors in semiconductor circuitry, and then you simulate the human brain.

This is the level of detail that is thought to be approaching the level that you need to actually reconstruct a circuit in a brain. This is sort of equivalent to that transistor image you saw before. I imagine this image makes pretty much zero sense to any of you... So we'll have a short course on how to interpret electron micrographs of brain tissue.

A crash course in interpreting electron micrographs of brain tissue

https://www.youtube.com/watch?v=Y524FbBaVHM&t=6m53s

Alright, you start with this. One of the more salient features in this image is the mitochondria. What's up?

Q: ...

A: Yeah, you... actually have some brain in my wallet right now. You, if you're going to prepare a brain for electron microscopy... This is some rabbit brain. So from the slide we saw earlier. I think it's from this brain, actually. So you prepare the brain, you dehydrate, you expose it to heavy metal stains, so that it can be prepared for the electron beam, you encase it in plastic resin. If you take a look at this, you see that the individual slices are about as thin as a piece of paper. If you could imagine slicing into a sheet of paper edge on about 10,000 times, that's about the size of the little pieces that you then put in.

Q: ...

A: ... That's right. So, the video that you saw earlier, there's two ways to cut through layers of brain tissue. One is that you can use a diamond knife that is sharpened to a molecule-thin blade. You can cut about 35 nm. For comparison, in the pixels here, these are about maybe 2-10 nm. A more recent technology, even better than diamond blade, is a focused ion beam which you could use to shave off just 2 nm of the surface of this and take another image. You can get isovolumetric pixels. You can get a lot of detail. You can also just get a really thin slice and shoot the electrons through it, or on it and have it bounce, and get pictures, with the side effect of it being a little bit more blurry but not too much. This image was taken that way. This image is about 900 atoms thick, this slice. You can still move it around. And then the electron beam shoots down.

So these are mitochondria highlighted in red. These long bacteria-looking things.. you can identify them because they have these holes in them. Generally they are a little bit more dense. These are mitochondria over here. These are providing power for the brain to function. You almost always find them when protein synthesis is happening, or where a synapse is.

Another really salient feature is this big dark circle around this one process. That's called myelin, a way to make the signals go faster, it's a way of highlighting a connection that it's a little more important.

Probably the most important features but probably less obvious is the synapses. These are synapses. I have highlighted them in blue and yellow. Each of these is a connection between one neuron and another neuron. And this is probably best understood in 3d. You can see how these neurons connect to each other. The brain is densely packed with all kinds of processes going everywhere.

I find that this really helps to convey the idea. What you're seeing is this guy is holding this silicon wafer, and on the silicon wafer there's these tiny really tiny pieces. Each piece is about 500-900 atoms thick. And then we can zoom into this even more (connectome-demo.mp4). This is going to do a 3d reconstruction of an extremely tiny region. So right here, these are individual neurons, arranged and aligned, they send out projections, and they kind of talk to each other. Can you see any mitochondria in this image? Come up and point to one. Alright, good.

This one was cut by the diamond knife. All these 2d images, got a ... and it's kind of... trace out one single process, and we can see how each layer is making the whole. And you can extrude out the 3d model from these details.

The green one is one neuron, the red one is from another neuron. The intersection is the synapse. All of these synaptic clefts are connecting to other neurons. A single neuron can have 10,000 synapses. It's one big core, sends out processes that branch over and over again, and they have synapses in all kinds of places.

These are the synapses. Each of the synapses are part of this huge structural neuron. Each one of these points is where it's talking to another neuron.

Highlighted here are vesicles. Neurons have three components that are really obvious at this level of resolution. They have mitochondria hanging around giving the neuron power, they have a bunch of vesicles and it has what's called a synaptic cleft. When the neuron sends information to a connected neuron, one of the vesicles fuses with this barrier and it releases neurotransmitters. These dots are full of neurotransmitters. This is sort of like fundamental level of neural computation. These neurotransmitters are going to get into receptors on the other side of the synapse, and this will influence the behavior of the neuron it's connected with.

