Transcriptome-Weighted Network Analysis of Glycosphingolipid Metabolism in Neuroblastoma

Background Introduction

Glycosphingolipids (GSLs) are a class of membrane lipids composed of a ceramide backbone linked to glycan moieties, widely present in the nervous system. They play crucial roles in cell signaling, cell-cell interactions, and tumorigenesis. Neuroblastoma (NB) is the most common extracranial solid tumor in children, characterized by high biological and clinical heterogeneity. Dysregulation of glycosphingolipid metabolism in neuroblastoma is closely related to tumor progression, prognosis, and treatment response, particularly GD2, a specific glycosphingolipid, which has become a target for immunotherapy in neuroblastoma. However, the complexity of glycosphingolipid metabolism makes its analysis extremely challenging, especially in inferring metabolic activity from transcriptomic data.

To address this challenge, a research team from the University Medical Center of the Johannes Gutenberg-University Mainz published a study in 2024 titled “Unraveling the glycosphingolipid metabolism by leveraging transcriptome-weighted network analysis on neuroblastic tumors”. The study proposed a novel method based on transcriptomic data and metabolic network topology to differentiate the various series of glycosphingolipid metabolism in neuroblastoma. This research not only provides new analytical tools for glycosphingolipid metabolism in neuroblastoma but also offers potential applications for metabolic studies in other cancers.

Source of the Paper

The paper was co-authored by Arsenij Ustjanzew, AnneKathrin Silvia Nedwed, Roger Sandhoff, Jörg Faber, Federico Marini, and Claudia Paret, and published in the journal Cancer & Metabolism. The research team is affiliated with the University Medical Center of the Johannes Gutenberg-University Mainz, the German Cancer Research Center, and the Center for Pediatric Hematology/Oncology at the University Medical Center Mainz. The paper was published online on October 11, 2024, and is open access under the Creative Commons Attribution 4.0 International License.

Research Workflow

1. Data Acquisition and Preprocessing

The research team utilized two publicly available RNA sequencing (RNA-seq) datasets: one from the GEO database (GSE147635), containing 6 ganglioneuroma (GN) and 15 neuroblastoma (NB) samples, and another from the UCSC Xena database, comprising 154 samples, including 25 ganglioneuroblastoma (GNB) and 129 neuroblastoma (NB) samples. Data preprocessing included normalization, log transformation, and pseudocount addition to ensure data comparability.

2. Metabolic Network Construction

Based on four glycosphingolipid metabolic pathways (hsa00600, hsa00601, hsa00603, hsa00604) from the Kyoto Encyclopedia of Genes and Genomes (KEGG), the research team constructed a metabolic network graph. The graph used metabolites as nodes and reactions as edges, integrating transcriptomic data with reaction activity scores (RAS) through Gene-Protein-Reaction (GPR) association rules.

3. Reaction Activity Score (RAS) Calculation

RAS was used to quantify the activity of each metabolic reaction across different samples. The research team calculated RAS values for each reaction based on gene expression data using GPR rules. For reactions involving multiple genes, RAS values were computed using logical operators (AND or OR). Ultimately, each sample’s metabolic network graph was assigned a weighted adjacency matrix representing the activity of each reaction.

4. Transition Probability Matrix (TP) Adjustment

To differentiate the four series of glycosphingolipid metabolism (0-, A-, B-, and C-series), the research team proposed three RAS adjustment methods based on transition probability (TP): - Simple TP Adjustment: Adjusted RAS values based on local network topology. - Recursive TP Adjustment: Adjusted RAS values by recursively calculating TP values of predecessor nodes. - Path TP Adjustment: Adjusted RAS values based on TP values along paths from a specific node (e.g., lactosylceramide) to target nodes.

5. Unsupervised Learning and Clustering Analysis

The research team used the Uniform Manifold Approximation and Projection (UMAP) algorithm for dimensionality reduction of adjusted RAS values and performed clustering using the Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) algorithm. By iteratively computing UMAP 1000 times, the team evaluated clustering stability and identified characteristic reactions for each cluster.

6. Differential Reaction Activity Analysis

The research team compared reaction activity differences between different tumor types (e.g., NB vs. GNB, MYCN-amplified vs. non-amplified NB) using the Kolmogorov-Smirnov (KS) test to calculate significance for each reaction, followed by Benjamini-Hochberg (BH) correction for multiple testing.

Key Findings

1. Differentiation of Glycosphingolipid Metabolic Series

By adjusting RAS values, the research team successfully differentiated the four series of glycosphingolipid metabolism. The results showed that GNB tends to express complex glycosphingolipids of the A-series, while NB (especially MYCN-amplified NB) tends to express simple glycosphingolipids of the B-series (e.g., GD2). This finding aligns with previous studies, indicating that MYCN amplification is associated with simplified glycosphingolipid metabolism.

2. Correlation Between MYCN Expression and Glycosphingolipid Metabolism

The study found significant correlations between MYCN gene expression and glycosphingolipid metabolism-related genes (e.g., B3GALT4 and ST8SIA1). In MYCN-amplified NB samples, ST8SIA1 expression was higher, while B3GALT4 expression was lower, suggesting that MYCN may regulate glycosphingolipid metabolism through these genes.

3. Unsupervised Learning Reveals NB Subgroups

Through unsupervised learning, the research team identified two subgroups within NB samples: one characterized by high expression of fucosyltransferase (FUT) genes and the other by high expression of sulfatide metabolism-related genes. The existence of these subgroups highlights the significant heterogeneity in glycosphingolipid metabolism in NB.

4. Differential Reaction Activity Analysis

Differential reaction activity analysis revealed that reactions involved in GD2 synthesis (e.g., R05940) were significantly more active in MYCN-amplified NB samples compared to GNB samples. Additionally, reactions involved in A-series glycosphingolipid synthesis were more active in GNB samples, further supporting the mature phenotype of GNB.

Conclusions and Significance

The study proposed a novel method based on transcriptomic data and metabolic network topology, successfully differentiating the various series of glycosphingolipid metabolism in neuroblastoma. The research not only revealed the relationship between MYCN amplification and simplified glycosphingolipid metabolism but also identified NB subgroups through unsupervised learning, providing new insights for personalized treatment of neuroblastoma. Furthermore, the method has broad applicability and can be used to study other metabolic pathways involving low-specificity enzymes.

Research Highlights

1. Innovative Methodology: The research team, for the first time, integrated transcriptomic data with metabolic network topology, proposing three RAS adjustment methods to successfully address the challenge of differentiating glycosphingolipid metabolic series.
2. Clinical Relevance: The study revealed the relationship between MYCN amplification and simplified glycosphingolipid metabolism, offering new biomarkers for prognosis assessment and targeted therapy in neuroblastoma.
3. Broad Applicability: The method is not only applicable to neuroblastoma but also holds potential for metabolic studies in other cancers, demonstrating wide-ranging applications.