Massively Parallel Optimization of the Linker Domain in Small Molecule Dimers Targeting a Toxic r(CUG) Repeat Expansion

Abstract

RNA contributes to disease pathobiology and is an important therapeutic target. The downstream biology of disease-causing RNAs can be short-circuited with small molecules that recognize structured regions. The discovery and optimization of small molecules interacting with RNA is, however, challenging. Herein, we demonstrate a massively parallel one-bead-one-compound methodology, employed to optimize the linker region of a dimeric compound that binds the toxic r(CUG) repeat expansion [r(CUG)^exp] causative of myotonic dystrophy type 1 (DM1). Indeed, affinity selection on a 331,776-member library allowed the discovery of a compound with enhanced potency both in vitro (10-fold) and in DM1-patient-derived myotubes (5-fold). Molecular dynamics simulations revealed additional interactions between the optimized linker and the RNA, resulting in ca. 10 kcal/mol lower binding free energy. The compound was conjugated to a cleavage module, which directly cleaved the transcript harboring the r(CUG)^exp and alleviated disease-associated defects.

Targeting RNA with synthetic molecules was long thought to be challenging because of RNA’s limited local diversity from only four nucleotides and its flexible and dynamic nature. Many RNAs, however, adopt stable three-dimensional (3D) structures, which are potential binding sites for small molecules.¹⁻⁶ The discovery of selective small molecules targeting disease-causing RNA has historically been accomplished through selection methods⁷ or in various targeted screens for specific RNA structures.^8,9 Suitable RNA targets for small molecules were identified in a myriad of diseases, such as cancers caused by oncogenic microRNA (miR)-21^10,11 or spinal muscular atrophy.¹² To identify and optimize novel chemical matter targeting toxic RNA structures, robust combinatorial methods are required. Advanced combinatorial chemistry methods, such as DNA-encoded libraries,¹³⁻¹⁶ one-bead-one-compound (OBOC) libraries,^17,18 phage display,^19,20 and affinity selection mass spectrometry,^21,22 have been successfully applied to proteins and can be adapted for RNA. For example, Patel and co-workers reported a one-bead-two-compounds (OBTC) strategy to discover macrocycles targeting the long noncoding RNA growth arrest specific 5 (GAS5), with implications for the treatment of type 2 diabetes mellitus.²³ The application of combinatorial methods to disease-causing RNAs enables the discovery of novel chemical probes and optimization of ligands targeting these RNA structures in a massively parallel fashion.

One class of toxic RNAs is repeat expansions that cause greater than 40 genetic diseases including myotonic dystrophy type 1 (DM1). DM1 is caused by a r(CUG) repeat expansion [r(CUG)^exp] located in the 3′ untranslated region (UTR) of the dystrophia myotonica protein kinase (DMPK) mRNA.²⁴ DM1 is an incurable neuromuscular disease with a prevalence of ∼1 in 8000 people. Those afflicted with DM1 have symptoms that include muscle weakness and myotonia, heart abnormalities, cataracts, and insulin resistance.^25,26 The r(CUG)^exp folds into a highly structured hairpin, the stem of which comprises a periodic array of 5′CUG/GUC 1 × 1 internal loops. These loops bind and sequester the pre-mRNA splicing regulator muscleblind-like 1 (MBNL1) protein as well as other RNA-binding proteins in nuclear foci (Figure 1A). Sequestration of MBNL1 by r(CUG)^exp prevents its interaction with its natural substrates, therefore leading to dysregulation of alternative pre-mRNA splicing and manifestation of disease.^27,28 Different types of modalities can bind to r(CUG)^exp and improve disease defects in DM1-affected cells,²⁹⁻³¹ including monomeric small molecules,³¹⁻³³ modularly assembled dimeric compounds that occupy multiple internal loops simultaneously,^30,34 pseudopeptides,³⁵ and small molecules that cleave r(CUG)^exp.³⁶ These compounds were discovered by various approaches, such as a bead-based screening method called resin-bound dynamic combinatorial chemistry³⁵ and by fragment-based target-guided synthesis.³⁴

Figure 1.

