Probing PAX6-DNA interactions using high-throughput yeast one-hybrid assays and deep mutational scanning

Abstract

PAX6 is a highly conserved transcription factor essential for the correct development of the central nervous system, the pancreas, and the eye.

Heterozygous deletions, nonsense, and frameshift mutations are generally well characterised as causing aniridia, while most missense variants produce a broad range of other discrete ocular pathologies. The interplay between PAX6 and its DNA targets is complicated by multiple functionally distinct subdomains, co-factors, and divergent spectra of disease phenotypes. Knowledge of the contributions made to binding by each residue and the impact of missense variants is likely key for understanding the role of PAX6 mutations in disease. Current methods of exploring this interaction grammar have been limited by relatively low throughput techniques that are resource intensive and can only feasibly be performed on a handful of residues.

Here I used a combination of yeast one-hybrid and deep mutational scanning in competitive growth assays to probe the functional consequences of almost all the possible single amino acid variants in the paired domain of PAX6. The results showcase the capabilities of the assay to reproducibly generate functional DNA-binding information that correlates with aspects related to PAX6 structure and proximity to DNA. Numerous variants demonstrated substantial shifts in functional consequence from gain-of-function to loss-of-function to worse-than-null that were exquisitely dependent on the DNA target sequence providing a plausible route for pleiotropic effects on genome-wide PAX6 gene target expression. Among the PAX6 variants identified in humans, pathogenic and benign variants showed significantly different fitness scores yet overlapped in their distributions suggesting mechanisms of pathogenicity exist beyond PAX6-DNA interaction dynamics.

Additionally, no correlation was found between DNA affinity and disease phenotype. The assay was also able to predict PAX6PD function independent of the yeast one-hybrid reporter system through suspected promiscuous binding to regions of the yeast genome.

It is hoped that this deep mutational scan of PAX6 will aid in the modelling of existing and novel variants, and in the development of in-silico methods for pathogenicity prediction. The data will also contribute to the growing online protein variant repositories that are increasingly becoming an invaluable resource for clinicians and researchers.