 And what I want to do in this talk is to approach this question... Well, let me backtrack a little bit. The idea, of course, ultimately is to make the connectome work in a functional sense. So people imagine... And there are lots of skeptics, of course. People imagine, though, that we might be able to put the connectome into a computer model, put all the neurons that we've mapped into a computer model, put all the connections between them into the model, and just run the model. And it will produce accurate behavior, functional behavior. So what I want to do in this talk is to approach that same question, which actually we've been working on in simple systems, mostly in verbat systems, for many, many years, in a somewhat historical way, following up a romance introduction, and ask the question whether the connectome is really sufficient to do the job. Okay, we've got slides finally. All right, good. So this is a slide that you don't need to see. So those of us working on simple systems, invertebrate systems, as I said, have faced this question already, and there are, as Roman pointed out in his introduction, known connectomes for some simple networks in invertebrates, such as C. elegans and the stomatogaster ganglion of lobsters and crabs, due mainly to the work of Eve Bodder and her colleagues and intellectual descendants. So we can actually, and this has been done already, ask the question, if we put these networks into a computer, in particular the stomatogaster ganglion network, which arguably is the most complete connectome, if we put that into a computer and run it, will it produce the pattern of behavior that the real system produces? And the answer is no. And in this talk, I want to highlight some of the pieces that are still missing from the connectome. Basically I take the view that the connectome is the basis for performing that, carrying out that project, but it's a necessary component, a necessary basis, but it's not as sufficient. It's not quite sufficient to accomplish the task, and I want to highlight some of the pieces that still need to be added to the connectome to make it possible to perform that task. So here we have a basic wiring diagram of connectome, and I want to talk a little bit about reconfiguration by synaptic plasticity, which is well known in the stomatogaster ganglion. I want to talk about equivalent connectivity patterns, and in particular I want to focus on neuromodulation. So reconfiguration by synaptic plasticity enables the connectome to be divided up into separate networks or conversely separate networks to be joined together, and this can be done on a dynamical basis from one motor act to the next in a rhythmic behavior, for instance, such that the network that's carrying out each act is different. I want to show you some examples of equivalent connectivity patterns. Some of this work actually, again, comes from the lab of E-model originally, originally by Astrid Prince, and then more recently by others in E-model's lab. So here we have a computer model, actually, which shows the same two computer models, two connectomes, which show the same pattern of activity, yet they have completely different connectivity patterns, and they have different synaptic strengths between the different neurons, and they also have different intrinsic properties, as shown by these conductances of the different ion channels in the neurons. So we have different connectomes performed in the same function. Conversely, we can also show that the same connectome performs very different patterns of activity, for instance, in different conditions of synaptic plasticity and neuromodulation. So the connectome is a very plastic thing. It's not a unique thing. This down here shows some experimental evidence. As I said, this was a model, but here is some experimental evidence for the same thing using dynamic clamp, where the connectome, the wiring diagram, was perturbed through dynamic clamp approaches to change the synaptic strengths, and here we have four different effectively connectomes with different synaptic strengths that produce the same pattern of activity. So this is an experimental demonstration that Astrid showed in the model here. Now really I want to focus, as I said, on neuromodulation in this talk, and here is the picture that Roman already showed of the somatic astraganglion, showing all the different modulators, all the different transmitters, hormones that impinge on this network, this basic connectome, and changes patterns of activity. Again, in C. elegans, the other animal that I've alluded to already that where the connectome is known, we have numerous neuropeptide genes, modulator genes, coding many, many different neuropeptide modulators. It's ironic that in the simplest organisms, such as crab and lobster, which is not that simple, but C. elegans certainly is a simple organism. Where the connectome is relatively simple, we have the greatest, possibly the greatest richness of neuromodulators that then change the way the connectome produces functional activity. And they may be, in my view, it's still an open question whether invertebrates and vertebrates solve the same problems in different ways or the same way. And this question has been around for decades also. But it's certainly true that this richness of neuromodulation is not unique to invertebrates. We have here the respiratory network in the work of Nino Ramirez, again showing different neuromodulatory inputs onto this basic network. Here we have an example from the work of Microsomo, of neuromodulation impinging on pyramidal cell output, changing the output in the brain, of vertebrate brain, and shows also that neuromodulators are one of the