There has been a lot of concern in the autistic community around a new study, Spectrum 10K, which aims to collect DNA samples from 10,000 autistic people. I have been keen to use my position as a scientist (not a geneticist, and not involved in this specific study) to support the autistic community. I’ve been focused on actions that put as much power into autistic people’s hands as possible, and don’t centre my own perspective or voice.
One barrier for many members of the autistic community, in considering the Spectrum 10k study, is the relative inaccessibility of genetics research. Both in terms of academic papers not always being free to read, but also in terms of jargon and technical language.
To that end, I am taking a recent pre-print* with many authors from the Spectrum 10K team and providing a lay summary of its abstract, as well as a glossary of key terms. I hope this will help people outside academia feel more informed about the “state of the art” in terms of genetic research into autism. Thanks to @AutSciPerson who drew my attention to this paper and wrote their own thread about it.
The pre-print* is available at this link. There is a long list of authors – but the first and last listed names (indicating leadership positions in the research team) are both people who are also part of Spectrum 10k. We can therefore reasonably assume that the work presented here is indicative of the work that is planned as part of Spectrum 10K.
In my lay summary I have tried to stay as close as possible to the original language while losing the jargon. I have used “they” throughout, because I was not one of the researchers involved in this study. Where I have added more detail, sometimes this is based on looking for a bit more detail in the paper itself (e.g. finding out what is meant by “associated features of autism”) or based on standard definitions or practices in clinical research (e.g. in defining “intellectual disabilities”).
However, in other places I have made quite lengthy additions, when I think it is hard to understand the original without a detailed description and / or metaphor. When it comes to the results, I have tried to add a bit of value-neutral detail about what they might mean, and have labelled these as “extrapolations” to make it clear that this is my interpretation of the science. Again, these are intended purely to help readers understand what’s been done as clearly as possible, and make their own judgements about that.
NB: I hope the metaphors work, I find them useful to understand complex science but I know sometimes inadvertent implications (in this case, from comparing autistic people to cakes) can arise and I can only apologise if these cause any harm.
Title: Genetic correlates of phenotypic heterogeneity in autism
My summary of the abstract:
The big differences between autistic people in terms of their clinical profile (i.e. features that we can measure with questionnaires or table-top tasks, like IQ tests) make it harder to understand the genetic causes of autism. To address this gap, they investigated genetic differences between autistic individuals. Their maximum sample size was 12,893 people (but in some of their statistical analyses there were fewer people included). They related genetic information from this sample of autistic people with four aspects of the clinical profile of autism:
(1) core features used in clinical diagnosis, which were “social communication difficulties” and “restricted and repetitive behaviours”
(2) associated features of autism, which were: i) general developmental level across multiple domains, including social, physical, communication and daily living skills, measured using a parent-report questionnaire; ii) IQ, overall and also split into verbal (mostly language based) and non-verbal (mostly visual & spatial based) scores; iii) features of dyspraxia
(3) co-occurring developmental disabilities, which were i) language disabilities, including being mute, and having physical difficulties that impede speech ii) disabilities related to physical development including developmental co-ordination disorder; and iii) intellectual disabilities, normally defined as having an IQ < 70.
(4) sex, which in this study could be male or female
They put all the data they had from measures of “core features” into something called a “factor analysis”. This kind of analysis looks for how much two different questionnaires (or other sources of data) seem to be measuring the same underlying thing. When two questionnaires seem to be measuring much the same underlying thing, one can say they “load” onto the same “factor”.
You could think of this a bit like classifying paint colours. You might have scarlet, vermillion and cerise, and they could be grouped into a factor and you could call that factor “Red”. Ochre, mustard and lemon might make a factor called Yellow, and violet, mauve and lavender could make a factor called Purple. We have taken nine individual sources of data – nine paint colours – and by analysing them we have created three factors – Red, Yellow, Purple – that describe all the available colours. This description has less detail, but it is also useful when you want to take a lot of fine-grained information and analyse it, without analysing every item individually.
Their factor analysis produced six factors. This means that for the data they were looking at there were six clumps of information – six factors – which provided a good summary of all the original data.The six factors they used to describe the original data on clinical profiles were labelled as follows: 1. Insistence on sameness; 2. Social interaction at age five; 3. Sensory-motor behaviour; 4. Self-injurious behaviour; 5. Idiosyncratic repetitive speech and behaviour; 6. Communication skills.
Then, they looked at whether genetic information was related to these factors. There are some “common genetic variants” related to autism – these are patterns in people’s genes which occur more often than chance, in autistic people. If you count up all of these common variants, you can get a “polygenic score” – basically the more and more of these genetic ingredients you have, the more likely you are to be autistic. They found that these common genetic variants and polygenic scores were indeed related to the six factors.
You can think of this a bit like trying to guess what someone is cooking from the ingredients in their shopping bag. But instead of buying those ingredients from a shop, they inherited them from their parents. If they have eggs (a common genetic variant), you might think there’s a chance they’re making a cake (a chance they are autistic), but it could also be an omelette or a quiche (not autistic). If they have eggs and sugar, you think there is more chance they are making a cake. If they have eggs and sugar and baking powder and flour, you might be pretty confident they are making a cake. BUT still, all of these ingredients are also used in other recipes, so you still don’t know for sure. This is a bit like polygenic scores – the more genetic ingredients of autism you have, the more likely it is that you will be autistic. In this analysis, they looked at the relationship between the genetic ingredients and the final recipe – not just “are you making a cake or not” (i.e. are you autistic or not?) but “what kind of cake are you making?” (i.e. what sort of clinical profile does this autistic person have?)
