How the Brain Regulates Aggressive Behavior with Ann Kennedy, PhD

 Erin Spain, MS [00:00:09] This is Breakthroughs. It's a podcast from Northwestern University Feinberg School of Medicine. I'm Erin Spain, host of the show. Today's guest is the winner of the 2022 app Eppendorf and Science Prize for Neurobiology. Ann Kennedy is an assistant professor of Neuroscience at Feinberg and the 21st recipient of this international prize, which is awarded jointly by the life science company Eppendorf and the journal Science. She joins us today to talk about this honor and her work as a theoretical neuroscientist investigating neural computation and the structure of behavior such as her recent research in the area of aggression and how it's regulated in the brains of animals. Welcome to the show, Dr. Kennedy.   

Ann Kennedy, PhD [00:00:56] Thank you. Thank you so much for having me.  

Erin Spain, MS [00:00:57] Explain the practice of theoretical neuroscience and what drew you into this field.   

Ann Kennedy, PhD [00:01:03] So theoretical neuroscience, the name comes from the analogy to physics, where you have theoretical physicists and experimental physicists. So some people say theoretical neuroscience, some people say computational neuroscience. The basic idea is that rather than doing wet lab experiments, we're building computational models of neurons or networks of neurons or synapses, and we're using mathematical models to try to gain insight into how the brain works. My background in undergraduate was in engineering, and I worked a lot with methods in computational biology and learned about mathematical models of the circulatory system, the nervous system, networks of gene regulation. And I've always had a love of neuroscience and just thinking of it as a thing that you could approach from a mathematical perspective. And I think that you could model, if you could just describe what all of the neurons are doing and how they're interacting with each other. That really appealed to me.  

Erin Spain, MS [00:01:55] You came to Feinberg in 2020 and set up your lab here. Tell me about your lab. The vision for your lab is to understand how behavior shapes and is shaped by the brain. Tell me about some of the theories, models, and methods used in your lab.   

Ann Kennedy, PhD [00:02:11] My lab kind of grew out of my postdoctoral work, where I was collaborating with experimentalists who are studying these regions deep inside the brain that are involved in the control of survival behaviors, things like aggression and reproduction and fear. There are theories about control of survival behavior, but they tend to be very old. They're things that have come out of ethology in classical studies of behaving animals. And as a theorist, what I wanted to do was take these classical models of how behavior might be structured, how animals might escalate an aggressive encounter or optimally forage for resources, and see how those normative models of behavior relate to what we actually see when we record from the brains of behaving animals. So with modern neuroscience methods, we can characterize the structure of animals movements in much more detail than we could before. We can use machine learning tools to estimate the postures of animals in two or three dimensions and identify a recurring movements that those animals are performing. And then we can relate that detailed characterization of animals' moment-to-moment actions with recorded neural activity of populations of cells in different brain areas that we think are involved in the regulation of these motivational states. So my lab is aimed at developing the methods that will help us to do this, to relate neural activity to behavior in an animal that's just running around wildly interacting with its environment, and also relates this data analysis side to the normative modeling side to classical theories of how animals make decisions about their behaviors and see if those classical models actually hold up in the face of real data.  

Erin Spain, MS [00:03:45] Use of artificial intelligence and machine learning in your work is really exciting because this field is absolutely exploding right now. Can you just talk to me a little bit about that and sort of what's on the horizon in the area of artificial intelligence and your lab?  

Ann Kennedy, PhD [00:04:00] It's a big part of the lab and it's something that we're very excited about. One of the directions that we're trying to push is rather than just trying to reproduce the human labels of what an animal is doing to kind of let the animal tell us itself what it's doing. So look at how the animal's moving and try to identify the repeated motions and patterns of movement that the animal produces and detect those rather than relying on what a human recognizes as behavior. So this is called unsupervised machine learning. And we're doing some work in this area to try to build very detailed phenotypic profiles of animals so that we can then examine how their behavior and how the structure of that behavior changes under different experimental settings. We're applying this right now to a Parkinson's model from the Laboratory of Jim Surmeier. These animals have a progressive loss of dopamine neurons and dopamine production that extends over the course of a month or two. So with our automated phenotyping, we can characterize what's changing in how these animals move and how they behave over these longer time ranges of experiments without needing to manually score hundreds of hours of behavioral data. We're also working to build better standards for how these kinds of methods are evaluated and how they're distributed in the neuroscience community. Right now, we're assembling what we call the Multi-agent Behavior challenge, which is we're creating this giant benchmark data set of social interactions from where we've reached out to about two dozen different labs and asked them to share their annotated videos of mouse behavior with us. And we're planning to assemble kind of a cross lab data set and a challenge to the machine learning community to see if they can develop methods that can recognize the same behaviors in arbitrary experimental settings. And we're planning on releasing this data set sometime in the coming year, and it will be accompanied by a Kaggle competition, will run a one or two month contest on Kaggle. It's kind of like the Netflix challenge. It's a site where anybody who's studying machine learning can kind of try to solve a problem and make submissions that'll be accompanied by a $50,000 prize pool for the teams that are able to do the best job of recognizing all of these different behaviors that experimenters have annotated.   

