ABSTRACT
The corpus callosum – the main pathway for long-distance inter-hemispheric integration in the human brain – has been frequently reported to be smaller among autistic patients compared with non-autistic controls. We conducted a meta-analysis of the literature which suggested a statistically significant difference. However, the studies included were heavily underpowered: on average only 20% power to detect differences of 0.3 standard deviations, which makes it difficult to establish the reality of such a difference. We therefore studied the size of the corpus callosum among 694 subjects (328 patients, 366 controls) from the Abide cohort. Despite having achieved 99% power to detect statistically significant differences of 0.3 standard deviations, we did not observe any. To better understand the neuroanatomical diversity of the corpus callosum, and the possible reasons for the previous findings, we analysed the relationship between its size, the size of the brain, intracranial volume and intelligence scores. The corpus callosum appeared to scale non-linearly with brain size, with large brains having a proportionally smaller corpus callosum. Additionally, intelligence scores correlated with brain volume among controls but the correlation was significantly weaker among patients. We used simulations to determine to which extent these two effects could lead to artefactual differences in corpus callosum size within populations. We observed that, were there a difference in brain volume between cases and controls, normalising corpus callosum size by brain volume would not eliminate the brain volume effect, but adding brain volume as a covariate in a linear model would. Finally, we observed that because of the weaker correlation of intelligence scores and brain volume among patients, matching populations by intelligence scores could result in a bias towards including more patients with large brain volumes, inducing an artificial difference. Overall, our results highlight the necessity for open data sharing efforts such as Abide to provide a more solid ground for the discovery of neuroimaging biomarkers, within the context of the wide human neuroanatomical diversity.
