Structural Features of Small Molecules Targeting the RNA Repeat Expansion That Causes Genetically Defined ALS/FTD

Abstract

Genetically defined amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD), collectively named c9ALS/FTD, are triggered by hexanucleotide GGGGCC repeat expansions [r(G₄C₂)^exp] within the C9orf 72 gene. In these diseases, neuronal loss occurs through an interplay of deleterious phenotypes, including r(G₄C₂)^exp RNA gain-of-function mechanisms. Herein, we identified a benzimidazole derivative, CB096, that specifically binds to a repeating 1×1 GG internal loop structure, 5′CGG/3′GGC, that is formed when r(G₄C₂)^exp folds. Structure−activity relationship (SAR) studies and molecular dynamics (MD) simulations were used to define the molecular interactions formed between CB096 and r(G₄C₂)^exp that results in the rescue of disease-associated pathways. Overall, this study reveals a unique structural feature within r(G₄C₂)^exp that can be exploited for the development of lead medicines and chemical probes.

Graphical Abstract

Amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) are fatal neurodegenerative diseases characterized by a loss of motor function and cognitive impairment.¹ The lack of effective treatments, despite multiple clinical trials,² can be attributed to heterogeneous manifestation of the disorders³ and an incomplete understanding of their etiology and development.^4,5 Recently, long stretches of hexanucleotide GGGGCC repeat expansions [hereafter r(G₄C₂)^exp] in intron 1 of chromosome 9 open reading frame 72 (C9orf72) were identified as the main cause of genetically defined ALS and FTD, collectively named c9ALS/FTD.⁶ Whereas neurologically healthy individuals have up to 30 r(G₄C₂) repeats,¹ affected individuals house hundreds to thousands of such consecutive repeats in C9orf72. Neuronal death in c9ALS/FTD patients is triggered by an interplay of pathological mechanisms⁷ originating from both C9orf72 protein loss-of-function⁸ and r(G₄C₂)^exp-mediated RNA gain-of-function.⁹ In particular, r(G₄C₂)^exp accumulates in the nucleus of a subset of central nervous system (CNS) cells in nuclear foci by sequestering various RNA binding proteins (RBPs) such as heterogeneous nuclear ribonucleoprotein H1 (hnRNP H1).¹⁰ The depletion of the RBP pool triggers RNA splicing defects, as well as dysfunction in RNA metabolism and nucleocytoplasmic transport.¹¹ Moreover, when exported to the cytoplasm, r(G₄C₂)^exp undergoes an unconventional translation process, named repeat associated non-ATG (RAN) translation, which occurs in the absence of an AUG start codon.¹² RAN translation generates toxic dipeptide repeats (DPRs), which can cause formation of stress granules, defects in autophagosome formation, and induction of endoplasmic reticulum (ER) stress.¹³ Thus, an efficient therapeutic approach to ameliorate the RNA-mediated pathological mechanisms is to directly target r(G₄C₂)^exp.

Thus far, antisense oligonucleotides (ASOs) and small molecules have been proposed as promising therapeutic solutions.^14–16 However, there are challenges associated with ASOs, such as limited cellular and tissue penetrance, off-target effects, suboptimal physicochemical properties, and toxicity in clinical settings.¹⁷ Alternatively, small molecules offer several advantages including (i) synthetic ease from commercially available starting materials, (ii) tunable physicochemical properties, and (ii) temporal control of biological activity via structure–activity relationships (SAR).^18,19 Indeed, recent studies showed that small molecules avidly bind two structures adopted by r(G₄C₂)^exp that are in equilibrium, a G-quadruplex²⁰ (stabilized in the presence of K⁺ ions) and a hairpin that forms a periodic array of 1 × 1 GG internal loops²¹ (stabilized in the presence of Li⁺ or Na⁺ ions), and alleviate c9ALS/FTD pathologies.^21–23 Great effort has been expended to determine how each of these two structures contributes to r(G₄C₂)^exp-mediated molecular defects. For example, hnRNPH1 has been shown to bind the G-quadruplex form in vitro and nuclear foci sequestering hnRNP H1 have been observed in patient-derived cells.¹⁰ We have previously shown that the r(G₄C₂)^exp hairpin, not the G-quadruplex, undergoes RAN translation in vitro and that this hairpin structure forms in cells.²¹ Moreover, we previously reported a selective small molecule that targets the hairpin form of r(G₄C₂)^exp and inhibits the production of poly(GP) DPR by stabilizing the 1 × 1 GG internal loops within r(G₄C₂)^exp.²¹ Interestingly, under pathogenically relevant lengths (≥30 repeats), the formation of the G-quadruplex comprises a minor population of the folding landscape, and the hairpin form predominates.²⁴ Additionally, a structural transition from the hairpin form to the G-quadruplex form is unlikely at pathogenic lengths due to its high activation energy barrier.^21,24 Collectively, these studies suggest that both structural forms contribute to c9ALS/FTD pathology, and small molecules selectively targeting each will allow these contributions to be teased out. Herein, we studied the ligandability of the r(G₄C₂)^exp hairpin form, identifying chemotypes that potently and specifically bind to structural features uniquely present within this structure and inhibit disease-relevant pathologies.

