New tool unravels biosynthesis pathways in plants without prior knowledge

Plants produce a large number of specialised metabolites. Think of terpenoids, alkaloids, and phenols. These contribute to the growth and protection of the plant. They help, for example, to attract beneficial microbes. In addition, these metabolites are used in medicines, cosmetics, and agricultural products. But how and when the plant produces these specialised metabolites is largely unknown. This hampers the use of these compounds.

Until now, researchers have needed relatively extensive prior knowledge to unravel the steps of the biosynthesis of a specific metabolite. For instance, a gene that researchers know plays a role in the production of the metabolite. Researchers then use this gene as bait to identify the next step in the biosynthesis. In this way, researchers can often manage to expose the biosynthetic route step by step.

Untargeted search

‘But we wanted to be able to conduct an untargeted search for biosynthetic routes in plants involved in attracting endophytic, beneficial, bacteria’, says Kumar Saurabh Singh, lead author of the paper and a researcher at Maastricht University. Singh began the research in the group of Professor Marnix Medema at Wageningen University & Research. ‘There’, Singh explains, ‘the recently graduated Hernando Suárez had already proposed in a concept how you could integrate transcriptomic and metabolomic data. Using this concept, I began building a tool for that untargeted search.’

This ultimately resulted in MEANtools (Metabolite Anticipation tools), a set of five scripts that the user runs sequentially. In addition to specific transcriptomic and metabolomic data, MEANtools also integrates general knowledge about chemical structures and enzymatic reactions. The latter two are retrieved by MEANtools from the LOTUS and RetroRules databases. The former is a repository of natural product structures, whilst RetroRules contains general descriptions of enzymatic reactions. Using all these data, MEANtools can then predict the steps of the biosynthesis. It presents these in the form of a table, but can also visualise them.

In short, MEANtools provides the researcher — who is searching for the steps of the biosynthesis of a particular compound — with candidate enzymes and their associated substrates and products. The researcher can then test these in the laboratory.

Hundreds of reaction steps

‘The biggest obstacle we encountered in developing MEANtools was ordering the predicted reaction steps’, says Singh. Because it is an untargeted analysis, MEANtools quickly predicts hundreds of reaction steps. ‘To bring order to this, we formulated a method to assign a probability score to reaction steps.’ The higher the score, the greater the likelihood that the reaction actually takes place. ‘We then organise the reactions based on correlation score, the number of enzymes associated with a substrate and product, and the probability score’, says Singh.

In addition, the team built in a targeted prediction function for reaction steps. This allows users to have MEANtools specifically predict the routes in which one or more metabolites or genes of interest participate. Here too, there is a good chance that the output MEANtools produces will explode. ‘We advise keeping the input number of genes or metabolites as low as possible’, says Singh.

The test

To test it, the team used the paired transcriptomic and metabolomic dataset from the work of Elizabeth Sattely’s group (Stanford). They had used it to determine the steps of the biosynthesis of falcarindiol. ‘We wanted to investigate whether MEANtools could predict the reaction steps of falcarindiol biosynthesis using this dataset’, says Singh. ‘Of course, Sattely already had knowledge about falcarindiol-related biosynthetic genes, so the research was somewhat more targeted. But MEANtools predicted five of the seven reaction steps without any prior knowledge.’ And not unimportantly, each of the predicted reaction steps had a high probability score.

But the true power of the untargeted approach really comes to the fore in the by-catch of this test. ‘Through the untargeted approach, we saw that many more biosynthetic routes were activated as a result of a bacterial or fungal infection’, says Singh. Such as those of hydroxycinnamic acid amide, α-tomatine, and chlorogenic acid.

‘As far as I know, MEANtools is the only tool that can predict the steps of a biosynthetic route in an untargeted and systematic manner’, says Singh. The software is open source and freely available. Furthermore, the use of MEANtools is not limited to predicting the steps of biosynthetic routes in plants. Because MEANtools makes use of linked transcriptomic and metabolomic data supplied by the user, researchers can also use it for other organisms.

Singh, K.S. et al. (2025) PLOS Biology, DOI: 10.1371/journal.pbio.3003307