Episode 2 | Transcript
Episode 2: "Mapping Connectomes with Ion Beam SEM"
Published Sun, 26 Aug 2018 | Transcript
Allen Sulzen: Hi there and welcome back. My name is Allen Sulzen, host of the Carboncopies Podcast, and for our second episode we'll be diving into methods for preserving animal and human brains with Dr Kenneth Hayworth. We'll also be looking at possible timelines for whole brain emulation and milestones to reach along the way. This presentation is the second in our workshop series here on the podcast, and we've got many more to come. Dr Keith Wiley, communications director at carbon copies wrote the following summary of Dr Hayworth's presentation. He wrote the session began with Dr Kenneth Hayworth from Howard Hughes Medical Institute at Janelia. Hayworth, runs the brain preservation foundation, whose goal is to promote and advance long term whole brain preservation brain preservation foundation aims for the eventual establishment of brand preservation as a standard medical practice offered in hospitals for life preserving or life stasis purposes. The brain preservation foundation is motivated by the likelihood that brain preservation might be achievable sooner than mind uploading and can therefore be used to bridge the intervening time span until mind uploading is possible.
Early success toward the brain preservation foundation's goals has been encouraging. The most noteworthy event being that in 2016, the brain preservation foundation awarded it's first cash prize for achieving new milestones and brain preservation. This prize was won by Robert Mcintyre's team at 21st century medicine for the electron microscopy verified preservation of a whole rabbit brain. Such a feat of peer-reviewed and verified high-quality preservation of macroscale neurophil. A whole brain in fact, has never been achieved previously. Not only does the brain preservation foundation motivate research by others through its cash prizes, but Hayworth himself has pioneered the technological development of slicing and imaging machines for the subsequent scanning of preserved brains. Such imaging is an obvious critical step in one of the more likely mind uploading procedures. A procedure in which a stasis preserved brain's structural scan is used as the basis for a subsequent whole brain emulation. With that summary we're ready to begin.
Randal Koene: Our very first speaker, Dr Ken Hayworth, Ken is a senior scientist at the Howard Hughes Medical Institute Janelia research campus. He's also the president and cofounder of the brain preservation foundation, and he's probably one of the scientists with the longest history of carrying out explicit research with the aim of making whole brain emulation possible and a longtime friend of mine.
Dr Hayworth: Uh, thanks Randal. Okay, so this, uh, this workshop is about a human mind uploading a great topic. Uh, I probably don't have to say this, but, uh, this, the goal of human mind uploading is, an overwhelmingly difficult, in fact, it is so overwhelmingly difficult that, uh, uh, it's achievement is at least many, many decades away as Randall's last slide was showing and therefore it doesn't make too much sense to talk about it except in so far as it can motivate today's research milestones. So with that in mind, I want to start off with a speculative timeline that I think can help organize some of these research milestones. So this is going to be quite a bit different from the timeline that randall just put up, but we can discuss that later. So here we are in 2018. If I had to speculate, I would say that the first successful human upload would not occur before the middle of this century and it will require a tens of billions of dollars of a research project on the order of an Apollo Moon shot.
And it will probably take till late the century until the cost comes down sufficiently for quote unquote normal people to have a chance at experiencing, uploading firsthand but there are many milestones along the way to this first human upload and these are what we should mainly be concentrating on now that is making sure that they occur and trying to accelerate their progress. Proof of memory decoding from a small connectome is probably the, uh, the top item that that should be concentrated on. And people are working on this very seriously today. Now, I expect that a convincing proof of this will be demonstrated early in the next decade following that imaging, a whole mouse connectome. Again, people are working seriously on this, but it is really beyond the bounds of today's connectomics imaging techniques. So there's some serious developmental work that needs to go on to address this.
And that's what I'll be talking about in my talk. The successful uploading of a mouse and I'm putting that at about 20 years from now because this is a lot, a lot of basic neuroscience knowledge will have to be gleaned in the intervening decades to make this possible. So this is really quite out there. And then imaging the first human connectome. I would put that out at about 30 years if I had to speculate because it is just such a monumental task. I see these as the major technical milestones leading to the first successful human upload. But there are a ton of neuroscience ones that I would love to go into detail, but I can't really in, in this talk. A question that should be asked of any, tens of billions of dollar projects should be a what good is this to humanity?
