The NIMH Director’s Innovation Speaker Series: Genetic Clues to Autism Heterogeneity
JOSHUA GORDON: All right, everyone. Good afternoon and welcome to this edition of the NIMH Director's Innovation Speaker Series. I am the NIMH director, Joshua Gordon, and it's really my pleasure to see all of you here today and to introduce Dr. Elise Robinson, who is our innovation speaker for today. Dr. Robinson is an assistant professor of epidemiology at the Harvard T.H. Chan School of Public Health and a member of the Broad Institute of MIT and Harvard. She's also an affiliated faculty member with the Analytic and Translational Genetics Unit at Massachusetts General Hospital. Dr. Robinson's research focuses on the genetic epidemiology of behavior and cognition, and she's interested in using this genetic data to understand the biology of neurodevelopmental variation and to study differences within and between neuropsychiatric disorders. She co-chairs the Autism Working Group of the Psychiatric Genomics Consortium and the program in Neurodevelopmental Disorders at the Broad Institute. And more personally, I've gotten to know Elise through several different interactions at conferences and meetings and visits to the Broad, and always found her to be incredibly thoughtful and also exceptionally skilled at being able to explain the more challenging concepts of genetic variation to those of us who, let's just say, could use that patient explanation. And so it's with great pleasure that I hand the microphone, as it were, over to Dr. Robinson and look forward to her talk.
JOSHUA GORDON: As always, please enter questions at any time into the Q&A function and Alex Denker will moderate the Q&A at the end of the talk. Elise, thanks for coming.
ELISE ROBINSON: Thank you. Hello, everyone. So today I'm going to speak about ASD, Autism Spectrum Disorders and genetics, providing an update on the state of the association work, specifically associating genetic differences between people with the diagnosis of ASD, but particularly focusing on how ASDs are not one thing, they are many different things, and how we're starting to be able to use genetic data to understand variability within the diagnostic category of autism, as well as more broadly variability in how human beings think and act. So this slide with the buckets is meant to provide an introduction to why we would want to think about diagnostic outcomes like autism or schizophrenia or bipolar disorder both within their little buckets, the way in which they're presented to us, but also between their buckets and heterogeneity within them. There are lots of examples. For example, people who have diagnoses of both schizophrenia and bipolar disorder, switch back and forth, or the same disorder occurring in multiple members or two different disorders occurring in multiple members of a family, that we believe that there is overlap between the genetic risk factors but also the presentation of different diagnostic categories in psychiatric illness. And then there are also situations like ASD where you have a bunch of things put into one bucket. Most people who are attending this talk are probably aware that people can have a diagnosis of autism, while able to grow up, live entirely independent and happy lives, be professionally successful and have children of their own, it's also quite common to have a diagnosis of autism if one doesn't develop speech as a child and requires full-time assistance for one's life.
ELISE ROBINSON: And that's obviously an enormous suite of possibilities to put into one bucket. And in studying them with genetic data - there we go - we typically take two approaches, at least these are the two approaches that have returned the largest amount of genetic association data. The first I'm going to talk about is de novo variation and de novo variation is a difference in a variant, let's start with that, a genetic variant is any difference between two people in their genome. And we have many, many variants in us, millions upon millions of them. Most of them are benign. The types of de novo variants we're going to be talking about here are not. A de novo variant is a change that occurs primarily in the sperm or an egg cell. So it's a genetic variant that's seen in a child that's not seen in their parents. And de novo variants as a class are very common. On average, people have about 40 to 60 of them in their genomes and in general they're quite well tolerated. The average person has one de novo variant in their exome, which is the protein coding portion of our genomes that occupies about 1 to 2 percent of the genome, and about 9% of people have a de novo variant that results in the loss of function of a gene. And that 9% makes it a fairly common event and reminds us that many genes in our genomes are quite comfortable with having or our bodies are quite comfortable with having only one functioning copy of many genes. But others, that it's not the case.