Small molecule targeting approach for the toxic RNA repeat expansion, r(CUG)^exp, that causes myotonic dystrophy type 1 (DM1). (A) Toxic r(CUG)^exp present in the 3′ UTR of the DMPK mRNA sequesters MBNL1 protein, a regulator of alternative pre-mRNA splicing, with high affinity, leading to splicing defects. (B) Modularly assembled RNA-binding modules bind with high affinity to r(CUG)^exp, liberating MBNL1 and rescuing DM1-associated splicing defects. (C) Chemical structure of the 5′CUG/3′GUC [loops present in r(CUG)^exp] RNA-binding module used in this study. (D) Chemical structure of a previously developed dimeric compound that rescues DM1-associated defects, 2H-4. 2H-4 was used as a starting point to develop the OBOC methodology described herein.

Our group has previously reported the modular assembly of RNA-binding modules targeting consecutive 5′CUG/GUC 1 × 1 internal loops to disrupt the toxic MBNL1–r(CUG)^exp complex and improve DM1-associated defects (Figure 1B,C).³⁷ A dimeric compound named 2H-4 was built by a N-propylglycine peptoid bridge to separate the binding modules at a specific distance to enable simultaneous binding (Figure 1D). Efforts to improve the binding affinity and potency of 2H-4 have involved changing the nature of its backbone including an N-methyl alanine linker,³⁸ a proline linker,³⁹ and macrocyclization.⁴⁰ These modifications have improved binding affinity and cellular activity; however, thus far, the optimization of the linker between RNA-binding modules has not been attempted on a large scale.

Peptoids, oligomers of N-substituted glycine building blocks, offer access to a large chemical diversity by simple incorporation of primary amine building blocks. Further, peptoids have desirable pharmacological properties, such as cellular permeability and resistance to proteolytic degradation.^41,42 As their syntheses are fully compatible with OBOC library approaches,⁴³ we envisioned a simple method to introduce a wide variety of functional groups in the linker domain of 2H-4, which is described herein.

Briefly, the OBOC library was synthesized on 90 μm Tentagel microbeads with four variable positions between the two 5′CUG/GUC internal loop binding modules, as this is the optimal linker length identified from previous studies (Figure 2A).³⁸ A peptide tag, BBRGYM, was incorporated at the C-terminal of each compound, allowing release from the beads by selective cyanogen bromide (CNBr) cleavage of the M residue for facile deconvolution of compound identity by tandem mass spectrometry (MS–MS). Indeed, fragmentation of the amide bonds generates a mass ladder that represents the nature of the different building blocks and their sequential order in the compound. The peptide tag provides a known starting point in the MS–MS spectrum from which unknown building blocks are identified. A set of 24 different building blocks were incorporated by split-and-pool methodology (Figure 2B), generating a theoretical diversity of 331,776 compounds. Building blocks were selected to cover various chemotypes, such as aliphatic, cationic, aromatic, or heteroaromatic moieties. As a quality control step, random beads from the library were selected and processed for compound deconvolution, and indeed MS–MS analysis enabled structure deconvolution unambiguously (Figure S1).

Figure 2.

OBOC library and screening method to optimize dimeric compounds targeting r(CUG)^exp. (A) Chemical structure of the library, with R_1–4 indicating variable residues. (B) Building blocks incorporated to generate the chemical diversity of the 331,776-member library. (C) Screening workflow using compounds supported on 90 μm Tentagel beads. (1) Incubation with biotinylated r(CUG)₁₂; (2) magnetic pull-down using streptavidin-coated Dynabeads; (3) bead isolation and cleavage; (4) hit structure deconvolution by tandem mass spectrometry. (D) Hit triage of the 142 hits by enrichment score and nuclear localization in DM1 myotubes.

To identify compounds that avidly bind r(CUG)^exp, the OBOC library was incubated with 150 nM of an established structural model of r(CUG)^exp, r(CUG)₁₂ containing a 5′ biotin tag,⁴⁴ in the presence of 15 μM of 2H-4 (100-fold molar excess as compared to the RNA’s concentration). The 2H-4 competitor was added to increase the stringency of the screen and to allow identification of compounds that bind more avidly to r(CUG)₁₂ than the competitor. Beads bound to 5′-biotin-r(CUG)₁₂ were isolated with streptavidin-coated Dynabeads (Figure 2C) and identified by MS–MS analysis, affording 142 hit compounds.