And Randal gave some ideas of that, but I think one that I would like to put forward is that I think the first successful human upload will usher in an era of what I might call experimental philosophy, a revolution in our understanding of what consciousness is, what the self is, what states of mind are possible, this is what really gets me excited about mind uploading, when we have complete access to an uploaded volunteers brain, the Neil Armstrong, if you will, of mind uploading, they will be able to explore the depths of consciousness in ways that we can't even imagine. These explorations of the uploaded mind will give us, all of humanity insight to the next stage of human evolution. So I think, I think crystallizing the impact of that first human upload, even if it is a tens of billions of dollar project, is probably a good a milestone to motivate the earlier ones that we really should be focusing on today.
Now, given the current pace of neuroscience and technology, I don't think this timeline is completely unreasonable. It may be overly optimistic by a factor of two or three. So why are so many neuroscientists reluctant to embrace that mind uploading is clearly the end goal of a successful neuroscience? Why are so few pursuing these early milestones? Well, here's my speculative take on this question. Joe neuroscientist here, we'll assume he's a middle aged or elderly, distinguished neuroscientist. He understands enough neuroscience to understand how monumental the task is. Mind uploading is just incredibly, incredibly a monumental task. And to understand, he understands that he will be long dead before mind uploading is achieved and we all have a tendency to say that things are impossible if they are only going to happen after we are dead, but by its very nature mind uploading has the potential to change this. If one could come up with a method of preserving the brain of terminally ill patients today for long term storage, then anyone alive today, could be expected to bridge that gap. A gap that might be 100 years long or more.
Thankfully the technology to bridge this gap is available today in the form of brain preservation by Aldehyde stabilized cryo preservation. The ASC technique was invented by Robert Macintyre and Greg Faye and published in this 2015 article. The ASC procedure is quite straightforward. The animals vasculature is first perfused with a deadly chemical, fixative Glutaraldehyde for about 30 minutes. Glutaraldehyde almost instantly stops the metabolic activity and begins cross linking all of the cellular proteins into a Sturdy Mesh. After 30 minutes, they start to slowly introduce cryoprotectant into the fixative profusing and ramp up its concentration to about 65 percent over the course of the next four hours. All of this is done at room temperature, allowing for uniform cryoprotectant concentration in all cells. This uniform high concentration of cryoprotectant allows the brain to be stored solid at extremely low temperatures minus 135 degrees Celsius without ice crystal formation.
Now, fixation by glutaraldehyde is known to preserve the patterns of synaptic connections among neurons and to preserve the ultra structural details of synapses. In fact, it has been the gold standard for electron microscopy for decades, and it is important to understand that glutaraldehyde is also known to preserve the primary structure and relative locations of most proteins as papers like this and this one describe. The key point is that glutaraldehyde fixation is simply the best method known to science to preserve both the structural and molecular information across an entire brain as a result of this cross linking. A glutaraldehyde fixed brain is immune to biological decay processes and will remain stable for months, but eventually diffusion would result in the slow dislocation of bio molecules. For example, membrane lipids that were not cross-linked, but even this slow diffusion is halted in an ASC preserved brains stored solid at minus 135 degrees Celsius time has essentially stopped for such a brain.
Here's a video of a rabbit brain preserved by ASC being lowered into a minus 1:35 degree Celsius cold storage chamber, and here's that same brain removed from storage the next day. Notice that it is completely solid. When this brain was rewarmed, we sliced it up and prepared the tissue for three d electron microscopy, the ultra structural details of the brain, where as well preserved as those in control brains that had undergone only the normal glutaraldehyde profusion fixation. Here's a 3D electron microscopy video of a small piece of that ASC preserved rabbit brain. This data set was acquired by me using focused ion beam scanning electron microscopy. There is no ice damage, seen, no cracks. The ultra structural details of St Francis are clearly visible and the connectome is clearly traceable. Now, the key points to note: ASC preserved the structural connectome across the entire rabbit brain, including ultra structural details of synapses that should make the determination of their functional strength possible.