Results and Discussion

High-Throughput Screening Identifies CB096 As a Specific r(G₄C₂)₈ Binder

The ligandability of the r(G₄C₂)^exp hairpin form was explored using a two-pronged screening strategy (Figure 1A). First, a TO-PRO-1 (TO1) fluorescent dye displacement assay^22,25,26 was used to identify small molecules within an RNA-focused library that bind a r(G₄C₂)₈ hairpin. The library comprises 3271 nitrogen-rich heterocyclic compounds featuring drug-like properties.²⁷ By monitoring the decrease in fluorescence upon incubation with the small molecules (tested at 100 μM), 26 derivatives were identified that induced ≥75% displacement of TO1 from r(G₄C₂)₈ (0.8% hit rate; Z-factor²⁸ = 0.6; Figure 1B). False positives were removed by applying stringent criteria: (i) compounds must lack interference with the signal readout due to intrinsic fluorescence; (ii) compounds must not aggregate at concentrations up to 100 μM; and (iii) compounds must exhibit dose-dependent displacement of TO1 (Supplementary Figure 1). A total of 13 hits were validated, seven of which were substituted benzimidazoles (Supplementary Figure 1).

Figure 1.

Identification of CB096, a lead molecule that binds to the r(G₄C₂)₈ hairpin. (a) Graphical representation of the two-pronged screening strategy leading to the identification of CB096. TO-PRO-1 (TO1) dye displacement assay was performed on 3271 compounds in 384-well format at 100 μM (single dose). Derivatives showing ≥75% TO1 quenching of the initial fluorescence (blue dots) were selected and further validated. A total of 13 derivatives met the criteria and were dose dependently evaluated for binding the hairpin form of Cy5-r(G₄C₂)₈ via microscale thermophoresis (MST), of which CB096 is the best binder. (b) Results of the high throughput TO1 dye displacement assay. Compounds that exhibited ≥75% TO1 displacement were considered hits (upper orange dashed line). (c) CB096 specifically binds r(G₄C₂)₈ (sense) over r(G₂C₄)₈(antisense) and r(GGCC)₁₀ (base pair) control RNAs. EC₅₀ values were determined by plotting the changes in normalized fluorescence upon dose-dependent addition of CB096. The data points are reported as the mean values ± SD and are representative of two independent experiments each performed in duplicate.

In the second step of the screening funnel, binding of the 13 hits to the r(G₄C₂)₈ hairpin was assessed via microscale thermophoresis (MST).^29–31 Of these, the most promising was benzimidazole derivative CB096, which had an EC₅₀ of 33(±9) μM in the TO1 displacement assay and an EC₅₀ of 19(±3) μM from binding studies (Figure 1C and Supplementary Figure 2A). Circular dichroism (CD) spectroscopy was performed to determine the binding stoichiometry of CB096 to the r(G₄C₂)₈ hairpin. The addition of CB096 triggered a dose-dependent increase of signals at 235 and 298 nm that reached saturation upon the addition of 4 equiv of CB096 (Supplementary Figure 4).

Interestingly, the RNA-focused library houses 262 benzimidazoles (8%), of which three have a nitro group (−NO₂) in position 5 (CB096, CB141, and compound 11) and two were hits in the primary screen (CB096 and CB141). Of the 47 benzimidazoles with a −NO₂ in a different position, none were hits in the TO-1 displacement primary screen nor were different scaffolds such as thiazoles, quinazolines, and indazoles that contain −NO₂ groups (n = 12).

The binding selectivity of CB096 was then assessed relative to r(G₂C₄)₈ (antisense) and the base paired construct r(GGCC)₁₀, for which only a minor change in thermophoresis and no saturation were detected, respectively (Figure 1C and Supplementary Figure 2B and C). Benzimidazoles are a privileged chemotype for binding RNA^26,27,32 and have been previously shown to bind 1 × 1 nucleotide internal loops.³³ We therefore studied the binding of CB096 to other 1 × 1 nucleotide internal loops, including 1 × 1 AA loops in Cy5-r(CAG)₁₂ and 1 × 1 UU internal loops in Cy5-r(CUG)₁₂. Indeed, no saturable binding was observed for either loop (Supplementary Figure 3). Collectively, these data indicate that CB096 is indeed specific for r(G₄C₂)^exp.

CB096 Alters the Dynamics of the 5′CGG/3′GGC Site within r(G₄C₂)^exp

To investigate the exact residues involved in CB096 binding to the r(G₄C₂)₈ hairpin, 1D imino NMR spectroscopy was employed. The r(G₄C₂)₈ RNA displays a total of six 1 × 1 GG internal loops containing two types of binding sites with different closing base pairs: three 5′GGC/3′CGG loops (highlighted in blue; G highlights the 1 × 1 GG non canonical base pair; Figure 2A) and three 5′CGG/3′GGC loops (highlighted in orange, Figure 2A). Each type of loop alternates to form a periodic array of 5′GGCCGG/3′CGGGGC. The imino ¹H NMR spectrum recorded at 25 °C exhibits a peak at 13.4 ppm, indicating a stable GC/CG base pair, and a less intense peak at 12.5 ppm, which could indicate that the base pair is more dynamic and hence less stable (i.e., faster exchange rate with H₂O; Figure 2A). The closing pairs of the two types of 1 × 1 GG internal loops within 5′GGCCGG/3′CGGGGC are not identical, and these differences could be distinguished by small molecule ligands. Indeed, the addition of CB096 only broadened the GC/CG base pair at 12.5 ppm dose-dependently, reaching saturation at 4 equiv of CB096, in agreement with the CD spectroscopy data (Figure 2A). Further, resonances at 13.4 and 12.5 ppm were also observed for the r(G₄C₂)₄ hairpin, and the resonance at 12.5 ppm also broadened upon the addition of CB096, saturating at 1 equiv of CB096 (Supplementary Figure 5). To confirm that the GG loop is required for binding, we studied the binding of CB096 to a fully paired RNA construct, r(G₂C₂)₁₀, and to a model of the C9orf72 antisense repeat, r(G₂C₄)₈. This RNA, complemented with a GAAA tetraloop for loop stability, adopts a rigid hairpin structure that maintains a similar nucleotide composition to r(G₄C₂)₄. No changes were observed in the 1D imino proton spectra of either RNA upon the addition of up to 3 equiv of CB096 (Figure 2B and Supplementary Figure 6).

