Hans Jacobs

Transcription

>> This is definitely a passion of mine and as well as a number of other people at Google. The whole aspect of how we perceived the world and cognition being a big part of that. So without further ado, I'm just going to let Mike and Em speak eloquently to their work. >> GOARD: Hello, everyone. So my name is Michael Goard and I'm a recent graduate of the Neuroscience Program at UC Berkeley. And so today--I'm going to talk today a little bit about the neural circuitry underlying the perception of the outside world. And so over the last five years, I've been studying how the visual cortex of rodents, mice and rats encodes information about visual scenes that they're seeing. And so, on the right here, I have a schematic of a rat brain and basically what happens as images coming into the eye get transferred to a relay in the thalamus to this area and base up in the upper right called the visual cortex. And what we can do as neuroscientist is drop an electrode into his visual cortex. And record the electrical activity of individual neurons and see how they are responding to the visual scenes that the rats are seeing. So this is usually how I start to talk and then I spend an hour going into all detail and data that's collected over the last five years. But today, I'm going to do something a little different. I'm going to take a big step back and address what I think is a very important question which is sometimes, as scientists, we don't get a chance to address. And it's a question of who cares? Why are we interested in the neocortex? I mean, first of all, what is the neocortex? Why do we care about the way in which it encodes sensory information? And if we can figure out how it works, what would we get out of it? So, first question, what is the neocortex? So shown here is a human brain, so if you'd seen a--you kind of know on the outside it's a big kind of a wrinkly mass, looks like a walnut that's kind of division down the middle. So we just kind of pop it in a half and we're seeing the inside. And you'll see it there's this kind of wrinkly rind on the outside, that's the neocortex. And on the inside, there's some smoother structures, called the basal ganglia, the thalamus, the brain stem and there's kind of cauliflower looking thing, this is the cerebellum. And so all the stuff in the inside have been referred to collectively as the "old brain." The neocortex is wrinkly stuff on the outside. So this comes from the Latin, cortex, meaning a rind or bark because it kind of envelopes the older structures and neo because it's new. So that--this is evolutionarily a recent invention, it's particular only to mammals. And it's only grown relatively, recently. So, why we are interested in this? So, I'm going to kind of layout a few brains for you. So starting at the upper left, we have a shark brain. Now, a shark brain is pretty much exclusively the kind of old brain structure, sort of the middle parts that I've showed you in the last slide. And there's only a very small part that's kind of cortex-like that's involved in sensory processing. Now, we go one down, and the next one over is a rat brain. So this one is kind of mostly the old brain but it's got a single sheet of a neocortex covering it. One more down, we have a cat brain. So here you can see that were starting to get more of this kind of a neocortical covering. And you can see it's also starting to get kind of wrinkled. So the neocortex is a sheet and it has kind of a fixed-width. And so, one way of kind of squeezing more of it in a fix volume is by kind of convoluting it and wrinkling it up so you get more surface area per volume. And these kind of compare to a logical extreme that human brain, which you can see, is a very kind of complex mass of contortions. And so, I'm almost completely taken up by this neocortex. So, well, let's say, in the rat, it's only about 25% of the total volume. And then human it's over 75% of the human brain is neocortex. And so, it seems to be doing something interesting. And in fact, if you kind of look at the size of the old brain compared to the size of the creature in general, you'll find it pretty kind of proportional, and so the bigger the creature, the bigger the kind of old brain. But the neocortex, you get this kind of unusual over expression of it in certain species, so humans are probably the best example of this. We have six or seven times what you kind of expect given our body weight, something like that, monkey or dolphin might have four or five times as much. A dog or a cat will have a little bit less. And so what we see is kind of the more neocortex has, the more kind of complex behavior and cognition you're getting from the animal. So this is kind of why we think it's interesting. So this is kind of summing everything up in broad strokes. So the old brain is kind of responsible for reaction to stimuli, the automated motor sequences, control of your peripheral nervous system whether it'd be temperature regulation, circadian rhythms, whether you're awake or asleep, arousal. Also emotion, we tend to think of emotion [INDISTINCT] terms but it's actually very much part of the old brain. And that, you know, it makes sense, even very simple organisms need to have emotional balance, you know, that tasty more so over there needs to be good and that thing would be a big teeth needs to be bad. Now, the neocortex, it's a--so the old brain kind of gives everything you need for a survival. You know, if you see something, you can chew it, you can bite it, you can reproduce, you can run away. But the neocortex is involved in a kind of the more complex cognition that we associate with intelligence. So complex sensory processing, it's not just kind of seeing something and reacting to it, but, you know, being able to understand, you know, and that is an object or what its uses might be or that is an individual and a certain individual. Motor planning, you know, the idea that you don't kind of just react to something, but, you know, we might plan to do something later on today and we'll actually do it. Abstract reasoning, theory of mind, the idea that we can understand what the other individuals might be doing; and all of these, by the way, you know, we tend to kind of think abstract reasoning and theory of mind, or things that only humans do. But there's abundant evidence that other animals do it as well. Consciousness is kind of a stickier one. But we do know that neocortex is somehow involved. There's this interesting phenomenon called "blindsight", just that if a person has damaged to the back of their brain where the visual part of the neocortex is and then you kind of flash a light in a certain location ask them to point at it. They can point directly to it but they have no conscious percept of the light. They'll tell you that they can't see at all. So it seems like consciousness is somehow kind of subserve by the neocortex but we're not sure how. So in order to kind of subserve all these different functions, the neocortex must be this very kind of complicated heterogeneous structure, right? And the answer is kind of yes and a no. So you may have seen these maps before, so this is, you know, the human neocortex and different areas kind to correspond to what they process. So in the back, there you have the visual areas and in the middle, you have the sensory and motor areas and the speech areas, et cetera, et cetera, and this is all true. It turns out that their cortical structure is actually remarkably homogenous kind of across the whole neocortical expanse. And we think they kind of define this different areas, it's not some different architecture of the [INDISTINCT]. >> Can you guys, please, mute their audio? Thank you. Thanks so much. >> GOARD: Thank you. So it seems to differentiate these different regions, it's not so much the structure of the neocortex in that region but rather