Erin Spain, MS [00:06:05] That's very exciting and innovative. What's been the response like from folks, your peers, who know what you're doing?   

Ann Kennedy, PhD [00:06:12] It's been really nice so far. A lot of people are really supportive of this idea, and I think they recognize that sharing data and establishing benchmark data sets is really important. I think that's one of the things that really drove progress in machine learning over the past decade. I mean, a lot of things have, but availability of large data sets that people could use and standardized evaluation metrics so that you could really look on the same data sets by the same metrics, what methods are working and what methods aren't. That was critical to early progress in machine learning, and it's something that I think neuroscientists appreciate. But most labs are working on these in-house experiments and these very small niche problems. You don't want to standardize everything across labs. So we're hoping that rather than forcing everybody to run the same experiment, this challenge will motivate people to develop methods that are more robust to changes in experimental setting in the first place. And so far the support has been really fantastic.  

Erin Spain, MS [00:07:08] I want to shift gears and talk about this prize that you recently won, the 2022 Eppendorf and Science Prize for Neurobiology. So you wrote an essay. The topic of the essay is about aggression and specifically your research on how aggression is regulated and ramps up in the brains of animals. Share the details of that work with me and what you wrote about in this essay.  

Ann Kennedy, PhD [00:07:32] So I guess this goes all the way back to the start of my postdoc. When I joined my postdoctoral lab in 2014, they had shown that you could optogenetically so experimentally activate a set of neurons and hypothalamus, and it would drive this time resolved aggression behavior. When you activate these neurons, the mouse runs over and attacks whatever mouse is in the home cage with it and fights with it. And when you turn off the activation, they stop fighting and go back about their business. So when I joined the lab, we said, Oh, we're going to image these neurons and identify the neurons that are activating to drive fighting. And so we did the experiment to do this imaging, which is we used a head mounted micro endoscope, which is this two gram microscope that the mouse wears on its head, attached to a green lens going into the base of the brain and imaged in this room neurons that were being activated. And surprisingly, we didn't really find any attack tuned neurons. We saw a lot of neurons that just kind of activated whenever another mouse was around. But we didn't see attack cells. And even stranger, when we did the same experiment in a neighboring brain region involved in control of mating behavior, we found a bunch of attack cells there alongside mating cells and sniffing cells and other behavior specific cells. So we had this weird disconnect between we know this area is involved in control of aggression. Its activation drives aggressive behavior, but you don't see attack tuned cells. So in the work I describe in the essay, I talk about our efforts to figure out why there's this disconnect between optogenetics and imaging. And what we ended up finding is that there aren't attack specific neurons. But if you look at the level of the neural population, if you look at a couple hundred neurons at the same time, you can extract out this one-dimensional signal from that neural population that is gradually ramping up over the course of an aggressive encounter where the level of activity along this kind of dimension is predictive of the animal's behavior. So when there's a little bit of activity, you see investigation a little bit more, you start to see sort of low level aggressive behaviors. And when there's a lot of activity, you see outright attack. So it's not something you can detect when you go one neuron at a time. But at the level of the population, we find this ramping signal that's present that is correlated with basically how angry the mouse is. And when we looked across a full set of like 14 different mice that we've done this imaging and we found that different mice have different baseline levels of aggression, some of them really just want to run in and fight whatever mouse is around and others are a little bit more chill. And the time constant of this ramp, the amount to which activity ramped up and the extent to which it was persistent after it had ramped up was really tightly correlated with how aggressive the mice are. So mice that were really determined, persistent fighters, you'd get a quick ramp up of activity and it would stay persistently elevated even for like a minute or so after you remove the other mouse from the cage versus the most that didn't really have a motivation to fight. You get little blips of activity along this dimension, but then it would decay away quickly. So it seems like you don't have attack neurons that turn on and off a decision to attack. But among the population as a whole, you have the signal you can read out that kind of encodes how aggressive the animal is and tuning the time constant of this ramping tuning, how persistent it is tunes, how likely the animal is to escalate aggression to these more intense forms of fighting.  

Erin Spain, MS [00:10:52] Fascinating work. And you've said that you're interested in finding out whether these ramping dynamics exist in other areas of the brain and associated with other behaviors. Can you tell me about that?   

Ann Kennedy, PhD [00:11:03] So another project I worked on as a postdoc was looking at a neighboring part of hypothalamus. So this is, I should say, all in hypothalamus. The ventral lateral portion of ventromedial hypothalamus,rolls right off the tongue, and there is a dorsal medial portion of the same nucleus that doesn't care about aggression behaviors. It cares about fear and defensive behaviors. And we know that this region has the same sort of pseudo architectural properties. It's a network of excitatory neurons. They're interconnected. And if you present a fearful stimulus to the mouse, it shows persistent activity the same way that the aggression neurons do. So we think that this idea of a population code where you have persistent and scalable neural activity is something that's not just going to be seen in aggression, but is going to be seen in all sorts of motivational states of animals that have a created level of intensity. So we've looked at this in the context of defensive behaviors in this VMHDN image region. I'm working with collaborators now to see if we see similar things in the neural encoding of pain-related cues and hunger-related cues in other subcortical nuclei. And our suspicion is this is going to be a general coding property of some nuclei of the brain, particularly nuclei where you have a bunch of glutamatergic cells that are recurrently connected with each other, which is a good basis for having reverberatory ongoing patterns of activation in the neurons.  