ELISE ROBINSON: And through de novo variation studies, scientists have been able to identify genes that are typically intolerant to what we call heterozygous loss of function variation, that is, having one copy of a gene disrupted. It's estimated that about 10% of genes fall into this category. 71 of them at this point have been associated to ASD at a genome-wide significant level. But the number of genes you see on ASD gene list varies widely. That 71 comes from the Autism Sequencing Consortium, which was kind enough to let me talk about their data a bit today. Typically, when you see lists of ASD genes, they have somewhere in the vicinity of 100 genes, most of which are overlapping. But there are different methods through which you can associate a gene with a disorder. I'll be talking about the 71 genes that, to this point, have been associated, with genome-wide associate level, with autism, by the Autism Sequencing Consortium. On the other hand, we have something we refer to as polygenic risk. Polygenic risk is the sum of many, many, many small effects from many, many variants across our genomes. And polygenic risk comes from the highly distributed nature of genetic influence on complex traits. A good example is height, where there's not 1 gene, 10 genes, 100 genes, 1,000 genes for height. The genome-wide influences are spread across our genomes. And for complex traits like autism and schizophrenia, we similarly have tiny, little effects that are very small in terms of risk creation for something like autism but can occur commonly in the population.
ELISE ROBINSON: So any common genetic variant, any variant that's seen in more than, let's say, 1 to 5 percent of people has to have a small amount of risk associated with it, specifically a small amount of risk for something like autism or schizophrenia, which substantially decreases the average number of children one has. Any common variant that creates a large amount of risk for an outcome that reduces the number of children one has would be rapidly pushed out of the population over time or could never rise to commonality if it was new. It's a fundamental feature of selection. So when looking at the common genetic influences on reproductively deleterious outcomes, the effect of common variants has to be small, typically in odds ratio of less than 10% for a fairly uncommon one percentage outcome. To this point, the Psychiatric Genomics Consortion, which I'm lucky to co-lead with Anders Borglum, has identified about 10 snips, common genetic variants, with risk for ASD, and the ambiguity doesn't reflect lack of attention to our results sheet. I'll explain it a little bit later. So in addition to identifying specific common genetic variants that one can associate with an outcome, you can add up, like I mentioned, the total amount of risk one carries across all of these many thousands of variants in the genome to come up with a polygenic risk score, which conceptually is the sum total of one's common variant risk distributed across the whole genome. Polygenic risk is not something that one can use diagnostically for a trait like ASD or schizophrenia or at the moment any other complex trait. But it's a type of risk that has a mean shift between cases and controls.
ELISE ROBINSON: In this case, you can imagine that the red distribution is ASD cases and their polygenic risk for autism and the blue distribution is controls. And this is actually a little bit generous. It's quite generous because in reality, these curves would be closer together. But there is a shift there that reflects that in aggregate, that cumulative polygenic risk is associated with a diagnosis of ASD. So beginning with these 71 genes and their association to heterogeneity, one of the things that we've known about a rare variant or de novo variant risk for ASD, and something we've known for a long time is that strong acting de novo variants that create risk for autism are more likely to be found in individuals who have autism and intellectual disability or autism and other phenomena that are indicative of broader neurodevelopmental impairment. This isn't surprising, given that intellectual disability, certain types of epilepsy and general global developmental delay are themselves, in or outside of autism, strongly associated with genetic syndromes. To put numbers to that, on the x-axis of this graph, we have groups of people in a dataset called Simons Simplex, who have differing accounts of other phenotypes that are associated with, yeah, global developmental delay. So the people who are in the zero category have autism but walked on time, don't have seizures and don't have intellectual disability. The people in the three have autism and all three of those things. And their rate of strong acting de novo variants is shown here.
ELISE ROBINSON: So people who have ASD and none of those three indicators of global developmental delay have a rate of strong acting de novo mutations that's about thrice that of controls. But once you go up to people that have ASD plus more than one feature of global developmental delay or neurodevelopmental impairment, it's more like 10 to 15 times. So that difference actually undersells the distinction in genetic architecture because with increases in the quantity of these sorts of strong acting variants comes increases or changes to the types of variants that are in each category, and specifically individuals who are in a category of people who have more strong acting variants. The ones that they have are often worse. They cause more developmental impact and they're more likely to be a diagnostically returnable genetic variant. So this is from some ongoing work with my friends, Summer, Audrey and Stefan, and using the same simple variables with the addition of sex, you can create a tree in ASD and identify groups of individuals with ASD for whom 60% will receive a clinical genetic result upon exome sequencing of them and their parents, and then individuals who have less than a 5% chance of receiving a genetic result upon clinical testing. And I say less than 5% just because this cutoff of IQ less than 70, which is the clinical, commonly used cutoff for intellectual disability, could be increased. For example, you could use average cognitive ability or people who graduated from college and that level would continue to drop in terms of the percent of returnables genetic result.
