WEBVTT 00:00:00.017 --> 00:00:07.677 This is Paul Verchure with Sam Wang. And Sam has been coming out of physics, 00:00:07.857 --> 00:00:14.077 dove into the cerebellum, using very advanced and novel technologies to actually 00:00:14.077 --> 00:00:16.437 study the properties of the cerebellum. 00:00:16.817 --> 00:00:22.057 And what you emphasized very much was this whole issue of how the cerebellum 00:00:22.057 --> 00:00:25.477 is so important in the processing of unexpected events. events, 00:00:25.537 --> 00:00:29.777 or in something you called about interrupt handling. 00:00:30.177 --> 00:00:33.417 So what does that exactly mean? How should I think about that? 00:00:33.537 --> 00:00:35.777 Well, probably, let's see. So when I talked about that in my lecture, 00:00:36.517 --> 00:00:37.797 there are a few things that I meant. 00:00:37.937 --> 00:00:43.017 So when I came out of physics coming into neuroscience, I first worked in cell 00:00:43.017 --> 00:00:46.137 physiology for my graduate degree, calcium release phenomena, 00:00:46.377 --> 00:00:50.237 and that was very much the biophysics of single cell phenomena, not even neurons. 00:00:51.637 --> 00:00:55.157 But there's something that's really attractive to a physical scientist about 00:00:55.157 --> 00:00:58.597 the cerebellum because it's got a really small number of cell types. 00:00:59.417 --> 00:01:03.457 It looks like it should be analyzable by some kind of simple means, 00:01:03.597 --> 00:01:04.517 even theoretical means. 00:01:04.857 --> 00:01:08.177 And people, of course, have been interested in this going back to Marr and Albus. 00:01:08.677 --> 00:01:13.877 So what I meant in my lecture about the cerebellum actings in Interrupt is that 00:01:13.877 --> 00:01:17.437 there are two major pathways of excitatory information coming into the cerebellum. 00:01:18.037 --> 00:01:23.357 One of them culminates in the production of a stream of spikes coming out of the Purkinje cells. 00:01:23.517 --> 00:01:27.197 So Purkinje cells are like many inhibitory neurons of the brain in the sense 00:01:27.197 --> 00:01:28.657 that they're firing all the time, tonically. 00:01:29.017 --> 00:01:32.797 And what I meant is that there's this second input pathway, a second excitatory 00:01:32.797 --> 00:01:36.157 pathway coming in via this funny structure called the inferior olive. 00:01:36.457 --> 00:01:40.377 An input that comes in through the inferior olive fires much less frequently. 00:01:40.557 --> 00:01:43.377 Each individual neuron of the olive fires maybe once a second, 00:01:43.557 --> 00:01:47.217 and then it comes in and innervates, each one of those neurons innervates a 00:01:47.217 --> 00:01:52.537 few dozen neurons in the cerebellum, these Purkinje cells, which are the output 00:01:52.537 --> 00:01:53.477 of the cerebellar cortex. 00:01:53.937 --> 00:01:59.417 And what I meant was that the system looks like it's a low-frequency firing 00:01:59.417 --> 00:02:03.357 pathway, and so right off the bat, it seems like it's probably not going to 00:02:03.357 --> 00:02:06.557 be encoding a lot of information in the rate of spikes. 00:02:06.797 --> 00:02:09.097 It's more like the timing of these things is important. 00:02:09.923 --> 00:02:14.503 Specifically, what I meant by the interrupt is that these individual spikes 00:02:14.503 --> 00:02:20.423 trigger events that can drive plasticity, so they can drive plasticity at the Purkinje cells. 00:02:21.103 --> 00:02:26.563 The thing I was showing today in my talk was also the possibility that we've 00:02:26.563 --> 00:02:31.523 demonstrated with our optical methods that a number of these cells can fire together at once. 00:02:31.523 --> 00:02:35.863 And that might be some kind of control signal that just sends a reset to the 00:02:35.863 --> 00:02:40.003 cerebellar cortex or to the deep nuclei of the cerebellum. 00:02:40.083 --> 00:02:44.523 And the idea is that it sends some kind of signal that can happen as an isolated 00:02:44.523 --> 00:02:50.003 event in time that maybe resets the output of the Purkinje cells or perhaps can teach plasticity. 00:02:50.183 --> 00:02:54.063 So the idea is that there's a unitary signal that can either reset in real time 00:02:54.063 --> 00:02:56.383 or can drive learning on the long term. 00:02:57.223 --> 00:03:00.103 But how should i think about this because then in some sense 00:03:00.103 --> 00:03:03.423 what you're saying is well i can have my my inferior 00:03:03.423 --> 00:03:06.363 olive neurons um through the climbing fibers talking to 00:03:06.363 --> 00:03:09.943 the cells but what they convey can be 00:03:09.943 --> 00:03:14.863 either like if you want a blind reset that is okay whatever was going on reset 00:03:14.863 --> 00:03:20.003 this or a more specific signal for learning it doesn't sound like a possible 00:03:20.003 --> 00:03:24.143 contradiction uh yeah so i guess one example that's going to be familiar to 00:03:24.143 --> 00:03:26.463 many listeners of this podcast is, um... 00:03:27.406 --> 00:03:30.886 Hebbian plasticity, right? So people talk about sequences of neurons firing 00:03:30.886 --> 00:03:34.146 in order, and somehow that sequence of neurons firing in order, 00:03:34.226 --> 00:03:39.526 which Hebb called the cell assembly, is some kind of re-remembered experience. 00:03:39.746 --> 00:03:41.926 Remember that Hebb suggested that when we have an experience, 00:03:42.066 --> 00:03:43.566 we have a bunch of neurons that fire in order. 00:03:43.966 --> 00:03:48.166 If we imagine just one chain of neurons, just as an idealized example, 00:03:48.426 --> 00:03:51.806 it would be A, then B, then C, then D, then E, right? So that would be a set 00:03:51.806 --> 00:03:52.866 of neurons that fire in order. 