the inputs that are coming into it. And so what does the structure of this neocortical kind of circuit look like? So here we have a picture of the human brain. So there's darkened kind of band here is the neocortical sheet that's all convoluted. And if you take--just kind of take a section of it and then turn it upright, so on the top we basically have, you know, towards the outside of the brain and on the bottom, we have towards the middle of the brain. You see this kind of layered structure, so the cells have kind of different densities and they're in different sizes in the different layers. And this is a very kind of stereotypical no matter which chunk you--of cortex you grab. So whether it'd be a visual part or a part up here involved in abstract reasoning, it really kind of looks the same. And then--so what's going on here, so you can see this little balls that kind of--little lines coming up. So each one of those is a single neuron, and there, you can see the neuron kind of blown up here in upper right. Unfortunately, the image hasn't totally come out but--so the kind of ball in the middle is the cell body. And then, there's a bunch of processes coming off of it in kind of a tree-like structure. Those are the dendrites of the neuron and that's where it collects all its inputs from other cells. Then, they get enough input at any one point in time. It gives off just a little blip of electrical activity called an action potential. And we'll send out activity to other cells. >> Is it all six layers [INDISTINCT]? >> GOARD: Yes, so all--so the neocortex is always 6-layered everywhere in the brain, everywhere in development. >> Is that layering the big difference between the neocortex in the old brain? >> GOARD: Yeah, so the old brain doesn't have any layers. It tends to be kind of discrete nuclei. There are kind of primitive areas of cortex like the archaic cortex or the hippocampus which have a few, like, three layers for example or, you know, some different number. But the 6-layered structure is kind of characteristic of the neocortex. And there's some other commonalities that I'll talk about. Also, I should emphasize, so I'm being a bit of a neocortical chauvinist here in kind of talking about the neocortex is the only kind of way of kind of getting intelligence, but there's actually at least one other solution that we know of. So if you look at bird brains, they've evolved very separately from us, if you remember your Jurassic Park, you know, they've divert some time around the dinosaurs. They actually have a different kind of structure covering their brain called the neopallium. It has a lot of commonalities with the neocortex that it is a layered structure. It doesn't have six layers, I believe. So that by the--so there's some similarities. But essentially, it's a different solution to doing the kind of processing that the neocortex does. And it turns out birds are actually quite intelligent, capable of all kinds of complex and interesting behaviors and sensation and, you know, creating songs that sort of thing. But back to what I was talking about, so these neurons here, they're all kind of wired to each other. And neurons are interesting because they're basically kind of collect inputs from other cells and then decides to either give off this action potentials, so a blip of a electrical activity which it sends to its partner or it doesn't. So you can almost think of it as like the transistor in the neural circuit. And this neural circuit kind of wires up in such a way to perform some kind of a computation. So, you know, is this really kind of a universal circuit. So, there's some experiments to address this about 20 years ago in which they took parrots very early in the development. And so normally, you have inputs from the eye go to a relay, you can see it there's labeled the LGN which then goes over to the visual cortex. And you have inputs from the ear go through it--different relay called the MGN and then over to auditory cortex. So what they did is they actually ablated the input from the ear and kind of reduced the input from the eye to go over into the auditory relay which, then one over to the auditory cortex. And what they found is very interesting, the auditory cortex or the "auditory cortex" now had visual responses. And the visual responses actually look very similar to what people see in the visual cortex. So it seems--at least early in development there's a fair amount of universality in the circuits. Now later in development that they come specialized for whatever inputs they're dealing within and they become kind of less--just kind of mix and match them in the same way. So, we've, you know, neuroscientists, have been trying to figure out kind of what is a circuit and what is it doing. And there's been a lot of work on this and we've made some progress even though we're not there yet. And I've kind of quickly schematized the kind of it, the general connections you see. And I'm not going to a lot of detail here. But basically, inputs coming in from lower regions of the cortex, so this can either be from the sensory periphery, so let's talk about visual cortex as an example. So this could be from the eye or you also have piece of cortex dock from top to each other, so you have the eye and then the primary visual cortex and then, you know, higher regions of visual cortex. So it's input, if you're in the higher region of visual cortex, it could be from, you know, the lower area visual cortex toward from the eye. You also have inputs from higher regions going down. And then, there's some processing that happens in the circuit and then it splits out inputs, you know, both up at the hierarchy and down the hierarchy. And these hierarchies can be quite complex. So neuroscientists like to show these slides that makes what we do and look terribly complicated, but we don't actually know on how what's going on here. But, so basically, each one of these blocks have--it represents one or more of these kinds of cortical circuits on parallel. And as you can see, they're kind of stack on top of each other, but we do know it's kind of--as you go up this hierarchy, you get more and more complex representations of the stimulus. So the visual stimulus going on here, this first level, basically, what you see are the neurons respond to a small kind of dots of light in a certain place in the visual field. Next level up, you see, they respond to lines kind of moving across the certain area of the visual field. Up here, you get kind more of complex contours and they also start to take up a bigger and bigger portion of the visual. >> [INDISTINCT] >> GOARD: Up here, they respond to a pretty big area and they respond some kind of as complex contour. And at the very top you get these very kind of abstract responses. You can get responses to, you know, only certain objects, responses only to people or faces or even certain people in some cases. And in this cases that usually takes up the entire visual field. So, you know, say, it's a person's selective neuron and if a person is anywhere in the visual field that would be active. And this property is very interesting because it's actually very hard to get. If any of you have ever work in computer vision problems, it's almost impossible to program a computer to be able to recognize an object, that's because there's infinite kind of, you know, ways that the object can present itself. You know, if I have this remote, I can turn it around and you know, depending on how I present it to you. You know, if I may be closer to you, it would be bigger on your retina. If I bring it back, it would be farther away. And yet, no matter what I do, you can recognize it, but trying to find some kind of archetypical feature which a computer could expect and say, okay, that's a remote is very difficult. So what it seems to be happening is the visual cortex we have, you know, the