Tools for mapping and repairing the brain [part 1] (25:33)
May 24, 2016
July 8, 2015
All Captioned Videos CBMM Summer Lecture Series
Ed Boyden, Professor of Biological Engineering and Brain and Cognitive Sciences at MIT, leads the Synthetic Neurobiology Group, which develops tools for analyzing and repairing complex biological systems such as the brain, and applies them systematically to reveal basic principles of biological function and to repair these systems. In this three-part lecture, he discusses tools for mapping and repairing neural circuitry using expansion microscopy (part 1), whole-brain imaging with light-field microscopy (part 2), and optogenetics (part 3).
Ed Boyden’s Lab website: The Synthetic Neurobiology Group Karagiannis, E.D. & Boyden, E. S. (2018)
Expansion microscopy: Development and neuroscience applications, Current Opinion in Neurobiology 50:56-63. Klapoetke, N. C., et al. (2014)
Independent optical excitation of distinct neural populations, Nature Methods 11:338–346. Prevedel, R. et al. (2014)
Simultaneous whole-animal 3D imaging of neuronal activity using light-field microscopy, Nature Methods 11:727-730. Tye, K. M. & Deisseroth, K. (2012)
Optogenetic investigation of neural circuits underlying brain disease in animal models, Nature Reviews Neuroscience 13:251-266.
ED BOYDEN: So what our group does, I like to say that we kind of work backwards from the big problems. They're big problems to understand how networks of neurons work together to implement a thought or an intelligent behavior. Can we figure out how cells and molecules within cells work together to store memory?
Can we figure out how even maybe neurons and non-neuronal body parts interact? People are now studying how bacteria in the gut might influence cognition and so on. So how do we wrestle with this immense complexity? Is this the room mic? This is the room mic. OK. What I'm going to do is, if I cough and I try to disconnect right away and then put it back on because I think it's faster than muting it. OK.
So one reason why the brain is so hard to understand is that the brain operates across vastly different spatial and temporal scales. A brain circuit is a large thing. We have many neurons that extend centimeters. If you count down the spinal cord, even a meter in length. And yet, the actual connections between cells and the molecules that those connections that govern what neural connections do, are nanoscale in dimension.
And so arguably, if you want to understand a brain circuit, you need to understand what those nanoscale building blocks look like and what they're doing. But across these vast scales which is really, really, really difficult. The other big problem with the brain is time. So the briefest events in the brain-- millisecond time scale, electrical events such as the action potential, millisecond time scale, releases of chemicals such as nerve neurotransmitters are really brief.
But if you want to understand how a memory forms or how Alzheimer's disease progresses or how different kinds of acquisition of a skill occur, these take years to occur sometimes. And so you have these vastly different temporal scales between the briefest events that make the brain do what it does and the long timescale events that especially in humans give us a lot of the attributes that make us interesting.
So in our group, our approach is really to think about how do we work backwards from the vastly different spatial and temporal scales that need to be understood to understand brain operation. But [INAUDIBLE] are the big classes of questions we think about. And then we survey all of engineering-- chemistry, photonics, micro fabrication, you name it.
And then, and only then, do we start actually inventing tools to analyze and fix brain circuits. So today, I want to tell you three short stories. One on mapping the brain. One on recording the dynamics of the brain. And one on controlling the dynamics of the brain and how these tools work and how they're being applied sometimes at very early stages to try to reveal the substrates of computation, such as those that can implement intelligent behavior.
So let's start with mapping. Mapping is one of the most difficult things to do because you, again, have to map these gigantic volumes, both nanoscale precision. And so many attempts have been made to try to tackle this. In fact, last year, some of the inventors of so-called super resolution imaging methods won the Nobel Prize in chemistry for their work.
But nevertheless, although it's now possible to see very small things, these technologies don't go fast enough to let you imagine large objects with that small degree of precision. So in our group, we started thinking about this about-- actually, about eight years ago. How do we possibly understand a large object with nanoscale precision?