If we zoom into the synapse a little bit more, here's the mitochondria, here's the vesicles, here's the synaptic cleft. Vesicles, synaptic cleft. Okay? So now, some of this might a little bit more sense. Here's some mitochondria, right. And, can anybody find a synapse in this image? Who wants to point them out?

You're looking for vesicles. This is zoomed out compared to the others. There's the synapse right there, with little dots. There's maybe 100 synapses or so in this image. You can see how moving this up in the z-depth creates a complex 3d structure.

Q: ... moving around.

A: This is like a 3d cross-section. It's the same as that cube you saw before. This is using the ion beam technology. It's the same deal, though. You're looking at x-ray goggles or something, static physical structure and you're moving through it layer-by-layer.

And in fact, it's that plastic you see before, every time you see a new layer here it's because there was an ion beam that destroyed the last layer that was imaged, revealing a new surface to image. Then you image the next layer underneath.

That's synapses. When you look at the default data, it's not very colorful. When I look at it, I sort of see this kind of colorful stuff here. And then another really interesting thing, are the things highlighted in green, which is the cytoskeleton. It's the individual filaments that exist in neurons that help give them shape and help transport machinery and other stuff to the synapse. If the synapse wants to become bigger, you'll see machinery being transported down to the synapse to get it to grow.

Hopefully this has given you guys something for being able to interpret electron micrographs. Does the video make a little bit more sense? Alright, good. You can see the immense complexity. That whole video was such a tiny section that you probably couldn't even see it, and that's just 100s and 100s of synapses in just one tiny single frame of it, and it's a 3d structure. That's what's going on in all your heads. It's alive, though. There's a quadrillion synapses. These words you're hearing right now are being transformed into signals for those synapses as you're forming memories. Pretty wild stuff.

Brain structural preservation

Going back to our analogy, the idea is that we want to interrogate the brain at the lowest level to understand how it does computations and if we want to do them eventually then maybe we can get to an abstract description of a complex neural network like the brain and potentially be able to emulate it. But if you want to actually do that, then you have to preserve the brain's structure first.

So now we can get back to the way this ASC procedure actually works in practice. One of the main concepts is this idea of perfusion. This is delivering chemicals through the circulatory system of the animal. The reason why we deliver through the circulatory system is because it's already designed to get chemicals everywhere to every cell very quickly. Every single neuron in your brain needs a constant supply of oxygen and sugar, or else it dies. Your blood vessels are designed to deliver that constantly and quickly. Every single neuron in your brain is at most some tiny distance away from some capillary which is going to be feeding that neuron.

So what you can do is install canula into those main blood vessels that supply the brain. Instead of blood, you can send whatever chemicals you want, and they will get everywhere almost immediately. This is kind of the first step of the ASC procedure. In a rabbit model, you would load the carotid arteries that are powering the brain, and the rest of the face, and then you could wash the blood out with a wash-out solution which is designed to approximate blood except it doesn't have red blood cells in it. You can wash out the red blood cells, control the temperature, control the pressure. The pump in this diagram is like the heart for the animal, to push the fluids through. The diagram was what it looks like in the abstract, and this is what it looks like in practice in the real world.

This is the pump, this is the filter, and the heat exchanger is down here. This is what we use to actually do the first stage.