circuit is kind of combining information from higher regions and from lower regions. And it's allowing it to do some kind of useful, informative task. So, if some of you may have seen this already, it's kind of a famous example, but if you haven't, I want you to try to figure out what is in this scene. And now I'm going to give you a hint if you haven't figured it out yet which if you haven't seen it you probably haven't. It's sort of other Dalmatian or is the Dalmatian is on the scene. So, some of you have probably seeing it now hopefully, if not, you know, there's a head in the front and this kind of tracing the contour. So now, even if I take it away, everyone sees the Dalmatian, right? And so what's happening here? So, it's--you're getting an ambiguous kind of not very clear input. And once you have, you know, the higher parts of your brain are kind of sending it back to useful information that, you know, there's a Dalmatian in the scene, then you kind of figure it out where it is and what's going on. And this is actually very interesting because I didn't tell you much about the Dalmatian. I didn't say which way he was facing. I didn't say whether it was big or small or whether it was, you know, down the ground or howling at the moon. All I said is that there was a Dalmatian. And yet somehow, your top down inputs kind of inform your kind of lower processing regions to kind of constraint the possibility of what you could look for in order help you find it. And this is what I think, and at this point I kind of leaving factual realm a little bit and kind of getting a little more speculative. But I think this is the kind of the point of it of hierarchy. And also, the point of this whole cortical circuit is to kind of combined this information about the high level and the kind of low level information that we're actually getting. And use it to kind of build up our perception of the outside world. And I think, this is kind of profound in a way. So the--our perception of the world is not actually what's coming into our senses but it's actually something we--our brain actively builds. And what we consciously experience is not so much what's coming in but the model that our brain has built. So I'm going to give you another example of this. So I have a little clip here of what sounds like R2D2 saying something. So I wanted you to try to figure out what the message is. >> The steady drip is worse than the drenching rain. >> GOARD: We'll play it one more time. >> The steady drip is worse than the drenching rain. >> GOARD: Okay, so has anyone actually able to hear what it said? Okay, one or two people. You guys are good but most people can't. Now, I'm going to give you a hint as to what R2D2 there was saying, "The steady drip," or you know, you can read it. So it's a little [INDISTINCT] I didn't come up with this example, but [INDISTINCT], maybe I don't know. But now, I'm going to play it again and I want you to see if you hear anything different. >> The steady drip is worse than the drenching rain. >> GOARD: So everyone hear it now? So, this is interesting so not only are you, you know, kind of abstractly recognizing that, it's the same as sentence I just showed you. But you are literally hearing it differently. You could point to, you know, which word occurred at what time and you would be right. And so, what's happening is your top down percept--your top down knowledge is changing your perception the outside world. I think this is pretty fascinating, and I think, it's one of the points basically of this neural circuit. So, you know, I have been kind talking in terms of perception but this, you know, it could also be useful for motor, if you're kind of comparing, you know, what you want, your motor actions could be, what the feedback you're getting from the outside world. I mean this is kind of general compared to a circuit. It seems to be doing something very important that allows humans to do all the kinds of--or other animals to do the kinds of cognition that we do. And so, this is kind of a major goal of neuroscientists to figure out how this all works and certainly, it's one of my major goals. And so, very brief, I just want to talk about the kind of implications if we could figure out how the circuit works. First, I think it's kind of, you know, have the normal set of implications from medicines. So, you know, autism, epilepsies, schizophrenia, you name it, almost any neurological disease is somehow a dysfunction of these cortical circuit. You know, whether it'd be, you know, miswiring or what have you. And so, you know, basically figuring out how to, you know, at this point, it's not like trying to fix a car when you don't know how it works and kind of remove parts and stuff in but you're probably not going to make it better. But if you actually know how it works I think what will--getting much closer to and I think able to find the cures for neurological diseases. Computing and artificial intelligence, there are a lot of things that computers are much better at than we are. So, for example, multiplying large numbers together, I think a $5 calculator could probably be pretty much anyone in this room. However, there are things that humans are much better at the computers. So for example, recognizing objects, your average 6-year old can be, you know, your most powerful super computer would, like the programmers, you know, kind of write computer vision programs would. So I think if we can, I mean, you know, we probably already have the processing power to kind of do these kinds of competitions that the neural circuits are doing. They are impressive but they are kind of within realms that if we haven't reach already, we can reach soon. But we really need to understand this architecture of circuit that one is doing. And once we do that, I think we should be able to build the kinds of circuits that allow, you know, things like computer vision and speech recognition to happen. And also, you know, we tend to think of inputs as being sensory inputs just because that's what we have, but the circuit does seemed to be pretty universal on it, and we can learn new tasks with it. For example, we learn how to read even though our evolutionary precursors never did that and so that--you could probably put some new types of sense information into the neocortical circuit and have it kind of intelligently kind of make inferences about it. And lastly, you know, kind of token, you know, human understanding I just think it's, you know, like what makes us different from caveman isn't just that we live longer and then we have cooler toys, but also that we come to understand our place in the universe that the Earth revolves around the sun, that we evolve from other creatures and so on. And so I think understanding the brain is very much part of, you know, kind of the--both fascination, humility of the human experience. So hopefully at this point I've motivated you as to why the neocortex is interesting and we're studying. And now I'm just going to talk briefly about some of my own research on this. And the--so as I mentioned the, you know, I'm kind of interested in studying how this kind of primary visual neocortical region is encoding information about the outside world. And this is my study, quite extensively for the 15--about 50 years or so and it's probably the best understood region of the neocortex. So, for example, it's known that neurons in this region will fire action potentials when minds cross the part of the receptive fields in a certain angle. So, what I want to do is actually not look so much at just the responses to the outside world but see how those responses change depending on the state of the animal. And we actually have good reason to think that they would change. So very early on, say, it was in 1930 that Hans Berger invented his EEG electrodes that you can, you know, put on your scalp and will