P-type ATPases constitute a superfamily of ubiquitous primary active transmembrane transporters. These ion pumps are critical for maintaining cellular homeostasis and exhibit complex multi-domain architectures, innately rendering them a subject of significant biological interest. The superfamily is subclassified based on structural and selectivity similarities into subfamilies. In this study, our focus is on the P1B-type subfamily which encompasses soft transition metal ATPases. To elucidate the key aspects that define selectivity for this subfamily, an interdisciplinary approach that combined computational and experimental techniques was employed. Our Collaborators employed unsupervised machine learning (ML) to generate a landscape of De Novo sequences derived from a multiple sequence alignment of around 13,500 P1B-type sequences. Subsequently, molecular dynamics (MD) was utilized to rationally filter and select a subset of those regenerated sequences to assess their functional viability as De Novo P1B-type ATPases. GPU-accelerated Amber software facilitated molecular mechanics (MM) and QM/MM simulations coupled with free-energy methods, aiding in the selection process. Besides, a final experimental validation of these MD-filtered regenerated sequences was used to evaluate their catalytic activity and transport properties. Remarkably, our findings demonstrate that despite undergoing approximately 200 mutations, the MD-screened De Novo sequences retain the characteristic behavior of P1B-type ATPases. The implications of this study underscore the emergence of computational methods as a tool for uncovering latent information within protein sequences and most significantly, the prospect of rationally engineering protein selectivity with in silico tools in the imminent future. Thanks to the advanced GPU-based technology, MD seamlessly brought together different disciplines, fostering an integrated approach.

Vladimir Volobouev

Appearances