The second stage of the procedure is the actual ASC part. Once the animal has had its blood removed, we transition to a fixative. This is a solution of glutaraldehyde, you pump it through in the same way that you did the wash-out solution. And almost immediately, upon contact, it is binding biomolecules together. Glutaraldehyde looks like a pair of handcuffs and it sort of works like handcuffs too, it grabs on to proteins and itself and all kinds of stuff and it converts something that is very dynamic and lots of movement... the nuclei of neurons are rotating around all the time, mitochondria can migrate from one side of the neuron to the other or attach to a cytoskeleton and move somewhere... gluturaldehyde glues everything together and stops that motion completely. Once you fix the brain, it's much more robust. The consistency of the brain goes from soft pourrage as it is in life, to a kind of soft rubber. At that point, you can re-circulate fixative. In the same way that blood is recirculated in your body, you can have this fluid leave out of a vein. Now you have a stable system. You can run a system like this for days, the circulatory system works for years with no problem, so a few days is no problem. We slowly add the ethylene glycol to the brain. Previous attempts to cryopreserve brains run into this problem where if the brain hasn't been fixed before, the cryoprotectant causes a lot of structural damage because the cryoprotectant is very toxic and it causes the brain to severely degrade. So, you always have to race, you can try to put in the cryoprotectant quickly, but if you add it too quickly then it will dehydrate the brain and crush everything. There's no good way to win that game. If you put the cryoprotectant in really rapidly, you crush the brain, and if you put it in very slowly then it dissolves the brain. You can lower the temperature and you can slow the rate at which it causes damage, but it also slows the rate at which it gets into the brain to begin with. But since we fixed the brain beforehand, we can slowly introduce the cryoprotectant, and it relaxes the time requirements completely. We can get it down to room temperature, hours, and there's no distortions in the brain that occur in other cryoprotectant protocols. And you can add quite a bit in, both from 0% cryoprotectant, to 65% ethylene gylcol and 35% water. So it's more ethylene glycol than water, after you're done with this.

And here's an image of what it looks like in real life. This is a recirculating radiant plumber... this handles the main pump that acts as the heart. This is where we do all the work.

Are there any questions about that, about how to physically do this perfusion, fixation, and then slow addition of cryoprotectant over the course of hours.

Q: ...

A: You can actually make plastics out of glutaraldehyde. You can make objects that are almost as hard as procelain. The solution we're using is 3% glutaraldehyde, so it doesn't quite have enough to polymerize everything. It's going to be making little changes, but it's not going to make things big enough to clog a blood vessel, at least not at the time scale we're dealing with here.

Q: ...

A: The animal is alive while the surgery is being done. The wash-out solution is compatible with the life of the animal, it cools the brain down as it is being perfused. If you had a surgical team prepared, you could actually reverse all of this up to this point, theoretically you could drain the wash-out solution and put blood back in and have a surgery team ready to get that animal back to normal and survive. The animal is alive exactly up to the point where glutaraldehyde is introduced.

Q: ... once you add the cryoprotectant .. say you got...

A: There's different classes of cryoprotectants. Normally used in cryobiology are unstable classes, these are cryoprotectants that don't have enough cryoprotectant to really properly avoid ice crystal formation. If you cool it quickly, there wont be any ice crystals, but the solution makes tiny sea crystals that you can't see, and when you warm it back up, the crystals grow as it heats back up. But if you have 65% ethylene gylcol then it's intrinsically thermodynamically stable, and even if you put in a seed crystal, it would just destroy the crystal. The other advantage is that we can go and use more cryoprotectant than was previously possible, and we get to a stable glass formation. It doesn't matter how slowly you cool, or how quickly you warm up. I like to cool brains down, put them in liquid nitrogen system, then we cool them for the next hour or two, and then to warm them up we put them just in room temperature and let them warm up for about an hour.

Q: ... and then.. blood... and then you..

A: You need access to the carotid arteries. It's two canulas. That's all the attachment you need.

Very similar to an extracorpeal austriaiagiigade machine. You attach them, it helps filter your blood.

Q: ...

A: Absolutely.

Timeline

To give a little bit of an idea on the timeline.. Brain Preservation Foundation was started in 2009. This research in ASC started around 2014. The prize was won in early 2016. It's around 2 years, but it took more like 1-2 years to deal with all the problems that occurred.

We won the Brain Preservation Prize for preserving a rabbit brain. It was a pretty big occassion. It appeared in a lot of publications too. The way that we demonstrated that we could preserve all the synapses in a rabbit brain was that we preserved a rabbit brain, we sliced many slices of the brain, we randomly picked a bunch of regions and looked to see if the synapses looked structurally sound. You saw the vesicles and synaptic clefts-- if we were to see a broken synapse or vesicles attaching to it, or even one vesicle, that would disqualify the brain. Everywhere we looked in the pruning videos, it all looked fine and generally accepted as good structure preservation. It was judged by Kent Hayworth and also Sebastien Seung who was at Princeton now he's at MIT.

