Keynote Address and Closing: NIMH 75th Anniversary Event 3

Transcript

SUSAN AMARA: Thanks, Dan. And I also want to thank the panelists and previous speakers for providing such a compelling and exciting vision of the future. I think it's given me a lot of optimism for the future. So, thank you all. I'm Susan Amara. And I'm the scientific director at NIMH.

And it's really my honor and pleasure to introduce Dr. Kafui Dzirasa. Dr. Dzirasa is the A. Eugene and Marie Washington Presidential Distinguished Professor in the Departments of Psychiatry and Behavioral Sciences, Neurobiology, Biomedical Engineering, and Neurosurgery at Duke University. I had to have notes to remember [laughs] all that.

He received his M.D. and Ph.D. in neurobiology from Duke in 2009. And completed his residency in general psychiatry in 2016. Dr. Dzirasa really has been at the forefront in the development of state-of-the-art approaches aimed at revealing the circuits and brain states that underlie neurological and mental disorders.

Through elegant studies using in vivo electrophysiological recordings in animal models and together with machine learning approaches. His work has really explored the changes in the large-scale patterns of brain activity -- brain network activity that can be linked to the presence or absence of depressive states.

His overarching goal is to combine his research, medical training, and community experience to improve outcomes for diverse communities suffering from neuropsychiatric disorders.

Among his many awards -- and I really mean many -- and honors, he's received the Presidential Early Career Award for Scientists and Engineers. He's also received the Society for Neuroscience Young Investigator Award.

And in 2021, he was elected to the National Academy of Medicine. Please, give a warm welcome to Dr. Dzirasa.

KAFUI DZIRASA: Well, good afternoon. It's a tremendous pleasure to be here. In many ways, this is home for me. I grew up not too far down the street, in Silver Spring, Maryland. And spent a ton of weekends at the Air and Space Museum. So, this is like it right here [laughs].

I met the NIMH in 2009. I was in my last year of medical school and I wanted to figure out what translational research was. So, I signed up for an array rotation at the National Institute of Mental Health. And I was in the laboratory of Carlos Zarate.

And while I was there, I got to meet the director. I emailed him and said, "I'd love to talk to you about futures and, you know, where neuroscience is going." And Tom took me out to lunch. And he played such a remarkable role in shaping much of what you'll see today.

He convinced me to do psychiatry residency. I see a bunch of other friends here from Maryland. People I've met along the way who have really shaped how I think about research. And how to bring a bunch of different disciplines in my experiences to bear on how I think about treating mental illness.

I'll warn you in advance, this talk is going to wind through a lot of different directions. My goal is to solve a massive problem. And I'm an engineer, which means I grab from all kinds of stuff to solve that problem. I'll do my best to orient you to the different disciplines and spaces that we'll move into as we go along.

All right. This is my status slide in here, but I'm working on it [laughs]. All right. So, this slide needs no background here, right? These illnesses impact people. Our whole country is totally familiar with them, particularly, as we've gone through the pandemic, the impact they have on individuals' lives.

If you haven't experienced these in a deep way, you see them in your kids, your family members, your loved ones. When we look at our young people, they're suffering tremendously. These sorts of plots on -- this is particularly looking at persistent feelings of sadness in our young people. These sorts of plots are alike all throughout the world.

And so, these illnesses are impacting people tremendously. Now, as Brittany talked about during her talk, I think this mental health model of preventing illness is something that's central to how we should be thinking about things. And [unintelligible] sort of that triangle outlined that way.

I didn't get an M.P.H. [laughs] in my many years of school. But I love this idea of prevention. And I got obsessed with it as an early career faculty member as well. Can we actually prevent mental illness? And there's sort of two things that I anchored the idea of preventing mental illness around. The first one is, can we predict who's going to get sick, right?

So, we need some sort of prediction mechanism. And then we need some sort of intervention, right? So, the talk is going to center on this idea of risk prevention or prediction and intervention.

All right. I hit this concept when I was in residency at Duke, how it was my chair at the time. And I had this observation that folks kept coming into the inpatient hospital. And there was this precipitating factor that showed up whether you had depression or bipolar disorder or schizophrenia, there was like stress, right?

And so, stress seemed to be there at the onset of illness, whether it was you sending your teenage -- emerging adults off to college. Or whether you're having some major family trauma. And again, we all get this as we've gone through the pandemic. So, stress was this major participating -- precipitating factor.

But not everyone who experienced stress has a major psychiatric illness, right? And so, what I really became interested in were, what were the biological factors that determined who, when they were stressed, would ultimately decompensate or have illness?

You see that the red squares there. And who would ultimately not develop illness? And we'll call that resilience through the top, right? And so, our idea was basically to track individuals. And those individuals not being humans. I'll get there in a second.

And see if there was a central biology that mediated vulnerability. So, vulnerability means not being resilient to future stress. And we're able to do this. We published that work in 2018. But we were able to find a biological signature -- in other words, a biomarker associated with vulnerability to future stress.

So, I'm going to show you how we did that now. We're going to sort of force into neuroscience and then into machine learning. So, we record electrical activity from many sites in the brain at the same time in mice. That's our model organism that we use.

And you've seen brain waves like this. You've seen them throughout the day. Antonio showed some slides with brain wave. Nicole showed some slides with brain wave. We're recording electrical activity from the tips of individual wires inside the animal's brain.

And then we can do standard engineering analysis. We can ask how much activity is there in that wave of electrical activity. These waves represent populations of neurons firing together. We can look at activity across different frequencies, which -- in each of those brain areas.

And then we can take advantage of engineering principles that say, "Things that change together over time tend to lie within the same system." So, we can quantify how much these brain rhythms in different brain areas change together over time.

And you'll hear me use the term synchrony or coherence, right? I'm just showing you two brainwaves here. And you can see how the red tends to align the peaks together, right? So, that's synchrony.

