 I was born in Sri Lanka, but I grew up in the United States from the age of three. And so I'm really an American. But I know that all the privileges I have had in my scientific career came from a roll of the dice, a rotation of the wheel of chance. If the various numbers of lucky dice didn't have come up, I would be in a very different position. So it's really a privilege to have an opportunity like this to do something to even out the balance that you get from the fluctuations in life that create haves and have nots. And I think it's just an incredible opportunity to do that. And I just want to thank you for the privilege of doing my tiny, tiny part. OK, so another thing I have to apologize. Immediately after this talk, I have to leave. I have to go get a plane back to the United States. So we won't have that much of an opportunity to talk afterwards. So please, if there's anything you want to talk about, email me or Skype me. I'm always happy to discuss. And it's also my privilege to, actually today, really, I'm just a mouthpiece for work that was done in my lab by three remarkable individuals. Balad, Postdoc, Vivek, a former Postdoc and now at Northeastern and Meijin, a collaborator of mine at the University of Toronto on courtship and mating in C. elegans. OK, so a number of years ago, we asked ourselves, most people in C. elegans, most like me, work on very simple behaviors, some simple sensory of both behaviors. This thermosensory neuron gives rise to thermotaxis. This chemosensory neuron gives rise to chemotaxis. This mechanosensory neuron gives rise to mechanosensory responses or individual tapping on the sensory periphery in some way and get a behavioral response. And that's hard enough. But actually, we asked ourselves, is there something more sophisticated? Because most of our perceptions, most of our real, interesting perceptions in our lives, what's going on? What? Our multisensory phenomena, right? We use many, many sensory percepts to form our recognition of objects in the world. And that's sort of illustrated in this sort of example, this classic illustration, right? So all these blind men are interrogating an elephant. One thinks it's a snake. One thinks it's a wall. Because each of them have a very narrow perception of the object, the elephant. And it takes a group effort to recognize this elephant. If they were able to talk and discuss and match all the different perceptions they're getting to some maybe internal image of what an elephant should look like, they would know that they were near an elephant. So then we asked ourselves, does a simple worm, does it see elegans, do something like that? Is there a case in which the worm uses many, many phenomena, sensory phenomena, to integrate a perception that's behaviorally relevant? And it does. It does in the case of this behavior, probably the most complicated behavior that see elegans does and possibly the most important, right? It's mating and courtship. What you see are two worms, all right? So worms come in, see elegans come in two sexes, hermaphrodites, which can fertilize themselves. And males, which is this guy, which can only reproduce by mating with hermaphrodites. The male is, they're both millimeter long animals. The male is trying to insert its spicule into the vulva of the hermaphrodite, which is a hole on its ventral side. The hermaphrodite doesn't want to be mated. Hermaphrodites, they can make their own sperm and eggs. And it's only when they deplete all of their own sperm that they're willing to be mated by males, by other males. So it's a very complicated behavior, as you can see. There are many things going on. And we thought we'd try to understand it better. One nice thing about mating behavior is that actually, and this is not a small thing in neuroscience, to know what the animal wants to do during the whole behavior. The male has one goal in sight throughout the entire process, which is mating. But it does it through using a whole set of different motor programs, forward movement, backward movement, turning movements, scanning movements, spicule insertion, and eventually ejaculation and rest. And it transitions between this whole complicated set of motor programs through continuous sensory perception of what the hermaphrodite is doing. Of course, the hermaphrodite is responding to the male, and it's very complicated, as you saw in the previous movie. No two mating events are the same. And yet they are all organized by the same neural circuit. And one advantage we have in C. elegans is that the neural circuits have been mapped at least anatomically. In the 80s, John White and colleagues established the wiring diagram of the hermaphrodite through serial section electron microscopy. Basically it took a whole worm. They cut it up like salami. They used electron microscopes to find every cell in synapse. The male has been recently reconstructed. It's actually now possible to do these kinds of reconstructions much more rapidly. It's a whole other subject that my lab and Mae Jens and Jeff Lictman's lab is involved in that I'm not gonna talk about today. What they took 30 years to do, we can now do in a couple of months. But anyway, but the field can do in a couple of months. But that, never