Molecular networks for the exploration of large libraries of plant extracts and the targeted isolation of new anti-infective natural products

Tuberculosis has proven to be a particular threat to humanity, claiming about 1.5 million human lives each year. While treatments for tuberculosis exist, there is still an urgent need for new molecules to overcome the challenges of antibiotic resistance. Plants offer a great source of chemical diversity to generate new drug candidates, provided one can easily map the chemical content of such complex matrices. In this context, a library of about 1600 plant extracts was analysed through untargeted fragmentation experiments in high-resolution mass spectrometry (HRMS/MS), to generate a Molecular Network (MN) compiling the entire chemical content of the sample library. Theoretical spectral database and taxonomically informed metabolite annotation were used to dereplicate structures based on HRMS/MS data [1]. This study led to the creation of a virtual library of 37 304 compounds. The first operation consisted in exploring this virtual chemical library to spot structural analogues of benz[g]isoquinoline-5,10-dione, an active azaanthraquinone derivative found in a previous screen with an infection model system using the amoeba D. discoideum as a host and Mycobacterium marinum [2], which is closely related to M. tuberculosis, as the pathogen. The extract of Cananga latifolia (Annonaceae) was prioritised and efficient targeted isolation with high-resolution chromatography [3], yielded 12 compounds, 5 of which are closely related structural derivatives of the active azaathraquinone structure. The small amounts of pure compounds generated (100 µg) were screened on our model infection system to reveal their bioactive potential, with compounds such as “onychin” standing out and confirming the initial hypothesis.