ELISE ROBINSON: So what about heterogeneity within the genes themselves that are associated with ASD and how can we start to think about this? So as with previous publications of ASD associated gene lists, neuronal communication and gene expression regulation, excuse me, remain most strongly implicated by the sets of genes that are associated with ASD. As the list continues to expand and its sequencing efforts and intellectual disability and schizophrenia continue to expand, we're starting to see a gradient emerging in the extent to which a gene can create risk for autism without very commonly also causing global developmental delay or intellectual disability. So of the 71 genes on this list, with thanks to Mark Daley for the slide, 61 of them are also associated with intellectual disability. And this is a very well established property. It's one of the first that became obvious about ASD associated genes. On average, they're enormously shared with the genetic influences on intellectual disability. In fact, most ASD associated genes to this point have a stronger association with intellectual disability than they do with autism. Another way of saying that is if you look in clinical groups of people who have been diagnosed with intellectual disability alone against people who have been diagnosed with autism, you're more likely to see disruptive variants in those genes in the people who have been primarily or exclusively diagnosed with intellectual disability.
ELISE ROBINSON: So quite excitingly, recently, with this new gene list, you also see that there is overlap, greater overlap than one would expect by chance with genes that are associated with schizophrenia in the schema consortium analysis. And then further, the genes that are associated with intellectual disability-- I should say, the autism associated genes associated with intellectual disability and the autism associated genes that are associated with schizophrenia are less overlapping than you'd expect by chance as well, which means we're beginning to be able to differentiate ASD associated genes in terms of their phenotypic properties. But we're also beginning to understand the relationships between those phenotypic associations and certain elements of mediating biology. The most obvious of which, at this point, is that genes associated with intellectual disability are on average expressed prenatally. Genes associated with schizophrenia are on average expressed postnatally, and genes that are associated with intellectual or, excuse me, autism are on average expressed in the middle. Now this isn't Bins by any means, it's distributions. And within the ASD distribution, the genes that are more likely to be associated with intellectual disability and autism are indeed expressed later. Those more likely would be associated with schizophrenia or those that can be associated with autism in the absence of intellectual disability are on average expressed later. And this is one particular example. The Gene GRIN2B is associated with intellectual disability and autism but not with schizophrenia. The closely related GRIN2A is associated with schizophrenia but not with autism.
ELISE ROBINSON: And the primary difference between them is that GRIN2B is involved in the NMDE receptor composition prenatally, but GRIN2A, postnatally. And there's likely to be more findings and clarity in the space emerging over the next few years as the data sets in which one can examine expression timing as well as the sequencing and associated gene list expand. And it could be a very profitable space, for example, to understand conditional on equivalent expression timing. What are certain differences and how genes create risk or what phenotypes they create risk for? What are we learning about how to differentiate genes beyond simply when they're expressed? What about common variation? So as a brief recap, rare variant association to ASD studies there of have been enormously productive. The gene lists are long. Productive studies of downstream biology are being pursued. We have biological mechanisms nominated and we would like to get there someday with common variation. So in 2019, the Autism Working Group of the Psychiatric Genomics Consortium published a GWAS that identified the first genome-wide significant loci for ASD. For any genome-wide association study, which is a study that looks for these common variants of small effect, what I was mentioning, there is a mathematical theorem that once one reaches an inflection point in the activity, you start to see a rapid increase in the number of associated loci.
ELISE ROBINSON: At this point, we don't believe we reached the sample size that will be associated with that inflection point in ASD GWAS returns. However, we have a GWAS that I would describe as both underpowered, meaning we do not have enough samples, which can sound odd given that we're now up to about 30,000 cases and 30,000 controls. So we have both an underpowered GWAS but we also have an underperforming GWAS, meaning for that number of cases, we don't have the return you might expect given the trajectory of other GWAS in, for example, schizophrenia. We also have some ongoing oddities regarding the stability of the results. So when one does a genome-wide association study, you subject yourself to fairly punitive multiple testing correction where you correct for a million independent tests specifically to avoid the identification of snips that are not actually associated with your outcome of interest. And because the statistical approach to genome-wide association studies is so conservative, what you typically see, for example, for the Crohn's disease or schizophrenia or for height, is once you identify snips with an outcome, you don't lose that many of them. Ideally, you don't lose any of them but you don't lose that many. That is not what seems to be happening with ASD. This is from the PGCASD working groups, unpublished ongoing analysis of our next wave of data. Simply by introducing another 10,000 cases of Danish ancestry, the original GWAS from Broad, that was largely Danish as well. We identified seven new loci. Yay. We lost three of the original five.