ASC Preservation is compatible with the later tissue processing steps needed to map the structural connectome, and this to me means that asc should be compatible with future mind uploading. The same ASC technique has now been applied to whole pig brains. These are pictures of an asc preserved pig brain that was rewarmed after being stored at minus 135 degrees Celsius. Again, there was no macroscopic damage, seen no ice crystal damage and no cracking and the ultra structure across the entire brain looks well preserved. This technique could just as easily have been performed on a human patient and the result would have been a preservation of these structural and molecular information contained in that person's brain and compatible with storage that can get that information to the distant future. So the implication of this that I want to stress is that ASC preservation of human patients will almost certainly support the possibility of future mind uploading. ASC is simply glutaraldehyde profusion of the brain combined with indefinitely longterm cryo storage. ASC has been proven to preserve the structural connectome. ASC should preserve the locations in primary structures of key neuronal proteins like ion channels. ASC Storage is resilient to short term rewarming because glutaraldehyde fixed brains are stables for a stable for months at room temperature, and when rewarmed ASC preserved, brains are in an optimal state to undergo the types of tissue preparation steps likely necessary for mind uploading. They can be reaper, fused with liquid fixatives, heavy metal stains, and potentially more targeted stains designed to highlight particular classes of proteins and they can be dehydrated and plastic embedded, allowing for precision sectioning and volume electron microscopy. The implications as I see it for today's discussion is that mind uploading is very difficult, very difficult, and may not be available till next century because of this lack of near term benefits, it is difficult to attract the interest in funding necessary to reach even the near term milestones. In contrast, brain preservation is easy and could be made available in every hospital today giving hopes to millions of terminal patients. I believe that once people believe, begin to realize this, it will create a virtuous cycle. More people interested in personally preserving their brains will lead to more funding of research to determine what would be required for mind uploading. This will in turn lead to better evidence that mind uploading will work which will close the circle, creating more people interested in preserving their brains. So with that preamble, let me spend the rest of my talk focusing on these two milestones. Imaging, mouse and human connectomes. This is a figure from Shawn Mccool is recent paper which highlights just how daunting a task we are facing. The field of connectomics is currently struggling to image a single cubic millimeter of cortex as part of the well funded IARPA microns project. It's a $100 million dollar project, I believe funding three different groups and they are all struggling to image the entire connectome of one cubic millimeter. To put this in perspective, a whole mouse brain is 450 cubic millimeters and a human brain is over 1 million cubic millimeters in volume, so it's a daunting task. Let's take a look at today's main connectomics imaging technologies and see if any of them are up to that task. They aren't by the way, but I will go through them anyway.
Each of these technologies has proven its ability to map the connectomes of small pieces of brain tissue. Only two of these in an atom SEM are currently in the running for the IARPA one cubic millimeter challenge because they are the only ones currently compatible with high current electronic imagery. Serial section transmission electromicroscopy allows fast imaging because the electron beam is spread across a tens of microns wide field of view, allowing an entire image to be taken simultaneously. This has the advantage of having orders of magnitude more electron current available for imaging then, for example, a single beam scanning electron microscope. But of course, the disadvantage is that the sections must be incredibly thin and such sections are inherently fragile and difficult to collect even when using automated tape collection. For example, an SEM collect sections on a solid tape, it allows fast imaging because it is compatible with the Zeiss multicell microscope, which is a scanning electron microscope, which has up to 91 electron beams, imaging the samples simultaneously.
And really this multibeam scanning electron microscope with, I think Sean McCullough, we'll talk about much, much more later, is the thing that I think will break open the mapping of connectomes. The current electron microscopes, the current Zeiss multisense have 91 beams, so that's about 91 times faster than a single electron scanning electron microscope. But Zeiss is working on hundreds of beams in a single microscope. So I think this is a rapidly advancing imaging technique, so atom SEM it's important to understand is compatible with the Zeiss multisense because it collects mini sections in these are spread across a wide wafer. And so you're able to image all of those, very efficiently. Again though it has to collect those incredibly thin sections and they're inherently fragile and difficult to collect. Now, the z resolution of both of these techniques is determined by the section thickness, and this represents a fundamental z resolution versus reliability tradeoff at the resolutions needed for tracing neurocircuits, which is less than 40 nanometers, probably considerably less than 40 nanometers. This means that these techniques must collect hundreds of thousands, and in the human case, millions of fragile ultra thin sections. For this reason, I don't believe either of these techniques can credibly be scaled up to whole mouse brain level, let alone a human brain level.