kind of measure the gross neurological activities. This is filtered over, you know, millions of neurons. They can kind of look like very crude idea of what's going on in the human brain during different behavioral states? And so this is from one of those patients. So, on the top you can see when the patient was excited, you see its very kind of low amplitude, high frequency activity. When he was relaxed you see this kind of higher amplitude, slow frequency activity. And when he's asleep you see like this very slow kind of high amplitude activity. Now we know kind of intuitively that we--we kind of perceive the outside world very differently in this different states. You know, when we're excited, we kind of pay attention more than when we're drowsy and we certainly paid attention much more than when we're asleep. And we also know that, you know, this kind of gross neurological activity is going to affect the neurons and this neocortical circuits will likely change the responses. So what I wanted to see is, you know, what is the biological mechanism of, you know, how are these changes happening and what effect is it having on the neocortical circuit. And so to investigate this I, you know, I picked a system which I had a strong hint was involved and this is the acetylcholine system. So it's absorbed by this nucleus here in the basal forebrain called the nucleus basalis, this small region here. It's a very small number of large cells that send acetylcholine basically all over the neocortex. And the reason I was interested in this nucleus is because it's known that if you stimulate it electrically, you basically change the state of the cortex. So this is an example here that--which I record the EEG from a rat. What you see is big slow wave oscillations. And then I stimulate this nucleus and now you see is much faster lower amplitude oscillations, so it's exactly the sort of state change that I showed you in the last slide. It's also known that this system is kind of active at different levels depending on the state of the animals. When we're asleep it's pretty much turned off. When we're awake it kind of 1459 fires tonically like a higher level, and then when we come really excited it kind of goes getting busters and it goes crazy. It's also known to be very important, if you damage these neurons you have major problems with perception, attention, memory. So, essentially a lot of the pathology from Alzheimer's and senile dementias had to come from damage to these neurons. And finally its known that if you just kind of drop acetylcholine on to the visual cortex it changes the response properties of the neurons. So what I wanted to do is kind of activate this area while measuring responses in the visual cortex to a natural scene and see how activating this area kind of changed the responses to the natural scenes. And so the way I did that, I'm not going to get into all the details, we just don't have enough time. But I use kind of a clever technique were we got this transgenic mice from lab at Duke. And you can actually specifically activate just the neurons that release acetylcholine by using a certain wavelength with a laser light. So we drop an optic fiber into the brain, activate these neurons, and cause the acetylcholine release all over the cortex. Then in the visual cortex I wanted to really record from the whole population of neurons rather than just one. So we used this thing, it's called a polytrode. And it's basically a silicon chip that has regularly space [NDISTINCT] sites on it. So this is the probe here and these are the individual sites, so they're kind of in an array. And then what happens is--so this is just kind of an example of there's some neurons in front of this chip. Everytime one has one of its little electro blips in action potential. You can see it as a kind of signature blip in the electrical activity. And each one, because they're in a different location will activate different channels. So this one over here will kind of cause that one or that pattern of responses and from that information you can use mathematical techniques to extract the firing properties about bunch of individual neurons and complicate populations of neurons and see how there responses are changing. So I have a ton of data on this that I unfortunately don't have time to tell you about. I'm just going to show you one quick result to kind of give you a flavor for it. So what are these? Colors didn't quite come out but--so what I've done here is I'm just taking a look at individual neurons. So I have three individual neurons in different colors here. So let's just look at this top one now, so the top two panels, and what I've done is I played the same movie over and over again for this neuron and look at its responses. And so basically each little hash mark you see is a response to somewhere in that movie. And what I've done is I've stacked the 30 responses to the 30 repeats of the movie on top of each other. And the reason why I've done these is that you'd expect that if the neuron is kind of giving you some useful information about the movie it's going to respond at the same time during every presentation. Otherwise, if it's responding at different times then it's not conveying anything useful to the upstream regions. And so what I wanted to see was it does the release of acetylcholine actually kind of increase the information and the response of the neurons. And so we can see that if you look at the bottom episode, the top three panels is the cell I'm going to show you, if you look at the bottom of those two panels that's the response when I cause the acetylcholine to release. And you can see that there's a couple of tiny points that would show very reliable responses to the natural movie. But if you look in the controlled situation where I didn't induce this acetylcholine release, so the panel above it you do not--you no longer see those kinds of reliable responses. So it seems like what's happening and, you know, this is just one example cell but we ran all the statistics so it's a very reliable fact. So it seems like what's happening is that the acetylcholine release is causing the cells to kind of better encode the visual environment. And as a kind of more quantitative way of testing this, what we did is we took the whole population of responses to the movies and we tried to predict what movie the or what part of the movie, which frame of the movie the animal is seeing base truly on the responses. And then we found when there's acetylcholine present, we're able to do it with much higher accuracy than when there wasn't. So essentially what's happening is acetylcholine is improving the signal processing in that region. And this is really cool to say what it suggest is that, you know, our brain is not like a silicon circuit were it kind of works the same way each time unless it's broken. It can actually change the mode in which it operates depending on the kind of the needs of the organism. So when you're kind of zoning out, you're kind of not paying attention to the outside world because you're doing something else; and then someone snaps your attention, you release, you know, acetylcholine, and suddenly you hear what they're saying to you which you sort of have ignoring a second before in those stimulus. So to sum everything up; so today, I talked about how the neocortex makes up the majority of the human brain and kind of underlies our complex behavior, motor planning, cognition, et cetera. And that the entire neocortical circuit or sheet is kind of composed of a somewhat stereotyped circuit and figuring out how the circuit works will be essential in terms of, you know, being--well, in terms of understanding how the brain works it makes up 75% of the human brain. And we'll also probably yield numerous practical benefits. And finally, I showed you kind of one example of how the neocortex is kind