ways in which synaptic plasticity can happen. They mediate, they modulate, but they even mediate synaptic plasticity, both presynaptically and post-synaptically in the vertebrate brain. So this is not... These examples that I'm showing you from the invertebrates really will probably translate to vertebrates, although that's still an open question to what extent. The importance is the same. Neuromodulators can, as I already mentioned, change the activity of the network. Here we have somatogastric ganglion output under the conditions of different modulators being added. And you can see the same network is producing completely different arrhythmic patterns of activity in different conditions of modulation. And this is a very common observation in invertebrates and presumably in vertebrates as well. Finally, I want to stress communication by neuromodulators. This is an interesting point where neuromodulators can link neurons that are not actually linked by the basic connectome at all. If the neuromodulators are released from one neuron, let's say this neuron here, and travel through the interneuronal medium to impinge on, to act on another neuron that has receptors for that neuromodulator, then it establishes a link, a communication link between those two neurons that is not represented in the original static wiring dagger. And so then it becomes a question of how far these modulators can travel, and the point has been made that a lot of neuromodulators, especially neuropeptides, travel through volume transmission rather than point-to-point connections between well-structured synapses, as in the traditional connectome, how far the neuromodulator can travel and where the receptors for that neuromodulator are and different. So we have patterns of communication established in the neural network that is through these biochemical pathways that are really unsuspected if you just look at the static connectome. I've put this in for Paul Katz's sake, showing that making the point that extrinsic modulation and intrinsic modulation are two kinds of modulation that can occur. Extrinsic just modulation is where the modulator comes from outside the network and can affect the network from outside. Here we have an intrinsic modulation where the actual elements of the network release the neuromodulator when they're active. And in both of these cases this kind of communication can be established. It then becomes important exactly what the extracellular matrix looks like in the brain. Obviously, it's going to be complicated by this dense packing of neurons, glial cells, blood vessels, what have you. And so the neuroanatomy that's, again, independent of the connectome is going to have great effect on what kind of communication through this biochemical network will occur. And that's probably responsible for findings such as this in the somatogaster ganglion where if you just apply exogenous modulators to the different neurons of the somatogaster ganglion you find that they have responses to some modulators but not others in different combinations here. But if you actually release the modulators and urgently from the neuron that normally releases them you find a different pattern of responses to those neuromodulators. So this is probably due to neuroanatomical constraints of this sort, even in the crustaceans tomatogaster ganglion. Here we have an example of computation carried out, very simple computation in Italy but computation carried out by these biochemical networks. In this case this is from the work of Nick Dale where he studied swimming in the Xenopus embryo. Swimming is controlled, the duration of swimming rather is controlled by the bursting, the length of the bursts of this neuron as shown here. This is where swimming starts, this is where swimming stops and the length of this burst is controlled by a complicated, somewhat complicated array of extracellular biochemical events where ATP is released from the neuron. So this is an intrinsic modulator of a sort. It's then degraded to ADP, AMB and adenosine. Each of these, in particular ATP itself and adenosine has an effect on potassium and calcium channels and there is some kind of structure to this network where feed forward inhibition controls the duration of the burst. So if you modulate, this is again a computer model now but this is a real data here, if you modulate the strength of this inhibition you change the duration of the burst. So there is memory and there is computation that's performed by this extracellular network and there are much more complicated examples that probably can be cited. Finally I want to show you the system that I think Romain was expecting me to talk about most of the time in my talk. The plesia feeding system, here's an aplasia feeding on seaweed. We've studied and others have studied for many, many years the musculature, the buccal musculature that performs these feeding movements. There are many muscles and many neurons and each of these neurons has a conventional transmitter but also has many intrinsic modulators. So just showing in this column here. Here is one muscle that we've highlighted, one muscle that we've worked on, the ARC muscle controlled by two neurons B15 and B16 and here is the network of actions of these modulators and superimposed on the sort of traditional connectome. The traditional connectome would be just these two neurons B15 and B16 both of which release astal choline and make the muscle contract. So the muscle here is taken as the output element of this network and it could have been neuron as well, in fact a lot of very similar things occur in the central nervous system of plesia as well. This muscle just happens to be a convenient output