Pathway Optimization in Escherichia coli through Systematic Design-Build-Test-Learn (DBTL) Cycle and Crispr-Cas9 Mediated Chromosomal Integration

Metabolic engineering have heavily relied on plasmid-based
expression systems that were optimized using try-and-error approaches guided
more by intuition than objective data. At JBEI, we envision biological
engineering as a discipline directed by increasingly complex and more
predictive metabolic models based on larger sets of omics
data. In the present work, we illustrate how our Design-Build-Test-Learn (DBTL)
cycle has been successfully applied to optimize the complex heterologous mevalonate pathway (MVA) in E. coli to produce a wide range of isoprenoids
in high yields (e.g. 2 g/L of isopentenol (C₅),
1.2 g/L of bisabolene (C₁₅) and 0.5-0.6 g/L
of limonene and 1,8-cineole (C₁₀), representing 55%, 35% and 19% of
its maximum theoretical yields (MTY), respectively).

To obtain these high productivities, several tools have been
created in order to improve the efficacy of our DBTL cycle. Analytical
methodologies, such as proteomics and metabolomics to quantify key enzymes and
intermediate metabolites of the MVA pathway as well as central carbon
metabolism, have been successfully developed at JBEI. More interestingly, mathematical
tools and models to interpret the data have been implemented to direct the
engineering in a systematic manner: 1) Principal component analysis of proteomics
(PCAP) aided the optimization of limonene and bisabolene
to up to 1.2 g/L and 2) a workflow integrating metabolomics, proteomics, and
genome-scale models was developed to study the effects of introducing the
heterologous MVA pathway into E. coli
and to identify candidate
genes to be engineered that increased isoprenoid productivity.

More recently, new engineering methodologies based on
CRISPR-Cas9 have been developed for the rapid integration and edition of the MVA
pathway directly into the E. coli
chromosome. These methodologies aim to avoid problems associated with plasmid
instability and the use of antibiotics. Hundreds of strain variants can be created
in this manner and fed into the DBTL cycle for optimization.

Automation of this optimization process is being developed
using robotics for 1) DNA parts assembly (j5 and Par-Par software created at
JBEI) 2) strain characterization in 48 well plates (Biolector)
and 3) sample preparation (proteomics and metabolomics). Furthermore, results
from the analysis are stored into a data repository (the experimental data
depot (EDD)) in a standardized fashion, facilitating the downstream
analysis. 

This optimized DBTL cycle is not strain or molecule
dependent and can, therefore be applied to optimize other target pathways and
molecules of interest to the metabolic engineering community.