mind. We have the wiring diagrams. We know where every cell is. We know the names of every cell. And we know most are all of their synaptic connections. The male is here. All right, so the male actually, so the head, tail. The male actually has more neurons than the hermaphrodite. But almost all of its sex-specific neurons are in the tail. So this organ is what was used to insert into the hermaphrodite and inject sperm. And it has its own very complicated set of sensory neurons, interneurons, and motor neurons. All right, so this wiring diagram was reconstructed a few years ago. And many of the cell types are here. The takeaway from the wiring diagram is shallow. It's shallow. There are many different sensory neurons. There are not that many different interneurons. And then it's motor neurons. So it's like only one or two synapses separate sensory inputs to behavior. And you can draw modular circuits if you want. You can try to guess at them just based on connectivity. Take that with a grain of salt as I'll tell you later. And we would like to understand the complete workings of this circuit in what we think of as a complicated, it is a complicated behavior. It's partly complicated because this actual behavior requires or has a huge sensory periphery. So these are all the sensory neurons that about half the neurons in this structure are sensory neurons. And all of them are working. So there are mechanosensory neurons, pheromone sensing neurons, chemosensory neurons, guzzatory neurons, proprioceptive neurons, and all of them are working to guide this behavior. All right, chemotaxis can be done with one chemosensory neuron, mechanosensory avoidance can be done with one mechanosensory neuron. But it has been shown that ablation or deletion of almost any of these sensory neurons leads to a degradation of the behavior. It really is like a whole bunch of blind men interrogating an elephant, a whole bunch of sensory neurons interrogating a hermaphrodite to recognize it's hermaphrodite and do what it should do with respect to that hermaphrodite body. Most many of these neurons are in these rays which are sort of fan out. There is these post cloacal and hooks sencilla which are near the actual mating apparatus. A spiccule comes out here to go into the hermaphrodite. And what we'd like to understand is a circuit level model of sensation to behavior. I will say that we don't know much about how to build the circuit model but I'll show you the data. And maybe you can tell me. All right, so the study of mating behavior has been going on for a long time. My friend and colleague Paul Sternberg when he started his lab at Caltech I think in the early 80s, decided to start studying mating behavior because it was the hardest thing to do and that's the kind of guy he is. But his work in the 80s was very observational and because it wasn't possible to record from neurons and even neurogenetics was hard. It's such a fragile, complicated behavior that it's not so easy to get mutants that are very informative, right? You get a mutant that just gets rid of the behavior and then you don't know what to do. But in the last few years, we've become possible, it's become possible to do whole brain imaging and freely behaving C. elegans. People have done Manuel Zimmer and company have do this immobilized animals. Andy Leifer at Princeton, Vivek Venkatachalam and I, we do it in freely behaving animals and it's through these sort of customized, volumetric, multicolor microscopes work that's been published, at least some of the early technology that's been published. So we applied whole brain imaging to the problem. And I'll show you a movie but this is kind of what it looks like, right? We can see the hermaphrodite. It's fluorescently labeled in one way so you can see the body. And this is the spicula, this is the tail. This is the thing we're looking at. It's got a couple of colors, a red color to tell us where the nuclei are and a green color that tells us whether neurons are active or not. Or we do calcium imaging. So if a cell is active, it gets brighter in the green channel. If a cell is inactive, it gets dimmer in the green channel. The red channel doesn't do anything. It always just tells us X, Y and Z. So these microscopes can run at 10 volumes per second. They can run a long time, actually. I think now we can make 12 hour recordings at 10 hertz and get a huge amount of data for whole brain imaging. And this is kind of what it looks like. So we watch these worms for minutes or tens of minutes while they mate and this is what it looks like. So it looks like a flat projection but this is actually volumetric. It's just a Z projection. And the male is looking all around the hermaphrodite. It turns around the head, turns around the tail. Looking for that. That's the vulva. And often misses and has to try again. When the male is mating, actually it's always moving backward in its body movement. So C. elegans are undulatory. And usually the undulations push the head forward. When it's mating, it almost always pushes itself backward. It pushes the spicule tail backward along the body to find the vulva. Sometimes it goes back forward again. I'll tell you a little bit more of that in a second. But there, look, it's