Then we can borrow from our friends. This is used in weather forecasting or in the stock market. It's essentially statistical forecasting. And what you can do is actually infer how information is moving through the system.

So, if the current activity in one brain area, let's say hippocampus, aligns with future activity in prefrontal cortex. In this case, you can infer that information is moving from hippocampus to prefrontal cortex. So, that's directionality. Or you can do the same thing and calculate that information in the other direction and infer that information is moving in the other direction.

So, for each mouse, we're recording about 10,000 data points per second, right? We've got activity in each of the brain areas. We've got activity across frequencies [laughs]. And then we've got these frequencies interacting with each other, basically creating a really big data problem.

And so, we started acquiring this information, and then we actually didn't know how to make sense of it. And so, one day, I found myself talking with a colleague. He was the chair of computer science and electrical engineering. And his graduate student at the time, now -- who's now a tenured faculty member at Duke, about this challenge of how you make sense of all this data.

And they were working across multiple disciplines, including like weather forecasting. And they said, "The problem you have in neuroscience sounds a lot like the problems we have in geothermal imaging." And so, they took their models. They built this model out, which was to make sense of our brain information that we were gathering.

And I'll explain to you what you see here through an analogy, right? So, when you hear music, right, you're basically hearing pressure waves traveling through the air, about 10,000 oscillations per second.

So, think of your LFPs as music, right? So, brain music -- you're hearing music. And what we're going to do is we're going to take that music and then we're going to chunk it up into notes, right? And those notes are going to be the oscillatory amplitude measures of power that I showed you, the synchronization or the leading and the lagging.

So, we've got all this music. We're going to break it up to notes. Makes sense? And then we're going to ask how those notes change together over time. In other words, we're going to figure out the chords that are playing and the instruments that are playing through a process called, using machine learning, a supervised auto-encoder.

And then we're going to relate those chords to behavior. So, what the animal's doing, right? And we're going to build one final piece onto this model that I think is the most important piece. We're going to ensure generalization.

And what that means is when we learn a network, we're going to make sure that that works more than just for the network, the animal that we've learned it on, right? And the principle here that you can appreciate when you're a clinician is if you have any sort of measure, like an EKG, when you go in the emergency room, you want to make sure that it works for the new person that comes in the door. Not just everybody that came in the door before.

So, we're going to build on this piece that makes sure if you hear that music with a new tempo or a new set of instruments, you still know it's the same music, right? So, we're going to ensure that it generalizes.

Okay. So, here's a paradigm that we're going to do this in. We're going to take a bunch of C57 mice. This is an inbred strain of mice. And then we're going to implant them with electrodes across a series of brain areas that have been implicated in stress behavior.

So, for the neuroscientists here, infralimbic and prelimbic cortex, nucleus accumbens, ventral hippocampus, amygdala, and ventral tegmental area. Everyone else ignored those words [laughs]. And we're going to record brain activity.

And then we're going to take these mice and we're going to put them through a stress paradigm. And the way the stress paradigm works is we put them in the same cage arena with an animal that's about 50 percent bigger. And these are going to be retired males, which makes them really aggressive.

And then we're going to pick the 10 percent most aggressive of these aggressive mice, right? So, if anyone has an older sibling, it was just like that growing up. So, this doesn't go well for the little mouse. It gets beat up.

And then after 24 hours, you put it in a new cage with a new older sibling who also beats it up. So, it gets 10 days of this in a row, getting beat up by different mice. This work was published, led by Rainbo Hultman, who's now in a faculty at Iowa.

All right. And what's fascinating about this is if you take these genetically identical animals and you put them through this paradigm, about 40 percent of them look just fine -- mostly fine. Sixty percent of them don't sleep well. They don't like rewarding substances. They don't like other mice. So, they have the stress-related behavioral problems, right?

So, now you've got the groups that we set up in this sort of prevention algorithm. We've got everybody before stress. We stress them out. We know how they respond to stress. And so, we can look at their brain activity before stress.

When we do that, we see something that looks like this. And we call these our electrical functional connect domes. I'll tell you the take home before I explain these. The take home is the picture on the right is different than the picture on the left, right?

[laughter]

All right. So, the picture on the right is the signature that shows up in the animals' brains after their stress. And the signature on the left is a signature that is there before stress when all the animals otherwise look exactly the same, right?

So, you're looking around the rim of the circle, we've got our brain areas there. If you see color around the tire, it means that brain area in that frequency contributes to the chord. It's part of the chord that's playing. The lines through the center means a brain area synchronizing.

And for the brain areas that are synchronizing, you see the information moving through the brain at the bottom. But again, the take home is, there's a signature that is different that predicts who's going to respond well to stress and who's not. So, we achieved the first part.

So, we're like, you know, nine minutes into the talk, [laughs] and prediction is done. Which means intervention is going to turn out to be way more complicated than we anticipated. This is the fun of science.

All right. So, we love using simple assays in the lab. The former director is probably cringing right now. He is saying, "This is not depression. This is not depression. I'm going to say it."

But we love using these simple tests. One, because there's a ton of literature running and other people have used it. And I'm not arguing that this is depression. I just need you to accept two things so we [laughs] -- before we move on.

The first is that this is stressful for the animal, right? I can objectively tell you this is stressful because we're going to measure the [laughs] animal's stress hormones and it goes up when we do this to the animal, right? So, we just need you to accept that this is objectively stressful.

The second thing is that this assay is sensitive to how much stress the animal has previously been exposed to. So, if I stress an animal out and then I put it on this test, it behaves differently, right?

So, the test is stressful and the assay is sensitive to prior stress. That's it. It's not depression. Not depression. All right [laughs]. So, we're going to take advantage of these two things by putting the animal on the test two days in a row, right?

And so, it's in the same context. But the difference between the first day and the second day is it's been previously stressed, which is the stressor of the first day. And what we're going to ask is, how do computations in the animal's brain change as a result of the prior stressor, right?