00:03:53.166 --> 00:03:56.446 And what Hebb suggested approximately, I mean, he didn't say all these things 00:03:56.446 --> 00:03:58.666 when he formulated his hypothesis, 00:03:59.086 --> 00:04:02.306 but what he was getting at was, okay, when you have the initial experience, 00:04:02.366 --> 00:04:06.886 you have A, B, C, D, E firing, and then you can have plasticity processes that 00:04:06.886 --> 00:04:11.046 strengthen the connections so that A, B, C, D, E is more likely to fire by itself. 00:04:11.486 --> 00:04:14.426 And so that's a familiar example, I think, to many people who are students of 00:04:14.426 --> 00:04:19.706 neuroscience, that somehow signals that go through the system can also instruct the system, right? 00:04:19.766 --> 00:04:22.866 The experience itself has an effect that's immediate, and then there's some 00:04:22.866 --> 00:04:27.006 kind of long-term plasticity. And when I was giving my lecture today, 00:04:27.126 --> 00:04:32.266 I pointed out the fact that these complex spikes could likewise play two roles. 00:04:32.366 --> 00:04:37.766 One, immediate role in driving processing right then and there to guide behavior. 00:04:38.226 --> 00:04:41.366 And then the other thing they could do is they could teach plasticity. 00:04:41.406 --> 00:04:45.026 So in some respects, I would say that this idea is not any different from Hebb's idea. 00:04:45.146 --> 00:04:49.346 So the idea that this interrupt signal can do something right away and can also 00:04:49.346 --> 00:04:53.186 drive a plasticity process. And I guess what I was getting at is that, 00:04:53.266 --> 00:04:55.246 at least given what people know 00:04:55.246 --> 00:04:59.646 now about this pathway, it's not really possible to separate those two. 00:04:59.746 --> 00:05:03.026 At least at the level of the Purkinje cells, they happen together. 00:05:03.286 --> 00:05:07.326 When one happens, the other happens. Okay, now I should say that there actually 00:05:07.326 --> 00:05:10.066 is an exception to this, and it's okay if I go into the exception. 00:05:10.526 --> 00:05:14.666 The exception occurs at the level of the synchrony. So I showed in my talk today 00:05:14.666 --> 00:05:20.946 imaging data that suggests to us that many olivary neurons can fire together, 00:05:21.226 --> 00:05:23.006 and therefore they evoke synchrony. 00:05:25.066 --> 00:05:29.726 Complex spikes in many Purkinje cells at once. And so the exception to this 00:05:29.726 --> 00:05:33.506 is that this interrupt signal, when it's in the form of synchrony across many 00:05:33.506 --> 00:05:38.446 cells, could be processed elsewhere, not in the cerebellar cortex. 00:05:38.706 --> 00:05:43.446 But when they fire all together, individually, they can drive plasticity in the Purkinje cells. 00:05:43.626 --> 00:05:46.766 But when they fire together, that might be something special. 00:05:46.926 --> 00:05:48.766 And I didn't get to talk about this much during the lecture. 00:05:48.986 --> 00:05:51.586 But there's an idea that we've been been very interested in my laboratory, 00:05:51.786 --> 00:05:55.526 which is that when these neurons all fire together, they can converge, 00:05:55.726 --> 00:06:01.366 and then that convergence can be detected as a special event in the deep nuclei. 00:06:01.706 --> 00:06:05.146 And that's another kind of event that can be encoded in this pathway. 00:06:05.426 --> 00:06:08.866 Okay, clear. Because that would mean that then you might be able to distinguish, 00:06:08.906 --> 00:06:14.706 let's say, a more localized reset versus really a plasticity controlling signal. 00:06:14.906 --> 00:06:18.826 Yeah. So I would say that if one olivary neuron fires and one Purkinje cell 00:06:18.826 --> 00:06:21.426 receives that, or sorry, a few dozen PNG cells receive that, 00:06:21.566 --> 00:06:27.526 then that's an interrupt to those neurons, and it's something that could drive 00:06:27.526 --> 00:06:28.626 plasticity in those neurons. 00:06:28.946 --> 00:06:33.626 If a bunch of olivary neurons fire at once because they're coupled by gap junctions, 00:06:33.866 --> 00:06:37.806 then that's the signal that has the additional property of maybe triggering 00:06:37.806 --> 00:06:39.246 something interesting in the deep nuclei. 00:06:39.606 --> 00:06:44.366 Okay. And that interesting thing could be, again, immediate readout or it could 00:06:44.366 --> 00:06:45.686 be teaching plasticity. 00:06:45.926 --> 00:06:48.246 And there's evidence in the literature. If you look in the literature, 00:06:48.246 --> 00:06:51.386 you can find evidence that there's something interesting that happens in deep 00:06:51.386 --> 00:06:55.706 nuclear cells, conditions under which you can get plasticity of the deep nuclei. 00:06:56.026 --> 00:06:59.566 And this could be an event that drives that plasticity. 00:07:00.543 --> 00:07:05.683 But that plasticity would then depend on this step over the Purkinje cells, 00:07:05.943 --> 00:07:10.023 or it would depend on the collaterals from the climbing fibers directly to the deep nucleus? 00:07:10.423 --> 00:07:14.963 Direct collaterals. So when I think about the cerebellum, what I visualize in my mind is this. 00:07:15.863 --> 00:07:20.883 I think about relatively direct arcs that are purely excitatory, 00:07:21.003 --> 00:07:23.903 for instance, input from the rest of the brain into the olive, 00:07:24.103 --> 00:07:27.043 and then an excitatory arc from the olive to the deep nuclei, 00:07:27.243 --> 00:07:29.663 and then the deep nuclei come back out to the rest of the brain. 00:07:29.663 --> 00:07:35.143 So think of that as a direct excitatory reflex arc, perhaps like what Sherrington 00:07:35.143 --> 00:07:39.943 said about the reflex arc with the knee reflex in the spinal cord. 00:07:40.003 --> 00:07:41.103 And the pen in the slip machine. 00:07:42.323 --> 00:07:47.203 I guess, yeah, right. Things that we don't think of as being too cognitively sophisticated. 00:07:47.683 --> 00:07:51.623 And then sitting on top of that is the second loop, and the second loop is some 00:07:51.623 --> 00:07:55.863 inhibitory thing that involves Purkinje cells, where you have excitation from 00:07:55.863 --> 00:08:00.323 the same axon, except it's a branch of that axon, and that branch is called a climbing fiber. 