Boyden: All right. I direct a group at the MIT Media Lab that’s working on neurotechnology, tools for mapping how the brain works. Can we have the slides please? Okay, good.
Our group at the MIT Media Lab is working on how the brain works, and this is a big mystery in part because the brain is incredibly complicated. So although this three-pound universe in our heads is generating our thoughts and feelings, our emotions, our movements, and decisions, we really don’t know much about how it’s actually doing these manipulations of information.
Now, neuroscience as a field is over 100 years old, and so we do know some things. And one thing we’ve learned is that if you can zoom into the brain, and our brains are made of 100 billion cells called neurons, you’ll find an incredibly intricate circuit that’s made out of these neurons that are connected in very dense operating networks. So within a cubic millimeter of your brain you’ll find perhaps 100,000 neurons with a billion connections between them and each of those cells and each of those connections computes using millisecond time-scaled pulses of electricity and chemical communication.
One of the big questions then is how do we possibly understand what’s going on? If you think about it, that cubic millimeter of brain tissue, and your brain has on the order of a million of those cubic millimeters, is doing on the order of some fraction of a trillion things per second. And so it’s incredibly dense, it’s incredibly high speed. Now, one approach, of course, is to make hypotheses about how the brain works and to try to test them. But the sheer brute force complexity of what’s going on really demands new technologies. There’s a long list of things we don’t know about the brain. We don’t have a parts list to the brain. We don’t know how many kinds of cells there are in the brain. We know the cells in the brain differ in their shape and size and how they are comprised of different molecules and how they change in different diseases. But we don’t even have a complete comprehensive atlas. We don’t know how these cells are connected and what molecules are at those connections. Finally, we don’t have the ability to record and control the dynamics, important if we want to understand how information’s flowing and to fix brain disorders by repairing them.
Now, why do we want to do this? Well, if we can understand how the brain’s computing, then it might be very helpful for us in our quest to understand something about the human condition. What is a feeling? What is a thought? Can we understand the basis of irrationality? Why do people make the decisions the way they do? It also could eventually help us understand new models of computation. Can you build artificial intelligences or algorithms that can do things that are sophisticated the way the human mind can? And, perhaps as importantly and pressingly, there’s an incredible medical need. So the number of people around the world who have sought some kind of attention for a brain disorder is going up over time, and if you look at the list of the diseases on this slide, they not only change the time we have to live, but they can actually change who we are, how we relate to our loved ones, our identities.
Well, one aspect of these brain disorders is that we know from the fact that the brain is this network of cells that all these brain disorders are linked into specific changes in regions of the brain. Although, again, we don’t know the parts list of the brain or the connections, and so the exact mechanism of a brain disorder, any brain disorder, is still something of a mystery.
Surprisingly then, the therapeutics that have been developed in the 20th century to treat brain disorders have been a little bit like bathing this complex circuit in a chemical. In fact, literally that’s what a lot of these therapeutics are. And not surprisingly, that means that the time that it takes to take a brain drug out of the lab and to help people is fairly long, almost a decade, and the failure rate’s very high, almost 100 percent actually, and the cost is enormous, a significant fraction of $1 billion.
And so, if you think about it and you look at the history of biology, which I think is a great way to learn about its future, you can see that a lot of the therapies developed in the 20th century were totally found by chance. You know, somebody was trying to figure out whether the urine from a schizophrenic patient could induce schizophrenia in an animal, they needed a control experiment so they bought some lithium urate, next thing you know lithium is a treatment for bipolar disorder. And that’s sort of how a lot of these therapeutics were discovered, total serendipity.
So our approach is a thing from ground truth principles. Can we build tools that will allow us to map the brain and figure out its fundamental molecular composition and structure—that’s sort of the yin of neuro technology—and the yang, from those maps, from those scientific ground truth principles, can we then devise optimal technologies that help us repair the brain. And this is a long-term vision. Right now we are engaged in the process of building the tools that will let us map circuits with nanoscale precision to find the connections in molecules. We’re working on new kinds of technology to let us observe the dynamics of the brain. We’re working on new strategies that will let us control brain dynamics in order to repair the computations that go awry in conditions like Parkinson’s and epilepsy. So I want to tell you about a couple of these technologies that we’re developing and why they’re important.
Now, one fundamental tension about understanding the brain is you have to understand all these different scales simultaneously. The connections between cells in the brain that exchange information in the form of chemicals are miniscule, nanoscale, and yet, the circuits that process a sensation, or that drive a movement, or that helps us process thoughts and feelings are enormous. They can span centimeters. And so you have to deal with this enormous vast difference in spatial scale all at once.