Comparison of brain preservation techniques

https://www.youtube.com/watch?v=Y524FbBaVHM&t=30m

Here's some comparison. This is a control brain. This is a brain that was just fixed in glutaraldehyde using the standard way of preserving brains for study and what is generally believed to approximate what the brain looks like in real life. At a similar level of magnification, this is an ASC-prepared brain. This is zoomed out a little bit. But here you can actually see individual neurons. There's more micro detail. This right here is a capillary, this is where the nutrients and blood cells would go. If you hadn't prepared the brain too well, then this is where you might see a red blood cell. And over here, these are the actual neuron cell bodies. This one goes like this, here, and around, and that's one of them. And then we have these huge nuclei in them, full of DNA, and this very dark structure here is the actual DNA in that nucleus. And so you can see that there's not much of a difference between the two images, which is good.

This other image is zoomed in to a similar extent... does this image make sense to any of you? We have synapses, here's a synapse, it has vesicles here, so we've got one there, one here, here, that's a synapse, there's one, there's one, there's one, you see the dark lines on the border..

Q: ...

A: They are mostly spherical. There's some interesting stuff there. We have in this area of brain two different types of synapses, one that is excitatory and one that is inhibitory. The excitatory use glutamate in their vesicles. The inhibitory ones have GABA in them. When the brain is alive, they are both spherical. But aldehyde-fixation actually causes the GABA ones to become ellipsoidal, and we can differentiate between them after the fixation process.

Q: ...

A: You mean, you normally know where you are, because you start from a whole brain. So you know generally where the slice was taken from.

Q: ...

A: Oh, you can do a tremendous amount of.. Yeah, so I think the question is, is it possible to selectively label parts of these neurons? Can you label the excitatory synapses or something. Is that the question?

Q: ... may well.. differentiated.. some other... specific part of the brain...

A: So you're wondering the different regions of the brain and the different types of neurotransmitters. Yes, absolutely. So where we are right now, in this image, is the hippocampus. These are pyramidal cells, named such because they kind of look like pyramids, I'm not so sure about that but I guess they look sort of similar. The pyramidal cells always use glutamate, knowing that this is the hippocampus and you trace the synapses back to this, then you suddenly know that this vesicle is going to have glutamate. And similarly, there's some inhibitory cells that only use GABA. There's 100s of types of neurotransmitters and different subtypes of neurons use them. Some neurons have more than one type of neurotransmitter. Depending on where you are, it's kind of like pokemon, and there's a bunch of subtypes everywhere, and they look quite different. The cells in the motor cortex for example are very recognizable and different to the cortex, or even the thalamus, the thalamus is this huge myelinated tracts so you don't see as much.

One of the things I definitely didn't appreciate until I actually started looking at real brains... coming from a computer science perspective, was just how variable the brain was. I thought the brain was homogenous continuum of stuff. But there's very distinct macroscopic structures. The genetics is such that there's a line of neurons here, a different line over there, and then it's reproducible in each brain, then it grows the projections up here and to this other one.. but the overall topology is layed down. It's a very beautiful architecture. I would encourage you to take a look at even coronal sections of the human brain. It's quite interesting to look at.

So this is a control brain- there are synapses here, and this is ASC in this image. This one in addition to being fixed was vitrified and rewarmed. The important part is that the vitrification would allow it to be stored for 100s or 1000s of years. The fixation alone wouldn't allow that, and to win the Brain Preservation Prize we had to be able to argue that it would hold for at least 100 years.

Who can find the synapse in this one? Can you find one? You think so? Alright. The dark lines between them, perfect. Great job. This is great. You guys know how to read this stuff already.