This brings us to the two blockface imaging techniques. SBEM uses a diamond knife to scrape an ultra thin layer off the surface of a sample block prior to each, a scanning electron microscope image being taken. That's how it gets its 3D volumes and FIBSEM essentially works the same way except the diamond knife is replaced with a focused ion beam of gallium ions. Both of these techniques avoid the z resolution versus reliability trade off because they do not require handling fragile sections. In principle either might be able to be scaled up all the way, but in practice it is still difficult to see how either could be scaled up to whole mouse brain level, let alone a whole human brain level. Personally, I think the focused ion beam approach has the most potential to be modified into something that could be scaled up all the way, which is why I will concentrate on it for the rest of the talk. Now because it is based on ion milling, which means it has essentially no moving parts, FIBSEM focused ion beam scanning. Electron microscopy can easily achieve isotropic resolutions below 10 nanometers as this volume through an optic lobe of the fruit fly shows. We have even a at at Janelia research campus. We have even acquired FIBSIM datasets with four by four by four nanometer resolution. So no matter what resolution is eventually required for mind uploading, it seems that ion milling approaches will be able to meet it.
A significant limitation, however, is that the width of the sample in the direction of the focused ion beam is generally limited to less than 50 microns. This is because the FIB beam is hitting the sample surface at a glancing angle, which is necessary for the smooth polishing to occur. We get around this limitation by cutting larger samples into thick slabs using a heated and lubricated diamond knife. This is a video of this hot knife cutting procedure. The hot knife sections are typically 20 microns thick, which means they are incredibly reliable to cut and there are comparatively few sections to handle compared with ultra thin sectioning.
And this is an overview of the procedure. A large block of tissue is cut into 20 micron thick sections. These are individually mounted on tabs so that they can be focused ion beam imaged with their thinnest dimension oriented in the direction of the ion beam. And this is what an image of one of these tabs looks like looking down on the surface of it. And you can see that the hot knife cut section's edges are extremely smooth. I've been cutting whole fly brains using this technique and the individual sections are then imaged to cross multiple FIBSEM machines in parallel and the individual imaged sections are then computationally flattened and stitched back together to create a full volume. This is a zoom in movie on one such stitch between two separately imaged hot knife thick sections. You can see that there, there is a boundary right down the middle where there's a stitch line. You can see that, that, that the neuronal processes are easily traced across this cut.
And here's a side view of seven such sections spanning the entire central complex of a fly brain all stitched together. And here's just an example of tracings performed by Brand Jane's group at at Google automated tracing of this dataset. But the Zeiss multicell has a field of view of over 200 microns in diameter, so FIBSEM cannot even clear off a single multicenter field of view and FIBSEM is also many orders of magnitude slower than the multisense imaging that's FIBSEM seems fundamentally incompatible with multicenter imaging, so that, that, that limits us for existing techniques. So I believe that we need a new technique, a new technique for connectomics imaging. I believe what is needed is a top down broad ion milling approach that can mill a few nanometers thick layer off of a hot knife cut thick section.
So for example, you would cut the whole, in this case, human brain, coronally at 20 micron thick slices. Those 20 micron thick slices would each be a put onto a silicon wafer abroad. Ion milling beam would be used to mill just 10 nanometers off of its surface and it would be transferred into a Zeiss multi SEM, surface would be imaged in its entirety, and then it would be transferred back to perform more broad ion milling. And this cycle would be repeated up to 2000 times across this 20 microns of depth in order to create a 3D image of this 20 micron thick hot knife section at 10 nanometer, isotropic resolution. To image a whole human brain this way, it would first be hot knife sectioned at 20 microns thickness into about 5,000 slabs each mounted on its own wafer 5,000 is a large number, but not incredibly large. And these wafers would be shuttled between multisem machines and ion milling machines in industrial scale robotic imaging machines.
It would require several thousand multi sems operating for several years straight to image an entire human brain this way. It would only require one operating for about a year to image an entire mouse brain. So the entire operation in this case for a human brain, well, a modern semiconductor fabrication plant, complete with automatic shuttling of wafers. This is by the way, where I get the $10,000,000,000 figure estimate for the first human upload. Now I've been nurturing this idea of broad ion milling for several years, but I could not figure out a way to get the broad ion milling to work. But over the past year I've finally built a small test rig that I think is finally getting some promising results. And I'm going to show those first here. This is the first proof of principle tests of this broad area ion milling sem technique. So on a single beam system, it's early days on this, but I think this is good proof of principle. This is a volume acquired at less than 10 nanometer isotropic resolution spanning a series of three, one micron thick sections.