of different in animal from some of the circuits that we're used to. You know, it can change how it processes information depending on the state of the animal. Then actually despite these kinds of things, we just make it seem like its going to be a very complicated problem. We're very optimistic that we're going to be able to solve this problem hopefully within our lifetime. So there's a lot of new tools for, you know, basically--so like what I did recording from large population of neurons. There's ways of which you can image the activity of large populations of neurons. We also have new tools which allow you to stimulate certain neurons and not others. So basically I have all the tools now to kind of dissect parts of the circuit and really kind of take it apart and see how it works. So finally I just wanted to thank especially my adviser Yang Dan other members in my lab and Guoping Feng, professor, he just moved from Duke to MIT, who--he provided the transgenic mice we used. And--anyway, talk about the work but these other the people kind of helped me with various parts of getting the techniques worked, and finally, big thanks to Peter for organizing this. It's been awesome and--so I think before we take questions I'm going to hand it over to Emily and she's going to go for a short trip talk on her work and then afterwards we can do some Q&A. >> JACOBS: Yeah. And certainly, of course, if I can set-up. >> GOARD: Okay. >> Any quick questions? >> Sir, [INDISTINCT]. The chemical acetyl--... >> GOARD: Acetylcholine. >> Is there any studies that show that your brain has more of that like while you're performing at high, like, intellectually or physically or like in athletic sense? Is that something that's related to higher performance in general or...? >> GOARD: It's definitely thought to be related to higher performance in levels of arousal. Fortunately, it hasn't been studied carefully in humans. It's mostly been animal studies, which is a little more difficult to quantify the amount of arousal but it seems like we--so there's been studies which they found that if the animal is performing some task it basically becomes--the nucleus releases acetylcholine during important parts of the task. >> Thanks. >> JACOBS: Okay. And hopefully, there'll be time for more Q&A after. So I'm going to talk about something pretty different. I think that's one of the joys of doing a tag team talk is that you can see the breadth of neuroscience research. We're both at Berkeley but in pretty different areas. So I'll begin by saying that the study of neuroscience is devoted to understanding how the brain functions uniformly across members of a species. For example, Mike explored the way, systems neuroscientists are tackling the profound problem of how cortex works. But a critical question is how neural in cognitive processes differ between members of the species or even within an individual in different environmental conditions? But the question becomes not how our brains similar, but how do we differ? Thus, the purpose of this talk is to introduce you to two concepts that or two factors that contribute to individual differences in cognition, that is genes and hormones; and because I study estrogen, apologies to the men in the room because a discussion of the female menstrual cycle is imminent. I don't want to get too far field but I will say that the second loftier goal of this talk and really this research in general is to stress the importance of considering the role of hormones and sex differences in cognition. From a basic science perspective, an intimate understanding of how our endocrine system impacts neural and cognitive processes is interesting. From a woman's health prospective, I'll argue it's fundamental. The estrogen cognition literature and in particular I'm thinking of the big population health studies on the use of hormone replacement therapy. That data is famously inconsistent and we can ask at the level of the brain, why? So before we start thinking about genes and hormones, we should talk a little but about neurochemistry. Mike was talking about acetylcholine; I'm going to talk about dopamine. I assume many of you have heard of dopamine somehow in the room. So like in what context? You shook your head. What is it? >> [INDISTINCT] motivation. >> JACOBS: Exactly. Yeah. >> [INDISTINCT] >> JACOBS: Yeah. [INDISTINCT] is great. Yeah, so the reward pathway, okay. That's good. So it turns out that there are three main ascending dopaminergic pathways in the brain. So you pointed out one which is the mesolimbic pathway so it starts here in the VTA or ventral tegmental area and projects to the limbic region, the nucleus accumbens and such. There's two others so the nigrostriatal pathway which originates in the substantia nigra. Another group of cells in the midbrain and projects to the striatum, when the cells in the substantia nigra die, you got Parkinson's disease. And finally, the pathway in red is the mesocortical pathway; it also originates in the VTA. These projects to the frontal lobes; so prefrontal cortex is a part of the brain that's really important for things like planning and logical thinking and abstract reasoning, as my talk about, and that's the pathway that I'm talking about today. They're distinct pathways and they carry distinct cortical purposes. Now it's been just over 50 years since Arvid Carlsson first defined dopamine as neurotransmitter. At that time in the late 1950s, it was thought to be nearly a precursor in the biosynthetic pathway of norepinephrine, another neurotransmitter. His basic science work was really focused on the nigrostriatal pathway and his work led to the development of using dopamine drugs to treat Parkinson's disease. He was awarded a Nobel for that in 2000. A really great work. But fast-forward 20 years from the late 1950s to the late 1970s and we meet Tom Brozoski who really demonstrated for the first time the functional importance of this red trace, the mesocortical pathway. And in a now landmark study he was able to removed dopamine from the prefrontal cortex using this neurotoxins 6-hydroxydopamine and he tested this monkeys on short-term memory task: pre and post off. And here's what he found, so in this task monkeys are asked to maintain a bit of information over delayed period and in some trials the delay is really short and then in other's its really long. So the intact animals at--represented by this white trace and you can see there's this very expected or predicted increase in performance as to delay period like--again to be expected given the nature of the task. But what happens when we remove dopamine from the prefrontal cortex? This monkey, in the black trace, performance begins to tank almost immediately. What's striking is that the deficit seen in dopamine depleted animals is almost as bad as those monkeys who had direct structural ablation to the prefrontal cortex to literally, you know, morphologically there's damage to that region of the brain which is shown in here. And importantly when dopamine was added back into the depleted animals, defect went away. So the moral of the story is that dopamine is really important for proper functioning of the prefrontal cortex. This region of the brain is exquisitely sensitive to its neurochemical environment. So you modulate dopamine, you're going to modulate cognitive functions that rely on that region of the brain. So the general relationship between dopamine and prefrontal cortex is clear. I've given you one example and there's, you know, a whole 50 years worth of work to show you more. But the intricacies of this association are not straightforward. So if I were to give everyone in these room 30 milligrams of Methylphenidate, Ritalin, a common drug, I may see a quarter of you show cognitive improvement on particular tasks, a quarter of you shows cognitive impairment, and then half of you maybe a null