High-accuracy nanopore consensus sequencing for protein engineering

Predicting the effect of mutations and subsequent evolutionary trajectories of proteins is hindered by our limited understanding of large sequence spaces which are shaped by frequent non-additive interactions between mutations. The study of these interactions requires that we accurately sequence many variants of an entire gene to capture mutations that occur simultaneously. Current sequencing technologies, however, are expensive, practically demanding or limited to short reads. Here, we introduce a simple and low-cost single-molecule tagging strategy that leads to long-reads with accuracies >99.99% using a commercial nanopore sequencer. We validate this approach by studying three rounds of directed evolution of an amine dehydrogenase via ultrahigh throughput screening in microfluidic droplets (10^5/h), identifying sign epistasis in one lineage and generating a biocatalyst with over 10-fold increased turnover number (kcat) based on sequencing data alone. This accessible sequencing workflow will enable evolution to be mapped from high-throughput experiments and could facilitate the development of better prediction tools and machine learning algorithms.