A new paper in Nature reports the results of a large genome-wide association of autism. After some fairly heroic data analysis, the researchers have managed to tag one region of the genome as containing a common variant that contribute to the disease, with odds ratios on the order of 1.2 (the paper actually reports six variants in the same region, but these all appear to be tagging the same underlying causal variant).
The finding is getting fairly glowing press coverage, but let's keep it in context: an odds ratio of 1.2 means that individuals carrying the variant have their risk of the disease increased by around 20%; their probability of actually suffering from autism is still much less than 1%. This is not a finding that will lead to powerful new diagnostic tools for the early detection of autism, and it's consistent with a generally fairly disappointing yield from studies of common variants in psychiatric diseases.
While the new variant is unlikely to have much utility for risk prediction, it should provide some insight into the molecular mechanisms of autism (and there is some good preliminary work in the paper heading in this direction). However, the biggest lesson from the paper is simple: the hunt for common variants underlying psychiatric diseases remains largely an exercise in frustration. This was a whole lot of work for a very slender yield in terms of potential causative variants.
Several recent studies (including one from the same group in the very same issue of Nature) have had much better luck looking at a different type of genetic variant: rare but large insertions or deletions of DNA, known as copy number variants. It seems likely that other types of rare variants - much smaller events, largely invisible to current genome scan approaches - will play an important role in the etiology of this disease.
This study was certainly worth performing, and the mechanistic work on the associated region will be interesting to follow - but autism may be one case where the advice of David Goldstein to abandon genome-wide association studies in favour of large-scale sequencing may be worth following as soon as possible.
Coverage elsewhere: Ed Yong has a characteristically thorough run-down on the paper, expanding on a number of important points I haven't covered here.
(As an aside: New Scientist goes with the somewhat misleading headline "Gene variant found in 65% of autism cases", which appears to be based on a quote from lead author Hakor Hakonarson (whose back story is an intriguing tale in itself). That headline might give the impression that this variant actually causes 65% of autism cases, which is absolutely not the case: if you read a little further down you will discover that the variant is also present in 60% of healthy controls.)
Uhhh... New Scientist, that's a terrible headline. I thought the 65% figure must be dodgy in some way...
And to put New Scientist in perspective, there's the Daily Mail who led with massive front page letters saying "IS THIS THE KEY TO AUTISM?"
No. No it isn't. It's interesting, but no. Now shush.
Ed - oh, my aching brain...
I wouldn't say this is a disappointing result. Nobody doing a GWA study these days should reasonably expect to see results indicating an odds ratio much higher than this.
I guess the only people that will be disappointed are those who see sensational headlines for this paper then realize that it got very average results.
I have read the supplementary data published for the article:
Male female ratios in autism has been reported as between 3:1 to 4:1 with some recent reports suggesting the male female ratio is even higher (9:1 for Asperger Syndrome).
No male female ratio was reported for the AGRE cohort. In the ACC cohort the authors reported a high male female ratio (Males=83% Females=17%), consistent with the high male female ratio in autism. In the control group of 7,000 individuals, the authors reported a near equal male female ratio (Males=52% Fwmales=48%)
The supplementary data did show that a male only autism group and a male only control group was selected to look for common genetic variants, but only for the Y chromosome scan.
For the genome wide scan, the results were not segregated by gender. Since common (and rare)genetic variants often tend to occur more frequently in males than females and vice-versa, do you consider the control goup in this study to be representative of the autistic group.
Perhaps I am missing something as I do not have any pretexts about understanding anything about the complexity of the human genome. Any comments would be appreciated.
I'm a little confused - what did you mean by this?
Since common (and rare)genetic variants often tend to occur more frequently in males than females and vice-versa...
Did you have specific examples in mind?
Perhaps I was a little unclear. There are any number of common and rare genetic variations that occur more or less frequently depending on gender. A few examples:
Type 1 Diabetes:
Spanish schizophrenic patients:
The autism study did not control for gender even though the male:female ratio in autism is between 4:1 to 9:1 hence the control group may not be a representative control. The gender difference between the autism group (Males = 83%) and controls (Males = 52%) was so large and the difference for 5P14 variants between autistics (65%) and control (60%) was so small as to question any interpretation given.
There is another common genetic variation that still has the highest risk for autism than any common genetic variation that has been proposed, the presence of a 'Y' chromosome.
Genome Wide studies should control carefully for both ethnicity and gender or any other variable that may be informative.
The next time you hear about a 'new autism gene discovered' headline, you might want to consider the AGRE database containing several thousand families with two or more autistic children. The AGRE data has been accumulated for more than a decade. The data that was acquired and placed in the database does not include IQ scores and was the primary source for the 5P14 study. Between 25 to 80% of autistic children are mentally retarded (depending on the institutes publishing the data and the concept of autism held by various groups).
A year ago similar headlines were produced by a discovery of genetic mutations in the region of 16P.1. The authors claimed that the discovery accounted for at least 1% of all autism cases.
This report has not been replicated:
The difference between these two studies is that the first one used the AGRE data set and the report that failed to replicated the original finding did not. The original study did look at another group at the Boston Children's Hospital, but the group included children with developmental delay or mental retardation that did not meet criteria for autism.
One of the strongest risks for autism is the presence of mental retardation. In fact, a number of mental retardation syndromes are also associated with autism (Fragile X, Tuberous Sclerosis, Downs Syndrome) with a significant minority possessing enough isolated secondary symptoms to qualify for an 'ASD' diagnosis.
The AGRE data set may include children with genetically influenced mental retardation, but by definition, includes only the small subset that qualifies for an autism diagnosis. When researchers claim to have identified an autism susceptability gene, what they may be looking at is a previously unrecognized susceptability gene for mental retardation and other nonautism developmental problems. We will never know since IQ scores are not avaiable in the AGRE data base.
Quality research in autism controls for age, IQ and gender. The 5P study only controlled for European ancestry.