They also looked at whether their six factors were related to “de novo” variants. These are changes in the genetic material that arise for the first time in a child, unpredictably, instead of being inherited from a parent. These “de novo” variants are described as “high-impact”, which I think means that people who have these variants are autistic but also have a lot of co-occuring conditions. In fact, “de novo” variants are really only discovered when someone has a diagnosis first (e.g. autism + epilepsy + intellectual disability), and then doctors decide to look for a possible genetic cause. To continue my shopping & ingredients metaphor, “de novo” variants are more like someone has a whole shop-bought cake, instead of a load of ingredients, in their bag.
Anyway, the researchers found that their six factors were NOT related to these kinds of genetic ingredients. This is even though, when they compared the six factors themselves between participants with de novo genetic variants or common genetic variants, the two groups looked about the same. So even though autistic people with de novo variants, and autistic people with common genetic variants, have the same patterns of clinical features (as measured in this study), in the first group, their specific genetic profile doesn’t systematically relate to their clinical profile. In the second group, their specific genetic profile does systematically relate to their clinical profile. So [I am extrapolating here] a genetic screen would tell me more about the clinical profile of an autistic person with a range of common genetic variants, than it would if they had a de novo “high-impact” genetic variant.
Specifically, the more of these common genetic variants someone has, the less likely they are to have a co-occuring developmental disability, of the kinds measured in this study (see above)….”which reflects both a true protective effect and additivity between rare and common variants.” I think this part means that the common genetic variants associated with autism actively make it less likely that you would have a developmental disability at the same time.So [I am extrapolating again] I guess that a lot of the co-occurrence of autism and developmental disability might be driven by people with de novo variants.
Also, for autistic people who don’t also have an intellectual disability, autistic females (their analysis is based on sex, not gender) have more of these common genetic variants than males.
And finally, the amount of variability in the clinical features measured, that is apparently explained by genetics, is larger for males, and for autistic people without an intellectual disability. So the relationship between genes and clinical profile is stronger for these two groups. In other words [another extrapolation by me], if I knew your genetic profile I could more confidently guess your clinical profile if you were male and / or had an intellectual disability. If you were female – despite the fact above that women have more genetic variants, or perhaps even because of it – then I would make a less accurate guess about your clinical profile based on your genetic information. However, one type of clinical profile (according to the six factors found at the start of the study) is not more explicable at a genetic level than another.
NB: I am not suggesting that it is possible to do this – to get someone’s genetic information and guess their clinical profile. I am just trying to translate the statistical findings into language that is more accessible. But all of these associations are actually operating at a group level in a big sample, and don’t imply precision at an individual level.
The authors end by saying that more detail on those clinical profiles – more questionnaires, more direct assessments with autistic people (my extrapolation: perhaps measuring stuff like IQ, sensory profile, and co-occuring diagnoses such as epilepsy, Ehlers-Danlos syndrome, ADHD) – is needed to be able to understand how genetic variance (differences between people’s genes) explains differences in thinking processes, behaviour and whether someone has another diagnosis as well.
The original abstract:
The substantial phenotypic heterogeneity in autism limits our understanding of its genetic aetiology. To address this gap, we investigated genetic differences between autistic individuals (Nmax = 12,893) based on core (i.e., social communication difficulties, and restricted and repetitive behaviours) and associated features of autism, co-occurring developmental disabilities (e.g. language, motor, and intellectual developmental disabilities and delays), and sex. We conducted a comprehensive factor analysis of core autism features in autistic individuals and identified six factors. Common genetic variants including autism polygenic scores (PGS) were associated with the core factors but de novo variants were not, even though the latent factor structure was similar between carriers and non-carriers of de novo variants. We identify that increasing autism PGS decrease the likelihood of co- occurring developmental disabilities in autistic individuals, which reflects both a true protective effect and additivity between rare and common variants. Furthermore in autistic individuals without co-occurring intellectual disability (ID), autism PGS are overinherited by autistic females compared to males. Finally, we observe higher SNP heritability for males and autistic individuals without ID, but found no robust differences in SNP heritability by the level of core autism features. Deeper phenotypic characterisation will be critical to determining how the complex underlying genetics shapes cognition, behaviour, and co- occurring conditions in autism.
Rather than attempt my own I found this excellent online glossary of genetics terms, which also includes audio clips of each term’s definition. Their definition of autism might not be perfect, but for the technical jargon of genetics research in general I think it can’t be matched.
- A “pre-print” is a manuscript which has been posted up by the authors before undergoing peer review or publication in a specific academic journal. This is increasingly common practice and considered to be a positive thing to do, as it permits widespread access to the science (for free) in a timely manner. Pre-prints are no less high-quality than journal articles, but they do not represent the peer-reviewed final article, which will have been closely examined by experts in the field and adjusted in response to their comments.