So our group is developing new kinds of microscopes that can scan with nanoscale precision, but at rates that range from thousands to millions of times faster than existing microscopic techniques. And so our hope is that we can actually capture these large-scale circuits with nanoscale precision in ways that have never been possible before.
So what you see here is a preserved piece of brain tissue—this is from a mouse, one of the most common models in neuroscience. At the top you can see the neurons in green, and the pink and blue dots are representative of the connections between cells, and of course you can’t see the actual connections because the screen cannot display them. At the bottom though, we’ve zoomed in on a tiny, tiny part of the brain and you can actually now start to see those individual dots are discrete. And so our hope is that we can start to build technologies that will let us map entire circuits in the brain, and maybe even entire brains, but with molecular precision.
We also have to have labels to make the invisible into the visible. A lot of the molecules in the body from an optical standpoint look very similar. We have 30,000 or so genes in our genome, at least 30,000 gene products, or proteins, or enzymes, and yet, if you look at a cell under a microscope, it just looks like a little blob. So we need to build tags to allow us to barcode these molecules and make them identifiable. And so we’re working on a wide variety of tags that allow us—here I’m zooming in to one of the tags we’ve been working on that labels the boundaries of cells in order to make them standout. And by multiplexing them, by using barcoding technologies that we and others are developing in order to make these things standout, we hope to be able to look at thousands and maybe even tens of thousands of different molecules at once to tell them apart.
Now, the nanoscale structure of the brain we think can be very useful. Imagine loading this into a computer and trying to simulate the dynamics that result. Could you actually try to replicate the sequences of electrical and chemical activities that might yield a pattern of thought or movement or feeling? But if you think about it, a static nanoscale snapshot is great, but you also need to know the dynamics, the initial conditions that send it on its way. To do that, we’ve been working on a collaborative project, this is with the Vaziri Lab in Vienna, to build microscopes that can see in three dimensions, but at speeds fast enough to capture the millisecond time-scale activity patterns of neurons in the nervous system. And to do this we take advantage of a really old trick. In fact, it’s been around in various incarnations for decades, and it basically it means to copy the fact of how our visual system sees in three dimensions. So we have two eyes and they’re separated, and so each eye captures a different view of the world. And so our brain can extract a three-dimensional image from static snapshots that our eyes are constantly taking. So what we’re doing here is actually using this in a microscope. We can take a microscope that can image, insert a couple of lenses, as you can see in the middle there, into the middle of the microscope, and much like how we have multiple eyes that allows us to see in three dimensions, we can now take, with single camera snapshots, pictures of these brain circuits in three dimensions.
So recently we collaborated to actually apply this technology to try to record all the neural dynamics of a small organism, the worm C. elegans, which has yielded many insights into how neurons compute including, for example, a Nobel Prize for the discovery of apoptosis, how cells undergo controlled cell death, which is important from topics from development to cancer. And what you see here is a worm, C. elegans, which has 302 neurons, and they’ve been genetically engineered to contain molecules that will fluoresce, that is they will emit light when neurons are active in certain conditions. And so if you look carefully, I think you can see that these neurons, these little dots are blinking at you. And that’s because we can now capture neural activity throughout entire organisms because we have a fast 3D imaging system.
So one of our hopes, of course, is to scale this up. What about the human? Well, human brains are enormous, right? If you were to scan a human brain with a certain degree of nanoscale precision, that might be on the order of 10 million terabytes of data. If you took one-terabyte hard drives and stacked them up, that tower would reach into outer space. So one strategy, of course, is, again, taking a cue from the history of medicine, is to look at model organisms, organisms that recapture some of the things we want to know, but are far simpler. And so one project we’ve launched at MIT is to start studying the world’s smallest mammal. And so you can see it here, that’s the Etruscan shrew. We’re collaborating with the Michael Brecht’s group, which has been studying their behavior for a decade in Berlin, and one of the things we’re trying to do is to figure out if we can make tools to allow us to record all of the activity through its entire brain and also to allow us to map it with molecular precision.
