Skip to main content
Bioinformatics logoLink to Bioinformatics
. 2018 Feb 23;34(13):2319–2321. doi: 10.1093/bioinformatics/bty108

MARSI: metabolite analogues for rational strain improvement

João G R Cardoso 1, Ahmad A Zeidan 2,, Kristian Jensen 1, Nikolaus Sonnenschein 1, Ana Rute Neves 2, Markus J Herrgård 1,
Editor: Jonathan Wren
PMCID: PMC6022549  PMID: 29949953

Abstract

Summary

Metabolite analogues (MAs) mimic the structure of native metabolites, can competitively inhibit their utilization in enzymatic reactions, and are commonly used as selection tools for isolating desirable mutants of industrial microorganisms. Genome-scale metabolic models representing all biochemical reactions in an organism can be used to predict effects of MAs on cellular phenotypes. Here, we present the metabolite analogues for rational strain improvement (MARSI) framework. MARSI provides a rational approach to strain improvement by searching for metabolites as targets instead of genes or reactions. The designs found by MARSI can be implemented by supplying MAs in the culture media, enabling metabolic rewiring without the use of recombinant DNA technologies that cannot always be used due to regulations. To facilitate experimental implementation, MARSI provides tools to identify candidate MAs to a target metabolite from a database of known drugs and analogues.

Availability and implementation

The code is freely available at https://github.com/biosustain/marsi under the Apache License V2. MARSI is implemented in Python.

Supplementary information

Supplementary data are available at Bioinformatics online.

1 Introduction

Genome-scale metabolic models (GEMs) describe the biochemical reactions in an organism and their relation to the proteome and genome (McCloskey et al., 2013). These models comprehensively represent natural metabolism and they are useful for predicting the effect of metabolite analogues (MAs) as therapeutics (Agren et al., 2014; Kim et al., 2014).

Non-rational strategies such as mutagenesis and selection or laboratory evolution can be used to develop industrial strains when the use of recombinant DNA technology is not allowed due to regulations (Derkx et al., 2014; Hansen et al., 2017). MAs, inhibiting the enzymatic conversion of the target metabolite, act as metabolite knockouts and can be used as the selective pressure in non-rational strategies to shape the metabolism of microorganisms (Sørensen et al., 2016).

Here, we present software that implements workflows to identify metabolite knockouts instead of gene or reaction knockouts (Figure 1A). We also provide a pipeline to identify structural analogues for those targets.

Fig. 1.

Fig. 1.

Metabolite target identification workflow and examples of MA targets. (A) The workflow for identifying for metabolite knockouts and candidate MAs. (B) Comparison between the known MAs (columns 1 and 2) and the best MARSI hits (columns 3 and 4) used to calibrate the search parameters. We show the TC and the SS. We highlighted rows where the best MARSI hit and the known MA are the same

2 Materials and methods

The first workflow consists of systematically replacing reaction knockouts (identified by other strain design methods) by metabolite knockouts, until we can find metabolite targets that result in a similar flux distribution. The second workflow consists of searching for metabolite targets using heuristic optimization, without the need to specify reaction knockouts a priori. A metabolite knockout consists of blocking all reactions consuming a given metabolite, excluding transporters.

After identifying the metabolite targets, we search for MAs similar to them. We compiled a database of potential MAs from publicly available sources (see Supplementary Material). We use OpenBabel (O’Boyle et al., 2011) and RDKit (2017) (http://www.rdkit.org) to calculate the features used to compare candidate MAs to the target metabolite: number of atoms/bonds/rings, MACCs fingerprints, Tanimoto coefficient (TC) and structural similarity score (SS).

3 Results

We implemented a software package containing algorithms to generate strain design strategies using MAs. Our software could generate metabolite targets for a published knockout-based design (Harder et al., 2016). We also provide the tools to identify candidate MAs that could be used for implementation of the designs.

