Aline Lefebvre1, Anita Beggiato1, Thomas Bourgeron2, Roberto Toro3. 1. Human Genetics and Cognitive Functions Unit, Institut Pasteur, Paris; Department of Child and Adolescent Psychiatry, Assistance Publique-Hôpitaux de Paris, Robert Debré Hospital. 2. Human Genetics and Cognitive Functions Unit, Institut Pasteur, Paris; Unité Mixte de Recherche 3571, Genes, Synapses and Cognition, Centre National de la Recherche Scientifique, Institut Pasteur, Paris; Human Genetics and Cognitive Functions, University Paris Diderot, Sorbonne Paris Cité, , Paris; Foundation Fondamentale, Créteil, France. 3. Human Genetics and Cognitive Functions Unit, Institut Pasteur, Paris; Unité Mixte de Recherche 3571, Genes, Synapses and Cognition, Centre National de la Recherche Scientifique, Institut Pasteur, Paris; Human Genetics and Cognitive Functions, University Paris Diderot, Sorbonne Paris Cité, , Paris. Electronic address: rto@pasteur.fr.
Abstract
BACKGROUND: Patients with autism have been often reported to have a smaller corpus callosum (CC) than control subjects. METHODS: We conducted a meta-analysis of the literature, analyzed the CC in 694 subjects of the Autism Brain Imaging Data Exchange project, and performed computer simulations to study the effect of different analysis strategies. RESULTS: Our meta-analysis suggested a group difference in CC size; however, the studies were heavily underpowered (20% power to detect Cohen's d = .3). In contrast, we did not observe significant differences in the Autism Brain Imaging Data Exchange cohort, despite having achieved 99% power. However, we observed that CC scaled nonlinearly with brain volume (BV): large brains had a proportionally smaller CC. Our simulations showed that because of this nonlinearity, CC normalization could not control for eventual BV differences, but using BV as a covariate in a linear model would. We also observed a weaker correlation of IQ and BV in cases compared with control subjects. Our simulations showed that matching populations by IQ could then induce artifactual BV differences. CONCLUSIONS: The lack of statistical power in the previous literature prevents us from establishing the reality of the claims of a smaller CC in autism, and our own analyses did not find any. However, the nonlinear relationship between CC and BV and the different correlation between BV and IQ in cases and control subjects may induce artifactual differences. Overall, our results highlight the necessity for open data sharing to provide a more solid ground for the discovery of neuroimaging biomarkers within the context of the wide human neuroanatomical diversity.
BACKGROUND:Patients with autism have been often reported to have a smaller corpus callosum (CC) than control subjects. METHODS: We conducted a meta-analysis of the literature, analyzed the CC in 694 subjects of the Autism Brain Imaging Data Exchange project, and performed computer simulations to study the effect of different analysis strategies. RESULTS: Our meta-analysis suggested a group difference in CC size; however, the studies were heavily underpowered (20% power to detect Cohen's d = .3). In contrast, we did not observe significant differences in the Autism Brain Imaging Data Exchange cohort, despite having achieved 99% power. However, we observed that CC scaled nonlinearly with brain volume (BV): large brains had a proportionally smaller CC. Our simulations showed that because of this nonlinearity, CC normalization could not control for eventual BV differences, but using BV as a covariate in a linear model would. We also observed a weaker correlation of IQ and BV in cases compared with control subjects. Our simulations showed that matching populations by IQ could then induce artifactual BV differences. CONCLUSIONS: The lack of statistical power in the previous literature prevents us from establishing the reality of the claims of a smaller CC in autism, and our own analyses did not find any. However, the nonlinear relationship between CC and BV and the different correlation between BV and IQ in cases and control subjects may induce artifactual differences. Overall, our results highlight the necessity for open data sharing to provide a more solid ground for the discovery of neuroimaging biomarkers within the context of the wide human neuroanatomical diversity.
Authors: Jason J Wolff; Guido Gerig; John D Lewis; Takahiro Soda; Martin A Styner; Clement Vachet; Kelly N Botteron; Jed T Elison; Stephen R Dager; Annette M Estes; Heather C Hazlett; Robert T Schultz; Lonnie Zwaigenbaum; Joseph Piven Journal: Brain Date: 2015-05-03 Impact factor: 13.501
Authors: Yuliya N Yoncheva; Krishna Somandepalli; Philip T Reiss; Clare Kelly; Adriana Di Martino; Mariana Lazar; Juan Zhou; Michael P Milham; F Xavier Castellanos Journal: J Am Acad Child Adolesc Psychiatry Date: 2015-11-26 Impact factor: 8.829
Authors: Carolyn Klocke; Joshua L Allen; Marissa Sobolewski; Jason L Blum; Judith T Zelikoff; Deborah A Cory-Slechta Journal: Neurotoxicology Date: 2017-10-24 Impact factor: 4.294