Skip to content

Using the RDoC Framework in Developmental Research

Watch on YouTube.

Transcript

Julia Zehr 00:02:46.263 Okay, welcome everyone. My name is Julia Zehr, and I'm a program officer and branch chief in the Division of Translational Research at NIMH, and it's my pleasure to welcome you, the audience, and our speakers, Dr. Sarah Karalunas, Dr. Charles Nelson, and Dr. Nim Tottenham, to today's webinar. The purpose of the webinar is to emphasize the central role of neurodevelopment within the neuro RDoC framework and provide examples of how some investigators are using the RDoC framework in their research on children and adolescents. To set the stage for our speakers, I will briefly talk about NIMH, RDoC, and development.

Julia Zehr 00:03:27.138 So the NIMH vision is a world in which mental illnesses are prevented and cured. As a result, our mission as a funding institute is to transform the understanding and treatment of mental illnesses through basic and clinical research, paving the way for recovery, prevention, and cure. NIMH has a strategic plan for how we can achieve this mission and our goals, the full text of which you can find on the web. Right now, I wanted to draw your attention to strategic objective number two: to chart mental illness, mental illness trajectories, to determine when, where, and how to intervene. This covers the neurodevelopmental underpinnings of psychopathology, as well as the clinical course and risks across a lifespan, from early development to childhood and adolescence, into adulthood, and then into aging populations. This slide provides contact information for a few of the branches and programs supporting child and adolescent research in NIMH. In the Division of Translational Research, we have two branches, one supporting research on mechanisms and trajectories of psychopathology, and a second supporting biomarker and novel intervention development. Other divisions also support child and adolescent research, including research on treatments, interventions, services, and HIV prevention in youth. Although not listed here, our basic sciences division also supports a wide range of research on neurodevelopment. And NIMH as a whole supports research on aging and development across the lifespan.

Julia Zehr 00:04:55.852 So what is RDoC? The goal RDoC is to develop, for research purposes, new ways of classifying mental disorders based on dimensions of observable behavior and neurobiological measures. Within that framework, we want to identify fundamental components that span multiple disorders, for example, executive function, affective regulation. We want to understand the full range of variation, from normal to abnormal. We want to be able to integrate genetic, neurobiological, behavioral, environmental, and experiential components, and ultimately develop reliable and valid measures to these fundamental components for use in basic and clinical studies. This is an illustration of the framework showing five domains of mental health functioning that can be investigated over multiple units of analysis, from neuro systems to behavioral dimensions. Overlaid on these domains and measures are the influence of the environment and of neurodevelopment-- in the course of neurodevelopment. So to put this into plain language in terms of development, we know that from perinatal development through to adolescence, there's significant interactions among a variety of biological systems that are all developing with their own rates and dynamic processes. These include the neural circuits, endocrine axes, immune function. Our genetic makeup also intersects with our metabolism. We also know that many different aspects of the environment can increase or decrease risk, including maternal stress or psychopathology, experiences of extreme adversity, such as trauma and neglect, or positive factors in the environment, such as sensitive parenting or social support. All of these can impact behavior or risk in different domains of function.

Julia Zehr 00:06:44.543 So as I said before, the purpose of this webinar is to emphasize the central role of neurodevelopment within the RDoC framework. Our speakers will provide concrete examples of how understanding psychopathology from a dimensional perspective and multiple methods of analysis enhances their work. Through each talk, you'll see how understanding heterogeneity of different disorders provides important insights into the mechanisms and trajectories of neurodevelopment, as well as increases our understanding of risk and potential avenues for new interventions and treatments. With that brief introduction, I will turn the stage over to our three speakers. We do encourage you to ask questions by using the Q&A button in your zoom window. You can do this at any time. We should have time to take one or two questions at the end of the talk, and then we'll have a longer period of question and answer after all of the talks are done. You can also send your questions during or after the talk to the email address on your screen, RDoCadmin@mail.nih.gov. So the first talk will be by Dr. Sarah Karalunas, who will talk to us about using developmental heterogeneity to predict clinical outcomes in ADHD.

[silence]

Dr. Sarah Karalunas 00:08:15.965 Great. Thanks so much for that introduction. Hopefully, everyone can see the slides here. So as Julia said, I'm going to talk about using developmental heterogeneity to predict clinical outcomes in ADHD. Whoops. I have no conflicts of interest or disclosures to report. Just a quick note on current ADHD diagnosis. So ADHD is a common, chronic, impairing neurodevelopmental disorder that's characterized by symptoms of inattention, hyperactivity-impulsivity, or both of those things. And the diagnosis remains quite controversial for many reasons, but at least in part because, just like all DSM diagnoses, it's really based on these surface-level symptoms without a clear understanding of what drives those symptoms for different children with the disorder. Also, again, just like all DSM diagnostic categories, ADHD suffers from the problems associated with diagnostic heterogeneity. So there's substantial heterogeneity even just in the symptom presentations. Some kids may be more likely to have inattentive symptoms, others more likely to have hyperactive-impulsive symptoms. The symptom course for kids also varies widely across development. So what we see is that for some children with the disorder, they continue to experience persistent symptoms really throughout their lifetime. For other children, you see a decline in symptoms across adolescence, with some even achieving full remission by the time they're in late adolescence or early adulthood, and then everything in between. And one of the issues is that we really don't know what drives those differences in symptom changes.

Dr. Sarah Karalunas 00:09:59.031 There's also substantial heterogeneity within the diagnostic category in the other features that are associated with ADHD. So things like comorbid mood or anxiety disorders, social problems, peer relationship problems, learning difficulties, all vary substantially between different kids who all meet criteria for ADHD. And this diagnostic heterogeneity makes it extremely difficult to predict clinical course for any individual child. So when I see kids and families in my clinic, I can tell them some kids will have this happen and some kids might have this happen. But what parents really want to know is, "What's going to happen to my kid, my individual child?" And that's a question that we really just can't answer well at this point. So all of the research that I'm going to talk about today really focuses on better characterizing this heterogeneity within the ADHD diagnostic category, with the ultimate goal of being able to both better understand mechanisms that drive these differences, as well as to improve our ability to clinically predict for individual children what their clinical course is going to be. And I'm going to give a couple different examples talking about different domains in which we see heterogeneity and also different approaches to characterizing it, to give a kind of idea of the range of the work that we're doing.

Dr. Sarah Karalunas 00:11:16.866 So the first example that I'm going to talk about focuses on emotional heterogeneity in ADHD. Emotional symptoms and emotional dysregulation are common in the disorder. They used to be part of the course core kind of diagnostic criteria way back when ADHD was thought of as minimal brain dysfunction, so quite a while ago. They've sort of worked their way out of the core diagnostic criteria. So right now, the symptoms that we use to diagnose the disorder don't include any reference to emotionality or emotional components. That said, we know that many children with ADHD experience difficulties with emotional ability and emotion dysregulation. I mean, there's a growing literature on the role of chronic irritability as one type of emotion dysregulation that may be important both within ADHD for understanding clinical course for kids, and then also broadly across a whole range of DSM diagnostic categories. So we think that this domain may be particularly important for understanding what's going on, at least with some kids with the ADHD diagnosis. So we wanted to better understand and characterize emotional heterogeneity and to do that, we're focusing on measures of temperament. When I talk about temperament, I'm talking about it from the perspective Mary Rothbart's group, who think of these measures as capturing individual differences and emotional reactivity and emotional regulation abilities across both positive and negative affective domains. And there's a number of reasons that I focus on measures of temperament when looking at emotional heterogeneity. One of the ones that I'll just highlight here is that we have already some well-validated measures of temperament that can be used just with parent report. So if we find things that are informative using this approach, these parent report measures are easy to translate over into clinical practice.