00:08:00.543 --> 00:08:04.583 Then it goes up to the Purkinje cells, and then that guy is an inhibitory neuron 00:08:04.583 --> 00:08:06.463 that again comes to the deep nuclei. 00:08:06.763 --> 00:08:10.883 It's a direct excitatory arc and an indirect inhibitory arc. 00:08:11.403 --> 00:08:16.783 One can even think of the mossy fiber pathway as being, with some details different, 00:08:17.003 --> 00:08:21.043 the same kind of system, where you have a direct arc that's excitatory and an 00:08:21.043 --> 00:08:22.503 indirect arc that's inhibitory. 00:08:22.503 --> 00:08:28.203 Basically, I guess when I look at cerebellar cortex, I see this inhibitory neuron. 00:08:28.363 --> 00:08:32.083 These Purkinje cells are basically inhibitory interneurons that got out of control. 00:08:32.203 --> 00:08:34.163 They hypertrophied over the course of evolution. 00:08:34.543 --> 00:08:40.763 They became so big and important that they started developing this massive convergence of parallel fibers. 00:08:41.083 --> 00:08:44.403 And they're so big and important, they get their own interneurons. 00:08:44.523 --> 00:08:47.143 And so you have an interneuron that's got its own assistance. 00:08:47.523 --> 00:08:49.243 So you've got layers of interneurons. 00:08:49.843 --> 00:08:51.883 But these interneurons you would see as... 00:08:52.503 --> 00:08:55.963 Controlling the function of deep nucleus 00:08:55.963 --> 00:08:58.743 and downstream motor pathways or you see it 00:08:58.743 --> 00:09:01.483 as an inhibitory control over if you 00:09:01.483 --> 00:09:05.823 want the rest of the brain as well well everything that they do has to get filtered 00:09:05.823 --> 00:09:10.003 through the deep nuclei because they only synapse onto the deep nuclei and so 00:09:10.003 --> 00:09:13.623 ultimately everything has to go through the deep nuclei so i would think of 00:09:13.623 --> 00:09:18.503 them either as getting back to the immediate versus long-term thing either as 00:09:18.503 --> 00:09:20.683 neurons whose output shapes the deep nuclei, 00:09:20.783 --> 00:09:23.283 the activity in the deep nuclear neurons, 00:09:23.483 --> 00:09:25.443 or perhaps as a source of instruction. 00:09:25.743 --> 00:09:30.643 So again, we have these two branches from, say, the olive going directly to 00:09:30.643 --> 00:09:35.483 the deep nuclei, and another branch going up to the Purkinje cells and then into the deep nuclei. 00:09:39.848 --> 00:09:46.488 Present one branch of input to the uh to the deepness yeah is that right okay yeah but then, 00:09:47.128 --> 00:09:49.808 if so if we if you focus a bit on 00:09:49.808 --> 00:09:53.448 these on these climbing fibers okay here we go with climbing fibers inferior 00:09:53.448 --> 00:09:58.428 olive going off one hertz more or less yeah and then we might have some modulation 00:09:58.428 --> 00:10:03.608 of this signal now on top of this we have a negative feedback on those responses 00:10:03.608 --> 00:10:07.728 going back from the perkin yourself to the deep nucleus with inhibition onto 00:10:07.728 --> 00:10:09.668 the inferior olive which can maybe 00:10:09.728 --> 00:10:13.308 add some jitter now to this one Hertz activity. 00:10:13.508 --> 00:10:18.128 Yeah. And on top of this, we can have now, let's say special kinds of events coming in. 00:10:18.168 --> 00:10:24.188 Let, let's say I, um, air air puffs to the, to the eye. If we talk about eye blink conditioning. 00:10:25.148 --> 00:10:29.968 So it's not a lot of mixing of, of possible states onto these, 00:10:30.028 --> 00:10:35.708 these oscillating, slow oscillating neurons, how do perky yourselves disentangled is right. 00:10:35.848 --> 00:10:41.368 So, so one, let's see, that's a great question. Um, I, uh, I'm not necessarily 00:10:41.368 --> 00:10:45.028 going to give a very clear answer about this, but one thing I'm thinking about is this. 00:10:45.208 --> 00:10:50.068 If you look at an individual olivary neuron that sends its climbing fibers to 00:10:50.068 --> 00:10:53.328 the Purkinje cells, it fires on average about once a second. 00:10:53.468 --> 00:10:56.928 What's observed is that it seems like most of the time it fires about once a 00:10:56.928 --> 00:11:01.288 second. Maybe transiently it can go faster or slower, but basically at one hertz. 00:11:01.468 --> 00:11:05.268 And that's a hard coding problem, because if you imagine, just imagine that 00:11:05.268 --> 00:11:09.028 you're receiving input from the inferior olive, and you're getting about one spike a second. 00:11:09.068 --> 00:11:13.308 Pop, pop, pop, pop, pop, right? 00:11:13.648 --> 00:11:17.408 Five of them came in five seconds and now you're supposed to extract, 00:11:17.548 --> 00:11:20.688 you know, you're the Purkinje cell or you're the deep nucleus and you're trying 00:11:20.688 --> 00:11:21.668 to extract meaning from that. 00:11:21.868 --> 00:11:25.508 And that's kind of a hard problem. But now if you add synchrony on top of that, 00:11:25.628 --> 00:11:29.528 now imagine that two different Purkinje cells are receiving those kinds of inputs. 00:11:29.968 --> 00:11:33.888 And then at certain points, this magical event happens where a number of them 00:11:33.888 --> 00:11:36.648 all receive an input at the same time. 00:11:37.088 --> 00:11:41.988 That can extract some olivary spikes from a background of ongoing activity. 00:11:42.308 --> 00:11:44.968 And so I think the synchrony is interesting because it allows the possibility 00:11:44.968 --> 00:11:50.168 of extracting features from this ongoing stream of slow asynchronous spikes. 00:11:51.228 --> 00:11:54.708 And again, coming back to the deep nuclei, I mean, maybe one answer to your 00:11:54.708 --> 00:11:58.728 question is that the deep nuclei are one way for this information to get read 00:11:58.728 --> 00:12:02.348 out, and the Purkinje cells are another way. So let's back up and think about it a little bit. 00:12:02.808 --> 00:12:05.608 Think about what the olive sounds like to these two structures. 