effect. That's interesting, right? But why is that? Well, it turns out that the effects of dopamine on performance are not linear. In fact, there's this inverted U-shape function or parabolic curve that defines dopamine's actions in the prefrontal cortex. Now this can be seen at the single unit level, if we're looking at this spatial tuning of prefrontal neurons engaged in the working memory task. We see that at optimal level of dopamine stimulation it really good spatial tuning but in--if we have too little dopamine receptors, stimulation are excessive. Stimulation, we see these effects be degraded. Now, importantly, this is also seen at the human cognitive levels. So it's just a measure of a behavior or neural efficiency. Dopamine works in this inverted U-shape curve. So there's this theoretical optimal amount at which you get the fast performance in an either end performance begins to tank. So what becomes critical is taking into account this Law of Initial Value or where did these individual lines occur? Only knowing that can we predict how additional dopamine modulation is going to affect your performance, right? If you're starting up here on the curve then you benefit from a dopamine dose from that morning coffee or even working well under stress. Both of which you get from a nugget. But if you're already optimal then you don't want any modulation. So the natural question is what does influence dopamine? Well, I already gave up ago since that genes and in hormones in particular estrogen, but also things like stress and coffee, drugs of abuse like cocaine and prescription drugs like Ritalin. So I'm going to focus on the first two in this talk. So first, a quick note about the genes story and to kind of approach that topic we need to first zoom in to the level of the synapse and understand dopamines biosynthetic and metabolic pathways. So dopamine is produced in the synapse and it gets released after--so Mike talked about the action potential and we know that neurons communicate by electrochemical signaling. So there's the flood of ions down to axon and then there's a release of chemicals which, as you've know from you internal neuroscience that cells don't actually connect. There's a small space in between and it's the job of neurotransmitters to relay that information. Well, so here's the instance of dopamine, its being relayed in or released into the synaptic cleft. But there have to be some way for it to get removed after it's done its job. And it turns out in prefrontal cortex the job of removing dopamine falls to an enzyme called Catechol-O-methyl transferase or COMT and it's really important. So in the prefrontal cortex COMT accounts for over 60% of all dopamine turn over. If you look out mice who don't express COMT, they show increased prefrontal dopamine and enhanced memory. Similarly in humans we can use particular drugs that inhibit COMT activity and we see increased of local prefrontal dopamine and an enhanced working memory. Now just a side note that in those other dopamine pathways that I showed you, the primary mechanism for dopamine degradation is something totally different. It falls to the--something called the dopamine or DAT, dopamine transporter protein. So this COMT enzyme is really particularly crucial for that mesocortical pathway for the frontal lobe. That's interesting because the gene that coats for that enzyme is polymorphic. So there's a naturally occurring difference in the way that gene is structured in everyone in this room. So it's a single nuclei type polymorphism. What that means is that one of the base pairs is switched, and if a guanine to adenosine flip. And this is very evolutionary recent. So this polymorphism is first observed in humans. All animals have the G, the guanine which is that base paired difference leads to the amino acid shift of valine to methionine. And the point I want to make is that--that, sort of flip fundamentally alters the way the protein works. So in individual, so about quarter of the population, you will have these two copies of this val/val gene. It's an issue of thermo stability, so that enzyme is physiologically under normal, like human body temperatures, it's very active. So it clears dopamine from the synaptic cleft really well and we kind of colloquially refer to these people as having low baseline dopamine. There's not a lot that's left hanging out in synaptic cleft. The more recent allele, again first observed in humans, that enzyme is actually thermolabile. So, it begins to breakdown. It doesn't do a great job. And so dopamine's allowed to accumulate in a cleft and we sort of refer to these people as having increased prefrontal baseline, prefrontal dopamine levels. And there's functional consequences, so val/val individuals actually have worse performance on this working memory of short term memory type tasks and reduce neural efficiency. Whereas met/met subjects are--have better memory and better efficiency in terms of a neural response. So, yeah? >> Just in general or after [INDISTINCT]? >> In general. So just--if we would have to split this population into val/val and met/met, we could see differences emerged without tweaking the system at all. So it's interesting. In this line of work is, you know, at a big picture level, we can see that variation in genes leads to variation in cognition, but clearly there are many steps in between. So we go from a genetic variation, altering how protein works, altering how some neural circuit works and maybe that bubbles up to the level of behavior, maybe it doesn't. So the point of what we do as human cognitive neuroscientist is to try to look at this endophenotypes or intermediate phenotypes and we can probe brain function to see how genes affect brain function and maybe behavior, but we can kind of get like an early biomarker of how that gene is having an effect on the brain. Okay, so I'm going to skip quickly to the realm of endocrinology. In a breast, estrogen amplifies dopamine transmission. There's strong evidence in animals to suggest that estrogen enhances dopamine synthesis release and turn over and modifies basal firing rates of dopaminergic neurons. We know that stimulated dopamine release actually fluctuates in a rat's estrous cycle. And finally, the PFC has amongst the highest concentration of estrogen receptors in the human brain. And hormones fluctuate throughout the menstrual cycle in this various stereotypical predicted way. So estrogen levels are lowest in the first few days of the cycle when women are menstruating and it spikes to about four times the typical value required for ovulation, so around day 11 to 12 of the menstrual cycle. So given this fluctuation of estrogen during the menstrual cycle, and given estrogen's ability to promote dopamine transmission, here's what we predicted; if we could get some index of where an individual lies on that inverted U-shape curve, we could predict whether estrogen was going to be helpful or harmful. So in order to do that study and a series studies, actually; this is kind of a schematic of the experiment, we pre-screened a whole bunch of subjects from--for this COMT enzyme, it's spitting into a cup, very simple. And then we tease apart the two extreme groups, so the val/vals and the met/mets, below dopamine and the high dopamine. And then we brought them into the lab and we gave them--we tested them on two occasions, so when estrogen levels are naturally low, when estrogen levels are naturally high and on each occasion we gave them a battery of cognitive test, we gave them a function of magnetic resonance imaging scanning while they're performing different cognitive tasks and we took blood samples for biomarkers. And this was the prediction that val/val subjects known from previous work, they're thought to have decreased prefrontal dopamine and thus reduced performance on this very highly dopamine sensitive tasks where