found the vulva. I think it fails to really attach at that time. But you probably saw some little bit in your life. And sometimes it falls off altogether and is to find it again. Typical mating behavior takes 10 or 15 minutes. We always like to look at completely unrestrained hermaphrodites, right? I mean, you can. They will mate with, actually now it's found it. And I think at some point it's inserted a spicule and then it ejaculates there. So it's done. And then everything goes quiet. Yeah, so you can look at male hermaphrodites that are immobilized in some way. And then the mating behavior is much, much quicker. The male doesn't have to fight to mate. But we wanted to elicit the whole richness that we could see. All right, so what do we do? We take those movies and then we analyze them, right? This is like a 10 minute long movie. So before the male has found the hermaphrodite, there's a search phase looking for the hermaphrodite. It finds it. Some neurons go off and other neurons go on. And all kinds of interesting structured activity happens until ejaculate sperm release and then something else. These movies are easy to get, but they're hard to analyze. Most of what we've done is sort of manual tracking. Now it's computer-assisted manual tracking. I think we're quite a bit faster than we were at start. But even so, we don't have that many movies right now. I think we have seven completely annotated. I'll tell you more about that later. But I think the important thing here is that we get to arrange not just the full set of neural activities or most of the neurons in the tail, but we also get the behavior in all of its complexity. So it's a complex behavior, but there are a finite number of variables that can define what the animal is doing at any point in time with respect to the hermaphrodite. Distance from the head, distance from the tail, forward, backward movement, curvatures, and so on. And we have a full record of that because we have the behavioral movie and the neural activity movie and we try to relate them. We know all the sensory inputs going in. We know all the motor outputs going out and we have the activity in between. And if that isn't enough for neural modeling, then we're in trouble. All right, so these are some of the movies we've gotten. Freely moving animal, freely moving animal, freely moving animal, freely moving animal, freely moving animal. This actually is a much less interesting movie because that animal's been locked down. That's a male, which is just glued, immobilized. And you can see there's rather a remarkable depletion of activity. And actually, I'd love to say we've discovered that, but we didn't. Andy Leifer at Princeton found this to be true. Many people have looked at immobilized animals and found rather very boring kind of, low dimensional activity that can be described with a small number of principal components. Just up, down, up, down, forward, backward, forward, backward, I think people think it is. When you look at a freely moving animal, so these are hermaphrodites here, neurons in a freely moving animal, the activity is much more chaotic and high dimensional and interesting. And then when you actually immobilize animal, it goes into this sort of on, off, on, off, on, off, on, off, very rhythmic, slow oscillatory behaviors that are quite a bit more boring. What I'll say is that what we learned from this is that behaving animals have fundamentally different neural dynamics than immobilized animals. When all of the feedback loops that guide behavior are intact, even in something as simple as C. elegans. And so it underscores for us why we do what we do and the way we do it. All right, so, yes. Well, people think it's forward, backward. Because there are certain neurons we can think of as command motor neurons for forward movement. When those are active on, those are, we think, okay, the animal might be in some fictive forward state. And when those are off, people think that it's in a fictive backward state. Yeah. All right, so it's easy to record from all the neurons, one by one in the volumetric image. What's hard, really hard, is putting the names on the neurons. So every neuron in C. elegans has a name. But we've been able to do it for almost all the neurons in the tail because of the stereotypy of the cell locations and also because we have friends who have very good cell-specific promoters that Tile, the male tail, Paul Sternberg has this whole Gal 4 collection that he's been letting us look at. So we've been able to put names on almost all the neurons and almost all the data sets. So seven data sets have almost all the neurons we've been able to identify. And that's extraordinary, although hard to do because it is done manually to go from the atlas to the names. Hopefully I'll get to tell you why, how we might be able to make that faster. But all right, so now we're in the position of having whole-brain imaging data sets of a complex behavior with all the sensory inputs defined, all the motor outputs defined and everything in between. And that, I think puts us in a rather interesting position. And we're trying to understand. And we're just doing the simple things that we think we know how to do to do it. So this is a simple