We built a larger machine and did this on pig. Because these solutions are delivered by perfusion, the only design criteria for scale-up to larger animals is more solution and bigger pumps. It took 7 months to design the machines and get everything working for the rabbit, but it only took one evening to design the machines for the pig. This is what we got as a result, pig brain. This is very promising. The pig brain is still being evaluated by the Brain Preservation Foundation. We're going to be hearing some good news from them in the next month or two, I think.

This is zoomed in a lot. I like this one because there's like six different synapses here, on one neuron. I think that's just pretty sweet. It's very difficult to tell what species you are at this level, whether a human brain or pig brain or a rabbit brain. One of the most striking differences is that mitochondria look different depending on which brain you're in. The synapses are a bit more challenging.

Q: ...

A: Uh. I do. I could bring that up I guess. Hold on one second. Vitrified brain micrographs.... So this, is equivalent to.... this. About. That's about the same magnification. This is what it looks like without the fixative. You can see the profound dehydration effect in Alcor's images. We can zoom in, so, let's see.... sure this one is about the highest mag. This is roughly equivalent in magnification to ... uh... this. So you can see, like maybe this process here. That would be equivalent to this process here, which is dehydrated, pulled away and retracted from the surrounding myelin. You should be able to see probably a good 20 synapses in this image. My best guess is that maybe this is a synapse, but maybe not.

Q: After you take a brain.. cells..

A: We're not bringing it back to life. We're preserving the structure. Starting from a fixed brain, you can remove the cryoprotectant, and you can rewarm the whole brain, and use perfusion to remove the cryoprotectant. You're left with a fixed brain. There's no way to undo the gluing reactions. Say you have a book and it's a precious book, it has a lot of important stories written on it, you might not understand the language it's written in, and you want to store it nonetheless. You can mix a clear epoxy resin and pouring it on the book. You're gluing the book together, the ink into the pages, everything, into a giant solid plastic object and you will never open the book again without some extreme techniques. But the information is still in this book, and maybe you slice the book into slices and scan the pages. Maybe you use an x-ray and the x-ray can let you see into the book without cutting it up. But if you're very motivated, you can get the story back and read it again. And even if you insisted on un-epoxying the book, I would say it's easier to figure out the chemistry to dissolve the epoxy between the pages, than it would be to fill in missing details from a hole. You could imagine that with billions of dollars of funding you might be able to figure out how to un-epoxy a page... but I don't know if there's, if there's a hole in that page, and you lost a whole paragraph of text, no matter how much technology you have then you will not be able to recover that information.

"Biologists fix tissue by pumping a glutaraldehyde solution through blood vessels, which allows glutaraldehyde molecules to diffuse into cells... This commonly shackles a protein molecule to a neighboring molecule.. These cross-links lock molecular structures and machines in place; other chemicals then can be added to do a more thorough or sturdy job... The vitrification process packs the glassy protectant solidly around the molecules of each cell... Fixation and vitrification together seem adequate to ensure long-term biostasis." -- Eric Drexler, Engines of Creation, 1986

One of the interesting things- and there's a lot of low-hanging fruit in the world of brain preservation.. this was announced 30 years ago. In 1986, this guy Drexler described glutaraldehyde and ethlyene glycol to preserve brains and that it might be sufficient and nobody followed-up on it. I actually learned about this after I had started doing it. I thought yeah, it's 30 years old, it's older than I was.. I asked Drexler why didn't he do this, well basically he was interested in other stuff.

Nectome

I started a company, Nectome. The goal of Nectome is to continue this work. We acquired an NIMH grant to work on connectomics technologies in collaboration with Ed Boyden's lab at MIT. The goal of this is to design methods to take brains and preserve them with ASC and then compare them for whole-brain perfusion and whole-brain perfusion expansion, which are some really cool technologies. This is a picture of our lab. We have a lab down in San Jose. We have been doing whole-brain perfusion and whole-brain expansion similar to expansion microscopy.