And we decided that the problem with nanoscale imaging is that a lot of the tricks people use are very slow, but regular imaging-- I think, if you look on a phone, there's a very inexpensive camera that's part of it. And that thing can acquire some significant fraction of a giga pixel every second if you run it in video mode.
So if you allow yourselves to use classical diffraction limited optics like lenses and so on, you can go very, very fast. So then the question became how could you image at really high speeds using cheap optics that's scalable but still be able to see nanoscale things? Now, if you take your camera like the one on a phone, it can't see nanoscale things because light has a finite wavelength around a couple hundred nanometers.
And so that means you can't really see things that are under that size, which is right where synapses and neurotransmitters and ion channels or receptors get interesting. So we thought, this is work that's been driven in our group by two graduate students, [INAUDIBLE] Chen and Paul Tillberg. Why don't we just blow everything up and make it bigger? And so the concept really borrows from 1960s, 1970s polymer chemistry.
Back in the day, there was a very interesting and exciting field called responsive polymers or smart gels. And these are polymers that vastly changed their shape or size when you exposed them to an appropriate environment. So one of these polymers is sodium polyacrylate, which some of you might know as the active ingredient in baby diapers.
And it's a very dense polymer. When you add water, the sodium gets washed out, and then what happens? Well, acrylate has a carboxyl group, which has a negative charge. And so if you wash the sodium away, you get all the positive charge out and the negative charge is stuck behind. And negative charges that are anchored to each other are going to repel each other, but there's really nowhere for them to go except away from each other.
And so what happens is, if you have a lot of negative charge on a polymer, it's going to expand and repel. OK. And realizing that the tape's going to get everything. So some editing might be required.
So what we did, was to take a piece of brain tissue and embed it in a dense polymer of sodium polyacrylate. And by embed, I mean that the polymers are winding their way through the tissue, between the molecules, inside the cells, outside the cells, everywhere in the sample. This is obviously a preserved tissue not a living tissue.
And then when we added water, it was pretty striking. This kind of worked on the first try almost. It grew. It physically expanded and became about 100 times bigger by volume, and it was pretty amazing actually. Now, there were many problems. It took us a couple of years to debug this because although the growth could occur, we often had cracking or other problems. And that makes sense, right? The brain has a lot of structure to it.
It's not going to just let some piece of baby diaper push it around, right? It's going to resist. And so it took us a couple of years to develop and refine the chemistry and validate it until we finally arrived at what we call expansion microscopic. And here's how expansion microscopy works.
The first step is very traditional. You preserve the tissue. You do that with traditional chemicals. Then you bring in a somewhat complex looking label shown here. And what we have here are a couple of components. One is an affinity tag. That is, it'll bind to a target, a neurotransmitter receptor, some kind of bio molecule. The second thing is a fluorophore, a molecule that glows so you can see it.
And the final thing is a polymer linking group, something that looks a lot like the actual monomers that make up the polymer chain, but allows us to bind to the polymer. And so if you think about it, what this means is, we can anchor a fluorophore. I'm losing my voice very rapidly right now. We can anchor fluorophore at a site in the polymer chain that's defined by where the affinity tag bound to its target.
So you can imagine bringing in an antibody that binds to a neurotransmitter receptor, anchor the fluorophore to the polymer, and now you can do the polymerization and the formation of the polyacrylate mash. And by the way, the mesh size, the spacing between polymer chains is really tiny, maybe only a couple nanometers, the size of a bio molecule.
Next step then, we use an enzyme to chop up the tissue gently. And so now, all of the chemical structure has been homogenized. There's nothing to resist the expansion. Finally, we add water and the polymer will swell.
So let's go through that with a slightly different set of slides. Here's our, as you can see, very artistic rendition of a neuron. We bring in that antibody or affinity tag to label a particular biomolecule. We form the polymer and we anchor the fluorophore to the polymer. We digest the original biomolecules to homogenize the mechanical properties. And then we add water, the polymer swells taking all of those anchored labels with it.