cross-correlation of the neural activity patterns across the whole data set. Paralyzed cross-correlations for all the neurons across the data sets. And we see this sort of block-like structure. So these neurons are co-active, those neurons are co-active, those neurons are co-active, those neurons, those neurons, those neurons, and so on. It's more interesting than hermaphrodite head in an immobilized animal, which has maybe two blocks forward, backward. And because we have the movies, we can sort of assign, figure out what behaviors these different blocks correspond to, and they correspond to observable, recognizable modes, the searching phase before the animal finds the hermaphrodite is attributed to the activity of these neurons. The scanning phase, all the time the animal spends going around the hermaphrodite body, looking for their vulva, are active there. Turning, this is when it goes over the head or over the tail and tries to stay on the animal. Refractory period, the neurons that go on after the ejaculation, sperm release and vulva detection. Not you can see that there's some correlation between sperm release and vulva detection as you would imagine. These variables are all motor variables. Okay, so yeah. Before you do this slide, I'm just curious, so this is instantaneous correlation, right? Yeah. So do you think that the indicators that you're using are, it's a kind of a slow filter on a faster process, because you could have delay, you could have functional modules that come from delays too, right? Yeah, so this is GCAM 6M, so I know it's got a time constant in the order of 100 milliseconds, so it's a filter on top of that level. Right, so I mean, would you give any sort of benefits and also shifting around your correlations that you're looking for correlations in time too? Or you could get the same overall picture? Almost certainly something like that would be useful and any advice you have, we'll take. But I'll give you the big picture on the simple-minded things that we know how to do because we're simple-minded experimentalists. Okay, so. All right, so one interesting thing is that each of these modules, like the search module, is comprised of a full set of neural architecture from sensory neurons to interneurons to motor neurons, different sensory neurons, interneurons and motor neurons for searching behavior, looking for the, from aphrodite. A different set of complete sensory, interneurons, motor neurons, a whole brain in and of itself is for scanning, looking for the vulva once you're there. First, look at the hermaphrodite. Once you're in the hermaphrodite, look for the vulva. You have your whole little mini brain for that. That's an interesting architectural principle that might be at work. And then you have a set of sensory neurons, mostly sensory neurons and one interneuron, just to recognize the vulva. Okay, I'll tell you more about that in a second. And we've been taking, we have these cross correlation, population level things, and we learn a lot just looking at the, like maybe what you were getting at, looking at the detailed neural dynamics of individual neurons with respect to the behavior. All right, so, yes. Should you have a three-dimensional matrix? Three-dimensional matrix? Yeah, for the correlation, because here on your matrix, it seems that, I mean, you need to separate each event, do the searching, the mating, and so on and so on. I mean, when you're making the experiments, you're doing the correlation, but each correlation has a positive answer at a given time. After a while, there is no correlation, right? After a while, yeah, it's a correlation time, yes, yes. It's a correlation time, so may I? Could you build a three-dimensional one that has the time constants? Can you fix them so you can do that? Yes, you could do that. I'm just looking at t equals zero, no delay, correlation. All right, and you're right. You could do a delay looking at, I think what Greg was saying. There's probably information there, yeah. And, yeah, no, you're right. All right, so, I mean, like these neurons, it's sort of looking at the vulva-sensing neurons, right? So these neurons, actually, Paul Sturmburg identified these long ago. Actually, it's not. We want one more question. That cross-correlation stuff you showed was, you said it's averaged over animals, like how? It was averaged over animals, seven. It's different as the correlation matrix. Not different at all. Yeah, so, yeah, it's very, very stereotyped at the level of the correlation matrix. So these neurons, actually, these were hypothesized, these post-cluacal neurons right here. It's not a giant leap hypothesis that Paul Sturmburg had that they might have something to do with recognizing the vulva, because that's where they recognize the spicule. But, you know, in our activity patterns, we see exactly that. So the behavioral metric is here. Every time it gets to the vulva, it's there. Boom, boom, boom, boom, boom. Every time, or most of the times, you hit the vulva, the post-cluacal neurons go on. Actually, right here, actually, you're still in contact, but they go off. And so something happens there. There's some state-dependent behavior, right? Before, after mating, whether these neurons are on and off. I