Identification of Rocaglate Acyl Sulfamides as Selective Inhibitors of Glioblastoma Stem Cells

Abstract

Glioblastoma (GBM) is the most aggressive and frequently occurring type of malignant brain tumor in adults. The initiation, progression, and recurrence of malignant tumors are known to be driven by a small subpopulation of cells known as tumor-initiating cells or cancer stem cells (CSCs). GBM CSCs play a pivotal role in orchestrating drug resistance and tumor relapse. As a prospective avenue for GBM intervention, the targeted suppression of GBM CSCs holds considerable promise. In this study, we found that rocaglates, compounds which are known to inhibit translation via targeting of the DEAD-box helicase eIF4A, exert a robust, dose-dependent cytotoxic impact on GBM CSCs with minimal killing of nonstem GBM cells. Subsequent optimization identified novel rocaglate derivatives (rocaglate acyl sulfamides or Roc ASFs) that selectively inhibit GBM CSCs with nanomolar EC₅₀ values. Furthermore, comparative evaluation of a lead CSC-optimized Roc ASF across diverse mechanistic and target profiling assays revealed suppressed translation inhibition relative to that of other CSC-selective rocaglates, with enhanced targeting of the DEAD-box helicase DDX3X, a recently identified secondary target of rocaglates. Overall, these findings suggest a promising therapeutic strategy for targeting GBM CSCs.

Figure 1

(A) Nature-produced (1–4) and synthetic (5–9) rocaglates. (B) Comparative dose–response for killing of CSC (red) and non-CSC (blue) populations for selected rocaglates.

Scheme 1. (A) Synthesis of Rocaglaic Acids and Derived N-Acyl Derivatives; (B) Substrate Scope for β-Lactone Ring Opening (One-Step Yields from Precursors Are Denoted in Parentheses)

Scheme 2. Chiral Resolution of (+)- and (−)-Rocaglaic Acids; The Absolute Stereochemistry of the Derived Salt 24 Was Confirmed by X-ray Crystallography (Inset)

Figure 2

Comparative dose–response for killing of CSC (red) and non-CSC (blue) populations for selected rocaglaic acid and N-acylated derivatives showing enhanced selective CSC killing for C4′-brominated congeners (right column) over their C4′-methoxy-substituted counterparts (left column).

Figure 3

(A) Comparative PISA profiling. Volcano plots depicting differentially stabilized GBM0308 lysate proteins in the presence of the natural rocaglate (−)-1 (left), derivative (−)-7 (middle), and derivative (−)-20 (right). Compounds were tested at a concentration of 10 μM. Significance thresholds were set to FDR < 0.01 with |log₂(FC)| > 0.25. (B) Extracted individual log₂(FC) (N = 4) from PISA assays for DDX3X and eIF4A1 stabilization in the presence of compounds (−)-1, (−)-7, and (−)-20. (C) Structures of compounds (−)-1, (−)-5, (−)-7, and (−)-20 and their corresponding EC₅₀ values against GBM CSCs. (D) Change in polarization (ΔmP) obtained with DDX3X:FAM-labeled r(AG)₈ and eIF4A1:FAM-labeled r(AG)₈ by FP assay in the presence of (−)-5, (−)-7, and (−)-20 (10 μM). Background-subtracted values were generated using a DMSO control (N = 3; lines indicate mean ± SEM). (E) In vitro translation experiments testing the bicistronic mRNA reporter construct FF–HCV-Ren. The translation of the FF cistron is cap-dependent, while translation of the HCV-Ren cistron is driven off an HCV-IRES, rendering translation of this cistron cap-independent. (F) Extracted individual log₂(FC) (N = 3) for DDX3 and eIF4A paralog stabilization in the presence of compounds (−)-10, (−)-11, (−)-16, and (−)-20. Horizontal bars indicate mean values. (G) Comparative PISA profiling of compounds (−)-10, (−)-11, (−)-16, and (−)-20 in GBM0308 cell lysate, with volcano plots depicting differentially stabilized lysate proteins in the presence of C4′-methoxy ((−)-10, (−)-16) (left) and C4′-brominated ((−)-11, (−)-20) (right) compounds tested at a concentration of 10 μM. Significance thresholds were set to FDR < 0.01 with |log₂(FC)| > 0.25.

Figure 4

(A) X-ray cocrystal structure of RocA(1):eIF4A1:poly(AG) complex (PDB entry 5ZC9). (B) Glide docking pose for (−)-20 into the 5ZC9 eIF4A1:poly(AG) structure (Glide G_score = −11.7 kcal/mol), showing a “canonical” rocaglate binding pose. (C) Top-ranked induced-fit docking pose for (−)-20 (G_score = −7.0 kcal/mol), modeled from an X-ray crystal structure of DDX3X bound to a RNA:DNA hybrid (PDB entry 7LIU), modified with a single-residue mutation (C704A). (D) IFD pose (G_score = −6.1 kcal/mol) for (−)-20 and the 7LIU-C704A receptor showing interaction between the ionized rocaglate acyl sulfamide and the cationic DDX3X residue Arg363. Movement of the Arg363 side chain creates a cleft into which the C4′-brominated B ring inserts.

Figure 5

(A) Representative images showing propidium iodide (PI) and Hoechst staining in GBM0308 neurospheres treated with either DMSO or varying doses of compound (−)-20 for 72 h, acquired using the Celigo image cytometer. Merge images (PI + Hoechst) are also shown. BF, bright-field. (B) Representative images showing caspase 3/7 staining in GBM0308 neurospheres treated with DMSO, 0.1 μM (−)-20, or 1 μM (−)-20 acquired at the indicated times using the Celigo image cytometer. (C) Representative flow cytometry dot plots showing annexin-V/PI staining following treatment with either DMSO or (−)-20 (1 mM) for 12 and 24 h, respectively (top), and quantification of percent annexin-V- and PI-positive cells normalized to respective DMSO controls (bottom). Data are presented as mean ± SD from two independent experiments, and p values were calculated using a two-tailed unpaired t test (**, p < 0.01). (D) GSEA multibubble plot. The color of the bubble represents the comparison, the size of the bubble represents the significance, and the x axis represents the normalized enrichment score (NES). All gene sets with FDR < 0.05 and |LFC| > 0.585 in any of the five comparisons are included here. (E) Z-score heat map visualizing differentially expressed genes (DEGs) involved in CSC stemness. DEG signatures in GBM0308 stem cells treated with either DMSO or compound (−)-20 for the indicated times and concentrations are shown. Only significant genes (FDR < 0.05 and |LFC| > 0.585) are presented here. Hierarchical clustering was performed using the complete linkage method and 1-Pearson as the distance. The genes were then classified into two clusters based on the dendrogram.