To triage the hits to a more manageable number for further investigation, we calculated the enrichment and the statistical confidence thereof (Z-score) of each building block at each variable position (Figure 2D). (A Z-score of 1.96 corresponds to P = 0.05 and hence statistical significance.) The Z-score at each position was also summed for each hit compound (Figure S2). Several trends could be observed in selection among the building blocks, including (i) a depletion of tertiary amines, particularly building blocks #19–22, at positions 1–3 (P = 0.013 to 0.039) and (ii) enrichment of secondary and primary amines, building blocks #13 and #15, at all positions (P < 0.0001) (Figure S3).

We observed that the sum of Z-scores for the building blocks comprising hit compounds correlated with positive charge (Figure S4). As the RNA backbone is negatively charged, this is not surprising. To further investigate this observation as well as to maintain chemical diversity, we binned hit compounds by their linker charge and then selected those in each bin with Z-scores in the top 30%. This first triage provided a reasonable number of compounds (n = 32) to synthesize by solid-phase parallel synthesis on Rink amide polystyrene resin, to measure avidity for bind r(CUG)^exp, and to study for rescue of DM1-associated defects.

In previous studies of dimeric compounds targeting r(CUG)^exp, we observed that cell uptake and localization can change as a function of the linker’s structure.^39,40 Furthermore, charged backbones are known to affect localization.⁴⁵ Because nuclear localization is crucial for biological activity as r(CUG)^exp is sequestered in nuclear foci, we studied the compounds’ cellular permeability and localization.

DM1 patient-derived myotubes containing 1300 r(CUG) repeats were treated with 5 μM of each compound, and localization was monitored using the inherent fluorescence of the H RNA-binding modules. Interestingly, 2H-4 localizes almost completely in the nucleus, as do 12 of the 32 hit OBOC compounds (Figure S5), which were thus prioritized for further study. A closer inspection of building block distribution provided insights into the relationship between compound structure and nuclear localization (Figure S6). Among the nuclearly localized compounds, only the polar building block #14 (N-hydroxyethyl substituent) was found to be significantly enriched at position 2 (17%, P = 0.03). Within the compounds with poor nuclear localization, building blocks #2 (N-propyl substituent) and #13 (N-methylaminobutyl substituent) were found to be significantly enriched at positions 3 (20%, P = 0.03) and 2 (33%, P = 0.049), respectively. These data suggest that the position where polar and hydrophobic moieties are incorporated within the dimer linker influence localization.

MBNL1 protein self-regulates the alternative splicing of its exon 5, which is included too frequently in DM1 cells.⁴⁶ Thus, we evaluated which of the 12 compounds rescue the MBNL1 exon 5 splicing defect in DM1-patient-derived myotubes and compared their activities to 2H-4 (Figure 3). DM1 cells were treated with a 5 μM dose for 2, 5, 10, 13, 15, 18, 21, and 25 for 48 h. Changes in cellular morphology were observed upon treatment with 5 μM of 7, 14, 22, and 32, and therefore these compounds were evaluated at 1 μM, where no changes in morphology were observed. Of the eight compounds evaluated at 5 μM, only two slightly improved the MBNL1 exon 5 splicing defect, by ∼20%, which is less than the effect observed upon treatment with 2H-4 (∼30% improvement) (Figures 3A and S7). The four compounds evaluated at 1 μM all improved the MBNL1 exon 5 splicing defect more potently than did 2H-4, which is inactive at the 1 μM dose (Figures 3B and S7), although the percent rescue by 32 was not statistically significant. Notably, 7 and 22 improved MBNL1 exon 5 splicing dose-dependently, with 22 appearing to be the more potent of the two (7% rescue for 22 vs 2% rescue for 7 at 0.2 μM) (Figure 3C). As 14 improved splicing similarly at all concentrations tested, it was eliminated from further study (Figures 3C and S8). Since 22 was the most potent, it was studied in more detail, both in vitro and in DM1-patient-derived myotubes.

Figure 3.