So, is it computing different between day one and day two? All right. And when you do this again to a group of animals -- I'm showing you a bunch of animals here. I'm showing you their behavior on the first day and the second day.

As a population, the animals show greater immobility on the second day. That's what I mean when I say sensitive to this -- the prior stress. But it's not all the animals. There's some variability across the population of animals.

Okay. So, we've got our animals implanted. We're recording from a series of brain areas. And what we want to know is, do any of the computations in any of the brain areas change? It's sort of a brain screen of what's changing as a result of prior stress exposure that computes how you deal with stress.

So, recording all these brain areas, we've got day one in gray, day two in red. I'm showing you a bunch of single cells that we've recorded. And the thing to take home from here is there are two brain areas that the computation looks a lot different, right?

One is infralimbic cortex, which is in the mouse, what we call mouse prefrontal cortex. And one is in the medial dorsal thalamus. So, the computations in the other brain areas look fine. But these two areas have changed.

Again, we're not recording comprehensively all the brain areas. But it's highlighting that something interesting might be going on here. So, now, we're going to look at each individual mouse, right? We are recording before -- we're pulling a bunch of cells.

Now, we're going back to the oscillations. And the oscillations, we get an oscillation from every single mouse, right? So, we're going to ask the simple questions; are -- is the power in these brain areas different? Right? Prefrontal cortex, infralimbic cortex, or medial dorsal thalamus? The answer is no.

Is there synchrony between them? No [laughs]. Is the flow of information different? No. This is again -- the rabbit hole gets deeper. And we were sort of reminded at the time that there was another interesting pattern of synchronization that could happen in the brain, right?

So, everybody's thinking of things changing together. You can also have a pattern of synchronization where things are changing across frequency. So, it's another pattern of synchronization. I practiced to get that right in talks [laughs].

So, it's another way the brain can synchronize. And this is called cross frequency phase coupling, where you have the amplitude of the higher oscillation synchronized with the phase of the low oscillation.

So, we calculated this property. And I'm showing you a plaque here. The phase of infralimbic cortex is on the bottom. The amplitude of medial dorsal thalamus is on the y-axis. And red is the -- is coupling. Blue is no coupling. So, we can actually find coupling between these two brain areas across frequencies.

And this coupling changes between day one and day two. And importantly enough, the amount that this coupling changes is related to the individual behavioral change of each of the animals. So, it looked like we found a signature that we can go after that's involved in how the animals are adapting to stress.

Okay. So, I'm an engineer. So, I immediately think of, you know, two things, right? Whenever we observe something, it could be the thing, right? So, this is the thing causing our stress problem, right? And what we want to do then is prevent it, right?

So, it's a primary problem. And if it's a primary problem, we want to get rid of it. The issue is, it may be compensatory, right? And what I mean by compensatory is if you run up a flight of stairs, your heart rate starts going faster, right?

What you don't want to do is stop your heart rate from [laughs] going faster. Because what it's trying to do is get more blood to your muscles, right? So, the heart rate increase is compensatory.

So, we need to know whether it's primary compensatory because it determines whether we want to create the signature or get rid of the signature, right? So, the way we're going to do this is we need to figure out some way to create cross frequency coupling in the brain.

You heard some talk -- Antonio mentioned this. And Nicole mentioned this as well, about closed-loop modulation, right? So, we're essentially going to use the same tools they talked about to create closed-loop modulation. And what we're going to do in principle is we're going to read electrical activity from infralimbic cortex.

So, the slow away from the infralimbic cortex. And then we're going to stimulate the medial dorsal thalamus at the higher frequency, essentially coupling them across frequencies. And we somehow got to figure out how to read the information and put it back in the brain in about 30 milliseconds.

I had a really awesome undergrad in the lab who figured out how to do this. And so, we're just going to extract information from the mouse's brain, process it in a computer, and figure out what the wave is in infralimbic cortex.

And then stimulate a high frequency burst in thalamus that's timed to it. And we're going to use a tool called channelrhodopsin. The simple sort of take home of channelrhodopsin is blue light makes cells fire [laughs], and yellow light doesn't, right?

So, blue light's going to be our experimental group. Yellow light's going to be our control. And we're going to create this within the animal's head and see what happens. When we did that, again, blue light's our experimental group, yellow lights our negative control. I'm just showing you the amount of data mobility the animals show when we do this. When we basically play that signature, read from infralimbic [phonetic] cortex right into mediodorsal thalamus, it makes the mice less immobile.

In other words, it looks like we're pushing some degree of resilience into the animal's brain by doing this, right? And this argues that the signature isn't actually primary. It is compensatory giving us a pathway to intervene now in specific animals. So, I'm just showing you what this looks like on the top here. In other words, when we time the signatures to the wave, the animals, green is good, right? It shows more behavior that looks like resilience.

But we did two important controls that really highlighted what we were running into. The first thing we did was we delivered the exact same number of light pulses into the brain, right? So, maybe this is just light going in the brain. This is just stimulation.

So, we delivered the exact same number of light pulses. We just delivered them in a fixed pattern where it wasn't related to the oscillations. Now, it's not timed and we were shocked. The animal's behavior actually went in the exact opposite direction. In other words, they got worse. And so, this is -- for an exam, this is like signal to noise for us. We're just putting noise in the brain and canceling out how the circuit is normally computing.

And then we did one final control which is where we said maybe it's the threes that are important and not, you know, the timing. So, we just had another experiment where we played threes, pattern of threes in the brain and the animal has no behavior change whatsoever. And the reason this is likely happening is some of the threes ended up at the bottom of the peak and some of the threes don't. So, it all basically cancels itself out.

But what this really highlights is that if we're going to intervene, it's not that we just need to sort of find the area of the brain and stimulate it, we need to have approach where we can record or read and write in a closed-loop way that gets the timing right. The timing's critical.

Okay. So, we're really excited about this. This was somewhere around 2016. And we were like, "We solved all the world's problems. We're going to cure mental illness." All we need to do is figure out how to create massive numbers of post-loop stimulation devices, deploy them across the world, make everybody resilient, and then we will retire and enjoy ourselves.