Dr. Sarah Karalunas 00:13:00.856 So for this first study I'm going to talk about, our question was really whether there are differences between children with ADHD in their emotional response and regulation profiles, and if there are, what could these tell us that we wouldn't already know by just knowing that they had ADHD? To answer this question, we're using the Oregon longitudinal ADHD sample, which includes 368 well-characterized children with ADHD. These children all entered the study between the ages of 7 and 12 and then were assessed annually. The study is ongoing, so the data I'm going to talk about today come from the first three years of that study but we are continuing to see these kids over time. To answer our question about emotional heterogeneity, we're using our temperament measures as input features into a community detection analysis. Community detection is really just an exploratory clustering algorithm. So based on the features that you put in, in this case, things related to emotional response and regulation, the community detection will tell us whether there are groups of children, subgroups within our broader ADHD sample, who share particular common emotional features.

Dr. Sarah Karalunas 00:14:06.555 So what did we find? We did, in fact, find that there were three distinct subgroups within our ADHD sample who differed from each other in terms of their emotional response and regulation profiles. So just to orient you to the figure on the screen here, the zero line is the typically developing sample, kind of average score. Anything that deviates above or below that line would be sort of a difference from typical development. What you see in this first set of domains here, these are really core ADHD symptom domains. All three of our emotional subgroups differ from typical development in their ADHD symptoms. That's not surprising because we know that they have ADHD. Where it gets really interesting is when you move across to look at their response in both positive and then negative affective domains. So focusing first on this blue line here, what you see is that these children actually look quite similar across both positive and negative affective domains to typically developing children. So we call this our mild ADHD group. They have clear ADHD symptoms, particularly problems with attention, but they really don't differ in their other type of emotional responding from typical development.

Dr. Sarah Karalunas 00:15:18.396 If you look next at the red line, what you see is that this is a group of children who differs primarily from typically developing kids as well as from other kids with ADHD in the way they handle positive emotions. So this is a group who's really low on shyness. They're really outgoing, really high on thrill-seeking, or high-intensity pleasure seeking, really high on activity level. So we sometimes call this our exuberant group of kids. They're the kids that kind of run into my office, rearing, ready to go, excited for whatever the day brings, but so much so that it may be dysregulating. As they get older, these are kids who we think of potentially engaging in sort of maybe more risk-taking, skydiving, sort of thrill-seeking behaviors. The last group that I'll talk about is the green line here. And you can see, this group differs from typically developing kids and other kids with ADHD, primarily in the way they handle negative emotions. So this is a group who is really high in things like anger, fear, sadness. They're really difficult to soothe. I think of this as the kids who come into my office and parents describe that they have tantrums pretty regularly, meltdowns, that once they do have those tantrums, the whole day is kind of thrown off. They have a really hard time recovering from that. And we call this our irritable subgroup.

Dr. Sarah Karalunas 00:16:37.278 So we've found these three different groups of children with ADHD who all differ in their emotional response profiles and what we wanted to look at next, then, is what does this tell us that we wouldn't know simply by knowing they have ADHD? And we looked sort of in two directions. So first, we looked at neural correlates and particularly at resting-state functional connectivity, so differences in patterns of brain connectivity between these three groups. There's a whole number of findings I could talk about, but I'm going to highlight just one here, which is that our irritable group showed particularly weak connectivity between the amygdala and the anterior insula, suggesting that these regions, which are really important for emotion regulation, are not functioning the same as they are in either, again, typically developing kids or the same as they are in other children with ADHD.

Dr. Sarah Karalunas 00:17:26.648 We also wanted to look at prospective clinical prediction because, as I said, this is really the crux of what parents often want to know when they come into the office. We looked at two things, overall ratings of impairment, as well as onset of new comorbid disorders, which are kind of an indication of worsening course. So we looked one to two years after the grouping analysis was done, followed up with these kids, and determined whether they met criteria for a new DSM diagnosis that they hadn't had before. And what we found is that our irritable group, in particular, showed increased risk for onset of mood, anxiety, and disruptive behavior disorders compared to either typically developing children or either of our other two emotional subgroups. So there seems to be something important about this irritable group that's helping us understand the potential course that kids are on.

Dr. Sarah Karalunas 00:18:14.058 So just a quick summary of what I talked about so far. So there does appear to be substantial emotional variation in ADHD, significant heterogeneity in both positive and negative affective domains. And these differences are related to differences in brain connectivity. And importantly, they predict differential clinical course, at least for onset of comorbid mood and disruptive behavior problems in the differential impairment. We did look, I should say, at the fact that this predicts over and above symptom severity. So these are better predictors than just knowing their ADHD symptoms.

Dr. Sarah Karalunas 00:18:50.297 Right, so switching gears slightly. I'm going to talk about another example now where we look at cognitive heterogeneity in this same subgroup of children using a slightly different approach. So cognitive processes have been of interest for a long time within ADHD. Again, Russell Barkley kind of has this famous theory that the core symptoms that you see in ADHD and many of the associated features are driven by problems with inhibitory control. As a field, we've sort of moved beyond inhibitory control to consider other domains that are likely relevant as well. Things like working memory, attention, as well as things like reward processing. But the idea is that some of these cognitive processes are related to symptoms. And there's sort of two competing theories for how these things may be related to symptoms in ADHD. So the first is a core deficit or liability model. Here, we think that these cognitive impairments are markers of liability for the disorder. So as symptom change across development for these children, we don't expect to see changes necessarily in the impairments in executive functioning or reward processing because they sort of always-- the child always carries that risk for the disorder, regardless of what's happening over time to their symptoms.

Dr. Sarah Karalunas 00:20:06.790 The other model is sort of a compensation or severity model, and here the focus has been primarily on executive functioning or sort of top-down control processes. And the idea is that development in these processes through middle childhood, and adolescence, and early adulthood may actually be related to symptom course. So in this case, development and top-down control processes is thought to be compensatory and to help children better regulate their kind of core ADHD symptoms. So it's a mechanism driving symptom change in this case. The problem is that we don't know which of these models really captures things. There haven't been very many longitudinal studies. The brain imaging literature is also mixed in terms of finding some regions that maybe serve more as liability markers and others that show kind of this developmental maturation pattern. And kind of the crux of what I'm going to talk about today is that no studies have looked at cognitive processing or development of cognitive processing from the perspective of heterogeneity. So we've always just treated ADHD as a single group instead of thinking about how children may differ.

Dr. Sarah Karalunas 00:21:11.865 So again, we use the Oregon longitudinal ADHD sample here. In this case, we have children both with and without ADHD included in analysis. It's an accelerated longitudinal design. So we're able to map development between the ages of 7 and 13 using the data we currently have available. We're using latent growth-curve models, and we're really using two different approaches. One that focuses on sort of group level, diagnostic differences in cognitive development, and the other that looks for heterogeneity within diagnostic groups. So at the group level, what we're-- and I'm going to focus on working memory today. So we've looked at a whole variety of cognitive processes, but the working memory, our results are the most interesting so that's what I'm going to talk about. So we have this working memory task, which is one example of an executive function or top-down control process that's often impaired in ADHD. When we mapped development between the ages of 7 and 13 at the group level, what we see is that children with ADHD are impaired at age 7 compared to their typically developing peers and they remain impaired across the span of early adolescence. So they kind of were persistently impaired. But of course, I wouldn't be presenting today if there weren't some heterogeneity there since that's what we're really interested in.