ELISE ROBINSON: Definitely not yay. And there doesn't seem to be anything untoward regarding methodology happening here. It's almost certainly a function of heterogeneity. And to talk about that a little bit, as many of you likely know, there's been an enormous shift in what is diagnosed with ASD over the last 30 years, and it's become a much more commonly assigned diagnosis. And this shift is actually highly relevant to genetic studies, particularly the type of genetic studies that are aggregating data over time. And that is because people who participated in autism studies in the '90s or in the early '2000s, which is when the large collections really started for genetic analysis purposes, had on average a substantially different behavioral and cognitive profile than the people who are being ascertained into studies today. For example, the ongoing SPARK study that's funded by the Simons Foundation. You can see this in the data in a way that is entirely consistent with one's intuition about the changes that have been going on. For those of you not familiar with the epidemiologic changes, there's inflation of ASD diagnosis at both tails of the severity distribution over the last 30 years. But the inflation at the independent living, high IQ, high educational attainment tail is substantially greater, to an extent that about 30 years ago, 70 to 80 percent of people who were diagnosed with autism met criteria for intellectual disability.
ELISE ROBINSON: In the United States, England and other parts of Western Europe, that rate has fallen below 30% over the last few years, which is an enormous shift in what's been diagnosed with ASD on average at a population level. And to the extent that ASD on average, as it is diagnosed in 2021, is not genetically identical to the average phenotype being diagnosed 1990, we're going to see shifts across our building collections that influence the probability with which results are going to replicate. In the context of two specific collections that I talk about and use a lot, the SPARK data set, which is a phenomenal ongoing activity, as I mentioned, funded by SFARI, that will collect data from more than 50,000 people with ASD over the next several years. And you can reach out to SFARI if you're interested in participating. The SPARK data set is quite different from an earlier SFARI-funded effort and a lot of ways. But two of them reflect inflation at both tails. So SPARK is heavier than SSC in terms of genetic syndromes, and that is because the Simons Simplex Collection had an exclusion criteria and that was very common for the time. So it didn't enroll individuals with ASD who had syndromic features that could be severe intellectual disability, severe seizure disorders, craniofacial dysmorphia, complex medical complications, things of that nature. But SPARK will enroll those individuals. And so there are a number of genetic syndromes that appear in SPARK that you don't see in the SSC data set or others that had similar exclusion criteria.
ELISE ROBINSON: On the other tail, you see a massive increase in the number of participants who are living independently as adults or are children who are being enrolled by their parents, who are in school, in typical schools and doing well. And kind of both of those changes not only make themselves apparent through the phenotype data but they make themselves apparent through the rare variant data as well. So what does this actually mean for someone who's engaged in a genome-wide association activity? We've done some data simulations to try to understand the quantitative impact of this. And the idea is imagine, for example, that we're doing a study of schizophrenia and we have some people who've been diagnosed with schizophrenia as cases and people who do not have a diagnosis of schizophrenia who are controls. However, there are, in our schizophrenia cases, in fact, two ongoing phenomena. We'll call them schizophrenia A schizophrenia B, and we can't tell them apart phenotypically. So we've called them both schizophrenia. But in reality, these things have a genetic correlation with each other of less than one. Less than one just means that on average, their genome-wide genetic influences are not identical. So we asked the question of how punitive is that going to be when you're trying to associate genetic risk factors to an outcome if you've hidden sub-phenotypes?
ELISE ROBINSON: Without going into detail on these figures, the answer is there is an algebraic bound on the effect of phenotypic heterogeneity within a diagnostic category, and that algebraic bound relates to the average genetic correlation between these phenotypes that we can't distinguish. And when thinking about how that affects specific disorders, for something like schizophrenia, which has a very high what we call genetic correlation to itself, the impact is likely to be fairly low. And by genetic correlation to itself, what I mean is if you take many of the different cohorts that contributed to the huge international schizophrenia genetics effort and you just do a genetic correlation between schizophrenia and cohort one and schizophrenia and cohort two and three, etc, the genetic correlations are very high. They're consistently above 0.9 and average above 0.95. So in that way, schizophrenia is kind of the shining star of psychiatric genetics. We do not have that level of internal consistency with ASD. The American samples that are contributing to the Psychiatric Genomics Consortium have a genetic correlation of about 0.75 with those from Denmark, which could reflect a variety of different sources of heterogeneity. But moreover, the real potential problem when doing an ASD GWAS is that intellectual disability and autism have a genetic correlation of zero in common variant space.