This is a picture of my family and the open chair is my uncle's. And my family on the bottom row, four out of five of the siblings carries a major psychiatric diagnosis whether depression or bipolar disorder, schizoaffective disorder. And this is my grandmother's funeral. The open chair is my uncle's who's -- who has bipolar disorder and totally decompensated when his mom passed away.

And I used this picture because I had a conversation with him about a year and a half before as we come up with this beautiful closed-loop stimulation device. And my motivation for getting into science and getting into this area of work is entirely selfish. Like I want to help my loved ones. Everyone else will benefit as a process of this but I fear for my loved ones. And I just like to be upfront about that.

And I was talking to him and he was like, "This is wonderful. It's amazing. The whole family is so proud of you. You've done it. Tell me more about how this is going to get to the third world because this isn't God." And I remember sitting there excited and then ashamed because there was no way it was going to get to the third world.

And in looking at those initial graphs that I put up, seeing the burden of illness worldwide with depression being a bit top of the list, it occurred to me that I was working on making interventions that was not going to move the needle on the burden of illness in the world at all because it just wasn't going to get to enough people, right? It was going to take, you know, eight billion down to -- let me not do the calculating, 7,999,999,999. In other words, it wasn't going to move things enough. So, we needed to go back to the drawing board and think about how would one make a scalable intervention that could improve resilience.

All right. So, we started thinking -- this is all conceptual, right? It's a concept of a plan. And we started thinking about what is like a scalable intervention? So, it occurred to us like two things are for it to be scalable. Antibiotics and vaccines, and they could be deployed worldwide. My thoughts on vaccines changed a little bit after the pandemic but sure. Before then, I thought they can be deployed worldwide relatively quickly.

Now, antibiotics have an interesting thing. What they do is they target non-self, right, to make self, to protect self, right? So, that's not what we're trying to achieve with -- in the case of mental illness. Vaccines on the other hand target self to make itself more resilient to the environment. So, in principle, the concept is what we will want to achieve is a vaccine, right? So, let's see if we could turn this like brain stimulation device into a vaccine.

So, in 2018, I was on the BRAIN Initiative work group and I heard about like two exciting technologies being developed that made this concept a little more concrete. The first one was a human brain cell atlas that's now underway. And the idea is we were going to sequence all of the cells in the human brain and find out the gene expression profiles, each individual cell type, and we're going to do this across all ancestries so it'll work for everybody. And what this would ultimately provide is a GPS system that you could use to express things in specific cell types. So, it's a locator signal, right?

The second thing that was being created were delivery systems that would cross the blood brain barrier. There's a colleague of mine at Caltech, Viviana, who's working on AAVs that cross the blood brain barrier, right? If you could take these two technologies and combine them, you could create an AAV that you could deliver a package to specific brain cell types that you could package into a vaccine that could be scalable. And all you needed to do was take that closed-loop electrical stimulation system based in silicon and turn it into proteins. And so, the entire rest of this talk is going to be focused on our goal of turning our closed-loop device into proteins.

So, I'd spent time in medical school and as part of medical school, we rotate through a bunch of rotations. One of which is cardiology. And so, I've always been obsessed with the heart. And it turns out this heart has an interesting what we call [unintelligible] system that takes information. Ions from one cell and pass it directly to the next cell and it's dumped through gap junctions which are called electrical synapses. You could use those terms interchangeably.

So, it passes ions from one cell to another, ions are electricity. So, it reads and writes from one cell to the next. That's why your heart stays synchronized. These gap junctions are made of connexin proteins. They are expressed in most of the tissue throughout the body and 20 isoforms in humans. There are some challenges with them which is they're bidirectional. You ever see a heart defibrillate, it's like current moving in all kinds of directions. There are all kinds of things going on.

All right. What was interesting about gap junctions is about six or seven years earlier, somebody had taken these connexin proteins and put it in worms. They expressed them in neurons in worms. And what they showed you could do is just by expressing these proteins, you can create an electrical synapse or gap junction between them, synchronize these neurons, and it changes the behavior of the worm. So, you can essentially rewire or edit part of a circuit just by putting gap junctions in there.

Now, this is great in worms because worms have one cell of each cell type so your GPS system is like sort of really optimal. You can appreciate the challenges of trying to do this in a mammalian nervous system. Mostly because you have many cells of the same cell type. So, let's suppose you wanted to have a closed-loop system that read from the presynaptic neuron on the left and wrote into the postsynaptic neuron on the right, you can insert gap junctions in between them that you see there in pink.

So, it's sort of like an electrical synapsis that you put between them. But you would immediately have two problems. The first of which is you'd also express gap junctions on the other presynaptic neuron because it's the same cell type. And what that would do, it would ultimately scramble the signal between the presynaptic neurons because current will be flowing in directions that it ultimately shouldn't.

The second thing is because the gap junction is essentially poor, you'd have current moving in both directions. So, this is sort of like talking to somebody when they're yelling at you. It's not really useful.

And so, we had a concept of a plan and so we made a viral vector with gap junctions just to try this out to see what would happen. We put it in the ventral hippocampus in mice. I'm showing you two doses which we did this and we injected the mice. This is -- my grad students said we really don't have behavior. I said, this is the clearest behavior that I've ever seen. It was totally lethal. The animals had seizures and died. Because we were essentially hyperconnecting their hippocampus, exactly the picture that you would see on your left.

And so, we had a -- what I call my dream team in science. It's generally some really smart people. And here's the idea that we came up with. Instead of having the gap junctions work like stickers, what if we made them work like magnets? Where there's a positive end and a negative end. Excuse me for everyone -- any one that's color blind. The colors will have relevance later.

But the idea is the green, if it's a positive event, it won't interact with green. If red is negative, red won't interact with red. But the green will interact with red. And just for kicks, let's make it so current preferentially flows in one direction. So, this is conceptually what we're trying to achieve.