Figure 2.

CB096 mode of binding investigated via 1D imino proton NMR. (a) CB096 broadens the imino proton signal of GC/CG base pairs at 12.5 ppm within the r(G₄C₂)₈ hairpin. Blue labels indicate 5′GGC/3′CGG internal loops, whereas the orange labels indicate 5′CGG/3′GGC loops (G represents the G residues within the 1 × 1 GG internal loops). (b) CB096 does not affect the 1D imino proton signal of GC/CG base pairs at 12.5 ppm within the base pair control r(GGCC)₁₀ RNA. Unique GC/CG base pairs within r(GGCC)₁₀ are highlighted in green and yellow. The 1D NMR spectra are representative of two independent experiments. (c) Simplification of r(G₄C₂)₈ secondary structure to the r(G₄C₂)₂ duplex featuring two 5′GGC/3′CGG (blue squares) and one 5′CGG/3′GGC (orange square). (d) The 1D imino proton spectra of r(G₄C₂)₂ duplex recapitulates the 1D imino proton peak pattern of r(G₄C₂)₄ and r(G₄C₂)₈. CB096 (1 equiv) broadens the peak at 12.5 ppm. 1D NMR spectra are representative of two independent experiments.

Interestingly, imino proton signals of equal intensities were observed for r(GGCC)₁₀ at 13.4 and 12.5 ppm, but the addition of CB096 did not lead to the broadening effect observed for r(G₄C₂)₄ and r(G₄C₂)₈. Since the secondary structures of r(G₄C₂)₈ and r(GGCC)₁₀ share the same array of GC/CG base pairs and vary solely by the 1 × 1 GG noncanonical base pairs in the former, the loops may trigger unique conformational features within the r(G₄C₂)₈ hairpin. These data provide structural support for specific targeting of the c9ALS/FTD r(G₄C₂)^exp RNA hairpin by CB096 observed in binding studies. Moreover, the substoichiometric ratio between CB096 and the number of 1 × 1 GG internal loops in r(G₄C₂)₄ and r(G₄C₂)₈ indicates that this compound might specifically bind to one type of GG loop and its closing pairs (5′GGC/3′CGG vs 5′CGG/3′GGC) and/or exhibit multiple binding modes. Indeed, previous studies have shown that both the loop nucleotides and closing base pairs influence conformation.^34,35

We therefore designed and studied a simplified duplex derived from r(G₄C₂)₄ and r(G₄C₂)₈, r(G₄C₂)₂, that features two 5′GGC/3′CGG and one 5′CGG/3′GGC and thus has two full r(G₄C₂) repeat sequences on each strand (Figure 2C). The imino-proton spectrum of r(G₄C₂)₂ was similar to r(G₄C₂)₄ and r(G₄C₂)₈ exhibiting peaks at both 13.4 and 12.5 ppm, with the latter completely broadened out upon the addition of 1 equiv of CB096 (Figure 2D). Notably, free energy minimization using the RNAstructure software³⁶ predicted that r(G₄C₂)₂ can also form a 2 × 2 GG internal loop containing structure with the same folding energy as the 1 × 1 GG internal loop (Supplementary Figure 7A). However, the 1D imino proton spectrum of a 2 × 2 GG containing duplex model did not recapitulate either the peak pattern or the specific signal broadening upon the addition of CB096 (Supplementary Figure 7B).

To precisely assign the two imino proton signals to their corresponding GC/CG base pairs within r(G₄C₂)₂, cytosine (C) residues were exchanged for 5-fluoro C (^5FC). Indeed, ¹⁹F has been used as a label in numerous 1D and 2D NMR spectroscopic studies of nucleic acids^37,38 due to its comparable size to ¹H, high detection sensitivity, and straightforward incorporation into oligonucleotides.^39–41 The ^5FC substitution (i) decreases the stability of GC/CG base pairs due to diminished electron density at the N(3) of C, which reduces the strength of the hydrogen bond with the N(1)–H of the paired G, and (ii) shifts the imino proton signal of the G’s N(1)–H to which it is paired upfield by ∼0.4 ppm (Figure 3A, left).^37,40 By systematically substituting nonequivalent C residues, the 1D imino proton peaks at 13.4 and 12.5 ppm can be deconvoluted, allowing for identification of CB096’s binding site (Figure 3A, right).

Figure 3.

CB096 affects the dynamics of 5′CGG/3′GGC, but not 5′GGC/3′CGG within r(G₄C₂)₂. (a) Graphical representation of the strategy to assign the imino proton signals emanating from guanine (G) residues in base pairing interactions within r(G₄C₂)₂. Briefly, replacing cytosine (C) by 5-fluorocytosine (^5FC) shifts imino proton signals of G residues in GC base pairs upfield by ∼0.4 ppm (orange triangle), compared to unlabeled base pair peaks (blue triangle). (b) 1D imino proton spectra of ^5FC labeled r(G₄C₂)₂ duplex constructs. Blue color highlights the imino proton peak corresponding to the GC base pairs while those labeled in orange are from ^5FC-labeled pairs. The 1D NMR spectra are representative of two independent experiments. (c) 1D imino proton spectra of ^5FC labeled r(G₄C₂)₂ duplex constructs in complex with CB096 (1 equiv). Black arrows indicate the imino proton peaks that broadened out upon CB096 treatment. The data indicate that CB096 binds 5′CGG/3′GGC and increases the dynamics of the base pairs (likely do not form or form transiently). The 1D NMR spectra are representative of two independent experiments. (d) CB096 binds to 5′CGG/3′GGC within Cy5-r(CGG)₁₆ (orange squares) with an EC₅₀ of 20(±3) μM, as measured by MST. The EC₅₀ value was determined by plotting the changes in normalized fluorescence upon dose-dependent addition of CB096. The data points are reported as the mean values ± SD and are representative of two independent experiments performed in duplicate.

