If you blend several fruits together, you can measure the overall nutritional content of the shake, but not how much one fruit contributed to each nutrient. Metabolomics works in a similar way: researchers grind up tissue samples, such as tumours, then profile the molecules in the resulting ‘soup’ to look for changes that could be used as diagnostic markers or treatment targets.
“Conventional bulk metabolomics is sensitive but tells us nothing about where each metabolite comes from,” said Lit-Hsin Loo, a Senior Principal Investigator at the A*STAR Bioinformatics Institute (A*STAR BII).
To address this lack of spatial information, newer techniques like mass spectrometry imaging can visualise the distribution of molecules on a sample. However, existing analysis methods often face difficulties extracting patterns from the mixed compositions of metabolites on these images. That was the issue the Precision Medicine in Liver Cancer across an Asia-Pacific Network (PLANet) programme ran into, having generated bulk and spatial metabolomics data from hepatocellular carcinoma (HCC) samples.
The PLANet team, led by Pierce Chow at the National Cancer Centre Singapore and Yulan Wang from Nanyang Technological University, Singapore, then collaborated with Loo to develop a new multi-modal approach called Spatially guided MEtabolomics (SgME) profiling. “We overcame the problem by focussing on metabolites that were co-detected in mass spectrometry and co-localised in tissue regions with clear histological features,” said Joey Lee, a Research Officer in Loo’s group.
The researchers trained a machine learning model to learn these spatial patterns and map tumour samples into six ‘metabolic neighbourhoods’ on the mass spectrometry images. Their analysis showed that low-grade tumour regions expressed markers for cell proliferation and rapid growth. In the aggressive, high-grade areas, HCC cells had reduced proliferation and lipid metabolism while immune signalling increased around them.
Moreover, SgME profiling revealed that over half of the top metabolites identified in ground-up tissue actually came from the samples’ non-cancerous parts, challenging their suitability as biomarkers. “We also found metabolites that arise in early-stage tumour areas but collapse in the dead tissue areas, which would have cancelled each other out in bulk profiling,” added Lee.
Instead of treating HCC as a uniform entity, Loo noted that these metabolic maps could guide more precise and effective treatment selection. For example, a drug that targets lipid metabolism could hit the low-grade tumour areas but miss the aggressive ones, whereas immunotherapies might work better for these high-grade regions.
The researchers next aim to apply SgME to more tumour types and improve its resolution to enable profiling at the single-cell level. “Our long-term goal is to develop a virtual liver tissue model that combines multi-omics data to capture how cells behave and test out different treatment combinations, accelerating drug discovery and development,” said Loo.
The A*STAR-affiliated researchers contributing to this research are from the A*STAR Bioinformatics Institute (A*STAR BII).