Evaluation of hit compound activity in DM1 patient-derived myotubes. (A) Activity of hit compounds with no toxicity at 5 μM, as assessed by rescue of the MBNL1 exon 5 splicing defect in DM1 myotubes (n = 3). (B) Activity of hit compounds with no toxicity at 1 μM, as assessed by rescue of the MBNL1 exon 5 splicing defect in DM1 myotubes (n = 3). (C) Dose–response analysis of compounds with statistically significant activity at 1 μM (n = 6). For all panels: error bars represent standard deviation, *P < 0.05, **P < 0.01, ***P < 0.001, as determined by one-way ANOVA. (D) Chemical structure of compound 22.

We first measured the binding of 22 to r(CUG)₁₂ in vitro, affording an EC₅₀ of 106 ± 4 nM (Figure 4A), ∼10-fold more avid than 2H-4 (EC₅₀ = 1140 ± 31 nM).⁴⁰ Importantly, no saturable binding was observed for 22 and a base-paired control RNA that does not contain the 1 × 1 nucleotide UU internal loops (Figure S9). This avid binding of 22 to r(CUG) repeats translated to inhibition of the formation of the r(CUG)^exp–MBNL1 complex in vitro with an IC₅₀ of 2.8 ± 0.2 μM in a TR-FRET assay (Figure S10),⁴⁴ an ∼10-fold improvement over 2H-4 (IC₅₀ of 32.2 ± 4.3 μM; determined by the same TR-FRET assay),⁴⁰ which correlates with its enhanced affinity.

Figure 4.

Compound 22 binds avidly to r(CUG)₁₂ in vitro and selectively improves disease-associated defects in DM1 myotubes. (A) Representative direct binding curve of 22 and r(CUG)₁₂ hairpin using the inherent fluorescence of 22 (n = 3). (B) Evaluation of DMPK levels in DM1 myotubes treated with 22 via RT-qPCR (n = 3). (C) Representative confocal microscopy images of nuclear foci. MBNL1 was imaged by immunofluorescence, whereas r(CUG)^exp was imaged by RNA fluorescence in situ hybridization (FISH). (D) Quantification of the number of foci/nucleus in DM1 myotubes treated with 22 (n = 3; 40 nuclei counted/replicate). ***P < 0.001 Student’s t-test. (E) Interactions between 22 and two 5′CUG/3′GUC internal loops present in r(CUG)^exp from a model generated by molecular dynamics simulation. Dashed lines represent interactions between the linker residues of 22 and the RNA.

To rationalize the improved in vitro binding of 22 to r(CUG)₁₂ when compared to that of 2H-4, molecular modeling studies were carried out (Tables S2–S5 and Figures S11–S14). Models of dimer 22 bound to a model RNA with two 1 × 1 nucleotide UU internal loops in r(CUG)^exp [r(5′-CCGCUGCUGCGG/3′GGCGUCGUCGCC] were generated using molecular dynamics simulations and compared with a previously published model of 2H-4. The best model, as defined by the lowest free binding energy, generated for 22 was 10 kcal/mol lower than that of 2H-4 (−61.7 vs −51.7 kcal/mol).

Inspection of the interactions of 22 with the internal loops shows an increased number of interactions between the dimer’s linker region and the RNA backbone, in addition to the interactions of RNA-binding modules with the RNA, which are similar to those of 2H-4 (Figure 4E). The imidazole group of the building block in position 1 hydrogen bonds with G21’s NH₂ [the bolded nucleotide in r(5′-CCGCUGCUGCGG/3′GGCGUCGUCGCC)] and stacks with the RNA-binding module. The carbonyl groups in the backbone undergo a number of hydrogen bonding interactions, including intramolecular hydrogen bonds with the NH–CH₃ side chain of the building block in position 3 and NH₂ of the side chain of the building block in position 4. Intermolecular H-bonding interactions of the dimer backbone with the RNA also contribute significantly to the binding of the dimer, including (i) interaction of building block 2’s carbonyl with 2′ OH of G6; (ii) H-bond formed between the terminal amide CONH₂ with 2′ OH of C7; (iii) interaction of building block 3’s carbonyl with G6’s and G8’s NH₂; and (iv) H-bond between the linker’s terminal amide NH₂ with the phosphate backbone of U8. These interactions, in addition to the van der Waals and π-stacking interactions formed throughout the dimer–RNA complex, contribute to the observed 10 kcal/mol improvement in the binding energy when compared to that of the bound structure of 2H-4.

