Characterizing Strain Variation in Engineered E. coli Using a Multi-Omics-Based Workflow

Abstract

Understanding the complex interactions that occur between heterologous and native biochemical pathways represents a major challenge in metabolic engineering and synthetic biology. We present a workflow that integrates metabolomics, proteomics, and genome-scale models of Escherichia coli metabolism to study the effects of introducing a heterologous pathway into a microbial host. This workflow incorporates complementary approaches from computational systems biology, metabolic engineering, and synthetic biology; provides molecular insight into how the host organism microenvironment changes due to pathway engineering; and demonstrates how biological mechanisms underlying strain variation can be exploited as an engineering strategy to increase product yield. As a proof of concept, we present the analysis of eight engineered strains producing three biofuels: isopentenol, limonene, and bisabolene. Application of this workflow identified the roles of candidate genes, pathways, and biochemical reactions in observed experimental phenomena and facilitated the construction of a mutant strain with improved productivity. The contributed workflow is available as an open-source tool in the form of iPython notebooks.

Figure 1. A workflow for bridging the genotype-phenotype relationship with multi-omics data and genome-scale models of E. coli metabolism expressing heterologous pathways

(a) Multi-scale data types that are generally collected to elucidate changes in metabolic phenotypes of different strains. (b) Our workflow involves a hierarchical staging of computational analysis methods: (i) basic strain differences; (ii) relevant patterns and correlations in the data; (iii) mechanisms of action in the context of a genome-scale network that can explain apparent differences in strain behavior.

Figure 2. Pathway assembly, strain selection and multi-omics data generation

(a) This study characterizes three versions of a heterologous mevalonate pathway engineered to synthesize isopentenol, limonene, and bisabolene. (b) Over a 72-hour time course, the engineered strains show various levels of fuel production due to changes in heterologous pathway architecture and expression. Each strain is indicated by its respective color, shown in the legend to the right. (c) The nine strains were further analyzed using a multi-omics approach during batch fermentation to generate detailed omics profiles.

Figure 3. Systems-level multi-omics integration and analysis of batch fermentation dynamics

(a) The first stage of the workflow filters, maps, and identifies system level differences between control (e.g., WT) and test (e.g., engineered strain) conditions through the construction of dynamic difference profiles. (b) The differences for each data point relative to the control were calculated, and the errors of the measurements were propagated to determine the range of change (from significant to not changing) between the control and test conditions. The plots in the left column refer to positive (“+”) shifts, or points where the test condition is greater than the control in terms of concentration or flux. Those in the right column refer to negative (“−”) shifts. Standard deviations for the test and control condition for each data point were calculated from triplicate measurements or estimated based on the percent root-squared deviation (%RSD) of representative triplicate measurements. (c) A cartoon depiction of the data types included in this analysis: protein level measured by proteomics (top left), substrate and product metabolites measured through metabolomics (top right and bottom right), and computed flux (bottom right).

Figure 4. Integrating multi-omics data with genome-scale models of metabolism

Stages two and three of the workflow combine multivariate analyses and genome-scale models of metabolism. (1) Standard metabolomics and growth measurements for over 80 metabolites were taken for nine different strains over a 72-hour time course. (2) Applying PCA on this dataset, we find three distinct metabolic phases that align with different phases of the growth. (3) These pseudo-steady state phases are modeled using constraint-based methods, such as Markov-chain Monte Carlo based sampling using extracellular measurements as inputs to the model. We compute and cluster perturbed reactions in host metabolism (illustrated by the red colored nodes in the network). (4) Perturbed reactions are assessed with other omics datasets, like proteomics.

Figure 5. Genome-scale modeling revealed perturbations in TCA cycle and pentose-phosphate pathway activity associated with certain engineered phenotypes

Reactions colored by the shift (absolute value) in flux in a top-producing strain, I3, compared to wild-type in different pathways in central carbon metabolism: (a) the pentose-phosphate pathway; (b) glycolysis/gluconeogenesis; (c) TCA cycle. Shown in (d) are significant reaction flux shifts (p < 0.05) corresponding to various reactions in these pathways in phase I (0–6 hrs) and those for phase II (6–20 hrs) are displayed in (e). Here, shifts in metabolic flux represent overall changes (both positive and negative perturbations) from wild-type behavior.

Figure 6. Constraint-based modeling elucidates pathways that allow for coupling of NADPH metabolism and biofuel production

(a) The sum of flux through these main NADPH-producing and consuming reactions is significantly higher in top-producing strains over WT. (b) Increases in cofactor (box A), glycolysis/gluconeogenesis (box B) and TCA (box C) metabolite concentrations (relative to wild-type E. coli) indicate regions in metabolism that are perturbed in different engineered strains. (c) Dynamic difference profiles identify changes in protein levels for isopentenol-producing strains. As shown in the lower left panel, key glycolysis (yellow), PPP (orange), and TCA (red) proteins shift above WT levels in higher producing strains (I2 and I3). On the lower right panel is an example of how progressive engineering efforts change the dynamic difference profile for the protein acetate synthase (ACS).

Figure 7. Model-driven predictions discover a gene-knockout that increases the specific production of isopentenol

(a) Single gene knock-out simulations were performed using genome-scale modeling to identify candidate targets that increase the production of isopentenol. Knock-outs were experimentally tested and deletion of the gene in bold, ydbK, was found to increase specific production of isopentenol. (b) Growth-normalized isopentenol titer (mg/L/OD600) is displayed for strain I3 (black) and I3 with ΔydbK knockout (gray). At every non-zero time point, the knockout variant produces significantly more isopentenol than the highest producing strain, I3 (stars denote p-values: 4 hrs p = 0.0058 (**), 8 hrs p < 0.0001 (****), 24 hrs p = 0.002 (***), 48 hrs p = 0.0037 (**), using an unpaired two-tail t-test). At 48 hours, absolute isopentenol titers are 920 mg/L versus 800 mg/L for strains I3ΔydbK and I3, respectively.