Next, we have Dr. Butch speaking to us about increased dynamic modularity of the frontotemporal limbic network precedes enhanced task performance. Thank you all. Thank you to the speaker so far, and thank you for giving me an opportunity to present some of my work as well. Next slide, please. Okay, so task performance really depends on adaptability, and this is for two reasons. Adaptability enables the sort of optimization of performance based on a changing task environment, and it also allows you to iterate your performance based on feedback. And so one of the keys is that the neural networks that underlie cognitive performance must also enable cognitive adaptability. So how can we frame that in a way that we can study? Well, we can study that with respect to neural network adaptability, and in particular, community structure, which is a graph metric, is thought to provide a network substrate for adaptability. And in community structure, you have groups of nodes, which are known as modules, that are more densely connected to themselves than to the rest of the network. And the degree to which this occurs in a system is quantified by the metric modularity or Q. And the modularity or community structure of these rapidly fluctuating networks during online cognitive performance is largely unknown. One thing that we, or several things that we do know about community structure is that it does characterize many real-world observations, such as the governmental budget organization, U.S. gas transportation pipelines. And it's important in long-term cognitive development and function. My co-PI, Danny Bassett, out of her lab has shown several things with fMRI data, one being that as you increase modular segregation over time and over development, you also have a correlated increase in executive function. These are linked. And also that when somebody is starting to acquire a new skill over days or weeks, the degree to which their networks show modular segregation before they start can actually predict how well they're going to learn that task. So the questions I have, though, is, does modularity or community structure fluctuate on a trial-by-trial basis, though that's currently not known? And is there a distinction between community structure based on how you've framed the network, whether it's high-frequency or low-frequency connectivity? And finally, can this community structure antecedently correlate with behavioral performance? So obviously with what we are allowed to access with our epilepsy patients gives us insight into this that we can't get with fMRI. So we have just started with a small subset of patients, 10 patients, undergoing S-stereo EEG, as well as one non-human primate that are both doing the same task, which I'll go over in just a minute. But the network of interest is really the global frontotemporal limbic network. And there's two bands of interest. One is a low-frequency narrowband and a high-frequency broadband. And the task itself is a type of temporal expectancy task. And what's nice about temporal expectancy is that it's something that many species can perform. And so the networks that underlie it are probably rudimentary networks that are involved. So the task itself is, like I said, very simple. There's a presentation stimulus that shows up, which is a white box, that then changes after a certain amount of time to a yellow box, which is a response stimulus, after which point the subjects press a button as fast as they can, or the primate moves a joystick as fast as he can. But I'm actually not interested in what's going on during the trial. I'm actually interested in what's going on in the networks before each trial starts. And in particular, is there a way that the network architecture might be organized that is optimal for the subsequent task performance? So we analyzed this task in three different ways. One was looking at all trials in that pre-Q period. Then we broke down the trials in the fastest one-third versus the slowest one-third of trials and subtracted the network metrics during those periods. And then the same thing is in internal control. We looked at the latest versus the earliest trials. And in terms of the analysis itself, the task performance was just reaction time, like I mentioned. And we used phase locking to create our network constructs and then do our graph analysis on it. Behaviorally, so behavioral results were variable. So we do have some subjects that tend to improve a little bit over time. We have others that are pretty consistently good throughout time. And then we have others that are actively trying to sabotage our data. But it's important because overall, when you look across all subjects, if you divide all the fastest versus all the slowest trials into two groups, you can see there's a significant difference between them across all subjects. However, latest versus earliest does not have that same difference and does not have any difference. So what that means is that this task is really a measure of online cognitive performance, not learning, which will be important later. So I'm going to start with just a series of images. We'll quantify this later. But just I want everybody to be able to see visually how these things are different. So starting with the low-frequency narrowband transformed network across all trials, you can see this is just two subject examples I'm showing. You can see that there is some structure to this network connectivity in these two subjects, and they're relatively similar between the two. But if you compare that to their networks during the fast versus slow comparison, you can see that there's a lot more intricate dynamics that are going on in the networks. And if you look at the graph visualization of these two in the all compared to the fast minus slow condition, you can see that the same nodes are interacting in very, very different ways. And when you look at the actual community structure within these networks, you can see that basically structure itself will be indicated by the 3D organization of all the nodes and their connections. And the strength is indicated by the weight of the color between each nodal connection. And if you look at all compared to the fast minus slow condition, you can see that the modules themselves are, first of all, much more intricate in their architecture and structure. And then on top of that, the weights of the connections between the nodes within the modules are also far greater in the fast minus slow condition compared to overall trials. So what that would perhaps mean is, as we talked about earlier, modularity is sort of the quantification of how densely connected these nodes are to each other compared to everything else around it. And you would suspect that the fast minus slow condition would have a higher modularity compared to the all condition. Again, these are the same nodes, the same patient, same everything. But just the way that they're interacting with each other is very different. And when we do this across all subjects, we do see that the fast minus slow condition does have a significantly higher modularity in the pre-Q period before better versus worse task performance. And as I mentioned earlier, the late minus early condition is important as a control because we didn't see a behavioral effect there. And again, in modularity, it doesn't really look much different than just overall trials in the low frequency band. And then in terms of the frequency dependency, we can see that the high frequency band actually doesn't really have a lot of community structure that's apparent in that architecture in any of the conditions. And interestingly, our non-human primate, Logan, also showed us the same pattern basically in his modularity. So in conclusion, we do see that community structure is actually a dynamic property of functional network architecture. And it fluctuates in very rapid fashion on a trial-by-trial basis. And we do see that there is a frequency dependency to that. And most importantly, we see that really that increased low frequency narrow band modularity in the pre-Q period is characterizing fast versus slow trials. And really that's sort of the first evidence we have that's linking antecedent connectomic fluctuations to enhanced online cognitive performance. So our goal from here now is to expand this, obviously more subjects, and try to work towards developing this technology in a way that can become closed loop, and in a way that we can affect these network metrics with stimulation to see how stimulation could be used to create these types of network fluctuations. And with that, I'd like to thank, it takes a village, obviously, so I'd like to thank all the members of Dr. Lucas's lab and Dr. Bassett's lab, who are my co-PIs, and then obviously our neurosurgery department and the AANS for allowing me to be here.

adaptability

neural networks

cognitive adaptability

community structure

frontotemporal limbic network