The selectivity and activity of 22 were further investigated in DM1 patient-derived myotubes. Importantly, the observed rescue of MBNL1 exon 5 pre-mRNA splicing defect was not due to transcriptional inhibition of DMPK, as its RNA levels were unaffected (Figure 4B), suggesting direct r(CUG)^exp target engagement. Importantly, 22 specifically improved DM1-associated defects as MAP4K4 exon 22a splicing, a NOVA-regulated splicing event,⁴⁷ was not affected in DM1 myotubes (Figure S15). Furthermore, MBNL1 exon 5 splicing was not affected in wild-type myotubes (from healthy donors) treated with 22 (Figure S16). Compound 22 also reduced the number of r(CUG)^exp–MBNL1 foci in DM1 myotubes by ∼20%, which correlates with the observed improvement in splicing (Figure 4C,D). Thus, through specific binding to r(CUG)^exp, 22 potently and specifically improves DM1-associated defects in patient-derived myotubes.

We have previously demonstrated that conjugation of bleomycin A5 to repeat-targeting compounds allows for specific cleavage of the target RNA as well as increases potency.^36,48 This direct cleavage by bleomycin A5 also enables target identification and analysis of selectivity as cleavage will be observed for all targets that the compound engages. Thus, we appended 22 with bleomycin A5 using its terminal amine as attachment at this site is known to decrease bleomycin’s affinity for DNA and ablate off-target effects (Figure 5A).³⁶ The compound 22-Bleo was first evaluated for its ability to selectively bind and cleave r(CUG)₁₂ in vitro. Indeed 22-Bleo selectively bound r(CUG)₁₂ with an EC₅₀ of 120 ± 7 nM (similar to the EC₅₀ of 22, 106 ± 4 nM; note: Fe²⁺, required for cleavage, is absent in binding assays), with no binding observed to a base-paired RNA (Figures 5B and S17). The RNA degrader also cleaved the r(CUG) repeats in vitro, and >70% of the RNA was cleaved at a 5 μM dose (Figures 5C and S18). Thus, 22-Bleo can selectively recognize r(CUG)^exp and cleave it in vitro.

Figure 5.

Compound 22-Bleo cleaves r(CUG)^exp in vitro and in DM1 myotubes, rescuing disease-associated defects. (A) Conjugation of 22 to bleomycin A5 afforded cleaving compound 22-Bleo. (B) Direct binding assay of 22-Bleo with r(CUG)₁₂ using the inherent fluorescence of 22 (n = 3). Note: assays were completed in the absence of Fe²⁺, required for cleavage. (C) In vitro cleavage activity of 22-Bleo using radioactively labeled r(CUG)₁₀ and analysis of fragments by gel electrophoresis (n = 3); ****P < 0.0001, as determined by one-way ANOVA. (D) Evaluation of DMPK levels upon compound treatment via RT-qPCR (n = 6); **P < 0.01, as determined by one-way ANOVA. (E) Ability of 22-Bleo to improve the MBNL1 exon 5 splicing defect in DM1 myotubes (n = 6); **P < 0.01, ***P < 0.001, as determined by one-way ANOVA. (F) Quantification of the number of foci/nucleus in DM1 myotubes treated with 22-Bleo (n = 3; 40 nuclei counted/replicate); ***P < 0.001, as determined by a Student’s t-test. (G) Competitive cleavage experiment between 22 and 22-Bleo where excess 22 prevents cleavage of DMPK by 22-Bleo. Levels of DMPK mRNA were measured by RT-qPCR (n = 6); ***P < 0.001, as determined by one-way ANOVA. (H) Abundance of r(CUG) repeat-containing transcripts upon treatment with 22-Bleo, as measured by RT-qPCR (n = 6); ***P < 0.001, as determined by a Student’s t-test. For all panels, error bars represent standard deviation.

