RBS Library Calculator: Efficient Search for Optimal Enzyme Expression Levels

The addition of synthetic metabolic pathways into engineered organisms enables the production of a wide variety of high-value chemicals from low-value feedstock. However, maximizing a pathway's productivity and optimizing its enzyme expression levels remains a major challenge. In bacteria, a common approach is to randomly mutate ribosome binding sites (RBSs) to vary the enzyme expression levels in a pathway, followed by functional screening or selection. Using our recently developed RBS Calculator (Salis et. al., Nature Biotech., 2009), we show that these libraries are highly inefficient and do not guarantee pathway optimization, due to exceedingly large library sizes and non-uniform sampling.

To overcome this challenge, we have developed the RBS Library Calculator, a predictive design method that enables efficient searching for optimal protein expression levels by substantially reducing the combinatorial search space. The method identifies a degenerate RBS sequence that will uniformly vary a protein's expression level across a selected range (from 0 to 100,000 on a proportional scale), while minimizing the resulting library's size. The RBS Library Calculator combines our predictive model of bacterial translation initiation with a genetic algorithm, using the 15 letter degenerate nucleotide code.

A typical optimized RBS library will uniformly vary a protein’s expression level across a 100,000-fold range using only 12 to 36 sequence variants; a similar randomly mutagenized RBS library with 6xN nucleotides contains 4096 variants. Multiple optimized RBS libraries may be combined to search for the optimal enzyme expression levels of a multi-protein pathway, yielding a significant reduction in the combinatorial search space compared to random mutagenesis.

We have experimentally tested the RBS Library Calculator’s predictions using fluorescent protein reporters, spectrophotometry, flow cytometry, and steady-state measurements. Optimized degenerate RBS libraries were constructed for individual proteins (mRFP1, GFPmut3b, and an alcohol dehydrogenase-GFPmut3b fusion protein), and for a synthetic operon consisting of three fluorescent proteins (mRFP1, GFPmut3b, CFP). Using experimental measurements, we demonstrate that optimized RBS libraries uniformly sample a single or multi-protein expression level space.

Further, the RBS Library Calculator may generate smaller optimized RBS libraries by “zooming in” on a selected subset of the expression level range or by uniformly scanning enzyme expression levels at a lower resolution. When low-throughout assays are necessary, one may use these features to iteratively zoom in on a smaller enzyme expression level range with an increasingly higher resolution, allowing one to efficiently pinpoint the optimal enzyme expression levels of a many-protein pathway.