think that if I had to hypothesize why that might be, once you ejaculate, you want it to inactivate the trigger. This is the trigger you pull. PCB, PCC, you've hit the vulva, insert your spicule. Once you ejaculate, you want to turn the trigger off. You want to do it again. Maybe that has something to do with that. I'm just speculating. But actually, right after ejaculation, almost all the neurons in the male tail, actually even the male head, go off for a while. The rest period, which I assume has something to do with resetting the whole system, right? The male has interrogated this hermaphrodite for a long time, gotten a lot of information that's useful for this particular mating event, clear it before you do the next thing. Again, speculating. But it's like delete all in programming in basic or something. But anyway, so the whole brain imaging allowed us to find actually more sensory neurons than those sort of obvious anatomical ones that might be. And in fact, so these are the rays. We discovered that two of the rays, ray 2, specifically ray 2, is also activated in the proximity of the vulva. This is a behavioral coordinate activity. And you can see this particular ray, ray 2B, one of the neurons in those is two neurons in each ray. One of the neurons in this ray is activated by the vulva. So the rays are mechanosensory. So this is our first indication maybe that there is a multisensory integrative event that happens on recognition of the vulva. Touch, taste, smell, all of these things, all of these sensory inputs are used collectively to recognize the target. We know from ablations that the sensory neurons are involved. And we now know from ablations that the rays are involved as well. If you ablate the rays, stopping at the vulva is strongly compromised. And the interneuron, one neuron, PVX, which is the only interneuron we found involved in vulva recognition, the deletion of that also dramatically compromises vulva recognition. It's not easy to see the sensory connections between PVX and all this diversity of sensory inputs in the wiring diagram. There are secondary paths and so on, which breaks up some of the modularity that was easy to see in the connectome. But the functional connectivity, as I will say later, is a little more subtle than that. So it's a multisensory event. And so that's one kind of glimmer of interesting stuff. Another glimmer of interesting stuff is when we discovered that there are neurons, you think that detecting the vulva is obvious. You want to do that. We also discovered a neuron, PHA, which is activated in a very different place than the vulva, right here, in the anterior ventral side. Why might that be? So if you ask a worm biologist, a worm biologist is better versed in worm anatomy than I was or am. They would say, oh, that's immediately that they don't know what's there. There's an excretory pore there. It's like a tear duct near the head. It's too small to easily mate with, so it's not something like that. But this chemisensory neuron is activated only there. It's not activated when it happens to brush against a male excretory duct, if it self touches itself or touches other males. It only gets activated when it touches a hermaphrodite excretory pore. So there's something specific about that particular target. That tells the animal it's on the anterior ventral side. And we believe that this is an extra cue, right? So one thing that has been in my mind is that for the male to successfully mate with a hermaphrodite, it effectively has an internal model of what the hermaphrodite looks like. It has some internal model that it has to match a template of sensory inputs to that in order to know it's not an hermaphrodite and actually deal with the hermaphrodite in a proper way. The male knows that the hermaphrodite is sort of a millimeter long, skinny thing. The male knows that the hermaphrodite has an excretory pore on its anterior ventral side. And that excretory pore tells you where it is with respect to the vulva. And we have some behavioral evidence, actually, that is used in a possibly strategic way with respect to such a model. When the male, for example, is going this way along the hermaphrodite, hits the excretory pore, it often pauses and then does the uncharacteristic thing of going back this way again, right? Going from reverse movement, which is its state, which is what it ordinarily does, and going back to forward movement. And that makes sense because if you've traced a long way along the body and you hit the excretory pore, then you know the vulva is behind you. And what you ought to do to get there is change your direction. So it's a reasonable thing. It's a rational thing to do. It's an appropriate strategic choice of multi-sensory integration that integrates something about having traversed a long distance, hit the excretory pore, and gone back. Exactly how it does that calculation, I don't know. But I think that the excretory pore provides an additional male guiding cue that has something to do with which you reflect some knowledge of the hermaphrodite body to use. Just tail, head, excretory pore, everything else, vulva. Just to rope a question, it's confusing to me a little bit because the videos that you're showing me look