Nanoscale human brain banking project

Recently we started a very interesting new project which I am pleased to announce to you guys tonight which is to scale ASC to a human brain. This is our setup. We'll be receiving a donated human brain. We've been working closely with third-party organizations so that we can obtain tissue samples extremely rapidly potentially 15-30 minutes post-mortem. We will see if that's adequate for preserving a connectome. In any case, this is going to be a major advance compared to normal brain banking technology which is only gets individuals about 6-20 hours after death, using immersion fixation in glutaraldehyde to do the preservation process. So we're very excited about the nanoscale human brain banking project that we're embarking on. We'll see if this looks good. We're doing our first trial experiments this weekend. We'll see.

Philosophy, consciousness, etc.

Q: Seems like the key assumption is... number of.. not all thought is computation... peripheral things.. Chaumers... inherent self-experience, subjective, ...

A: Well, you know, I can give you some intuition on this. I wouldn't be too quick to say that just because it's intuitive to you doesn't necessarily mean that it has elevated you to a universal truth or a universal principle. Most people think that the concept of a self is a thing, that you're one entity, but in reality, it's more like you're the CEO of a gigantic corporations and you don't even know what all your subdepartments are. The "self", in neuroscience, is not nearly a coherent clean concept as you might imagine. In split-brain patients, most people are two separate brains stapled together and you have to pilot it together and we sort of don't realize this. The thing that we're talking about, this language, this is responsible for telling stories about us doing things because we were angry or a CEO making a mission statement or something-- but in reality, it's an incredibly complex vast computational system that has depths that aren't really aware of your conscious mind. So the same way that the self becomes slippery and maybe not a complete description of what you are, this idea of saying well it feels like you have a first person experience and therefore that's the fundamental organizing principle of the universe, I would caution against saying that's true. It's very easy to assume you have a privileged view of yourself-- but that's a tiny part of the vast organism that is telling that story to yourself. It can feel sometimes like the leader of a company is the company, but at best it's an abstraction, which is not to say that abstractions aren't useful, it can be very useful to say there's a self, in the same way that saying there's an object is useful, but objects are a thing that we project on the world- you can't take a thing and say where is the objectness of this thing, how do I measure that... well if you start measuring it, maybe it become stwo separate objects, how does the objectitity come from, you can keep doing that as much as you want, maybe it's a different thing than it was before, maybe you say it's the same as before, but you can't argue if you drop it whether it will fall, you can talk about that one empirically. I try not to criticize people for their internal philosophy.. but your mind physically exists, your brain does this. What's the center of mass of an object, can you find the atom that is the center of mass? That's a nonsense question to ask. Given that the world is computable, that physics is computable, and the brain, being a part of the physical world is also computable, and the only way out of saying therefore the brain is computational thing, is to say the brain is not an entirely physical thing-- you can bite that bullet, but then why? What do you need to explain about the brain or human behavior that can't be explained by neuroscience or physics or computation? There's a lot of evidence that would convince me that we have souls, like if we opened up a human brain and there was one atom of cobalt in there surrounded by a bunch of salt water and yet we still did all of our behaviors, then I think that would necessitate some sort of very metaphysical explanation for how the mind works. But in our world, we open up the human brain and we see a quadrillion synapses, we see machines that are constantly working, changing sizes when you learn something, and if you invoke the machinery to induce the synapses that had just increased, it perfectly erases that memory, and when you create false memories it works, it's all based on synapses. There seems to be plenty of machinery in the brain that doesn't need more explanation from an incredible huge machine that seems to be doing all of this.

Structural information and memory preservation requirements

Q: Kind of on... struct the brain.. what other information... need to know if you wanted to...