00:12:05.608 --> 00:12:11.248 If you look at the olive-to-deep-nuclear pathway, that's an excitatory projection, 00:12:11.528 --> 00:12:14.368 and the only thing that's coming in on it is olivary spikes. 00:12:14.728 --> 00:12:18.648 So that's an easy detection problem. Okay, now if you think about... 00:12:19.674 --> 00:12:22.634 Olive, to the Purkinje cells, to the deep nuclei. 00:12:22.914 --> 00:12:27.134 That's a hard detection problem because those same spikes are coming in, 00:12:27.214 --> 00:12:31.274 but they're against this continuous wash of simple spikes that are being driven 00:12:31.274 --> 00:12:33.314 by the parallel fiber pathway, right? 00:12:33.474 --> 00:12:37.054 And so there's this wash of spikes coming through with these little guys riding 00:12:37.054 --> 00:12:40.014 on top of it. And it's not so obvious how you would detect that. 00:12:40.214 --> 00:12:42.934 Right. But you gave a pretty good imitation of that in your lecture. 00:12:43.034 --> 00:12:44.654 So what does it sound like in the lab? 00:12:45.494 --> 00:12:48.594 Well, so it's in the, you know, these things are recorded optically. 00:12:48.654 --> 00:12:50.854 And so So actually, it turns out it doesn't sound like anything. 00:12:52.834 --> 00:12:57.154 But the thing I was doing in front of the, you know, in the lecture today is 00:12:57.154 --> 00:12:59.994 I was having each of my fingers be one Purkinje cell. 00:13:00.954 --> 00:13:05.754 And when they're firing asynchronously, it looks like you're playing some kind 00:13:05.754 --> 00:13:07.414 of random set of keys, say, on a piano. 00:13:08.274 --> 00:13:12.374 But when you have a synchrony event, what that looks like is like playing a chord on a piano. 00:13:12.514 --> 00:13:15.354 And so you have many fingers all coming down at the same time. 00:13:15.354 --> 00:13:20.734 And so what I was demonstrating for people is the idea that somehow these synchrony 00:13:20.734 --> 00:13:24.334 events are like chords being played in the olive. Right. 00:13:24.914 --> 00:13:29.754 So then, already now we look at what you call this hard coding problem for the 00:13:29.754 --> 00:13:31.474 Purkinje cells with respect to the inferior olive. 00:13:33.114 --> 00:13:38.814 But apparently this became more complex because now with this imaging work that 00:13:38.814 --> 00:13:40.034 you have been developing, 00:13:40.854 --> 00:13:43.794 here we have the mouse on the styrofoam ball 00:13:43.794 --> 00:13:47.034 running around and you're imaging the climbing fibers 00:13:47.034 --> 00:13:51.254 and the Purkinje cells in particular and 00:13:51.254 --> 00:13:56.934 suddenly what we see is that it's fairly complex synchronized responses among 00:13:56.934 --> 00:14:03.554 the subgroups of Purkinje cells so is it still in the same ballpark as you just 00:14:03.554 --> 00:14:07.314 described the dynamics of the system or do you think it adds a whole new layer 00:14:07.314 --> 00:14:08.574 of complexity complexity. 00:14:11.158 --> 00:14:18.938 Oh boy. Let's see. So I don't know how to answer that because I know that people 00:14:18.938 --> 00:14:23.858 in the field have been very interested in the idea that somehow the signals 00:14:23.858 --> 00:14:26.498 themselves are used in processing. 00:14:26.718 --> 00:14:29.158 So for instance, people are interested in the fact that when a Purkinje cell 00:14:29.158 --> 00:14:33.238 fires a complex spike, there's a brief pause of tens of milliseconds right after 00:14:33.238 --> 00:14:36.658 it fires the complex spike where there are no sodium spikes, 00:14:36.878 --> 00:14:38.798 and then the sodium spikes begin again. 00:14:38.798 --> 00:14:44.758 So it sounds like, if you slow it down, it sounds like where the complex spike 00:14:44.758 --> 00:14:50.518 comes in, and there's a pause, and then the sodium spikes go and start up again. 00:14:50.698 --> 00:14:54.598 And so there's this characteristic shape, and one focus of research has been 00:14:54.598 --> 00:14:58.698 the idea that that and the pause somehow encode information, 00:14:59.078 --> 00:15:04.358 and that's a salient feature that then presumably gets picked up by the deep nuclei. 00:15:04.358 --> 00:15:07.618 So that's a feature that can sit on top of the simple spike stream. 00:15:07.898 --> 00:15:12.658 And that's a candidate for what could get processed by the deep nuclei. 00:15:13.158 --> 00:15:16.998 So I think I'm giving you a little bit of an unclear answer, 00:15:17.158 --> 00:15:20.858 but I'm saying that essentially it could be a feature that could be processed 00:15:20.858 --> 00:15:22.698 on top of the sodium spike stream. 00:15:22.698 --> 00:15:29.978 So are you confident that there's enough evidence available that would show 00:15:29.978 --> 00:15:34.298 us that indeed the deep nucleus could make sense from such a response? 00:15:34.418 --> 00:15:38.278 Because in some sense, the deep nucleus has to invert this now. 00:15:38.758 --> 00:15:42.098 If the deep nucleus wants to report to the rest of the system, 00:15:42.218 --> 00:15:47.158 something happened in the cerebellar cortex, it has to convert this pause into 00:15:47.158 --> 00:15:50.278 an action potential. So how do we do that? 00:15:50.838 --> 00:15:52.818 That's a great question. I think 00:15:52.818 --> 00:15:55.358 that biophysically, it feels like it's a little bit of a hard problem. 00:15:56.638 --> 00:16:03.678 This pause, when people have recorded it in vivo, has only been a few tens of milliseconds. 00:16:04.058 --> 00:16:06.598 If you think about synaptic mechanisms that could read that out, 00:16:06.798 --> 00:16:08.638 that's a little bit of a hard problem. 