We next studied whether the in vitro cleavage activity of 22-Bleo translated to cleavage of r(CUG)^exp in DM1 myotubes. Indeed, a ∼35% reduction in the abundance of DMPK was observed after treatment with 1 μM of 22-Bleo (Figure 5D). Importantly, this cleavage was specific to the disease-causing, DMPK-allele-harboring r(CUG)^exp as DMPK levels were not affected in wild-type cells (Figure S19). A competition experiment between 22-Bleo and parent compound 22 was then completed to confirm the former’s mode of action (cleavage rather than transcriptional inhibition) and the latter’s direct engagement of r(CUG)^exp. As expected, co-treatment of DM1 myotubes with varying concentration of 22 and a constant concentration of 22-Bleo afforded a dose-dependent rescue of DMPK levels (Figure 5G).

Specific cleavage of r(CUG)^exp resulted in improvement of DM1-associated defects including ∼30% rescue of the MBNL1 exon 5 splicing defect and ∼30% reduction in the number of r(CUG)^exp–MBNL1 nuclear foci (Figures 5E,F and S20). A small but statistically significant improvement in splicing was observed at 0.2 μM, indicating that 22-Bleo is more potent than 22 (Figure 5E). Similar to the parent compound, 22-Bleo did not affect MBNL1 exon 5 splicing in wild-type myotubes or MAP4K4 exon 22a splicing in DM1 myotubes, indicating specific effects (Figure S21).

Importantly, the cleavage mode of action of 22-Bleo allows for direct profiling of potential off-targets. We assessed the levels of all transcripts containing short, nonpathological r(CUG) repeats upon treatment of DM1 myotubes with 22-Bleo. None of these genes was significantly affected, indicating that 22-Bleo can specifically recognize and cleave the disease-causing repeat expansion (Figure 5H). We have previously shown that these RNAs containing shorter r(CUG) repeats do not fold into a structure containing repeating 1 × 1 U/U internal loops, the source of the observed selectivity.³⁶

In conclusion, the OBOC library methodology provides a facile means to optimize the linker domain of a dimeric compound targeting r(CUG)^exp. A simple affinity-based selection strategy enabled the screening of >330,000 compounds and subsequent hit identification via MS–MS sequencing. Through subsequent analysis of the bioactivity of hit compounds, we identified compound 22 which bound r(CUG) repeats 10 times more avidly than 2H-4 by forming additional interactions between the target and optimized linker and more potently rescued disease-associated defects in DM1 patient-derived myotubes. A derivative of 22 attached to the natural product bleomycin A5 selectively cleaved r(CUG)^exp in cells and improved DM1 defects at concentrations lower than that of the parent binder. Thus, OBOC library synthesis and screening can be used to identify high-affinity binders to r(CUG)^exp. Importantly, the methodology developed herein is likely to be general and applicable to numerous other RNA targets to aid in the identification of high-affinity small molecules.

Glossary

Abbreviations

CNBr

cyanogen bromide

DM1

myotonic dystrophy type 1

DMPK

dystrophia myotonica protein kinase

EC₅₀

concentration of compound that affords half-maximal response

MAP4K4

mitogen-activated protein kinase kinase kinase kinase 4

MBNL1

muscleblind-like 1

miR

microRNA

MS

mass spectrometry

OBOC

one-bead-one compound

UTR

untranslated region

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.1c00027.

• Supplemental tables and figures and methods (PDF)

Author Present Address

^§ Expansion Therapeutics, 555 Heritage Dr., Suite 150, Jupiter, FL 33458.

Author Contributions

M.D.D. directed the study. S.V.-D., A.J.A., S.C., and K.W.Y. carried out the experiments. I.Y. oversaw the computational experiments. S.V.-D. and A.J.A. contributed equally.

This work was funded by this work including the National Institutes of Health (R35-NS116846 to M.D.D. and F31-NS110269 to A.J.A.), the Department of Defense Peer-Reviewed Medical Research Program (W81XWH-18-0718 to M.D.D.), Myotonic U.S. Fellowship Research grant (to S.C.), and the Fonds de Recherche du Québec, Nature et Technologies (B3X scholarship to S.V.D.).

The authors declare the following competing financial interest(s): M.D.D. is a founder of Expansion Therapeutics. A.J.A. is a current employee of Expansion Therapeutics.