Dr. Sarah Karalunas 00:22:27.571 So if we look even just within the control sample, we see that there are different trajectories of development for different children. So the control sample, or the typically developing sample, is not heterogeneous. We have one group here who starts off doing pretty well on the task and develops sort of normally across that age range, so we're kind of calling them our normal control group. We also have a group of typically developing kids who start off impaired early on in development and remain impaired across that span of middle childhood and early adolescence. If we look within our ADHD group, we also see substantial heterogeneity. So we also get a group of children who are impaired relative to typically developing controls and remain impaired across the full age span that we look at. But we also have this group who start off just as impaired as the other two impaired groups but are showing really rapid improvement in their working memory abilities, such that by the time the time they're 13, they've really normalized completely. They're no different than that normal developing group in their working memory performance. So we're really seeing a developmental kind of catch up here. And then most interestingly, this group that shows that rapid improvement in working memory skills also shows a faster improvement in their ADHD symptoms over time than either of the other ADHD groups. So this is our first kind of evidence that potentially there's something going on with working memory here that may be directly related to symptom change across this age range, at least for some children.

Dr. Sarah Karalunas 00:24:01.953 So what we see from this study is that there's substantial cognitive variation in ADHD in addition to the emotional variation that we talked about earlier, that this differences in cognitive development across time predict differential symptom course. These things are not predictable from baseline symptoms, or from baseline working memory, rather. So all the groups started off equally impaired, and it's only by looking at repeated assessments across time that we can detect this trajectory class that's showing rapid improvement in cognition and also rapid improvement in their symptoms. And really, one of the most important things here is the focus on the group level patterns. So if we didn't dig down into those diagnostic groups would really lead us to incorrect conclusions about the role of working memory and what's happening in ADHD.

Dr. Sarah Karalunas 00:24:48.835 So just a couple of comments, sort of overall comments and then talking about where we're going with this research next. So first of all, I think, one of the big takeaways for me from the work that we've been trying to do so far is that it is possible to develop alternative diagnostic schemes or typologies using a combination of these exploratory clustering methods and external validation. So using a variety of input features in an exploratory fashion to identify subgroups of children within a single diagnostic category, or even within just a population broadly. And then looking at what those groups tell us that we wouldn't know it by just looking at diagnosis, either back at sort of biological mechanisms or forward to clinical prediction. The relevance of any specific domain is likely to vary depending on goals. So it's not to say that emotional heterogeneity is the most important thing, or cognitive heterogeneity is the most important thing. It depends what you want to predict. In this case, one predicted symptoms better and the other predicted sort of things about impairment and onset of new comorbid disorders better.

Dr. Sarah Karalunas 00:25:48.615 The importance of these domains is also likely to vary across development, and for us, the next step is to really start to integrate these things. So rather than focusing separately on emotion and cognition to really think about how they interact, which is of course, how things work in the real world. And we're starting to do that using some EAG methodology, where we're looking at children's response to emotional stimuli in the context of a cognitive task and finding that their response to those things, both early reactivity and later regulatory response, really depend on that emotional profile group that they're in. All right. A quick thank you to my lab and all the staff. A number of people who've helped with this research, funding, which I guess it comes from NIMH, and then all the children and family who participated. And I'm happy to take any questions, Julia, if you have some there.

Julia Zehr 00:26:38.213 Yes. We have some questions. So one question is how the working memory-- if you could talk about working memory and gender differences in ADHD.

Dr. Sarah Karalunas 00:26:52.525 Yeah. So briefly, except for our findings, we don't see any differences in the gender distribution across these different subgroups. So there's some debate about whether particularly girls with ADHD actually may need to have more severe symptom profiles and more severe cognitive impairments in order to end up being diagnosed, and there's variety reasons both sort of socioculturally and that may be true. But at least for our particular results, we're not seeing differences in the distribution of boys or girls across these different developmental trajectories.

Julia Zehr 00:27:37.674 Hold on [laughter]. Okay, and then another question. It was very interesting to see the profiles of emotional heterogeneity within the diagnosis of ADHD, and you talked about the neural correlates as well as the perspective prediction in disorders. Within the emotional heterogeneity, can you also make predictions about the trajectories of brain development?

Dr. Sarah Karalunas 00:28:03.296 Yeah. So that's a really good question and we're in the midst of those analyses now. So we do have repeated resting state data on all of the kids in that analysis. And Damien Fair, the collaborator here at OHSU where I am, who's working on that, right now we've only done prediction to a single time point. So we can't really say anything about the developmental trajectory. But that's definitely a question we have, definitely, one that we want to be looking at as whether-- and actually, another person here, Alice Graham, has looked a little bit very early on. So infants and looking at whether their temperament profile predicts something about their brain development over a six-month time period. Those results I think were just recently published, but I couldn't tell you exactly what they were. So a number of people from our group here though are looking at whether we can predict on that trajectory pattern more than just to a single [inaudible] point.

Julia Zehr 00:29:00.926 And I think we have time for one last question, which is kind of related to that, which is when in development-- well, so we know the diagnosis of ADHD in the preschool period is relatively unstable. So when does ADHD really start and how early-- actually, you just answered that I guess. How early do differences you're reporting first make their appearance?

Dr. Sarah Karalunas 00:29:27.793 Yeah. So when does it start? I think there's still probably a lot of questions. Some people would even say we even have some prenatal studies where we're looking at differences prenatally that may predict that diagnosis later. Whether that means the diagnosis started then or these are just more likely early risk factors that interact with lots of other things to produce a diagnosis potentially, later on, is up for grabs. We've looked at the emotional profiles using the subtyping approach that I talked about only starting at age seven. So that's the youngest where we've used that particular approach. There are these other studies that I mentioned where you're looking at an infant's prediagnosis, but potentially at high risk for things like ADHD based on their family history. And we are seeing differences in both temperament profile, or temperament responding and brain development, but it's not quite using the same approach. So it's a little different.

Julia Zehr 00:30:18.545 Okay. Thank you very much. I think at this point we'll turn it over to Dr. Chuck Nelson, who's going to talk to us about predicting autism from infancy, the development of neural endophenotypes. Chuck, you're muted.

[silence]

Dr. Charles Nelson 00:30:59.346 There we go. Okay. So as I was saying, I'd like to start with a brief review of the last eight or nine years worth of work and the context for this work, and I'll begin with some background. So as a rule, a diagnosis of autism is based on a set of behaviors that typically appears in the second year of life. That said, the average age of diagnosis in the US remains stubbornly at three to four years of age. So even though this is an early appearing disorder, it's typically not diagnosed until later. And of course, the milder the case, often the older the diagnosis. And without access to state of the art care that would often occur in major medical centers, the older the diagnosis. But what's important to notice is that there are many populations that are very high risk for developing autism, and a very brief list would include the following: an infant with an older sibling with autism. The prevalence estimate is actually 20%. In our studies, it's actually closer to 30%. Children with a variety of single gene mutations. So for example, children with tuberous sclerosis complex. Roughly 60% of them develop autism. Children raised in institutions. 5 to 10 percent of those children have autism. Children with Down's syndrome. It's hard to pin down figures, so the estimate is roughly 1 to 10 percent. And in fact, even children with Duchenne muscular dystrophy, a subset will develop autism. And if you look at this sort of from 30,000 feet, it seems very unlikely that we're talking about the same disorder. This is very heterogeneous disorder. And given the multiple risk factors for this, it begs the question of, even though they meet diagnostic criteria for the disorder, mechanistically, is it the same disorder? And if that's the case, if the underlying mechanisms are different, this could account, in part, why treatment efficacy really is limited to a subset of kids. So for any child who meets criteria, who, for example, is referred out to ABA therapy, only a subset of those children will actually benefit from ABA therapy. So right now, there are no reliable behavioral indicators of autism that consistently predict outcome in infants less than 18 months of age. So there's a need to develop reliable what I'm referring to as neuro endophenotypes, or biomarkers, that can more finely splice the disorder into subtypes that can be used translationally, for example, mouse to human, that can shed light on underlying mechanism that can be used in clinical trials.