element. So the connectome, the traditional connectome B15 and B16 releasing astal choline on the muscle, very, very simple. Yet there is also this very complicated network of effects of these modulatory co-transmitters, in this case just two of them are shown as CPMI modulin through different second messenger systems, effects on ion channels, change the relaxation rate of the contraction size of these contractions such that each modulator has a different effect on the shape of the contraction. So if you take that as the output element of the network, you can see we produce very different patterns of contraction and patterns of activity in the nervous system. It would be depending on which modulator is released and also what other modulators are being co-released because if we then combine the modulators, SCPMI modulin in this case, in all possible ways, we can see how that space of modulator inputs projects into the output space of the relaxation rate and contraction size effect. And we can see it covers a two-dimensional space in which many, many points can be reached by combinations of modulators but not the individual modulators alone. And here some relaxation phase is shown from different muscles under different conditions of SCPMI modulation. So we have a... This is still a fairly linear network, actually, but we already have the idea that multiple modulators jointly will change the network activity and the output of the network in ways that individual modulators could not. So we have combinations of modulators that become important. And presumably, and certainly that's true in the plesia, but presumably in most of the other systems, such as Smartagaster, Ganglion, and vertebrate systems, modulators will be released as cocktails. There's going to be cocktail on your modulators, not just one modulator at a time. And it's that cocktail that's going to affect, it's going to determine the final activity. And finally, I want to make the point that the... I haven't mentioned dynamics yet, so I want to mention dynamics. These effects in a plesia and elsewhere, in other systems, have powerful dynamics which differ from each other for the different effects. So some are slow, some are fast, and in particular, this creates a history dependence for the system. So these are some of the elements in that network that you just saw up here in the CPG and here in the muscle. And the size of these arrows indicates the correlation from one cycle to the next. This is a rhythmic cyclical behavior, so we can measure these effects for many, many cycles. Actually, what was done in this model was to feed in real feeding behavior, which is very variable, over a couple of hours. So hundreds of cycles, and these are just the correlations between one cycle and the next. And you can see that the CPG has some effect on the current cycle, in particular on the fast effect, the potassium current effect, but much more powerful are the effects of the previous cycle on the current cycle. So we have carryover of history, carryover of the activity, depending on what the nervous system has done before. And this is something that, again, the static connector on the traditional connector does not represent very well. I want to put in a plug for some recent work that we've been doing. I've already mentioned that combinations of neuromodulators are going to be important, and very often the combination of the neuromodulators has unexpected effects on the network. As I said, that one I showed you was relatively linear, but others are not, where the combination of modulators suddenly produces a bifurcation in the behavior of the system. One modulator alone, or the other modulator alone, let's say that just two modulators, for the sake of argument, will produce rather expected patterns of activity, but the two together will produce some completely radically different pattern of activity. So how do we study this experimentally? So traditionally the experiment is to take the network and apply the modulator exogenously. And that works up to a point. What if you have a cocktail of modulators, and you want to understand the relationship between the modulators in that cocktail with respect to the activity, the output of the network? So you can try them pairwise or tripletwise, but nobody's actually done that very systematically because there's an infinite number of experiments, an astronomical number of experiments you have to do to actually test all possible pairs of all possible triplets. So what we've done is develop a method that uses basically a block experimental design approach where we can put on cocktails of modulators where this is showing it for some hypothetical number of modulators at different concentrations, where in each cocktail sample the modulator of interest, the pair of modulators of interest or the triplet of modulator of interest is combined with all others over the whole set of cocktails. So basically we have a block experimental design and this radically reduces the number of experiments that we actually need to carry out, as shown here at the bottom, to get at least some initial screen of which neuromodulators, which neuromodulator combinations are going to produce non-linear effects that perhaps lead to radically different changes in the behavior of the system. So I want to just summarize what I've said. I really want to make the point that the Connectome, as this work in invertebrate so-called simple systems has shown, the Connectome is necessary but not sufficient to understand how the brain works. So yes, of course it's necessary. We will be much further and we already are much further in the invertebrate systems when we have some kind of Connectome