Following this strategy, C5, C6, or C11 within the r(G₄C₂)₂ duplex were replaced by ^5FCs (Figure 3B). The ^5FC5 (green label, Figure 3B) and ^5FC11 (yellow label, Figure 3B) substitutions triggered the emergence of new imino proton peaks at 13.0 ppm, whereas the introduction of ^5FC6 (purple label, Figure 3B) shifted the 12.5 ppm peak to 12.1 ppm. Thus, the imino proton signal at 13.4 ppm in r(G₄C₂)₂ was from the guanine (G) residues of the base pairs formed between C5–G9 and C11–G3, both forming closing pairs for the 5′GGC/3′CGG loops (bold lettering indicates the base pair referred to in the motif). Likewise, the peak at 12.5 ppm was assigned to G8, which forms a base pair with C6, closing the 5′CGG/3′GGC loop (bold lettering indicates the base pair referred to in the motif).

We next determined which peaks changed upon the addition of CB096 to each construct. For the ^5FC5– and ^5FC11–r(G₄C₂)₂ duplexes, the addition of 1 equiv of CB096 broadened the signal at 12.5 ppm, assigned to G8, which forms the closing base pairs of 5′CGG/3′GGC, while having no effect on the signals at 13.0 and 13.4 ppm (Figure 3C). Likewise, the addition of CB096 to ^5FC6-r(G₄C₂)₂ causes broadening of the G8 resonance (12.1 ppm), without affecting the other peaks (Figure 3C). Collectively, these data suggest that CB096 binds and selectively affects the dynamics of the 5′CGG/3′GGC but not those of the 5′GGC/3′CGG. Interestingly, nearly complete broadening of G8’s resonance occurred with 1 equiv of CB096, consistent with the construct containing only one 5′CGG/3′GGC and two 5′GGC/3′CGG motifs. This observation was further validated by binding studies of CB096 to Cy5-r(CGG)₁₆ and Cy5-r(GGC)₁₆ using MST. Experimental results indicated that CB096 binds Cy5-r(CGG)₁₆ with an EC₅₀ of 20(±3) μM (Figure 3D), whereas no change in thermophoresis was observed for Cy5-r(GGC)₁₆ (Supplementary Figure 8). Thus, binding of CB096 to r(G₄C₂)^exp is exclusively due to its interaction with the 5′CGG/3′GGC motif within the r(G₄C₂)^exp hairpin.

Previously reported experimental data^42,43 indicate that the 5′GGC/3′CGG is more stable than 5′CGG/3′GGC, which could make the latter more amendable to structural rearrangement required for CB096 binding. The free energies, ΔG°₃₇, of 5′GGC/3′CGG and 5′CGG/3′GGC are −1.89 and −1.40 kcal/mol, respectively, calculated from a model duplex of each loop and a duplex lacking the loop.^42,43 This ∼0.5 kcal/mol difference, or 3.6-fold in an equilibrium constant, can likely be traced to more favorable stacking interactions of the 1 × 1 GG internal loop on its closing base pairs. Indeed, our molecular dynamics (MD) simulations⁴⁴ confirmed more optimal stacking of loop nucleotides on their closing pairs in 5′GGC/3′CGG than in 5′CGG/3′GGC. In 5′CGG/3′GGC, the loop nucleotides stacked on the 5′ and 3′ closing pairs with 5.5 ± 1.1 Ų (5′CGG/3′GGC′) and 3.5 ± 0.5 Ų (5′CGG′/3′GGC′) of surface area, respectively (Figure 4A and B, black and red, respectively). In contrast, the stacking surface area of the GG loops on the 5′ and 3′ closing pairs within 5′GGC/3′CGG was 4.4 ± 1.0 Ų (5′GGC/3′CGG) and 11.8 ± 1.4 Ų (5′GGC/3′CGG) of surface area, respectively (Figure 4A and B, blue and green, respectively). Therefore, the computational study highlights a ∼2-fold increase of the stacking surface within 5′GGC/3′CGG compared to 5′CGG/3′GGC′. Consequently, this might affect the conformational ensemble of the 5′GGC/3′CGG motif and the ability to accommodate small molecule CB096.

Figure 4.

The 5′CGG/3′GGC loop is thermodynamically less stable than 5′GGC/3′CGG. (a) Overlap areas between 5′CGG/3′GGC and 5′GGC/3′CGG (G represents the G residue within the 1 × 1 GG internal loop) GG internal loops and the closing GC base pairs as a function of time, obtained by molecular dynamics (MD) simulations. Overlap areas are represented by underlining of two residues, which highlights the G residue part of the 1 × 1 GG internal loops and the G or C residue of the closing base pair. (b) Stacking observed for 1 × 1 GG internal loops on 5′ and 3′ closing base pairs in 5′CGG/3′GGC and 5′GGC/3′CGG, extracted from MD simulations. Figures display averaged structures calculated from the MD trajectories. Blue colored base pairs represent the 1 × 1 GG internal loops in syn-anti orientations. Red and orange colored base pairs highlight the closing Watson–Crick GC base pairs.

