Mapping, analyzing, and emulating brain computations
December 2, 2022
November 4, 2022
All Captioned Videos Advances in the quest to understand intelligence
Ed Boyden - MIT BCS, MIT MAS, MIT BE, MIT Quest
Jim DiCarlo: We're now going to start shift to the first talk of this session. The next several talks will give you a look at some of the research coming out from Quest and CBMM. And the first one I'm going to give you is Ed Boyden, who's the Y. Eva Tan Professor of Neurotechnology, with appointments in Brain and Cognitive Sciences, Media Arts and Sciences, and Biological Engineering. And among many other activities, he leads the Synthetic Neurobiology group.
ED BOYDEN: Thank you so much. It's really great to be part of this community and see all the synergies between different approaches, trying to understand, build, and really get to the heart of intelligence, with all of its profound insights into the human condition, and also the payoff for humanity in so many realms of science engineering in daily life.
I lead a group here at MIT which works on technology for mapping and controlling the brain. And so one of the efforts that we are participating in is to try to help our community to understand brain circuits at their most fundamental levels. Part of the problem, of course, is that brain circuitry is incredibly complicated.
If you look at the spatial dimensions of the brain, we're talking about enormous biological systems made of tiny parts. So brain cells are so big. In the human brain, they can be centimeters in spatial extent, by far the largest cells in the body.
But the wiring of the brain is nanoscale, huge number of orders of magnitude difference in spatial scale. And if you zoom in to an individual connection between brain cells, a synapse, it's chock full of biomolecules, which are even smaller. Now we're getting down to single digit nanometer size building blocks.
So the brain really presents a unique challenge in terms of understanding how molecules contribute to cells, cells contribute to circuits, circuits contribute to phenomena of great importance like intelligence. So the other issue, of course, in terms of understanding the brain, is not just the spatial complexity, but the enormous temporal complexity.
So brains of course can develop many of their functions, and for that matter dysfunctions, over long periods of time, months, years, even decades sometimes, whether it's learning a certain thing, or a skill, or a language, or there's lots of things that take time. But the quantile building blocks of brain computations are very, very short events, millisecond time scale electrical pulses within brain cells, and chemical exchanges between brain cells.
So how can we map and control across space and time so we can understand how brain circuits will work? So I'll tell you two short stories today, one about space and one about time, where we at MIT and in this wonderful collaborative environment have been working on strategies to try to help the understanding and repair of the brain across these spatial and temporal dimensions.
Let's start with space. So, of course, there are ways of imaging the brain. I think we've all seen images like this, brain scans. They're amazing. They can be, of course, noninvasive, which is one reason why they're so powerful. And, of course, you're hearing from many other people here involved with Quest about such things.
But they can't get down to the fundamental, nanoscale wiring of the brain, the individual connections, individual synapses, and so forth. At the other extreme are microscopes, which can see very tiny things. But even they can't see the most fundamental building blocks because light has a finite size, or wavelength, and you can't see things that are much, much smaller than that. People have tried to overcome that limit, but these techniques are difficult to apply at the scale that the brain operates on.
So here we often try to think of the opposite of what people do. People have been zooming in on biological systems for literally 300 years by magnifying images. What if we magnify the brain itself? And so we started thinking about swellable polymers like the stuff in baby diapers. Add water, and as this cartoon shows, a baby diaper will swell, an experiment that millions of babies do every day. And those white threads, the polymer threads, will expand apart from each other.
And, importantly, those polymer threads are nanoscale and spaced by nanoscale distances. So we started wondering if we did install this chemical spiderweb-like mesh of baby diaper polymer inside the brain just right, add water, could we physically magnify the brain?
So that's the basic idea is to chemically weave this baby diaper material inside brain cells and outside brain cells, in between the biomolecules, and around the biomolecules. And, amazingly, in early 2015, we discovered that we could do this.
So in panel B is a piece of the mouse brain, a few millimeters on a side. Panel C, the same piece of the mouse brain about a day, day and a half later. We physically expanded the thing by 100 times in volume about 4 and 1/2 times in each direction.
And the polymer thread, I guess we can use the mouse pointer-- it starts out very densely packed. The spacing is around the size of a single biomolecule. And at the end of the day, it has become dilated in size. So here's a little movie of a piece of the mouse brain which had the baby diaper material installed earlier.
And this is a sped up movie, so it's half an hour in half a minute. But you can see this polymer-sized piece of brain tissue from a mouse expanding as we add water. And expansion, we verified through many studies, is accurate down to the nanoscale, down to, effectively, the size of individual molecules we think.