Dr. Charles Nelson 00:33:36.698 So I'm going to sort of review now a line of research that my colleague, Helen Tager-Flusberg at Boston University, and I have been doing for almost a decade, focused on infants with an older sibling with autism. So the baby sibling search consortium, which is a consortium of more than 18 groups around the world rather, funded initially by Autism Speaks, has been working on this problem for many years. It focuses on infants with an older sibling with ASD. So now, as we said, there's a risk of one and five who might develop autism. The way this might distribute is one would develop autism, another may develop a speech delay. Another may develop a problem with anxiety. And although I illustrated here, two will develop typically, even that's unclear because the longitudinal studies are very limited, both in the number of such studies and how long the studies go on. Yesterday or the day before, Matt State and I were giving talks, and Matt is a really wonderful autism geneticist. We know that the genes that are identified in autism are shared by other disorders, in particular, schizophrenia. So we are curious about the very-long term outcome of infants with older siblings. So we believe that early biomarkers can bridge the gap between the molecular level and the behavioral level, as this slide will illustrate here. Here's an infant undergoing EEG testing. Here's an infant in our lab undergoing near-infrared spectroscopy. And this shows you where autism sits, but you can see the comorbidities with other disorders, in particular, of course, ADHD, which we just heard about, and anxiety disorders.

Dr. Charles Nelson 00:35:22.087 So I'm going to focus on the first part of my talk on our use of EEG. So the first one or two studies I'll focus on have been conducted in my lab with two former postdocs, April Levin, a child neurologist, and Candice [inaudible], a clinical psychologist. And in this work-- before I go to that lovely picture of me, what we do is it's a longitudinal study starting at three months of age, we see infants at 3, 6, 9, 12, 18, 24, and 36 months of age. We have a group of infants at low risk, no family history, no siblings with the disorder, and a group at high risk where there's at least one affected older sibling. And as you'll see in the few slides, those at high risk, we can eventually separate into two groups. Those at high risk but who at three years of age do not have autism versus those at three years of age who do have autism. So the test I'm going to talk about is, we'll call it a resting state or a baseline EEG task. I spend every day in my lab blowing bubbles at infants as they sit quietly. This is not quite how it's done. And the idea is to keep them generally occupied but without imposing an actual task on them. And during this time, we're recording the brain's electrical activity, the EEG, from 128 electrodes. And here's what a typical spectrum looks like. The EEG is a multiplex signal that is comprised of multiple different frequencies of varying amplitudes. And using simple math, we can deconstruct that multiplex signal into constituent frequency. So here you would see a frequency of 4 hertz, and here you would see a frequency of 25 hertz. And you can then compute a power spectrum and see the different bands. And probably in life, most of these frequencies that we see typically are in the sort of 1 to 10 hertz band. But we are going to talk a little bit also of that activity that sits out here in the gamma band.

Dr. Charles Nelson 00:37:24.032 So let me start with this slide. These are data from a longitudinal of a study obtained at three months of age. And if you look at the simple comparison at the group level of the high-risk infants in red and the low-risk infants in blue, and if you look at high-alpha activity 9 to 13 hertz, or beta activity at 13 to 30, or gamma activity at 30 to 50 hertz, consistently, the high-risk infants show less brain activity, that is reduced power. Now, what if we then divide out the high risk into those who do and do not have a diagnosis at age three? And so if you look at the beta band and to some degree the gamma band, here are the high-risk infants that have autism at age three. Now, I'm only showing a few data points here, but in fact, we actually now have 30 infants with a confirmed diagnosis at age three. So you can see this decline over time where the lowest power is observed among the infants with the diagnosis, even lower than those who are at risk but don't have a diagnosis. What's important about this is that that means the high risk without are showing the endophenotype, they have reduced power, but not as reduced as those who actually wind up with the diagnosis. Something we may want to come to in the question and answer period is, if you look at the scatter, here's an infant, for example, with the diagnosis, whose power is at the level of the low-risk children. So not all infants will show this power distribution, although most do.

Dr. Charles Nelson 00:39:00.273 So now, can we relate this to behavior? So we find that the three-month EEG power in the higher frequency bands, beta and gamma in particular, is associated with gross motor skills at 6 and 12 months. This is inferred from the Mullen Scales of Early Learning. And so here's the gross motor key score, and here's the gross motor key score, and here's the gross motor key score at the different ages. We see that as power increases so too does the gross motor score. We also find that three-month EEG power in the mid-frequency bands is associated with early markers of autism. So at nine months, we do something called the Autism Observational Scale of Infants, the AOSI, and again we find here that higher power is associated with lower scores on this measure. So the more symptoms of autism, the lower the power. Here we see that again at three months, at all frequency bans except delta, we see an association with language skills, expressive language skills at 12 months. So again, higher power is associated with better scores in expressive language. So differences in EEG activity are present by three months of age or by the high-risk infants of lower power across most frequency bans. The high-risk infants with autism have the lowest power in the high-frequency bans. So more power at an early age seems to be associated with good outcomes or better early development, at least for motor and social communication and language.

Dr. Charles Nelson 00:40:35.931 So now let me turn to a different approach to looking at EEG activity. This was done by Brandon Keegan, a former postdoc in the lab who's now at Purdue University. Here what we did is we did a test at looking at whether infants at risk for autism showing neurotypical pattern of hemispheric specialization for face processing during the first year of life. And we present familiar and unfamiliar faces to the infants that alternate while we're recording brain activity. And we're going to look at infants at 6, 9, and 12 months, and the prediction is that with development will come increased cortical specialization. And in typical development, as is the case with adults, we find right posterior activity is greater for face processing than left activity. If what we're looking at now is frontal parietal gamma power, so essentially, the coherence between posterior cortex and anterior cortex in the left hemisphere and the right hemisphere. If you look at the blue line, which are the low-risk infants, what you see is across age an increasing shift to the right hemisphere. So the zero mark, neither left nor right is greater than the other. Below zero would be left hemisphere dominance. Above zero was right hemisphere dominance. So the low-risk kids showed the typical pattern of an expected pattern of a shift to the right hemisphere, whereas, the high-risk infants show exactly the reverse. They show a shift to the left hemisphere. Now, I'm not going to discuss this, but we also have a variety of language tasks, and we see the same thing in language - a typical pattern of hemispheric specialization in the high-risk infants.

Dr. Charles Nelson 00:42:16.177 And now here, if we divide out the high risk into those do and do not meet criteria at 36 months of age, we've computed a laterality index. So for the low-risk kids, you see it's positive, which means it's shifted to the right hemisphere. The high risk who do not have autism are shifted to the left, and the high risk here who do have autism is dramatically shifted to the left. Now, this is a little disingenuous on my part because it's only one child, but this was one of the two infants in the lowest group who eventually developed autism. And notice where they sit. They sit exactly where these children sit as well. So lowest control infants show a trajectory of increasing right lateralization of intrahemispheric coherence for face processing across the first year of life. That is the expected and typical pattern. In contrast, the high-risk infants show an atypical pattern of increasing leftward lateralization. And the results suggest that infants at risk for autism do not show a neurotypical pattern of domain-specific hemisphere specialization. And as I said before, we see the very same thing in language.

