Discovery of compound 1105486 as a selective inhibitor of B4GALT1: potential for pancreatic cancer therapy

Abstract

Targeting aberrant β-1,4-galactosyltransferase 1 (B4GALT1) activity represents an unexplored therapeutic avenue for pancreatic ductal adenocarcinoma (PDAC). Guided by a concise active-learning structure-based workflow, we rapidly triaged 22.6 million compounds and singled out 1105486 for experimental characterization. In PANC-1 cells, the molecule suppressed proliferation with an IC₅₀ of 19.8 ± 1.3 µM, while hTERT-HPNE epithelial cells retained >95% viability at concentrations up to 80 μM, indicating an encouraging initial safety window. Mechanistically, 1105486 engages the UDP-galactose pocket through stable hydrogen bonds to ARG187 and GLU313, a binding mode corroborated by 1 µs molecular-dynamics simulations and MM/GBSA energetics. Unlike previously reported glycosyltransferase inhibitors, which often lack selectivity and may affect multiple family members, 1105486 specifically targets B4GALT1 with high selectivity, occupying its unique catalytic pocket. To our knowledge, 1105486 constitutes the first reported small-molecule inhibitor of B4GALT1 and establishes a tractable chemical scaffold for optimization toward sub-micromolar potency and in vivo evaluation. The compound's selective cytotoxic profile, promising physicochemical properties, and the potential for further development highlight its in vivo efficacy and its role as a lead candidate for the next-generation of glycosylation-directed therapeutics for PDAC.

Keywords: B4GALT1; PDAC; active-learning; computational-experimental integration; cytotoxicity safety window.

FIGURE 1

Structural modeling and molecular dynamics simulation of the native B4GALT1 complex. (A) Time-dependent profiles of Root Mean Square Deviation (RMSD), Solvent Accessible Surface Area (SASA), and Radius of Gyration (Rg) throughout the 500 ns molecular dynamics simulation of the native B4GALT1 complex. (B) Two-dimensional projection of principal component analysis (PCA) along PC1 and PC2, overlaid with the corresponding Gibbs free energy landscape. Red dots denote low-energy conformational states. (C) Representative minimum-energy structure extracted from the global free energy basin. (D) Binding interactions between B4GALT1 and the pentasaccharide ligand. (E) Binding interactions between B4GALT1 and the UDP-hexose donor (UDH) ligand. Key interacting residues are labeled for downstream docking constraint definition.

FIGURE 2

Iterative model optimization and compound prioritization through active learning-driven virtual screening. (A–C) Model performance metrics across three active learning iterations: R² (A), RMSE, and MAE (C). Progressive improvements were observed with each training round. (D-G) Summary of the fina lfiltering process applied to the top 1.6 million compounds screened, leading to the selection of four high-confidence B4GALT1 inhibitor candidates.

FIGURE 3

Physicochemical descriptor profiles of selected B4GALT1 inhibitor candidates. (A–D) Radar plots illustrating key physicochemical properties of compounds 1105486 (A), 826788, 910875 (C), and 1503232 (D). Properties include LogP, LogD, LogS, TPSA, molecular weight, and others relevant to drug-likeness evaluation. Shaded regions represent the optimal value ranges.

FIGURE 4

Molecular dynamics simulation of the B4GALT1–1105486 complex. (A,B) Ligand RMSD profiles for compounds 1105486 and 1503232 over 1 μs of simulation, indicating stability differences. (C,D) Time-resolved interaction analysis of 1105486 with B4GALT1, including total number of contacts and interaction type frequency. (E) Structural interaction model of 1105486 based on residues with interaction frequency >40% during the MD trajectory. Key residues involved in hydrogen bonds, π–π stacking, and water-bridged interactions are highlighted.

FIGURE 5

Structural and energetic validation of key residues involved in 1105486 binding via alanine scanning and binding pose metadynamics. (A) Per-residue MM/GBSA energy decomposition showing binding free energy contributions of six key residues. (B) CVRMSD trajectories of wild-type and alanine mutants during BPMD simulations, indicating the stability of the binding pose. (C) BPMD poseScore and persScore analysis for wild-type and mutant complexes. (D) Time-resolved MMGBSA ΔG values over 1 μs cMD simulations for wild-type and six mutant complexes, reflecting energetic stability.

FIGURE 6

Selective cytotoxicity of compound 1105486 against pancreatic cancer cells. (A) Dose–response curve showing cell viability of hTERT-HPNE (gray) and PANC-1 (blue) cells following 24 h treatment with compound 1105486 (0–80 μM), as assessed by the CCK-8 assay. Data are presented as mean ± SD (n = 3). (B) Bar graph comparing the viability of hTERT-HPNE and PANC-1 cells at each concentration of compound 1105486. Statistical significance was determined by independent-samples t-tests. *P < 0.05; ****P < 0.0001.