Figure 1.Building and using sequence‐expression‐activity maps (SEAMAPs)
A. The RBS Library Calculator designs a synthetic RBS library to efficiently search a multi‐dimensional protein expression space. A kinetic mechanistic model maps the relationship between protein expression levels and genetic system activity, using a minimal number of measurements for parameterization. The SEAMAP's predictions are repeatedly used for different design objectives. The RBS Library Calculator designs a new RBS library to zoom onto a region of targeted protein expression levels for optimized genetic system performance.
B. The RBS Library Calculator combines a biophysical model of translation with a genetic algorithm to identify the smallest degenerate RBS sequence (dRBS) with maximal search coverage for an input protein‐coding sequence (CDS). The biophysical model calculates the ribosome's binding free energy ΔGtot for an input mRNA sequence S, which is then related to its translation initiation rate and protein expression level P.
C–E Fluorescence measurements show that optimized RBS libraries in Escherichia coli DH10B searched a 1‐dimensional expression level space with 94, 79, and 99% search coverages at high, medium, and low search resolutions, respectively. Translation initiation rate predictions (red diamonds) are compared to measurements (Pearson R2 is 0.88, 0.79, and 0.89, respectively. All P‐values < 0.001). Data averages and standard deviations from 6 measurements.
Figure 2.Searching expression spaces in diverse bacterial hosts
Differences in bacterial 16S rRNA sequences lead to different mRNA translation rates. To overcome combinatorial explosion, the biophysical model predicts how changes in mRNA and rRNA sequences control translation rate.
A 14‐ to 16‐variant‐optimized RBS library controlling mRFP1 expression was characterized in Escherichia coli BL21, Pseudomonas fluorescens, Salmonella typhimurium LT2, and Corynebacterium glutamicum. The biophysical model accurately predicted translation initiation rates across a > 1,000‐fold range in E. coli BL21 (ΔΔGtotal is 1.18 kcal/mol, R2 is 0.93), P. fluorescens (ΔΔGtotal is 1.63 kcal/mol, R2 is 0.90), S. typhimurium LT2 (ΔΔGtotal is 1.83 kcal/mol, R2 is 0.89), and C. glutamicum (ΔΔGtotal is 1.81 kcal/mol, R2 is 0.88). All P‐values < 10−6. Data averages and standard deviations from three measurements.
Two RBS libraries were optimized to control the expression of a genomic single copy of mRFP1, incorporated into the amyE locus of Bacillus subtilis. Fluorescence measurements from 14 clones were compared to their predicted translation initiation rates (Pearson R2 is 0.81, P‐value is 2 × 10−6). The expression space was searched with 76% coverage. Data averages and standard deviations from three measurements.
A 12‐variant RBS library was optimized to control genomic lacZ expression. Predicted translation initiation rates are compared to measured lacZ activities (circles), including the wild type (diamond), showing a linear relationship below the activity plateau (Pearson R2 is 0.93, P‐value is 0.02). The expression space was searched with 84% coverage. Data averages and standard deviations from four measurements.
Figure 4.Sequence‐expression‐activity mapping of a multi‐enzyme pathway
Characterization of two libraries of neurosporene biosynthesis pathway variants, using the RBS Library Calculator for searching a large combinatorial space (Search mode, left) or a narrow targeted region (Zoom mode, right). Averages and standard deviations from at least three measurements of neurosporene productivities.
Measurement data and translation rate predictions (circles) from Search mode are used to parameterize a kinetic model of the pathway's reaction rates, showing the relationship between crtEBI translation rates and neurosporene productivity.
To design pathways with higher activities, a translation rate region (gray box) is targeted using the RBS Library Calculator in Zoom mode. Translation rate predictions from selected pathway variants are shown (circles).
A schematic of the bacterial operon‐encoding crtEBI, and the corresponding reactions, genes, and metabolites in the biosynthesis pathway. Cofactors are not shown.
To evaluate the design of pathways with intermediate activities, 19 additional crtEBI pathway variants were characterized, and the predicted neurosporene productivities (black bars) were compared to the measured productivities (green bars). Data averages and standard deviations from two measurements.
Figure 5.Using SEAMAPs to design multi‐enzyme pathways with desired activity response curves
A promoter's transcription rate r and the crtEBI translation rates (x, y, z) are inputted into SEAMAP's kinetic model to determine the pathway's productivity.
A slice of the CrtEB expression‐activity space is shown, where CrtI expression is 200,000 au. The effects of transcriptional regulation for four pathway variants are shown as diagonal lines at their respective translation rates. The productivity of a reference pathway variant in one condition (black square) was characterized to determine its location in expression‐activity space, which provides orientation for all other locations.
Left: The effect of the PlacO1 promoter's transcription rate on the pathway variants' productivities is calculated. The location of the global maxima depends on the promoter's transcription rate and the mRNA translation rates. Right: The productivities of the four pathway variants are measured as transcription rate is increased via IPTG induction. Changes in translation rate cause the global maxima to appear at lower transcription rates, consistent with model calculations. Data averages and standard deviations from two measurements.
Figure 6.Increasing precursor biosynthesis for optimally balanced versus imbalanced pathways
A. The relationship between crtEB translation rates and CrtE's flux control coefficient (FCC) is calculated using SEAMAP predictions. A lower FCC indicates that the enzyme is less rate limiting. Here, the crtI translation rate is 100,000 au. According to their FCCs, increasing precursor biosynthesis is predicted to improve the optimally balanced pathway variant (black circle) more than the imbalanced pathway variant (black diamond).
B, C An optimized RBS library is integrated to control genomic dxs translation initiation rate and systematically vary precursor biosynthesis, followed by productivity measurements using either (B) an optimally balanced pathway or (C) an imbalanced pathway variant. Predicted crtEBI translation initiation rates are (305,000 au; 17,120 au; 886,364 au) for the optimally balanced pathway variant and (1,046 au; 20,496 au; 200,300 au) for the imbalanced pathway variant.
Figure 7.Predicting the evolutionary landscape of a multi‐enzyme pathway
Histograms show that random mutations will more likely decrease a pathway's productivity.
A–C Either (A) one‐, (B) two‐, or (C) three‐nucleotide mutations are randomly introduced into the 35 nucleotide‐long ribosome‐binding site sequences of the crtEBI operon. Changes in enzyme expression levels and pathway productivities are predicted using the SEAMAP.