ELISE ROBINSON: So I mentioned that in rare variant space, the individual genes associated with both of those phenotypes, specifically those that can be or are that are associated through loss of function variation, the individual genes are highly overlapping. But the common polygenic influences on those phenotypes are unrelated to each other, at least under current models. And in some ways, that's not particularly surprising given the relative associations of those polygenic scores to cognition in the population. So some parts of the diagnostic category of intellectual disability are more heritable with others, that most of the heritability lies in mild and moderate intellectual disability. For the most part, severe intellectual disability is both epidemiologically and genetically predominantly a sporadic phenotype. But mild, moderate intellectual disability is both heritable and familial. And in common variants space, it behaves quite well as the tail of the general population IQ distribution. Autism polygenic risk, on the other hand, actually has a positive association with IQ of the general population, and I'll talk a little bit more about that later. But that is a very well replicated and it's been interrogated extensively, and is indeed the case. So these two things have different associations with cognition and, at least on average, no association with each other. And that is a situation that's potentially enormously punitive to the GWAS because you can quite easily imagine scenarios, particularly in large registry-based studies that are minimally phenotyped, where a lot of people being enrolled - and please excuse the casualties of my language - have something that is primarily intellectual disability, we'll say. Yeah, primarily intellectual disability.
ELISE ROBINSON: And if there are a substantial number of those cases included in an ASG GWAS, that has the potential to really reduce our power to associate common genetic variants to ASD, which is one of the reasons moving forward that the Psychiatric Genomics Consortium autism group is going to be pursuing studies of ASD with and without intellectual disability and comparing the average genetic influences. Okay, going back to this observation, we've noticed that common and rare genetic risk for ASD have opposing average effects on cognition. And this is true both in people who have a diagnosis with ASD and in the general population. So in people who have ASD, as I mentioned, those that also meet criteria for intellectual disability, have a higher rate of strong acting de novo variation. In the general population and people who don't have autism, that's also the case. Strong acting or those same types of strong acting geneic events are associated with intellectual disability and lower IQ. In people with ASD, higher polygenic risk, be it for either autism or the genetically correlated trait or educational attainment, is associated with higher measured IQ. And that's also true in the general population and people who don't have autism. So recently we've been trying to figure out what this means in terms of the potential relationship between ASD associated common and rare variation. We're developing methods by which you can split up the genome-wide common variant influences on ASD risk and look at it more specifically within certain methods or certain pieces of the genome.
ELISE ROBINSON: And in this case, Dan Winer, a very talented PhD student in my lab, MDPhD student looked at the common polygenic influences on autism risk that specifically live within these genes associated with ASD and intellectual disability through loss of function variation. And what he finds is that the common polygenic risk for ASD that lives in those genes remains positively associated with IQ and positively associated with educational attainment. Meaning, yes, there's some colocalization of common and variant risk in those genes. But at least in our current analysis, it doesn't appear to be disproportionately colocalized and it does appear to be operating in opposite phenotypic directions, vis a vis cognition, which is a mystery that people will continue to look into. Okay. Another area of heterogeneity in ASD is sex. So men are diagnosed or boys and men are diagnosed with ASD about three to four times as frequently as girls and women. And there's long been hypothesized a female protective effect against many sorts of risk factors for an ASD diagnosis. And that idea has been supported by studies of rare variation. So if you look at-- if you compare male and female ASD cases, female cases have on average a higher rate, about twice the rate of strong acting and rare de novo variation, which supports this idea that more risk is needed to get to the same place if you are a girl.
ELISE ROBINSON: The relationship between sex and common variation isn't as clear. And we were surprised to recently observe what I show here, specifically in both SPARK and SSC using data from an outside cohort, the moms of kidos with autism in that cohort have, on average, higher polygenic risk than the dads, higher polygenic risk for ASD. So they're carrying more polygenic risk than the fathers of those children. And this is something that could only happen as a function of ascertainment. In the general population, even for a very sex divergent phenotype like height, there isn't a male, female difference in autosomal polygenic risk for anything. So this could only happen as a function of ascertainment. And in this case, the likely contributing ascertainment factors are that when you recruit families who have a child diagnosed with autism into a research study, you are A, recruiting families that have on average more genetic risk for ASD in the family than the general population, and that expectation is met here. Here's the general population average for polygenic risk for autism. Here's the average level in kids with ASD, and then here's the average level of their parents. So that is what you'd expect in that there's a shift. However, you expect the shift to sit about here where the dads are because parents have a 50% genetic correlation with their children. So as expected, you would guess the dads would split the difference. However, you'd also guess the same of the moms, and instead, the mothers are 50% more deviated from the general population than the fathers.