A: Okay, so the question is, you want to get a brain scan. What sort of stuff do you need other than just these electron micrographs? This is still being debated. I'll give some comments. It's really interesting that if I know the geometry of the whole neuron then I can say it's a pyramidal cortical neuron and therefore the chemical in the vesicles is glutamate. This information exists in the electron micrographs, but if yo just saw the pic of the vesicle then you wouldn't be able to tell. We might be able to work out other pieces of information from the electron micrographs. You certainly need a lot of electrophysiological data. This is blocking us on getting celegans to work. They scanned the connectome, why can't we simulate it? We don't actually know for most of the synapses for celegans whether they are excitatory or inhibitory. We need a systemic study of those neurons and how they behave in order to figure out the characteristics of those neurons. If I gave you a circuit diagram of the brain but I failed to label which ones are resistors or capacitors, you're going to have a hard time figuring that out, or search in exponential space to make that work. There might be some things that are difficult-- like gap junctions, or electrical synapses where they are not like normal synapses but they are just against each other and they have proteins like connexins that blow holes through them and directly share proteins and ions with each other. If those neurons are firing, then that electrifcal connection is different from chemical synapses, which can be challenging to see with electron micrographs and under some preparations it's totally invisible. I think you're going to have trouble simulating if you don't have those electrical synapses. Some people say that they are more common between inhibitory neurons... so simply function just to dial back the excitability of the network. The brain is always on the edge between epilepsy and coma, and it has to be on the edge of criticality so that even small clusters of neurons can cause macroscopic behavior and amplify the signals. You have excitatory synapses, and maybe a mesh of inhibitory synapses, and it just generally dials the system down... and maybe it would be sufficient to know where the excitatory synapses are, and then we say the whole system has to be dialed down, which would simplify the problem. Say that we have a neuron with two different kinds of neurotransmitters that can't be distinguished from the electron micrograph, that would be a big challenge. In that case, you could label the individual proteins that matter, with immunolabels or other techniques to actually find these, bind them, and differentially stain them like with a heavy-metal stain. You can expose the brain tissue to whatever preparation techniques you need to highlight the different materials. Well what about a thousand different things to disambiguate? Well, then you need to go beyond the limits of current staining protocols. There's some debate about whether electron micrographs of the kind we use right now would be sufficient. But going a step back, is ASC insufficient itself? The proteins are still there-- a particular ion channel is going to be glued to the membrane with glutaraldehyde. Maybe the challenges associated with labeling that in a way that is consistent with labeling everything else; the proteins are still there, so the information is still there. What would really sink ASC is if there was a memory storage mechanism in the brain that is simultaneously so fragile that it is destroyed by rapid glutaraldehyde fixation, or maybe down to 10 degrees celsius would also destroy it, and it's hard to point to anything in the brain that has both of those properties.

Q: So... modeled.. the types..

A: This was a brain that was preserved under a euthanasia scenario... if you let the brain wait hours, then it's not going to be pretty. I'm sure I can find this, if you can give me a minute, I can find one that was left out for greater than 20 hours or something.

Memory preservation

Let me show you what brains look like after about 20 hours or so. They are basically gone. There is literally nothing left. You just see cytoplasm everywhere and a bunch of trash littered around. There's no synapses. It's 20x worse than the Alcor micrograph. To say the information is gone, what does that mean? If I take a book and just incinerate the book, there's the ashes, is the information gone? You might think that it is gone, but maybe a sufficiently advanced cryptographic approach could get the information. Neurons have genetic tags to avoid synapsing on itself, which would destroy itself, you really don't want those feedback loops. Neurons can tell whether it's self or not... maybe that could provide some extra information. The less you can rely on extremely ill defined exponential search processes in the future, then the better. Every bit of damage to the connectome is certainly going to make it less likely to recover the brain. I think you can make it quantitative. Say you take 10 mice, and you're going to label the location engrams in the mice, and they make projections into the amygdala. Say this is a red neuron and green neuron coming in, for two engrams. Say you do a fear memory induction in 5 of the mice and not in 5 other. You preserve all of the brains using whatever method you want. You have 10 brains, you have to sort them into two groups, and the two groups need to correspond to the two training groups. It's almost a zero-knowledge method to determine whether there was still sufficient information to determine the correct group identification, without actually knowing the exact memory detail. Otherwise you wouldn't be able to differentiate. Maybe you have some machine learning thing that tries to separate them. With one preservation technique if you always get it perfect, and with another technique you do better than chance but not always right, then you could say that one technique is better than another technique. Maybe a new technique would have better data extraction. There's a lot of opportunity to make this stuff more rigorous like "this preservation technique is better under these particular conditions" especially for certain aspects of brain capture.