met/met subjects have optimal dopamine and really robust performance. Now if we tweak that system and we say, [INDISTINCT] but we can get these subjects to be more optimal. So it looked like the met/met subjects, its estrogen levels are high. But in the met subjects, we'll actually see opposite effect, that their performance is best when estrogen levels are low and they kind of overdose, that dopamine gets pushed over the edge under high estrogen conditions. So kind of alluded to cognitive task and prefrontal--prefrontal kind of mechanisms but I haven't really talked about the kinds of tasks that we do, and I'm not going to--to give an example, specific example of what we do in this scanner because it's sort of long and boring. But it'll give you a kind of back of the envelop test of what we use in the field to test working memory or short term memory. Okay, so I'm going to give you a string of numbers and I want you to repeat them back to me or you can do it in your head backwards, okay? So here are the numbers, 2, 1, 7, 3, 4, 8. [PAUSE] Is this what you send? 8-4-3-7-1-2, great, so that's working memory, you have to keep information in mind over a short period of time and maybe manipulate it in some way. And that's a gateway cognitive function for almost anything you do certainly on a daily basis here at Google, right? So reading a paper and trying to relate one sentence to the next, doing math calculations in your head, that's all relying on this really basic cognitive ability and it's highly sensitive to the amount of dopamine you have in prefrontal cortex. So I'm going to cut right to the chase and show you a couple of data slides. The first one is just behavior, so how do women perform on this type of working memory task? Well, it turns out that if you look at val/val women, so low--just naturally low prefrontal dopamine. So in the red bar, performance is about 75%. Well, not great but if we look at this same women with estrogen levels are high, so right before ovulation, we see a boost of performance. Now the interesting things is the exact opposite effect is seen in met/met subjects who already have all dopamine. Their performance is peak when estrogen levels are low which is most of the cycles have actually performance goes down when estrogen levels are at the highest value. And to us that seemed to be a suggestion that this inverted U dopamine response was actually visible in humans, taking into account estrogen and genotype. Now, I talked a lot about, you know, you can run it down, you can measure somebody's genotype for this COMT gene and you're either val or met or your header as I guess, but what happens if we cut, you know, sort of leap-frog over DNA all together and go straight to the source, straight to the enzyme and we can measure that. We can take blood samples. It's a little bit more invasive but we can measure the active enzyme. And we do that because even within genotype of groups there's a lot of variance. And that's a whole other discussion of why that happens but it's really fascinating. The point is that, if this model is correct, then I should be able to take into account of this variant and predict at an individual level who's going to benefit and who's going to suffer. And the point of this graph is to show that even at the individual level we can see this inverted U shaped curve pop out by considering not only estrogen but active COMT enzyme levels at the time of FMRI scanning. So now unto the brain, so again we did this in the context of a Functional Magnetic Resonance Imaging or FMRI study. And the data that I'm just going to show you is something called a region of interest analysis. So we know exactly what part of the brain is really important for working memories. So we can essentially drill a circle around that region, it's a task related region in the brain. And just look at activity, mean activity across that region and compare it between groups. And essentially what we find is a representation of that behavioral effect but in terms of neural activity or bold FMRI bold response such that the less optimal dopamine group shows a sort of weak effect in this part of the brain, whereas the more optimal dopamine groups are really robust effect in this brain while you're performing the task. And importantly, the extent to which any individual shows this nice robust activation during the task was predictive of their performance. So we see that dopamine status predicts PFC activity which in turn predicts an individual's performance on this working memory task. So in summary, estrogen improves working memory in women which inadequately low prefrontal dopamine but impairs performance in women with naturally high prefrontal dopamine. And we see that this inverted-U dopamine effect is reflected in behavioral and neural indices of PFC function. So a few implications, first dopamine's influence on cognition cannot be fully understood without taking estrogen into account, I argue. Second, the functional impact of genetic variation can be explored via direct assays of brain function. And finally, this kind of lose concept of women's health can be addressed very specifically within the central nervous system. And there's actually a really great edition in scientific American Mind and talking all about this he's and her brain. Okay, and a final slide, why should we care? I think Mike and I both get our own share points of talking about this at the end. Often in neuroscience, sex is something to be controlled for, or scientists leap-frog over the matter all together by studying only males of species. To make that point clear, colleagues at Berkeley recently indexed all studies in mammals published in 2009 across 10 fields and over 40 journals and they found a wide spread bias towards using only male animals. The effect was especially striking in neuroscience with a ratio of male only to female only studies, was 5.5 to 1. Now there's a logic to that, it's cheaper, you don't have to track things like females menstrual cycles, so there is some rationale to that. But the interesting thing is that this bias was largest in neuroscience pharmacology and physiology fields for which the basic science research has huge implications for human health. If we only study male species, we're going to miss out on a few things is what I'm trying to argue. Sex differences are repeatedly found for rates of depression, vulnerability to stress, pain, Parkinson's disease, ADHD, and nearly all aspects of drug abuse from acquisition to treatment. We know that enormous endocrine changes happened during menopause but we know relatively little from a human neuroscience perspective about how this normal aging process affects specific neurochemical and cognitive systems. According to the news one day estrogen is good for you and the next day it's bad for you. So, yes, it is a messy problem. I think the good ones always are. But the problem gets thrown into relief only after it's subjected to careful scientific study. So hopefully we'll continue to do that and understand a little bit more. But in general hormones and genes play an interesting role in terms of brain function, I hope I've convinced of that today. Thank you. >> Emily, Thank you for fantastic and really exciting perspective on things. I just like to open up the floor to questions and I'll extend it to the patient people on DC first just to see--does anyone on DC have a question? [PAUSE] Last chance? All right, I'll come back to you, anyone in the room? [PAUSE] >> Thanks both. Those were both really great talks. I have a question for Michael first. It looks like you used the optical genetics to stimulate the acetylcholine? >> GOARD: Yeah, exactly. >> So that has really great timing patterns. Have you guys looked at all of that effects of when you stimulate? Does it really matter? >> GOARD: So with acetylcholine release, it works a little