Prioritizing Mouse Models for Autism Spectrum Disorder Using A Disease-Specific Phenotype Signature

Background: The high genetic and phenotypic heterogeneity of Autism Spectrum Disorders (ASD) poses an enormous challenge for understanding its etiology. Mouse models are excellent tools to study the effect of specific genetic or environmental alterations on ASD-related phenotypes. However, despite the rapid growth in both number and scope of ASD mouse models; there is no consensus about the phenotypic measures (phenoterms) to properly recapitulating this human condition in mice. 

Objectives: To address this issue, we investigated phenotypic data from hundreds of ASD mouse models with the goal to prioritize existing ASD mouse models according to their disease-specific phenotype signature.

Methods: Overall, data of 61 commonly used phenoterms belonging to 10 (7 auxiliary and 3 core) phenotype categories from 249 ASD mouse models from AutDB (http://autism.mindspec.org/autdb/AMHome.do), and 79 non-ASD mouse models from the Mouse Genome Informatics (MGI) resource (http://www.informatics.jax.org) were evaluated. Two-sided Fisher exact test was used to assess the differences in phenotype outcomes between ASD and non-ASD mouse models. Further, a forward stepwise procedure was employed to search for the best combination of phenoterms to separate ASD from non-ASD mouse models. Finally, we used these data to score and rank existing ASD mouse models.

Results: Of the 61 phenoterms, 16 demonstrated significant differences in outcomes between ASD and non-ASD mouse models (P < 0.05), and four (‘Brain Morphology’, P = 7.5 x 10^-10;  ‘Motor Coordination’, P = 1.7 x 10^-9; ‘Locomotor activity’, P = 1.8 x 10^-9; and ‘Cytoarchitecture’, P = 5.3 x 10^-8) remained significant even after using the conservative Bonferroni correction for multiple testing. Further, a subset of twenty-one phenoterms achieved the best distinction between ASD and non-ASD mouse models (Area Under the Curve [AUC] = 0.82).  An alternative algorithm using all 61 phenoterms and assigning high weights to the three core phenotype categories of ASD (Social, Communication, and Repetitive Behaviors), showed a slightly lower classification efficiency (AUC = 0.76). Remarkable variation was seen between model scores reflecting differences in both tested phenotypes and their corresponding outcomes. The highest ranked mouse model by both scoring schemes was the Pten conditional knockout, implying a good agreement between core and auxiliary ASD symptoms.   

Conclusions: This study provides an initial step towards establishing a standardized battery of phenotypic measures to allow efficient and accurate evaluation of ASD mouse models.