ELISE ROBINSON: So this brings us to a second ascertainment criteria that's likely contributing, specifically that in these studies, the parents generally have to be unaffected. In some studies like Simons Simplex, they're actually screening negative both for diagnosis and for aggregated autism-like traits. So what this is suggesting is that women are able to carry more risk for ASD and without meeting symptom criteria for an ASD diagnosis. And there are probably at least two things going on there, as many would say, but the exact mechanism is entirely unclear. There's likely both protection from general kind of neurodevelopmental disruption that's associated with being female. And there's also quite potentially an average difference in the expression of social communication difficulty between men and women. And there's a lot of ongoing research to look at both of those possibilities. What does this look like in cases? For time, we'll go quickly. Also consistent with a female protective effect, we see this step-like increase in how much polygenic risk for autism different cases are carrying. And the reason there are four dots here is because we have both males and females. But we've divided males and females under those that do have a strong acting de novo variant and those that do not have a strong acting de novo variant.
ELISE ROBINSON: And the increase in common polygenic risk carried by cases is consistent with the amount of liability space left to be filled where girls who do not have a strong acting de novo variant and are girls have the most space to be filled for on average meeting criteria for diagnosis, and the smallest amount of space to be filled was by the boys who do have a strong acting de novo variant. So lastly, I'm going to talk a little bit about the joy that is NeuroDev. NeuroDev is an ongoing data collection effort that I am very fortunate to participate in. To this point for predominantly funding reasons and capacity reasons, most genetic studies of any outcome, neuropsychiatric and otherwise, have been conducted in European ancestry populations in the United States and in Europe, which creates issues of both scientific value and downstream social justice. And NeuroDev is one of many ongoing projects designed to collect data from predominantly people of non-European ancestry. And NeuroDev is specifically targeting intellectual disability and ASD in now two sites in Kenya, one in Nairobi, one in Khaleefa, which is a rural area on the southeastern coast of Kenya, and then another site in Cape Town. And with some destructions thanks to COVID over the last year, NeuroDev has nonetheless collected data from now nearly 2,000 individuals, 2,000 total participants, and is beginning to do detailed genetic analysis of cases, which includes rare disease diagnosis and return of results.
ELISE ROBINSON: We're also collecting about three hours of cognitive, behavioral and medical data, all of which will, at the end of the study, be made public through the NINH, doing relevant genotype to phenotype analysis with those data and then critically collecting a large number of control samples that can be contributed to reference panels. A genetic reference panel is genetic data from people who are on average typically developing to try to get a sense of what kind of "normative variation" in the human genome looks like. And particularly for medical genetics, reference panels are enormously important because one of the reasons we are able to identify what deleterious or which variants are deleterious is we look for basically their absence in reference panels. Because if a variant is very, very bad, if it creates a severe syndromic Mendelian phenotype, it should be seen almost never in people who are typically developing. And because African individuals have substantially more genetic variation in their genomes than European ancestry individuals or many other groups or human groups, simply because they are the oldest human group, there is perfectly normative and benign variation in the African genome not present in the European genome that then makes medical genetic efforts more difficult in trying to diagnose people of African ancestry or of non-European ancestry more broadly. So to this point, with our COVID pause, we first focused our sequencing efforts on 99 trios.