So, I had a team of -- I call this my dream team. They're all -- they were all undergrads at the time. Two of them have PhDs now. One's a medical doctor. Just to show how long this has gone on. But I love using undergrads for projects because when you tell them stuff, they don't know there's no way it will work.

So, they searched through the literature. They searched through all kinds of connexins and all kinds of species. They searched for innexins which are invertebrates, pannexins, and came up with this pair that's in goldfish and perch fish. And connexin 34.7 and 35, their names are based on how much they weigh in kilodaltons.

And what was really fascinating about this pair was this pair could come together, these two different proteins would come together to form a gap junction. They also conducted current at exactly the amount that we would want for the mammalian nervous system. And they had this cool property they tended to rectify. In other words, they sent current more in one direction than in the other. So, this is our baseline pair that we're going to start with.

The issue with this pair is that while 34 and 35 will come together, 34 can also come together with itself and 35 can come together with itself. So, it does what we want but it also does things that we don't want it to do. And so, at that time, I'll be honest, I'm going to acknowledge this now, I barely passed biochemistry in medical school and cellular biology. But it was clear that we suddenly had to become biochemists and cellular biologists to figure out how to do protein engineering to make these cells, these proteins do what we wanted them to do.

So, it turns out that docking or the coming together between these connexin proteins, these halves, these hemichannels are controlled in part by extracellular loop 2. So, residues on extracellular loop 2. So, we just had the simple idea what if we could figure out which residues control the docking, mutate them, and then find a pair that does something that nature's never created before. This was our idea that we put into our concept of a plan that we put into a BRAIN initiative grant. And believe it or not, they funded it.

So, here's our strategy. A postdoc from my lab came up with this idea on how to achieve this. So, here's a connexin life cycle that happens between two cells. So, you have two cells, cell on top, cell on bottom. The connexins are synthesized, trafficked to the circuit's surface and then they combine to form a gap junction that you can see there on your right top left corner.

And when they do this, when you need to get rid of the gap junctions, you don't actually rip it in half. One cell swallows the whole gap junction. And so, what you realize you could do was just tag the connexins with the green fluorescent protein or red fluorescent protein. So, one cell expresses red connexins and one cell expresses green connexins.

And then you know you formed the gap junction because you get green and red in the same cell when it swallows it. So, you're simply asking question does a single cell have green and red in it. If it does, you form the connexin. So, I'm going to show you what this looks like. Pay attention to the upper left hand corner. I'm showing you a bunch of cells that we're just doing live imaging. And you'll see those sort of green and red little circles that show up. Really small green and red double layer circles that show up. That means that you formed a -- you formed a gap junction.

So, here's what we were able to do. We just take a connexin. We'll take one of our mutant pairs that we've created. We created 70 of each one. We tagged one with green. We tagged one with red. We put them together. We let them hang out. And then we see if their cells with green and red in it using a technique called flow cytometry.

And all it's going to do is tell me is whether individual cells have green and red in both of them. You see that on the bottom right hand corner. Pay attention to the plot all the way on the bottom right. One axis is how much green do you have. One axis is how much red do you have. Each dot is an individual cell. And you could see that there's nothing really in the upper left hand -- upper right hand corner. There's only green on one axis and red on the other. So, those connexins proteins do not interact.

If you look at the other axis, the plot right on top of it, this is gap junctions are being formed. In other words, you have blue dots that are off the axis. They are all sort of filling out the center there. So, it's a really simple quick assay that we could do when we have pairs to see if they form a gap junction. So, here's what we're going to do. We're going to mutate a bunch of proteins. We're going to test them against ourselves. And we're looking for proteins that don't interact with themselves. We don't care why it doesn't interact with itself. We don't care if the protein never gets formed or it doesn't corrupt. We just want to make sure it doesn't interact with itself. So, that's step one.

Then we're going to take all the proteins that don't interact with themselves and we're going to see if it interacts with the other protein. If it does interact with the other protein, it means it's getting formed correctly. And then because for kicks, we're not interested in curing mouse depression. We're going to test this against all of the other human connexin proteins just to make sure we have a translational path for it.

Okay. And you can see on the bottom right, basically 0.2 percent. So, none of the cells really have both. Here, 17 percent of the cells have both when it's positive. So, we go through this process and again, here's our three steps. We're going to test them against themselves. We're going to test them against the other one. And then we're going to screen them against human connexin proteins particularly those in the brain but we ultimately did all of them to make sure it doesn't do that.

When we do that, with 15 proteins make it past around when they're screening. So, we started with 150. We ended up with 15. We screen them against each other. We're really excited. We ended up with three pairs. These three pairs, the halves don't dock with themselves but they dock with the other one so it's a really cool property that hasn't been seen before in nature.

And then we took the last step. Again, because selfishly, my goal is to help my loved ones so we want this to work with humans. And none of the pairs survived interacting with human connexin proteins. So, they all interacted with other human connexin proteins. So, we had what may have been a reasonable mouse tool but not a useful translational tool in the long run.

And so, we felt good but progress was sort of stifled. And then it got really bad. This was all in March of 2020. And the lab shut down. And at that point in time, we took -- we all took totally new career directions. We were all sort of making bread and cookies, playing around with yeast. And I just had a new graduate student join my lab who's -- he finished up at Berkeley a little earlier and he's back on his computer science. And he convinced us to get him a supercomputer that he would put in his house. And -- because what else are we going to do when the world shuts down.

And he started doing protein modeling. And so, he took all of the protein sequences, all the means that we created. Stuck it in the computer. Figured out how to build them into an artificial membrane and then calculate how these connexin proteins were interacting with each other. Basically, running through all of these pairs and seeing what their patterns of interactions were and what controlled them.

And not to soon after that, he sort of found this interesting code and determined how connexin proteins interacted with each other and docked. They're sort of four residues with a second one down interacted with the three residues on the other side. So, it's sort of this mechanism where the whole thing locked together. And so, what we realized we could do from this computational model was mutate the residues at select sites and create versions of the protein that only docked with each other. Not with themselves and didn't interact or dock with human connexin proteins.