Dr. Charles Nelson 00:43:31.206 Finally, a newer tool that we've been using for a number of years now, is to look at neurometabolism using near-infrared spectroscopy. And again, the goal is, do we see differences in the high versus low-risk infants? So here's one way to do this. We have two neurosystems. This is in a conscious system. We also have a system built in the UK. And we have roughly 40 something [inaudible] on the head. Sorry, the way this would work is that we pass light through the skull, it refracts over the cortical surface, and then we have a receiving unit and a detecting unit and we can measure the degree of oxy or deoxyhemoglobin and total hemoglobin from the cortical surface. So it is a little similar to fMRI but the major difference is that we can only visualize the cortical surface. So as you heard a little bit from Sarah, and you will hear from Nim, fMRI, of course, allows you to visualize the whole brain, whereas NIRS allows you only to visualize the cortical surface. So it's a tool that is useful if the functions you are interrogating, in a sense, live on the cortical surface. Its advantage is that you can move around. There are no constraints on movement because the headgear moves with the child's head.

Dr. Charles Nelson 00:44:49.871 So I'm going to present some data from a language paradigm. We, and many others, believe that the language acquisition develops in part through detecting the statistical regularities in the speech train. So we do a statistical learning paradigm. It's an auditory processing paradigm where we present sets of artificial syllables in an ABB versus an ABC design. So the infants will listen to 28 blocks of artificial words. It's syllables in either the ABB or ABC pattern. So what we're looking for is things that repeat, as you can see in the first word here, versus things that don't repeat. So we're going to look again at infants at high risk and at low risk. We have too few kids. We added this procedure late to our longitudinal study so we don't have really enough children with autism to divide up a high-risk sample. So we're going to really just be looking at endophenotypes and we data at 3, 6, 9, and 12 months of age. So we're going to compute a functional connectivity analysis. This is very similar to what many others, including probably both Sarah and I know Nim do with fMRI, to do resting state functional connectivity. We're going to average time course over four regions of interest, and we're going to look at the correlation between the average time course for each region of interest. So we're going to show anterior and posterior regions of interest for each hemisphere, as you can see here. So in the left and the right hemispheres, we can look at within each region the correlation, and of course, we can look at the correlation for anterior to posterior.

Dr. Charles Nelson 00:46:28.545 Now, this is a bit complicated, but let me walk you through this slowly. The top are the high risk-- sorry, the top are the low-risk children, the bottom are the high-risk children. We're starting at 3 months, 6 months, 9 months, and 12 months. For the low-risk kids, if you're looking at this matrix, which is the correlation between left posterior and right posterior cortex, as you move in this direction, you see increasing, as you can see here, increasing coherence. So that is, with increasing age, you find that the correlation among brain areas is getting tighter. But for the high-risk infants, there are two notable differences. First of all, at three months, the high-risk infants actually have a higher coherence than the low risk do. But in contrast to the low risk, with development, they show increasing less coherence. So the correlation among brain regions or regions of interest is declining with development rather than increasing with development.

Dr. Charles Nelson 00:47:31.176 So briefly, high-risk infants evince an altered trajectory of connectivity across the first years of life, increased connectivity in the high-risk infants at three months of age. However, by the second half of the first year, we see reduced connectivity in high-risk infants. And [inaudible] and colleagues years ago doing an fMRI study reported the same thing. With development, they find decreased coherence. So as was the case with our speech task, we observe an atypical functional specialization by year of age. And a dominant theory in autism is one that posits two things. One is an overabundance of local connections and a reduction in long-distance connections. And I think what we're seeing here is with development, either the loss of or the failure to develop these long-distance connections. And the second that relates with the EEG data is an imbalance in excitation and inhibition. And this relates to the translational work that my colleagues, Takao Hensch and Michaela [inaudible] have been doing. Them with the mouse, me with the human, where we're trying to go back and forth to look, doing the same task, whether we can actually look at EI imbalanced in both mice and humans.

Dr. Charles Nelson 00:48:46.841 The future. So a big question for us is are these endophenotypes specific to autism or simply sensitive to features of autism? What's important about this is that we don't know. Even though we're seeing strong associations with autism outcomes, perhaps these relate to other outcomes as well. So in that context, I should add that right now my lab sees roughly a half a dozen populations of children with rare genetic disorders, all of whom have a very high risk of developing autism. This includes tuberous sclerosis, Fragile X, CPK05, PTEN, PMS, and FOXG1, and tuberous sclerosis. And the reason for this is they all have different underlying single-gene mutations, and the molecular pathways might differ, but they all develop the disorder. So what we want to know is are we predicting autism or are we predicting an atypical pattern of brain development? The second is can these tools help us to identify the connectopathy that underpins autism? And if so, when errors in brain wiring first make their appearance. If we're seeing evidence at three months of age already, atypical wiring in infants who later develop autism, that begs the question of is this truly a postnatal phenomenon or is this something that starts before birth? How does the information we're obtaining inform us about the molecular biology of the disorder? And that's why our animal models are so important because that's the only way to really get deep into the biology.

Dr. Charles Nelson 00:50:19.340 Can we use this information to develop more targeted interventions? If we can identify the circuits, and when there is sort of a circuit connectopathies develop, can we target those circuits with specific interventions? So in our work with tuberous sclerosis with [inaudible] and [inaudible] at UCLA, we're looking at-- in the first year of life, we're doing a behavioral intervention focused on joint detention in kids with TSC in the expectation that will improve social communication and maybe will actually prevent autism from developing in some cases. And then finally, can we use these tools to evaluate the efficacy of interventions? Gerry Dawson reported some years ago, changes in EEG activity based on the early start Denver model. And so what we want to know is rather than use something like a clinical severity score, or parent-reported reduction, or clinical-reported reduction in symptoms, do we actually see changes in the brain? And on that note, I'll end, and open the conversation to questions.

Julia Zehr 00:51:27.080 Okay. Thank you very much, Chuck. We have a couple questions for you, one of which is how would you translate these measures of EEG in humans back to model systems? Can you provide an example of what you consider to be a good correlate in animal models?

Dr. Charles Nelson 00:51:50.088 Sure. So let's use our work with Rett syndrome. So Rett syndrome is a MECP2 mutation. It's a devastating neurodevelopment disorder whereby kids develop typically, we think, for the first 18 or so months, and then they start to regress. There are a number of animal models on this as well. Now, one of the things that MECP2 does is it has a big effect on synaptogenesis, and we think the brain of girls with Rett syndrome is sort of underpopulated by synapses. If that's the case, EEG activity, which reflects the sort of synchronous activity of multiple large populations of neurons, should show a very different pattern. So what we did in the human is we did a study of the visual evoked potential. We simply present alternating black and white checkerboards to drive the visual system, and in a healthy brain we would see a typical evoked potential at about 100 milliseconds. Michaela [inaudible] and Takao Hensch and did exactly the same paradigm, visual evoked potentials, differing spatial frequencies in the mouse and we did it in the human, and we found surprising concurrence across a couple of levels. We find that with the mutation, in their case the knockout versus our girls, we see a greatly altered VEP. Second, if we sort the girls, or we look at the wild types based on different sub-mutations, we see that clinical severity and different severity of mutations shifts the VEP waveform. And finally, we and they observe that a very, very low spatial frequencies we see a perfectly normal VEP, but right after you start to move to higher spatial frequencies, the VEP becomes very abnormal. So in a paper we published in Annals of Neurology a few years ago, we show remarkable concurrence between mouse and human. So if we think of the VEP, and maybe even the auditory [inaudible] potential as a read out of cortical function, then here we can go back and forth. And then in the mouse, they can very quickly try different treatment protocols. So we're at the end of a clinical trial using IGF-1. But in humans, clinical trials take forever. In the mouse, they can run a clinical trial in months. And so this is a way for them to test the clinical trial on the mouse, and then we can try to translate it to the human.