So this works. Here is just a cell in a dish, and we've labeled the cytoskeleton with an antibody against tubulin. So these are the microtubules. The blue scale bar means pre-expansion, and this is after expansion. That's what the orange scale bar means. So this is 4 and 1/2 times bigger in all directions, but 100 times bigger in volume than this picture. But we shrink it down so you can compare visually what the two images looks like.
So they look pretty similar, but we want to be quantitative. And so we adapted from image analysis, this vector field calculation, and the way this works is you take your pre-expansion image and your post expansion image and you overlay them. And then you nudge them just a little bit until they overlay. And the amount of nudging that you do is a local measure of distortion, and the amount of nudging, we plotted as this little vector fields. These are tiny little arrows.
Then if we want to calculate the error of a biological measurement, all we have to do is to pick two points in the image, and then integrate over that vector field. And then we do that, we can calculate the root mean square error of a file of a biological measurement as a function of the measurement length. And it's very, very small, around 1%.
So for almost all biological and medical measurements, the 1% change in the length is not that important. Most people want to know whether things are nearby, co-localize, the organization, the topology. And I should point out the dotted line here is the resolution of our microscope. So it's quite possible that we're taking more blame than we need to, but it's a new technology. We want to be as conservative as possible in our claims. And so even if we take all the blame for the distortion, it's only about 1% or so.
Now, what about down to the nanoscale? So to do nanoscale comparisons, we have to take a pre-expansion image using an older super-resolution method. These are, of course, are very slow, but they work. Then after expansion, we take an image on a regular microscope, and then you can actually zoom in and analyze the distortion the same way.
So if we calculate the root mean square error as a function of measure at length, we get an error of around 60 nanometers or so. We can also, because we know the size of these microtubules, compare before and after. You draw a line through the microtubules is in this area before expansion, looks really blurry.
But if you draw the same line after expansion, you see three peaks. That's because they're three microtubules there. And since we know how wide the microtubules are, we can calculate the distortion of our energy method. And again, you get around 60 nanometers or so. This works in intact brains as well, of course. Otherwise, I wouldn't be telling you.
So as you can see here, this is a piece of mouse brain. And the label that you see that's glowing is yellow fluorescent protein being expressed in a random subset of the cells. And there's a little bit more distortion around 2% to 4% because the brain has more structure than the cells in a dish.
But still, 2% to 4% is negligible for the vast majority of biological and medical measurements. That said, it is important for some measurements and we are working to make the polymers even more even. So this is now where the fun part begins. We can kind of zoom in. And then maybe turn the lights off here. Is that the right one? I'm not sure.
Well, here's a piece of brain again. Now we're going to zoom into that little square, which is what you see here, and now you can see two neurons and some purplish fuzz. And then we're going to zoom in on this little square and now you can see one neuron branch, and slightly larger purplish fuzz.
That's pre-expansion Post expansion though, you can see that the purplish fuzz is not actually purplish fuzz. There's a blue part and a magenta part. Blue, magenta, blue, magenta, blue, magenta, blue, magenta. Those are synapses. And in fact, we can zoom all the way and see a single synapse. The blue is the presynaptic protein bassoon. The magenta is the post synaptic protein homer 1A. And if you measure the distance between the centers of those, you get exactly what you'd expect based upon the classical literature, except that now you can do all these imaging methods using cheap scalable optics.
So the take home message of this first prototon is that now we can do 3D, large volume, scalable imaging with nanoscale precision. And so here's a piece of the mouse brain hippocampus. And we've label, again, a random subset of neurons. Now shown in green.
And then blue and magenta are, again, pre and postsynaptic molecules, bassoon and homer 1A, and this is a $30 billion Voxel image in three colors taken on a regular microscope just over at the Whitehead core facility across the street. And we can zoom in onto some branches of neurons, and then zoom in onto a single branch.
And now, if you look carefully, these are dendritic spines. They're one of the major components of extendatory synapses in the hippocampus and cortex amongst other places. And you can see the magenta post synaptic densities there. And if you look carefully, the blue presynaptic densities in opposition.