So just because it's a weird way of doing microscopy, we made it a little animation just to emphasize what we are doing at the molecular scale. So we can take a brain cell like the one that left, this golden neuron. And we're pulling the building blocks of life apart from each other.
So we end up with a constellation of biomolecules that have been pulled apart by this baby-diaper polymer, sodium polyacrylate, and we can then see their organization. So one of the things we're very interested in is, indeed, looking at brain circuits and trying to map out how they are connected, how molecules are organized at those connections, and along the wiring of the brain.
And that could lead to several kinds of interesting contributions to this community. One, of course, is to look for interesting motifs or organizations of neural circuits. One could also try to work on modeling what neural circuits are doing at different levels of granularity, informed by different hypotheses and top down ideas to test. And the hope, of course, is that we can use such detailed mapping strategies to help out with different kinds of missions that the Quest for Intelligence is spearheading.
Now, of course mapping is only able to see things. You can't really prove causality only by a static map. These are not even living systems by the time you've expanded them. And so the second short story I want to talk about is about control of the brain.
Suppose you have a map. And you want to figure out whether a certain part of the circuit contributes to a certain function. You could try to use different modeling strategies to infer where in the circuit to perturb. That's one of the beauties of making such maps. But, in the end, you probably have to test it with an experiment. The brain is such a complex mess, we want to prove and not just speculate as to what is mediating a given function.
And so one of the things that, of course, we have to confront is the high speed time dynamics that underlies brain functionality. And to do that, we've been working for many years now on this technology where we can activate specific cells in the brain with pulses of light.
So brain cells compute using many kinds of signal. But one of the most prominent is electrical signals that are very, very fast, biologically speaking. So if we convert light into electrical signals, say by installing little solar panel-like molecules in brain cells, shine light on the brain, which we can bring in using optical fibers or other techniques-- and many of us in this community are working on different ways of delivering light into the brain with interesting patterns and so forth-- then we could control specific cells, turn them on or off, and figure out what kinds of processes they could trigger.
So the basic idea is that there are all sorts of different cells in the brain. Some are big. Some are small. Some excite. Some inhibit. The list goes on and on. And we don't even have a full list of all the cell types of the brain, of almost any species except for a small worm, which has only a couple of neurons.
But suppose that we can take this technology which we call optogenetics, opto for light, and genetics is genetically encoded, we can express it in certain cell types using all sorts of genetic tricks pioneered by many people in this building and around the world. Then you can aim light at a single cell, or a set of cells as you see in this cartoon, and turn those cells on or off, depending upon the identity of the molecule.
As this strategy is being very, very widely used throughout neuroscience to activate and silence brain cells and figure out how they initiate, sustain, or are needed for different kinds of behavior. I'll just show one example because it's got a cool movie associated with it, but one of the things that many people are interested in is about what facilitates learning. And can you modulate it? Can you control it? What are the parameters that are important?
And so here's this a very simple experiment where dopaminergic neurons deep in the brain of a mouse are made sensitive to light. An optical fiber was implanted into them. Every time the mouse goes to the right side of the box and pokes its nose at a sensor, it will get a pulse of light. If it goes to the left hand side of the box, then nothing will happen.
So here's the movie. The mouse pokes its nose, gets a pulse of light. You can see the optical fiber lighting up. And basically this mouse is working for light. It will do this over and over again for a very long period of time, actually.
So one of the hopes here is that through causal perturbation, one can then investigate what a model or what a map might predict in terms of an interesting point to intervene, and allow you to prove out a specific consequence, which then, of course, could help close the loop and feed back upon new ideas so that the cycle continues.
But it's also interesting because just last summer, interesting plot twist occurred, was that a European team used one of the molecules that our group reported in 2014 and showed that it could be expressed in human nervous system cells. They took a person who is blind, who lacks the photoreceptors of the eye, and could make the eye see again by making the spared cells of the eye sensitive to light, basically installing a genetically-encoded camera, if you will.
And so it's very intriguing to think about how the kinds of tools that we all are developing here could potentially not only help drive the science of intelligence, but we know a lot more about the eye, of course, than we know about the rest of the nervous system and the brain.
As the knowledge of maps and models matures, could we have other kinds of optical-neural interfaces to the brain that can be used to drive other functions? And with that, I think I'm out of time. And I will just acknowledge that this describes a huge amount of work by many people.