So in conclusion, I think there is a path to whole brain, a whole mouse brain and whole human brain connectome mapping that involves hot knife sectioning with Ian Milling. And with the Zeiss multisem, I would like to think that such a technology might actually be the one that enables the first human upload, but we are talking quite a few decades before that would occur. I want to stress again here that every step along this path to the first human upload is a step closer to proving that persons preserved by Aldehyde stabilized cryo preservation can be revived in the future. Thank you.
Allen Sulzen: Immediately following the presentation, there was a question and answer section which was directed by Dr Randal Koene. Let's listen.
Randal Koene: I wanted to start right away with a question that came in from the online viewers. In this case, it's a Michael Andrew actually, um, he's asking the question with the fibsem approach, how much damage to the sample is there from ion scattering?
Dr Hayworth: Uh, yeah. So, um, in terms of damage, uh, I'm not exactly sure what you mean by damage. Uh, it destroys the entire sample while it's being imaged. So there is no going back and imaging. So is there a question from this audience? Daniel asks, would it be possible to collect molecular information from that which is being scattered during the process so that you could use that to further identify what you're scanning at that moment? Uh, yeah, that's a great question. The answer is in principle, yes. But how I like to think of this is that the structural connectome is probably sufficient for mind uploading and that could lead into a very, very long discussion. But in essence, the idea is that the structural connectome has so many, is so high resolution is, has so many correlative features that you should be able to infer molecular information. Now, let's assume that that's not the case though. Let's assume that we go through and do the initial experiments and show that we really do need additional molecular information. That additional molecular information is probably going to be a very specific specific types. It's going to be of a, we need to know the density of ion channels on these particular neurons.
That is almost certainly what is going to come down to. Give me the density of ion channels on particular classes of neurons, and that's the additional information. If that is the case, then I think the more proper way of doing this would be to design some type of tags, molecular tags using immunostaining or something to that effect that would stain the original brain. And this is why I think Aldehyde, stabilized cryo preservation is probably a very good technique to allow this because you preserve the brain with Glutaraldehyde, but then you stop it at that stage before going through any of the other staining steps. And so in the future, 100 years from now, they can rewarm that brain and put through whatever staining they need to highlight those particular proteins as well as the membranes, et cetera. And so I think that is really the way that molecular information will be brought out. It will be highlighted with particular tags that are visible under the electron microscope and during the scanning for the structural connectome, those additional tags will be able to be read that gives these very specific ion channels that are necessary, the ability to see what their densities are in particular places in the brain.
Randal Koene: Thank you. I've got another question. This one is from the online audience. This is a question by Oriseo and he's asking: "Are you joining efforts with Max Moore from Alcor?"
Dr Hayworth: So I, I don't know what's happening with Alcor, haven't talked to them in a while.
Here's my take. I'm not joining efforts with anybody. Actually, I'm trying to get the best preservation possible in hospitals. Whoever does that is fantastic. What I do require for my stamp of approval, if you will, is that, the whatever technique is used, needs to be a demonstrated and published in the literature to preserve the connectome of the brain and whatever technique is applied to human beings needs to have a whole list of quality control checks built in. And I don't see that happening at Alcor. Maybe they could make that happen, but I don't see that happening at Alcor. I don't see that happening anywhere else. And so what I'm really looking for is, I'm looking for the larger mainstream scientific and medical community to buy into this, to say, yes, this makes sense. We need to do it right and we need to integrate it into mainstream medicine. That's what I'm looking for before this is ever applied to human beings. So Aldehyde, stabilized cryo preservation has not been applied to any human beings. I wouldn't recommend it being applied to any human beings before those basic medical procedures have been published and the medical community has had a chance to review them for the quality checks that are necessary.
Randal Koene: Thank you, Ken. We have time for one more question for you. And before I check for another one from the online community, like to, oh, Ted Berger has a question. Okay. Ted Berger says he wasn't sure if he missed something in your presentation, but what he saw was evidence for structure but not connectivity.
Dr Hayworth: So I've looked at these 3D electromicroscopy volumes and I've shown them to many of my colleagues that trace neurocircuits in normally prepared tissue and all of them have said that the connectivity is being preserved in this aldehyde stabilized cryo preservation of these samples.
Allen Sulzen: That concludes today's episode. Videos of these presentations from our workshop series are available at Carboncopies.org under Past Events. If you enjoyed this podcast, please visit us at Carboncopies.org to learn more and get involved.