to work with. But it's not enough. We can't just plug the Connectome into a computer and run the computer and it produces behavior and we're done. We understand how the brain works. That's not going to turn out to be the case. I think invertebrates, as it has not turned out to be the case, in invertebrates. I kind of think of this as analogous to the Human Genome Project, which I think has been mentioned already, where again the idea was if we understand the whole genome, if we have the whole genome mapped and we can just to read off all the instructions for making the organism. And that will be very simple. And the Human Genome and genomes of other species have turned out to be fundamental for advancing science, but not in that way yet. Because between the genome and the phenotype, there is a very complicated network of cellular interactions that mediate the translation from the genome to the phenotype. So I think it's going to turn out to be very similar for the Connectome Project, where we need to understand the Connectome, we need to have it in front of us, and the Connectome, in a way that I've tried to show you in the invertebrate systems, actually suggests what else we need to know and incorporate into the computer models in order to make the thing work. Neuromodulation, synaptic plasticity, dynamics, and so on. So there is a complicated network between the two levels and we need to understand that network, at least build that network into our models in order to make the whole thing work. And I've talked about synaptic plasticity a little bit, not very much, probably in connective patterns, but mostly neuromodulation. So, and really the most interesting idea there is neuromodulation. Neuromodulation could certainly change these two things and others in the network, but most interesting idea to my mind is that there is communication by neuromodulators that is independent, superimposed on the Connectome, the static wiring diagram, where the neuromodulators establish channels of communication between the neurons that are not in the static Connectome. And it's these, both in terms of structure, neuro-atomical structure, location, and dynamics, these channels are delocalized from the Connectome and they alone, of course in conjunction with Connectome, but even alone, they can store memories, process information, and perform computations. So I want to thank some of my past collaborators and present collaborators, and thank NIH and INDS NSF for funding this work. Thank you. Hello. Hi. I was wondering if it would be possible somehow to have a model of the computation of the extracellular network by, I don't know, maybe taking into account the association constants or speed of reactions between different neuromodulators or how this can be, how this can have a model computationally. Right. So yes, that would be the idea. So the failure of these simple point-to-point Connectomes suggests that you need something like that, and you need to incorporate that into your model to make it work. And of course, biochemical models like that exist. The problem with that biochemical model is that it's three-dimensional and delocalized in space and time. It's not a simple connection from A to B that you can basically just draw an arrow in the graph. You need to know a lot more about where the neuromodulator goes, the association constants, the effects of the neuromodulator on other components in the network, and so on. So it becomes a much more complicated problem to do that, but in principle that is perfectly doable, and it should be done. Okay. Thank you. Very interesting talk. I think it made a very convincing case that simple structural connection is not enough, and we need to know also the dynamics, neuromodulation, and things like that. But I think it's interesting to consider, even if we know all of this, will that be enough? Because the interesting case to consider is the lesson we learned from studying the artificial neural network. In that case, we know structure, we know dynamics, we know pretty much everything. But still, in many cases, especially the early multilayer perceptron, it's still black box, it's still very hard to explain, detail how does the network work. Yeah. I agree entirely. There are other... Well, it depends what you mean by explain. Explain in terms of a computer model has its own problems, obviously. Computer models tend to be rather brittle. Probably related to some of the things I talked about, they work fine for your data, but then they break down if you try to model other situations. So that's a limitation of computer modeling and understanding also. But beyond that, yes. The things I talked about are not, you know, be a whole and end all. There are many more things that I think we will need to know, just don't know what they are yet, as Donald Rundfeld said. Yeah, thank you. I just think probably it's also important to consider what type of question that we can ask, or as you mentioned, what is understanding that... what do we mean if we try to understand the network, right? Right. So there's, again, the argument between computer modeling, kind of brute-force computer modeling. Yes, we can reproduce the pattern, but do we really understand it and understanding? Which usually, for most people, that means reducing it to some simpler conceptual, perhaps, or computer model. And both are valid. I tend to be on the simpler side rather than the complicated side because for one thing, we haven't been able to make the complicated models work. But even if we do, I'm not entirely sure the understanding will emerge suddenly from the model network. Yeah, okay, thank you. Okay. Thank you very much. Our next speaker is...