And so, we created these in the computer. Eventually, the lab opened up again and then we made these proteins in the lab. So, some of them we had real proteins that did this. All right. We quickly tested them just to make sure that they were functional. So, we put them in oocytes. These are basically frog eggs that had been used, sort of things that had been used classically in neuroscience. We put them in the oocytes, one in each one and then you see if electricity moves back and forth between them.

On the top, I'm showing you the wild type pair. On the bottom, I'm showing you the mutant pair. The take home here is there is current moving thorough those connexin proteins. So, check one.

Okay. So, then we went back to the C. Elegans and we loved the C. Elegans. Reached out to my colleague, Daniel Colon-Ramos, who is at Yale. I see Yale represented here. And who had been doing really cool work in C. elegans. And the C. elegans is a really nice system for testing out these connexin proteins. I'll sort of orient you as I go throughout. I never worked with C. elegans before, so all of this is like really new stuff trying to solve a problem for me.

But C. elegans is nice because when it's clear, and you'll see why it's really useful, that it's translucent. It has 302, exactly 302 neurons and neuroscientists have done an amazing job of mapping behavior onto those specific neurons. So, they know what those neurons do. And so, it's a really nice system to test out if we could change how neurons are interacting with each other because C. elegans also don't have connexin proteins. So, it's a nice clean system to test our proteins out.

So, here's our assay that I'll orient you to. So, C. elegans, I say, are a lot like graduate students. Whatever you feed them, they will return. So, if you feed a C. elegans at cold temperatures, it migrates towards cold temperatures the next day. If you feed it warm temperatures, it migrates towards warm temperatures. And that behavior in part is controlled by the interaction between two neurons. One called AFD, which is a sensory neuron. It senses temperature. And then one called AIY, which is an interneuron.

When you feed the C. elegans at cold temperatures, you can sort of see the -- see where it says calcium rises. Look under that. I don't have a laser pointer. Do I have a laser pointer? All right. So, you could see as the temperature rises on the plot middle on the bottom, the activity goes up. This is different at 61 animals and that's just showing you how much activity you have in each of the cells. In other words, the cells fire more as the temperature goes up because it's a sensory neuron.

On the right, you see what happens with the postsynaptic neuron when the connection between them is weak. Temperature goes up, that the neuron doesn't respond. That's when you feed the animals cold. When you give them a chance to go to find food, they go towards the cold. Makes sense?

All right. If you feed the animals at warm, that connection between the two neurons becomes much stronger. All right. And now, you can see that when you rise the temperature, the second neuron also increases activity because the connection between them is strong. And then the animals migrate towards the warm when you give them a chance to eat food.

So, you can assay how strongly are those two neurons connected by -- if the animal floor is warm and there's a calcium rise in the second cell. So, what we're going to do experimentally is we're going to take our proteins and we're going to put them in the two cells. In case number one, we're going to take the lock and put it in both cells. In case number two, we're going to put the key in both cells. In case number three, we're going to put the lock in one and the key in the other.

So, here's what it looks like when you have the lock and the lock. This is the calcium imaging. In other words, when you look on the right, the lock and the lock, 34.7M1 and the key and the key, 35, 35. There's no postsynaptic calcium rise. In other words, the connection between those two cells is weak as temperature goes up. It's exactly what you would expect. The lock and the lock and the key and the key aren't doing anything.

But when you put the lock and the key in, suddenly, that neuron starts behaving really different. Now, with the lock and the key in it, it's as if you've edited or changed the connection between those two cells and you have the postsynaptic calcium rise. So, you've change how that cell is functioning. But the question is what's going on behaviorally.

All right. So, the wild type cases are normal C. elegans. They like the cold. All these are cold trained. We put the lock and the lock in. The animals go to the cold. The key and the key, the animals go to the cold. But when you put the lock and the key in, suddenly, the behavior of the animals changes and the animals immediately prefer the warm temperature.

So, you can change the physiology and behavior of the animal by inserting or editing the circuit between them. Okay. And we also tested this against the connexin proteins in the human brain. They do not interact with our proteins. So, again, we're just -- at every step, we're trying to make sure that we have a translational pathway.

Okay. So, what kind of stuff can we do with this? All right. So, we got really excited and we're like, "Well, let's go back to the beginning and see what happens with these little puppies and mice." So, back to the beginning. We have this two-day assay that we're going to subject our animals to. And when we recorded the animals in this assay, we found two brain areas that were important and the interaction between them, infralimbic cortex and medial dorsal thalamus.

And so, we took our viruses and we basically put them in infralimbic cortex and medial dorsal thalamus. We put the lock in the infralimbic cortex. Sort of waited it for it to express along the axons going to the medial dorsal thalamus. We put the key into the medial dorsal thalamus. Again, you could see the red and the green. We could see that we could get sort of these colors coming together in the medial dorsal thalamus. We went through and did some physiological assessments of this circuit and we showed that we could strengthen the physiological interactions between these areas. And then we wanted to know what would happen to the behavior of this animal.

So, again, as I showed you, if you put an animal through the two-day task normally, the exposure to the stress causes their behavior to change and they are more sort of immobile on the second day as a result of the exposure to stress. So, this is what happens when we put the lock and the lock or the key and the key in. They are sort of -- they have circles or X's there. The populations basically overlap and are pooled together there. And animals show increased --f they show adaptation to the stress. So, they are changing in the way that unedited animals are.

But when we put the lock and the key in the circuit, suddenly, we saw that the population of animals weren't changing. We repeated this a few times. We've actually looked across a bunch of different assays at this point in time. But we are changing the physiology and the behavior of this circuit. So, for us, we were really excited about this because we sort of started with our huge closed-loop pacemakers, left the country, went to Ghana, went through a pandemic, came back. All sort of trying to come up with something that would help us find a scalable approach to improving resilience to mental illness.