Large search spaces and digital repair

Q: Even if...

A: Even if the brain is... there's this idea of single... if you start to get to this limit of are the structures inferrable or not, you have to invest tremendous amounts of computation, and maybe it's theoretically possible to figure out a whole connectome with a quadrillion synapses, but the computational requirements would be crazy, especially if you had to simulate each one, which would grow exponentially with the number of synapses that you're trying to get correct. Even in a perfectly preserved brain, you have to scan a whole liter of material at 2 nm resolution. Even doing that with good staining is a challenge. Not impossible, I don't know. But we should be putting more resources into proving that the synapses are preserved without scanning a whole brain, but now we need to put more resources into proving that there's memory preservation. The incident rate of damaged synapses has to be extremely low. Same thing can be done with memory in the amygdala, or memory in different brain regions. We'll put 100 bits into a mouse, and extract the bits out. I can't even find a single synapse in an image with this degraded brain from this micrograph of a decayed brain... it's all very challenging.

Physical evidence of synaptic theories of memory

Q: ...

A: I got some cool slides that show the process a little bit more. I enjoy it. Here's a picture of some brain preserved in plastic. This is a picture of some of the... at Nectome.. where we use perfusion to actually do the entire process of embedding. We put in cryoprotectants but also heavy-metal staining, and plastic resins, and htis is a late stage picture, and also tetrium oxide to blacken it. Here's a whole brain, made plastic from these techniques. Here's some close-ups. Here are some rabbit and pig brains. I just find this fascinating. Here's a vile of tetroxide, some nasty stuff. But yeah, whatever questions you have yeah.

Q: ...

A: So the question is, what do we really understand about the brain and what do we not, with regards to simulating the brain? Great. There was some references from Ken Hayworth. It's amazing that in the literature people say it's just synapses. If you understand the system then you might be able to manipulate the system. There's impressive work coming out in the past few works about manipulating synaptic structures related to false memories. If there was too much more complexity beyond that, we shouldn't be able to selectively knock out memories like this... it seems somewhat implausible. I think there will be some interesting surprises. Do I think synaptic basis of memory is completely wrong, no I don't think that's true. But in any case, I'll send out some resources.

Motivation

Q: ...

A: What is a just world. Martha Nausbaum.. we're talking about, what is the path of humanity to take, to take this world that doesn't really care about us, and replace what's natural with what's just, and we decide what justice is. That's a beautiful thing. We got dealt a bad hand with nature, but we're able to care about each other, we can change the world with technology and the world can begin to care about us in the same way that we care about each other. In the same way that we care about life, liberty and the pursuit of happiness, life is number one, and most people who die don't want to die, and even the people who are on the fence about dying or not, they are on the fence because they have suffered as their body has falled apart. If it could be demonstrated to preserve memories, then eventual digitization would be a tremendous benefit to humanity both at an individual level and at a societal level. With each generation, we get better technology but I don't think we get better wisdom because wisdom is something that I think you need to spend a lifetime developing for yourself. Language isn't powerful enough to communicate wisdom. The computation in your mind is powerful enough to do wisdom, but not language. Stories do an okay job, but we need a good solution for this. Digitization of memories is one of those potential technologies I think.

More philosophy

Q: .. philosophical discussion... does this project...

A: Humans having free will or not, or you read some archane branch of physics where you start thinking everything is alive or whatever, please rethink your life. But bookkeeping in your head, that's not real, it's just an incredibly useful construct that you use to navigate the world. It's going to be related to your capability to make choices. If free will is saying that there's some metaphysical thing that exists that is immaterial, that's an entirely different thing to be talking about... The common use of the idea of free will, as to whether you are capable of making choices based on the information you receive, then yes I would say that's something that is mechanically possible.

1h 12min