Frullania genus is the most complex among liverworts, known for its rich chemical diversity. Its most distinctive compounds are sesquiterpene lactones and bibenzyls, but their biosynthetic pathways are mostly unknown. One of the first steps of terpenoids biosynthesis are catalyzed by terpene synthases.

Recently, plant terpene synthases called microbial terpene synthase-like enzymes (MTPSLs) have been identified. My thesis will investigate terpene biosynthesis in Frullania liverworts, focusing on the functional and biochemical characterization of their MTPSLs. This includes assessing enzyme activity in heterologous systems like yeast and in planta, as well as in vitro, followed by detecting their products. I will also explore the structure-function relationship of these enzymes by introducing mutations in their conserved regions.

Better understanding of liverwort terpenoid biosynthesis could offer insights into the evolutionary path of MTPSLs in non-seed plants.

Terpenes are synthesized from the precursors isopentyl diphosphate (IPP) and dimethylallyl diphosphate (DMAPP), which are naturally derived from the methylerythritol phosphate (MEP) pathway or the mevalonate (MVA) pathway.

This project aims to enhance terpenoid production through optimization of the MEP pathway with systems biology tools. The expression of the improved terpenoid pathway in combination with terpene synthases (TPS) will then create a highly efficient microbial production host. This host can then be adapted to different TPS and even previously engineered TPS to expand the terpene product range or enhance terpenoid production further.

This project explores the potential of terpene synthases (TPS) for designing terpenoid scaffolds via a combination of computational and experimental methodologies. Specifically, it seeks to uncover new biocatalysts within the sesqui- and diterpene synthase families and investigate their physicochemical properties by integrating laboratory work and in silico analysis.

Our group has deve-loped EnzyDock which tends to dock neutral and carboca-tionic ligands into en-zyme’s binding po-ckets. This tool would be used in ModBioTerp to un-derstand and im-prove TPSs: terpene synthases.

The goal of my research is to model the mechanisms of terpene biosynthesis using deep learning methods. I will achieve this by assembling a dataset and designing deep learning models for predicting the reaction substrates and products directly from the amino acid sequences of terpene synthases.

To this end, I will start with building a comprehensive database describing the reaction mechanisms of terpene synthases characterized to date. The database will contain all the required data for the deep learning models which will be designed followingly. I will then develop a deep learning model using state-of-the-art methods such as transformer neural networks for predicting the substrates, products, and reaction mechanisms of terpene synthases directly from their amino acid sequences. The model will consider vector representations of each protein generated by large protein language models, as well as structural domains detected in each protein structure.

I expect this model to predict if a protein is a terpene synthase given its amino acid sequence. In the case of TPS, the model will also predict the details of the enzymatic activity, such as substrates, products, and reaction mechanisms. Last but not least, I will also be working on developing a generative algorithm for designing artificial TPSs with a desired function.

Apis mellifera. ©Georges-Alexandre Cotnoir

Terpenoid secondary metabo-lites are widely distributed in nature across diverse life forms and are generally utilized for defense and communication. They have many interesting industrial applications in phar-maceutics, cosmetics, in the food industry and others.

Terpene synthases (TPSs) produce a wide variety of natural compounds used in medicine, agriculture, and cosmetics, among other industries. However, predicting the products of a given TPS is challenging due to the diversity of their structures and mechanisms. Traditional methods for product prediction are time-consuming and often require a curated experimental approach for each novel enzyme.In contrast, state-of-the-art machine learning models offer a solution by leveraging structural and sequence features from previously characterized TPS enzymes.

In my project, I combine data from structural conformations, sequences, substrate docking, and other computational simulations to identify subtle structural differences between TPS variants that result in specific products. These enhanced predictions of TPS specificity, based on structure-function relationships, can improve our understanding of terpene synthases and their potential for biotechnological applications.

As part of the ModBioTerp project, I focus on evolutionary studies and the design of terpene synthases. These enzymes can synthesise a wide range of products, a trait linked to the complexity of their reaction mechanisms. The impact of protein structure on function remains largely unknown for terpene synthases, limiting their industrial applications.