Now, this brain has a lot of similarities to the human brain. It has a cerebral cortex like we do with six distinct layers. It has regions like the hippocampus, which is involved with episodic memory, and there’s a lot of homology between its brain and our brain and it can make fairly intelligent decisions as well. So one of the things we want to figure out is whether we can use this as a way to learn the principles of mammalian brain operation and possibly learn a lot about the algorithms the brain is using to compute. I mean these animals have to sense, they have to make decisions, they have to hunt and eat insects bigger than them, and so one hope is that we can actually learn a lot about how small neural networks in mammals are making decisions and extracting from the analog visual world symbolic representations and other kinds of complex computations that are important for intelligence.
Finally, we want to be able to control the brain in order to fix its computations. And so one technology we’ve been working a lot on, which has come to be known as optogenetics, allows us to control neural computations using light. So if you recall, I mentioned that neurons compute with electrical impulses within them. If we can take little tiny solar panels and install them in neurons, then shining light on them will allow us to control their electrical processes within. And the brain doesn’t feel pain. Hundreds of thousands of people have neural implants of some kind to stimulate neurons, and similarly, we can put an optical fiber in the brain in order to deliver light into the brain.
Now, it turns out that these little solar panels exist all over the world, all over the tree of life, and they mediate various kinds of photo sensation and photosynthesis. For example, there’s a single cell plant, a green algae, that has a little eyespot that lets it swim towards light. And if you zoom in, you can find molecules that act a little bit like those solar panels. They convert light into electrical signals. It’s a protein, so we can take the gene that encodes for it, put it into a neuron, or set of neurons, and now if we shine light on those cells we can activate them.
I’ll give you just one example of how thousands of groups around the world are now using our technology in order to study how the brain works and potentially prototype therapeutics in order to fix brain dynamics in disease states. One example of an area that many people are interested in is how learning occurs, and so many groups are trying to activate different sets of cells in the brain and to see if you can drive learning, can you make the brain do more of what it was just doing. So collaborating with Chris Fiorilla’s group in Korea, they’re very interested in a small cluster of cells deep in the brain known as dopamine neurons. These are neurons that are commonly referred to as the pleasure center of the brain. They’re implicated in phenomena ranging from learning to addiction, and if you can activate them directly with light, you might be able to figure out what they actually can do.
And so here’s a little movie from this collaboration. This is a mouse. Its dopamine neurons have been made sensitive to light with our technology and a fiber implant in the brain, and every time the mouse pokes its nose into a little sensor it gets a pulse of light from a laser, down an optical fiber aimed at this little tiny cluster of cells. And so as you can see, this mouse is basically working for light. It will poke its nose in the sensor, get a pulse of light, think “Hmm that was good,” and it will do that over and over again. So you can really try to figure out what specific kinds of cells in the brain do. And potentially, if you can turn them on or shut them down, they’ll cancel out pathological activities. So many groups are using our technologies, for example, to try to shut down the overactive neurons in animal models that have epileptic seizures in an attempt to make a human therapeutic that would let you shut down seizures in human patients with epilepsy. That’s just one of many examples out there.
In the long run, if we have maps of the brain and dynamical models of the brain and the ability to control brain circuits, we’re hoping to develop a generation of tools that we call brain coprocessors, technologies that can read from the brain, compute information the brain needs, and write to the brain. And so we’ve been building nanofabricated devices that can record from the brain in three dimensions and stimulate the brain in three dimensions in an attempt to prototype this.
When it comes to humans though, we also want to be very pragmatic, and so one thing we’ve recently begun is to try to figure out whether these probes can be introduced into the brain without neurosurgery. So these tens of thousands of patients I mentioned earlier could have electrodes implanted, for example for Parkinson’s disease, through the skull. Those electrodes will displace brain tissue. But what if we could take this electrodes and bring them into the brain through the bloodstream and park them in the vasculature of the brain so that we don’t actually have to drill a hole in the skull, we don’t have to splice brain tissue, and in contrast to the thousands of people who get brain implants each year through the skull, millions of people have things inserted like stents into their bloodstream. So we think that the vascular access path could potentially be a very powerful way to go.
So where does this leave us? Well, we’re very excited about the possibility that we’ll be able to open up some frontiers on some of these philosophical questions that since the dawn of humanity have been occupying us, and also have very pragmatic insights into maybe how brain maps lead to new models of computation and may help us pinpoint the control switches in the brain that we need to stimulate in order to fix brain disorders.
I’ll just close on this slide to acknowledge that this is all a very large and collaborative endeavor. This is an omni-disciplinary field and almost every field of engineering potentially could have something to contribute to this field of understanding and fixing the brain, and so we’re very happy to open up this discussion to try to figure out how we can help move humanity forward. Thank you.