00:16:08.638 --> 00:16:13.038 And I think the way that that problem becomes easier is to get back to the synchrony 00:16:13.038 --> 00:16:18.038 is if a lot of, it may be a pretty short pause, but if a lot of Purkinje cells 00:16:18.038 --> 00:16:20.738 do it at once, then it has the advantage of summing, right? 00:16:20.778 --> 00:16:24.338 So if you think about, I don't know, 100 Purkinje cells all converging on one 00:16:24.338 --> 00:16:27.738 deep nuclear neuron, and they're all pounding away with their sodium spikes, 00:16:28.158 --> 00:16:32.698 if those pauses come at different times, that is not going to be a very impressive 00:16:32.698 --> 00:16:33.598 event to the deep nucleus. 00:16:33.598 --> 00:16:38.378 But the moment you synchronize them all with one another, now you have everybody pausing. 00:16:38.478 --> 00:16:42.058 It's like those moments in the orchestra when the orchestra is playing and playing 00:16:42.058 --> 00:16:43.738 and playing, and then everybody stops all at once. 00:16:43.938 --> 00:16:46.318 That's a really interesting event when you're listening to the orchestra. 00:16:46.638 --> 00:16:50.538 And so it's something like that, where synchrony buys you something that the 00:16:50.538 --> 00:16:53.558 pause itself might not be very good at without the synchrony. 00:16:54.441 --> 00:17:01.081 Would you say that we need something like a rebound polarization to invert the pause into activity? 00:17:02.621 --> 00:17:07.281 That is certainly the feeling that's been prevalent in the field. 00:17:07.801 --> 00:17:12.781 Not all deep nuclear neurons rebound, but the phenomenon, to remind people, 00:17:12.941 --> 00:17:17.161 is that when you hyperpolarize a deep nuclear neuron and then you let go, 00:17:17.461 --> 00:17:21.261 at the moment that you let go, there's a rebound and there's a little burst 00:17:21.261 --> 00:17:24.521 of spikes, and then the neuron resumes spiking again. 00:17:25.241 --> 00:17:30.421 So one major belief in the field of people who study cerebellum is that that 00:17:30.421 --> 00:17:31.461 could happen, for instance, 00:17:31.621 --> 00:17:35.341 when you say, when the Purkinje cells pause, they provide inhibition, 00:17:35.461 --> 00:17:39.781 they stop providing inhibition, and that's the equivalent of release from inhibition, 00:17:39.881 --> 00:17:43.161 and then you get a rebound. So that's the prevailing view in the field. 00:17:44.061 --> 00:17:48.681 I think that is a candidate for how information could be encoded. 00:17:49.161 --> 00:17:53.001 Another possibility is that there might be other interesting biophysical events 00:17:53.001 --> 00:17:54.781 that nobody's observed yet. 00:17:54.981 --> 00:17:59.621 So one possibility we're very interested in in my laboratory is the idea that 00:17:59.621 --> 00:18:02.301 at the time of that rebound, there might be something biophysically interesting 00:18:02.301 --> 00:18:03.761 happening in deep nuclear cells. 00:18:03.921 --> 00:18:07.221 Something like, say, a dendritic action potential, something like that. 00:18:07.221 --> 00:18:12.641 And so I think there are levels of that that haven't yet been explored because 00:18:12.641 --> 00:18:15.001 people have not had the imaging technology available. 00:18:16.884 --> 00:18:22.164 Some of the data you showed us showed this very curious phenomenon where, 00:18:22.244 --> 00:18:25.804 for instance, you might initially be able to trigger climbing fiber responses 00:18:25.804 --> 00:18:29.624 driven by some metasensory stimulus like an air puff to the snout. 00:18:30.584 --> 00:18:37.004 And later, you might see the same climbing fiber responding in a way that seems 00:18:37.004 --> 00:18:39.084 to be correlated with motor action. 00:18:39.704 --> 00:18:41.944 So how should I interpret this? 00:18:43.084 --> 00:18:46.624 Well, okay. So first, the empirical observation is this. the empirical observation 00:18:46.624 --> 00:18:51.524 is that these climbing fibers are active the animal is under the microscope 00:18:51.524 --> 00:18:56.064 it's a mouse and it's also standing on a foam ball and it can walk freely and 00:18:56.064 --> 00:18:58.984 it's an awake animal and under this condition, 00:18:59.144 --> 00:19:03.764 this is collaborative work that we've done with David Tank, under this condition 00:19:03.764 --> 00:19:07.984 what we can see is that the climbing fibers sorry, the complex spikes that we observe. 00:19:10.204 --> 00:19:15.284 Are activated when the animal's resting when we give a stimulus to the animal. 00:19:15.364 --> 00:19:20.164 Like if we give it a clapping sound, or if we apply an air puff to its hindquarters. 00:19:21.064 --> 00:19:23.644 And we originally did this to try to get the animal to walk. 00:19:24.024 --> 00:19:27.044 And because we were trying to get the animal to do something, 00:19:27.164 --> 00:19:30.504 we've got the animal on the ball, we think, well, if we could poke it, 00:19:30.564 --> 00:19:33.224 maybe it'll start walking, and then that'll be interesting. So that was our 00:19:33.224 --> 00:19:34.304 original reason for doing it. 00:19:34.804 --> 00:19:40.004 So we give the animal a poke, and we can see that a bunch of these complex spikes 00:19:40.144 --> 00:19:42.584 fire across the population of Purkinje cells at once. 00:19:42.604 --> 00:19:45.864 So a chord is played to follow the musical metaphor. 00:19:46.864 --> 00:19:50.484 And that happens when the animal's resting, when there's a stimulus given. 00:19:50.884 --> 00:19:54.564 But then when the animal starts walking, there's something that happens that. 00:19:55.594 --> 00:19:59.334 There's some kind of switch that happens. And the switch that happens is that 00:19:59.334 --> 00:20:02.514 the same population of cells that we've been observing all along, 00:20:02.654 --> 00:20:08.534 dendrites that we've been imaging, start generating lots of cords without our input. 00:20:08.734 --> 00:20:13.994 So it's some kind of cords that are synchronous firing events that are self-generated. 00:20:14.434 --> 00:20:19.034 Somehow by the animal's movement, there's something about the animal's movement 00:20:19.034 --> 00:20:20.454 that generates lots of these. 