And then finally you can zoom into one synapse, and this is a very interesting synapse. This is one presynaptic terminal. You can see multiple blue parts though. This is one presynaptic terminals talking to multiple post synaptic terminals. And you can see multiple blue spots with multiple magenta compatriots next to it.
Now, one nice thing about expanding issues is that since you're filling them with water, they become completely clear. So here's a piece of mouse brain before expansion, and here it is after expansion. You can't even see the boundary. It's somewhere in here. That's not the brain boundary. That's where the air meets the sample.
So somewhere in there is the brain. And it makes sense, right? You're expanding it and filling it with water. So it becomes clear. And when you clear things, then you can even scan faster because now you can do interesting methods microscopy like so-called light sheet microscopy. You scan a whole sheet of light through and then you take a picture at 90 degrees to the sheet. So it's like you just-- suppose you have a sample here, we illuminate it from the side and we just illuminate an entire sheet through it, and you take a picture from the top. So we can take an entire 2D image at once. There's all sorts of fun things to look at now.
Here are two neurons in the cortex. And the red stands for GAD65/67, which is one of the enzymes that make inhibitory neurotransmitters in the brain. And you can see the inhibitory synapses. And you can see how they're arranged on the cell body. Some people think that's because it gives the inhibitory neurons a kind of veto power over the neural activities that others have hypothesized this allows for certain kinds of oscillations or dynamics that are important for attention or for feature selectivity.
I always like this image. This is the nucleus of the neuron. So in genomics, most people think of a genome as a long string of letters. But every one of our cells has this meter long string of DNA all wound up in a tiny compartment. And here we've labeled the nuclear envelope, the boundary of the nucleus.
And it has this really weird shape, part of it is smooth, part of it is ruffled. And so one idea now, and people have seen these kinds of things in populations of cells before, but now we want to see how it happens at the nanoscale is how does the genome coil up in three dimensions? Does it help us understand how memories are encoded or other changes that are known as epigenetic changes?
Now, one of the most exciting-- so expansion is exciting for two reasons. One is because you can get nanoscale imaging across 3D volumes. That by itself is pretty cool, if I say so myself. But there's another thing that is only starting to be appreciated. When you move all the molecules in a sample away from each other, you make a lot of room around often.
And with all that room, you could bring in little barcodes and attach them to different molecules, and so that lets you have essentially an infinite number of colors for your imaging. So most imaging systems only see a couple of colors, maybe three, four tops. The really advanced ones maybe you can get up to 10, but it's very difficult to do that. Now, let's go back to this picture. I glossed over it before, but now I want to go back to it and tell you a different story about it.
We have an affinity tag like an antibody that combined to a neurotransmitter or a receptor or an ion channel. This molecule here that binds it all together, is actually DNA. And if you think about it, DNA encodes information. So imagine you have 1,000 different tags to bind to a bunch of different ion channels, receptors, transmitters, you name it, everything you want to know about the brain. And each one has a different barcode on it.
And there's a lot of room around all the molecules. So you can take advantage of that room and bring in all these barcodes. So you anchor the barcode to the polymer and expend all the barcodes away from each other, and you can read the barcodes. So therefore, kinds of DNA base ATCG, if you have 10 bases long barcodes, there are four to the 10th power, or one million approximately, different barcodes. You could tell a million different things apart from each other.
Now, how do you read the barcode? That's obviously a difficult thing to do. So it turns out, when you sequence DNA, suppose my arm is DNA, when you sequence DNA, you're copying it. And as you copy it, when you add a new base to the DNA and you copy it, it's fluorescent. And so as you copy a piece of DNA, over time, it's basically blinking out its code.
And so something we are now exploring is whether-- so this is in collaboration with the church lab at Harvard, is whether you could take the barcodes and anchor them, expand them, and then sequence the DNA right there in the sample, not like regular DNA sequencing where you grind everything up, and you put into a machine and it emails you the results a week later. Here, you want to image it. You want to see the sequences blink out over time.