Structure–Activity Relationships Reveal Functional Groups Critical for RNA Binding

A SAR study was performed to identify the structural features that are responsible for CB096 binding to the r(G₄C₂)₈ hairpin (Figure 5A and B). First, the importance of the nitro (−NO₂) group at position 5 in CB096 was investigated (Figure 5A). Ten CB096 derivatives were synthesized according to literature procedures, maintaining the 2-methoxyphenyl unit of the parent compound, and binding to Cy5-r(G₄C₂)₈ was assessed via MST (Supplementary Figure 9). The removal of the −NO₂ group (derivative 1), or placement of the −NO₂ group at position 4 (derivative 2), completely abolished binding to Cy5-r(G₄C₂)₈ hairpin, suggesting this functional group’s placement at position 5, i.e., a directional noncovalent interaction, is key to the small molecule–RNA interaction (Figure 5A). Moreover, replacing the −NO₂ group with electron-donating functional groups such as methyl (−CH₃) and methoxy (−OCH₃; derivatives 3 and 4, respectively) and electron-withdrawing substitutions such as fluoro (−F), chloro (−Cl), methanesulfonyl (−SO₂CH₃), trifluoromethyl (−CF₃), or carboxyl (−COOH; derivatives 5–9, respectively) led to either no change in thermophoresis signal or no saturable binding to Cy5-r(G₄C₂)₈ (Figure 5A and Supplementary Figure 9). Replacement of −NO₂ with its isostere −CN (nitrile) as in derivative 10, yielded a ∼2-fold lower binding affinity compared to CB096. These results suggest that neither the electronic properties of the functional group nor the steric size at position 5 dramatically contribute to r(G₄C₂)₈ hairpin binding (further investigated by MD calculations, vide infra). In conclusion, the −NO₂ group at position 5 of the benzimidazole core is both unique and essential for CB096’s interaction with r(G₄C₂)₈ (Figure 5B).

Figure 5.

Mode of binding of CB096 to 5′CGG/3′GGC motif within r(G₄C₂)^exp hairpin revealed via SAR and MD simulations. (a) The SAR study included 17 small molecules (1–17; shown below each structure), closely related derivatives of CB096. N.D. indicates not determined. (b) Summary of the SAR study performed with 17 structurally similar derivatives of CB096. (c) Structure of CB096 in complex with the 5′CGG/3′GGC motif within the r(G₄C₂)-model duplex, where G represents the G residue within the 1 × 1 GG internal loop, as determined by MD simulations. Atoms of O, N, C, and H are represented in red, blue, green, and white, respectively. (d) Structure of CB096 in complex with the 5′CGG/3′GGC motif within the r(G₄C₂)–model duplex, highlighting the noncovalent interactions that are responsible for molecular recognition. Hydrogen (H) bonds are represented with dark blue dashed lines, π–π stacking interactions with light blue, C–H−π interactions with orange, π-hole interactions with purple, and van der Waals interactions with green dashed lines. (e) Schematic diagram of the CB096–5′CGG/3′GGC complex summarizing the noncovalent interactions that stabilize the small molecule-RNA complex with colors as indicated in panel d.

Next, derivatives that varied the 2-methoxyphenyl moiety (derivatives 11–17, Figure 5A and Supplementary Figure 10) were synthesized, while maintaining the −NO₂ functionality at position 5. Binding studies of the seven derivatives showed that (i) methylation of the −OH group led to a ∼1.5-fold lower binding affinity compared to the parent compound (derivative 11); (ii) removal of both −OH and −OCH₃ completely ablated binding to r(G₄C₂)₈ (derivative 12); (iii) removal of −OCH₃ reduced binding affinity by ∼2-fold (derivative 13); (iv) addition of −OCH₃ at position 6 reduced affinity by ∼1.5 fold (derivative 14); (v) methylation of derivative 14 at the phenol moiety ablated binding (derivative 17); and (vi) sterically constraining the 2-methoxyphenol moiety, as in 1,3-dioxolane (derivative 15) and 1,4-dioxane (derivative 16), also ablated binding (Figure 5B).

MD Simulations to Further Interrogate CB096’s Binding Mode

In addition, we investigated the molecular recognition of r(G₄C₂)^exp by CB096 using MD simulations. Given the rapid exchange of tautomeric forms within benzimidazoles,^45–46 computational studies considered both tautomers of CB096 (CB096 and CB096_T in Supplementary Figure 11 and Supplementary Tables 1 and 2). The tautomers are structurally different because of the −NO₂ substitution at position 5. To determine the most preferred bound state of CB096 to 5′CGG/3′GGC, we employed an approach previously utilized to study small molecule-RNA binding.^21,47–49 MD simulations combined with Molecular Mechanics Poisson–Boltzmann Surface Area (MM-PBSA)⁵⁰ calculations revealed that the 5′CGG/3′GGC RNA motif prefers one of the tautomeric forms (CB096) over the other (CB096_T) by ∼5 kcal/mol in binding energy (38.31 vs −32.93 kcal/mol; Supplementary Tables 3 and 4).

We therefore focused our attention on CB096’s (not CB096_T’s) interaction with the 5′CGG/3′GGC motif (Figure 5C–E). The computational calculations predicted that CB096 directly interacts with one of the loop’s G residues while the other loop nucleotide is flipped out of the helix (Figure 5C and D). Various noncovalent interactions stabilize the complex including: (i) the −OH group within the 2-methoxyphenol moiety forms a hydrogen bond (H-bond) with the phosphate group of the flipped-out G4 residue (highlighted in blue in Figure 5D and E); (ii) the methoxy group forms a van der Waals interaction with G11 base in the major groove; (iii) the imidazole’s NH group in CB096 forms a H bond with the phosphate group of G5 while the imidazole nitrogen acts as a H bond acceptor for G11’s amino (−NH₂) group; (iv) CB096’s imidazole and G5 form a π–π stacking interaction; (v) the oxygen atom within the nitro group at position 5 H-bonds with the −NH₂ group of the G5. Interestingly, the −NO₂ group could also serve as an excellent π-hole acceptor^51–53 for the electron rich hydroxyl (−OH) group at the C2′ position of G11, which is favorably located on top of the nitrogen (N) atom within the −NO₂ group. Indeed, the N atom of nitro-aromatics has recently been validated as an excellent π-hole acceptor;^53,54 and (vi) C–H−π interactions occur between the nitrobenzene moiety and the H 1s of G5 and G12. Collectively, these MD simulations support our SAR studies, both of which highlighted the key importance of the −NO₂ group at position 5 and the 2-methoxyphenol moiety.

