Integration of behavioral and biological variables using penalized regression: an application to the maternal immune activation model of autism

Maternal immune activation (MIA) during pregnancy increases the odds of developing neuropsychiatric disorders such as autism spectrum disorder (ASD) later in life. In pregnant mice, MIA can be induced by injecting the viral mimic polyinosinic:polycytidylic acid (poly(I:C) to pregnant dams resulting in altered fetal neurodevelopmental and behavioral changes in their progeny. Although the murine MIA model has been extensively studied worldwide, the underlying mechanisms have only been partially elucidated. Furthermore, the murine MIA model suffers from lack of reproducibility, at least in part because it is highly influenced by subtle changes in environmental conditions. In human studies, multivariable (MV) statistical analysis is widely used to control for covariates including sex, age, exposure to environmental factors and many others. We reasoned that animal studies in general, and studies on the MIA model in particular, could therefore benefit from MV analyzes to account for complex phenotype interactions and high inter-individual variability. Here, we used a dataset consisting of 26 variables collected on 67 male pups during the course of several independent experiments on the MIA model. We then analyzed this dataset using penalized regression to identify variables associated with in utero exposure to MIA. In addition to confirming the association between some previously described biological variables and MIA, we identified new variables that could play a role in neurodevelopment alterations. Aside from providing new insights into variable interactions in the MIA model, this study highlights the importance of extending the use of MV statistics to animal studies.