The importance of verifying biological activity predictions in metabolomics studies of natural products: A case study on antimicrobial properties of Cannabis sativa (hemp)

Informatics-guided approaches involving untargeted mass spectrometry metabolomics to predict active compounds in natural products mixtures have become increasingly common. With such strategies, it is sometimes possible to target active compounds for isolation early in the fractionation process, thereby reducing effort and increasing hit rate success. However, such approaches require follow up studies to address the potential problem of false correlations. To demonstrate this, we employed the botanical Cannabis sativa (hemp) as a test case. A C. sativa extract was fractionated in several stages, and the ability of the fractions to inhibit the growth of Methicillin-resistant Staphylococcus aureus (MRSA) was evaluated in broth microdilution assays. Metabolomics data were collected for the extract and fractions using high performance liquid chromatography coupled to high resolution mass spectrometry on an Orbitrap mass spectrometer, and selectivity ratio analysis was employed as a statistical tool to predict the active compound from two different rounds of fractionation of the C. sativa extract. From the early stage of fractionation, we predicted that the cannabinoids cannabidiol (1, CBD) and cannabidiolic acid (2, CBDA) were major active constituents. These predictions, when verified with follow up minimum inhibitory concentration (MIC) assays of the pure compounds, proved to be accurate. However, in a later stage of fractionation using the same statistical approach, the previously reported C. sativa constituents N-trans-p-coumaroyltyramine (3, pCT) and N-trans-feruloyltyramine (4, FT) were predicted to be responsible for activity. Follow up assays with these pure compounds revealed that they possess no direct or synergistic antimicrobial activity against MRSA. This study highlights how statistical predictions of active compounds in untargeted metabolomics studies are inherently correlative and emphasizes the need for follow up studies to verify the accuracy of such predictions.