Friday, December 27, 2013

Autistic brains: Under- or over-connected?

In Garcia Dominguez et al.'s EEG study, children with autism had increased "coherency" between electrodes overlying the visual cortex  Source

Note: A Spanish translation of this post can be found at Autismo Diario: Cerebros con Autismo: ¿Sub o Hiper Conectados?

For a while now, the prevailing view has been that the brains of people with autism are "underconnected" and that this lack of communication between different parts of the brain causes at least some of the features of autism. It's an intuitively plausible idea but it was always likely to be an over-simplification.

Recently, a number of studies have pointed to the exact opposite - an increase in connectivity in autism. The emerging story is that kids with autism have "hyper connected" brains but they become less connected as they get older.

But this is still just a story. It takes the new brain imaging data, which comes from kids, and tries to reconcile it with the older brain imaging data which mainly involves studies of adults. There are, however, lots of other differences between studies in terms of the methods they used and, perhaps most importantly, the way the data were analysed.

On top of that, we still don't really understand what many of these "connectivity" measures actually signify. And it's not really clear how changes in connectivity might relate to the development of autism at the behavioural level. Certainly, people with autism change as they get older, but I don't think anyone would argue that autistic adults are the opposite of autistic kids. It's a fascinating area of research but there are still a lot of holes in the plot.

Earlier this month I contributed to a "Cross Talk" feature on the SFARI website, in which Damien Fair, Tal Kenet, John Rubenstein, and I gave our thoughts on an earlier post by Vinod Menon, discussing precisely these issues.

I was also interviewed by Emily Anthes for a SFARI article on two recent papers providing evidence for hyper connectivity in autism: an fMRI study by Kaustubh Supekar and colleagues at Stanford; and an EEG study by Luis Garcia Dominguez and colleagues in Toronto.

Inevitably, my fairly lengthy answers to Emily's questions were edited down to a couple of short quotes. So, with Emily's permission, I'm posting the full email Q&A. I'd be very interested in other people's thoughts.

Regarding the Supekar et al paper: The researchers's use of three independent cohorts seems impressive. Is that notable/unusual?
This is a really important step, particularly as there are often conflicting findings across studies. It shows that the effect is robust and not just a quirk of the particular sample of children being tested. However, we shouldn't interpret the fact that there are group differences in all three studies as meaning that this is true of all people with autism - or even all people with autism in the studies. 

Also regarding the Supekar paper: What do you think of the suggestion that connectivity may change as children with autism age? Could that help explain some of the contradictory findings in the literature?
It's possible although I think we need a direct comparison with data from adults analysed in exactly the same way as the data from children. Given their argument, I'm surprised Supekar et al didn't look to see if there was a correlation with age within the study. If there really is a change from hyperconnectivity in young children to underconnectivity in adolescents and adults then we would expect to see that hyperconectivity is most pronounced in the youngest children. That wouldn't itself be conclusive evidence - we'd need a longitudinal study to test the idea properly, but it would be the obvious starting point. 

The PLoS One paper uses EEG to study connectivity--is this unusual? Does EEG have any particular advantages or disadvantages over MRI when it comes to studying connectivity?
To me it makes much more sense to use EEG or MEG to look at connectivity. There are lots of studies that have done this before although like the fMRI data the results are quite messy and contradictory. The big advantage of EEG and MEG is that they have very good time resolution so can pick up on the rapid and dynamic changes in connectivity that are involved in cognitive processes. 
While people often talk about fMRI connectivity in terms of "communication" between the different parts of the brain, it's more accurate to talk about "coactivation". We're really just talking about brain regions increasing and decreasing their activation together at the roughly the same time, roughly once every second or so. In fact, Supekar et al. make a point of filtering the data to look only at very low frequency changes (0.01 to 0.05 Hz) - so they're looking at the extent to which different brain regions are coactivated at the timescale of minutes. 
One of the problems with EEG is that the signal from any part of the brain is picked up (to different degrees) by all of the electrodes on the scalp. If you find evidence for synchronization between the response at two different electrodes, it's tempting to assume that this means that the parts of the brain underneath the two electrodes are acting together in synchrony. But it could just be that the two electrodes are measuring the same activity from the same source in the brain. The Dominguez paper uses a relatively new mathematical technique that is claimed to eliminate this problem. My understanding is that it effectively looks at the parts of the brain response that are nearly but not perfectly synchronized (on the assumption that perfect synchronization is almost certain to reflect the same brain response being measured at two locations).

Going forward, how can we help clear up some of the contradictions in the literature?
Researchers need to use the same techniques across different populations. They also need to subject data from the same samples to different analyses. That way we can start to work out how much of the contradictions are due to differences in participant characteristics (eg age, autism severity) and how much is down to the different methods and analyses being used by different research groups. We also need to look at individual as well as group differences - and for that we need to know how reliable these measures are (ie if you test the same person twice, do you get the same result).


García Domínguez L, Stieben J, Pérez Velázquez JL, & Shanker S (2013). The imaginary part of coherency in autism: differences in cortical functional connectivity in preschool children. PloS one, 8 (10) PMID: 24098409

Supekar K, Uddin LQ, Khouzam A, Phillips J, Gaillard WD, Kenworthy LE, Yerys BE, Vaidya CJ, & Menon V (2013). Brain hyperconnectivity in children with autism and its links to social deficits. Cell reports, 5 (3), 738-47 PMID: 24210821

SFARI links:


  1. Common genetic polymorphisms may contribute to the disruption of synapse formation and under or over connectivity. Connectivity studies in ASD compare ASD and TD controls. None included a control a group of unaffected siblings. In schizophrenia, Whitfield-Gabrieli et al (2009) examined under and over connectivity in schizophrenia patients, unaffected siblings and TD controls. Schizophrenia patients and their siblings presented with the same connectivity issues present but relatively milder in unaffected sibling controls in contrast to typically developing controls. ASD and Schizophrenia are multifactorial disorders and including unaffected parents and siblings is warranted. The over and under connectivity may not be related to strict autism but may reflect the brain connectivity phenomena in the broader autism phenotype often seen in unaffected family members.
    Connectivity issues are also found in bi-polar. Under and over connectivity seems to be a feature of all the psychiatric conditions thus can’t be predictive of outcomes. First degree relatives need to be included in all these studies that might offer greater insight. So far, only in schizophrenia have first degree family members been included fMRI studies. Connectivity issues may be a familial phenomenon found in both affected and to a lesser degree in unaffected parents and siblings.
    Under or over connectivity in ASD, bi-polar disorder and schizophrenia may be a background familial effect that is always reliant on other risk factors including other genetic influences, inherited and/or de novo, environmental risk factors and epigenetics that cumulatively are involved in the transition to strictly defined ASD, bi-polar disorder and schizophrenia.
    Chen et al (2011). A quantitative meta-analysis of fMRI studies in bipolar disorder. Bipolar Disord. 2011 Feb;13(1):1-15. doi: 10.1111/j.1399-5618.2011.00893.x
    Whitfield-Gabreili et al (2009). Hyperactivity and hyperconnectivity of the default network in schizophrenia and in first-degree relatives of persons with schizophrenia. Proc Natl Acad Sci USA. 2009 Jan 27;106(4):1279-84. doi: 10.1073/pnas.0809141106. Epub 2009 Jan 21.

  2. What's your take on this review article? which suggests that local connectivity is high while long-distance connectivity is low?

    Resting state EEG abnormalities in autism spectrum disorders

  3. I certainly agree to some points that you have discussed on this post. I appreciate that you have shared some reliable tips on this review.