00:20:20.634 --> 00:20:24.194 And under that condition, there's lots of these synchronous events happening. 00:20:24.194 --> 00:20:28.654 But the other thing that happens is that now when we apply stimuli, no response. 00:20:28.974 --> 00:20:33.834 It just becomes insensitive to those stimuli. So the same set of Purkinje cells, 00:20:34.074 --> 00:20:37.834 whose complex spikes were previously sensitive to external input, 00:20:38.054 --> 00:20:43.394 now seem to be sensitive to self-generated activity, something that the animal itself is doing. 00:20:43.594 --> 00:20:48.454 And so there's some kind of gating where it gates from external to internal events. Right. 00:20:48.634 --> 00:20:52.314 So how can you be sure that it's really internally generated? 00:20:52.834 --> 00:20:58.354 You could argue like, well, maybe it is just another form of somatosensory stimulation 00:20:58.354 --> 00:21:00.394 because of the movement. 00:21:01.414 --> 00:21:03.554 It could be. It could be things like joint movement. 00:21:06.134 --> 00:21:09.174 We've done a little bit in that regard to try to tease that apart. 00:21:09.414 --> 00:21:13.674 One thing we've done is when the animal's standing there passively, we rotate the ball. 00:21:13.954 --> 00:21:19.294 When we rotate the ball and the animal's standing there, we're inducing some 00:21:19.294 --> 00:21:23.554 degree of passive movements, and we do not see synchronous firing events under that condition. 00:21:23.554 --> 00:21:26.934 So that suggests to us that it's something more than just, say, 00:21:26.934 --> 00:21:31.754 a joint movement or some kind of somatosensory or proprioceptive event. 00:21:34.534 --> 00:21:38.874 Honestly, I'm not sure we know exactly what it is, but we do know that we can't 00:21:38.874 --> 00:21:40.754 get it by the animal passively being moved around. 00:21:42.208 --> 00:21:47.968 But then, so in the concepts that you're developing, we started with this notion 00:21:47.968 --> 00:21:50.908 of climbing fibers as interrupt signals. 00:21:51.228 --> 00:21:56.068 Yes. And now we move to this notion of gating, which more has to do with the 00:21:56.068 --> 00:22:01.988 different input streams can be sort of re-channeled into driving the same set 00:22:01.988 --> 00:22:02.848 of these interrupt signals. 00:22:03.448 --> 00:22:08.648 So what are the functional consequences of that? How should I think about gating and interrupt? 00:22:09.328 --> 00:22:14.528 That's a great question. I think that for the time being, tentatively, 00:22:14.568 --> 00:22:18.868 I might imagine that the system is sensitive to different kinds of signals depending 00:22:18.868 --> 00:22:21.328 on what is behaviorally needed. 00:22:21.548 --> 00:22:25.488 So when the animal's at rest, what it needs is to be sensitive to external events 00:22:25.488 --> 00:22:27.608 that say, okay, it's time for you to walk. 00:22:27.748 --> 00:22:30.968 Something surprising has happened. It's time for you to react in some way. 00:22:31.448 --> 00:22:34.568 And then once the animal's walking, well, under that condition, 00:22:34.908 --> 00:22:37.828 then there are a lot of internal signals being generated that are more important 00:22:37.828 --> 00:22:40.188 for the animal to deal with when it's handling its walking. So, 00:22:40.308 --> 00:22:44.628 I think that I was talking to you or someone else earlier today when I made the joke. 00:22:44.828 --> 00:22:48.428 It's something like, don't talk to me now, I'm tying my shoes. 00:22:48.828 --> 00:22:53.048 And it's like that, where when you're tying your shoes, what you don't want 00:22:53.048 --> 00:22:54.288 is someone telling you stuff. 00:22:54.388 --> 00:22:57.588 What you want is to focus on your shoes, or whatever it is that you're doing 00:22:57.588 --> 00:22:58.648 in order to tie your shoes. 00:22:58.648 --> 00:23:04.068 So what I suspect is that some kind of gating whose behavioral function could 00:23:04.068 --> 00:23:10.388 be to let in relevant information and not let in information that's of less 00:23:10.388 --> 00:23:11.528 use at that particular moment. 00:23:12.228 --> 00:23:17.328 But in some sense, this raises the next question fairly automatically, no? 00:23:17.408 --> 00:23:22.568 Because then you're implying that the circuit that is affected by these climbing 00:23:22.568 --> 00:23:28.028 fiber signals is playing a functional role in both behavioral contexts. 00:23:29.288 --> 00:23:34.028 Yes, that is what I'm implying. There's some work that precedes this from a 00:23:34.028 --> 00:23:37.348 single-neuron recording in The Olive and also upstream of The Olive, 00:23:37.508 --> 00:23:41.548 I think mainly in The Olive, from this fellow in the UK, Richard Apps. 00:23:41.908 --> 00:23:46.808 And Apps and his collaborators have been very interested in that kind of switched information. 00:23:47.308 --> 00:23:51.248 And I believe, if I recall correctly, that they can see switching on a pretty rapid timescale. 00:23:51.988 --> 00:23:57.048 One result that comes to mind is that when a cat is walking on a treadmill, 00:23:59.328 --> 00:24:03.248 if you deliver a tactile stimulus to the cat's foot, 00:24:04.146 --> 00:24:08.706 How much climbing fiber response you get depends on what stage of the stride. 00:24:08.806 --> 00:24:12.506 So if the animal's in stance, then if I remember correctly, if the animal's 00:24:12.506 --> 00:24:13.786 in stance, then there's no response. 00:24:14.106 --> 00:24:16.546 If the animal's in stride, then there is a response. 00:24:17.106 --> 00:24:20.666 And the functional difference there is that when the animal's in stance, 00:24:20.786 --> 00:24:24.906 it's getting tactile impotence. So of course, something that touches it might not be so interesting. 00:24:25.046 --> 00:24:28.446 But if the animal's in stride, then touching it is going to be a more interesting event. 