3.1 Identification of replacement targets

We used an experimentally validated strain design for itaconic acid production in Escherichia coli (Harder et al., 2016) and the E.coli GEM iJO1366 (Orth et al., 2014) to demonstrate the use of MARSI. MARSI identified acetyl-phosphate as a metabolite knockout target that can replace the Phosphotransacetylase (PTAr) reaction knockout and sustain the same flux for itaconic acid production (Table 1). Using a SS cut-off of 0.5 (see Supplementary Material), we found 182 MAs for acetyl-phosphate (Supplementary Table S1 shows the top 10 hits). More examples of replacement targets in other E.coli strain designs can be found in Supplementary Material.

Table 1.

Knockout replacements for the strain design

Non-replaced knockouts Replaced reaction Metabolite Original fitness New fitness
PTA2, ICL, ALDD2x, PYK, SUCOAS, GGGABADr PTAr Acetyl-P 0.001 0.001

We use Biomass Product Coupled Yield (Patil et al., 2005) as fitness measure. Reaction Ids: Phosphate acetyltransferase (PTA), Isocitrate lyase (ICL), Aldehyde dehydrogenase (ALDD2x), Pyruvate kinase (PK), Succinyl-CoA synthetase (SUCOAS) and Gamma-glutamyl-gamma aminobutyric acid dehydrogenase (GGGABADr).

3.2 Query calibration with known MAs

In order to validate the ability of MARSI to find known analogues for a target metabolite, we selected 42 known metabolite-MA pairs from the literature (Supplementary Table S3). We compared the structural features between the MAs and their target metabolites (Supplementary Fig. S1). We used a distance of 4 for the number of atoms, 3 for the number of bonds and 2 for the number of rings as our query cut-off. The TC cut-off changes dynamically with the size of the metabolites (see Supplementary Material). In Figure 1B, we show the SS and TC for different targets and their known analogues as well as the best hit analogue in the database. For most targets MARSI found candidate MAs that showed higher structural similarity to the target metabolite than the known analogue.

Supplementary Material

Supplementary Data

Acknowledgements

We would like to thank Miguel Campodonico for input on chemoinformatics tools.

Funding

This work has been supported by the Novo Nordisk Foundation. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 686070.

Conflict of Interest: none declared.

References

  1. Agren R. et al. (2014) Identification of anticancer drugs for hepatocellular carcinoma through personalized genome-scale metabolic modeling. Mol. Syst. Biol., 10, 721–721. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Derkx P.M. et al. (2014) The art of strain improvement of industrial lactic acid bacteria without the use of recombinant DNA technology. Microb. Cell. Fact., 13(Suppl. 1), S5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Hansen A.S.L. et al. (2017) Systems biology solutions for biochemical production challenges. Curr. Opin. Biotechnol., 45, 85–91. [DOI] [PubMed] [Google Scholar]
  4. Harder B.-J. et al. (2016) Model-based metabolic engineering enables high yield itaconic acid production by Escherichia coli. Metab. Eng, 38, 29–37. [DOI] [PubMed] [Google Scholar]
  5. Kim H.U. et al. (2014) Integrative genome-scale metabolic analysis of Vibrio vulnificus for drug targeting and discovery. Mol. Syst. Biol., 7, 460–460. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. McCloskey D. et al. (2013) Basic and applied uses of genome-scale metabolic network reconstructions of Escherichia coli. Mol. Syst. Biol., 9, 661. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. O’Boyle N.M. et al. (2011) Open Babel: an Open chemical toolbox. J. Cheminform., 3, 1–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Orth J.D. et al. (2011) A comprehensive genome-scale reconstruction of Escherichia coli metabolism–2011. Mol. Syst. Biol., 7, 535. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Patil K.R. et al. (2005) Evolutionary programming as a platform for in silico metabolic engineering. BMC Bioinformatics, 6, 308. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. RDKit. (2017) Cheminformatics and machine Learning Software. http://www.rdkit.org (31 August 2017, date last accessed) [Google Scholar]
  11. Sørensen K.I. et al. (2016) Enhancing the sweetness of yoghurt through metabolic remodeling of carbohydrate metabolism in Streptococcus thermophilus and Lactobacillus delbrueckii subsp. bulgaricus. Appl. Environ. Microbiol., 82, 3683–3616. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Data

Articles from Bioinformatics are provided here courtesy of Oxford University Press

RESOURCES