like there was a continuous circular stretching along the same direction. So if you get past the vulva to the tear duct, do they not go backwards or do they continue around? They could. I mean, it'd be a shorter circuit this way. Yeah, you're right. Most of the time, the animal is just going around, around, around, around, around. But it's capable of going around and then doing that. The hermaphrodite's constantly running away, and the male's constantly pushing against it. That's how it sticks together. That's more than that, I don't know. What kind of natural environment does the hermaphrodite undulation of their little particles and sort of make them? What kind of a natural situation have they found? We, I believe that sea elegans can be reliably found on the surfaces of snails. And rotting fruit. And rotting fruit, yeah. And so, yeah, this gets. What are the two-dimensional surfaces? They're happy to crawl on two-dimensional surfaces. They can borrow through three-dimensional surfaces. They can swim in water, but they don't mate very well in water or at all. Yep. So here's another example of a neuron that I think connects with my notion of an internal model. PVV. This is a neuron, an interneuron, that is reliably activated every time the animal goes, flips around the head and flips around the tail. So it's a useful neuron if you know that your target is a long-skinny thing, because when you get to the head, you need to flip around and stay on. Otherwise, you lose it, and same for the tail. And at first I wasn't, honestly, I don't know much about, and I don't know a little bit more. But when Vlad started this project, I didn't know anything about males, and I didn't know to be excited about PVV, because I thought, well, maybe it's just a proprioceptive neuron or a mechanosensory neuron. Turns out it's an interneuron. And when I talked about it, when I mentioned this to Paul, who knows Sturmburg, who knows much more than I do, he was excited about it, because PVV is an interneuron, which is downstream from these ray neurons and these finger-like things. Actually, not ray 2B. Remember ray 2B? 2B is for the vulva. It's downstream of every other one, but ray 2B. And Paul, from some phenomenological observations and just intuition, had hypothesized that these neurons were fingers that sensed the taper of the animal. When you get to the head, it tapers. And when you detect taper, then you know you'd better turn soon, or shortly thereafter. You get to the head, same thing. Taper-sensing neurons would be useful for initiating subsequent turn by an interneuron like PVV. And in fact, these neurons are also co-active at the near the head and the tail. And in fact, just killing PVV does something that makes sense, actually. So an animal without PVV tends to, instead of reliably turning around near the head of the tail and flipping around, often overshoots and has a problem with the head-tailed transition. Actually, it was interesting when it overshoots. The error correction is characteristic as well. If it overshoots, the error correction is to, again, change the direction of motion. So that's another interesting aspect of this whole thing. There's not just circuitry to enact the behavior. There's circuitry to correct errors in an appropriate way. All of that's encoded in this little mating robot. OK, so again, we have similar vignettes for different aspects of the behavior, single neuron to characteristics of behavior. We have another approach to modeling is to try to take the whole population activity and do things like predictive of behavior. Again, so we've tried lots of different approaches in our lab. And actually, none of them worked anywhere near as well as what Andy Leifert does at Princeton, actually. So he was my former graduate student and he had a much better idea than we've been able to have so far in terms of modeling. So we just copied it, which is just build a sparse linear model that converts the neural activity into our finite, much lower dimensional behavioral traces. So people in the room know much more about sparse linear modeling than I do. But just sparsify a linear model. Impose sparsity with some tunable parameter and then figure out which neurons are most relevant to weighting particular behavioral outcomes. And what's nice about what we can do is because we have the names of all the neurons, we can actually integrate all the data sets. So you can use like six data sets to predict the seventh. And just like Andy has found in the hermaphrodite, this works better than PCA based models and so on and so forth. And we get a significant amount of the variance for behavioral output. There are errors in prediction. But for some of the velocity does very well. Curvature does pretty well. And I think r squared is 0.4 or something like that. It's not terrible. So this is kind of where we're trying to do. We're trying to figure out how do we understand this whole neural circuit on the fly or on the worm in a complete way. All right, so I should say a word about what we do. We have a functional dynamics. And so this is just an animation of neurons are active over time in one of the movies connected together by lines that are strengths of correlation. We can build a functional