DS0507 - Biotechnologies  et valorisation des bio-ressources :

Metabolic Engineering Machine – MEM

Metabolic Engineering Machine

Despite the growing number of chemicals successfully engineered in host organisms, bioproduction R&D is slow and expensive, as the process is mostly based on trial-and-error. To overcome this critical hindrance, we propose to implement a generic automated design-build-test and learn cyclic pipeline for the production of targeted chemicals. As an illustration, we will apply the pipeline for the metabolic engineering of a library of new antimicrobials against Gram-positive bacteria.

An automated design-build-test- and - learn pipeline for metabolic engineering

Getting from a proof of principle strain (a few mg/L) to a pre-industrial optimized strain (a few g/L) for producing a chemical of interest is a long and tedious process based on trial-and-error. In this project, we aim at showing that a supervised machine learning approach can be used to reduce the development time and costs, replacing trial-and-error by knowledge-driven and model-driven experimental plans. <br /><br />Precisely, we propose to use machine learning to determine the next set of experiments that needs to be performed to increase production titers of chemicals in chassis strains. In the specific approach we will be using, named active learning, a growing training set is acquired on the fly in an iterative process between learning and measurements. The remarkable advantage of active learning is to yield performances comparable to classical supervised learning with training sets sizes that can be several orders of magnitude smaller. <br /><br />The machine learning approach we propose to develop is embedded in an automated design-build-test and learn cycle. That cycle makes use of original design and learning tools, a robotized platform for gene cloning, pathway assembly, strain transformation and bacterial cultures, and an automated high-throughput screening for metabolite identification and quantification. We are using the pipeline to produce a variety of novel flavonoids from the pinocembrin and naringenin precursors. To achieve this goal, the pipeline is run for four research objectives: (RO1) learn enzyme sequences that maximize flavonoid titers, (RO2) learn enzyme expression levels to increase final product yields and avoid or limit intermediate metabolites accumulation, (RO3) -learn up and down gene regulation in the chassis strain genome that maximizes both growth and flavonoid titers, and (RO4) learn flavonoid structures maximizing cytotoxicity toward B. subtilis, the model organism of Gram-positive bacteria.<br />

Our scientific program is composed of four scientific tasks and two supporting activities (management and exploitation tasks). The scientific tasks are the four steps of the MEM pipeline, design, build, test, and learn. The scientific tasks are applied to four research objectives, which will make use of the pipeline for the bioproduction flavonoids in E. coli (see above section).

For RO1 and RO2, all the genes encoding the biosynthetic enzymes are being cloned in E. coli using a robotized workstation, giving rise to several plasmid libraries. Intermediate and final product titers are determined using biosensors. The design-build-test and learn pipeline is repeated several rounds to establish relationships between product titers and enzyme sequences (RO1) or enzyme production levels (RO2). To avoid plasmid load burden, we will engineer in RO3 the best pinocembrin and naringenin producing pathways in the genome of E. coli..

The best performing strains of RO3 will then be used to produce multiple flavonoids selected for their antibacterial activities (RO4). The pipeline will be used again, this time to identify and insert specific heterologous enzymes producing compounds from pinocembrin and naringenin. Their antibacterial activities will be measured using a toxicity assay against B. subtilis. The toxicity measurements will be utilized to learn a structure-activity relationship enabling toxicity prediction from chemical structures. The compounds, which are highly toxic for B. subtilis and displaying low toxicity for E. coli (as predicted by our structure-activity relationship), will be selected for the next cycle round.

As planned in our initial proposal we have focused during the first 18-month period on the bioproduction of two flavonoids precursors: pinocembrin and naringenin.

For each of the four Design, Build, Test and Learn tasks of the MEM pipeline we have completed our milestones and deliverables.

- We have developed and deployed the RetroPath2.0 scientific workflow. This workflow, which is open-source, enables one to search pathways for a target product based on retrosynthetic reaction rules. The workflow has already been used to produce retrosynthetic maps for various flavonoids. The workflow can be downloaded at MyExperiment.org

- We are completing the construction of combinatorial libraries producing pinocembrin and naringenin. In these libraries we have varied gene sequences , promoter and RBS sequences, along with plasmid copy numbers. The product produced by the libraries (i.e., pinocembrin and naringenin) is to be measured using biosensors.

- We have developed and benchmarked biosensors for our final products (pinocembrin and naringenin) and intermediates (malonyl-CoA and tyrosine).

- We have developed a novel machine learning method to search enzyme sequences catalyzing metabolic reactions. That method is based on Gaussian processes enabling to compute a variance to each prediction made. The method is being used to search for sequences catalyzing the flavonoids pathways.

MEM is a 4-year project. Despite the change of laboratory of the coordinator (from the iSSB research unit to the MICALIS laboratory), which has led to a delay in the provision of funds, the project has not lagged behind in its scientific milestones and deliverables. We do not anticipate delays in our future milestones and deliverables.

While the project is being illustrated by the construction of a library of flavonoids, the «design-build-test-learn« cycle is generic and can be applied to other molecules of interest for various industrial sectors, commodities, fine and speciality chemicals. As a matter a fact, the cycle may be directly used for the development of any biologically active product as long as an activity screen is available.

Beyond the bioproduction of molecules, the cycle can also be useful for carrying out the engineering of commercially attractive nutrient degradation pathways, or for the development of biosensors of environmental pollutants or medical biomarkers (see Delépine B, Libis V, Carbonell P, Faulon JL, SensiPath: computer-aided design of sensing enabling metabolic pathways, Nucleic Acids Res., 8: 44: W226-31, 2016).

