Optimization of heterologous gene expression for in vitro evolution.

Proteins can be evolved in vitro by randomly mutating the corresponding genes, expressing the resulting libraries in a population of E. coli cells, and screening for clones that exhibit a desired phenotype (16). In our experience, the most time-consuming part of in vitro or “directed” evolution is the development of the high-throughput assay or “screen”. In particular, the wildtype gene should initially be expressed so that the desired activity is at the lower limit of detection, allowing for maximal improvements in phenotype during the course of the experiment. Adjustment of gene expression to achieve this basal activity level is often labor intensive, requiring trial and error with multiple expression vectors, strains, and induction conditions. For example, we attempted to engineer an expression vector that produces low levels of β-glucuronidase (GUS) for an in vitro evolution experiment. Standard expression vectors, based on the lac or T7 promoters, proved unsuitable because they yielded activities near the upper limit of the dynamic range of our assay (data not shown). We used standard recombinant DNA techniques to subclone the EcoRI-AflIII fragment of gusA-pBS∆ containing the lac promoter (lacp) and the β-glucuronidase gene (gusA) (10) into the lower copy number pET20 plasmid (Novagen, Madison, WI, USA). Unfortunately, this lacp-gusA-pET20 construct proved to be phenotypically identical to the higher copy number, parental gusApBS∆ plasmid (data not shown). We considered modulating gusA expression by varying inducer concentrations and induction times, but had previously found that these measures provided limited control. Most transported inducers, including those that de-repress the lac and PBADpromoters, yield “all-or-nothing” expression (7) and thus can lead to mixed populations of expressing and non-expressing cells within clonal cultures (2,7). This cellto-cell variability in turn hampers screens and selections in which enzyme activity within individual cells is assayed (2,13). The growing popularity of these high-throughput screens (12) demands a more systematic and general approach to adjusting expression levels. Rosenberg and Court (14) have shown that 75% of all mutations that affect promoter function fall within the -35 or -10 hexanucleotide regions upstream of the RNA start site. Mutations within the -35 region tend to greatly diminish transcription, whereas those in the -10 region can have mild to deleterious effects (18). Previous workers have randomized these conserved elements to learn about promoter structure and function (4,11). Jensen and Hammer (6) used a randomization approach to optimize the expression of a Lactococcus lactis gene for industrial fermentation. They generated a library of synthetic L. lactis promoters that contained conserved consensus elements flanked by 24 nucleotides of random sequence, isolated 38 clones, and found that gene expression levels among these clones varied over a 7000-fold range (5,6). While this approach might work for other promoters, it is technically easier to identify and manipulate core promoter sequences. Therefore, we chose to randomize the six conserved nucleotides in the -10 region of the lac promoter instead of 24 non-conserved ones. By limiting the size of the library (46 vs. 424), it proved possible to examine every combination of nucleotides using the same high-throughput assay developed for the in vitro evolution experiment. The -10 promoter region was randomized by PCR-based mutagenesis, although any site-directed mutagenesis protocol (reviewed in Reference 9) would likely have worked. We replaced the gusA of lacp-gusA-pET20 with a 1kb XbaI-EcoRI stuffer fragment using standard techniques. The whole lacpstuffer-pET20 expression vector was amplified using Vent DNA polymerase (New England Biolabs, Beverly, MA, USA) (1) and the primers 5′-NNNNNNCGAGCCGGAAGCATAAAGTGTBenchmarks