Autism Spectrum Disorder Liability Is Modulated Along a Gender Continuum from the Female to Male Neuroanatomical Brain Phenotype
Objectives: Here, we aimed to (1) identify neuroanatomical patterns that are characteristic of the male and female brain phenotype, and (2) test the hypothesis that having a male or female brain phenotype represents a risk or protective factor for ASD, respectively.
Methods: 98 right-handed adults with ASD (49 males, 49 females) and 98 matched neurotypical controls (51 males, 47 females) aged 18-42 years were recruited and assessed at the IoPPN, London, and the ARC, Cambridge. A Gaussian Process classifier (GPC) (Rassmussen 2006, Marquand 2010) was initially trained to make probabilistic predictions for each control participant based on binary classes of biological sex using eight surfaced based measures of cortical neuroanatomy (cortical thickness, grey & white matter surface areas, cortical volume, grey:white matter signal intensity ratio, curvature, sulcal depth, and metric distortion). This approach allowed us to derive a probabilistic prediction for each individual based on the binary categories dictated by biological sex, and to represent each brain along a gender continuum with individuals being most confidently classified representing the prototypical male or female brain phenotype. The performance accuracy of each model was estimated using leave-one-out cross validation and tested for significance via permutation of class labels (n=1000). Subsequently, we applied each predictive model to the males and females with ASD in order to estimate probabilistic predictions for individuals with ASD along a gender continuum. ASD liability for each model as a function of neuroanatomical gender phenotype was estimated as the ratio of males (females) with ASD to the total number of males (females) in eight bins along the axis of predictive probabilities of gender.
Results: Across all eight vertex based cortical features, GPC was able to separate male from female controls at accuracies ranging from 68% for metric distortion to 84% for G:W intensity ratio (p<=.001). When predicting individuals with ASD we found that female ASD cases were allocated more frequently to the male category across all eight models (as opposed to the female category), significantly above chance level (p-values ranging from <.001-<.002). When examining the proportion of ASD cases as a function of predictive probability for the male phenotype, we found that increased probabilities for the male brain phenotype were associated with an increased prevalence ratio for ASD within our sample.
Conclusions: Our results support the hypothesis that having a male brain phenotype constitutes greater risk for the development of ASD regardless of biological sex, hence implying that a female brain phenotype may provide a protective effect for ASD.