Animal Population Imaging - Linking Molecular Pathways to Autism Clusters in the Mouse Using the Neuroanatomy.

Background: Autism is extremely heterogeneous, not just in behaviour, but also in genetics, and neuroanatomy. Over the past 10 years, we have established a magnetic resonance imaging (MRI) database of mouse models related to autism, which allows the investigation of the neuroanatomy in a large autism population in the mouse. Addtionally, this dataset provides us with a metric by which we can cluster these models and examine commonalities and differences between them in the hope of subcategorizing autism.

Objectives: The purpose here is twofold. 1) To further our previous work, which examined 26 different mouse models of autism (Ellegood et al. 2015), in an effort to characterize and cluster our current dataset of 92 different mouse models, and 2) to look for existing molecular pathways which may explain these clusters.

Methods: The data used in this study was collected from 92 different autism related mouselines and includes over 3,700 mice. Imaging was performed ex-vivo using a 7T MRI with a T2 weighted, 3D fast spin echo sequence which acquires data at an isotropic resolution of 40 μm (Spencer Noakes et al. 2017).

Data Analysis – To visualize and compare any differences, the images are registered together (Lerch et al., 2011). From this the volumes of 182 different regions (Dorr et al. 2008, Ullmann et al. 2013, and Steadman et al. 2014, Richards et al. 2011, Qiu et al. 2018, Beera et al. 2017) were calculated. Group differences in each of the 182 regions across the different mouse models were calculated (measured as effect size) and used to group the different models using hierarchical clustering algorithms.

Bioinformatics – Online databases, including StringDB (string-db.org), the Molecular Signatures Database (software.broadinstitute.org/gsea/msigdb) and the KEGG Database (www.genome.jp/kegg), were used to link specific molecular pathways to our clusters to provide more information about potential common determinants that underlie the models.

Results: Our data suggests that the autistic phenotype both preferentially affects key regions of the brain, but also divides the autism neuroanatomical phenotype based on directionality and localization of the differences throughout the brain. In total, four different clusters were found from the 92 different models. Additionally, what appears to link these clusters together, outside of the neuroanatomy, is a shared molecular pathway. The first cluster, which includes the models Chd8 and Dvl1, links significantly to the Wnt signaling pathway. The second cluster, which includes the models Arid1b and Tsc1, links to MapK and mTor signaling pathways. The third cluster, which includes the models Nlgn1, Nlgn3, Nrxn1, and Shank3, links to cell adhesion. The final cluster does not link significantly to any specific pathway, which may indicate the need for further refinement of our clusters.

Conclusions: The heterogeneity of autism is problematic. Therefore, finding ways to link autism models or genetic modifications together can be very powerful. The hope with techniques such as this is ultimately to subcategorize and increase the diagnostic specificity of autism.