ELISE ROBINSON: And of the 75 trios analyzed from South Africa, we were able to identify a causal gene in nearly 20% of them. And in Kenya, again, these are small samples, but about 40%, which is about the rate expected for a sample that's predominantly intellectual disability, which this Kenyan sample is. The South African sample is primarily autism but it's severe autism. And one of the unique things about NeuroDev is certainly not just its genetic aspects but the fact that in different communities and different cultures, not just around the world but within the United States as well, but in this case in Kenya and South Africa, the type of autism that's on average ascertained is much more similar to that which was ascertained in the US about 30 years ago, on average, it's much more severe. And this is more the case in Khaleefa County, which is a rural area of Kenya, than it is in Cape Town, for sure. But at the same time, [Cursi Donnel?], our South African PI's clinic has both formal and kind of informal severity thresholds that are needed to be seen. And since we primarily recruit from that clinic, we're primarily recruiting very behaviorally severe autism, both with and without intellectual disability but that which is behaviorally severe. And in speaking about ASD heterogeneity, the fairly low syndromic rate we found is a reminder that behaviorally severe autism or very impairing forms of autism are not necessarily the same thing as genetically syndromic autism, which is something that one can lose sight of in kind of the more modern US-based cohorts.
ELISE ROBINSON: Another joy of NeuroDev is we've been able to participate in a clinical genetics program called Matchmaker Exchange, which was built by our colleagues at the Broad. We submitted 14 candidates to Matchmaker Exchange. That is basically variance in children from the first nearly 10 NeuroDev trios, variants that were possibly causal but we weren't sure. There wasn't a strong enough literature base or clinical genetic base to say so. But you can submit them to Matchmaker Exchange and other clinical geneticists and genetic researchers from around the world are also submitting their-- I don't know if this variant is significant but it looks curious, sorts of genetic data along with the relative phenotype data for the case. And through that, we've already been matched with seven groups working on the description of novel genetic syndromes. And so the team is collaborating on all of those in-process publications. And with that, I think we are at 3:45. So thank you so much to the people in purple in my lab who did that beautiful work, to the Autism Sequencing Consortium for letting me share theirs, and to my many wonderful collaborators.
ALEXANDER DENKER: Thank you so much, Dr. Robinson. So we have many questions coming in. So I will do my best to combine and get through them. One of the first questions, can you go back a little bit and tell us more about the criteria that separate intellectual disability from autism and how those items are parsed?
ELISE ROBINSON: Well, sure. So there's a very big phenotypic difference on average between the two. So intellectual disability historically has been defined primarily by IQ-measured cognitive ability that's more than two standard deviations below the population mean. Clinically, it's typically ascertained through a variety of things. Those could include global developmental delays. So, for example, a child who's very young could be diagnosed with a global delay because they aren't sitting up, they aren't walking, they aren't meeting standard milestones. But classically, intellectual disability means cognition measured to standard deviations below the general population mean. Cognition and autism can range from very low to extremely high. The diagnostic criteria for ASD are social and communicative and behavioral. They don't have an inherent cognitive component but the social, behavioral and communication phenotypes that are associated with ASD are more commonly found in people who have intellectual disability than the people who have measured IQs in the middle of the curve.
ALEXANDER DENKER: Thank you. What are your expectations about the mapping of various rare and common ASD risk combinations onto clinical phenotypes such as ASD, cognitive features, but also related psychopathology? And if assortative mating might boost the clinical significance if it's enriched with common or rare risk combinations.
ELISE ROBINSON: So I think the first part of that about the mapping of genotype to phenotype, I think that very few things are going to be specific. So I don't believe we're going to find many genes that just create risk for autism or just create risk for schizophrenia. And to this point, we haven't really found evidence that that's going to be common at all. If we had infinite sample size, I guess there would be some degree of non-specificity identified for every neuropsychiatrically involved gene. That being said, I do think we're going to continue to identify average differences in phenotypic affinity, if you will. We'll continue to, for example, within autism, be able to better clarify genes that create risk for ASD that are more likely to create risk for schizophrenia and those that aren't really very likely to create risk for schizophrenia. Which is, again, not to say there's going to be specificity but I think we're going to be able to come up with different average phenotypic preferences for genes and that those different average preferences could be quite informative when linked to biology. Regarding assortative mating, I'm not sure if I understood the between common and rare part. We don't usually think about rare very much when it comes to assortative mating because the vast majority of rare variants that are related to ASD and intellectual disability at this point are de novo, meaning they aren't inherited. But when it comes to common variation, we don't really know.
ELISE ROBINSON: We don't have a sense very clearly of the extent to which that is occurring for a trait like autism at kind of a meaningful genetic level across the genome. And then to the extent it is how much it would influence phenotype or understanding of genetic architecture.
ALEXANDER DENKER: How do you think we might design a study to get at the gene environment effect implied in the hypothesis that females can carry more genetic risk? Because, as you mentioned, the socialization of many females is often different and making for a protective effect.