connectome with all the neurons. And to the right of it was what happened. This is the actual anatomical connectome. We have from the wiring diagram, we know who's connected to, we know who's coactive or who, we know who precedes who, and so on and so forth. And I'll tell you in the next slide, we'll just tell you something about despairing of modeling in terms of figuring out how to match a functional connectome to an anatomical connectome. The functional correlations between neurons is the top half of this. The synaptic connections is down here. And this does not look particularly mirror symmetric. It's not easy to see a great deal of homology between synaptic connectivity and functional connectivity, at least in this population-wide metric, which is all encompassing and rather blurry. And we can de-blur it by maybe separating chemical synapses from electrical synapses. So chemical synapses, you look at the pairwise correlations between all the neurons in terms of activity and the strengths of synapses in terms of numbers of synapses or something correlated with that. Virtually no correlation is the level of kinetics. Chemical synapses and the functional correlation of neuronal pairs, which is kind of shocking. But there is rather actually a strong correlation, well, I mean, a correlation, in electrical synapses, though. So there are both chemical synapses and electrical synapses in biology or in neuroscience. Electrical synapses are just little holes between neurons. And most of them we think of as an excitatory or positive correlations, activity from one leads into activity of the other. So positive correlation here is comforting. I should say that we probably should redo the connectome. The connectome that we have was done for the male tail, the data was collected 30 years ago. We're much better now at scoring these things than it might be worth revisiting. Some of this might be a simple error in flag and chemical synapses. There really doesn't have to be a strong correlation between the activity side and the mathematical side. So I'm not so surprised. No. Shape rate is a multiplexed idea. No, I'm not entirely surprised. And I'll say more about that in a second. Now, one thing we do know, all right, so one of the things that makes the chemical synapses much more complicated is there are many different types. Acetylcholine, glutamate, GABA, serotonin, dopamine, tyromine. So the worm has almost all the neurotransmitters or all the neurotransmitters of any vertebrate. But actually even here when we try to separate, and all those these patterns of expression are no. But actually even when we try to even be more accurate or just look at those pairwise correlations, nothing anywhere. Maybe some negative correlation in GABA, maybe very slight negative correlation in GABA or glutamate. But again, this complexity actually could be because there are a whole bunch of receptors. And these are the neurotransmitters, but they're both excitatory inhibitory receptors. But I think, though, by saying, why should you not necessarily expect all this coming? Probably, and this is just a speculation because I haven't had time to think about it, that a great deal of the correlation in the system-wide activity is not just because sensor neurons are connected to each other synaptically, but because they are connected to each other by the coherence of the environment. The hermaphrodite presents a coherent environment in terms of all of the sensory patterns of sensor stimulation. Those correlated patterns at the level of sensory inputs correlate the downstream activity in a way that isn't reflected in the wiring diagram, if that makes sense. Which suggests that the correlation structure of the brain, the true correlation structure that's behaviorally relevant in the brain, will only manifest in a freely moving animal doing its thing, at least in this context. And maybe that's also just because the way this thing is set up. This particular behavior is a shallow network with a whole bunch of sensory neurons, 36 or so more, I think, and just a couple of synapses downstream. So it's the correlations amongst this whole set of sensory neurons with respect to the hermaphrodite body, which are probably responsible for most of the correlations downstream. So that said, the electrical synapses probably play a strong role. I don't know if I left the slide, but remember the block-wise correlations, the searching phase, the scanning phase, the turning for all the detection. Those groups of neurons that came out of the functional diagram, functional connectome, are more prominently connected by gap junctions than any other. The chemical synapses go everywhere. Those blocks, though, could not easily be seen in the raw wiring diagram, but only in the electrical wiring diagram. So that is interesting. And the chemical synapses between those blocks might have something to do with lateral inhibition, be in this mode, but inhibit all the other modes. Or a sequence generation. Once you're here, trigger the next one, that kind of thing. And that said, just like you saw, there's a great deal of information, though, also in the microscopic details of individual neurons within those blocks that do very specific things. So