In-Situ NMR based Metabonomics of Microbial Secondary Metabolites

Microbial communities are known to produce an abundance of natural products, including secondary metabolites. Parts of the microbial community have been studied under in vitro conditions, which has led to a greater understanding of specific elements attributed to set experimental conditions. However, in vitro conditions fail to mimic metabolites being produced as a consequence of microbial interactions within natural niches. Therefore, if one were to gain an understanding of natural microbial interactions, it may lead to the uncovering of novel secondary metabolites. However, when biological complexity increases, so does the chemical complexity, leading to an abundance of challenges within In-situ detection.
Throughout the work of this thesis, nuclear magnetic resonance (NMR) spectroscopy was applied to generate metabolomic data which was utilized within a targeted approach. Two specific challenges were undertaken within targeted NMR in-situ detection. The first challenge was to reduce operator bias by developing automatic detection and uncertainty evaluation of complex metabolomic spectral samples. The second challenge was to ensure that a robust workflow with respect to data generation and analysis of data could be set up via a standardized metabolomic pipeline.
The first challenge led to the creation of the Python/Pytorch-based NMR-Onion framework (paper 2). The framework allowed for automatic detection, quantification, and uncertainty evaluation of detected peaks within complex spectral NMR samples. It was concluded that NMR-Onion could detect and evaluate signals across multiple signal-to-noise ratio values. In addition, the algorithm would make users aware of potential sample-to-sample variations in the form of potentially resolved peaks, reducing the risk of drawing false conclusions. The program was developed with a targeted approach in mind but has the potential to be utilized within nontargeted studies as well.
The second challenge was addressed by creating a metabolomic workflow. The workflow was generated by combining design of experiments (DoE), statistical quality control (SQC), minimal preprocessing, automatic detection (NMR-Onion), and statistical analysis. Methods for DoE and SQC were reviewed in paper 1, which resulted in two recommend workflows for metabolomics experiments involving DoE and SQC. Finally, the generated metabolomics workflow was utilized within a case study of pseudo-in-situ data. Here it was found that deconvoluting DoE-based NMR spectral data with NMR-Onion and subsequently analyzing the deconvoluted results via general linear effect models could be utilized to link targeted amplitudes to that of specific amino acids.
In conclusion, this thesis has developed a new tool of NMR-Onion to be utilized within automatic detection, quantifying, and evaluation of NMR signals. When generating metabolomic data, the framework should be paired with steps of the generated workflow to ensure optimal results. The future of the workflow generated in this thesis may be utilized to explore metabolomic in-situ data from natural microbial communities, potentially aiding in uncovering the diversity and functions of secondary metabolites. In addition, the NMR-Onion framework may be utilized within any area of in-situ detection such as disease diagnostics, agriculture, or food science.