All right. So, I'm here with one of my favorite slides in the world. I saw this at BRAIN camp. Tom invited me at the time. I was -- last year, I was a PhD student to meet a bunch of leading and eminent neuroscientists in [inaudible] put this up as a mouse researcher. So, I was hurt by this slide. I will say that people show their facial expressions in rodents are predictive since that beautiful paper a few years ago by Nadia.

But this is a real problem. Because as you think about the work we're doing in mice, whether it's guilt, or suicidality, or sadness, we don't have great ways of measuring that in these preclinical model organisms. And, in fact, if any of us were to see a mouse or a rat run across the room, most of us will jump on our chair or run out the room. Because -- we might even call an exterminator. Because we're not entirely convinced deep down inside that mice have feelings. It's unclear if they pass the Turing [phonetic] test. We surveyed the American population. It's unclear if they were -- believe that they passed the Turing test.

I'll ask a question. Right hands up. How many of you have a dog? Okay. Left hand up if you believe your dog has feelings. Yeah. Right. I gave a talk at a vet school and I was immediately convinced. Frankly, I've never had a dog. But I was convinced that dogs passed the Turing test, right? So, if at its core, psychiatric illness is so sure of occupational dysfunction that could cause all the illnesses, it's a common thing. It occurred to me that dogs are sort of the only species that live, and socialize, and work with us the way we live, and socialize, and work with each other.

So, they both pass the Turing test and they embedded themselves within the structure that is disrupted in psychiatric illness. And so, someone asked me and said, "Could you do this work in dogs? Wouldn't that be cool?" I said, "Absolutely not. There's no way I would go anywhere near that." And the person said, "No, you don't understand. You're giving your mice fluoxetine. We give dogs fluoxetine." I never knew the dogs got Prozac. Dogs get Prozac, believe it or not. Because they have behavioral disorders that look a lot like the behavioral disorders that we have in humans.

And anybody who bought a dog during the pandemic and then went back to work suddenly saw their dog had separation anxiety. Just like your kids when you're leaving them at kindergarten the first day. They look exactly the same. And I came across this fact that really changed the way I thought about this. And I was talking to a bunch of vets and there are all these sort of mixed breed pit bulls in Raleigh.

And they get anxious and when they get anxious, they get aggressive. And when they get aggressive, they get euthanized. Because in dogs, it turns out, anxiety and aggression could be terminal. And so, the case that this person, this vet that I was talking to made was that if you could figure out how to do this, you'd actually be saving dog's lives.

So, I got really motivated around this question. How could we use neuroscience and cutting-edge neuroscience, as I sort of think about the future and the opportunities to help dogs. Not simply because it's like probably a really good translational approach to understand human neuropsychiatric illness and how encoding happens and get to an endpoint. But also, because it's just really valuable and useful to help dogs.

And people in America care about dogs. So, it took about two or three years to build out my first team of collaborators to take on something like this. And I was able to ultimately bring in the humane society as collaborator to figure out how we can ask questions. Like if we can understand aggression in dogs, could we use neuroscience to minimize it so that they get adopted from shelters.

And how can we understand pain in dogs in a way that decrease osteoarthritis? How can we understand social behavior in a way that help dogs live in our homes better? And in doing so, could we understand neuroscience in a way that gets us closer to ultimately develop therapeutics and interventions that help people like my family members.

So, I'm extremely grateful to be here and with -- I'll take this last two-minute privilege just to say this. Because I think it's a real statement about American science. So, I want to talk about the people who do this work.

Liz Ransey was a postdoc in the lab who tagged the connexin protein with the different colors. She's a first -- she was a first-generation college student and she's now on faculty at Carnegie Mellon. Ryan was one of those really amazing Montessori kids from San Francisco. Came to my lab as an undergrad. He was on the dream team. Did the master's at Carnegie Mellon.

Dalton Hughes was a Meyerhoff scholar in Baltimore, Maryland. I met him as an undergrad. He's now finished his MD PhD. Elise Adamson was an undergrad in Biomedical Engineering, came to my lab later. Did her PhD in Biomedical Engineering and is now consulting. Daniel Colon-Ramos who grew up in Puerto Rico is on faculty at Yale, my collaborator on the work with C. elegans.

Tatiana Rodriguez, in the upper left-hand corner grew up in the Appalachian Mountains in Maryland, is now doing her PhD in Toxicology at NYU. Catherine, who's right below her was -- grew up in Iowa. Had two kids during the pandemic. I literally saw her in lab the morning before she gave birth. And a week and a half later, she's zooming in the lab. And I was like, "Please don't do that." But it really highlighted for me the importance of taking care of young people and young families and giving them good supportive childhood care. Because they play an important role in the discoveries that we make.

Steve joined my lab after finishing up his PhD at University of Pennsylvania. It's been fantastic with his expertise in addiction research. Eli who's now an MD PhD student in Oregon. Hannah Schwennesen who's on the undergrad dream team is now a medical doctor finishing up residency in internal medicine.

Gwenaelle Thomas who was an undergrad at UMBC as well joined my lab, who experienced countless members in inner city New York who were murdered while she was growing up. And really had deep insights about how that impacts a person going through science. She became the second black female to get a PhD in neuroscience from Duke. In the middle, Nenad Bursac, who migrated here from Serbia. Next to him, the whiz kid from Berkeley whose family migrated here from Russia.

And then finally, in the bottom right-hand corner is Rainbow who migrated to my lab from Iowa. And Rainbow's story is one of the most unique ones. Some of you know her. In graduate school, she started suffering from debilitating migraines and couldn't look at the computer.

She joined my lab as a postdoc because she wasn't sure that there were places in science for people who had disabilities and challenges. She ultimately finished her PhD. She's now on faculty at Iowa and she won the NIH New Innovator award to study the neural mechanisms underlying migraines.