this is kind of where we are. I think we're in a good place. It took those seven functional pole-brain imaging across many animals. It was hard to find data. The first one that Vlad did, he did it by literally going through every movie. Neuron by neuron, 10 hertz, 30 volumes, whatever, volumes. And it was painstaking. And we got faster and faster and faster. Now we use computer-assisted methods, so on, to do this. One thing that we were very slow at still is identifying the neurons, right? So where you just take your 3D constellation and then try to map it to a static constellation. Atlas, just like looking at the stars and figuring out where who's who. That was very slow. But actually, this summer, I think, that we're now able to get past that as well. I think we're at automatic. We are at automatic cell identification. Not perfect, but proofreadable in a rapid way. And that's due to a collaboration with Columbia. Paul collaborates with the sound of mail. May collaborates with the sound of every day. And Oliver Holbert collaborated with us to build this animal, which is a rainbow animal. This paper's actually already on bio-archive. So he made an animal with five different colors, deterministically located in the individual neurons of the C. elegans. So on our end, we equipped our microscope to simultaneously take five colors at a time. So we don't just take red and green, but we take all the other colors. And by referencing the deterministic color of each neuron, we can put a name on it. Liam Paninsky at Columbia has built just a simple machine, well, simple, a CNN that automatically puts the labels on any volume. So everything should get faster. We should be able to get these whole brain images as much more rapidly. And now we're in a position to look at animals missing particular neurons, animals with different mutations. There's lots of interesting phenomenology of males mating with different hermaphrodites. Actually, Paul has done a lot of this. I think we haven't done it yet, but Paul was telling me that males have trouble mating with hermaphrodites that have been modulated in some interesting way. You make dumpy hermaphrodites, which are too short. You can make long hermaphrodites, which are too long. You can make hermaphrodites that don't have vulva. You can make hermaphrodites that have too many vulva. And the male has trouble with all of them, but interesting different problems with all of them. And I would love to test those and to understand how the strategies might change and adapt. One, I think to me, a fascinating phenomena that Paul was telling me about was that when a male tries to mate with a hermaphrodite that lacks a vulva, it will try very, very hard going around and around and around and around to look for the vulva. And eventually it'll fail, of course it has to fail, but eventually it switches and just stops and just tries to stick his spicule anywhere. And that's fascinating to me. It's sort of given up, but not given up entirely. It will just do the next best thing, which is just randomly guess. And the behavioral state transition in the mind of the male that coincides with that will be fascinating to understand. There's other phenomenology we have as well. I think males get better and better at mating with hermaphrodites. The first time a male might mate with hermaphrodite of fumbles, but after a while it gets, the time it takes to actually do it, it gets smaller and smaller. Actually there's dopamine. There's dopaminergic neurons in the male tail. So maybe there's some sort of reward circuit. So there's some sort of learning, right? I think most of this is innate, obviously it's innate. But you could imagine there's, it might be some useful learning that might happen when the male is trying to adjust its mating strategy to the particular environment, right? Two dimensional auger pads, right? Do things in a particular way and get good at it. I don't know. And there's a whole bunch of other stuff to do too. Males with different species hermaphrodites, so on and so forth. What if you have two males that are combined for the male, what happens then? One will win. Males aren't interacting particularly. Males aren't interested in one another, no. No, males don't do that. Well, again, if there are no hermaphrodites, actually another phenomena, there are no hermaphrodites available, males will eventually just start trying to mate with anything that's long and skinny, right? Its first goal is to presumably match the whole template. Taste, smell, pheromone, touch. But when variables are eliminated and they continually fail, the requirement of matching the whole template is relaxed and then just portions of the template are required. Yeah, so that's an interesting thing too, right? The male will do that. There's just different states, there are different requirements. So this is a sort of a taste of the whole thing. The data sets are unbelievably rich and there's all sorts of information in them. And we're doing our best to extract what we can. But from individual neuron coding to behavior, or population level coding to behavior, I think that's everything you want to know in neuroscience at some level is represented by this model. So we're excited about it.