Julia Zehr 00:54:19.656 Okay. And then another question from the audience. Can you expand more on the enhanced coherence found in the high-risk group early in life?

Dr. Charles Nelson 00:54:28.420 Yeah. So that's interesting. It may be that, going back to the idea of an overabundance of local connections, it could be that the brain is populated by too many synapses, and then there is overturning. Now, of course, there is no way to prove this in the human. This would only be done in the animal. But I have to say that while we were not surprised with the longitudinal data, I was a little bit surprised by the three-month data that the high-risk kids showed higher coherence. And what we now want to look at is that's not something we see in EEG, but remember in NIRS, we're looking at metabolism and not physiology.

Julia Zehr 00:55:17.644 Okay. I think with that, we probably should turn to our next speaker. A couple of these questions that have come in are also good. We're going to save them for the end of the session. So with that, let's turn it over to Dr. Nim Tottenham, who's going to be talking to us about the neurodevelopment of emotion regulation circuitry following early caregiving adversity.

[silence]

Dr. Nim Tottenham 00:56:01.169 So good afternoon everyone. The work that we focus on in our laboratory is really trying to understand this enduring and robust length between early experiences and later emotional functioning. And we are very focused on the connections between the amygdala and medial prefrontal cortex, which are bidirectional in adulthood, and really form the foundation of many of our emotional regulation abilities. However, as I'm going to describe, this connection between amygdala and prefrontal cortex takes a very long time to develop in the human, and so a lot of our work characterizes the nature of that maturation during childhood and adolescence. So if we look at a number of mental disorders that are associated with emotion dysregulation, we can see that very often, the peak age of onset is at the end of childhood, beginning of adolescence. And so that makes the antecedent period, namely infancy and childhood, very important to focus on when thinking about the construction of the circuitry supporting individual differences and emotion regulation. And so in some of our findings, what we have repeatedly seen is that the amygdala, in terms of reactivity to emotional stimuli, in this case, fear faces, shows tremendous change across development. So age here on the x-axis, an amygdala bold signal on the y, we can see that the younger we go, although there's a lot of variability, amygdala response is quite strong in the youngest children and then declines thereafter. And that's been replicated by many other groups, as well. So we get this strong early amygdala response, but when we look at the amygdala functional connections with medial prefrontal cortex, we get quite a different picture. So if you look at the adults, we get this nice, strong anticorrelation between amygdala and medial prefrontal cortex, which is consistent with a regulatory relationship. So when one region increases its activity, the other decreases. In the younger and older adolescence, we still see evidence of that anticorrelation, but it's certainly weakened in terms of magnitude.

Dr. Nim Tottenham 00:58:26.331 What we were most interested in was the observation that in childhood, we get a very different pattern of connection between amygdala and prefrontal cortex, where they're positively coupled with each other, suggesting at the very least that the nature of the communication between amygdala and medial prefrontal cortex is different in childhood than it is at points thereafter. And we've seen that the switch in the connectivity valance is associated with normative anxiety changes across development. And so this really speaks to the importance of understanding the relationship between brain and behavior within normative development to understand atypical paths. So putting these together, if we have age on the x-axis here, we see that in childhood, if not earlier, we get a very strong, robust amygdala response, but it's happening without the connections between amygdala and medial prefrontal cortex that we see in adults. So we don't see the onset of that until the beginning of adolescence, which makes childhood a very interesting time to consider when thinking about events that happen that could conceivably impact the construction of the connections between amygdala and prefrontal cortex. Moreover, we've gone on to hypothesize that the nature of the functioning in this immature state may be necessary prerequisites for the instantiation of mature healthy phenotypes.

Dr. Nim Tottenham 00:59:54.256 So to start to address that question, we collected connectivity between amygdala and medial prefrontal cortex in two different modalities, both that stimulus solicited connectivity in response to fear faces and also connectivity at rest. And we brought all of the participants back in two years later to collect the same two modalities. Now, when controlling for all possible errors here, what we observed is that across a two-year span, the strongest prediction went from stimulus solicited connectivity to resting state connectivity. So that is the nature of the way that amygdala and prefrontal cortex were communicating at point one was the best predictor of this resting state connectivity between the two, suggesting that stimulus solicited connectivities are good predictors of future functional architectures of the circuitry. Moreover, the strength of that prediction was stronger the younger and younger that we saw. So what I'm plotting here are the beta waves that change across a two-year period, and you can see that there are better predictions the younger we go. So the interpretation is that we move from this more plastic connectivity pattern early in life that switches to more stability later on in life.

Dr. Nim Tottenham 01:01:15.483 And so, we've gone on to ask, what are the variations in early environments that are happening that can produce some individual differences in amygdala-prefrontal connectivity and associated behavioral phenotypes. And so, certainly, caregiving is an environment that children experience for quite a prolonged long time under most typical circumstances. However, we've also had the opportunity to work with families that have adopted children from a very unusual caregiving circumstance, which is institutional caregiving, which, even in the best of circumstances, is really a suboptimal caregiving situation for many reasons. Most importantly, there is the lack of a stable parent in the child's life. And this is a potent stressor for a human infant. And many studies in the animal research has shown that early life stress, always, in this case, the removal of the parent, can accelerate the development of emotion regulation neurobiology, which, at first, was counterintuitive to us because early life stress is often associated with impairment. But instead, many studies are showing that there can be a facilitation of function, at least initially, following the early adversity when we looked during development. And the interpretation is that differences that we see in brain development often reflect developmental adaptations, and these adaptations are changes in brain development that help meet immediate needs. And if we understand these brain differences, in terms of adaptations, then we can better understand trade-offs that are being made between costs and benefits.

Dr. Nim Tottenham 01:03:05.478 So children that have experienced institutional care and then subsequent adoption into families have a very high-likelihood, at first, of exhibiting a number of developmental delays and yet, can show tremendous rebound in a number of behavioral domains upon adoption, in part because the families that adopt internationally are a very special group of families, who, for instance, have a great desire to provide care, we infer, because it's not easy to adopt internationally. There's tremendous heterogeneity, echoing Sarah's point. Some children are thriving by every metric that we can come up with and some children are really struggling. And if they're struggling, they tend to struggle most commonly in the area of dysregulated affect and behavior. And we have worked with two separate cohorts. One in New York City in BJ Casey's lab, and then in my own lab in Los Angeles. And previously institutionalized children in the black bars will show elevated dimensionally-scored anxiety across a number of subtypes. And when we look in both cohorts, we see evidence of hyperactivity in the amygdala in response to emotional stimuli like fear faces.

Dr. Nim Tottenham 01:04:35.772 When we look at connectivity between the amygdala and prefrontal cortex, we also see differences. But here we see age by caregiving group differences in functional connectivity. So what I'm showing you here are the data I showed you earlier where children with a typical caregiving history show a more childlike, positive correlation, that non-regulatory pattern, that switches in adolescence to that more regulated anticorrelation. We've been able to replicate that in a separate cohort. But importantly, in the children with a history of institutional care, we don't see that childlike pattern, but instead see that more adult-like anticorrelation more consistent with what we're seeing in the adolescent group. However, this difference in the brain seems to be an adaptation in that it's associated with reduced adversity-associated anxiety. So in other words, the children whose brains most differ from the typically raised children are those with the best outcomes for anxiety. So what we think we're observing is that in the context of amygdala hyperactivity early on related to this caregiver deprivation, might lead to an earlier instantiation of connections between the amygdala and medial prefrontal cortex through an activity-based process.

