Targeting Expanded CUG and CTG Repeats as a Therapeutic Approach for Myotonic Dystrophy Type 1 (DM1)

Abstract

Myotonic dystrophy type 1 (DM1) is caused by expanded CTG repeats, d(CTG)^exp, transcribed into toxic r(CUG)^exp RNA repeats that sequester splicing regulator MBNL1, leading to its loss‐of‐function. An emerging therapeutic strategy toward DM1 treatment relies on the inhibition of MBNL1 sequestration by using small molecules, oligomers, peptides, engineered proteins, or synthetic oligonucleotides that interact with CUG repeats at the RNA level and/or CTG repeats at the DNA level. This review covers ∼18 years of research in the field of CUG and CTG ligands that were identified or rationally designed as DM1 drug candidates, with an emphasis on their chemical structures, molecular design, RNA‐ or DNA‐binding modes, in vitro affinities and specificities, molecular mechanisms of action, and biological activity in DM1 models.

DM1 is an RNA gain‐of‐function disease caused by CTG repeat expansion, producing toxic r(CUG)^exp RNA that sequesters MBNL1 and impairs splicing. This review covers the field of CUG and CTG ligands identified or rationally designed as DM1 drug candidates, highlighting their molecular design, RNA‐ or DNA‐binding modes, in vitro affinities and specificities, molecular mechanisms, and activity in DM1 models.

1. Molecular Pathogenesis of DM1

1.1. General Aspects and Molecular Basis of DM1 Pathology

Myotonic dystrophy type 1 (DM1), or Steinert's disease, is a multisystemic neuromuscular disorder affecting about 1 in 8000 people worldwide [1, 2]. In 1992, it was found that DM1 is caused by the expansion of CTG trinucleotide repeats in the 3′ untranslated region (3′‐UTR) of the DMPK (dystrophia myotonica protein kinase) gene on chromosome 19q13.3 [3]. In unaffected individuals, the number of CTG repeats in DMPK ranges from 5 to 37. Alleles with 38–50 repeats are premutations, while those with 51–100 are considered protomutations. Both are unstable, meaning the number of CTG repeats can vary in somatic tissues or across generations, and are prone to further expansion. Pathogenic expansions exceed 50 and can extend to several thousand CTG repeats, with the repeat size tending to increase through several generations, especially via maternal transmission [4, 5]. Clinically, DM1 manifests in four major forms which depend on the CTG repeat length, with larger expansions associated with earlier onset and more severe symptoms (Table 1).

TABLE 1.

Clinical classification of DM1 based on CTG repeat size, symptoms, and disease progression [4, 5].

Group	Symptoms	CTG repeat size	Age of onset	Age of death
Healthy individuals	None	5–37	N/A	N/A
Premutation	None	38–50	N/A	N/A
Late onset (mild)	Cataracts, mild myotonia	51–150	20–70	Normal lifespan
Adult onset (classical)	Muscle weakness with respiratory failure, myotonia, cataracts, cardiac arrhythmias, excessive daytime sleepiness, brain abnormalities (structural and functional), diarrhea, incontinence	50–1000	10–30	48–60
Childhood onset	Psychosocial problems, intellectual impairment, incontinence	>800	1–10	N/A
Congenital	Reduced fetal movements, polyhydramnios, infantile hypotonia, respiratory failure, learning disability, feeding difficulty, mental defects, and retardation	>1000	Birth	45

CTG repeat expansions are easily detectable in blood cells but are often much larger in skeletal muscles, with significant variability between individual muscle cell nuclei. Current research suggests that the somatic expansion, corresponding to the progressive accumulation of repeats in somatic cells, plays the key role in triggering the disease onset. However, the correlation between expansion size and symptom severity varies between tissues [6].

At the molecular level, DM1 is characterized as an RNA “gain‐of‐function” disease. Expanded CTG repeats in the DMPK gene are transcribed into mutant mRNA containing expanded CUG repeats, r(CUG)^exp, which adopt dynamic secondary structures comprising hairpin‐like and single‐stranded forms (Figure 1A,B, see Section 1.2). This toxic RNA aggregates into nuclear RNA foci via phase separation into gel‐like droplets through RNA–RNA interactions [7, 8]. In addition, r(CUG)^exp interacts with RNA‐binding proteins, among which the splicing regulators muscleblind‐like 1 and 2 (MBNL1 and MBNL2) [9, 10] and CELF1 (previously called CUG‐binding protein 1, CUG‐BP1) are most relevant to DM1 pathogenesis [11, 12].

FIGURE 1.

Molecular pathogenesis of DM1. (A) Mutant DMPK is transcribed into mRNA containing expanded CUG repeats, r(CUG)^exp. (B) Toxic r(CUG)^exp adopts single‐stranded and hairpin conformations. (C) r(CUG)^exp aggregates into nuclear foci that sequester MBNL and exclude CELF1 proteins, leading to missplicing of multiple genes and disease onset.

MBNL proteins regulate alternative splicing, 3′‐end cleavage and polyadenylation of pre‐mRNA, mRNA localization and stability, microRNA biogenesis, and circular RNA generation, in particular during embryonic and postnatal development and terminal muscle differentiation [7]. MBNL proteins act in a compensatory manner: in the absence of MBNL1, MBNL2 is upregulated, and vice versa; the loss of one or the other leads to splicing defects [10]. MBNL1 is the predominant isoform in most tissues, except the brain, where MBNL2 predominates. Sequestration of MBNL1 by r(CUG)^exp into ribonuclear foci, detailed in Section 1.3, disrupts MBNL1's normal function as a splicing regulator and induces its proteasomal degradation, causing widespread missplicing of more than 20 pre‐mRNAs, including its own pre‐mRNA (Table 2) [13, 14, 15].

TABLE 2.

Key genes misspliced in DM1 and their associated symptoms [2, 13].

Impacted misspliced genes	Site(s) of splicing alteration	Related DM1 symptoms
Ankyrin 2 (ANK2)	Microexon (miE)	Autism‐related traits
Sarcoplasmic/endoplasmic reticulum calcium ATPase 1, calcium pump 1 (ATP2A1, or SERCA1)	Exon 22	Muscle degeneration
Bridging integrator 1 (BIN1)	Exon 11	Muscle weakness
Chloride channel 1 (CLCN1)	Intron 2, exon 7a	Myotonia
Calcium voltage‐gated channel, subunit alpha 1S (CACNA1S)	Exon 29	Muscle weakness, altered excitation–contraction coupling
Dystrophin (DMD)	Exons 71, 78	Muscle weakness, altered membrane integrity
Insulin receptor (INSR or IR)	Exon 11	Insulin resistance
Muscleblind‐like 1 (MBNL1)	Exons 5, 7, 10	Splicing defects
Nuclear receptor co‐repressor 2 (NCOR2)	Exon 10	Unknown
Cardiac troponin (TNNT2 or cTNT)	Exon 5	Cardiac conduction defects

Together with MBNL proteins, CELF1 regulates multiple aspects of mRNA metabolism, including splicing, translation, polyadenylation, and RNA stability and decay. CELF1 binds single‐stranded CUG repeats but, in contrast to MBNL1, is excluded from r(CUG)^exp ribonuclear foci [16, 17]. In DM1, the level of CELF1 is increased due to its hyperphosphorylation and subsequent stabilization, leading to its gain‐of‐function [5, 18]. Both the phosphorylated (active) and unphosphorylated (inactive) forms of CELF1 have distinct biological functions, making its simple downregulation inadequate as a therapeutic strategy [6]. MBNLs and CELF1 exhibit antagonistic effects on the splicing of key target genes such as CLCN1, INSR, and TNNT2: CELF1 promotes embryonic exon inclusion, whereas MBNLs favor exclusion of the same exons. During embryogenesis, nuclear MBNL levels are low, while CELF1 is abundant. As development progresses, MBNL levels rise and that of CELF1 declines, leading to the transition from embryonic to adult splicing patterns in several downstream targets (such as INSR exon 11, CLCN1 exons containing a stop codon, and TNNT2 exon 5). In DM1, a combination of MBNL1 loss‐of‐function and CELF1 gain‐of‐function reactivates embryonic‐like splicing in adult tissues, leading to expression of fetal isoforms and development of disease symptoms (Figure 1C) [18, 19].

Beyond dysregulated alternative splicing, other molecular mechanisms contribute to DM1 pathogenesis. Thus, bidirectional transcription of expanded CTG repeats leads to the synthesis of expanded r(CAG)^exp RNA transcripts, subjected to repeat‐associated non‐ATG (RAN) translation across all reading frames, with production of homopolymeric polyQ, polyS, and polyA proteins [5, 20]. Among these, polyQ aggregates in the nucleus and in the cytoplasm, leading to apoptosis. Dysregulation of microRNA (miRNA) pathways, which modulate the expression of multiple genes on the post‐transcriptional level and are involved in the control of many fundamental processes, has been linked to MBNL1 sequestration and/or to their direct interaction with r(CUG)^exp [21, 22]. Binding to r(CUG)^exp causes inappropriate redistribution of various transcription factors, leading to dysregulated expression of a large number of genes [5, 18]. Finally, it was proposed that r(CUG)^exp causes mitochondrial dysfunction, excessive production of reactive oxygen species and DNA damage, leading to premature senescence in DM1 cells [23].

1.2. Structure of r(CUG)^exp and d(CTG)^exp Repeats

1.2.1. Short Oligonucleotide Models of CUG Repeats

The formation of double‐stranded secondary structures in short oligonucleotide models of CUG repeats has been initially proposed on the basis of circular dichroism (CD) spectroscopy and gel mobility assays [24] and unequivocally confirmed by the X‐ray structure analysis of r(CUG)₆ and its follow‐up refinement [25, 26]. This oligonucleotide forms an antiparallel double‐stranded helix close to the canonical A‐form RNA (Figure 2A), with an average helical rise of 2.6 Å, a twist of 33.7°, and base pairs inclined 17° toward the minor groove. All uracil residues of the six U·U mispairs, separated by tandem G≡C base pairs, remain intrahelical, interacting via only one hydrogen bond between the carbonyl O4 atom of one uracil and the N3 atom of another. In addition, water molecules surrounding the U·U mispairs form a characteristic pattern of hydrogen‐bonding interactions with both uracils. This contrasts with the U·U mispairs in other sequence contexts, where they typically form two hydrogen bonds. This asymmetric base pairing motif, termed “stretched U–U wobble,” was attributed to the energetic penalty associated with the RNA backbone distortion required for a closer approach of the mispaired uracils, and suggests that U·U mispairs in the CUG duplex can be easily disrupted to accommodate the incoming ligands. NMR and restrained molecular dynamics (MD) study of an RNA oligonucleotide containing three CUG repeats, r(3×CUG), showed a similar overall A‐form geometry (rise 2.7 Å, twist 30.8°), consistent with X‐ray data, yet suggested that U·U mispairs are more dynamic and exist in an equilibrium between stretched U–U wobbles with one hydrogen bond and pseudo‐Watson–Crick, symmetric U–U wobbles with two hydrogen bonds (Figure 2B) [27].

FIGURE 2.

(A) Structure of [r(CUG)₆]₂ obtained from single‐crystal X‐ray diffraction data [25, 26]. Nucleotide colors: C yellow, G green, U red. Replotted from PDB ID: 3GM7 using UCSF Chimera. (B) Close‐up view of U·U mispairs featuring one (top) and two (bottom) hydrogen bonds (cyan) between uracil residues, as observed in the NMR/MD model or r[(3×CUG)]₂ [27]. Replotted from PDB ID: 5VH8 using UCSF Chimera. (C) Close‐up view of the water‐bridged U·U mispair observed in the X‐ray crystal structure of r[(3×CUG)]₂ [28]. Water molecules are shown as red balls. Replotted from PDB ID: 7Y2B using UCSF Chimera.

Additional X‐ray structures of oligonucleotide models of CUG repeats confirmed that U·U mispairs adopt multiple conformations with zero, one, or two hydrogen bonds between uracil residues [29, 30, 31, 32]. A systematic analysis of all available structural data revealed that U·U mispairs in the context of CUG repeats adopt six unique conformations differing by the number of hydrogen bonds, degree, and direction of inclination (toward the minor or major groove of the RNA duplex) [31]. Finally, a recent high‐resolution X‐ray study of a model containing three CUG repeats identified a symmetric, water‐mediated uracil–uracil mispair (U·H₂O·U) at the central 5′‐CUG‐3′/5′‐CUG‐3′ repeat (Figure 2C), which presumably corresponds to the zero‐hydrogen‐bond conformation observed in earlier studies [28]. The water‐bridged U·U mispair likely represents an intermediate state between the asymmetric stretched wobble conformations, highlighting the dynamic nature of U·U mispairs within the A‐form RNA duplex. Importantly, the dynamics of U·U mispairs disrupts the contiguous stacking interactions with the neighboring cytosine residues, which causes destabilization and increased flexibility of the whole RNA duplex structure, potentially allowing local melting and formation of slipped and branched structures [28, 33].

1.2.2. Long r(CUG)^exp Repeats

Magic angle‐spinning solid‐state NMR confirmed that long r(CUG)₉₇ RNA, similar to shorter models, principally adopts the A‐form double‐helical structure, characterized by a C3′‐endo sugar pucker and an anti glycosidic angle [34]. In line, the formation of straight, rod‐like structures with lengths close to those expected for perfect hairpins was observed for r(CUG)₇₅ and r(CUG)₁₃₀ by electron microscopy [35, 36]. However, early chemical and enzymatic probing experiments, combined with computer modeling, suggested that long r(CUG)₄₉ repeats fold into metastable, “slipped” hairpin structures resulting from alternative alignments of the strands and differing by the overhangs, with tetranucleotide (UGCU) or heptanucleotide loops (Figure 3) [37, 38]. A detailed chemical and enzymatic probing study suggested that sequences containing 69 or more CUG repeats form not only perfect or slipped hairpins, but also multi‐hairpins and multibranched loops (Figure 3), with the fraction of these branched forms reaching 15% for the longest construct studied, r(CUG)₁₉₇ [39]. This suggests the dynamic nature of pathologic r(CUG)^exp RNA, enabling MBNL1 binding at transiently single‐stranded sites.

FIGURE 3.

Alternative structures adopted by long r(CUG)^exp RNA.

In DM1 cells, mutated DMPK transcripts containing expanded CUG repeats form RNA foci, which accumulate in the nuclei and can be readily visualized via fluorescence in situ hybridization (FISH) using anti‐sense probes. These ribonuclear r(CUG)^exp foci represent the hallmark of DM1 [40, 41]. In DM1 myotubes (i.e., multinucleate cells formed by fusion of myoblasts and representing the precursors of skeletal muscle fibers), most foci (∼50%) contain a single copy, but rare large foci can contain up to 25 copies of DMPK mRNA. MBNL1 is not required for the formation of the foci, but contributes to their persistence and the formation of large, multimeric foci [42]. The capacity of expanded CUG repeats to form RNA foci has been reproduced in vitro. Thus, synthetic RNA containing 30 or more CUG repeats undergoes sol–gel transition at concentrations as low as 25 nM, with formation of micrometer‐sized spherical clusters through intermolecular interactions between RNA molecules [8, 43].

1.2.3. Structure of d(CTG)^exp

The formation of hairpin structures by CTG repeats in the coding strand and/or by CAG repeats in the template DNA strand are the key molecular events at the origin of genomic instability, ultimately leading to expansion of these repeats and the onset of DM1. Several mechanisms have been proposed for this expansion, including the formation of hairpins during DNA replication (Figure 4A), DNA repair, homologous recombination, or transcriptional stress [45, 46].

FIGURE 4.

(A) Replication model of CTG repeat expansion mediated by the formation of hairpins in the coding strand. (B) Overall structure and (C) close‐up view of the T·T mismatch in the NMR structure of the DNA hairpin formed by d(CTG)₄ [44]. Replotted from PDB ID: 8X4F using UCSF Chimera.

Intrastrand hairpin structures formed by short d(CTG) _n repeats are well characterized in vitro. Similar to r(CUG), these hairpins consist of tandems of G≡C base pairs interrupted by T·T mismatches. The loop of the hairpin depends on the repeat number (n): while even‐numbered repeats form blunt‐end hairpins with a TGCT tetranucleotide loop, odd‐numbered repeats form either blunt‐end hairpins with an energetically less favorable CTG trinucleotide loop, or hairpins with a TGCT tetranucleotide loop and an overhang [47, 48, 49, 50, 51, 52]. This conformational variability highlights the “slippery” nature of these hairpins, leading to genomic instability. A high‐resolution structure of the blunt‐end hairpin with a TGCT loop, formed by d(G(CTG)₄C), was solved by NMR (Figure 4B) [44]. According to NMR data and MD simulations, T·T mismatches are somewhat less dynamic compared to U·U mispairs in r(CUG)^exp, adopting predominantly the symmetric wobble pairing geometry with two hydrogen bonds between thymine residues (Figure 4C). Similar to U·U mispairs, they can adopt multiple local conformations without altering the global conformation of the double helix [27].

1.3. r(CUG)^exp–MBNL1 Interaction: The Core of Molecular Pathogenesis

The sequestration of MBNL1 by expanded CUG repeats is the critical molecular event leading to MBNL1 loss‐of‐function and dysregulated splicing in DM1. MBNL1, along with its paralogs MBNL2 and MBNL3, belongs to the family of Muscleblind‐like proteins, tissue‐specific splicing regulators that exhibit distinct subcellular distributions yet share structural similarities and functional compensation [53, 54, 55]. The human MBNL1 coding sequence is located in chromosome 3 and contains 10 exons, of which six (exons 3, 5, 6, 7, 8, and 9) are alternatively spliced, resulting in at least ten MBNL1 isoforms with molecular weights ranging from 35 to 43 kDa (Figure 5A) [56]. The expression of these isoforms is tightly regulated, including autoregulation by MBNL1 itself [14, 57]. Constitutively included exons 1, 2, and 4 encode four CCCH‐type zinc‐finger (ZnF) domains, organized in two similarly structured tandem pairs (ZnF1/2 and ZnF3/4) and critical for the recognition of the cognate RNA sequence. Exon 3, frequently included, encodes the flexible linker joining the second and the third ZnF domains, modulating MBNL1's affinity to its RNA targets. Exons 5 and 6 regulate the nuclear localization of MBNL1, whereas exon 7 is critical for MBNL1 multimerization [10, 36, 56]. Unstructured regions in exon 3 and in the C‐terminal tail (exons 6–8 and 10) enable membrane docking of MBNL1, thereby regulating mRNA localization and local translation [58].

FIGURE 5.

(A) Organization of the human MBNL1 gene. White boxes, constitutively spliced exons; gray boxes, alternatively spliced exons; hatched boxes, untranslated regions; tan boxes, zinc finger domains. (B) Close‐up view of the protein–RNA interface at the ZnF4 binding site of the MBNL1 ZnF3/4 fragment. The hydrogen bonds between the nucleobases and the protein (tan‐colored) are shown as thin cyan lines and Zn ions as violet spheres. (C) Coulombic potential surface representation of the same view as in (B). (D) Overall view of the MBNL1 ZnF3/4 fragment bound to two 5′‐CGCUGU‐3′ RNA strands (backbone, gray; nucleobases: G green, C yellow, U cyan). Panels (B–D) replotted from the PDB ID: 3D2S using UCSF Chimera. (E) Putative model of interaction of MBNL1 with a single‐stranded fragment of r(CUG)^exp. (F) Electron micrograph of r(CUG)₁₃₆–MBNL1 complexes showing the free RNA hairpin (arrow) and RNA‐bound MBNL1 molecules assembled in ring‐like structures. Panel (F) adapted with permission [36]. Copyright 2007, Y. Yuan et al.

The tandem ZnF domains of MBNL1 are highly evolutionarily conserved. Each ZnF motif recognizes a single GpC dinucleotide within the MBNL1's well‐established cognate RNA sequence, 5′‐YGCY‐3′ (Y = U or C) [59, 60, 61, 62], with ZnF1/2 showing greater specificity than ZnF3/4 [63]. MBNL1 interacts with this consensus motif both in introns and in 3′‐UTR, suggesting the same binding mode. The structural details of this interaction are known thanks to the seminal X‐ray study of ZnF3/4 domains of MBNL1 bound to r(CGCUGU) [64]. In this complex, each ZnF domain forms a base‐specific binding pocket, which includes the highly conserved aromatic (Phe or Tyr) and arginine residues that make stacking interactions with guanine and cytosine residues of the GpC step in RNA. The binding pocket, involving the zinc‐coordinated cysteine residues, forms multiple hydrogen bonds with the Watson–Crick edges of the three nucleobases, enabling their highly specific recognition (Figure 5B,C); in addition, the complex is stabilized by hydrogen bonding between the side chains of ZnF domains and the RNA sugar–phosphate backbone. The tandem ZnF domains adopt an overall symmetric fold, with RNA‐binding surfaces of ZnF3 and ZnF4 oriented away from each other. This results in an antiparallel orientation of the two single‐stranded RNA chains, suggesting a chain‐reversal loop trajectory for the MBNL1‐bound RNA strand (Figure 5D,E). This binding mode was largely confirmed by a more recent NMR solution study, which, however, suggested that the individual ZnF domains of each tandem are not equivalent and that RNA binding occurs only at the second zinc finger of each pair (i.e., ZnF2 and ZnF4) [65]. In either case, the presence of two tandem ZnF motifs allows for multiple contacts of MBNL1 along r(CUG)^exp.