The expertise that we will accumulate in the practice of the MEM project will enable the partner ABOLIS to substantially improve its platform of development of strains. In return, ABOLIS will introduce MEM technology in the industrial biotechnology market and will be proactive in research and procurement of public-private contracts.

In the longer term, the «exploitation« task will allow the commercialization of pilot projects using MEM technology. Indeed, the interactions that ABOLIS maintains with several major industrial groups already show the need to use a cycle such as the one we propose for the biosynthesis of active products. This service we bring to the industry will be a source of income and job creation.

Publications
1. Delépine B, Libis V, Carbonell P, Faulon JL. SensiPath: computer-aided design of sensing-enabling metabolic pathways. Nucleic Acids Res. 8; 44:W226-31, 2016
2. Carbonell P, Gök A, Shapira P, Faulon JL. Mapping the patent landscape of synthetic biology for fine chemical production pathways. Microb Biotechnol. (5): 687-95., 2016
3. Libis V, Delépine B, Faulon JL. Sensing new c hemicals with bacterial transcription factors. Curr Opinion Microbiol. 33: 105-112, 2016.
4. Delépine B, Duigou, T, Carbonell P, Faulon JL. RetroPath2.0: a retrosynthesis workflow for metabolic engineers, submitted 2017

Présentations
1. Faulon, J.L., WISBI Synthetic Biology conference, University of Warwick, March 2016.
2. Trabelsi, H., Libis, V. Jacry, C, Sieskind, R. Faulon, J.L. Congrès GDR BioSynSys, Bordeaux, June 2016.
3. Delepine B, Duigou, T., Carbonnel, P. Faulon, J.L., Congrès GDR BioSynSys, Bordeaux, June 2016.
4. Libis, V., Delepine, B. Faulon, J.L. SEED (Synthetic Biology, Engineering, Evolution&Design), Chicago, IL, USA, July 18-21, 2016.
5. Faulon, J.L., Centre for Synthetic Biology & Innovation & Department of Life Sciences, Seminar series, Imperial College, London, July 2016.
6. Faulon, J.L., 3rd Synthetic Biology Congress, 20th-21st October 2016, London, UK
7. Faulon, J.L., ET/SynbiCITE Engineering Biology Conference 13-15 December 2016, London, UK.
9. Faulon, J.L., UPHAR Natural Products Meeting, Paris, May 22, 2017.
10. Delepine, B., Duigou, T., Carbonnel, P. Faulon, J.L., SB7.0 (Synthetic Biology 7.0), Singapore, June 2017.

Despite the growing number of chemicals successfully engineered in host organisms, bioproduction R&D is slow and expensive, as the process is mostly based on trial-and-error. To overcome this critical hindrance, we propose to implement a generic automated design-build-test and learn cyclic pipeline for the production of targeted chemicals. As an illustration, we will apply the pipeline for the metabolic engineering of a library of new antimicrobials against Gram-positive bacteria.

The pipeline comprises state-of-the-art bioproduction pathway design tools, robotized strain engineering, and high throughput product quantification via biosensors. The whole process is driven by an original computational machine learning component that determines the next set of constructions that needs to be processed by the pipeline with the goal of increasing product yield. In the specific approach we will be using, named active learning, a growing training set of experimental results is acquired on the fly in an iterative process between learning and measurements. The remarkable advantage of active learning is to yield performances comparable to classical machine learning with training sets sizes that can be several orders of magnitude smaller. Active learning can thus drastically reduce the cost of performing measurements, and in the present application significantly reduce the number of iterations for strain optimization.

We propose to apply the pipeline for the production of nutritional and antimicrobial flavonoids. Precisely, the pipeline will be run for four research objectives that complement each other: (RO1) to learn enzyme sequences that maximize flavonoid titers, (RO2) to determine enzyme expression levels limiting intermediates accumulation and increasing final product yields, (RO3) to regulate the expression of the genes of the host strain to optimize both growth and flavonoid titers, and (RO4) to produce novel flavonoid structures with maximal toxicity against Gram-positive bacteria.

While moving toward optimizing strains and producing novel flavonoids, our project will offer a technological rupture to industrial biotechnology where machine learning is driving experimental implementation and measurement. We anticipate this innovative solution will bring tremendous gains in throughput and speed.

The project will be illustrated with the production of a library of flavonoids, but the design-build-test-learn pipeline is general enough to be applied to other molecules of interest to the health, food, chemistry and energy industrial sectors, including commodity chemicals, and fine and specialty chemicals. Our approach could for instance be extended to other pharmaceutical applications beyond the search for antimicrobial activity, as long as there exists a screening method relevant to the problem. Beyond small molecule bioproduction a similar pipeline could also be implemented to metabolize alternative but commercially attractive feedstock and to develop biosensors for environmental pollutants.

The expertise gained in the project will drastically improve our SME partner strain development platform and in return the SME partner will bring the technology to the market seeking for industrial collaborations through a specific exploitation task. While we plan to release our computational methods to the academic community through web services, for specific applications, our know-how and software products will be packaged in an integrated pipeline and commercialized as a service. We foresee large industrial groups will want to customize development of the pipeline for their own application. The service we will provide to the industry will generate revenues and will also be a source for job creation.

Project coordinator

Monsieur Jean-Loup FAULON (Institut de Microbiologie de l’Alimentation au service de la Santé)

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.

Partner

iSSB Institut de Biologie Systémique et Synthétique
MICALIS Institut de Microbiologie de l’Alimentation au service de la Santé
ABOLIS Abolis Biotechnologies

Help of the ANR 517,206 euros
Beginning and duration of the scientific project: September 2015 - 48 Months

Useful links

Explorez notre base de projets financés

 

 

ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter