Targeting bioactive compounds in natural extracts - Development of a comprehensive workflow combining chemical and biological data.

In natural product drug discovery, several strategies have emerged to highlight specifically bioactive compound(s) within complex mixtures (fractions or crude extracts) using metabolomics tools. In this area, a great deal of interest has raised among the scientific community on strategies to link chemical profiles and associated biological data, leading to the new field called "biochemometrics". This article falls into this emerging research by proposing a complete workflow, which was divided into three major steps. The first one consists in the fractionation of the same extract using four different chromatographic stationary phases and appropriated elution conditions to obtain five fractions for each column. The second step corresponds to the acquisition of chemical profiles using HPLC-HRMS analysis, and the biological evaluation of each fraction. The last step evaluates the links between the relative abundances of molecules present in fractions (peak area) and the global bioactivity level observed for each fraction. To this purpose, an original bioinformatics script (encoded with R Studio software) using the combination of four statistical models (Spearman, F-PCA, PLS, PLS-DA) was here developed leading to the generation of a "Super list" of potential bioactive compounds together with a predictive score. This strategy was validated by its application on a marine-derived Penicillium chrysogenum extract exhibiting antiproliferative activity on breast cancer cells (MCF-7 cells). After the three steps of the workflow, one main compound was highlighted as responsible for the bioactivity and identified as ergosterol. Its antiproliferative activity was confirmed with an IC50 of 0.10 µM on MCF-7 cells. The script efficiency was further demonstrated by comparing the results obtained with a different recently described approach based on NMR profiling and by virtually modifying the data to evaluate the computational tool behaviour. This approach represents a new and efficient tool to tackle some of the bottlenecks in natural product drug discovery programs.

Automatised pharmacophoric deconvolution of plant extracts - application to Cinchona bark crude extract.

We present a development of the "Plasmodesma" dereplication method [Margueritte et al., Magn. Reson. Chem., 2018, 56, 469]. This method is based on the automatic acquisition of a standard set of NMR experiments from a medium sized set of samples differing by their bioactivity. From this raw data, an analysis pipeline is run and the data is analysed by leveraging machine learning approaches in order to extract the spectral fingerprints of the active compounds. The optimal conditions for the analysis are determined and tested on two different systems, a synthetic sample where a single active molecule is to be isolated and characterized, and a complex bioactive matrix with synergetic interactions between the components. The method allows the identification of the active compounds and performs a pharmacophoric deconvolution. The program is freely available on the Internet, with an interactive visualisation of the statistical analysis, at https://plasmodesma.igbmc.science.

Automatic differential analysis of NMR experiments in complex samples.

Liquid state nuclear magnetic resonance (NMR) is a powerful tool for the analysis of complex mixtures of unknown molecules. This capacity has been used in many analytical approaches: metabolomics, identification of active compounds in natural extracts, and characterization of species, and such studies require the acquisition of many diverse NMR measurements on series of samples. Although acquisition can easily be performed automatically, the number of NMR experiments involved in these studies increases very rapidly, and this data avalanche requires to resort to automatic processing and analysis. We present here a program that allows the autonomous, unsupervised processing of a large corpus of 1D, 2D, and diffusion-ordered spectroscopy experiments from a series of samples acquired in different conditions. The program provides all the signal processing steps, as well as peak-picking and bucketing of 1D and 2D spectra, the program and its components are fully available. In an experiment mimicking the search of a bioactive species in a natural extract, we use it for the automatic detection of small amounts of artemisinin added to a series of plant extracts and for the generation of the spectral fingerprint of this molecule. This program called Plasmodesma is a novel tool that should be useful to decipher complex mixtures, particularly in the discovery of biologically active natural products from plants extracts but can also in drug discovery or metabolomics studies.