So, I think it's a real statement on American science in how we create spaces for everyone to bring their disabilities and perspectives in. We can bring things and ultimately have a potential for helping people throughout the world. So, thank you so much for having me here.

Almost all of this was supported by NIMH. So, I want to thank you all. I met Josh at Society for Neuroscience in 2007. We're sort of the only two really talking about brain oscillations in mice and how you might be able to decode emotions and behavior from them. So, this is home for me. Thanks.

FEMALE SPEAKER: That was fantastic. I have many questions but I'm just going to ask one. Can you talk more -- so, okay. I love the idea of hacking into the biology and figuring out ways that we can change electrical activity in the brain.

But I want to know how much you understand how that electrical activity is actually tied to the computations that lead to the behaviors. Because we have been talking all day about how complex behavior is, how complex mental disorders are. So, what are your thoughts on what we need to do to figure that out or should we or can we?

KAFUI DZIRASA: Yeah. I have a concept of a plan. I don't know. We have to test these things out empirically. I'm not by any means arguing that human beings go playing around with their connexin proteins tomorrow. I honestly haven't figured out what resilience is. It may be that these animals are resilient to some things and not resilient in others or vulnerable to some things and not vulnerable to others.

So, I think we do have to learn that as what making an animal sort of more resilient to one stressor, we're sort of creating problems in another domain. So, we have to learn all of that. The idea of getting upstream is to figure out what is common or sort of one convergent place that we can target to ameliorate the likelihood that many different psychiatric disorders would emerge rather than thinking about how do we get rid of hallucinations, how do we get rid of sadness moving forward or upstream for that.

But in the words of the great Desmond Tutu, don't let him jump in the water.

FEMALE SPEAKER: And if I can just follow up that. In the way that you're conceptualizing things and thinking about moving upstream, you really highlighted the importance of stress. And stress as being a driver, and trigger, and exacerbator of a number of different sorts. So, do you think you want to focus on understanding the mechanisms of -- that explain how this neural activity leads to stress resilience or can we bypass that in a way and then still leverage these techniques?

KAFUI DZIRASA: It's a great question. You're sort of scratching on my multiple areas of training. So, I'm an engineer by origin. All right. So, my immediate gestalt is to say, "I just want to fix it. And if I can make it better and my family members aren't suffering, I am okay and somebody else can figure it out," right? So, it's me gathering as much information as possible that's necessary to solve the problem.

At the same time, I'm certainly a neuroscientist who's curious and is taking a lot of tools and terms to understand things. But it's understanding to a degree that I need to solve the problem and hopefully in the process, training some graduate students who will figure the rest out. If I could figure out how to modify neural circuits in a way they increase resilience and prevent the emergency psychiatric illness, I would be equally satisfied as the folks who created Prozac, and Haldol, and all of the other things that we still don't understand exactly how they work.

FEMALE SPEAKER: Thank you.

FEMALE SPEAKER: Okay. We just have one comment from the participants online from Lauren Hill at NIMH. "That was awesome."

KAFUI DZIRASA: Hello, Lauren. I actually thought it was a different Lauren Hill one. But we love you too, Lauren. [laughs]

FEMALE SPEAKER: Do we have more time?

KAFUI DZIRASA: Yeah.

FEMALE SPEAKER: Thank you for that fantastic talk. Such a fun one to end this long and interesting day. So, my question is coming from a clinical psychology background, interested in mechanisms, how specific or general do you think this mechanism is? I have two levels of that question so on the one hand, how important is it that your assay is the same day one and day two? And secondarily, what other kinds -- you mentioned other assays. Like a swim test. I'm never good at -- that's my question. Thank you.

KAFUI DZIRASA: Yep. No, that's exactly right. What our strategy has been is now, can we put animals through a bunch of different stress paradigms and with the same circuit beacon [inaudible]. Yeah. Let me just say this because I see Brianna sitting up there and I was super proud giving a talk. So, I met her when I went to the postdocs at Stanford and they invited me to give a talk. And I went out there.

And I'll never forget her pulling me aside, "I've never met a Black faculty member." So, for me, I have incredible sense of pride seeing her as a faculty member and all that she's done to create space for other people. So, I wanted to just give her a special shout out. And I thank you all for having me here. It's been a tremendous honor and a pleasure.

SHELLI AVENEVOLI: Wow. It's been an incredible day on so many levels. I can honestly say I've learned something new from every talk I heard today. And I know it's going to shape the way all of us in this room and those online think about the kind of science we've talked about.

So, first and foremost, thank you so much for being here. We know you're incredibly busy people. But thanks for sharing the day with us. Traveling here, sharing the day, sharing your ideas, and letting us push you a little further. I acknowledge that the talks we asked you to give are much different than the talks that you would traditionally give. So, we appreciate that and acknowledge that.

It's also my privilege to thank so many other people who were making -- involved in making this day possible. I especially want to acknowledge Megan Kinnane sitting in the front row. Megan led the work group that really put together the agenda for today, the organization, and the implementation. And with her is Phyllis Ampofo up in the back. Syed Rizvi, Syed Rizvi and Nichole Cook are also in the back. And Elizabeth Sekine who is sitting up here next to Megan. Thank you all so much.

I also want to acknowledge that it takes a village to plan our 75th anniversary year celebration. And these are the people. We had three major symposium but we also had several different events throughout the year. We also have an amazing team that put together materials online. I do hope you check out our 75th anniversary website. It's not going to stay there forever. But please check it out. We have videos. We have podcasts. We have feature stories. And I want to thank Natalie Zeigler, call her out especially for leading that aspect, creating the materials for us.

And lastly, I just want to thank members of the entire NIMH community which includes our investigators, our patients and families who contributed, our advocates and professional organizations that support everything we do. The celebration is really about all of us. So, we should give all ourselves a round of applause. And thank you for coming and enjoy the -- I think we can stay here until 5:30 p.m. We have to be out by 5:30 p.m. So, enjoy yourselves and thank you again.