CB096 Inhibits RAN Translation in r(G₄C₂)₆₆-Expressing HEK293T Cells, a Model of c9ALS/FTD

The inhibition of RAN translation by CB096 was assessed in HEK293T cells cotransfected with plasmids that encode (G₄C₂)₆₆-No ATG-GFP and ATG-mCherry (the latter used for normalization).²¹ We first defined the therapeutic window for CB096 by using a cell viability assay, which showed that CB096 is not cytotoxic up to 25 μM upon 24 h of treatment (Supplementary Figure 12A). Further, CB096 did not alter the proliferation of r(G₄C₂)₆₆-transfected HEK293T cells (Supplementary Figure 12B). Treatment with CB096 for 24 h elicited dose-dependent inhibition of RAN translation with an IC₅₀ of ∼20 μM, as assessed by measuring GFP and mCherry fluorescence (Figure 6A). The reduced GFP signal caused by RAN translation of the repeat expansion was not due to transcriptional inhibition as GFP mRNA levels were not altered upon CB096 treatment, as measured by RT-qPCR (Figure 6B). In agreement with these observations, CB096 had no effect on the canonical translation of GFP, assessed by transfecting HEK293T cells with plasmid encoding GFP with an ATG start codon (Figure 6A). These results were further confirmed by measuring levels of a dipeptide repeat protein generated by RAN translation of r(G₄C₂)^exp, poly(GP) (Figure 6C). Notably, CB096 did not inhibit RAN translation of r(G₂C₄)₆₆, the C9orf72 antisense strand (Figure 6C). As expected, compounds 1, 2, 15, and 16 were unable to inhibit RAN translation of r(G₄C₂)₆₆ as they did not bind to the Cy5-r(G₄C₂)₈ (Supplementary Figure 13). Collectively, these in cellulis results indicate that CB096 binds r(G₄C₂)^exp and thereby inhibited RAN translation.

Figure 6.

Biological activity of CB096 in a HEK293T cellular model of c9ALS/FTD. (a) CB096 selectively inhibits repeat-associated non-ATG (RAN) translation of r(G₄C₂)^exp. Data indicated with blue bars were collected from HEK293T cells cotransfected with plasmids encoding r(G₄C₂)₆₆-No ATG-GFP and ATG-mCherry, while data indicated with purple bars were collected from HEK293T cells cotransfected with plasmids encoding ATG-GFP and ATG-mCherry. Data are reported as the mean ± SD from three independent measurements each with three technical replicates. (b) Effect of CB096 on r(G₄C₂)₆₆ mRNA levels, as measured by RT-qPCR. Data are reported as the mean ± SD from three independent measurements with three technical replicates each. “n.s.” indicates not significant. (c) Effect of CB096 on the levels of toxic dipeptide repeats of poly(GP) upon 24 h of treatment. Poly(GP) levels were measured by an electroluminescent sandwich immunoassay. Data are reported as the mean ± SD and are representative of two independent measurements with two technical replicates each. (d) Rescue of nucleocytoplasmic transport dysfunction by CB096. Representative images of HEK293T cells stably expressing Lentiviral-S-tdTomato cotransfected with plasmid encoding a (G₄C₂)₆₆-No ATG-firefly luciferase (RAN translation), SV40-Renilla (canonical translation), and ATG-GFP (for distinguishing transfected from non-transfected cells in imaging studies). Cells were treated with vehicle, KPT-350 (KPT at 10 μM; a selective inhibitor of Exportin-1-mediated nuclear export), or CB096 (25 μM). (e) Quantification of the percentage of tdTomato localized to the nucleus or cytoplasm (n = 3 biological samples; 200 cells evaluated per biological sample). (f) Effect of CB096 on r(G₄C₂)₆₆-induced or chemically induced stress granules. For the former, HEK293T cells stably expressing Lentiviral-S-td-Tomato were transfected with (G₄C₂)₆₆-No ATG-GFP and treated with CB096 or vehicle (25 μM). For chemically induced stress granules, HEK293T cells stably expressing Lentiviral-S-td-Tomato were transfected with ATG-GFP and treated with 0.5 mM sodium arsenite (NaAsO₂). Stress granules were visualized by Ataxin-2 (ATXN2) staining and imaged by confocal microscopy. (g) Quantification of average number of stress granules per nuclei upon treatment with vehicle or CB096 (n = 3 biological samples; 200 cells counted per biological sample). *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001, as determined by one-way ANOVA. Data are reported as mean ± SD.

CB096 Improves Nucleocytoplasmic Transport and Reduces Stress Granule Formation in r(G₄C₂)₆₆-Expressing HEK293T Cells

Accumulation of toxic DPRs can cause impaired nucleocytoplasmic transport.⁵⁵ To study whether CB096 alleviates nucleocytoplasmic transport dysfunction, we used a cellular model in which HEK293T cells stably express NLS-NES-tagged tdTomato [where NLS denotes nuclear localization signal (N-terminus) and NES denotes a nuclear export sequence (C-terminus)] or Lentiviral-S-tdTomato.⁵⁶ These cells were cotransfected with plasmids encoding (G₄C₂)₆₆-No ATG-firefly luciferase, which solely undergoes RAN translation, Renilla firefly, used to measure canonical translation, and a plasmid encoding ATG-GFP (used to distinguish transfected cells for imaging studies).⁵⁶ As expected, expression of r(G₄C₂)₆₆ disrupted the nuclear localization of tdTomato (Figure 6D), which was rescued upon treatment with CB096 (Figure 6D,E). In particular, 25 μM of CB096 rescued ∼75% of tdTomato mislocalization. Rescue can be traced to inhibition of RAN translation, as a dose-dependent decrease in firefly luciferase was also observed, with no effect on Renilla luciferase levels (canonical translation; Supplementary Figure 14).