Despite this well‐established binding mode, ambiguity persists in the literature regarding the binding of MBNL1 to r(CUG)^exp. Indeed, early studies, considering the propensity of CUG repeats to readily form hairpin structures (cf. Section 1.2), attributed MBNL1 sequestration to its interaction with the RNA hairpin structure adopted by r(CUG)^exp [9]. Major support for this hypothesis came from the observation that the presence of MBNL1 protects GpC sites in r(CUG)₅₄ from RNase T1 cleavage, interpreted as the persistence of the double‐stranded RNA structure upon MBNL1 binding [36]. However, the same result may also be interpreted as a footprint of MBNL1 binding to the consensus sequence and protecting the GpC sites from cleavage. The image of MBNL1 bound to r(CUG)^exp hairpins has spread in the literature, with broad‐readership highlights [66], influential reviews [67, 68, 69], and even recent specialized studies depicting this binding mode. The alternative model, that is, sequence‐ rather than structure‐driven binding of MBNL1 to the single‐stranded form of r(CUG)^exp, was also postulated in the early studies [59, 70], but struggled to become widely accepted. Nevertheless, multiple facts accumulated over two decades of research strongly suggest that MBNL1 interacts with the single‐stranded form of r(CUG)^exp and melts the hairpin structure upon binding, at least locally. First, and most importantly, MBNL1 engages the Watson–Crick edges of GpC nucleotides as shown above, which requires the opening of G≡C base pairs and disruption of the hairpin structure [64, 65]. Second, MBNL1 binds U‐rich RNA strands containing as few as one or two 5′‐UGCU‐3′ sites, which cannot form stable hairpins, with a low‐nanomolar affinity (1 GC, 12 or 45 nM; 2 GC, 11 nM, EMSA), which is one order of magnitude stronger than its affinity for the relatively stable (CUG)₄ and long (CUG)₉₀ hairpins (K _d = 170 and 260 nM, respectively) [61, 70, 71]. Same trend was observed in a more recent study [63]. Third, MBNL1 binding negatively correlates with thermal stability of r(CUG) hairpins [71], and stabilization of the hairpins via introduction of pseudouridine or 2′‐OMe‐uridine modifications dramatically reduces MBNL1 affinity [72]. Fourth, MBNL1 binding to an RNA hairpin containing two 5′‐CGCU‐3′ sites in the stem part and mimicking its binding site in TNNT2 pre‐mRNA leads to the opening of the hairpin, as unambiguously demonstrated by FRET, circular dichroism (CD) spectroscopy and NMR experiments [65, 70]. Fifth, MBNL1 binds to RNA hairpins regardless of whether its cognate sequence is located in the loop or stem, as demonstrated by an unbiased Bind‐n‐Seq study [62]. Collectively, these results speak in favor of a model where MBNL1 locally disrupts r(CUG)^exp hairpins and binds the separated strands, which ultimately leads to multimerization of MBNL1 molecules via their C‐terminal tails as schematically shown in Figure 1C. Of note, this model agrees with the early electron microscopy study that observed the formation of ring‐like MBNL1 multimers, with a diameter of ∼18 nm, bound in the middle of rod‐shaped RNA hairpins (Figure 5F) [36].

In healthy cells, MBNL proteins are found both in the nucleus, where they regulate alternative splicing and polyadenylation of pre‐mRNA, and in the cytoplasm, where they control mRNA localization by docking to membranes and RNA transport granules. In DM1 cells, interaction of MBNL1 with r(CUG)^exp results in its sequestration into ribonuclear foci. MBNL1 is not required for the formation of r(CUG)^exp foci; however, the presence of MBNL1 increases their size and persistence, presumably due to MBNL1 multimerization via its C‐terminal tail [43]. In particular, MBNL1 contributes to the formation of large foci containing multiple copies of the mutated DMPK transcript [42]. Interestingly, recent studies suggest that r(CUG)^exp foci sequester only a fraction (<20%) of nuclear MBNL1; in combination with other pathogenic mechanisms, this depletion is sufficient to impair the splicing and induce DM1 pathology [42].

Considering the critical role of MBNL1 sequestration in the molecular pathogenesis of DM1, most therapeutic strategies aim to inhibit the r(CUG)^exp–MBNL1 interaction and thereby release MBNL1 from the ribonuclear foci and rescue the impaired splicing. These strategies employ diverse ligands that bind CUG repeats and, most typically, either stabilize the hairpin form, less prone to MBNL1 binding, or directly compete with MBNL1 via binding to the single‐stranded form of r(CUG)^exp. In addition, some of the ligands induce degradation of the toxic r(CUG)^exp by various mechanisms, preventing repeated MBNL1 sequestration. Section 2 focuses on these ligands and their effects in vitro and in DM1 models. It should be noted that other therapeutic strategies for DM1 are being explored, such as enhancing endogenous MBNL1 expression to counteract its sequestration by r(CUG)^exp [73, 74, 75, 76, 77, 78]. These approaches are out of the scope of this review.

2. r(CUG)^exp and d(CTG)^exp Ligands

2.1. Triaminotriazine‐Based Ligands

2.1.1. Triaminotriazine–Acridine Conjugates

Triaminotriazine‐based ligands, extensively developed by Zimmerman and coworkers since 2009, represent one of the most well‐studied families of ligands targeting d(CTG)^exp and r(CUG)^exp repeats. The choice of the 2,4,6‐triamino‐1,3,5‐triazine unit, also known as melamine, is governed by its presumed capacity to form a perfect set of six “Janus‐wedge” hydrogen bonds [79] with two mispaired residues of uracil in RNA (or thymine in DNA), enabling highly specific recognition of U·U or T·T mispairs (Figure 6A). The chronicle of the development of these ligands through a rational design approach, rooted in the principles of supramolecular chemistry, has been partially described by Zimmerman in his autobiographical account [83].

FIGURE 6.

(A) Putative “Janus‐wedge” hydrogen‐bonding interactions of a 1,3,5‐triamino‐2,4,6‐triazine moiety with U·U or T·T mispairs. (B) Structure of the prototype triaminotriazine–acridine ligand Z1 and its schematic binding mode to r(CUG) or d(CTG) repeats. (C) Schematic representation of a four‐way junction structure observed in crystals of Z1 (violet) bound to d[^BrUT(CTG)₃AA]. (D) Close‐up view of the ligand binding site in the structure schematically shown in (C). (E) N‐Methylated triaminotriazine–acridine conjugates. (F) Macrocyclic triaminotriazine–acridine conjugates and a molecular model of the ligand A3D3 bound to d(CTG)₆; K _d values are from ITC experiments. (G) Structure of the d(CTG)‐alkylating agent 82.1. (H) Correction of the missplicing of INSR pre‐mRNA in DM1‐modeling HeLa cells treated with 82.1. Panels C and D reproduced with permission [80]. Copyright 2020, American Chemical Society. Right part of panel F reproduced with permission [81]. Copyright 2019, Elsevier Ltd. Panel H reproduced with permission [82]. Copyright 2022, American Chemical Society.

In simple triaminotriazine–acridine conjugates, such as Z1 (Figure 6B), the triaminotriazine moiety is linked via a short yet flexible linker to acridine, a prototypical intercalator known to interact with double‐stranded DNA and RNA [84, 85]. Stacking of acridine and triaminotriazine moieties decreases nonspecific binding of the resulting ligands to DNA and RNA, by allowing intercalation of the acridine unit only at the sites adjacent to U·U (or T·T) mispairs recognized by the triaminotriazine unit (Figure 6B). Ligand Z1, a prototype of this family, induces moderate thermal stabilization of a DNA sequence containing two 5′‐CTG‐3′/5′‐CTG‐3′ sites (ΔT _m = 4.5°C), without affecting the stability of the fully matched DNA duplex [86]. Binding of Z1 to T·T mispairs in DNA is characterized by a dissociation constant K _d = 0.39 μM per single T·T site (isothermal calorimetry, ITC) and a 13‐, 169‐, or 85‐fold lower affinity for duplexes containing C·C, A·A, or G·G mispairs, respectively. Z1 binds to U·U mispairs in the homologous RNA sequence with lower affinity than to DNA (K _d = 2.1 μM for a single U·U site) but maintains the selectivity with respect to other mispairs. Furthermore, Z1 disrupts the r(CUG)₄–MBNL1 and r(CUG)₁₂–MBNL1 complexes in vitro with IC₅₀ values of about 50 μM (K _I = 6–7 μM, EMSA). The activity of the ligand in this assay was essentially unaffected by the presence of tRNA as a nonspecific competitor, confirming its high selectivity to r(CUG)^exp sequences. However, despite its promising in vitro activity, ligand Z1 was poorly water‐soluble, hardly cell‐permeable, and rather toxic, inciting further optimization of this scaffold.

Aiming to elucidate the structural details of T·T mismatch recognition, Zimmerman, Hou, and coworkers solved the crystal structure of Z1 in a complex with a DNA fragment containing three CTG repeats [80]. Intriguingly, this study revealed that binding of the ligand induced strong deformation of the DNA backbone with formation of a four‐way DNA junction structure, reminiscent of Holliday junctions (Figure 6C). The stacking of triaminotriazine and acridine units was well‐preserved in the bound ligand, and the acridine units intercalated next to T·T mispairs as initially hypothesized (Figure 6D). However, the triaminotriazine units flipped out one of the mispaired thymine residues of each T·T mismatch, resulting in DNA backbone bending, while forming direct hydrogen bonds only with intrahelical thymines (e.g., T10 in Figure 6D). Whether DNA backbone distortion and formation of four‐way junction structures take place upon binding of Z1 and related ligands to CTG repeats in solution and in the cellular context is not known; however, the formation of such structures could contribute to the genomic instability of CTG repeats induced by related ligands such as 82.1. ¹

Despite its poor biological properties, the molecular design of Z1 proved to be extremely well‐suited for recognition of U·U and T·T mispairs, since changing the length of the linker connecting the triaminotriazine and the acridine moieties, as well as the replacement of triaminotriazine unit with another heterocycle, led to a dramatic loss of the affinity and/or selectivity, highlighting the critical importance of the triaminotriazine moiety [86, 87]. Notably, methylation of the triaminotriazine moiety strongly influences the DNA‐vs.‐RNA selectivity of these ligands. Thus, mono‐ and dimethylated ligands 88.2 and 88.3 (Figure 6E) weakly bind the RNA sequence containing a U·U mispair (K _d > 200 μM, ITC), while maintaining high affinity to the DNA counterpart (K _d = 0.8 and 0.12 μM, respectively) [88]. This was attributed to the modification of their binding mode to T·T mispairs from the Janus wedge‐type triplet (cf. Figure 6A) to a stretched wobble base pair, favored in the DNA context. In line with their reduced affinity to RNA, ligands 88.2 and 88.3 are unable to disrupt the r(CUG)₁₂–MBNL1 complex in vitro. Methylation of more than one amino group of the triaminotriazine moiety dramatically reduces the affinity of the ligands not only to RNA, but also to DNA mismatches. At the same time, methylation of the triaminotriazine moiety enhances the intramolecular stacking in these ligands, resulting in even lower nonspecific binding to double‐stranded DNA. Thus, these ligands represent promising candidates for molecular therapy of DM1 at the DNA level.

To further decrease nonspecific binding to double‐stranded DNA and cytotoxicity of acridine conjugates, the Zimmerman group designed macrocyclic ligands (e.g., A3D3 and A4D3, Figure 6F), in which the molecular scaffold enforces the intramolecular stacking of triaminotriazine and acridine units and, at the same time, provides the ligand with the necessary flexibility to adapt to its binding site [81]. MD simulations suggest that A3D3 binds to d(CTG)₆ by inserting its acridine unit between the T·T mispair and the adjacent G≡C base pair, with the triaminotriazine moiety displacing one of the mismatched thymines and forming three hydrogen bonds with the intrahelical thymine. In contrast, these simulations predicted no stable complex with the RNA counterpart. In line, ITC demonstrated moderate affinity of these ligands to the DNA duplex presenting a T·T mispair (K _d = 18.1 and 19.8 μM for A3D3 and A4D3, respectively) and no binding to the fully matched DNA duplex or the RNA duplex presenting a U·U mispair. A3D3 and A4D3 inhibited in vitro transcription of a plasmid containing 74 d(CTG·CAG) repeats (∼60% and 85% inhibition at 25 μM, respectively), presumably via stabilization of the DNA hairpin formed by the d(CTG)₇₄ strand; in contrast, the transcription of a control plasmid lacking these repeats was barely impacted. While A3D3 was minimally cytotoxic, the more efficient ligand A4D3 showed significant cytotoxicity in HeLa cells (IC₅₀ ≈ 20 μM), highlighting the necessity of further optimization.

Building on the high affinity and selectivity of 88.3 to T·T mismatches in DNA, Zimmerman and coworkers designed a DNA‐alkylating agent 82.1 selectively targeting d(CTG)^exp repeats [82]. In addition to acridine and N ²,N ²‐dimethyl‐2,4,6‐triamino‐1,3,5‐triazine units, the structure of 82.1 (Figure 6G) comprises a residue of chlorambucil as a DNA‐alkylating warhead as well as a biotin moiety, initially added as a handle for DNA pull‐down but which has proved to be essential for the ligand's selectivity. Ligand 82.1 selectively alkylates the d(CTG)₁₂ hairpin (51% alkylation in the presence of 100 μM of 82.1, as per PAGE analysis), with negligible alkylation of a control DNA sequence lacking T·T mismatches. At concentrations of 12.5–25 μM, 82.1 selectively inhibited in vitro transcription of the d(CTG)₇₄ template. Treatment of DM1‐modeling HeLa cells expressing 960 interrupted CTG repeats with 12.5–25 μM of 82.1 resulted in a significant decrease of r(CUG)^exp transcript levels (by 45–72%) as well as a reduction of the total area of r(CUG)^exp–MBNL1 foci by 49–70%, confirming the interference with the transcription of CTG repeats. Furthermore, treatment with 82.1 fully corrected the DM1‐characteristic missplicing of the INSR gene in a cellular DM1 model (Figure 6H). Most intriguingly, 82.1 was able to modulate the length of CTG repeat expansion in DM1 patient‐derived fibroblasts. Thus, repeated treatment with 25 μM 82.1 resulted in a change of distribution of the length of CTG repeats, with odd numbers of 48‐hour treatment rounds inducing repeat contractions and even numbers of treatment rounds leading to repeat expansion. This genomic instability may be attributed to the dysfunctional DNA repair triggered by DNA alkylation at CTG repeats or to the ligand‐induced formation of higher‐order secondary structures (cf. Figure 6C). Although the origins of this oscillating effect are not totally clear, five rounds of treatments with 82.1 were sufficient to decrease the fraction of expanded (i.e., 66 and 125) CTG repeats from 4–5% to 0.1%. These results suggest the potential of this compound to reduce the length of d(CTG)^exp repeats in the genome of DM1 patients and thus to cure the very origin of the disease.

To harness the capacity of triazine–acridine conjugates to act at the RNA level, that is, to bind the hairpin form of r(CUG)^exp and to disrupt the r(CUG)^exp–MBNL1 interaction, Zimmerman and coworkers deployed a medicinal chemistry program aimed to improve aqueous solubility, cellular permeability, nuclear influx and RNA affinity, while at the same time limiting the cytotoxicity of the ligands. The polyamine‐conjugated derivative 89.1 (Figure 7A) moderately stabilized the hairpin form of r(CUG)₁₂ in UV melting experiments (ΔT _m = 2.5°C and 5.5°C in the presence of 1 and 3 molar equivalents of ligand, respectively) and disrupted the r(CUG)₁₂–MBNL1 complex even in the presence of a large excess of competitor tRNA (IC₅₀ = 15 μM, surface plasmon resonance (SPR)), demonstrating a modest improvement over the prototype Z1 [89]. However, the polyamine remnant significantly enhanced cellular uptake of the ligand, as observed by microscopy due to the fluorescence of the acridine moiety. Treatment of DM1‐modeling HeLa cells with 75 μM 89.1 reduced the fraction of cells presenting r(CUG)^exp–MBNL1 foci by ∼86% and improved the missplicing of the INSR gene from 35% to 45% of isoform B (with the normal level being 57%), without significant cytotoxicity.

FIGURE 7.

(A) Structures of polyamine‐substituted and dimeric triaminotriazine–acridine conjugates. ΔT _m values refer to the ligand‐induced stabilization of r(CUG)₁₂ hairpin at a 1:1 ligand‐to‐RNA ratio; IC₅₀ values indicate the capacity of ligands to inhibit the r(CUG)₁₂–MBNL1 complex (SPR) [89, 90]. (B) Schematic representation of polypeptide‐assembled multivalent ligands. K _I values refer to the capacity of ligands to disrupt the r(CUG)₁₂–MBNL1 complex as per fluorescence anisotropy [91].

Further optimization of this scaffold was achieved by dimerization of triaminotriazine–acridine “modules,” aiming to attain multivalent recognition of r(CUG)^exp via simultaneous binding to two U·U sites. Of note, the multivalence effect may be mitigated by the conformational rigidity of the dimers, which decreases binding efficiency; in turn, increased conformational flexibility of dimers raises the entropic cost of the binding. In addition, the increased molecular weight and the size of dimers reduce their drug‐likeness. To optimize the molecular design of dimeric ligands, Zimmerman and coworkers synthesized and evaluated a library of ten dimers differing in the length and chemical composition of the linkers, as well as their attachment points to acridine units [90]. Among these, oligoether‐linked dimers, such as 90.7 (Figure 7A), suffered from poor aqueous solubility and showed modest stabilization of the r(CUG)₁₂ in UV melting experiments, suggesting that they interact with U·U sites through only one of their binding modules. In contrast, oligoamine‐linked dimers, such as 90.9, strongly stabilized r(CUG)₁₂ in UV melting experiments (e.g., ΔT _m = 9.3°C for 90.9), suggesting a bivalent binding to the target, likely promoted by electrostatic interactions of the protonated linker with the RNA backbone. The values of dissociation constants, obtained from fluorimetric titrations with TAMRA‐r(CUG)₆, confirmed these conclusions (K _d = 23 and 0.32 μM for 90.7 and 90.9, respectively). In line with its higher affinity, the oligoamine‐linked dimer 90.9 was a much more potent inhibitor of the r(CUG)^exp–MBNL1 interaction (IC₅₀ = 1.1 μM, SPR) than oligoether 90.7 and the monomeric analog 89.1 (Figure 7A). The effect of 90.9 on the r(CUG)^exp–MBNL1 interaction was studied at the single‐molecule level, suggesting that the ligand does not merely block MBNL1 binding sites on r(CUG)^exp via a simple competitive mechanism, but rather binds to r(CUG)^exp concurrently with MBNL1, thereby reducing the protein's affinity and promoting its dissociation [92]. This mechanism is compatible with the ligand‐induced conformational shift of r(CUG)^exp toward a hairpin form characterized by lower MBNL1 binding (cf. Section 1.3). In the HeLa cell DM1 model, the dimer 90.9 induced a dose‐dependent dispersion of r(CUG)^exp foci, with a full dispersion observed in less than 10 h upon treatment with 50 μM of the ligand; no cytotoxicity was observed at concentrations up to 75 μM [90].

To further amplify the r(CUG)^exp affinity and selectivity of triaminotriazine–acridine conjugates without compromising their cellular uptake, Zimmerman and coworkers developed conjugates of triaminotriazine‐based ligands with cationic cell‐penetrating peptides [91]. These synthetic peptides represent polymers of d‐ or l‐glutamic acid (or of their racemic mixture), decorated with guanidium units (providing cationic charge and aqueous solubility) and ligand residues attached via triazole linkers, in a ratio from 1/2 to 1/10 (Figure 7B). The resulting polymers inhibit the r(CUG)^exp–MBNL1 interaction at low‐nanomolar concentrations, with polymers bearing higher ligand loading being most active (e.g., PLG ₅₀ ‐1/2, K _I = 2.5 nM), but also more toxic and less water‐soluble. This represents more than two to three orders of magnitude increase in potency with respect to monomeric ligands such as Z1. Both d‐ and l‐glutamic acid‐based peptides, such as PLG ₅₀ ‐1/2 and PLG ₅₀ ‐1/5, are rapidly taken up by cells at a sub‐micromolar concentration and decrease the number and intensity of nuclear r(CUG)^exp–MBNL1 foci in DM1‐modeling HeLa cells. Accordingly, treatment with 0.5–1 μM of peptides fully rescued the missplicing of the INSR gene without inducing cytotoxicity. Finally, in a functional larvae crawling assay performed with the Drosophila DM1 model, PLG ₅₀ ‐1/2 showed a dose‐dependent improvement of larval mobility, with full recovery to the nonpathological level observed at 10 μM [91]. Of note, cationic polycarbonates decorated with only guanidinium units were also shown to bind to r(CUG)₉₆ in vitro; however, their selectivity with respect to other RNA sequences or the activity in DM1‐relevant models have not been assessed [93].

2.1.2. Groove‐Binding Triaminotriazine‐Based Ligands

Despite the high affinity of triaminotriazine–acridine conjugates to r(CUG)^exp and d(CTG)^exp repeats, the intrinsic drawbacks of this family of ligands include their poor aqueous solubility and the cytotoxicity of the acridine unit. To eliminate these drawbacks, Zimmerman and coworkers developed an alternative family of 1,3,5‐triaminotriazine derivatives comprising a residue of 1,4‐diamidinobenzene as the central nontoxic unit, providing aqueous solubility and RNA affinity via minor‐groove binding and two triaminotriazine units enabling multivalent recognition of r(CUG)^exp via simultaneous interaction with two U·U mispairs (Figure 8A). Initial molecular modeling studies predicted that in ligands such as 94.3, each triaminotriazine unit could interact with both uracil residues of a U·U mispair by forming a total of up to five to six hydrogen bonds (cf. Figure 6A) [94]. However, more recent MD simulations demonstrated that one uracil residue of each U·U mispair undergoes base flipping, leading to the formation of hydrogen bonds between the triaminotriazine residue, the intrahelical uracil residue, and the phosphate group of the flipped‐out uracil (Figure 8B,C) [95].

FIGURE 8.

(A) Structure of the 1,4‐diamidinobenzene–triaminotriazine conjugate 94.3 and schematic depiction of its binding mode to r(CUG)^exp. Reproduced with permission [94] Copyright 2014, American Chemical Society. (B) Molecular model of 94.3 bound to r(CUG)_6×2 duplex, according to MD simulations. The U·U mispairs are shown in green. (C) Hydrogen bonding interaction of triaminotriazine unit with uracil residues of r(CUG)_6×2. Panels B and C reproduced with permission [95]. Copyright 2020, Canadian Science Publishing. (D–F) Structures of groove‐binding triaminotriazine conjugates 96.5 and 97.2a–b (D), cationic polymer 98.4 (E), and the triaminotriazine‐based RNA‐cleaving ligand 100.9 (F).

The prototype of this family, 94.3 (Figure 8A), binds to r(CUG)₁₂ with a micromolar affinity (K _d = 8 μM, ITC), without detectable binding to the d(CTG) counterpart, and disrupts the r(CUG)^exp–MBNL1 interaction with an apparent inhibition constant K _I = 8 μM (EMSA), which is close to its thermodynamic dissociation constant. In terms of r(CUG)^exp affinity and r(CUG)^exp–MBNL1 inhibition, 94.3 is very close to the acridine conjugate Z1 (K _I = 7 μM, Figure 8B). However, in contrast to Z1, ligand 94.3 is not cytotoxic in mammalian cells at concentrations up to 100 μM, and has a maximum tolerated dose of 50–100 mg/kg in mice (single injection). In DM1‐modeling HeLa cells, treatment with 100 μM 94.3 disperses ribonuclear r(CUG)^exp–MBNL1 foci and partially rescues the missplicing of the TNNT2 and INSR genes (by 46% and 56%, respectively). In a Drosophila DM1 model, flies fed with 200–400 μM 94.3 demonstrated a significant improvement of the DM1‐characteristic glossy‐eye phenotype [94].

A charge‐neutral analog of 94.3, ligand 96.5 (Figure 8D) was designed by replacing the 1,4‐diamidinobenzene unit with a thiazole‐based peptidomimetic linker as a generic minor‐groove binder [96]. This modification resulted in binding of 96.5 not only to r(CUG) repeats, but also to d(CTG) repeats in DNA (K _d = 16 and 141 μM, respectively, ITC), suggesting that it could also act at the transcriptional level. Indeed, ligand 96.5 selectively inhibited in vitro transcription of the d(CTG)₉₀‐containing plasmid (IC₅₀ = 300 μM). In HeLa cells modeling DM1, 96.5 reduced the level of the toxic r(CUG)^exp transcript (IC₅₀ = 400 μM) and, albeit at a rather high concentration of 800 μM, decreased the area of ribonuclear r(CUG)^exp–MBNL1 foci by more than 80%. In line, treatment with 400 μM 96.5 resulted in a full correction of the missplicing of the INSR gene. Comparing with 94.3, the charge‐neutral analog was slightly less active in the splicing correction assay; however, this was mitigated by its substantially lower cytotoxicity, as no cell death was observed at up to 1 mM of 96.5.

To improve the r(CUG)^exp affinity of 94.3, Zimmerman and coworkers applied the same multimerization strategy as for the dimeric acridine conjugates described above, which resulted in conjugates 97.2a and 97.2b presenting two 1,4‐diamidinobenzene and four triaminotriazine units each (Figure 8D) [97]. According to in vitro, cellular, and in vivo assays (Table 3), dimers 97.2a and, to a lesser extent, 97.2b are much more potent r(CUG)^exp ligands than the prototype 94.3. In particular, the K _I values (Table 3) indicate that the dimer 97.2a is a more than 1000‐fold potent inhibitor of the r(CUG)^exp–MBNL1 interaction than 94.3 in vitro. However, the activity of 97.2a in the cellular DM1 model was limited by its cellular permeability, and a rather high concentration (100 μM) was required for full correction of the INSR missplicing. In addition, the maximum tolerated dose in mice was much lower for 97.2a than for 94.3 (3 vs. 50–100 mg/kg, respectively).

TABLE 3.

Comparison of ligand 94.3, dimer 97.2a, and polymer 98.4 in in vitro, cellular, and in vivo assays [94, 97, 98].

Assay	94.3	97.2a	98.4
r(CUG)16–MBNL1 inhibition in vitro (EMSA)	IC50 = 189,000 nMa	IC50 = 290 nM	Not studied
	K I = 16,000 nM	K I = 25 nM
Cellular assays in DM1‐modeling HeLa cells:
‐ Dispersion of r(CUG)exp–MBNL1 foci	Dispersion of 20% of foci after 48 h (100 μM)	Dispersion of 72% of foci after 48 h (1 μM)	Dispersion of 76% of foci after 48 h (0.5 μM)
‐ INSR splicing correction	56% recovery (100 μM)	Full recovery (100 μM)	Full recovery (200 nM)
Improvement of DM1 phenotype in Drosophila flies:
‐ Glossy eye	Partial improvement at 400 μM	Partial improvement at 50 μM	—
‐ Larvae crawling assay	84% improvement at 400 μM	Full recovery at 100 μM	—
‐ Adult climbing assay	31% improvement at 400 μM	—	54% recovery at 80 μM
Cytotoxicity in HeLa cells	None (100 μM, 72 h)	10% (100 μM, 96 h)	IC50 ≈ 14 μM (72 h)
Maximum tolerated dose in mice (i.p. injection)	Between 50 and 100 mg/kg (one‐time)	3 mg/kg (daily)	40 mg/kg

^a These values differ from the ones presented in Figure 8A due to a different length of the r(CUG) repeat.

The cellular uptake issues observed with oligomers 97.2a and 97.2b are unsurprising, considering their high molecular weight (e.g., 1166 Da for 97.2a) and large number of hydrogen bond‐donor and ‐acceptor groups, which illustrates the inherent limitation of the multimerization approach. To overcome this issue, an untrivial solution explored by Zimmerman and coworkers consisted in a further increase of the degree of oligomerization, which resulted in the cationic polymer 98.4 (Figure 8E) reminiscent of cationic cell‐penetrating peptides [98]. Indeed, cationic polypeptides such as oligoarginines and aromatic foldamers are readily taken up by cells despite their high molecular weight, which makes them promising delivery agents [99]. Polymer 98.4 was obtained and studied as a mixture of oligomers ranging from 4‐ to 8‐mers and an average molecular weight of 2 kDa. At high concentrations, the interaction of 98.4 with r(CUG)₁₆ resulted in the formation of aggregates, precluding detailed investigations by classical in vitro techniques. However, polymer 98.4 was cell‐permeable and showed high activity in cellular and in vivo DM1 models (Table 3). Thus, a 68–76% decrease of ribonuclear r(CUG)^exp foci was observed in DM1‐modeling HeLa cells treated with 100–500 nM 98.4, even though a full dispersion could not be achieved; a similar effect was observed in DM1 patient‐derived fibroblasts. Despite being moderately active with respect to disruption of r(CUG)^exp foci, 98.4 fully rescued the INSR missplicing at a concentration as low as 200 nM. In the Drosophila DM1 model, flies fed with 80 μM 98.4 showed a significant improvement in their climbing ability, whereas a fivefold higher concentration of the monomer 94.3 was required to achieve a comparable effect. Moreover, polymer 98.4 was well tolerated in mice, with a maximum tolerated dose of 40 mg/kg. Upon peritoneal administration in the liver‐specific DM1 mouse model, 98.4, at a daily dose of 2 mg/kg, strongly decreased the level of toxic r(CUG)^exp RNA as well as the number r(CUG)^exp foci in liver biopsies, and corrected the splicing defects of several MBNL1‐regulated genes. The high activity of 98.4 in vivo demonstrates the potential of oligomeric cell‐permeant drugs for DM1 treatment.

The groove‐binding triaminotriazine scaffold has also been exploited in the design of the “RNA degrader” 100.9 endowed with two tris(2‐aminoethyl)amine (TREN) residues as RNA‐cleaving units (Figure 8E) [100]. As a side effect of this modification, 100.9 acquired strong binding to CTG repeats in DNA (K _d = 5 μM, ITC), unlike the parent ligand 94.3. Cleavage of r(CUG)₁₆ in vitro required rather high concentrations of 100.9 (50–100 μM) and extended incubation time (24–48 h); nevertheless, the cleavage was specific with respect to the r(CUG)^exp sequence. Conversely, neither the parent ligand nor TREN, nor their mixture, cleaved r(CUG)₁₆. In DM1‐modeling HeLa cells, treatment with 50–150 μM 100.9 not only decreased the number of ribonuclear r(CUG)^exp foci (as was already observed for 94.3 and related derivatives), but also strongly reduced the level of r(CUG)₉₆₀ transcript, while 94.3 was inactive in this regard. This supports the putative mechanism of action of 100.9 via ligand‐induced degradation of r(CUG)^exp. Finally, in the Drosophila DM1 model, 100.9 induced a dose‐dependent improvement of the DM1‐characteristic phenotypical features such as impaired larval mobility and eye defects; in both cases, the parent ligand 94.3 was less efficient. Altogether, 100.9 represents a paradigm of a multifunctional DM1 drug acting at several levels: (i) inhibition of transcription of CTG·CAG repeats via binding to d(CTG)^exp hairpins; (ii) release of the sequestered MBNL1 via binding to the double‐stranded form of r(CUG)^exp; and (iii) ligand‐induced degradation of the r(CUG)^exp. Yet, the relative contribution of each of these processes to the observed phenotypical effect is difficult to estimate.

2.1.3. On‐Target Self‐Assembly of Triaminotriazine Ligands

Template‐assisted self‐assembly of drug molecules from small fragments is a promising strategy in drug discovery, where a biological target (such as protein, DNA or RNA) drives the selection or the synthesis of its ligands from a pool of reactive fragments, allowing a straightforward identification of high‐affinity ligands among a large number of potential candidates [101, 102]. This strategy may be implemented via two distinct methodologies. In the kinetically controlled target‐guided synthesis (KTGS) approach, the biomolecular target acts as a template that accelerates an irreversible reaction between the fragments by binding them at the optimal distance and stabilizing the pre‐reactive ternary complex [103, 104]. Conversely, in the dynamic combinatorial chemistry (DCC) approach, the biomolecular target binds the strongest ligands from the dynamic covalent libraries (DCLs) of products formed through a reversible reaction between the fragments and thereby shifts the equilibria within DCLs in favor of strong ligands [105, 106, 107].

KTGS approaches most often exploit the in situ, copper‐free 1,3‐dipolar alkyne–azide cycloaddition (‘click reaction’) which produces both 1,4‐ and 1,5‐triazoles, in contrast to the copper‐catalyzed reaction yielding exclusively 1,4‐triazole products. The Zimmerman lab explored the application of the in situ click reaction to d(CTG)^exp and r(CUG)^exp targets [108]. A one‐pot reaction of a mixture of 23 azides, 4 alkynes, and 6 amines (including several derivatives of triaminotriazine‐based ligands described above) was monitored by mass spectrometry in the absence of a template, as well as in the presence of d(CTG)₁₆, r(CUG)₁₆, or non‐cognate DNA and RNA targets (Figures 9A and 10A). Despite a large number of possible combinations, only a few products, all derived from the alkyne Z‐I, were detected when the reaction was performed in the presence of d(CTG)₁₆ or r(CUG)₁₆, with the dimer D‐III formed through a reaction of Z‐I with the azide Z‐II (Figure 9B) representing one of the most prominent peaks. Indeed, molecular modeling suggests that binding of monomers Z‐I and Z‐II to the contiguous U·U mispairs in r(CUG)^exp brings the alkyne and the azide groups into proximity, facilitating their reaction (Figure 9C). The dimer D‐III was extemporarily resynthesized, and its binding to d(CTG)₁₆ and r(CUG)₁₆ was investigated in detail. Similar to the prototype Z1, both the monomer Z‐I and the dimer D‐III preferentially bind d(CTG)₁₆ over its RNA counterpart; remarkably, the affinity of the dimer D‐III to d(CTG)₁₆ (K _d = 23 nM) and r(CTG)₁₆ (K _d = 230 nM) is ∼21‐fold and ∼15‐fold higher with respect to the monomer Z‐I. In line with its high affinity to d(CTG)₁₆, dimer D‐III inhibits the transcription of the d(CTG·CAG) repeat at 1–5 μM concentration, while neither monomer shows significant inhibition at ≤5 μM (Figure 9D). Despite its preferential binding to DNA repeats, dimer D‐III also efficiently inhibits the r(CUG)^exp–MBNL1 interaction in vitro (IC₅₀ = 180 nM), revealing a more than two‐orders‐of‐magnitude increase in potency with respect to the monomer Z‐I (Figure 9D).

FIGURE 9.

(A) Principle of the template‐assisted selection of r(CUG)₁₆ ligands with MALDI‐TOF detection. (B) Structures of the monomers Z‐I and Z‐II and the dimer D‐III. (C) Molecular model of Z‐I and Z‐II bound to an r(CUG) fragment and reacting with the formation of D‐III. (D) Inhibition of in vitro transcription of d(CTG·CAG)₇₄ by Z‐I, Z‐II, and the dimer D‐III. (E) Inhibition of the r(CUG)₁₂–MBNL1 complex by Z‐I and D‐III in EMSA. The K _d values in panel B refer to apparent dissociation constants from fluorimetric titrations. Panels A and C–E reproduced with permission [108]. Copyright 2020, American Chemical Society.

FIGURE 10.

Comparison of one‐pot and parallel screening strategies for kinetic template‐guided self‐assembly of ligands. (A) In the one‐pot screening, all reactive fragments are incubated with the template, and the dimers formed in the presence of the template are detected by MS and/or HPLC. (B) In the parallel screening approach, the reactive fragments are incubated pairwise with the template, and MS and/or HPLC analysis detects the formation of dimeric products.

While the one‐pot selection method allowed the identification of a novel, potent ligand of d(CTG)^exp and r(CUG)^exp, the outcome was limited in terms of structural diversity, as all identified dimers contained the same triaminotriazine–acridine module of Z‐I. Indeed, the product D‐III is very similar to the rationally designed dimers 90.7 and 90.9 (cf. Figure 7A). To enhance the structural diversity of dimeric and trimeric ligands, Zimmerman and coworkers implemented the parallel template‐guided synthesis, where the building blocks were combined pairwise in the presence or in the absence of templates (Figure 10B) [109], allowing the detection of high‐affinity hits obtained from less competitive fragments. From the 32 parallel reactions performed in the presence of d(CTG)₁₆, 19 resulted in the formation of di‐ and/or trimeric products detected by mass spectrometry and HPLC analysis, with the azide Z‐II giving the highest signals of products upon reaction with alkynes 109.1, 109.3, and 109.8 (Figure 11A). The corresponding dimers D(1+II), D(3+II), and D(8+II) were studied with respect to their capacity to inhibit the synthesis of r(CUG)₉₀ and r(CAG)₉₀ upon bidirectional in vitro transcription of the d(CTG·CAG)₉₀ template (Figure 11B). All three dimers inhibited the synthesis of r(CAG)₉₀ and, with a ∼2‐fold lower efficiency, r(CUG)₉₀ transcripts, with IC₅₀ values in the low‐micromolar range (Figure 11C); unfortunately, the activity of the related dimer D‐III in this assay has not been evaluated. The preferential inhibition of r(CAG)₉₀ synthesis has been attributed to the strong binding of ligands to the hairpin form of d(CTG)^exp in the template DNA strand, interfering with the progression of RNA polymerase. However, it also inhibits the transcription of the opposite d(CAG) strand, leading to reduced synthesis of the toxic r(CUG)^exp.

FIGURE 11.

(A) Dimeric self‐assembled d(CTG)₁₆ ligands identified in parallel screening experiments. (B) Principle of the bidirectional in vitro transcription assay and (C) IC₅₀ values for inhibition of transcription of r(CUG)₉₀ and r(CAG)₉₀ by dimeric ligands [109].

In parallel to the KTGS methodology described above, the Zimmerman group also explored the DCC approach to on‐target self‐assembly of triaminotriazine ligands. Toward this end, the structure of the groove‐binding ligand 94.3 was modified to include benzaldehyde and aniline groups, to give compounds 110.2 and 110.3 (Figure 12A) that can engage in a reversible reaction producing the imine dimer 110.(2 + 3) and higher oligomers [110]. Molecular modeling suggested that upon binding of both monomers to r(CUG)^exp, the aldehyde group in 110.2 and the aniline nitrogen in 110.3 would be located at a short distance, suitable for their reaction (Figure 12B). Indeed, incubation of a mixture of both monomers in the presence of d(CTG)₁₆ or r(CTG)₁₆ resulted, in as few as 2 min, in the formation of the dimer 110.(2 + 3) and well as higher oligomers, detected by mass spectrometry and HPLC analysis following the reduction of unstable imines into stable amine analogs; notably, no products were detected in the absence of the templates. The capacity of self‐assembled imine products to bind to d(CTG)₁₆ was assessed in an in vitro transcription assay with a (CTG·CAG)₉₀ template. An equimolar mixture of 110.2 and 110.3 inhibited the synthesis of r(CUG)₉₀ with an IC₅₀ value of 20.8 μM, which was further reduced to 9.8 μM if the mixture was pre‐incubated with the DNA template for 24 h prior to the addition of RNA polymerase, to favor the template‐guided formation of the imine. In contrast, a nonreacting mixture of 110.2 and 110.3 would inhibit RNA polymerase with a theoretical IC₅₀ value of 49 μM, calculated by taking into account the inhibitory activity of each component. Treatment of DM1‐modeling HeLa cells with an equimolar combination of 110.2 and 110.3 (100 μM each) resulted in a 65% rescue of the INSR gene missplicing, while the individual monomers were not active at this concentration. This indicates that monomers 110.2 and 110.3 reach d(CTG)^expand/or r(CUG)^exp targets in cells and react on‐target with the formation of dimeric and/or oligomeric products, more potent than individual monomers. Of note, the observed gain in potency is modest compared with the extemporarily synthesized dimers such as 97.2a and 97.2b (cf. Figure 8D), which may be due to the poor yield of the imine dimer upon reversible reaction of monomers, as well as to the differences in cellular uptake of monomers, resulting in their suboptimal on‐target ratio. However, the major advantage of this approach resides in the use of highly cell‐permeable and essentially nontoxic monomers. In addition, the reversibility of imine formation may favor the hydrolysis of the off‐target self‐assembled products, thereby enabling a “recycling” of reactive monomers; in contrast, target‐bound imine products should be, at least partially, protected from hydrolysis (Figure 12C). Thus, this approach could avoid long‐term toxicity issues related to off‐target effects of ligands and lead to an attractive therapeutic strategy.

FIGURE 12.

(A) Triaminotriazine derivatives 110.2 and 110.3 used for template‐guided oligomerization. (B) Molecular model of ligands 110.2 and 110.3 bound to r(CUG)^exp before (left) and after formation of the imine 110.(2 + 3) (right). (C) Principle of the in‐cell, template‐assisted formation of oligomeric d(CUG)^exp and/or d(CTG)^exp ligands from reactive monomers. Panel B reproduced with permission [110]. Copyright 2022, Wiley‐VCH GmbH.

To increase the molecular diversity of ligands obtained via the DCC approach, the Zimmerman group explored a large combinatorial library of imines, generated in a one‐pot reaction of 8 aldehydes and 26 amines including multiple triaminotriazine fragments as well as other nucleobase mimics and nonspecific DNA and RNA binders [111]. Mass‐spectrometric analysis of the library incubated in the presence of d(CTG)₁₆, r(CUG)₁₆ and other nucleic acid templates, followed by reduction of imines into stable amine analogs prior to analysis, resulted in the identification of 332 (15%) of products out of 2173 possible dimeric and trimeric combinations. Of these products, 64 were formed only in the presence of one of the templates, suggesting that the self‐assembly of these ligands was not random but rather template‐guided. To confirm the hits identified by mass spectrometry, several aldehyde/amine combinations, as equimolecular two‐ or three‐component mixtures, allowing for in situ self‐assembly of imine products, were assessed with regard to their capacity to inhibit in vitro transcription of the d(CTG·CAG)₉₀ template. Several of these combinations (Figure 13) inhibited the transcription of the d(CTG·CAG)₉₀ template with IC₅₀ values in the range of 25–44 μM, whereas the respective individual components were inactive at these concentrations. Intriguingly, the combination A5/N16, which showed the most potent inhibitory effect (IC₅₀ = 25.9 μM), was identified by mass spectrometry in the “blank” library incubated in the absence of templates, but not in the presence of nucleic acid targets. This anomaly points out the possible bias of mass‐spectrometric screening when applied to large “one‐pot” libraries. Nonetheless, this study highlights the tremendous potential of the in situ template‐guided self‐assembly of efficient d(CTG)^exp and r(CUG)^exp ligands from smaller monomeric fragments.

FIGURE 13.

Structures of selected hits identified upon one‐pot template‐guided screening of the imine library. The structures of the corresponding aldehyde (A) and amine (N) building blocks are framed; the imine bonds are shown in red. The IC₅₀ values refer to the inhibition of in vitro transcription of the d(CTG·CAG)₉₀ template by equimolar mixtures of the respective components [111].

2.2. Monomeric and Dimeric Base‐Recognizing Ligands

To rationally design ligands selectively targeting d(CTG)^exp and/or r(CUG)^exp repeats, several teams explored heterocyclic derivatives with a hydrogen‐bonding pattern D‐A‐D (where D is a hydrogen‐bond donor and A is a hydrogen‐bond acceptor group), complementary to that of thymine or uracil residues (i.e., A‐D‐A). Of note, this hydrogen‐bonding pattern is more challenging in terms of molecular recognition than classical Watson–Crick pairings because of multiple unfavorable, repulsive interactions [112]. Along these lines, Nakatani and coworkers developed a 2,9‐diaminoalkyl‐1,10‐phenanthroline derivative DAP (Figure 14A) [113, 115]. A molecular model (Figure 14B) suggests that the phenanthroline unit of DAP interacts with one of the uracil residues of a U·U mispair via hydrogen bonding, with the second uracil being flipped out of the duplex; in addition, the phenanthroline core provides strong stacking interactions with the flanking G≡C base pairs. According to SPR experiments, DAP interacts with r(CUG)₉ and r(CCG)₉, as well as their deoxyribonucleotide analogs (i.e., d(CTG)₉ and d(CCG)₉), but not with r(CAG)₉, r(CGG)₉, or a fully matched RNA duplex. Likewise, in UV‐melting experiments, DAP preferentially stabilized r(CUG)₉ and r(CCG)₉ hairpins (ΔT _m = 7.8°C and 5.3°C, respectively, at 20 μM DAP). Native ESI mass spectrometry confirmed that DAP binds to r(CUG)₉ with a 4:1 stoichiometry, consistent with the number of available U·U mispairs. The interaction of DAP with the DNA counterpart, d(CTG)₆, was evidenced by ITC (K _d = 23 μM) and FRET‐melting assays (ΔT _m = 6°C with 2 equiv. DAP). DAP inhibited mRNA translation downstream of the r(CUG)₂₀ sequence, pointing to an interaction with CUG repeats in RNA, but also was shown to induce stalling of DNA polymerase at CTG repeats in primer extension experiments, highlighting its multi‐targeting ability [44, 113].

FIGURE 14.

Base‐recognizing CUG ligands. (A) Structures of monomeric ligands DAP and PQA‐19 (along with its putative interaction with a thymine residue, highlighted in gray). (B) Molecular model of DAP (green) bound to a U·U mispair in r(CUG)^exp. Reproduced with permission [113]. Copyright 2016, Wiley‐VCH & Co. KGaA. (C) Structure of the dimeric phenanthroline DDAP and its proposed binding mode to r(CUG)^exp. (D,E) Chemical structures of JM642 (D), 119.1 –3 (along with its putative interaction with uracil residues, highlighted in gray) and 114.1a (E). (F) “Janus‐wedge” interaction of one of the pyrido[2,3‐d]pyrimidin‐7‐(8H)‐one moieties in 114.1a (blue) with two uracil residues of a U·U mispair. (G) Molecular model of 114.1a bound to r(CUG)^exp according to MD simulations. (H) Structure of heterodimeric ligand iQN and its putative binding mode to d(CTG) sites in DNA. Panels F−G reproduced with permission [114]. Copyright 2021, R. Ondono et al.

The dimeric derivative DDAP, in which two DAP moieties are connected through a long, flexible linker, was designed to improve the affinity to the RNA target via simultaneous binding to two U·U mispairs (Figure 14C) [115, 116]. DDAP inhibits the r(CUG)₂₀–MBNL1 interaction much more efficiently than DAP (K _I = 0.26 and 30 μM, respectively, EMSA). In DM1 mouse myoblasts, DDAP was not cytotoxic at concentrations of up to 40 μM, in contrast to DAP which was strongly toxic at > 10 μM. Furthermore, at a concentration of 40 μM, DDAP dramatically reduced the number of nuclear r(CUG)^exp foci and rescued the MBNL1‐dependent missplicing of the Atp2a1 gene almost to the normal level. In a murine model DDAP, administered at 20 or 50 mg/kg via daily intraperitoneal (i.p.) injections, partially rescued the missplicing of Clcn1 and Atp2a1 genes; however, full recovery could not be achieved due to toxicity at higher doses.

Another series of monomeric ligands designed by Nakatani and coworkers is based on the 1H‐pyrrolo[3,2‐h]quinoline‐8‐amine (PQA) scaffold, whose hydrogen‐bonding pattern is also complementary to thymine or uracil residues (Figure 14A) [117]. The leading candidate of this series, PQA‐19, selectively binds to d(CTG)₉, and, to a lesser extent, d(CCG)₉ repeats (K _d = 20 and 33 μM, respectively, SPR), without detectable binding to d(CAG)₉ and d(CGG)₉ counterparts, and does not stabilize fully matched DNA duplexes. The binding of this compound to r(CUG) repeats or its activity in DM1 models have not been assessed.

More recently, the same team reported a dimeric derivative, JM642 (Figure 14D), in which two pyridyl‐substituted 1,3‐diaminoisoquinoline moieties act as uracil‐recognizing units, enabling binding to two neighboring U·U sites [115, 118]. SPR studies confirmed strong binding of JM642 to r(CUG)₉ in the nanomolar concentration range and selectivity with respect to the r(CCG)₉ counterpart. Treatment of DM1 patient‐derived myoblasts with 30 μM JM642 strongly reduced the number of cells presenting nuclear r(CUG)^exp foci. In line, in a murine cell model, JM642 (albeit at a relatively high concentration of 80 μM) restored the Mbnl1‐dependent missplicing of the Ldb3 (LIM domain binding 3) gene almost to the normal level, whereas the monomeric analog (not shown) had only modest activity. Finally, in the transgenic HSA^LR murine model of DM1 expressing 220 CUG repeats in the 3′‐UTR of the human skeletal actin (HSA) transgene, JM642 administered at 10−20 mg/kg partially rescued the missplicing of Clcn1 and Atp2a1 genes, without inducing toxic effects.

A related dimeric ligand 119.1‐ 3 (Figure 14E) containing two pyrido[2,3‐d]pyrimidin‐7‐(8H)‐one fragments, each having a hydrogen‐bonding pattern complementary to that of uracil, and connected by a p‐xylylene linker, was identified by Estrada–Tejedor and coworkers through an extensive in silico approach followed by experimental validation [119]. Ligand 119.1‐3 binds to r(CUG)₂₃ in the concentration range of 125−250 μM (as per fluorescence anisotropy measurements). Its effect on the r(CUG)^exp−MBNL1 interaction has not been studied in vitro; however, treatment of human DM1 myoblasts with 100 μM 119.1–3 led to a partial release of MBNL1 from ribonuclear foci into the cytoplasm and nucleoplasm, without affecting the average number of the foci. In a Drosophila DM1 model recapitulating muscular atrophy and impaired locomotion, 119.1‐3 significantly alleviated locomotion defects. The pyrido[2,3‐d]pyrimidin‐7‐(8H)‐one motif was further exploited in the ligand 114.1a (Figure 14E), recently developed by the same team [114]. In 114.1a, each pyrido[2,3‐d]pyrimidin‐7‐(8H)‐one moiety can establish “Janus wedge”‐type interactions with both uracil residues of every U·U mispair (Figure 14F), similar to 1,3,5‐triaminotriazine derivatives (Section 2.1). MD simulations, using an r(CUG)₁₆ model, showed that the five‐carbon linker in 114.1a was optimal for binding to the consecutive U·U mispairs (Figure 14G), while the dimers with longer linkers could reach more distant mispairs. In vitro TR‐FRET assay demonstrated that 114.1a inhibited the r(CUG)₁₂–MBNL1 interaction by 33% at a concentration of 0.1 μM, while the analogs with longer linkers were less active. Finally, in DM1 patient‐derived dermal fibroblasts, 114.1a, at a concentration of 10 μM, reduced the sequestration of MBNL1 into ribonuclear foci by 23%, without affecting the total number of these foci; however, it did not affect the MBNL1‐dependent missplicing of the ATP2A1 (SERCA1) and INSR genes, nor the expression level of DM1‐related splicing factors, such as MBNL1 or CELF1.

Beyond homodimeric scaffolds combining two identical base‐recognizing moieties, the Nakatani lab recently reported a heterodimeric ligand iQN (Figure 14H) which combines the moieties of 3‐acylaminoisoquinoline, making two hydrogen bonds with a thymine residue, and 2‐acylamino‐1,8‐naphtyridine, specifically recognizing guanine residues [120]. In thermal denaturation experiments, iQN selectively stabilized DNA duplex bearing a T·T mismatch in the 5′‐CTG‐3′/5′‐CTG‐3′ context. ITC measurements provided the K _d value of 1.9 μM and a 2:1 binding stoichiometry, in agreement with the proposed binding mode where two iQN molecules bind to a single d(CTG/CTG) site by making hydrogen bonds not only with mispaired thymines, but also with the neighboring guanines, leading to flipping‐out of the complementary cytosines (Figure 14H). Despite its promising affinity and unusual binding mode, the effects of this ligand in more relevant DM1 models have not yet been investigated.

2.3. Benzimidazole‐Based Ligands

2.3.1. Monomeric and Multimeric Bis‐Benzimidazoles

Bis‐benzimidazole Hoechst 33258 (Figure 15A) is a well‐known, widely used, cell‐permeable DNA stain that binds in the minor groove of duplex DNA, with a marked preference to AT‐rich regions [122, 123]. In 2000, Cho and Rando studied the interaction of Hoechst 33258 with several RNA constructs containing 1 × 1 internal loops (i.e., base mispairs) in the stem part and observed strong binding to 5′‐CUG‐3′/5′‐CUG‐3′ and 5′‐CCG‐3′/5′‐CCG‐3′ sites (K _d = 57 and 60 nM, respectively, from competitive fluorescence anisotropy titrations). In contrast, no binding to the RNA hairpin devoid of mispairs in the stem part was observed, suggesting that Hoechst 33258 and its analogs do not interact with double‐stranded RNA [124].

FIGURE 15.

Bis‐benzimidazole derivatives developed as r(CUG)^exp ligands: (A) monomeric bis‐benzimidazoles; (B) first‐generation multimeric ligands. (C) Molecular model of a single bis‐benzimidazole moiety (H) bound to a 5′‐CUG‐3′/5′‐CUG‐3′ site in an RNA duplex according to MD simulations. (D) Molecular models of the dimeric ligand 2H‐4 bound to an RNA duplex presenting two 5′‐CUG‐3′/5′‐CUG‐3′ sites. Panels C–D reproduced with permission [121]. Copyright 2020, Wiley‐VCH GmbH.

Building on these results, Disney and coworkers developed a series of modular, multimeric ligands, in which bis‐benzimidazole residues H (differing from Hoechst 33258 only by the location of the substituent in the phenyl ring, Figure 15A) are linked to the peptoid backbone via triazole groups and separated by propylamine spacers [125]. These multimers follow the general formula (a + 1)H‐b (Figure 15B), where (a + 1) is the total number of bis‐benzimidazole groups and b the number of spacers between the adjacent H residues. The monomeric bis‐benzimidazole H‐N₃ (124.1) binds to RNA hairpin with a single 5′‐CUG‐3′/5′‐CUG‐3′ site with an affinity comparable to that of Hoechst 33258 (K _d = 130 nM) and a 13‐fold selectivity over a fully paired RNA duplex, but only ∼2‐fold over a DNA duplex. Its affinity to r(CUG)₁₀₉ (K _d = 150 nM) is close to that observed for a single 5′‐CUG‐3′/5′‐CUG‐3′ site, suggesting noncooperative binding; in addition, the binding stoichiometry (54 ligands per r(CUG)₁₀₉) indicates that the ligand can occupy every 5′‐CUG‐3′/5′‐CUG‐3′ site. Multimerization of H units drastically increases the r(CUG)^exp affinity and selectivity of the resulting ligands. Thus, the K _d values for r(CUG)₁₀₉ binding decrease from 150 nM for H‐N₃ and 100 nM for the dimer 2H‐4 down to 13 nM for the pentamer 5H‐4 (Table 4). In contrast, binding to double‐stranded DNA weakens upon multimerization (e.g., K _d = 700 nM for 5H‐4), leading to a dramatic increase of the r(CUG)₁₀₉‐vs.‐dsDNA selectivity from 0.73 (for H‐N3) to 54‐fold (for 5H‐4) and suggesting that such multimeric ligands operate at the RNA level. The binding stoichiometry values (Table 4) suggest that each bis‐benzimidazole residue of a multimeric ligand interacts with a single 5′‐CUG‐3′/5′‐CUG‐3′ site. The impact of the number of propylamine spacers separating the adjacent H residues (b) was systematically studied; it was found that four to five spacers are optimal for binding to CUG repeats, whereas shorter (e.g., as in 2H‐2) or longer spacers (e.g., as in 2H‐6) result in ligands with reduced affinity [127].

TABLE 4.

Dissociation constants (K _d) and binding stoichiometry (N) for the interaction of bisbenzimidazole derivatives with r(CUG) and IC₅₀ values for inhibition of the r(CUG) _n –MBNL1 interaction in vitro.

Ligand	Number of ligand modules (a + 1)	Number of spacers (b)	K d (CUG) (N)	IC50[r(CUG) n –MBNL1]
Hoechst 33258 [124]	1	—	57 nMa	110 μMb
H‐N3 (124.2) [125]	1	—	130 nMa
			150 nM (54)c
			2100 nMd [126]
First‐generation dimers
2H‐2 [127]	2	2	4000 nM (2.3)e	22 μMb
2H‐4 [125, 127, 128]	2	4	100 nM (18)c	11 μMb
			90 nM (1.2)e	71 μMf
			70 nMg [128]	32.2 μMh [121]
2H‐6 [127]	2	6	140 nM (1.1)e	>30 μMb
3H‐4 [125]	3	4	65 nM (16)c	0.96 μM,b 0.41 μMi
4H‐4 [125]	4	4	35 nM (11)c	0.39 μM,b 0.21 μMi
5H‐4 [125]	5	4	13 nM (8)c	0.22 μM,b 0.077 μMi
			25 nM (1.9)j
Second‐generation dimers [126]
2H‐SPM	2	—	—	3 μMh
2H‐3G	2	3	—	4 μMh
2H‐4G	2	4	—	4 μMh
2H‐6G	2	6	—	30 μMh
2H‐3NPr	2	3	—	∼5 μMh
2H‐K4NPr	2	4	—	∼30 μMh
2H‐K4NH	2	4	—	∼35 μMh
2H‐K4NMe	2	4	13 nMd	∼20 μMh
2H‐K4NiBi	2	4	—	∼55 μMh
2H‐K4	2	4	—	∼30 μMh
2H‐K4NMeS [129]	2	4	12 nM (23)k	∼14 μMh
			280 nM (3)l [130]
Third‐generation dimers [130]
2H‐K4‐D‐Ala	2	4	—	∼11 μMh
2H‐K4‐Pro	2	4	—	∼9 μMh
2H‐K2‐Pro	2	2	150 nMl	5.3 μMh
Other dimers
131.22 [131]	2	4	106 nMl	2.8 μMh
2H‐2C2 [121]	2	2	81 nMl	3 μMh
Functionalized derivatives
2H‐4‐CA [132]	2	4	—	5 μMf
2H‐4‐HPT [128]	2	4	1500 nMg	64 nM (dark)f
				10 nM (post UV)f
2H‐K4NMeS‐Bleomycin (Cugamycin) [133]	2	4	365 nMl	—
DeglycoCugamycin [134]	2	4	610 nMl	—
2H‐K2‐Pro‐Bleo [130]	2	2	280 nMl	—
131.22‐Bleo [131]	2	4	120 nMl	—

^a (CUG)_2×1 hairpin.

^b Enzyme fragment complementation assay with immobilized r(CUG)₁₀₉.

^c r(CUG)₁₀₉.

^d SPR assay with r(CUG)₁₂.

^e (CUG)_2×2 hairpin.

^f HTRF assay with r(CUG)₁₀.

^g SPR assay with r(CUG)₁₀.

^h HTRF assay with r(CUG)₁₂.

ⁱ Assay with a 3.4‐fold lower amount of r(CUG)₁₀₉ substrate.

^j (CUG)_2×12 hairpin.

^k Fluorescence titrations with r(CUG)₁₀₉.

^l Fluorescence titrations with r(CUG)₁₂.

In line with their high affinity to r(CUG)^exp exceeding that of MBNL1, bis‐benzimidazole multimers inhibit the r(CUG)^exp–MBNL1 interaction in vitro. Multimerization dramatically increases the inhibitory activity, as illustrated by the IC₅₀ values obtained in the ELISA‐like assay with r(CUG)₁₀₉, from IC₅₀ = 110 μM for the monomer H‐N₃ to 11 μM for the dimer 2H‐4 and 0.22 μM for the pentamer 5H‐4, the most potent ligand from this series (Table 4).

Cellular uptake of multimeric ligands was studied in mouse myoblasts by fluorescence microscopy, taking advantage of the fluorescence light‐up of bis‐benzimidazoles upon RNA binding. Multimers up to 5H‐4 were all detected in the nuclei, demonstrating that their high molecular weight does not impair cell permeability, and none of the ligands was toxic at concentrations up to 5 μM [125]. In DM1‐modeling HeLa cells, a dose‐dependent rescue of MBNL1‐dependent missplicing of the TNNT2 minigene was observed upon treatment with 2H‐4, 3H‐4, and 4H‐4 (5–50 μM), but not 5H‐4, despite the latter being the most potent ligand in vitro. Notably, while the monomer H‐N₃ was inactive at concentrations up to 100 μM, treatment with 5 μM 2H‐4 restored TNNT2 splicing to the wild‐type level, demonstrating the potential of the modular assembly to design biologically active compounds from non‐bioactive units. The poor efficacy of longer oligomers, such as 4H‐4 and 5H‐4, could be due to their poor solubility in the cell culture medium. Multimers 2H‐4, 3H‐4 (at 25 μM), and, to a lesser extent, 4H‐4 (at 50 μM) decreased the number and intensity of nuclear r(CUG)^exp foci in DM1‐modeling cells. To confirm the capacity of the oligomers to disrupt the r(CUG)^exp–MBNL1 aggregates and enable cytoplasmic export of the RNA transcript, they were evaluated in murine C2C12 myoblasts stably expressing the firefly luciferase gene containing a d(CTG)₈₀₀ expansion in its 3′‐UTR. Treatment with 2H‐4 (25 μM), 3H‐4 or 4H‐4 (10 μM) induced an over 150% increase in the luciferase activity, indicating the capacity of these ligands to correct the DM1‐related translational defects, with 3H‐4 being the most potent drug [135].

The binding mode of bis‐benzimidazole ligands to r(CUG)^exp has been recently investigated by a combined dynamic docking and MD approach [121]. Molecular modeling suggests that each bis‐benzimidazole moiety (H) intercalates at a 5′‐CUG‐3′/5′‐CUG‐3′ site, leading to the opening of the U·U mispair, with phenyl and piperazine groups located in the minor and the major grooves of the RNA duplex, respectively (Figure 15C). This intercalative binding mode is surprising, considering the well‐established minor‐groove binding mode of bis‐benzimidazoles with double‐stranded DNA [122, 123], but explains the capacity of dimeric bis‐benzimidazole ligands such as 2H‐4 to interact with each 5′‐CUG‐3′/5′‐CUG‐3′ site of r(CUG)^exp. The peptoid linker provides additional stabilization of the complex via interactions with the minor groove of the RNA duplex (Figure 15D).

Considering the promising biological activity of the dimer 2H‐4, the Disney team invested massive efforts into the scaffold optimization of dimeric ligands. Several molecular scaffolds were explored, including polyamines, α‐ and β‐peptides, peptoids and peptide‐tertiary amides, all bearing two identical bis‐benzimidazole modules H each (Figure 16A) [126]. The dimers were initially screened with respect to their capacity to disrupt the r(CUG)₁₂–MBNL1 interaction in the HTRF assay, in comparison with the prototype 2H‐4 (Table 4). The polyamine 2H‐SPM (IC₅₀ = 3 μM) and α‐peptides 2H‐3G and 2H‐4G (IC₅₀ = 4 μM) were the most potent; an increase of the number of spacer units led to a loss of activity (e.g., 2H‐6G, IC₅₀ = 30 μM). 2H‐3NPr (IC₅₀ ≈ 15 μM) and 2H‐K4NPr (IC₅₀ ≈ 30 μM) were most active among the peptide‐tertiary amide derivatives. The best compounds from each class were subsequently assessed with respect to their capacity to correct the missplicing of the TNNT2 minigene in DM1‐modeling HeLa cells. Despite its moderate in vitro activity, 2H‐K4NPr was the most active in the cellular assay, surpassing the first‐generation dimer 2H‐4 at a concentration of 10 μM. In the second round of optimization, the structure of 2H‐K4NPr was systematically modified by changing the nature of the substituent at the amide nitrogen (2H‐K4NH, 2H‐K4NMe, 2H‐K4iBu) and at the α‐carbon atom (2H‐K4, Figure 16A). The N‐methyl derivative 2H‐K4NMe was the most active in the HTRF assay (IC₅₀ ≈ 20 μM, Table 4), suggesting that bulky alkyl substituents (e.g., isobutyl) hinder the tight binding of ligands to r(CUG)^exp. Following the assessment in the HeLa cell DM1 model, the dimer 2H‐K4NMe was selected as the optimal ligand with regard to its cellular activity (marginally improved with respect to 2H‐K4NPr) as well as stability, toxicity, and synthetic accessibility [126].

FIGURE 16.

(A) Second‐generation and (B) third‐generation dimeric bis‐benzimidazole ligands designed by the Disney lab. The structure of the H unit is shown in Figure 15A.

To confirm the cellular target of the lead 2H‐K4NMe, the biotinylated derivative 2H‐K4NMe‐Biotin (Figure 16A) was used in pull‐down experiments with total RNA extracted from DM1‐modeling HeLa cells, demonstrating high enrichment in r(CUG)^exp [126]. In the murine HSA^LR DM1 model, 2H‐K4NMe, administered at 100 mg/kg/day for 7 days, partially yet statistically significantly rescued the missplicing of the Clcn1 and Atp2a1 genes. Further improvement of this scaffold concerned its metabolic stability, resulting in the compound 2H‐K4NMeS with minor differences in one of the linkers connecting the bis‐benzimidazole moiety (highlighted in Figure 16A) [129]. The optimized lead 2H‐K4NMeS was found to bind r(CUG)₁₀₉ with a K _d value of 12 nM (fluorimetric titrations), without binding to double‐stranded DNA or competitor RNA. In cellular assays, 2H‐K4NMeS improved MBNL1‐dependent missplicing of the MBNL1 pre‐mRNA and other genes at a concentration as low as 100 nM and reduced the number of r(CUG)^exp–MBNL1 foci by half at a concentration of 1 μM. This optimized lead was subsequently exploited in the design of functionalized derivatives 2H‐K4NMeS‐CA‐Biotin and Cugamycin (Section 2.3.2) and for the self‐assembly of bis‐benzimidazole ligands on the r(CUG)^exp template (Section 2.3.3).

Further research led to additional optimization of the linker domain connecting the two bis‐benzimidazole moieties in dimeric ligands. In the third generation of dimers, linkers containing four d‐alanine, tyrosine, hydroxyproline, or two to five proline spacers have been exploited (Figure 16B) [130]. Among these, 2H‐K4‐D‐Ala, 2H‐K4‐Pro, and 2H‐K2‐Pro strongly inhibited the r(CUG)^exp–MBNL1 interaction in vitro (Table 4). In particular, 2H‐K2‐Pro, with only two proline spacers in the linker, represents about 3‐fold improvement over 2H‐K4NMeS. This improvement was also reflected in the high affinity of 2H‐K2‐Pro to r(CUG)₁₂ (K _d = 150 nM, fluorimetric titration). Despite improved in vitro properties, the biological activity of 2H‐K2‐Pro in DM1 patient‐derived fibroblasts was similar to that of 2H‐K4NMeS in terms of splicing rescue and the effect on the number of r(CUG)^exp–MBNL1 foci. The lack of improvement at the cellular level has been attributed to the changes in subcellular distribution, resulting in accumulation of 2H‐K2‐Pro both in the nucleus and the cytoplasm, in contrast to the predominately nuclear accumulation of 2H‐K4NMeS.

Considering the potential of linker optimization for delivering better r(CUG)^exp ligands, Disney and coworkers exploited the massively parallel, one‐bead‐one‐compound (OBOC) library approach [131]. An OBOC library of dimeric bis‐benzimidazoles was designed to include four variable N‐propylglycine residues (R¹ to R⁴, Figure 17A) among the 24 building blocks, resulting in 331,776 unique bead‐bound compounds. This library was initially screened using a competitive binding assay, where the bead‐bound compounds were incubated with biotinylated r(CUG)₁₂ in the presence of 2H‐4 as a competitor, to suppress the weakest binders. After magnetic pull‐down and MS/MS decoding, 142 hits were identified (Figure 17B), of which 32 top‐scoring ones were selected for preparative synthesis. Further screening relied on direct assessment of the biological activity of the candidates in DM1 patient‐derived myotubes. Twelve dimers demonstrated the desired nuclear localization as per fluorescence microscopy and were further evaluated for their capacity to improve the MBNL1 missplicing, resulting in a selection of four active compounds. Among these, the dimer 131.22 (Figure 17C) was the most potent and studied in detail. In vitro assays provided a K _d value of 106 nM for binding to r(CUG)₁₂ and an IC₅₀ of 2.8 μM for inhibition of the r(CUG)₁₂–MBNL1 complex. MD simulations suggested that 131.22 binds to r(CUG)^exp with its peptoid linker in the minor groove of RNA and bis‐benzimidazole units penetrating into the open U·U mispairs (Figure 17D); additionally, the side groups of the linker establish several electrostatic and hydrogen‐bonding interactions with the RNA backbone. Given the potential of 131.22, it was further exploited to design an RNA‐cleaving bleomycin conjugate.

FIGURE 17.

Selection of optimized dimeric bis‐benzimidazole r(CUG)^exp ligands from one‐bead‐one‐compound (OBOC) library [131]. (A) Primary screening workflow. (B) Generic structure of the members of the OBOC library. (C) Structure of the lead compound, 131.22. (D) Molecular model of 131.22 bound to an RNA presenting two 5′‐CUG‐3′/5′‐CUG‐3′ sites. The parts of the ligand are colored as in panel C. Panel D reproduced with permission [131]. Copyright 2021, American Chemical Society.

Another optimization strategy relies on macrocyclization of the linker domain of dimeric ligands, resulting in more rigid, pre‐organized scaffolds that enable higher affinity and selectivity for the target. Among the seven peptoid macrocycles differing by the number of N‐propylglycine spacers between two bis‐benzimidazole units, the smallest, 2H‐2C2 (Figure 18A), was the most potent in terms of inhibition of the r(CUG)₁₂–MBNL1 interaction (IC₅₀ = 3 μM, HTRF), showing a ∼10‐fold improvement over the non‐macrocyclic prototype 2H‐4 [121]. Macrocycle 2H‐2C2 was also characterized by strong binding to r(CUG)₁₂ (K _d = 81 nM), and MD simulations revealed that its complex with an r(CUG)^exp model (Figure 18B) was energetically more favorable by about 5.6 kcal mol^–1, comparing with 2H‐4. Furthermore, 2H‐2C2 showed improved cellular uptake in DM1 patient‐derived myotubes, resulting in a 10‐fold higher intracellular concentration than 2H‐4, albeit with a loss of the exclusively nuclear subcellular localization. Treatment with 2H‐2C2 reduced the number of RNA foci and improved MBNL1‐dependent splicing of the MBNL1 (by 25%) and NCOR2 genes in DM1 myotubes, with about 10‐fold higher potency than 2H‐4. This improvement underscores the potential of macrocyclic scaffolds for the design of r(CUG)^exp ligands.

FIGURE 18.

(A) Structure of the macrocyclic bis‐benzimidazole dimer 2H‐2C2. The structure of the H unit is shown in Figure 15A. (B) Molecular model of 2H‐2C2 bound to RNA duplex presenting two 5′‐CUG‐3′/5′‐CUG‐3′ sites. Panel B reproduced with permission [121]. Copyright 2020, Wiley‐VCH GmbH.

The variations of the bis‐benzimidazole unit as the RNA‐binding motif have also been explored. Symmetric bis‐benzimidazole H1 (Figure 19A) with a tail‐to‐tail arrangement of benzimidazole units (in contrast to the head‐to‐tail arrangement in Hoechst 33258 and its derivatives described above) was identified by Disney and coworkers upon virtual screening of chemical libraries for similarity to Hoechst 33258 and pentamidine (cf. Section 2.4.1) [137]. In vitro validation confirmed the capacity of H1 to inhibit the r(CUG)₁₂–MBNL1 interaction with IC₅₀ = 50 μM (HTRF assay), making it significantly more active than Hoechst 33258 (IC₅₀ > 1 mM). The selectivity, assessed by competition dialysis, revealed the preferential binding of H1 to the 5′‐CUG‐3′/5′‐CUG‐3′ motif over other RNA structures, and the binding was correlated with the number of CUG repeats, rendering this compound selective for long r(CUG)^exp. In the HeLa cell DM1 model, H1, albeit at high concentrations (500 μM to 2 mM), decreased the number of r(CUG)^exp foci and improved the splicing of the TNNT2 minigene, whereas Hoechst 33258 was inactive. In the HSA^LR mouse model, H1, at doses of 80–100 mg/kg (i.p. injections), partially rescued the missplicing of the Clcn1 and Serca1 genes.

FIGURE 19.

(A) Symmetric bis‐benzimidazole derivatives studied as r(CUG)^exp ligands. (B) Molecular model of 136.2b bound to a 5′‐CUG‐3′/5′‐CUG‐3′ site in duplex RNA. Panel B reproduced with permission [136]. Copyright 2020, Elsevier Ltd.

To enhance the RNA‐binding properties of bis‐benzimidazole H1, Disney and coworkers developed several guanidinylated derivatives of this scaffold, of which the bis‐guanidine 136.2b (Figure 19A) was the most active in terms of r(CUG)₁₂–MBNL1 inhibition in vitro (IC₅₀ = 4.6 μM, HTRF) [136]. Binding of 136.2b to r(CUG)₁₂ was characterized by a K _d value of 40 nM, with no binding to other RNA motifs as per microscale thermophoresis (MST). Dynamic docking simulations provided a molecular model of 136.2b binding to 5′‐CUG‐3′/5′‐CUG‐3′ sites, in which intercalation of the ligand induces an opening of the U·U mispair, leading to π‐stacking of at least one of the mispaired uracils with the phenyl group of the ligand. Stacking of benzimidazole groups of the ligand with the adjacent G≡C base pairs and hydrogen‐bonding interactions of the guanidine groups with the RNA backbone additionally stabilize the complex (Figure 19B). In DM1 patient‐derived myotubes, 136.2b improved the MBNL1‐dependent splicing of the MBNL1 gene and reduced the number of r(CUG)^exp–MBNL1 foci at concentrations of 1–10 μM, without affecting the level of the DMPK transcript. Target engagement by 136.2b was demonstrated in DM1 fibroblasts via a competitive chemical cross‐linking and isolation by pull‐down (C‐Chem‐CLIP) approach using the RNA‐crosslinking ligand 2H‐K4NMeS‐CA‐Biotin (Figure 20B, cf. Section 2.3.2 below). In this experiment, cells treated with increasing concentrations of 136.2b yielded a lower amount of r(CUG)^exp pulled‐down by 2H‐K4NMeS‐CA‐Biotin than untreated cells (IC₅₀ = 38 nM), indicating that 136.2b effectively competes with the crosslinker for binding to r(CUG)^exp.

FIGURE 20.

r(CUG)^exp RNA degraders and cross‐linking agents based on (A) first‐generation and (B) second‐ and third‐generation bis‐benzimidazole ligands.

2.3.2. “RNA Degraders” Based on Bis‐Benzimidazole Ligands

The dimeric bis‐benzimidazole scaffold has been exploited by Disney and coworkers in several strategies toward “RNA degraders,” that is, ligands able to cleave, or chemically damage, toxic r(CUG)^exp upon binding, and thereby prevent the repeated sequestration of MBNL1. The first approach toward such RNA degraders relies on the formation of covalent cross‐links between the ligand and r(CUG)^exp. Toward this aim, the first‐generation bis‐benzimidazole dimer 2H‐2 was conjugated with chlorambucil, a DNA‐ and RNA‐alkylating drug [132]. The resulting conjugate 2H‐2‐CA (Figure 20A) was about 14‐fold more potent than 2H‐2 in inhibiting the r(CUG)₁₀–MBNL1 interaction in vitro (Table 4), suggesting that the chlorambucil unit does not interfere with the binding to r(CUG)^exp. Gel electrophoresis experiments in combination with mass spectrometry revealed that 2H‐2‐CA forms intrastrand cross‐links with r(CUG)₁₀ at r(GpC) sites, most likely via alkylation of N7 of guanine and N3 of cytosine residues. In the HeLa cell DM1 model, treatment with nanomolar concentrations of 2H‐4‐CA restored the MBNL1‐dependent splicing of the MBNL1 pre‐mRNA with IC₅₀ = 4 nM, making it about 2500‐fold more potent than the parent compound 2H‐4 (IC₅₀ = 10 μM in the same assay) and one of the most potent compounds in terms of missplicing correction. Intriguingly, RT‐PCR analysis of RNA extracted from cells treated with 2H‐4‐CA demonstrated no degradation of r(CUG)₉₆₀‐containing DMPK mRNA, suggesting that the formation of crosslinks alone, and not degradation of this RNA, was responsible for the high activity of 2H‐4‐CA. A derivative endowed with an additional biotin handle, 2H‐4‐CA‐Biotin (Figure 20A) was further exploited for chemical cross‐linking and isolation by pull‐down (Chem‐CLIP) experiments that confirmed that the compound formed cross‐links with r(CUG)^exp in cells. Additionally, competitive (C‐Chem‐CLIP) experiments with the unreactive ligand 2H‐4 demonstrated that micromolar concentrations of 2H‐4 inhibited the formation of crosslinks between r(CUG)^exp and 2H‐4‐CA‐biotin, indicating that both ligands bind to the same target in cells. A similar chlorambucil and biotin conjugate, 2H‐K4NMeS‐CA‐Biotin (129.2, Figure 20B), was employed to validate target engagement of the second‐generation lead candidate, 2H‐K4NMeS [129]. In this case, quantification by RT‐PCR revealed that in DM1 patient‐derived fibroblasts, 2H‐K4NMeS‐CA‐Biotin enriched the pull‐down of DMPK mRNA containing 500 CUG repeats by ∼13,000‐fold, while no DMPK enrichment was observed in non‐DM1 fibroblasts. Direct binding of 2H‐K4NMeS‐CA‐Biotin to the r(CUG)^exp region in DMPK mRNA was additionally confirmed via RT‐qPCR using a set of specific primers (Chem‐CLIP‐Map approach).

Two derivatives have been explored in a strategy aiming to induce degradation of r(CUG)^exp upon light irradiation. In the first case, the dimer 2H‐4 was endowed with an N‐hydroxypyridine‐2(1H)‐thione (HPT) unit, generating a hydroxyl radical upon UV irradiation [128]. The conjugate 2H‐4‐HPT (Figure 20A), at a concentration of 10 μM, induced partial cleavage of r(CUG)₁₀ upon irradiation with 365 nm light, evidenced by gel electrophoresis. Interestingly, the parent ligand 2H‐4 also cleaved r(CUG)₁₀ upon irradiation due to the intrinsic photoreactivity of bis‐benzimidazole moieties, albeit with about two‐fold lower efficiency. The introduction of the HPT moiety reduced the affinity of 2H‐4‐HPT to r(CUG)₁₀ about 20‐fold with respect to 2H‐4 (Table 4). Nevertheless, 2H‐4‐HPT was almost as active as 2H‐4 with regard to its capacity to inhibit the r(CUG)₁₀–MBNL1 interaction in vitro (IC₅₀ = 64 μM). Most importantly, the potency of 2H‐4‐HPT was increased more than sixfold when RNA and the ligand were irradiated prior to addition of MBNL1, suggesting that ligand‐induced RNA cleavage prevents MBNL1 binding. In the HeLa cell DM1 model, irradiation of the cells treated with 5 μM 2H‐4‐HPT resulted in a statistically significant improvement of TNNT2 missplicing, while non‐irradiated cells showed no change in the splicing pattern. Furthermore, quantification of r(CUG)₉₆₀‐containing DMPK mRNA by RT‐PCR demonstrated a ∼50% decrease in the r(CUG)^exp transcript level in cells that were treated with 2H‐4‐HPT and irradiated, indicating the cleavage of r(CUG)^exp in live cells.

A drawback of 2H‐4‐HPT is related to the fact that each HPT moiety generates a single hydroxyl radical upon irradiation. To enable catalytic production of reactive oxygen species, the Disney team exploited the photoactive tris(bipyridine)ruthenium(II) complex linked to the same scaffold [138]. The conjugate 2H‐4‐Ru (Figure 20A) oxidizes the guanine residues of r(CUG)₁₀ into 8‐oxoguanine (8‐oxoG) upon irradiation with visible (420 nm) light, as demonstrated by gel electrophoresis, mass spectrometry, and ELISA assay with an anti‐8‐oxoG‐antibody. Treatment of DM1‐modeling HeLa cells with 2H‐4‐Ru followed by irradiation led to the appearance of 8‐oxoG modifications in the r(CUG)₉₆₀ transcript, detected via Northwestern blotting and immunoprecipitation. The level of the r(CUG)₉₆₀‐containing transcript was not affected, indicating that 2H‐4‐Ru induces chemical modification (i.e., oxidation) of r(CUG)^exp without complete degradation. Selective generation of 8‐oxoG resides in the r(CUG)^exp transcript was also observed in the larvae of Drosophila DM1 model, treated with 2H‐4‐Ru and irradiated.

In the third strategy toward targeted degradation of r(CUG)^exp, the Disney team harnessed the DNA‐ and RNA‐cleavage properties of Bleomycin A5, a naturally occurring antibiotic whose iron(II) complex generates reactive oxygen species inducing nucleic acid cleavage [129, 133]. The conjugate of the second‐generation lead 2H‐K4NMeS with Bleomycin A5, 2H‐K4NMeS‐Bleomycin (also termed Cugamycin, Figure 20B), binds to r(CUG)₁₂ in vitro with an apparent K _d of 365 nM (fluorimetric titration) and cleaves r(CUG)₁₀ between U and C nucleotides. Although the DNA‐cleavage activity of Cugamycin was significantly reduced in comparison to Bleomycin A5, non‐negligible DNA cleavage was observed in vitro at concentrations above 500 nM [133]. Treatment of DM1 patient‐derived fibroblasts with 250 nM Cugamycin selectively reduced the level of mutated DMPK mRNA by ∼30% without affecting other mRNAs containing shorter CUG repeats, and markedly improved MBNL1‐dependent splicing defects [129]. In DM1 patient‐derived myotubes, representing a more relevant DM1 model, treatment with 1 μM Cugamycin cleaved ∼60–80% of the expanded DMPK mRNA and rescued the splicing of MBNL1 by about 40%; at this dose, a minimal increase of DNA damage markers was observed. In the HSA^LR mouse DM1 model Cugamycin, administered at 10 mg/kg via bidiurnal i.p. injections, induced a ∼40% reduction in the expression of the r(CUG)₂₅₀‐containing HSA transgene in muscles, indicating that the drug was able to reach muscular tissues. In addition, Cugamycin induced a partial rescue of the Mbnl1 and Clcn1 missplicing, accompanied by an increase in the CLCN1 protein level in muscles and a reduction in myotonia. On the transcriptome‐wide level, treatment with Cugamycin decreased the composite missplicing score from 0.3 (in vehicle‐treated HSA^LR mice) to 0.15; conversely, no significant effect on the transcriptome‐wide splicing was observed in wild‐type mice. Administered at the same dose, compound 2H‐K4NMeS, devoid of the bleomycin fragment, failed to induce a significant change in the HSA transgene level or improve myotonia, demonstrating that the high efficacy of Cugamycin to relieve the DM1‐associated defects in vivo is due to its RNA‐cleaving activity [133].

Analogs of Cugamycin based on optimized bis‐benzimidazole derivatives, such as the smaller dimer 2H‐K2‐Pro and the ligand 131.22 identified from the OBOC library (2H‐K2‐Pro‐Bleo and 131.22‐Bleo, respectively, Figure 20B), have also been studied [130, 131]. Similar to Cugamycin, these RNA degraders cleave r(CUG)₁₀ in vitro at low‐micromolar concentrations. In DM1 myotubes, both conjugates were more active than Cugamycin with respect to reduction of the mutant DMPK level (45% reduction for 2H‐K2‐Pro‐Bleo vs. ∼30% for Cugamycin, both at 5 μM; 35% reduction for 131.22‐Bleo at just 1 μM), which has been attributed to the improved cellular uptake. Treatment with 2H‐K2‐Pro‐Bleo rescued the MBNL1 missplicing and decreased the number of r(CUG)^exp–MBNL1 foci in DM1 myotubes at drug concentrations of 1–5 μM, whereas 131.22 was active at a concentration as low as 0.2 μM. These results highlight the therapeutic potential of bleomycin conjugates and the room for further optimization.

To reduce the undesired DNA‐cleavage activity, the Bleomycin A5 moiety of Cugamycin was replaced with deglycobleomycin, devoid of the carbohydrate domain [134]. The resulting compound, DeglycoCugamycin (Figure 20B), showed a slightly reduced affinity to r(CUG)₁₂ (K _d = 610 nM, Table 4), while conserving the same r(CUG)₁₀ cleavage activity in vitro (∼35% cleavage at 1 μM after 24 h). While Cugamycin showed undesired DNA‐cleavage activity at concentrations above 500 nM, DeglycoCugamycin did not cleave DNA at concentrations up to 2 μM. In mouse C2C12 myoblasts, Cugamycin induced DNA damage (manifested as the appearance of γ‐H2AX foci) at a concentration of 25 μM, whereas no DNA damage was observed for DeglycoCugamycin in the same conditions. Finally, in DM1 patient‐derived myotubes, DeglycoCugamycin was roughly as active as Cugamycin with respect to the decrease of the mutated DMPK level, reduction of the r(CUG)^exp–MBNL1 foci and correction of MBNL1‐related splicing defects. This demonstrates that the removal of the carbohydrate domain of bleomycin allows for elimination of the undesired DNA damage without compromising the cellular uptake or activity of the drug with respect to the DM1‐associated molecular mechanisms.

2.3.3. Self‐Assembly of Bis‐Benzimidazole Ligands

Similar to the on‐target self‐assembly of triaminotriazine derivatives (cf. Section 2.1.3), bis‐benzimidazole ligands have been harnessed in the KTGS approach based on the in situ copper‐free alkyne–azide cycloaddition reaction. Toward this end, the optimized dimer 2H‐K4NMeS has been functionalized with reactive azide and terminal alkyne groups (Figure 21A) [129]. A reaction between equimolar concentrations of the alkyne 2H‐K4NMeS‐Aak (129.4) and the azide N ₃ ‐2H‐K4NMeS (129.5) in the presence of r(CUG)₁₂ gave an ∼80% yield of the tetramer product (Figure 21B), while almost no product was detected in the absence of RNA and ≤10% in the presence of alternative RNA templates. The formation of the tetramer was also detected in DM1 cells (expressing 500 CUG repeats), but not in non‐DM1 cells, treated with 2H‐K4NMeS‐Aak and a biotinylated derivative of N ₃ ‐2H‐K4NMeS, where the biotin handle was introduced to facilitate isolation of the products via affinity precipitation. The template‐driven formation of tetramers was further exploited in a FRET‐based approach to detect and visualize r(CUG)^exp in DM1 cells. A reaction between the alkyne FAM‐2H‐K4NMeS‐Aak (129.9) endowed with fluorescein (FAM) as fluorescence donor and the azide N ₃ ‐2H‐K4NMeS‐TAMRA (129.5) bearing TAMRA as fluorescence acceptor in the presence of r(CUG)^exp template produces a FRET signal due to formation of the tetrameric product combining both donor and acceptor fluorophores (Figure 21C). Thus, an ∼12% increase of FRET signal was observed after 48 h in the presence of r(CUG)₁₂ template, but not in the presence of a fully matched RNA, r(GC)₂₀. When DM1 cells were incubated with FRET sensors FAM‐2H‐K4NMeS‐Aak and N ₃ ‐2H‐K4NMeS‐TAMRA, a decrease in the fluorescence lifetime of the donor fluorophore was observed, indicating the formation of the FRET product in cells, while no decrease was observed in non‐DM1 cells. Fluorescence lifetime imaging microscopy (FLIM) revealed that the FRET signal originated from the cytoplasm of DM1 cells, despite the predominant localization of r(CUG)^exp in ribonuclear foci. Thus, the formation of the tetrameric FRET product on r(CUG)^exp may facilitate its cytoplasmic translocation, which is in line with the enhancement of DMPK mRNA translation by bis‐benzimidazole dimers (see above).

FIGURE 21.

(A) Structures of azide‐ and alkyne‐modified dimeric bis‐benzimidazole ligands for in situ self‐assembly via copper‐free click reaction. The structure of the H unit is shown in Figure 15A. (B) Template‐assisted synthesis of the tetramer. (C) Principle of the FRET sensor based on the template‐assisted synthesis of the tetramer from fluorophore‐modified bis‐benzimidazole ligands. (D) On‐template oligomerization of the dual‐functionalized ligand N ₃ ‐2H‐K4NMeS‐Aak.

In an extension of the KGTS approach, the dimeric bis‐benzimidazole scaffold was endowed with both azide and alkyne groups (N ₃ ‐2H‐K4NMeS‐Aak or 129.6, Figure 21A), thereby allowing its self‐oligomerization on the r(CUG)^exp template [129]. Although the formation of the oligomeric products has not been observed in vitro, N ₃ ‐2H‐K4NMeS‐Aak had high biological activity, which may be considered as an indirect yet solid evidence of intracellular synthesis of the strongly binding oligomer on the r(CUG)^exp template (Figure 21D). Thus, treatment with as little as 100 pM of N ₃ ‐2H‐K4NMeS‐Aak had a statistically significant effect on the missplicing of MBNL1 pre‐mRNA in DM1 cells, and the IC₅₀ value of ∼10 nM obtained in this assay represents a 100‐fold gain with respect to the dimer 2H‐K4NMeS‐Aak and a 50‐fold gain with respect to an equimolar mixture of the reactive dimers 2H‐K4NMeS‐Aak and N ₃ ‐2H‐K4NMeS.

The weak point of most KGTS implementations based on the Huisgen azide–alkyne cycloaddition is the slow rate of the uncatalyzed reaction. To address this issue, the Disney lab developed an alternative approach to self‐assembly of bis‐benzimidazole ligands based on the bioorthogonal tetrazine ligation [139]. Dimeric bis‐benzimidazole ligands functionalized with a 1,2,4,5‐tetrazine (139.2) and 1,2‐dimethylcyclopropene units (139.6, Figure 22A) selectively react in the presence of r(CUG)₁₂ to form detectable amounts of the ligated product already after 1 h (Figure 22B), in contrast to 24 h required for the azide–alkyne cycloaddition reaction. The tetrazine ligation was further exploited in a fluorogenic reaction allowing the detection of r(CUG)^exp. In the derivative 139.8, the fluorescence of FAM is quenched due to the proximity of the tetrazine residue, but restored upon reaction with the cyclopropenyl ligand 139.6 in the presence of r(CUG)^exp (Figure 22C). Indeed, green fluorescent signal was detected in live DM1 patient‐derived fibroblasts treated with 139.8 and 139.6 (but not in non‐DM1 fibroblasts), as well as in muscle fibers from the HSA^LR mouse (inset in Figure 22C). This method allows a direct visualization of r(CUG)^exp in live cells, in contrast to the classical FISH technique that requires cell fixation and permeabilization.

FIGURE 22.

(A) Structures of tetrazine‐ and cyclopropene‐modified dimeric bis‐benzimidazole ligands for in situ self‐assembly via tetrazine ligation. The structure of the H unit is shown in Figure 15A. (B) Template‐assisted synthesis of the tetramer. (C) Principle of fluorogenic detection of r(CUG)^exp upon reaction of the quenched fluorescein‐conjugated tetrazine ligand 139.8 with cyclopropene ligand 139.6. Inset: Image of HSA^LR mouse myofibers treated with 139.8 and 139.6; the fluorescent product is seen as green dots. (D) On‐template oligomerization of the dual‐functionalized ligand 139.7 via tetrazine ligation. Inset in panel C reproduced with permission [139]. Copyright 2020, American Chemical Society.

Template‐guided oligomerization of the bivalent ligand 139.7 bearing both tetrazine and cyclopropene reactive groups (Figure 22A) has also been explored [139]. In vitro, incubation of 139.7 with r(CUG)₁₂ (containing five binding sites) led to the formation of the ligated tetramer (Figure 22D, n = 2), whereas longer r(CUG)₂₄ led to the formation of the hexamer (n = 3), detected by mass spectrometry. Treatment of DM1 fibroblasts with as little as 100 pM 139.7 led to a reduction of r(CUG)^exp–MBNL1 foci and an improvement of MBNL1 missplicing, whereas the dimers 139.2 and 139.6 showed similar levels of activity only at a 10,000‐fold higher concentration. Remarkably, 139.7 was active already after 24 h, while the corresponding azide–alkyne analog N ₃ ‐2H‐K4NMeS‐Aak used at the same concentration of 100 pM showed an effect on the splicing only after 48 h, illustrating the advantage of the tetrazine ligation for on‐target self‐assembly.

2.3.4. Miscellaneous Benzimidazole Derivatives

Considering the potential of bis‐benzimidazoles as r(CUG)^exp ligands, Disney and coworkers screened a library of 320 smaller, drug‐like benzimidazole derivatives with respect to their capacity to inhibit r(CUG)₁₂–MBNL1 interaction in the HTRF assay [140]. This led to the identification of 28 hits with IC₅₀ values in the 30–150 μM range, of which three benzimidazole derivatives (140.1, 140.16, and 140.17, Figure 23A) were further confirmed in a cellular assay based on the splicing of the TNNT2 minigene in DM1‐modeling HeLa cells. Interestingly, all three compounds are uncharged at physiological pH, and ligands 140.16 and 140.17 both contain a 2‐phenyl‐substituted benzimidazole fragment (highlighted in Figure 23A) in combination with a furan‐2‐carboxamide group, suggesting a common pharmacophore. Competition dialysis assay confirmed the selectivity of these three ligands for r(CUG)^exp over other RNA motifs in vitro, with 140.17 being the strongest binder. Target engagement by these ligands was confirmed via a competitive Chem‐CLIP (C‐Chem‐CLIP) approach using the RNA‐crosslinking ligand 2H‐4‐CA‐Biotin (Figure 20A; cf. Section 2.3.2). These experiments demonstrated that ligands 140.1, 140.16 or 140.17 compete with the crosslinker 2H‐4‐CA‐Biotin for binding to the r(CUG)^exp target in DM1‐modeling HeLa cells. Altogether, the activity of 140.16 or 140.17 can be related to their direct binding to r(CUG)^exp and competition with MBNL1, whereas ligand 140.1 not only binds to r(CUG)^exp, but also reduces the level of this toxic transcript by ∼40% via other mechanisms.

FIGURE 23.

(A) Benzimidazole derivatives studied as r(CUG)^exp ligands. (B) Structure of the ligand 141.1 discovered by the DEL approach and its derivatives. (C) Effect of DEL1 and DEL‐Bleo on the number of nuclear r(CUG)^exp–MBNL1 foci in DM1 patient‐derived myotubes [141]. Panel C reproduced with permission. [141] Copyright 2020, American Chemical Society.

Another benzimidazole derivative, 142.p2 (Figure 23A), was identified upon pharmacophore search in a ∼4.3 million virtual compound library [142]. This compound was moderately active in vitro, disrupting the r(CUG)₁₂–MBNL1 interaction by ∼30% at a concentration of 50 μM (HTRF). It was also active in cellular assays, increasing the nuclear export of luciferase mRNA containing 800 CUG repeats in its 3′‐UTR and modulating, albeit to a limited extent (<10%), the splicing of the TNNT2 minigene in DM1‐modeling HeLa cells.

In 2022, the Disney lab developed a DNA‐encoded library (DEL) approach to identify novel r(CUG)^exp ligands [141]. A bead‐supported DEL composed of three building blocks enriched in nitrogen heterocycles and containing 12,672 unique members was screened with respect to binding to Cy5‐r(CUG)₁₂ as the target and Alexa750‐r(GC)₈ as a control substrate, via two‐color fluorescence‐activated bead sorting. Subsequent decoding of DNA tags and building block frequency analysis led to the identification of 16 hits, which were resynthesized and directly assessed in a cellular assay with respect to their capacity to enhance the expression of the luciferase reporter whose mRNA contains 800 CUG repeats in its 3′‐UTR. Benzimidazole derivative 141.1 (Figure 23B) was the most active in this assay, whereas its enantiomer was inactive, highlighting the importance of chirality for interaction with r(CUG)^exp. Further investigations were performed with the N‐propyl derivative DEL1 (Figure 23B), whose binding to r(CUG)₁₂ was characterized by a K _d value of 6.3 μM (MST). Target engagement in vitro and in DM1 patient‐derived myotubes was confirmed via the Chem‐CLIP approach using the photo‐crosslinking derivative of DEL1, which demonstrated about threefold enrichment of DMPK mRNA in the pulled‐down RNA fraction. Conversely, no DMPK enrichment was observed in non‐DM1 myotubes, consistent with binding of the ligand to long r(CUG)^exp repeats. DEL1 partially rescued the MBNL1‐dependent missplicing of MBNL1 pre‐mRNA and modestly reduced the number of r(CUG)^exp foci in DM1 myotubes at a concentration of 5 μM. To improve the biological activity of DEL1, it was endowed with an RNA‐cleaving Bleomycin A5 moiety. The resulting conjugate, DEL1‐Bleo (Figure 23B), was characterized by increased affinity to r(CUG)^exp (K _d = 1.1 μM, MST in non‐cleaving conditions) and cleaved r(CUG)₁₀ in vitro (∼60% at 10 μM DEL1‐Bleo), as evidenced by gel electrophoresis. In DM1 myotubes, DEL1‐Bleo reduced the level of the mutated DMPK transcript by 25% and rescued the aberrant splicing of MBNL1 by 40% at the 5 μM dose, without significant DNA damage. The comparison of the effect of DEL1 and DEL1‐Bleo on the reduction of nuclear r(CUG)^exp–MBNL1 foci in DM1 myotubes (Figure 23C) nicely illustrates the increase in the ligand's potency by the appendage of an RNA‐cleaving moiety.

2.4. Aromatic Diamidines and Related Ligands

2.4.1. Pentamidine and Analogs

Pentamidine (Figure 24) is a well‐known, AT‐specific DNA minor groove binder, approved as a drug for the treatment of trypanosomiasis (sleeping sickness), leishmaniasis, and pneumocystic pneumonia. Back in 2009, upon screening a set of 26 well‐established DNA and RNA binders with respect to their capacity to disrupt the r(CUG)₄–MBNL1 interaction in EMSA, Berglund and coworkers found that pentamidine was the most active drug with an IC₅₀ value of 58 μM and suggested that it could bind the minor groove of the double‐stranded RNA structure formed by r(CUG)^exp and thereby prevent MBNL1 sequestration [146]. In line, Artero and coworkers found that pentamidine did not significantly alter the level of r(CUG)^exp transcript in a Drosophila DM1 model, suggesting that it acts primarily at the RNA level [147]. However, this mechanism was questioned in another study by Berglund and coworkers, who found that pentamidine disrupts the r(CUG)₄–MBNL1 interaction only in the presence of Bromophenol Blue, suggesting that the dye acts synergistically with pentamidine [143]. In addition, pentamidine, as well as its shorter‐linker analog, propamidine (Figure 24), strongly reduced the level of r(CUG)^exp in HeLa cells transfected with a plasmid containing 960 interrupted CTG repeats (IC₅₀ ≈ 35 μM), which was attributed to the strong and selective binding of pentamidine and its analogs to d(CTG)·d(CAG) repeats in duplex DNA, presumably leading to stalling of RNA polymerase and transcription inhibition at this locus. This hypothesis was further confirmed using an in vitro transcription assay, where pentamidine inhibited the transcription of d(CTG)₅₄ and d(CAG)₅₄ templates (IC₅₀ = 14.2 and 13.2 μM, respectively) more efficiently than would be expected with regard to their AT content. Titrations with fluorescent pentamidine analogs provided another evidence of their direct binding to d(CTG)·d(CAG) repeats in duplex DNA [143].

FIGURE 24.

Aromatic diamidines studied as r(CUG)^exp ligands. The EC₅₀ values refer to the activity of compounds in the HeLa cell‐based INSR splicing correction assay [145, 147, 150].

Regardless of their precise mechanism of action, pentamidine and its analogs decrease the formation of r(CUG)^exp−MBNL1 foci, which results in the release and diffusion of MBNL1 throughout the cell nuclei, and rescue the MBNL1‐dependent missplicing of the INSR and TNNT2 minigenes in the HeLa cell DM1 model, leaving the MBNL1‐independent splicing unaffected [146]. The efficiency of missplicing correction correlates with the length of the linker connecting the two benzamidine fragments (Figure 24). Thus, hexamidine (n = 6) and heptamidine (n = 7) were most efficient in the splicing correction assay, with half‐maximum effective concentration (EC₅₀) values of 9−15 μM. Unfortunately, the higher efficiency of splicing correction also correlates with increased toxicity for the longer‐linker derivatives [143]. In the murine DM1 model, pentamidine, and, even more, heptamidine partially rescued the missplicing of Clcn1 and Atp2a1 (Serca1) pre‐mRNA; full recovery of the splicing pattern could not be achieved, as higher doses of the drugs were lethal [143, 146]. In addition, heptamidine, administered at 20–30 mg/kg, decreased the severity of DM1‐related myotonia in mice from grade 3 to grade 1 or 0 [143]. In parallel, Artero and co‐workers studied the effect of pentamidine in a Drosophila model recapitulating DM1‐specific heart dysfunction and found that pentamidine, added to the nutritive media at a concentration of 1 μM, partially rescued some of the abnormal heart parameters and increased the median lifespan of the flies almost to the normal level [147].

Blankert and coworkers studied the interaction of pentamidine with r(CUG)₅₀ in vitro using affinity capillary electrophoresis (ACE) and found a K _d value of about 70 μM, albeit with unusually high stoichiometry (∼121:1, i.e., ∼4.8 molecules of pentamidine per every 5′‐CUG‐3′/5′‐CUG‐3′ site). The same method was applied to a small library of 13 drugs, resulting in the identification of the structurally close 1,2‐ethane bis‐1‐amino‐4‐benzamidine (EBAB, Figure 24) as a ligand with a slightly lower r(CUG)₅₀ affinity (K _d = 83 μM), yet much lower toxicity than pentamidine [148]. Of note, a more recent study using a longer r(CUG)₉₅ sequence suggested that EBAB has a much higher affinity for CUG repeats (K _d = 18.7 μM), along with a 5.6‐fold selectivity with respect to the DNA counterpart (d(CTG)₉₅, K _d = 105 μM) [93]. Similarly to pentamidine, EBAB reduced the sequestration of MBNL1 into ribonuclear foci in DM1‐modeling HeLa cells (by 43% at 100 μM EBAB), without inducing cytotoxicity [148].

Aiming to improve the efficiency and reduce the toxicity of diamidines, variations of this scaffold have been explored. Berglund and coworkers studied a number of pentamidine analogs differing by the number of benzamidine groups (one or two), the substitution pattern of benzamidine moieties, as well as the presence and the nature of N‐substituents in the amidine groups. However, following the assessment in a HeLa cell‐based INSR minigene splicing assay, none of the analogs containing a flexible hydrocarbon linker was found superior to pentamidine in terms of the efficacy/toxicity ratio [144]. Instead, two rigid diamidines, 4,4′‐diamidinobiphenyl and furamidine (Figure 24), were roughly as active as pentamidine in the missplicing correction assay (EC₅₀ = 30−31 μM), while being significantly less toxic. While the effects of furamidine were highly specific to the DM1 model, 4,4′‐diamidinobiphenyl also affected the splicing of reporter minigenes in a control system devoid of r(CUG)^exp repeats, indicating some nonspecific effects on RNA splicing. In line, furamidine strongly decreased the number of r(CUG)^exp−MBNL1 foci and released the sequestered MBNL1 in the HeLa cell DM1 model. Finally, in a murine DM1 model, furamidine, administered at 10 or 20 mg/kg, partially rescued missplicing of the Clcn1 and Atp2a1 genes without inducing toxicity, demonstrating an improvement over pentamidine [144].

To understand the mechanism of action of furamidine, Berglund and coworkers undertook a detailed study both in the murine HSA^LR model and in DM1 patient‐derived myotubes [149]. In the mouse model, furamidine specifically decreased the expression of the HSA transgene containing 220 CUG repeats in its 3′‐UTR, suggesting a mechanism involving DNA binding and inhibition of transcription at the level of d(CTG)·d(CAG) repeats. Accordingly, furamidine strongly binds to a double‐stranded DNA oligonucleotide containing four d(CTG)·d(CAG) repeats in vitro (K _d = 0.49 μM, ITC). However, the decrease of the r(CUG)^exp transcript level was insignificant in human DM1 myoblasts treated with 0.5−1 μM of furamidine, despite the fact that these doses induced the maximal rescue of MBNL1‐related and several other DM1‐linked splicing defects. In line with the mechanism operating at the r(CUG)^exp level, in vitro assays demonstrated that furamidine binds to r(CUG)₄ stronger than to DNA repeats (K _d ≈ 0.1 μM, ITC) and disrupts the r(CUG)₈–MBNL1 complex (IC₅₀ = 40 μM, EMSA). Furamidine also increased, albeit to a small extent, the expression of splicing regulators MBNL1 and MBNL2 in human DM1 myotubes, both at the transcript (∼1.5‐fold and ∼2‐fold, respectively) and the protein (≤1.2‐fold and ≤1.3‐fold, respectively) levels; this effect may also contribute to the correction of MBNL1‐related splicing defects in vivo. Altogether, these data suggest a multi‐level mechanism of action for furamidine (and, likely, other diamidines) where the ligand interacts both with d(CTG)·d(CAG) repeats in DNA (leading to reduced synthesis of r(CUG)^exp) and with the r(CUG)^exp transcript per se (leading to the release of MBNL1 and other splicing regulators), with concomitant upregulation of MBNL1 expression via direct or indirect mechanisms. Despite (or thanks to) this multitargeted mechanism, the effect of furamidine in DM1 models appears to be highly specific, as the global gene expression pattern is only slightly affected. Thus, in a murine DM1 model treated with furamidine, the expression of only 2.9% of all genes has been modified, and 21% of these differentially expressed genes were rescued back to wild‐type levels [149]. Interestingly, a combination of 0.25−1 μM of pentamidine with 25−50 μM of erythromycin, a clinically used macrolide antibiotic that is also active in DM1 models, presumably via binding to r(CUG)^exp RNA and disruption of r(CUG)^exp−MBNL1 interaction (cf. Section 2.5 below), displayed an additive effect with regard to splicing correction in DM1 myotubes, without inducing cell toxicity. In a murine DM1 model, a combination of an oral prodrug of furamidine, pafuramidine (Figure 24), with erythromycin resulted in a supra‐additive (synergistic) effect, evidenced by rescuing more than a twofold number of missplicing events compared to either drug alone, and led to an alleviation of the DM1‐related myotonia from grade 3 to grade 2 [150].

Eight analogs of furamidine were recently studied in DM1 and DM2 patient‐derived fibroblasts, resulting in the identification of more potent extended diamidines DB1247 and 145.7 (Figure 24) [145]. Among these, DB1247 rescued the missplicing not only of INSR (EC₅₀ = 0.50–0.75 μM), but also of two other genes misspliced in DM1, FLNB and SYNE1; unfortunately, this effect was compromised by the cytotoxicity observed at concentrations above 1 μM. In turn, 145.7 was slightly less efficient in terms of splicing correction (EC₅₀ ≈ 1 μM for INSR missplicing) but had no toxicity up to 64 μM, highlighting its potential for further therapeutic development. Another extended diamidine P1 (Figure 24), along with several related analogs, was identified by Disney and coworkers upon virtual screening of over 8 million compounds with respect to their similarity to Hoechst 33258 and pentamidine as well‐established (CUG)^exp ligands [137]. Diamidine P1 inhibits the r(CUG)₁₂–MBNL1 interaction with an IC₅₀ value of 10 μM (HTRF), being 100‐fold more active than pentamidine. This compound also shows preferential binding to the 5′‐CUG‐3′/5′‐CUG‐3′ motif compared to other RNA motifs in competition dialysis experiments, although the binding does not correlate with the number of repeated CUG units. Despite its in vitro activity, P1 does not modulate the splicing of the TNNT2 minigene in the HeLa DM1 model cells at concentrations up to 500 μM.

2.4.2. Miscellaneous Aromatic Amidines and Related Ligands

Diguanidinobenzoate 151.1, similar bis‐imidazoline 151.2 and 6‐amidinoquinoline 151.3 (Figure 25A) were selected by Disney and coworkers from an RNA‐targeted small‐molecule library [151]. All three compounds inhibit the r(CUG)₁₂–MBNL1 interaction in vitro, with 151.1 being the most active (IC₅₀ = 8 μM, HTRF). The interaction of these ligands with a 13‐mer RNA duplex containing a single 5′‐CUG‐3′/5′‐CUG‐3′ site was studied by NMR spectroscopy and MD simulations to determine binding modes. All three ligands bind near the central 5′‐CUG‐3′/5′‐CUG‐3′ site, with the RNA globally maintaining its A‐form. While aromatic diamidines such as pentamidine are typically considered minor‐groove binders, ligands 151.1 and 151.2 were located in the major groove of the RNA duplex. Remarkably, despite their high activity in terms of r(CUG)₁₂–MBNL1 inhibition, both 151.1 and 151.2 make rather few contacts with the RNA. Ligand 151.1 disrupts the hydrogen bonding within the U·U mispair and is held mainly by hydrogen bonds between its protonated guanidine units and the phosphate groups of the RNA (Figure 25B,C). In contrast, in the complex with 151.2, the mispaired uracils (red in Figure 25B) remain hydrogen‐bonded, but are displaced toward the minor groove by the incoming ligand, which is held mainly by weak π‐stacking interactions with these uracils. Finally, ligand 151.3, despite being much less efficient in terms of r(CUG)₁₂–MBNL1 inhibition (IC₅₀ = 100 μM), binds with its larger aromatic surface nearly perpendicular to the RNA duplex axis and its polar amidine group extending into the major groove. In contrast to 151.1 and 151.2, 151.3 forms efficient stacking interactions with the two G≡C base pairs adjacent to the disrupted U·U mispair, resulting in a binding mode reminiscent of classical intercalation (Figure 25C). These three complexes are among the very few experimental structures of r(CUG)^exp–bound ligands.

FIGURE 25.

(A) Chemical structures, (B) NMR‐based structural models, and (C) close‐up views of binding sites of ligands 151.1, 151.2, and 151.3 with 13‐mer RNA duplex containing a single 5′‐CUG‐3′/5′‐CUG‐3′ site. The CUG nucleotides constituting the binding site are colored in panel (B): C yellow, U red, and G green; the ligand is shown in golden‐yellow color. Replotted from the PDB ID: 9CPG, 9CPI, and 9CPD, respectively, using UCSF Chimera.

2.5. Aminoglycosides and Macrolides

Aminoglycosides represent an important family of antibiotics known to interact with a wide range of RNA and DNA structures, primarily bacterial 16S rRNA, but also viral RNA, microRNA, and triple‐helical structures [152, 153, 154, 155, 156]. Berglund and coworkers discovered the capacity of neomycin B (Figure 26) to disrupt the r(CUG)₄–MBNL1 interaction in vitro (IC₅₀ = 280 μM) concurrently with pentamidine's [146]. In line, affinity capillary electrophoresis experiments provided a K _d value of 17 μM for the binding of neomycin B to r(CUG)₅₀, indicating roughly four times higher affinity than pentamidine [148]. However, unlike pentamidine, neomycin B, even at high concentrations (200–500 μM), did not affect the MBNL1‐dependent missplicing of the INSR and TNNT2 minigenes in DM1‐modeling HeLa cells or in the murine DM1 model [146, 157]. This lack of activity in vivo is most likely due to the binding of neomycin B to other cellular RNA targets, i.e., poor selectivity for r(CUG)^exp.

FIGURE 26.

Structures of simple aminoglycoside and macrolide antibiotics studied as r(CUG)^exp ligands.

Sisomicin (Figure 26) is a broad‐spectrum aminoglycoside antibiotic structurally related to gentamicin, which is also known to interact with the trans‐activating response (TAR) element of HIV‐1 RNA. Sisomicin binds to r(CUG)₁₀ with a dissociation constant K _d ≈ 0.9 μM, as per fluorescence titrations with a fluorescein‐labeled oligonucleotide [158]. However, the binding was also observed at temperatures above the T _m of the r(CUG)₁₀ hairpin, suggesting an interaction with the unfolded RNA repeat. The selectivity of sisomicin for r(CUG)₁₀ with regard to other RNA structures, or its activity in DM1 models have not been explored.

Kanamycin A (Figure 27), another aminoglycoside antibiotic, inhibits the r(CUG)^exp–MBNL1 interaction only at concentrations above 200 μM [146, 159]. A trimer of kanamycin A, 3K‐4 (Figure 27) containing three 6′‐N‐linked aminoglycoside moieties assembled on a peptoid scaffold, spaced by four propylamine modules and endowed with a fluorescein tag (F in Figure 27), was initially synthesized by Disney and coworkers for targeting expanded r(CCUG)^exp repeats involved in the pathogenesis of myotonic dystrophy type 2 (DM2) [160]. In addition to strong binding to CCUG repeats (K _d = 8 nM for r(CCUG)_2×12), this construct also showed high affinity to CUG repeats, with K _d = 0.21 μM for r(CUG)_2×12 and a 4:1 stoichiometry, suggesting that the binding site comprises three 5′‐CUG‐3′/5′‐CUG‐3′ repeats for each trimer molecule (Table 5). The affinity of 3K‐4 for r(CUG)^exp is thus comparable with that of MBNL1 (K _d = 0.25 μM in the same conditions); accordingly, 3K‐4 disrupts the r(CUG)₁₀₉–MBNL1 interaction in a surface‐based in vitro displacement assay (IC₅₀ = 0.28 μM).

FIGURE 27.

Structures of kanamycin A and kanamycin‐decorated peptoid oligomers (a + 1)M‐b and (a + 1)M‐b‐DR9 designed as r(CUG)^exp ligands by Disney and coworkers.

TABLE 5.

Dissociation constants (K _d) and binding stoichiometry (N) for interaction of kanamycin and kanamycin‐decorated peptoid oligomers ( a  + 1)M‐ b and ( a  + 1)M‐ b ‐DR9 (structure: Figure 27) with r(CUG)_2×12, and IC₅₀ values for inhibition of the r(CUG)₁₀₉–MBNL1 interaction in vitro.

Oligomera	M	a	b	K d (r(CUG)2×12) (N)	IC50 (r(CUG)109–MBNL1)
FITC‐K	K	—	—	1.0 μM (11)	210 μM
					>250 μMb
3K‐4	K	2	4	0.21 μM (4)	0.28 μM
2K‐2	K	1	2	0.05 μM (5.3)	0.3 μM
3K‐2	K	2	2	0.02 μM (3.3)	0.04 μM
4K‐2	K	3	2	0.004 μM (2.6)	0.007 μM; 16 μMb
Kaz monomer	Kaz	—	—	0.92 μM (11)	190 μM
2Kaz‐2	Kaz	1	2	>0.50 μM	3 μM
4Kaz‐2	Kaz	3	2	>0.25 μM	>25 μM
2KazL‐2	KazL	1	2	0.075 μM (5.2)	0.9 μM
4KazL‐2	KazL	3	2	0.020 μM (2.4)	0.05 μM
2K‐2‐DR9	K	1	2	0.0035 μM (3.7)	0.24 μMb
4K‐2‐DR9	K	3	2	—	1.43 μMb

^a Data from refs. [159, 160, 161, 162].

^b IC₅₀ values for inhibition of r(CUG)₁₀–MBNL1 interaction measured by qTR‐FRET assay.

To optimize the binding of kanamycin derivatives to r(CUG)^exp, Disney and coworkers systematically varied the design of the peptoid oligomers described by the general formula (a + 1)M‐b , where (a + 1) is the total number of kanamycin moieties, b is the number of propylamine spacers between the adjacent kanamycin residues, and M is the variant of kanamycin attachment (Figure 27 and Table 5) [159]. The derivatives containing kanamycin moieties separated by two propylamine spacers (instead of four in 3K‐4), such as the dimer 2K‐2, the trimer 3K‐2 and the tetramer 4K‐2, are the most efficient r(CUG)^exp ligands, in agreement with the shorter distance between the binding sites in r(CUG)^exp than in r(CCUG)^exp. The affinity of these oligomers to r(CUG)^exp strongly increases with the number of kanamycin residues, with each additional residue amplifying the affinity 2.5‐fold to 50‐fold. Thus, the affinity of the tetramer 4K‐2 (K _d = 4 nM) is 250‐fold higher compared to monomeric kanamycin A (FITC‐K, Figure 27) and 63‐fold higher with regard to MBNL1 protein. An even stronger multivalence effect was observed in the capacity of oligomers to disrupt the r(CUG)₁₀₉–MBNL1 interaction in vitro, with the tetramer 4K‐2 (IC₅₀ = 7 nM) being 7500‐fold more potent than FITC‐K (Table 5). Intriguingly, this effect was strongly reduced when the kanamycin residues were replaced with another aminoglycoside, neamine, indicating that kanamycin moieties are well‐suited for interaction with CUG repeats. The presence of the fluorescein tag allowed monitoring cellular uptake of the oligomers. In cultured C2C12 mouse myoblasts, oligomers 2K‐2, 3K‐2, and 4K‐2 were two‐ to threefold more cell‐permeable than FITC‐K, indicating the beneficial effect of the peptoid scaffold on cellular uptake. The dimer 2K‐2 and the trimer 3K‐2 were not toxic, and the tetramer 4K‐2 marginally toxic at a concentration of 5 μM in mouse myoblasts.

Variants of the peptoid oligomers differing in the attachment mode of kanamycin moieties, such as those derived from 6″‐azidokanamycin A (Kaz, Figure 27), were also explored [161]. As in the previous series, the oligomers in which the aminoglycoside moieties were separated by two propylamine spacers, such as 2Kaz‐2, were the most efficient r(CUG)^exp ligands. However, the derivatives bearing more than two Kaz units were less potent inhibitors of the r(CUG)^exp–MBNL1 interaction (e.g., 4Kaz‐2, IC₅₀ > 25 μM), presumably due to lesser flexibility of the short linkers connecting the aminoglycoside moieties to the peptoid scaffold. Conversely, when Kaz moieties were conjugated via a flexible bis(ethylene glycol) linker (KazL, Figure 27), the efficiency of the resulting oligomers was dramatically improved, both in terms of r(CUG)^exp affinity and inhibition of the r(CUG)^exp–MBNL1 complex. In this series, the potency of the ligands expectedly increased with the number of aminoglycoside moieties, with the dimer 2KazL‐2 (IC₅₀ = 0.9 μM) and the tetramer 4KazL‐2 (IC₅₀ = 0.05 μM) being 106‐ and 950‐fold more potent inhibitors than Kaz monomer, respectively. A similar trend was observed with regard to the affinity of ligands to r(CUG)_2×12 (Table 5). Yet, this series failed to outperform the best candidate from the 6′‐N‐linked oligomer series, 4K‐2, both in terms of potency and selectivity, since KazL derivatives displayed non‐negligible binding to the fully paired RNA duplex (e.g., K _d = 0.5 μM for 4KazL‐2). In contrast, cellular uptake (in mouse myoblasts) or toxicity (<5% at a concentration of 5 μM) were essentially not impacted by the nature of kanamycin functionalization or the number of aminoglycoside units per oligomer.

Despite their promising in vitro activity and cellular uptake, kanamycin oligomers of the general formula (a + 1)M‐b were inactive or only marginally active in cellular DM1 models, presumably due to their localization in the perinuclear region and not in the nuclei. To improve the nuclear influx of these oligomers, Disney and coworkers modified their structure by conjugating a cationic nona‐d‐arginine (DR9) tag as a cellular and nuclear transporter [162]. The resulting conjugates, such as 2K‐2‐DR9 and 4K‐2‐DR9 (Figure 27), were over 10‐fold more cell‐permeant than the analogs devoid of the DR9 tag, without being more toxic. The affinity of 4K‐2‐DR9 for r(CUG)_2×12 was only slightly higher compared with the 4K‐2 analog (K _d = 3.5 vs. 4.0 nM, respectively); however, the increase of the potency in terms of inhibition of the r(CUG)₁₀–MBNL1 interaction was nearly 70‐fold (Table 5). In the HeLa cell DM1 model, the treatment with 2K‐2‐DR9 (2–20 μM) rescued the missplicing of the TNNT2 minigene by ∼50%, whereas the treatment with 4K‐2‐DR9 (10 μM) restored the splicing to the normal pattern, without affecting MBNL1‐independent splicing. In addition, treatment with 4K‐2‐DR9 dramatically reduced the number and the size of ribonuclear r(CUG)^exp foci. In a cellular model recapitulating the impaired translation of the DMPK gene in DM1, this oligomer stimulated the expression of a luciferase reporter whose mRNA contained 800 CUG repeats in its 3′‐UTR by up to 90%. Finally, in the murine HSA^LR DM1 model, treatment with 80 mg/kg 4K‐2‐DR9 partially rescued the missplicing of the Clcn1 and Atp2a1 genes, demonstrating the therapeutic potential of cell‐permeant kanamycin oligomers.

In 2016, the macrolide erythromycin (Figure 26) was identified by screening a panel of 20 FDA‐approved antibiotics with respect to their capacity to inhibit the interaction of MBNL1 with surface‐attached r(CUG)₁₀₀, and further confirmed by EMSA (IC₅₀ ≈ 5 μM) [157]. Direct binding of erythromycin to CUG repeats was demonstrated via fluorimetric titrations with Cy3‐r(CUG)₁₀ (K _d = 0.78 μM). In a murine‐cell DM1 model expressing r(CUG)₈₀₀ erythromycin, at concentrations of 25–50 μM, strongly decreased the fraction of cells displaying ribonuclear r(CUG)^exp foci and rescued the missplicing of the Atp2a1 gene. The treatment did not modify the r(CUG)^exp transcript level, suggesting that erythromycin does not act on the transcription level, but rather only via inhibiting the r(CUG)^exp–MBNL1 interaction. In vivo experiments in the HSA^LR mouse model demonstrated that erythromycin, administered either via i.p. injections (50 to 150 mg/kg/day) or per os (300 to 900 mg/kg/day), rescued missplicing of the Clcn1, Atp2a1 and other DM1‐related genes and reduced the severity of myotonia from grade 3 to grades 2 or 1, without inducing toxicity.

Considering the encouraging results obtained in the murine DM1 model and the fact that erythromycin is a nontoxic, clinically used antibiotic, its efficacy has recently been assessed in a double‐blind, placebo‐controlled yet small‐sized phase 2 clinical trial involving 30 adult DM1 patients [163]. Despite its favorable safety and tolerability profile, erythromycin, administered at doses of 500 or 800 mg/day for 24 weeks, failed to modify the six primary DM1‐specific alternative splicing biomarkers assessed by RT‐PCR of muscle biopsies, or to improve DM1 symptoms in the patients. However, a notable improvement of the CLCN1 splicing was observed in a subgroup of patients, and a statistically significant improvement of two secondary splicing biomarkers (MBNL1 exon 7 and CANA1S exon 29) was observed post‐hoc in the whole treated arm, suggesting that the study design and/or the treatment regimen were not optimal. Further evaluation of the erythromycin's efficacy is expected from a well‐powered phase 3 clinical trial. Regarding the clinical prospects of erythromycin, it is important to recall its synergy with furamidine and its prodrug pafuramidine, which has been observed both in DM1 patient‐derived myotubes and in the murine DM1 model (cf. Section 2.4.1) [150].

