The first work package was to generate mutants of our target genes and to measure the phenotypic impact of mutations. The target genes are associated with desiccation resistance in tardigrade, their source organism but also in bacteria and yeast (Boothby et al 2017). First, we optimised the sequences for expression in yeast/bacteria. We used the optimised sequence as template in error-prone PCR to generate libraries of random mutants. The libraries were cloned into expression plasmids and transformed into bacteria/yeast cells, ready for screening. In parallel we attempted to reproduce the results of Boothby et al 2017 but could never observe the reported phenotype. At this point, our lab closed due to Covid-19. When lab work resumed we chose to investigate desiccation-resistance proteins from the Late embryogenesis abundant (LEA) family published in Liu et al., 2019. These sequences are extremely rich in repeated regions, which creates difficulties for gene synthesis, assembly and sequencing. Still, our preliminary experiments with these genes showed a promising phenotype in bacteria. We are near finishing the preparation of the new libraries.
The second work package was to develop a pipeline to analyse the experimental data. Our collaborators at the Kondrashov lab shared a dataset of mutants vs phenotype for 4 green fluorescent proteins (GFPs). The results were published (Sommermeyer et al., 2022) and the code, available on Github can be reused on intrinsically disordered proteins. In the paper, we characterized the fitness peaks of four GFPs with a broad range of sequence divergence. Two studied fitness peaks were highly sensitive to mutations and epistatis, two were mutationally robust. Interestingly, mutationally robust proteins are not optimal templates for machine-learning-driven protein design. Instead, predictions were more accurate for mutationally fragile proteins. This conclusion gives useful insights for protein engineering.
