Transcription factor binding and individual genetic risk for valproate teratogenicity

Valproate (VPA) use during pregnancy is associated with a wide range of birth defects and adverse neurodevelopmental outcomes, but not all exposed children are affected and there is evidence for a genetic predisposition. We hypothesised that genomic variants that impact on the binding affinity of transcription factors (TFs) are integral to VPA-associated teratogenicity and a plausible explanation for variance in interindividual risk. We interrogated maternal exomes from women recruited through international epilepsy genomics consortia. The variant burden within genes associated with 32 different birth defect types was higher for those exposed to VPA as compared to other antiseizure medications (OR 1·73 [95% CI 1·40 to 2·14], p = 2·25E-07). Variants in women exposed to VPA were predicted to impact the binding affinity of 359 TFs and network analysis of encoded proteins indicated that a master regulator, EP300, interacts with 42% (151/359) of all variant sensitive TFs. We then profiled coexpression between EP300 and other TFs in differentiating neurons derived from human embryonic stem cells (hESCs) exposed to VPA at 300µM and 700 µM, or unexposed, and a reference map generated using public data. Almost all coexpressed pairs observed in the cell assay data were also observed in the reference map. In contrast, only 30% (119/400) of the coexpressed pairs observed in unexposed cells were present in cells exposed to VPA; suggesting that VPA activates alternative biologically valid, but context inappropriate, signaling cascades. Our findings suggest that VPA-induced disruption of EP300-related regulation is common across birth defect types and that genetic variation can modify subsequent transcriptional dysregulation, explaining why only some pregnancies are affected. The results have implications for the development of genetic risk biomarkers and safer drugs.