Erin Spain, MS [00:12:24] And again, this work is done in animals, but it may give us insight into human behavior as well. Is that right?  

Ann Kennedy, PhD [00:12:30] Yeah. These brain regions, these are all nuclei of the hypothalamus. They're present in humans the same way they're present in mice. There have been patients where they've stimulated in these regions and seen the same sort of emotions elicited that you see in the animals, seen, people who have feelings of rage or anxiety or fear when these regions are surgically stimulated. So they're present. They're under a lot more, I think, top down control in a human than in a mouse. If you're angry, you're able to kind of clamp that down and keep control of your actions. But it's definitely a part of our brain and it's an evolutionarily ancient thing that we really need to structure our behavior in an adaptive way.  

Erin Spain, MS [00:13:07] And not only is it fascinating to learn about this and this ramping up, but there could also be some application to possibly neurological diseases and disorders where aggression is a symptom of one of these diseases or disorders. Do you think there could be a way to look at maybe a therapeutic approach in some distant future?  

Ann Kennedy, PhD [00:13:26] Well, I think it does point to the fact that just because people feel aggression doesn't mean that there are aggression neurons that are on when you're angry and off, when you're not. It’s a more subtle population level signal, which might change the way that you think about trying to target it in a medical treatment. Also say that this kind of scalable encoding a motivational state we don't think is just associated with aggression. It's also involved in things like anxiety and appetite and other poor motivational drives in people. And there are plenty of disorders of these motivational drives present in the patient population.  

Erin Spain, MS [00:14:02] And again, this work that you've just explained to us and the essay that you wrote and that was published in Science led to winning this prize from Eppendorf and Science. What did it mean to you to win this prize?  

Ann Kennedy, PhD [00:14:15] It was a really nice validation of the work that we've been doing. I think we're a little bit atypical because this project was fully computational. We'd worked with existing imaging data, but it was about trying to make sense of neural population coding in this data. So it was nice to see that you could explain it in a way that it registered with the judges. And also it's nice to have a chance to communicate this story with an audience that wouldn't normally be seeing it.  

Erin Spain, MS [00:14:38] Your essay, it could not exceed 1000 words. That was part of the rules of the competition. Is it difficult to explain your work in such tight parameters?  

Ann Kennedy, PhD [00:14:48] Yeah, I think it's a nice challenge. It's like the Mark Twain quote. If I had enough time, they would all be short stories. I think writing a long elaborate description of what you do is doable. Maybe not easy, but getting it trimmed down to capture the kernel of the story is trickier, and I think we were just lucky here that there's been so much theorizing over the years about how the control of motivational states like aggression might be structured that we kind of tie things to. And there was a narrative of the experiments to kind of communicate in the essay.   

Erin Spain, MS [00:15:18] So in a lot of ways you are just getting started. You just established your lab at Feinberg a few years ago. This award you won is for early career scientists. So tell us what's on the horizon next for you.  

Ann Kennedy, PhD [00:15:31] Like you said, this award, you have to be 35 or under to get it. So it's really targeted towards senior postdocs starting faculty. And most of the work that I described in my essay was built out of my postdoctoral work. In my own lab we're pursuing a couple of different routes. Part of it is building better tools to understand the structure of animals actions and how that structure is modulated by the animal's motivational state. If you apply a manipulation that makes an animal more anxious or more aggressive or more hungry, how is that shaping the animal's moment-to-moment decision making? So what does that tell us about the translation between your motivational state and your moment-to-moment behavior? We're also interested in a sort of computational level at how these motivational states get set. When I see the time constant of integration of this ramping dimension changes, what does that actually mean in terms of a neural circuit? What is changing to scale this ramping up and this persistence? And we think that this could be a sort of synaptic plasticity between cells, but it could also be a change in the expression of a neuropeptide or a neural modulator within the system. And so we're interested in building models of behavioral control at the level of neural populations to try to understand what kinds of control a neural circuit can have over an animal's motivational state and its resulting behavior.  

Erin Spain, MS [00:16:48] Thank you so much, Dr. Ann Kennedy, for your time today and telling us about your winning essay.  

Ann Kennedy, PhD [00:16:55] Thank you so much. It was great to talk to you.  

Erin Spain, MS [00:17:07] Thanks for listening. And be sure to subscribe to this show on Apple Podcasts or wherever you listen to podcasts and rate and reviews. Also for medical professionals, this episode of Breakthroughs is available for CME Credit. Go to our website feinberg.northwestern.edu and search CME.