Genome scale metabolic network Model (GEMs) reconstruction
De novo speci c GEMs from the three genomes used for the experiments are built follow-ing these steps : genome annotation, draft reconstruction, gap lling and curation, including speci c assessment of the organoleptic production pathways.
Interest of three di erent annotations tools
The goal is to get the most complete genome annotation possible as raw data for GEMs con-struction. Thus, three di erent annotation tools are used : Prokka, Eggnog and Microscope. The gure 12 illustrates the reason why three annotations tools were chosen. In this Venn diagram, in complement of a core set of 988 reactions that were consistently annotated by the 3 methods, each tool provides their own reactions that are not identi ed by the other tools, to be added to complete the network (126 from prokka, 336 from eggnog, 125 from microscope). In total, there is about 1600 reactions in the metabolic network of P. freudenreichii which is the expected order of magnitude for this kind of bacteria. Similar results are found for L. lactis and L. plantarum after adding the biomass objective function ( gures 13). The biomass objective function for each bacteria is added from VMH database and choosen according to the closest strain of each bacteria (see material and method for more details).
To assess the completeness of the draft model, the topological connectivity of the metabolic pathways involving milk compounds is systematically checked, and whether the draft model sustains metabolic uxes and growth is veri ed. To get ux, an objective function has been op-timized, in this case the biomass objective function. After this processing, di erences between topology and ux are observed which can be explainable because they do not use the same mathematical formalism. Indeed, topology is based on logical framework and ux is based on quantitative reasoning. In the model, metabolites must be evacuated into the extracellular compartment not to saturate and inhibit the growth of the bacteria. The table 4 illustrates the results after adding exchange reactions. For each step (draft merging, biomass objective function selection, exchange reaction de nition, gap- lling and curation), a new drafts is built to ensure traceability.
Gap- lling and curation
Some metabolites are producible in ux when the objective function is designed to maximize its production, but no growth is observed when the biomass reaction is set as objective func-tion. To improve the number of metabolites produced by ux, a completion step of metabolic network with Meneco is performed, and reactions from the union output are added ( gure 4). The union output corresponds to the union of minimal solution, that enables us to produce desired molecules. Some biomass compounds are still unproducible. It is likely due to the lack of information into the database MetaCyc for these metabolites or missing annotations in the genome. After removing them, bacterial growth is achieved for each bacteria FBA model. Before setting the community model, the production of the metabolites of interest is checked, i.e. organoleptic compounds as Propionate and 2MB for P. freudenreichii and Butanediol for L. lactis. After executing this pre-treatment, only 2MB was not produced. To solve it, a curation step is done using Meneco with 2MB as a target. Two reactions are found in the output and added to P. freudenreichii metabolic network.
Focus on organoleptic compounds At genome scale
Using Menepath and FVA, one path of synthesis of organoleptic compounds is computed and analysed at the genome scale ( gures 14, 15, 16). To obtain this result, the expected path of production of propionic acid is rstly checked from the literature (Wood-Werkman and TCA cycle). As the biosynthesis of 2MB and butanediol are not well known, we expected (based on biological experiences and biological knowledge) pyruvate to be a common precursor of these two compounds. Next step consists in retrieving these paths inside the metabolic network. For each path of production of an organoleptic compound, the table 5 shows the Menepath outputs, indicating the number of essential and alternative reactions (see introduction for more details about this notion). FVA analysis gives exactly the same essential and alternative reactions.
MisCoTo aims at characterizing exchanges within a bacterial community. It reveals two results: one was expected, unlike the other. First, we recovered the well known mechanism that P. freudenreichii is able to uptake lactate molecules produced by L. lactis or L. plantarum. Second, P. freudenreichii could produce propionate without lactate with an alternative pathway that is not the usually observed pathway : lactate -> pyruvate -> proprionate by the Wood-Werkman and TCA cycles. We could also say, from the postulate announced in introduction part, that pyruvate is a common precursor to produce 2MB and butanediol.
After retrieving metatranscriptomic data and averaging among replicates (see materials & methods part), I can obtain the proportion of reactions activated and the preferred time of propionate production in P.Freudenreichii. Results are presented in gure 18. There is one heatmap for each experiment (control, MT+. . . ), with, on the left, the heatmap showing the proportion of reaction activated in the di erent pathways of propionate production and on the right, heatmaps representing the preferential time steps of activation for each propionate pathway. Rows are Menepath solution, i.e. each metabolic pathway. Time steps are repre-sented on the bottom and on the right, the scale which represents the proportion of each time point in a solution at each time step. On the gure 17a, a metabolomics analysis of propionic acid was made (from TANGO project). Time is represented on the x-axis and the value of propionic acid on the y-axis. The di erent experiments are represented by the di erent colors (red for MT+, green for BT-, blue for ST- and violet for control). We can see that propionate appears after brining time step, consistently with the metabolomic data. The gure ?? shows that, after brining, there is no preferential time step to produced propionic acid, indicating that priopionate production goes on all along ripening, again consistently with the metabolomics. In average, the moulding and de-moulding steps are less represented (lower than 12%). It can con rm the trend of the curve between the moulding and after salting steps in the gure 17a. At the opposite, after brining and ageing four and seven weeks, genes are well represented (more than 24% by each time step). It is in accordance with the gure 17a.
Table of contents :
cheese manufactured at pilote-scale
Main features of bacterial metabolism during cheese production
Experimental design of the TANGO project
How to model them?
Modeling the bacterial metabolism
Metabolic network reconstruction
Mathematical models of metabolism
Dynamic modeling using ODE
Materials & Methods 20
FBA model construction and assessment
Omics data integration
Genome scale metabolic network Model (GEMs) reconstruction Interest of three different annotations tools
Gap-lling and curation
Focus on organoleptic compounds
At genome scale
At community scale
Integration of metatranscriptomics data
UNIVERSITY OF BORDEAUX