00:24:28.446 --> 00:24:35.226 And that's been demonstrated at the level of single cells using electrophysiological 00:24:35.226 --> 00:24:39.526 conventional electrode-based methods, but with pretty rapid time switching, 00:24:39.706 --> 00:24:41.126 like on the timescale of walking. 00:24:42.126 --> 00:24:45.446 Still seconds or hundreds of milliseconds. Yeah, definitely pretty fast. 00:24:45.586 --> 00:24:48.646 I mean, as you know, the pace of a cat walking on a treadmill. 00:24:48.726 --> 00:24:54.806 Okay. So then in some sense you're saying it's like the parsing of sensory information 00:24:54.806 --> 00:24:57.226 in the context of behavior. 00:24:58.046 --> 00:25:03.566 Yeah. So for instance, we're now getting pretty far away from any of the data that I showed. 00:25:03.786 --> 00:25:07.986 But think of one common thing that people talk about when they talk about cerebellar 00:25:07.986 --> 00:25:10.826 function, which is the phenomenon that you can't tickle yourself. 00:25:11.086 --> 00:25:16.106 There's some kind of sensory cancellation where when you stroke yourself in 00:25:16.106 --> 00:25:20.106 your abdomen, then that doesn't tickle. But when someone else does it, then that does tickle. 00:25:20.166 --> 00:25:22.946 And so there's very clearly this contextual processing of the 00:25:23.006 --> 00:25:25.806 same sensory information whether you're generating it or 00:25:25.806 --> 00:25:30.306 whether you know whether somebody else is doing it under some silly condition 00:25:30.306 --> 00:25:34.806 right and so something like that is very clearly something we have to do every 00:25:34.806 --> 00:25:39.846 day at every moment and so one possibility here is that this is one neural correlate 00:25:39.846 --> 00:25:45.426 of that okay so now um to switch a little bit topic um. 00:25:46.250 --> 00:25:51.570 So coming from physics, in some sense, you've made a rather dramatic transformation 00:25:51.570 --> 00:25:55.790 because now you're really a very much data-oriented physiologist, if you want. 00:25:56.130 --> 00:25:59.530 And you would expect as a physicist, you still be looking for, 00:25:59.590 --> 00:26:01.530 let's say, theory models and so on. 00:26:01.630 --> 00:26:04.950 So what's your opinion of the current state of the art in our, 00:26:05.030 --> 00:26:08.650 let's say, theoretical understanding of this system? For instance, 00:26:08.670 --> 00:26:12.330 you have these theories floating around about, let's say, inverse models or 00:26:12.330 --> 00:26:16.150 forward models, adaptive filters, reinforcement learning models. 00:26:17.010 --> 00:26:20.230 How well are we doing there, in your opinion? 00:26:20.830 --> 00:26:23.710 Well, I think those ideas are all very interesting ideas. I mean, 00:26:23.710 --> 00:26:27.190 basically, we're talking about ideas in which people either take inspiration 00:26:27.190 --> 00:26:31.830 from physics or from control theory, from engineering, to try to understand 00:26:31.830 --> 00:26:34.730 how these neural systems work. So let's see. 00:26:34.750 --> 00:26:38.330 So I'll state the positive thing, which is that these are simplified ideas that 00:26:38.330 --> 00:26:41.790 give us some conceptual framework for thinking about how neural systems work. 00:26:42.690 --> 00:26:47.230 They come from engineering in many cases, which is a very positive thing because 00:26:47.230 --> 00:26:50.590 nervous systems are shaped by natural selection. 00:26:50.750 --> 00:26:55.450 And in some sense, natural selection is basically nature's engineer that's trying 00:26:55.450 --> 00:26:58.070 to get things to work well. So those are positives. 00:26:59.330 --> 00:27:04.510 I think that one thing that those theories do is they provide a framework for 00:27:04.510 --> 00:27:06.190 then doing experiments to test the ideas. 00:27:06.930 --> 00:27:10.370 And so I think that right now we're at a stage where there's lots of ideas, 00:27:10.470 --> 00:27:14.750 and then I think well-designed experiments can then go in and test those ideas. 00:27:15.330 --> 00:27:19.390 But if I had to really be frank, I would say that many of those ideas will probably 00:27:19.390 --> 00:27:20.590 end up being wrong, right? 00:27:20.650 --> 00:27:23.550 Because those are tentative things, those are frameworks that we have to work 00:27:23.550 --> 00:27:27.390 with. and we go in and if we do a well-designed experiment, if we do a good 00:27:27.390 --> 00:27:31.030 experiment, then we can start weeding out wrong ideas. 00:27:31.830 --> 00:27:43.930 I think that one strong role of a well-formed theory is that you go do an experiment, 00:27:44.050 --> 00:27:48.730 and I've slowly formed the impression over time that it's perfectly okay to throw out the theory. 00:27:48.990 --> 00:27:52.590 That one should not get too sentimental about a particular theoretical idea. 00:27:54.270 --> 00:28:00.810 Of course, I'm an experimentalist, and so I guess the theorists say the same 00:28:00.810 --> 00:28:02.490 thing about the experimentalists. Sure. 00:28:03.430 --> 00:28:08.370 Where do you see this experimental protocol or the paradigm going that you are 00:28:08.370 --> 00:28:09.830 developing, which is fairly advanced? 00:28:11.556 --> 00:28:14.756 So the things that we are very interested in right now, basically, 00:28:14.836 --> 00:28:19.356 to state the obvious, the thing that these imaging methods can buy you is the 00:28:19.356 --> 00:28:22.876 capacity to probe many neurons at once in a working neural system. 00:28:23.136 --> 00:28:24.396 And that's very exciting. 00:28:24.856 --> 00:28:30.196 I think that the kinds of things that are coming up in the near future are development 00:28:30.196 --> 00:28:32.376 of tools to get better temporal resolution. 