Dr. Nim Tottenham 01:06:08.960 So now, in the future, we have recently received renewal for this work from NIMH, and I'm adapting a figure here from insulin copper, which comes from this great paper describing the need to look at heterogeneity. And the questions that we wanted to ask in this renewal is can we - similar to the questions Sarah was asking - better predict outcomes for any particular child? And secondly, what findings from the previously institutionalized sample generalized to other forms of caregiving adversity that children can experience and what phenotypes are unique? So another way of saying this is that we can look for patterns of equifinalities. So regardless of what type adversity children might have experienced, there are some domains where we might see common outcomes. Or multifinality, wherein other domains, we see differing outcomes as a function of subtype of environment.

Dr. Nim Tottenham 01:07:11.798 So one approach would be to take a group of children with institutional care in their history and compare them to children, for example, who were domestically adopted from foster care and compare those two groups. These are traditional adversity categories that the experimenter might define. But as you can appreciate, there is significant heterogeneity even within any one of those groups. So an alternative approach is to invite that large heterogeneity into the analysis, this time the heterogeneity being based on the environments that children experience, or we can think of each child as having a unique environmental fingerprint. And for each child, we can collect a number of outcomes ranging from genetic and epigenetic, to brain structure connectivity and volume, and also collect behavioral phenotypes in the laboratory. And we do this longitudinally across a three or four-year period. And we do this iteratively on every child that comes in for the study. And we might observe that the patterns of phenotypes or outcomes cluster in more meaningful and homogenous ways. So we can then take each of these profiles and we can look on a domain by domain basis at the trajectories that we see for children. So we may see some children who start off high and well-functioning and remain so, and others who don't. And then we may get other paths where children crossover from a lower functioning point to a higher functioning point, and vice-versa. We can do that, for example, for the negative valence system, we can do it for the cognitive control system, and so on.

Dr. Nim Tottenham 01:09:06.978 And then the idea is that we get these data-driven brain behavior longitudinal clusters. So here, the phenotype is the longitudinal path for these different domains. And then we can apply these longitudinal brain behavior clusters to decision trees about the environment, and we can do this iteratively with machine learning and create random forests for the developmental experiences. And so we can more tightly link each longitudinal brain behavior cluster to the specific environmental fingerprint, so to speak, that each child has, which can include both adverse experiences but also later occurring more positive growth-promoting experiences. And so that we can start to disentangle and understand why paths diverge even though children might have started off at similar points. So I'll conclude by saying that amygdala medial prefrontal cortex circuitry continues to exhibit developmental change throughout childhood and adolescence, and it's for this reason that early caregiving experiences are so powerful in shaping emotion-regulating processes, which we observe at multiple levels of analysis. But importantly, so do later experiences. And we know much less about the ameliorative effects of later experiences. And so individual differences in mental health really take root in development, making it imperative to understand normative trajectories across multiple levels of analysis to understand when paths can go awry. So with that, I will just thank the NIMH as well as other funding sources. And I'm happy to take any questions at this point.

Julia Zehr 01:11:02.743 Thank you very much, Nim. One question is, you talked early in your talk about how plasticity moves to stability over the course of development. Could you talk a little bit about some of the mechanisms that may open or close this plasticity? And sort of as a related question, is there variability across individuals in the degree of plasticity that they may have later in development?

Dr. Nim Tottenham 01:11:34.524 Yeah. So this is, I think, a really exciting area of research and Chuck and his collaborator Takao had done a lot of this work, which I think is really moving the field forward in understanding what are the molecular mechanisms, both the go signals as well as the brakes, on plasticity during development. And so some of the major candidates that have emerged involve things like changes in GABA function across development, as well as myelination, as well as BDNF, and things that we typically think of as plasticity factors. And I think one of the more exciting areas of research is trying to understand how those go signals and brake systems can be manipulated at later points in development, which has certainly been done in animal studies and, to some degree, in some human studies as well. And I think that holds great promise for understanding how we can reverse the course of development for some individuals who might have experienced significant adversity early in life.

Julia Zehr 01:12:41.105 And then another question that came in right as you were finishing up your talk on your summary slide was about the immune system factors and whether or not you're incorporating those into your work.

Dr. Nim Tottenham 01:12:52.762 Yeah. So that's a great question. We have just begun starting to do this, but I really can't take the credit for that as much as my postdoc, Bridget Callaghan, who's become very interested in not only the immune system but also the brain-gut axis, really trying to understand extra central nervous system influences on the central nervous system. And certainly, early life experiences have pervasive effects, not only on the brain but also on the immune system and the gut as well.

Julia Zehr 01:13:27.541 Okay. And then I think at this point, we'd like to sort of open it up for questions for our panel in general. And I want to start with one question that came in earlier about the comorbidity and overlaps between ASD and ADHD. So I think there was some answer to this within the chat window, but I think it would be good to get Chuck and Sarah to talk a little bit about the neuroatypicalities between ASD and ADHD. So who wants to start?

Dr. Charles Nelson 01:14:04.717 Sarah, you want me to go first or do you want to go first?

Julia Zehr 01:14:07.142 Sure. Why don't you--

Dr. Sarah Karalunas 01:14:07.734 Sure. You go ahead.

Dr. Charles Nelson 01:14:09.118 All right. So I mean, what's interesting about this, if you look at this through our clinical lens is that if you're an autism specialist, then you see autism with ADHD as a comorbidity. And I think if you're an ADHD specialist, you see the ADHD with some features of autism. And one of the questions that someone asked was this overlap. Many kids with autism have ADHD, but is it the same ADHD that Sarah studied? So an example would be what if some of the hyperactivity-overactivity is due to anxiety, which is also another major comorbidity? I don't know if we know the variant of ADHD that kids with autism had is the same ADHD that you see in, we'll call it idiopathic ADHD. The second thing, that I responded to via text, is unfortunately for us, we haven't seen our sample beyond three years of age and there's very few studies that have seen kids beyond that. It's not until kids get to really be preschool age that it gets a little bit more obvious to see ADHD. So all 18 and 24-month olds have ADHD and bipolar if you do it based on their inability to regulate emotion in their overactivity. But we don't know yet what the long-term view of the comorbidity looks like in this population. So Sarah, what would you add to that?

Dr. Sarah Karalunas 01:15:36.298 Yeah. No, I think I agree with everything you said so far, especially that sort of that your lens shifts depending on where your specialty is for sure. We're doing some other work right now, kind of thinking about this overlap in features from a cognitive perspective since that's a lot of what I've done. And I have some work under review looking at cognitive impairments that crossover between ADHD and autism, and have them propose this kind of potential transdiagnostic phenotypes for those disorders, and finding that there's some that do seem to really be transdiagnostics, so impairments in working memory and attention regulation. But if you use some kind of novel statistical methods to decompose the cognitive components that go into successful cognitive performance, you actually can get distinct correlates, things that are associated only with autism symptoms, only with ADHD symptoms as well. So I think part of this is to say that there are likely some things that are really dimensional that exist that are probably the same symptom dimension that are cutting across both of these disorders, and they happen to co-occur at fairly high rates, and there may be some specificity that you can find too in terms of underlying, at least from a cognitive perspective where we have looked impairments also. So there may be a mix there.

Julia Zehr 01:16:53.105 Okay. So one of the questions that had-- I think across these talks, we've talked a lot about heterogeneity and where that heterogeneity can show us group level differences and inform us about the underlying methods. And then in Nim's talk, we heard more about how she's going to be using decision tree and random forest clustering to really start building better predictive models. Could the speakers talk more generally about how we might be able to use these different modeling approaches to really incorporate large amounts of data across a variety of levels of analysis from the environment to genes to neural circuits, especially within the context of the developing individual. And I don't know who might want to [inaudible] that.