So then, not only can you see large objects of nanoscale precision, but you might be able to do it with an infinite number of colors, enough to distinguish all the different molecules that make a neuron do what it does. And so one thing that we're starting to think about is whether we could actually map an entire brain. If we know where the key transmitters, receptors, channels, and so on are, we actually simulate it in a computer.
So can we really get down to the fundamental circuits and mechanisms and see how they actually operate? To do that, of course, we're going to need to expand much bigger. And so we are working to make the expansion bigger, although this is still very early stage.
So that's the end of part one, on mapping. Great question. Yeah. Can we catch information that we don't know we're looking for? It's very difficult, because if you don't have an affinity tag that binds it, how do you really know that you're going to get it? So I would say that's a very difficult problem. I think what's going to happen is suppose we can map a brain circuit.
We try to simulate in the computer, it can recapitulate some features but not others. OK. Now, we have to make a hypothesis. Maybe it's nitric oxide or gap junctions or-- there's a lot of stuff that the cellular neuroscientists have discovered. We have to then build a tag to label those, and then repeat. But at least though, the technology is extensible.
Why I like this technology is that unlike some methods of microscopy where you only see three colors and that's it, you can never go beyond that, or 10 or 12 whatever it is or some only one actually. But this you can eventually, as long as your imagination can help you build a tag, you might be able to label it and hunt it down and find out where it is. Also this is really, really cheap and it doesn't take much time.
So it's really easy to try and many, many groups have written to me to say, oh, yeah, this undergrad in our group tried it out, it worked on the first try. We're starting get data. And so if you do try it out and you miss something, OK, it's exploratory, but it doesn't require a huge investment and that's also important when doing high risk, high reward kinds of research.
So the church lab about a year ago, and also a group in Sweden, the Neilson Group, about a year before that published papers on what they call in situ sequencing. So the basic idea is take a cell, anchor the RNA or DNA or whatever you want to sequence to the molecules inside the cells so they're stuck, and then basically do what's called sequencing by synthesis. So the sequencing by synthesis is basically that concept that we were just talking about.
So you have your piece of DNA, you want to sequence it. So what you do is you add four bases each with a different fluorophore, blue, green, red, and infrared. And let's suppose that you have an A here. That's right. So we have an A coming in to bind. And that's red, let's say. So it will come here, this little dot will look red for a while. And then you destroy all the fluorophores by bleaching it, by shining bright light on it, or maybe you cut off the fluorophore, and then you wash with another one.
So the next base that comes in is T, that's blue. So now, the same dot that you saw that was red earlier, now it's blue. And so over time, you take a movie and the dots will change color-- blue, red, blue, green, blue, red, red, red, blue, purple, whatever. And so by doing this reaction many times in a row, a dot will change its color and you can read it over time.
So what we're doing now is working on incorporating these sequencing methods into the expanded state, which is important too because by making more room around these tags, we think the sequencing will go better. Right now, the sequencing is pretty low yield because when you're in the cell, it's so compact. There's so much stuff there.
The answer to the first question, can we watch it expand and look at the properties? You can do that. We haven't tried it though. And the answer to your second question, are you asking whether the brain can be alive after the expansion or-- so far we cannot figure out how to do this in the living brain. Because even if you could keep the cells alive after expansion, everything will be so far apart. They will not single properly.
So it's an interesting question, though. I mean, an idea that we have had is take a brain, expand it, map stuff, and then could you shrink it back down again and would it run again? I don't know. That's a very interesting question.
But it's interesting, you can imagine taking a brain, expand it, map it, upgrade a few key receptors by putting in receptor 2.0, shrink it back down, and now you've got an upgraded brains. But I don't know. This is going on the internet, right? Just kidding. No. I think it's a fun idea, and actually we thought about also, could you imagine expanding a brain and then bringing in computer chips or digital parts, and then shrinking it back down again, and seem like a digitally upgraded brain.
Associated Research Thrust: