A Predicted Brain-Connectivity Pattern in Autism Did Not Hold

What the research found

In 2005, in Current Opinion in Neurobiology, Eric Courchesne and Karen Pierce proposed that the autistic frontal cortex might be talking only to itself: dense, excessive communication between neighbouring patches of tissue, alongside weak communication across longer distances. The proposal fit the anatomical and post-mortem work available at the time, and it became the standard account of autistic brain organisation. The local half of it has never been confirmed by functional imaging.

The measure most often used to test it is regional homogeneity, usually shortened to ReHo. A child lies in the scanner for a few minutes doing nothing in particular, and the analysis asks how closely each small cluster of adjacent brain tissue rises and falls in activity in step with its immediate neighbours. A high score means a well synchronised neighbourhood. Studies applying it to autistic children have contradicted each other for a decade. Several report higher and lower values within the same brain, and the cerebellum and the insula have each been reported in opposite directions by different groups.

Donato Liloia and colleagues at the University of Turin, with a co-author from the Adult Autism Centre of the Turin health authority, took that inconsistency as their subject, publishing in European Archives of Psychiatry and Clinical Neuroscience in January 2023. A systematic search of PubMed to January 2022 returned 1,190 records, of which eight articles met their criteria, supplying eleven independent datasets: 455 autistic children and adolescents, 83 of them girls, and 474 typically developing controls, 110 of them girls. Mean ages were 11.76 and 11.94 years, and no participant was over 18. The pooling method combines the peak coordinates published by each study rather than the original images.

Three regions showed consistently lower regional homogeneity in the autistic groups, occupying 6,160 cubic millimetres in total: the right paracentral lobule, part of the sensorimotor system (Z = 4.708, p = .003); the medial superior frontal gyrus (Z = 5.082, p = .004); and the left posterior cingulate cortex (Z = 3.726, p = .032). Not one cluster of higher regional homogeneity survived statistical correction, although ten of the eleven datasets had individually reported at least one such peak. A bilateral cluster in the visual cortex reached p = .005 and fell away under correction. Heterogeneity between studies was near zero at all three peaks, and Egger’s tests gave no indication of publication bias, so the missing over-connectivity cannot be blamed on a few discrepant samples.

What it means for you and your child

The claim this unsettles is a general one. Local over-connectivity was proposed as a property of the autistic brain at large, and a pooled analysis of the pediatric literature found none of it. About any single region the evidence is thinner, since the authors worked from published coordinates rather than whole statistical maps, and they note that a set of only eleven experiments can bias estimated effects slightly towards zero.

There is a further reason this literature has failed to settle, and the Turin group could only partly address it. They tested whether age, IQ, the proportion of girls, spatial smoothing or slice thickness explained the disagreement between studies, and none of these did. They also observe that regional homogeneity is sensitive to whether participants kept their eyes open or closed during the scan, a variable several of the included studies did not record. Head motion, which is more frequent in autistic children and alters precisely this kind of correlation measure, was not among the variables they were able to examine, and the paper does not report how the original studies handled it.

What the three regions are thought to do comes from a separate step. Two of them, the posterior cingulate and the medial superior frontal gyrus, are core parts of the default mode network, the set of areas most active when a person is not attending to anything external. The Turin group matched their coordinates against a database of more than 15,000 published imaging studies, which associates that territory with mentalising and with autobiographical memory. That association describes what those coordinates are usually reported alongside in the wider literature, and it is not a measurement taken from any child in this analysis.

No imaging measure of local connectivity distinguishes one autistic child from another, and none is used in diagnosis. Whether the reduced homogeneity found here belongs to autism or to the conditions under which autistic children are scanned is not settled by this analysis, and the variable most likely to decide it was never tested. A field that cannot yet separate a finding about autistic brains from a finding about how still a child lay in a scanner has not arrived anywhere a parent needs to follow it.