Researchers at the University of Texas Southwestern Medical Center and collaborating institutions have developed a high-throughput untargeted metabolomics platform that integrates liquid chromatography-mass spectrometry with machine learning algorithms to characterize the metabolic landscape of bladder cancer. Published in the journal Cancer & Metabolism, the study demonstrates that this combined approach can identify distinct metabolic signatures that differentiate bladder cancer tissues from adjacent normal or benign tissues, as well as distinguish between muscle-invasive and non-muscle-invasive disease stages. The work represents a comprehensive effort to map metabolic vulnerabilities in bladder cancer, offering potential avenues for improved diagnostic stratification and the identification of therapeutic targets.
The study employed a robust analytical workflow using the Thermo Scientific IQ-X Orbitrap Tribrid mass spectrometer coupled with a Vanquish UHPLC system. To capture a broad spectrum of metabolites, including both polar and non-polar compounds, the team utilized four distinct separation and detection methods: reverse-phase chromatography in positive and negative ion modes, and hydrophilic interaction liquid chromatography in positive and negative ion modes. This multi-modal approach was designed to enhance detection sensitivity and analytical depth, allowing for the comprehensive profiling of the global metabolome in biological samples.
Clinical samples for the investigation consisted of human bladder cancer tissues and adjacent normal or benign tissues, obtained in a de-identified manner from the Cooperative Human Tissue Network, the University of Maryland Baltimore, and the University of Texas Southwestern Medical Center. The tissue samples were stored at -140 degrees Celsius until analysis. Approximately 10 milligrams of tissue were homogenized in a methanol-water solution, followed by the addition of methanol-acetonitrile to precipitate proteins. The resulting supernatant was dried and reconstituted for mass spectrometry analysis, with pooled quality control samples injected throughout the acquisition process to monitor system performance and ensure data reliability.
Metabolite identification was achieved by leveraging an in-house retention time-based MS/MS spectral library built from 630 reference compounds, alongside commercial databases such as mzCloud, NIST 2020, and the Human Metabolome Database. After rigorous data filtering and quality assurance, the platform identified 2,590 metabolites or compounds across all analytical methods. The identification process categorized metabolites based on the strength of their spectral matches, with higher priority given to those matching both retention time and MS/MS spectra in the in-house library as well as external databases. This multi-layered annotation strategy aimed to maximize the accuracy of metabolite identification in the complex biological matrix of tumor tissues.
When comparing adjacent normal or benign tissues, comprising 29 samples, to bladder cancer tissues, comprising 65 samples, the researchers identified a total of 1,475 altered metabolites with a false discovery rate of less than 0.25. Among these, 812 metabolites were decreased and 663 were increased in the bladder cancer tumor samples relative to the adjacent benign tissues. Pathway enrichment analysis revealed that fatty acid and xenobiotic metabolism showed the most significant enrichment, indicating major alterations in lipid metabolism. Other notably enriched pathways included glycolysis and oxidative phosphorylation, reflecting significant metabolic reprogramming, as well as adipogenesis and bile acid metabolism.
The study further examined the metabolic differences between muscle-invasive bladder cancer and non-muscle-invasive bladder cancer. In this comparison, which included 41 muscle-invasive samples and 10 non-muscle-invasive samples, a total of 16 altered metabolites were identified. The majority of these, 13 metabolites, were increased in muscle-invasive samples, while three were decreased. Pathway analysis highlighted significant enrichment in xenobiotic metabolism, reactive oxygen species signaling, glycolysis, and bile acid metabolism. Additionally, glutathione metabolism was prominently enriched across multiple pathway databases, emphasizing its role in detoxification and redox homeostasis in the context of more aggressive disease.
To complement the statistical analysis of differentially expressed metabolites, the team applied machine learning models to the untargeted metabolomics dataset. Three classification models were evaluated: k-nearest neighbor, random forest, and support vector machines. For the distinction between muscle-invasive and non-muscle-invasive bladder cancer, all models demonstrated robust performance, with the support vector machine achieving the highest area under the receiver operating characteristic curve of 0.813. The k-nearest neighbor and random forest models achieved scores of 0.750 and 0.766, respectively. The superior performance of the support vector machine led to its selection for determining the most informative metabolic markers for this specific classification task.
Feature importance analysis revealed that as few as eight metabolites were sufficient to effectively classify muscle-invasive bladder cancer from non-muscle-invasive disease. Among the top informative metabolites, 3-hydroxyglutaric acid, 3-methylglutaric acid, mesaconic acid, and 2-methylacetoacetic acid were found to be decreased, while guanine was increased in patients with muscle-invasive disease. The reduction of these organic acids may indicate disturbances in branched-chain amino acid catabolism and related pathways, potentially reflecting a shift toward alternative metabolic routes that support tumor growth and survival in invasive cancers.
In the comparison of bladder cancer tissues against adjacent normal or benign tissues, the machine learning models also performed strongly, with all three achieving an area under the curve above 0.90. The k-nearest neighbor model was selected for this classification due to its slightly superior performance, with an area under the curve of 0.938. Analysis of the top informative features identified metabolites such as adipoylglycine, S-adenosylmethionine, and 2′-O-methylguanosine as increased in bladder cancer patients. The accumulation of methylated nucleosides and elevated S-adenosylmethionine suggests increased methylation activity, which is commonly associated with epigenetic dysregulation in tumor cells. These findings point to reprogrammed metabolic pathways that may contribute to cancer progression.
The study acknowledges several limitations, primarily the relatively small sample size, which requires validation with larger cohorts to confirm the identified metabolic signatures. Additionally, the platform detected metabolites with high mass accuracy but limited spectral confirmation, which may require further validation in future studies. Despite these constraints, the research highlights the power of combining untargeted metabolomics with machine learning to map the metabolic landscape of bladder cancer. By identifying novel dysregulated pathways, including xenobiotic metabolism, glutathione metabolism, and glycolysis, the study offers critical insights into the metabolic vulnerabilities of the disease.
The integration of high-resolution mass spectrometry with advanced computational modeling addresses key challenges in metabolomics, such as high dimensionality and biological variability. The development of an in-house spectral library enhanced the ability to identify metabolites that might be missed by standard databases alone. This approach not only validates previous findings regarding the elevation of xenobiotic metabolism in bladder cancer but also provides a deeper understanding of the broader metabolic framework underlying the disease. The identification of specific metabolic markers with strong potential for early detection and characterization of invasiveness suggests that this integrative analytical pipeline could accelerate the development of precision diagnostics.
By elucidating the metabolic reprogramming associated with bladder cancer progression, the study provides a foundation for future therapeutic strategies targeting these metabolic vulnerabilities. The distinct metabolic signatures identified between different disease stages offer a basis for patient stratification, which could inform treatment decisions and improve clinical outcomes. As the field of cancer metabolism continues to evolve, the application of untargeted metabolomics combined with machine learning represents a significant step forward in understanding the complex interplay of genetic, environmental, and lifestyle factors that drive tumor biology.
The findings underscore the importance of comprehensive metabolic profiling in cancer research, moving beyond targeted approaches to capture the dynamic and diverse aspects of tumor metabolism. The robustness of the quality control measures and the reproducibility of the data, evidenced by high correlation coefficients in pooled samples, support the reliability of the identified metabolic alterations. This work contributes to the growing body of evidence that metabolic reprogramming is a hallmark of tumor progression and therapeutic resistance, offering new perspectives on how metabolic interventions might be leveraged in the fight against bladder cancer.
Subject of Research: Metabolic profiling and machine learning classification of bladder cancer tissues
Article Title: Leveraging untargeted metabolomics in combination with machine learning to uncover novel insights into bladder cancer
Article References: Kamal, A. H. M., Putluri, V., Gandhi, T., Ambati, C. S. R., Amara, C. S., Kami Reddy, K. R., Spradlin, M. L., Koirala, A., Jorvekar, S. B., Grimm, S. L., Fu, D., Parsawar, K., Jong, F. D., Beecher, C., Sen, S., Lerner, S. P., Siddiqui, M. M., Lotan, Y., Eberlin, L. S., … Putluri, N. (2026). Leveraging untargeted metabolomics in combination with machine learning to uncover novel insights into bladder cancer. Cancer & Metabolism, 14(1), Article 8. https://doi.org/10.1186/s40170-026-00427-4
Image Credits: AI Generated
DOI: 10.1186/s40170-026-00427-4
Keywords: bladder cancer, metabolomics, machine learning, mass spectrometry, metabolic reprogramming, biomarkers, xenobiotic metabolism, precision medicine, tissue profiling, Cancer & Metabolism, Leveraging, untargeted
News Source: Nathaniel Bowman. (November 10, 2026). Metabolomics and Machine Learning Map Bladder Cancer Metabolism. Scienmag.
