Methods

Cancer cell metabolite change upon CB-839 treatment

Drug treatment, metabolite extraction and LC-MS analysis

C4-2B prostate cancer cells were treated with 500 nM CB-839 for 24 hours. Following treatment, cells were harvested and washed with cold PBS. Cell lysates were prepared, and metabolites were extracted using a biphasic solvent system. Briefly, cells were lysed in cold methanol, followed by the sequential addition of chloroform and water. The mixture was centrifuged to achieve phase separation. Liquid chromatography - mass spectrometry (LC-MS) was performed using a high-resolution mass spectrometer in both positive and negative ionization modes. Peak intensities were normalized and used for downstream statistical and pathway enrichment analyses.

Principal component analysis

Principal Component Analysis (PCA) was performed to visualize the separation between control and treatment groups based on the metabolomics dataset, which includes over 1,000 metabolites. We selected PCA to reduce dimensionality and to visualize the data in two dimensions (PC1 and PC2), which are linear combinations of the original variables.

The 2D PCA score plot was generated using the web-based platform MetaboAnalyst 6.0. The statistical significance of the group separation was evaluated using PERMANOVA (Permutational Multivariate Analysis of Variance), based on Euclidean distance computed from the selected principal components.

Group distributions are visualized with shaded regions representing 95% confidence intervals.

Hierarchical clustering heatmap

Hierarchical clustering analysis was conducted using the MetaboAnalyst 6.0 platform to explore the similarity in metabolite profiles between the control and treated groups. The input metabolomics data were normalized and autoscaled to ensure comparability across features. Clustering was performed using Pearson correlation as the distance measure and the single linkage method to define cluster proximity.

A heatmap was generated using a red/green color contrast to visualize relative abundance patterns across samples. Both sample and metabolite (feature) names were displayed for clarity. An annotation bar was included to indicate sample groupings, and the overview mode was selected for visualization export. This analysis helped identify distinct clustering patterns corresponding to treatment conditions.

Bar plot of log2 fold changes of metabolites

Metabolite intensity data from control and CB-839-treated samples (4 replicates) were analyzed using Python (pandas, numpy, seaborn, and matplotlib libraries). The average intensity for each metabolite was calculated separately for the control and treatment groups. Log2 fold change (log2FC) was computed as the log2 of the ratio between the treatment and control group means. The resulting log2FC values were compiled along with metabolite identifiers (HMDB and KEGG IDs), sorted, and exported as a CSV file for downstream analysis including metabolite set enrichment. A bar plot was generated to visualize the top metabolites based on their log2FC values using Seaborn.

Metabolites set enrichment analysis (MSEA)

Metabolite set enrichment analysis was performed using a GSEA-style approach implemented with the Python package gseapy. Metabolites were ranked based on log2 fold change (log2FC) values calculated from CB-839-treated versus control samples. Metabolite names were standardized (case formatting, typo correction) and matched to curated metabolite sets corresponding to four major metabolic pathways: TCA cycle, glutaminolysis, glutathione (GSH) synthesis, and one-carbon metabolism. These sets were saved in a custom .gmt file format.

The ranked metabolite list and pathway sets were used as input for the prerank() function in gseapy, with 1,000 permutations, a seed of 42, and a minimum set size of 1. The minimum threshold was intentionally lowered to include small but biologically relevant pathways that would have otherwise been excluded due to limited metabolite coverage. Enrichment scores (ES), normalized enrichment scores (NES), nominal p-values, and FDR q-values were used to assess pathway significance.

Protein-ligand structure prediction with Chai-1

Chai-1: Deocding the molecular interactions of life

Among recent advances in macromolecular structure prediction, Chai-1 represents one of the most powerful multi-modal foundation models, capable of accurately modeling a broad range of biomolecular interactions, including protein-ligand, protein-protein, and protein-nucleic acid complexes. Unlike earlier methods that rely heavily on multiple sequence alignments (MSAs) or predefined structural templates, Chai-1 can operate in single-sequence mode without MSAs while maintaining high predictive accuracy. It also supports the inclusion of experimental restraints such as cross-linking mass spectrometry or epitope mapping, enabling further refinement of challenging binding interfaces. Notably, Chai-1 model is open-accessible, with both its modes are available via GitHub (https://github.com/chaidiscovery/chai-lab) and an interactive web server provided at (https://lab.chaidiscovery.com/). In this project, Chai-1 was employed to predict the ternary structure of GLS1 in complex with small-molecule inhibitors, focusing on three GLS1-targeting compounds that are currently under preclinical or clinical investigation.

GLS1 inhibitors in (pre)clinical trials

Three allosteric GLS1 inhibitors were used to predict protein-ligand interactions:

BPTES (top left)
CB-839 (bottom left)
IPN60090 (right)

Experimental structural data from X-ray crystallography are available for BPTES and CB-839, whereas no structural data currently exist for IPN60090.

Page updated

Report abuse