ELISE ROBINSON: That's a good question. So, again, there's no certainty here. I can provide a little bit more of an explanation of why I think there's probably at least two different things going on. The first, just across all sorts of developmental disorders and genetic risk factors that create risk for developmental disorders, there is commonly observed resilience in females, even in controls. And in the general population, you see a higher risk or a higher rate of certain sorts of intellectual disability risk creating rare variants in female controls compared to male controls. So there does seem to be a sort of neurodevelopmental resilience that we don't understand associated with being female. Whether that is all that's happening in ASD or if it's-- which I don't think it is because the sex ratios is obviously much bigger in ASD than it is in intellectual disability. But we don't know how much of the protective effect is biological versus differences in how the phenotype might be behaviorally expressed.
ALEXANDER DENKER: Thank you. One of our viewers points out that a new study shows that SPARK genes in individuals without ASD appear to reduce intellectual function. Does that not imply that even large effect ASD associated genes contribute to intellectual disability?
ELISE ROBINSON: It does, yeah. So ASD associated genes, which are associated predominantly through rare variants, de novo variants that eliminate the function of one copy are overwhelmingly on average associated with intellectual disability, both in cases and in the general population. It's only the common polygenic gentled genome wide influences that one studies in GWAS that are associated with higher IQ, both in cases and in the general population. But yes, being a carrier of a rare variant associated with ASD, even if one isn't diagnosed with ASD or ascertained for intellectual disability, does on average have a negative effect on cognition.
ALEXANDER DENKER: So the next question starts out with Kristen Gottrie mentioning that she's not sure if this is answerable. So I don't know if I'm doing a terrible job as a moderator by asking this. Anyway, you mentioned that GRIN2B is part of the neuronal receptor complex before birth and is associated with intellectual disability. Do you think imitation in GRIN2B results in improper brain development and this allows to sort of chance for there to be a treatment versus GRIN2A which is expressed in individuals postnatally? And do you think there is any chance of fixing these de novo mutations and improving life for these individuals? I guess a lot of this requires quite a bit of speculation.
ELISE ROBINSON: Oh, we don't know at all. Actually, Josh and I were talking about that very topic right before meeting. And but obviously it's an area of tremendous interest. If one could intervene, how early would you have to for it to be meaningful and effective and does that vary between different types of genetic syndromes and can we guess how early it has to be based on average when that syndrome makes itself apparent? But for our next kind of phase of genetic intervention considerations, it's the kind of critically important question.
ALEXANDER DENKER: Given the phenotypic heterogeneity in ASD, do you think there's a future for trait-based GWAS as opposed to case control GWAS in ASD research?
ELISE ROBINSON: There is and there are many such things ongoing. And there is an entire NIMH initiative known as RDoC focused on such things that recently had a intellectually vigorous discussion about that very question. So in brief, traits of some disorders are more genetically related to the diagnosis that may exist at the tail of that trait distribution than others. So ADHD has an extremely strong genetic correlation between the diagnosis and the quantitative distribution of ADHD-related traits in the general population. In autism, it's quite hit or miss. If you look at autistic traits rated in young children, it's about 30 to 40 percent but it drops precipitously as autistic traits skills are used in older adolescents and then adults, which is likely mostly reflective of measurement. And the trouble with using trait scales, they have enormous benefits for some things, and I'm certainly a believer in underlying continua, but I'm also a realist in that it's really hard to measure quantitative traits of behavioral variation at least if hard is compared to how easy it is to measure height or weight or something like that. There's hard things to measure.
ALEXANDER DENKER: Thank you. We have time for one more question, I think. So I'm just looking through them. Of the genes that you presented for ASD, do you think any are convincingly associated only with the ASD and not intellectual disability?
ELISE ROBINSON: No, no. I think in current sample size, we can't say that. Again, I think if we were to infinitely expand sample size, even beyond the number of humans currently on the planet, we'd be able to, with statistical clarity, identify genes associated with ASD that are in definable bins in terms of their average co-relationship with intellectual disability. But I don't think we're going to get to a level of specificity where you see a gene that, in the context of infinite sample size, is associated to autism and nothing else. That being said, since we're never going to get to infinite sample size, I guess I can't be proved wrong. So that's just a guess.
ALEXANDER DENKER: Okay. Well, thank you so much for joining us today and providing your expertise. And thank you to everyone who attended. We look forward to seeing you at the next NIMH Innovation Speaker Series talk. Thank you so much.