Studies by other groups have shown a link between defects in nucleocytoplasmic transport, formation of nuclear foci, and the presence of stress granules in ALS/FTD,⁵⁷ toxic depositions of membrane-less condensates containing RNA/protein assemblies caused by cellular stress. Indeed, stress granules are observed in an adeno-associated virus (AAV) mouse of model c9ALS/FTD,⁵⁸ and they are distinct from those caused by heat or chemical stress.^59,60 We thus sought to determine if CB096 could decrease the abundance of r(G₄C₂)₆₆-induced stress granules using the HEK293T stably expressing Lentiviral-S-tdTomato and forced expression of r(G₄C₂)₆₆-No ATG-GFP. Upon treatment with 25 μM of CB096, the number of stress granules was reduced by 81 (±11)%, as assessed by staining of the stress granule marker Ataxin-2 (ATXN2; Figure 6F,G). To verify that this effect is specific to r(G₄C₂)^exp, we chemically induced stress granule formation with sodium arsenite (NaAsO₂), a commonly used chemical stressor that causes both cytoplasmic mislocalization and stress granule formation.⁶¹ Notably, CB096 (25 μM) was unable to reduce the number of stress granules induced by NaAsO₂, indicating that the compound’s effect is r(G₄C₂)^exp-dependent (Figure 6F,G).

Summary and Conclusions

Collectively, the ligandability of the r(G₄C₂)^exp hairpin responsible for RAN translation in c9ALS/FTD was assessed by screening an RNA-focused library containing nitrogen-rich heterocycles featuring drug-like molecules. High throughput screening using a TO1 dye displacement assay identified 26 hit compounds that were further validated via MST. This two-step procedure yielded the benzimidazole derivative CB096 as a selective binder to the r(G₄C₂)₈ hairpin over r(G₂C₄)₈ (antisense), r(GGCC)₁₀ (base pair) RNAs, and other 1 × 1 internal loops. NMR structural studies showed that CB096 selectively binds the 5′CGG/3′GGC loops (and not the 5′GGC/3′CGG loops) within r(G₄C₂)^exp hairpin and disrupts its closing base pairs. Computational investigations revealed that 5′CGG/3′GGC has less optimal stacking interactions than 5′GGC/3′CGG and is thus less stable. These data indicate that 5′CGG/3′GGC may be more prone to structural rearrangement required for small molecule binding.

Additionally, SAR studies showed that a rather unique −NO₂ group at position 5 and the 2-methoxyphenyl moiety are crucial for binding the r(G₄C₂)^exp hairpin, which were confirmed by MD simulations. Interestingly, the contribution of the −NO₂ group does not appear to be electronic in nature, as various electron-donating and -withdrawing functional groups of various sizes did not promote binding to the r(G₄C₂)₈ hairpin. A known isostere of the −NO₂ group, −CN, yielded a less avid binder compared to CB096, further highlighting the uniqueness of the −NO₂. Rather, the −NO₂ group acts as a H-bond acceptor and has the potential to serve as a π-hole acceptor. The 2-methoxyphenyl moiety also plays a key role in molecular recognition, where the −OH group establishes a H bond with the phosphate group of the flipped-out G4 residue and the −OCH₃ group forms van der Waals interactions with the nearby G11 residue.

Notably, CB096 selectively inhibited RAN translation and decreased the production of poly(GP) DPRs without affecting r(G₄C₂)₆₆ mRNA levels in an HEK293T cellular model of c9ALS/FTD, alleviated dysfunctional nucleocytoplasmic transport, and selectively reduced the abundance of r(G₄C₂)₆₆-induced stress granules. In conclusion, the r(G₄C₂)^exp hairpin adopts multiple structural features that can accommodate or can be stabilized by small molecules, further substantiating the potential of r(G₄C₂)^exp ligandability for therapeutic benefit.

Supplementary Material

Supplementary Material

Figure S1. Hit validation of from r(G₄C₂)₈ binders via TO1 displacement in vitro. EC₅₀ values were determined by plotting the change in normalized fluorescence as a function of compound concentration. Data points are the mean ± SD of two independent experiments in which each concentration was measured in triplicate. The binding curve was fitted using a four-parameter nonlinear regression equation of the GraphPad Prism 8.1.2 suite.

Figure S2. Representative thermophoresis curves of CB096 binding to Cy5-labeled RNAs obtained. Dose dependent addition of CB096: i) increases the thermophoresis of the hairpin form of Cy5-r(G₄C₂)₈ (a); ii) leads to minor change of the Cy5-r(G₂C₄)₈ (antisense) thermophoresis (b); iii) triggers no saturable change of the Cy5-r(GGCC)₁₀ (base pair control) thermophoresis (c). The normalized fluorescence upon dose-dependent addition of CB096 was recorded in the Thermophoresis and T-jump channels. Data points are representative of two independent experiments each performed in duplicate.

Figure S3. CB096 specifically interacts with the 1×1 GG internal loops within the r(G₄C₂)^exp hairpin RNA. a) No saturable binding was observed in thermophoresis measurement for CB096 and RNAs featuring 1×1 AA and 1×1 UU internal loops within Cy5-r(CAG)₁₂ and Cy5-r(CUG)₁₂ RNAs, respectively, as assessed via MST. b-c) Representative thermophoresis curves of CB096 interacting with Cy5-r(CAG)₁₂ (b) and Cy5-r(CUG)₁₂ (c) RNAs. Dose dependent addition of CB096: i) triggers no saturable change of the Cy5-r(CAG)₁₂ thermophoresis (b); and ii) leads to minor change of the Cy5-r(CUG)₁₂ thermophoresis (c). The normalized fluorescence at each CB096 concentration was recorded in the Thermophoresis and T-jump channels. Data points in panel a were generated from two independent experiments each performed in duplicate.

