Thread by @deevybee

We've recently published a series of papers in @WellcomeOpenRes reporting findings from a study of the impact of an extra X or Y chromosome (ie. trisomy) on development. Data collection took 6 yrs! I'm going to try a mega-thread summarising the main findings

The children we studied included girls with trisomy X (XXX), boys with Klinefelter syndrome (XXY), and boys with an extra Y (XYY). These trisomies affect from 1/600 to 1/1000 births, but often go undetected bcs effects are relatively mild. /1

The reasons why these trisomies don't have more impact is fascinating - I discuss it a bit here: deevybee.blogspot.com/2011/05/x-and-… /(1a)

Why did we do the study? 2 reasons; 1) I'm interested in causes of language disorder and the literature suggested an extra X or Y was a risk factor - that's unusual because most genetic conditions that affect neurodevelopment have more global effects /2

The wonderful Pat Jacobs, who discovered several trisomies in the 1960s (see ncbi.nlm.nih.gov/pmc/articles/P…) urged me to do a study of children identified on prenatal screening. /3

Pat explained that clinical geneticists had to advise parents when a trisomy was discovered but felt they had inadequate information about outcomes. Most info was based on v old studies; Some was based on biased samples /4

The study of sex chromosome trisomies shows up the huge issue of how ASCERTAINMENT BIAS can produce misleading results. In short, the people you get to study are seldom representative of the whole population of interest /5

The biggest bias is if you only study those whose trisomy was discovered in the course of investigations for developmental disorders: obvs you then only find people with developmental disorders! /6

But there are other biases too, which we consider in the Discussion of this paper: wellcomeopenresearch.org/articles/4-32/…: e.g. those identified on prenatal screening will be children of older mothers. People who volunteer to take part may be those with concerns /7

Consistent with previous research, we found an increased rate of language problems in children with XXX, XXY and XYY chromosomes, as reported in this paper: wellcomeopenresearch.org/articles/3-143… /8

However, the average result masks considerable variation from child to child. A lot of the variation was explained by ascertainment bias. But even when such bias was low, the range of outcomes was substantial /9

There are genetic reasons to expect that the picture might be different for boys or girls with an extra X chromosome vs boys with an extra Y, but we found more variation within each trisomy group than between. /10

In reporting the data, I realised it was important to show individual data points, rather than just summary statistics, and I made good use of beeswarm plots for this purpose. Seeing the range this way really alters how you think about the data /11

In a previous study that just relied on parental report, we were surprised to find an increased rate of autism in boys with XXY and XYY, as that hadn't previously been reported. We surmised this reflected changing diagnostic criteria over time /12

In the current study we again found an increase in autism, but this time we could see this in girls with XXX as well as boys with XXY & XYY. Milder autistic features insufficient for a diagnosis were also increased. But, most children didn't have autism. /13

It's also been reported that high rates of social anxiety are found with extra X chrom. We found some support for this, but numbers were too small to be convincing. But again, we need to think of bias: kids with social anxiety may not agree to take part! /14

Our findings on autism and social anxiety are described in this paper: wellcomeopenresearch.org/articles/4-32/… . This has not yet been reviewed so should be regarded as preprint. /15

At this point, I'd like to thank Robert Goodman, who developed the instruments we used for diagnosis and was a pioneer of open science long before it was common. He set standards for epidemiology in child psychiatry /16

Bottom line for main findings on neurodevelopment: all 3 trisomies associated with increased risk of neurodevelopmental disorders, especially those involving language and communication. But huge range makes it hard to predict outcome for a specific child /17

And this is a classic case where the same risk factor (additional X or Y) can have a wide range of different consequences. No simple mapping of genetic constitution on to phenotypic outcome. Much comorbidity too.

In most cases, the child will have some difficulties that impact on school and family life, but most will attend mainstream school. /18

The remaining mystery we'd like to solve is why the outcomes are so variable. In collaboration with geneticist @diannenewbury we looked at this: wellcomeopenresearch.org/articles/3-85/… /19

We tested a 'double hit' theory which proposed that the impact of a sex chromosome trisomy might be enhanced by variation in autosomal genes involved in specific neural circuits /20

We had a problem: potentially the number of analysis we could do was enormous, & sample size was v small for genetic analysis. We preregistered our analyses so that if we found an association, we could have confidence it was not just obtained by p-hacking /21

Planning the analysis took several months. And after all that effort, we obtained null results! But I'd recommend this approach - if you have a specific hypothesis, it helps to think through the analysis b4 you look at the data /22

And we are cool with the null result - a nice idea didn't work out. It's good to get it out in public, if only to prevent others wasting time pursuing a false lead. If we only publish positive results, we get a distorted picture /23

We have other ideas to explain variation, but it may be there's no explanation - we cite some work in yeast that finds that trisomy disrupts development rather randomly. To find out more re random genetic effects, I recommend Innate by Kevin Mitchell. /24

In this area - and with other genetic disorders - it's hard to recruit a big sample for convincing genotype-phenotype analysis. By making results open, we can start to build a bigger picture; potential to combine information across research groups. Fin! /25