During this reporting period, we contributed to multiple work packages (WPs). In WP2, we developed a novel method for the simultaneous detection of metabolites from key metabolic pathways, which previously required different methods. We improved metabolite identification by creating a machine learning-based model, trained and evaluated on a large set of datasets. We conducted a study on authentic metabolic standards to improve detectability and metabolite identification. In WP4, we developed computational methods to build a single-cell atlas of metabolic states from cells treated with reference metabolic modulators. These methods help explain the metabolic mode of action of drugs by comparing the single-cell metabolic profiles from treated cells against the atlas. In WP5, we performed a comparative analysis of thousands of public datasets from METASPACE. As planned, we applied the computational methods for single-cell metabolomics data analysis developed in other WPs to this large single-pixel dataset and identified metabolic markers.
Overviewing the results of the entire project, our key achievement was the development, evaluation, and demonstration of SpaceM, a method for detecting metabolites from single cultured cells. Linking this work to specific WPs: In WP1, we developed cell culturing methods compatible with SpaceM for various cell types and created binary co-culture models to evaluate SpaceM's single-cell resolution. We also prepared CD4+ T cells for single-cell metabolomics, a feat not previously reported due to their non-adherent nature and small size. In collaboration, we cultured patient primary cells for SpaceM. In WP2, we developed several protocols for sensitive and high-coverage MALDI-imaging mass spectrometry from cultured cells. We improved metabolite identification by developing a machine learning-based method, enhancing our previous rule-based approach. We also integrated LC-MS/MS results into MALDI-imaging mass spectrometry for better structural elucidation and resolution of isomers and isobars indistinguishable with MALDI-imaging. In WP3, we streamlined and implemented procedures for correlating microscopy with MALDI-imaging mass spectrometry and developed software for automated batch analysis of this step in SpaceM. In WP4, we developed computational methods to assess variability between technical replicates and intensities of individual metabolites. We demonstrated that SpaceM achieves a coefficient of variation (CV) below 20% for many detected metabolites, an important milestone for reproducibility. Additionally, we developed a full data analysis pipeline covering quality control, unsupervised and supervised data analysis, including the search for subpopulations and markers. We evaluated different strategies for correcting batch effects and compensating for other confounding factors. In WP5, we conducted a comparative analysis of thousands of public datasets from METASPACE. Despite the challenges of dataset variability and quality, we performed cross-dataset single-pixel analysis on high-quality representative datasets with over 1 million single-pixel profiles. We successfully applied our computational methods for single-cell metabolomics data analysis and identified metabolic markers.
Regarding the exploitation of the results, this project helped us secure further funding from several sources: from Michael J. Fox Foundation to characterize the metabolism of single iPSC-derived neurons of Parkinson's patients; from the Swiss National Science Foundation (SNF) to develop a single-cell metabolomics approach for therapy selection in prostate cancer patients; from the La Caixa Foundation to map how nutrients affect brain function and behavior; and from the Chan Zuckerberg Initiative (CZI) to map fatty acid synthesis in the Drosophila fly.
Furthermore, we began commercializing the SpaceM method developed in this project with support from ERC Proof of Concept funding and a major grant (€2.4 million) from the BioInnovation Institute in Copenhagen, Denmark.
Finally, the journal Nature featured the SpaceM method in their "Seven Technologies to Watch" highlight in 2023.