different. So usually timing is very important for neurons and the exact kind of timing of assimilation you give them really effects how they fire. Acetylcholine is a little different in that respect and that--it's when the neurons releases acetylcholine, they release in this kind of bulk way, it goes all over, kind of, extracellular space and takes a long time to clear. So the term--the effects is much more kind of, it's like a spread out in time, basically. And so, I looked at--basically, what I found this when the light is on, the effect is there, when I turn it off it goes away relatively quickly, so it's not instantaneous. So giving shorter pulses, it tends but just not activated at all, it kind of have to give like a longer continuous pulse. >> One other question, so it looks like your recording B1 to see the effects? Did you have a particular hypothesis that should be there Do you expect it to be most parts of the brain? >> GOARD: It's probably going to be all over the neural cortex or at least the central neural cortex and it also projects to the prefrontal cortex but, you know, just with me I really know what's going on there. I pick the visual cortex just because we have pretty good models of what's happening in terms of sensory processing already. So it makes much easier to interpret the results. >> Thanks. >> Can you talk a little bit about the heterozygotes or the dominant behavior of whether you have seen...? >> JACOBS: Yeah, so, I mean, in my study I didn't look at the headers I get, which represent 50% of the population because I wanted to look at the extremes and see the effect. But in studies that do include them, you'll find basically that they fall right down the middle. So it's interesting because there's another group led by Danny Weinberger at the NIH who studies the effects of drugs and you can actually--in pharmacogenomics, that is you can predict the effect of a dopamine drug, he's used I believed methylphenidate and amphetamine. You can predict the effect of that drug based on COMT genotype. So, you see improved both neural efficiency in terms of FMRI bold and also improved performance in val/val subjects slightly improved performance in heterozygotes and actually impaired performance in met/met subjects. Super interesting, I mean, it speaks to the fact that we've, you know, at least in theory could begin to do individual as medicine and this is kind of one example. >> I'm assuming, for people who had too much dopamine, did you notice like increased anxiety and things like that? >> JACOBS:So I've measured sort of state and trait anxiety level, so during the day of the scan, because we really want it to control for that. Now it's an interesting point you bring up because people say we'll--now, obviously, met/mets the best right, you know, they better work in memory like we should just figure out who they are and hire them all at Google or something. But it's not quite that simple, I mean, the reason we have both genotypes is, you know, it could be residual effect. But in fact, met/met subjects have increased rates of anxiety. So it's hard to say that, you know, there's a better or worst, it's just different. It depends on what your measuring stick is. >> Did you look at serotonin as well as dopamine? >> JACOBS: I didn't, I didn't but that would be fascinating in next five years in my life. [PAUSE] >> Just a simple question, what's the prevalence of the two alleles and met/met? >> JACOBS: Yeah, so they're co-dominant. So about 25% of the populations are val/val, about 25% are met/met. In my sample it was slightly less. I don't think I had about 20% where met/met. And again, 50% are heterozygotes, and yes of course I genotyped my self. So when my estrogen levels are high, forget about it. >> Question. [PAUSE] For Michael, you're talking about how the brain effectively feeds to the top down abstract context into, say, visual perception, and that certainly was under the computer's lack, so we kind of struggle electronic and computer to have these types of things. I presume that--sort of develops during quite as a baby, it looks like the brain boot straps on these concepts to a high level? >> GOARD: Yeah, so it's not known exactly how that happens. What seems to happen is that the cortex can't develop hierarchically kind moving up. So like the earliest areas developed first so like the primary visual cortex for example at first level of visual cortex, it's very early in development when it kind of, you know, two, three years of age in a human while the prefrontal cortex is still kind of developing, you know, even teenagers at early 20s or so kind of developing their prefrontal cortex. So it seems like that the cortex is kind of getting progressively trained like in hierarchical fashion. >> What do you think computer vision would benefit from trying to sort of do the science of bootstrapping, perhaps or even more stop and [INDISTINCT]? >> GOARD: Yes, I think it could be--so I mean in some ways it's a very different strategy of computing and I think for doing the types of things like computers are currently good at, I think it's not going to be very useful, but for doing the kinds of things that humans are good at I think it will be very useful. So it's not--and it go almost like a different type of computer totally speculating here, but I think it will be good at the kinds of like object recognition type tasks and must good and kind of, you know, serial or straightforward tasks. >> I have a quick question for both of you. But I'm having a really hard time resisting hating you guys for being both really in a beautiful people so and so. With all of these findings, we get closer and closer to larger questions of can we control and should we control these things to enhance our own abilities? Perhaps there's a better way from just understanding our biorhythms or perhaps healthier ways of addressing better releases for either dopamine or serotonin. But I was just curious form both of you, what's your perspective on those things? >> GOARD: So I think the first kind of priority is to figure out cases in which these systems are working incorrectly, you know, in neurological diseases and fixed those first. Then comes a question with like, okay, we develop this, you know, way of kind of manipulating acetylcholine system in order to help people with Alzheimer's disease. It turns out you can kind of give yourself, you know, a little boost on test days, should you do it? That's kind of a good question form a bio--I mean, yeah, I don't know quite how to answer it, my--I guess I'd say, yes, under certain circumstances but I think that the circumstances are tricky. >> JACOBS: I would--I guess I would just agree that, I mean, this is kind of the question that always--that we always get especially with this--the COMT genotype. You know, the important point is that, I mean, I talked today about one single, single nucleonic type polymorphism right, but there are, you know, an infinite number in the brain and of like neural regions, genes don't operate in isolation right? They all work together and they are part of this larger system. So, you know, until we can understand how, what really that the ultimate good is like what is the best? It's not just met/met genotype, its met/met genotype plus, you know, D1 receptor polymorphism of the, you know, it's like going to be the whole concert. And so we're just literally, you know, scrapping the type--top of the iceberg off when we asked this question so, I mean, maybe that's a cope up answer because all scientist say we have to know more, we have to know more, whereas the general public says, where can I buy the drug. So at some point we're going to have to confront these questions pretty soon. >> I'll extend the invitation to DC, anyone out there with a question? All right, well thank you so much. You have given us a lot to think about and be inspired by.