Dr. Nim Tottenham 01:18:05.343 Well, I'll start. I think the ability to develop models can hold a lot of promise for applied settings, for example. So I mean, the way that we were thinking about it with children who experience early caregiving disruptions. For practitioners who are sort of at the front lines working with children, they may not have access to many of the tools that we would use in a laboratory. For example, they may not use fMRI, they may not be giving flanker tasks, so the question is can we develop a sturdy enough model that has been cross-validated enough to start to strip away dependent measures necessary to still make a decent prediction based on minimal amounts of information? In the example that I discussed, that minimal amount of information may be just the environmental fingerprint that a child has experienced. Can you, with much better certainty, make specific predictions with precision for a particular child?

Julia Zehr 01:19:20.730 The other speakers want to comment?

Dr. Nim Tottenham 01:19:22.653 Well, my answer may be answering a question that I wanted to hear, and not necessarily the question you asked. So you can push me a little bit more on this. We've made a lot of use of machine learning and other computer science-based approaches and so we can do this across multiple scales within a domain like EEG. The concern I have - and I've seen this same thing apply to MRI data - is it lacks biology. So for example, we can predict autism at three months of age using these approaches to a very high degree of sensitivity and specificity, but mechanistically, I have no idea what we're doing, meaning the computer's clearly picking up patterns that we can't pick up. And that's very important from a predictive biomarker perspective, but it doesn't necessarily tell us what's going wrong in brain development. And so I think that approach needs to be used in conjunction with more conventional approaches if we're going try to get a mechanism. But if the concern only is prediction, then that's another matter entirely. It may be a great tool to predict. But in terms of working across multiple domains, you have imaging data of all sorts. You have behavioral data of all sorts. The big issue is, at least in autism, and I know in what Nim and I do with early neglect, and I think in ADHD, the data sets are nothing like they are in schizophrenia and Alzheimer's. We do not have sample sizes of 20 or 30 thousand that allow us to do either deep sequencing or [inaudible] sequencing or things like that. And so I don't know when we're going to get there. I'm sorry. Maybe I misspoke about ADHD. Maybe there are samples like that, but I don't know.

Dr. Sarah Karalunas 01:21:19.694 No. I think we're in the same position. And I just want to echo what you already said, but I think that we've wrestled with a lot in our group too related to this is how to combine this sort of theory-driven and exploratory methods, right? Because there may be things that you pick up on with the more exploratory methods that are meaningful patterns in the data, they're predictably important. Potentially, depending on what they are, they even can point you to mechanisms. But it can be very hard. And we have certainly have many things that we can pick up on that way that are not obviously connected to mechanisms. And how we've tried to kind of balance this is by sort of selecting input features that have some theory behind them, so that therefore what we pick up on in the data is-- we already have some kind of theory for why those measures went in that would drive our interpretation of it. But that also means we may be missing things from measures we leave out. So in any case, I think it's a really complicated component to this ongoing work that we've wrestled with a lot.

Julia Zehr 01:22:18.212 Okay. And then I have one more question for the panel. So as we talk about the overlaps across disorders, what are the things that researchers should be focusing on so there can be more crosstalk between labs, and to determine the characteristics that might be unique to certain disorders versus the things that might be similar? And you may want to comment a little bit about measurement, little bit about technique standardization. But I guess the question is, how can we capitalize on the dimensional nature of the work rather than being stuck categorically within these lenses that we have?

Dr. Nim Tottenham 01:22:58.727 So there are many parts to that answer, and I think imaging has made very good progress in trying to harmonize across different sites. And so certainly large-scale studies like Human Connectome and ABCD are making their sequences available so that future harmonization is possible. And so with our renewal, for example, when we're selecting our scan frequencies, we're making them harmonizable for these larger-scale studies, because, as Chuck said, we will never have the sample sizes that we need for our particular sample. But certainly, many of the environmental factors, as well as the phenotypic factors, are going to be present in these larger-scale studies that will allow us to look at them dimensionally in a larger sample.

Dr. Charles Nelson 01:23:54.744 Yeah. I think, Julia, the way this has been dealt with in animal models makes it easier because you're looking at discrete behaviors, right? Whereas with the human, we tend to fall back on constructs like anxiety or autism or depression, rather than the behaviors that are associated with those disorders. But I don't know how to do that. So for example, this p factor that's been used in psychopathology, there's some efforts - Katie McLaughlin is now doing this - to use that with imaging data, not just psychopathology data. That might be a way to approach this, but I'm really not sure.

Dr. Sarah Karalunas 01:24:50.098 And I mean, I think a number of the things-- so these cognitive processes or emotional processes that we're looking at are-- even though I'm looking within ADHD, the thought is not that they're only relevant there, particularly if you think about some of the developmental outcomes we're looking at, things like onset of comorbid mood or anxiety problems. So the same risk factors may predict those outcomes regardless of what other diagnosis you meet criteria from. So the goal is to potentially sort of recruit these larger samples that are irrespective of diagnosis, and move in that direction. But it's a [inaudible].

Dr. Nim Tottenham 01:25:27.036 And NIH Toolbox is certainly one vehicle for synthesizing across labs. As someone who's really interested in emotion, I really look forward to more emotion tasks being built into NIH Toolbox. But that's certainly, to Chuck's point, a way to get to more discreet behaviors that can be tested across samples and labs.

Julia Zehr 01:25:54.184 Thank you. So one question that I was going to ask of the panel is how can we move-- how can we move these questions, also, into, potentially, global mental health?

Dr. Charles Nelson 01:26:17.497 Okay, so I'll field that one since I spend so much time flying around the world. So I have a large-- well, leaving aside the work that I've done for years in Romania with institutionalized kids, I oversee a large project in Dhaka, Bangladesh, in a slum there. And one of the things-- even though the focus of the work is use or neuroimaging tools in kids experiencing profound adversity, one of the things that's been very obvious to us is the significant mental health issues that many of the parents experience. Right now, we think that roughly 30% of the mothers have a depression. We know that, oddly, for a low resource country, in Dhaka anyway, kids diagnosed with autism actually have access to reasonably good services, but the parents do not. So there's a very high prevalence of anxiety and depression in the parents that goes untreated, even if their kids are getting treated for autism. And so I think that more attention needs to be paid to this, not just because-- I mean, there are clear scientific reasons for working globally and internationally, but there are also even better reasons to do so in terms of just the sheer prevalence of these problems. There's a surveillance study done in Bangladesh showing that prevalence of autism in Dhaka city is 1 in 30, and that's almost twice the US' rate. And if that's true, why is that? I don't know if that really adequately answers this question but [inaudible].

Dr. Nim Tottenham 01:27:59.870 I mean, I think that it's a really important question because ultimately, human brain development is a political and an economic issue because of the decisions and behaviors that we care about at the global level really take their root in what's happening during early development. So as Chuck's saying, we need to not only be thinking about what's happening to the developing individual, but to those caregivers that are responsible for scaffolding the brain development of those children.

Julia Zehr 01:28:37.222 Okay, I think we're pretty close to out of time. So I think with that, I'd like to really thank our speakers for their excellent talks and the audience for their excellent questions. If you have a question that has not been addressed, please feel free to email the RDoC office and/or any of our speakers individually. I'm sure they'd be happy to answer your questions or follow-up with anything that you might have. And I'd like to thank the RDoC office for organizing this seminar.

Dr. Charles Nelson 01:29:14.663 Thank you.

Dr. Nim Tottenham 01:29:15.662 Thank you.

Dr. Charles Nelson 01:29:16.842 Bye.

Julia Zehr 01:29:17.425 [inaudible] everyone. Thank you for this learning opportunity.