00:28:32.636 --> 00:28:35.896 So, for instance, if the tool we're talking about is use of a calcium-sensitive 00:28:35.896 --> 00:28:40.936 fluorescent dye to get those dyes to perform more quickly, Perhaps if a voltage-sensitive 00:28:40.936 --> 00:28:47.016 indicator ever becomes able to report single cells in vivo, then that would be the ultimate of that. 00:28:47.676 --> 00:28:51.796 Another direction that's exciting is using molecular biology to target these 00:28:51.796 --> 00:28:56.356 indicators so you can see specific cell types. That's obviously of interest. 00:28:57.456 --> 00:29:00.816 Another direction is optogenetics, which I believe came up during the discussion 00:29:00.816 --> 00:29:05.936 of my talk, where you can have light-activated ways of perturbing the tissue. 00:29:05.936 --> 00:29:09.416 So instead of the tissue telling you something, now you tell the tissue something, 00:29:09.596 --> 00:29:10.896 and you can perturb function. 00:29:11.716 --> 00:29:14.696 And then totally independent of all these things. 00:29:14.856 --> 00:29:18.056 So I think those are the natural outgrowths of imaging technology. 00:29:18.056 --> 00:29:22.116 And then there's this other thing that's sort of looming on the horizon that 00:29:22.116 --> 00:29:25.976 we haven't talked about at all, which is this other field of connectomics in 00:29:25.976 --> 00:29:30.056 which people are starting to do tracing with better and better technologies 00:29:30.056 --> 00:29:31.916 to reconstruct entire neural circuits. 00:29:32.176 --> 00:29:36.596 And I think that that's some distant future point of convergence between this 00:29:36.596 --> 00:29:42.076 kind of methodology to monitor and perturb function, and then anatomical methods 00:29:42.076 --> 00:29:45.056 to reconstruct the whole circuit. Right. So that would be the global... 00:29:46.610 --> 00:29:49.810 Very good. So to close off, I have two questions for you. 00:29:49.870 --> 00:29:55.890 So on the one hand, in your experience in this field, and this very sort of 00:29:55.890 --> 00:29:59.630 dedicated study of, in this case, the cerebellum, on the basis of the experience, 00:29:59.850 --> 00:30:03.530 what's the law of Sam Wang you would like to give to us we should adhere to? 00:30:03.890 --> 00:30:05.830 Oh, good Lord. Are you serious? Yeah, of course. 00:30:06.210 --> 00:30:09.390 A general rule for how to conduct oneself? Yes. Oh. 00:30:09.550 --> 00:30:14.610 How to study the brain, how to gain knowledge, how to explain brain function. You have freedom. 00:30:14.830 --> 00:30:17.970 It's not only about morality and ethics. I don't know. 00:30:18.230 --> 00:30:22.070 There are several things that I've been very interested in lately, 00:30:22.210 --> 00:30:24.630 and they were touched upon a little bit during Parthometra's talk, 00:30:24.770 --> 00:30:30.010 which he gave last Friday on September 3rd. 00:30:31.010 --> 00:30:38.130 He talked about principles that could guide a theorist as being development and evolution. 00:30:38.470 --> 00:30:41.890 I would say as an experimentalist, I'm very interested in that as well. 00:30:41.890 --> 00:30:45.530 So, so far, what we've talked about is experimental tools that one can bring 00:30:45.530 --> 00:30:47.190 to bear on understanding a neural system. 00:30:47.470 --> 00:30:52.350 But one thing I've been thinking about a lot lately is the role of natural selection 00:30:52.350 --> 00:30:56.930 in, say, conserving a system, namely that these neural tissues cost a lot of 00:30:56.930 --> 00:31:00.110 energy to operate, and the bigger they are, the more energy they take. 00:31:00.230 --> 00:31:03.210 And so maybe I want you to think a little bit about energetics as a guiding 00:31:03.210 --> 00:31:04.370 principle, because maybe neural 00:31:04.370 --> 00:31:07.270 systems are trying not to spend too much energy in doing what they do. 00:31:08.390 --> 00:31:12.370 And then evolution. Evolution can take any number of forms, but one thing that's 00:31:12.370 --> 00:31:15.510 very interesting to me these days is homology between different systems. 00:31:15.630 --> 00:31:20.010 Looking at different neural systems to look for guidance in how one, 00:31:20.210 --> 00:31:25.730 in my case, the mouse cerebellum works, maybe I should be interested in, say, electric fish. 00:31:25.730 --> 00:31:31.350 Or maybe I should be interested in, I don't know what, neuromodulatory systems 00:31:31.350 --> 00:31:32.450 in other parts of the brain. 00:31:33.150 --> 00:31:37.410 And I don't have a very good specific piece of advice to give, 00:31:37.550 --> 00:31:41.290 but the reason I'm bringing up these things is that these areas, 00:31:41.430 --> 00:31:44.510 development, evolution, and even neuroanatomy, 00:31:45.410 --> 00:31:49.150 are things that a physicist does not necessarily naturally take to. 00:31:49.330 --> 00:31:52.370 And so the reason I'm bringing these up is that as a former physicist. 00:31:53.330 --> 00:31:57.270 These have been unusually, unexpectedly interesting to me. 00:31:57.350 --> 00:32:00.790 I never thought that I would be interested in neuroanatomy, especially comparative 00:32:00.790 --> 00:32:04.610 neuroanatomy, but it turns out that against my will, I've become very interested in that. 00:32:05.010 --> 00:32:08.630 Okay, very good. And then the last one, if five years from now, 00:32:08.710 --> 00:32:12.990 I'm going to go visit you in your lab and say, look, Sam, five years back, 00:32:13.030 --> 00:32:17.550 you made this one prediction, and today I'm going to check whether it turned out to be false or true. 00:32:17.830 --> 00:32:21.590 What's this one prediction you would like to make today you really would stick your neck out for? 00:32:22.370 --> 00:32:26.830 A prediction? Yeah, one prediction. The inferior olive is a teacher to both 00:32:26.830 --> 00:32:28.490 the Purkinje cells and to the deep nuclei. 00:32:28.710 --> 00:32:32.810 And its main role is as a teacher of the Cerebellar circuit. 00:32:33.430 --> 00:32:35.650 Perfect. Sam Wang, thank you very much for this conversation. 00:32:36.150 --> 00:32:39.550 Okay, see you in five years. Good.