Figure S4. Representative CD spectrum recorded upon dose dependent addition of CB096 to r(G₄C₂)₈ hairpin. Compound addition triggers an increase of CD signals at 235 and 298 nm (a) that reached saturation at 4 equivalents CB096, i.e. a binding stoichiometry of CB096 to r(G₄C₂)₈ of 4:1 (b). The spectra are representative of two independent experiments.

Figure S5. CB096 mode of binding to the r(G₄C₂)₄ hairpin. CB096 broadens the imino proton peak of GC/CG base pairs at 12.5 ppm. Blue labels indicate 5’GGC/3’CGG (G represents the G residues within the 1×1 GG internal loops) internal loops, whereas the orange labels indicate 5’CGG/3’GGC loops. Further addition of CB096 triggered precipitation. The 1D NMR spectra are representative of two independent experiments.

Figure S6. Dose dependent addition of CB096 does not change the imino proton signals of the 5’GG/3’CC (blue squares) and 5’CC/3’GG (orange squares) base pairs at 13.1 ppm and 12.8 ppm within r(G₂C₄)₈ (antisense). The 1D NMR spectra are representative of two independent experiments.

Figure S7. CB096 does not interact with a 2×2 GG internal loop-containing duplex. a) Free energy minimization (RNAstructure software) of the r(G₄C₂)₂ duplex predicts secondary structures with 1×1 GG and 2×2 GG loops have the same free energy of folding at 37 ^oC (ΔG^o₃₇). b) Dose dependent addition of CB096 does not affect the imino proton signal of GC/CG base pairs at 12.5 ppm within the 2×2 GG internal loop containing model construct. The 1D NMR spectra are representative of two independent experiments.

Figure S8. Binding of CB096 to 5’GGC/3’CGG sites (highlighted in blue) within Cy5-r(GGC)₁₆ via MST. No change in thermophoresis, i.e. no binding, was observed upon dose dependent addition of CB096 to Cy5-r(GGC)₁₆. The secondary structure was predicted by RNAstructure software. The normalized fluorescence was recorded in the Thermophoresis and T jump channels. Data points indicate mean values ± SD from two independent experiments each performed in duplicate.

Figure S9. In vitro binding of compounds 1–10 to Cy5-r(G₄C₂)₈ hairpin, as measured by MST. No change in thermophoresis or no saturation, i.e. no binding, was observed upon dose dependent addition of derivatives 1–10 to Cy5-r(G₄C₂)₈ as shown by the normalized fluorescence signal recorded in the Thermophoresis and T jump channels. Saturable binding was observed for derivative 10 only, with EC_50,total value of 32(±6) μM, calculated by plotting the changes in normalized fluorescence as function of compound concentration. EC_50,total considered the EC₅₀ values for the two observed binding modes and was generated by calculating the square root of the first binding mode (EC_50,1) multiplied by the value of the second binding mode (EC_50,2). Data points are reported as the mean ± SD from two independent experiments each performed in duplicate.

Figure S10. In vitro binding of compounds 11–17 to Cy5-r(G₄C₂)₈ hairpin, as measured by MST. Saturable binding was observed for derivatives 11, 13, and 14 only, with EC₅₀ values of 30(±4) μM, 39(±5) μM, and 30(±3) μM, respectively. EC₅₀ values were determined by plotting the changes in normalized fluorescence as function of compound concentration. No change in thermophoresis or no saturation, i.e. no binding, was observed addition of derivatives 12,15–17 to Cy5-r(G₄C₂)₈. Data points are reported as mean ± SD from two independent experiments each performed in duplicate. Binding curves were fitted using a four-parameter nonlinear regression equation of the GraphPad Prism 8.1.2 suite.

Figure S11. Atom numbering of CB096 (a) and its tautomer CB096_T (b).

Figure S12. CB096 is not cytotoxic in r(G₄C₂)₆₆-transfected HEK293T cells nor affects cell proliferation. a) Cellular viability of r(G₄C₂)₆₆-transfected HEK293T cells upon incubation with CB096 for 24 h assessed via CellTiter-Glo. Data points are reported as the mean ± SD from two independent experiments each performed in triplicate. b) CB096 (25 μM) does not alter the cell proliferation of r(G₄C₂)₆₆-transfected HEK293T cells. Data points are reported as the mean ± SD relative to vehicle (0.5% DMSO) from two independent experiments each performed in triplicate. n.s. = not significant, as determined by a two-tailed Student t test.

Figure S13. Inhibition of RAN translation by CB096 and compounds 1, 2, 15 and 16. CB096 binds r(G₄C₂)₈ hairpin in vitro and inhibits RAN translation in cells in a dose dependent manner. Conversely, derivatives 1, 2, 15 and 16 do not bind r(G₄C₂)₈ and do not block RAN translation. Data are reported as the mean ± SD from three independent experiments each performed in duplicate.

Figure S14. CB096 selectively inhibits RAN translation of r(G₄C₂)^exp in HEK293T cells that stably express Lentiviral-S-tdTomato co-transfected with plasmind encoding a (G₄C₂)₆₆-No ATG-firefly luciferase (RAN translation), Renilla luciferase (canonical translation), and ATG-GFP. Relative translation was measured by normalizing the signal from firefly luciferase to Renilla luciferase (non-canonical translation) or the signal from Renilla luciferase to GFP (canonical translation). Data are reported as the mean ± SD (n = 4).