Autism spectrum disorder (ASD) is associated with high structural heterogeneity in magnetic resonance imaging (MRI). This work uncovers three neuroanatomical dimensions of ASD (N=307) using machine learning methods and constructs their characteristic MRI signatures. The presence of these signatures, along with their clinical profiles and genetic architectures, are investigated in the general population. High expression of the first dimension (A1, aging-related) is associated with globally reduced brain volume, cognitive dysfunction, and aging-related genetic variants. The second dimension (A2, schizophrenia-like) is characterized by enlarged subcortical volume, antipsychotic medication use, and partially overlapping genetic underpinnings to schizophrenia. The third dimension (A3, classical ASD) is distinguished by enlarged cortical volume, high non-verbal cognitive performance, and genes and biological pathways implicating brain development and abnormal apoptosis. Thus, we propose a three-dimensional endophenotypic representation to construe the heterogeneity in ASD, which can support precision medicine and the discovery of the biological mechanisms of ASD.

We propose SPHARM-OT, an enhanced spherical harmonic (SPHARM) surface modeling method using optimal transport (OT) for spherical parameterization. To demonstrate its effectiveness, we apply it to shape analysis of amygdala atrophy in Alzheimer's disease (AD). Of note, identifying morphological abnormalities of medial temporal structures such as hippocampus and amygdala is an important research topic for early diagnosis of AD. Our empirical study includes two steps: (1) the newly proposed SPHARM-OT method is used to model amygdala shapes of 101 participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI); (2) using the random field theory to perform surface-based statistical analysis for detecting shape changes between cognitively normal (CN) and AD participants. We demonstrate that (1) the new SPHARM-OT method could effectively reduce surface mapping distortion and lead to a more accurate shape reconstruction result; and (2) significant shape changes between CN and AD participants are identified on certain amygdalar surface regions.

BackgroundData‐driven unsupervised and semi‐supervised clustering methods have parsed neuroanatomical heterogeneity of Alzheimer's disease (AD) and/or mild cognitive impairment (MCI) into multidimensional neuroimaging representations (Vogel et al., 2021; Wen et al., 2021b, 2021a; Yang et al., 2021; Young et al., 2018; Zhang et al., 2016). However, whether this neuroanatomical heterogeneity is partly underpinned by genetic heterogeneity remains unanswered.MethodWhole‐genome sequencing (WGS) data were analyzed for individuals in the Alzheimer’s Disease Neuroimaging Initiative. WGS data underwent a standard genetic pipeline. This resulted in 1,487 participants (428 healthy controls, 489 MCI and 570 AD; age = 78.96 ± 7.72; 44.05% females) with 24,773,167 single nucleotide polymorphism (SNP). The deep learning Smile‐GAN model (Yang et al., 2021) generates expression scores (ES) across each of the four subtypes of MCI/AD (Fig. 2). The ES of each subtype were used as phenotypes in genome‐wide association studies (GWAS). Specifically, we performed multiple linear regressions controlling for age, sex, intracranial volume, disease diagnosis, and the first four genetic principle components using Plink (Purcell et al., 2007). We then performed clumping to define the independent significant variants (ISVs).ResultGWAS discovered 17 ISV‐ES pairwise associations. WGS data allowed us to identify four de no ISVs that were not included in dbSNP (Fig. 1). P1 and P4 detected similar AD‐related genetic variants (e.g., APOE, NECTIN2, and TOMM40, Fig. 1 and 2), but with opposite directions of effect, i.e., protective factors for P1 and risk factors for P4, consistent with respective patterns of neurodegeneration in these phenotypes. In particular, for rs429358 (APOE‐4), the P4 subtype had more C alleles (minor allele) than the P1 subtype (P‐value < 1e‐20). The four ISVs identified for P2 and P3 were not previously associated with any clinical trait.ConclusionThe current study identified genetic heterogeneity related to the expression of four MCI/AD subtypes previously defined via Smile‐GAN. Notably, our results confirmed previous findings that AD is genetically heterogeneous (Nacmias et al., 2018), identifying known genetic risk factors and several novel variants. Further research is needed to understand how these variants may affect AD pathophysiology and determine if any may provide a therapeutic target.