2.6. Intercalating Antibiotics and Related Compounds

Actinomycin D (ActD, Figure 28A; also known as dactinomycin) is a well‐established anticancer drug that strongly binds to double‐stranded DNA via intercalation at GpC sites and thereby interferes with DNA replication and transcription. Considering the fact that d(CTG)^exp repeats can adopt a hairpin form presenting an array of GpC sites separated by T·T mismatches (cf. Section 1.2.3), these DNA repeats may represent the preferred binding sites for ActD. Using spectroscopic methods, Chen demonstrated that ActD preferentially targets CTG repeats over other trinucleotide repeat sequences, with an apparent dissociation constant K _d = 85 nM and a stoichiometry of two ActD molecules per three T·T mismatch sites in the d(CTG)₃ duplex. The binding of ActD to d(CTG) duplexes is characterized by an extremely slow dissociation, nearly 10‐fold slower compared with fully matched duplexes [166, 167]. An early NMR study by Wang and coworkers demonstrated that ActD indeed binds via intercalation at GpC sites flanking a T·T mismatch, and suggested that N‐methyl groups of N‐methylvaline residues, which snugly fit at the mismatch site, are critical for the high‐affinity binding to CTG repeats [168]. This was further confirmed by the crystal structure of ActD bound to a DNA duplex containing a 5′‐GCTGC‐3′/5′‐GCTGC‐3′ site as a model of d(CTG)^exp repeats (Figure 28B) [169]. In contrast to many mismatch‐specific DNA ligands whose binding leads to the ejection of one or both of the mismatched bases into the grooves [170], the binding of ActD actually stabilizes both mismatched thymines in the intrahelical conformation. ActD stabilizes the oligonucleotide model of a single d(CTG) repeat by ΔT _m = 48°C, which significantly exceeds the stabilization of fully matched DNA duplexes [169]. In addition, ActD was shown to trap the hairpin structure of d(CTG)^exp and prevent its hybridization with the complementary strand, d(CAG)^exp, leading to the formation of cruciform structures that potentially could stall transcription and replication of such repeat sequences in cells [168, 169]. According to ITC, ActD binds to d(CTG)₄ with a cooperative 2:1 stoichiometry and K _d values of 10 and 0.5 μM [171]; of note, these values are at least one order of magnitude higher than the ones determined from spectrophotometric titrations [167]. In contrast, no interaction with the RNA counterpart, r(CUG)₄, was detected [171]. In line, Berglund and coworkers reported that ActD does not disrupt the r(CUG)₄–MBNL1 interaction in EMSA [146]. Taken together, these results strongly suggest that ActD does not operate at the RNA level in cells. Berglund and coworkers found that treatment of DM1‐modeling HeLa cells with low‐nanomolar concentrations of ActD (5–20 nM) decreased the level of the r(CUG)^exp transcript by 50%–80% and the number of nuclear r(CUG)^exp–MBNL1 foci by 50%, in particular the large ones. Moreover, ActD, administered at 0.125–0.25 mg/kg for 5 days, corrected the missplicing of the Clcn1 and Atp2a1 (Serca1) genes in the murine HSA^LR DM1 model. At these doses, ActD only marginally affected the global gene expression profile [171]. Collectively, these results suggest that ActD is a promising drug for DM1 treatment, operating at the DNA level by reducing the transcription of the d(CTG)^exp repeat, when administered at doses lower than those typically used for cancer treatment.

FIGURE 28.

(A) Structure of actinomycin D (ActD). (B) X‐ray structure of a 2:1 complex of ActD (orange) bound to an oligonucleotide model of the d(CTG) repeat in DNA (replotted from the PDB ID: 1MNV using UCSF Chimera). Methyl groups of the N‐methylvaline residues in ActD critical for the specific binding to T·T sites are indicated with red arrows. Nucleobase colors: A red, G green, C yellow, T blue. (C) Structure of echinomycin. (D) Structures of daunorubicin and doxorubicin; IC₅₀ value refers to the inhibition of r(CUG)₁₀–MBNL1 complex in EMSA [164]. (E) Structures of lomofungin and dilomofungin; IC₅₀ values refer to the inhibition of r(CUG)₁₂–MBNL1 complex in AlphaScreen assay [165].

Echinomycin (Figure 28C), another anticancer antibiotic known to bind to DNA and to interfere with many DNA‐related transactions in cells, shows preferential, cooperative binding to T·T mismatches at d(CTG/CTG) sites in DNA, as evidenced by preferential stabilization of DNA duplexes containing T·T mismatches and dissociation constants in the sub‐micromolar range (K _d = 170 and 42 nM, SPR) [172]. X‐ray studies revealed that, in contrast to ActD which intercalates at GpC sites, echinomycin binds to d(CTG/CTG) sites by intercalation of quinoxaline rings from two molecules between the T·T mismatch and the two adjacent G≡C base pairs, leading to encapsulation of two G≡C base pairs within each echinomycin molecule. Interestingly, a 1:1 combination of echinomycin and ActD shows stronger binding to T·T mismatches than individual drugs, and the structure of a ternary complex featuring the two drugs bound to both sides of a T·T mismatch has recently been solved by X‐ray crystallography [173]. Despite its promising affinity to d(CTG/CTG) sites, the binding of echinomycin to longer CTG repeats or to their RNA counterparts, or its activity in DM1 models have not been investigated. Nevertheless, the structure of echinomycin has been used by Miller and coworkers as the starting point for the design of disulfide peptides developed as r(CUG)^exp ligands (see Section 2.8.1 below) [174, 175, 176].

In contrast to ActD and echinomycin, daunorubicin (or daunomycin, Figure 28D), another DNA‐intercalating anticancer antibiotic, directly interacts with r(CUG)^exp RNA repeats. Daunorubicin was identified by Artero, Llamusi and coworkers in 2018 through an in vitro screening of ∼6500 compounds for their capacity to disrupt the r(CUG)₂₆–MBNL1 interaction, as assessed by fluorescence polarization assay and confirmed by EMSA (IC₅₀ = 100 nM) [164]. Daunorubicin stabilizes the hairpin form of r(CUG)₁₂ in thermal denaturation experiments (ΔT _m = 8.1°C at 2 μM daunorubicin), which likely leads to reduced MBNL1 binding. In the Drosophila DM1 model, treatment with daunorubicin reduced the accumulation of r(CUG)^exp–Mbl ribonuclear foci in cardiomyocytes (with Mbl being the Drosophila ortholog of the human MBNL1), rescued Mbl‐dependent splicing defects leading to cardiac dysfunction, and restored normal phenotype and survival of the flies. The same study demonstrated that, in contrast to ActD, daunorubicin had no effect on the transcription level nor the length of the expanded CTG repeat, which makes it a prototype of DM1 drugs acting exclusively at the RNA level. Intriguingly, a closely related anthracycline antibiotic, doxorubicin (Figure 28D), as well as its synthetic analog mitoxantrone, more than 20 years ago were shown to induce a contraction of expanded CTG repeat in DM1 fibroblasts, presumably via stabilization of the hairpin form of d(CTG)^exp and/or induction of DNA damage at this locus [177]. Direct interaction of doxorubicin with d(CTG)^exp and/or r(CUG)^exp repeats has not been studied; however, it was shown that doxorubicin strongly reduces the number and the size of r(CUG)^exp foci in DM1 fibroblasts, presumably by interfering with RNA phase separation [8]. At the same time, it is likely that strong cytotoxicity of intercalating antibiotics and their nonspecific effects on DNA‐ and RNA‐related cellular processes represent their Achilles’ heel in terms of prospective DM1 drugs.

Lomofungin (Figure 28E), a natural antibiotic that inhibits RNA synthesis, was identified in an in vitro HTRF screen of ∼280,000 compounds with respect to their capacity to disrupt the r(CUG)₁₂–MBNL1 interaction [165]. Interestingly, dilomofungin (Figure 28E), a dimer of lomofungin formed by its spontaneous oxidation, is even more potent in disrupting the r(CUG)₁₂−MBNL1 interaction (IC₅₀ = 717 and 42 nM for lomofungin and dilomofungin, respectively). Both compounds bind to r(CUG)₁₂ (K _d ≈ 1.0 and 0.61 μM, respectively, fluorimetric titrations) and show preferential binding to RNA sequences containing 5′‐CUG‐3′/5′‐CUG‐3′ sites over other motifs. In the murine C2C12 cellular model of DM1 lomofungin and, to a lesser extent, dilomofungin at a concentration of 10 μM corrected the MBNL1‐dependent missplicing of the Atp2a1 gene. However, dilomofungin dramatically increased the number and intensity of the toxic nuclear r(CUG)^exp foci in cells, presumably via their stabilization and preventing their decay, while lomofungin was devoid of this effect. This may imply that high‐affinity r(CUG)^exp ligands are not necessarily better in terms of cellular efficiency and as potential DM1 drugs.

2.7. Miscellaneous Small‐Molecule Ligands

2.7.1. Small RNA‐Intercalating Ligands

The derivative of 2,7‐naphthyridine 178.2 (also known as SID 3712249, Figure 29A) was identified upon screening of over 300,000 compounds with respect to their capacity to inhibit r(CUG)₁₂−MBNL1 interaction in a TR‐FRET assay (IC₅₀ = 2 μM) [178, 180]. Compound 178.2 directly interacts with r(CUG)₁₂ with a K _d value of 125 nM (from fluorescence anisotropy titrations) and high selectivity with respect to r(CAG)₁₂ and fully paired RNA. MD simulations suggested that 178.2 binds to 5′‐CUG‐3′/5′‐CUG‐3′ sites by taking the place of the two mispaired uracil residues, which are both flipped out into the minor groove of the RNA duplex. The resulting complex (Figure 29B) is stabilized by stacking interactions between 178.2 and two flanking G≡C base pairs, as well as hydrogen bonds between the amino and cyano groups of 178.2 with the unstacked uracil residues. In a cellular DM1 model, compound 178.2, albeit at rather high concentrations (125−500 μM), restored the MBNL1‐dependent missplicing of the INSR and TNNT2 genes to the normal level, without affecting the splicing of genes not regulated by MBNL1.

FIGURE 29.

Small RNA‐intercalating CUG ligands. (A) Chemical structure and (B) molecular model (side and top views) of compound 178.2 (SID 3 712 249) bound to a 5′‐CUG‐3′/5′‐CUG‐3′ site in RNA duplex. Reproduced with permission [178]. Copyright 2013, Springer Nature. (C) Structures of 2‐aminoperimidine derivatives 179.1a and 179.1b. (D) Molecular model of 179.1a bound to r(CUG)₁₂ hairpin and a close‐up view of the binding pocket. Reproduced with permission [179]. Copyright 2023, Q. M. R. Gibaut et al.

More recently, the Disney team identified the 2‐aminoperimidine scaffold (Figure 29C) through an in vitro chemical cross‐linking and isolation by pull‐down (Chem‐CLIP) screen of 13 RNA‐targeted, low‐molecular‐weight fragments endowed with photo‐crosslinking diazirine groups [179]. Ligand 179.1a binds to r(CUG)₁₂ with a dissociation constant K _d = 1.8 μM and a ∼5:1 stoichiometry (MST), consistent with the number of U·U mispairs in this RNA model, without binding to the fully paired RNA duplex. NMR spectroscopy suggested that binding of the ligand disrupts the hydrogen bonding between the mispaired uracil residues. This was confirmed by molecular modeling, revealing that the binding of 179.1a leads to displacement of one of the uracil residues and formation of a hydrogen bond between the incoming ligand and the second, intrahelical uracil residue (Figure 29D). Target engagement in DM1 patient‐derived myotubes was investigated via Chem‐CLIP approach using a photo‐crosslinking derivative of 179.1a and demonstrated significant enrichment (∼7‐fold) of DMPK mRNA in the cross‐linked RNA fraction; in contrast, no DMPK enrichment was observed using myotubes from a healthy donor, consistent with the selective binding of the ligand to r(CUG)^exp repeats. To decrease the level of toxic r(CUG)^exp RNA, the Disney team, harnessing the same strategy as used in the design of Cugamycin (cf. Section 2.3.2), developed an RNA‐degrading derivative 179.1b (Figure 29C), where the 2‐aminoperimidine ligand was linked to Bleomycin A5. In vitro studies demonstrated that the conjugate 179.1b maintained the parent ligand's affinity and selectivity to r(CUG)^exp and, in addition, cleaved r(CUG)₁₀ oligonucleotide, with primary cleavage sites corresponding to the G≡C base pairs 3′ to U·U mispairs. Treatment of DM1 patient‐derived myotubes with 5 μM 179.1b led to a 41% reduction of the DMPK transcript level; in contrast, no decrease was observed in the non‐DM1 myotubes whose DMPK transcript contained only 20 CUG repeats. Finally, RNA‐seq analysis indicated that treatment with 179.1b restored to the normal level 51% of the 211 missplicing events associated with the DM1 phenotype. This was consistent with the reduction of the number of r(CUG)^exp−MBNL1 foci in myotubes treated with 179.1b, suggesting partial release of MBNL1. Altogether, despite an apparent simplicity of its RNA‐binding unit, the conjugate 179.1b appears as a potent and specific r(CUG)^exp RNA degrader with low toxicity and high therapeutic potential.

2.7.2. Groove‐Binding Ligands

Chromomycin A3 (ChroA3, Figure 30A) is a glycosidic antitumor antibiotic that strongly binds to GC‐rich regions of DNA and inhibits RNA synthesis. In the presence of Mg²⁺, ChroA3 forms a dimeric complex that binds to DNA in the minor groove, making multiple specific contacts with guanine residues, which explains the high specificity of this binding (Figure 30B) [182]. In a phenotypic assay based on the quantification of nuclear r(CUG)^exp foci in DM1 patient‐derived cell lines by FISH and high‐content imaging, ChroA3 appeared as one of the few compounds that decreased the intensity of r(CUG)^exp foci without inducing significant toxicity. Furthermore, at a concentration as low as 40 nM, ChroA3 reduced the fraction of nuclear MBNL1 in DM1 cells to a level similar to that in non‐DM1 fibroblasts, suggesting a release of MBNL1 from the nuclear foci. In the splicing correction assay, ChroA3 corrected the missplicing of the INSR, but not of the ATP2A1 gene [183]. The detailed mechanism of action of ChroA3 in DM1 cells is unknown; however, considering its strong affinity to GC‐rich DNA loci, it is possible that its tight binding to d(CTG)·d(CAG) repeats in DNA leads to inhibition of transcription at these sites and, as a consequence, reduces the level of r(CUG)^exp. Alternatively, it is possible that ChroA3 directly interacts with the RNA transcript, leading to the stabilization of the hairpin form of r(CUG)^exp and the release of the sequestered MBNL1.

FIGURE 30.

(A) Chemical structure of chromomycin A3 (ChroA3) and CWG‐cPIP. (B) X‐ray structure of [Mg²⁺(ChroA3)₂(H₂O)₂] complex bound to d(TTGGCCAA)₂ duplex (replotted from the PDB ID: 1MNV using UCSF Chimera). Two ChroA3 moieties are shown in orange and pink, Mg²⁺ in light blue, and coordinated water molecules as red balls; nucleobase colors: A red, G green, C yellow, T blue. (C) Molecular model of CWG‐cPIP bound to a fragment of d(CTG) repeat. Reproduced with permission [181]. Copyright 2023, Ikenoshita et al.

In 2023, Sugiyama, Shioda and coworkers described the cyclic pyrrole–imidazole polyamide CWG‐cPIP (Figure 30A) as a ligand specifically targeting CWG (W = T or A) repeats in DNA via binding in the minor groove and formation of specific hydrogen bonds with nucleobases (Figure 30C) [181]. In thermal denaturation experiments, CWG‐cPIP strongly stabilized the Watson–Crick duplex formed by d(CAG)·d(CTG) repeats (ΔT _m = 51.6°C) as well as the mismatched hairpin structures formed by individual d(CAG)₁₀ and d(CTG)₁₀ strands (ΔT _m = 45.1°C and 41.4°C, respectively); in contrast, it had no effect on the RNA analog, r(CUG)₁₀. In line with its strong DNA binding, CWG‐cPIP inhibited the synthesis of r(CUG)₇₃ RNA from the DNA template in the in vitro transcription assay, whereas the transcription of the shorter sequence, r(CUG)₁₀, was much less affected. Reduction of the pathogenic r(CUG)^exp level and a decrease of the number of ribonuclear r(CUG)^exp foci were also observed in Neuro‐2a mouse neuroblasts (transfected with the corresponding plasmids) as well as in DM1 patient‐derived fibroblasts, both treated with 1 μM CWG‐cPIP. The in vivo effects of CWG‐cPIP were assessed in a murine model expressing in the hippocampus 300 CUG repeats, delivered via gene transfer using adeno‐associated virus 9 (AAV9). In this model, intracerebral administration of CWG‐cPIP (83 μg/kg) reduced the accumulation of nuclear r(CUG)^exp foci and MBNL1 sequestration in the hippocampal tissue, improved the memory‐related cognitive deficit, and partly restored the neuronal dysfunction associated with overexpression of CTG repeats. According to RNA‐Seq analysis, the treatment led to the recovery of 63% of the top differentially spliced genes, compared with control animals expressing only 10 CUG repeats. These results demonstrate that CWG‐cPIP is a strong therapeutic candidate for DM1, despite its drawback related to the poor blood–brain barrier permeability and requiring intracerebral administration.

2.7.3. Other Small‐Molecule Ligands

Several aromatic benzamides, including 119.2–5 (Figure 31A), were identified by Estrada–Tejedor and coworkers through in silico screening based on scaffold analysis, similarity searching with respect to the pentamidine pharmacophore (cf. Section 2.4.1) and druggability analysis [119]. Compound 119.2–5 binds to r(CUG)₂₃ in vitro with ∼3‐fold higher affinity than pentamidine (as per fluorescence anisotropy measurements; the exact K _d value has not been determined) and, at a concentration of 40 μM, releases MBNL1 from ribonuclear foci in human DM1 myoblasts, despite an ∼1.7‐fold increase in the average number of foci per cell. Administered at a concentration of 40 μM, it significantly alleviated locomotion defects in the Drosophila DM1 model.

FIGURE 31.

Miscellaneous r(CUG)^exp ligands. (A) Aromatic benzamides. (B) Plant‐derived alkaloids. (C) Flavonoids. (D) Azacryptand TrisA. (E) Quinazoline derivative VLT037.

In a related approach, two other benzamides, 142.p1 and 142.p7 (Figure 31A), were identified upon the pharmacophore‐search screening of a ∼4.3 million compound library, followed by in vitro confirmation in the r(CUG)₁₂–MBNL1 inhibition assay [142]. Both compounds were moderately active in vitro (IC₅₀ > 50 μM, HTRF), yet increased the nuclear export and expression of a luciferase mRNA containing 800 CUG repeats in its 3′‐UTR in a cellular assay, indicating their capacity to disrupt r(CUG)^exp foci. In line, both compounds, at a concentration of 10 μM, improved the splicing pattern of the TNNT2 minigene in the HeLa cellular DM1 model. Direct binding of 142.p7 to r(CUG) and selectivity over a fully paired RNA duplex, tRNA and DNA competitors were confirmed via fluorimetric titrations (K _d = 4.6 μM for the RNA duplex presenting a single 5′‐CUG‐3′/5′‐CUG‐3′ site).

Alkaloids berberine, coralyne and harmine (Figure 31B) were identified upon screening of natural products, plant and fungi extracts with respect to their capacity to inhibit the r(CUG)₇₈–MBNL1 interaction in an ELISA‐like assay [184]. Although all three alkaloids are known to interact with a variety of DNA and RNA structures, their biological activity was further assessed in cellular and animal DM1 models. In human DM1 myoblasts, berberine, at concentrations of 20–80 μM, rescued the MBNL1‐dependent missplicing of the TNNT2, but not of the INSR gene, whose missplicing was further worsened by berberine. Coralyne, despite its higher inhibitory activity with respect to the r(CUG)^exp–MBNL1 interaction in vitro, was inactive in cellular models. In contrast, harmine, despite its lower in vitro activity, partially rescued missplicing of both TNNT2 and INSR genes, without affecting the splicing of the genes not regulated by MBNL1. Both berberine and harmine induced significant cytotoxicity in mouse myoblasts (IC₅₀ = 212 and 123 μM, respectively). Treatment with 80 μM harmine significantly reduced the number of ribonuclear r(CUG)^exp foci in human DM1 myoblasts and increased the total level of MBNL1 protein (through a mechanism presumably not related to r(CUG)^exp binding), while berberine was inactive in this regard. In the HSA^LR mouse DM1 model, berberine was not active at sub‐toxic doses, whereas harmine partially rescued the missplicing of the Atp2a1 (Serca1) gene at a high dose (40 mg/kg), accompanied by secondary effects. The discrepancy between the results obtained in vitro, in cellular and in animal models, as well as the toxicity issues are likely related to off‐target effects of these promiscuous RNA and DNA binders and illustrate the challenges in molecular recognition of r(CUG)^exp.

The flavonol quercetin (Figure 31C), belonging to the large family of polyphenol natural products, was identified by Reddy, Berglund and coworkers in a cellular screen of 390 natural products with respect to their capacity to decrease the level of r(CUG)₄₈₀ transcript in HeLa cells [185]. The authors proposed that quercetin inhibits the transcription of r(CUG)^exp by interacting with the double‐stranded DNA template. This was supported by fluorescent intercalator displacement experiments that demonstrated direct interaction of quercetin with a short DNA duplex containing two d(CTG)·d(CAG) repeats (K _d = 1.41 μM), in line with the known capacity of quercetin to intercalate into double‐stranded DNA [186]. However, this mechanism has not been confirmed through an in vitro transcription assay, and possible interactions of quercetin with hairpin structures formed in the template (i.e., d(CAG)^exp) or the coding strand (i.e., d(CTG)^exp) have not been explored. Irrespective of its precise mechanism of r(CUG)^exp transcription inhibition, quercetin induced significant, dose‐dependent rescue of missplicing of the INSR and FLNB genes in DM1 patient‐derived fibroblasts, with a ∼100% rescue of both splicing events at a concentration of 64 μM and negligible toxicity; similar effects were observed in patient‐derived myotubes. In addition, treatment with quercetin decreased the expression of DMPK in DM1 patient‐derived cells, in line with the proposed transcription inhibition mechanism. In the murine HSA^LR DM1 model, a water‐soluble derivative of quercetin, enzymatically modified isoquercitrin (EMIQ, Figure 31C), administered in drinking water at 15 g/L, led to a significant reduction of the level of the HSA mRNA containing 220 CUG repeats, and rescued missplicing of the Clcn1, Atp2a1, and several other genes contributing to DM1 pathology. Moreover, mice treated with EMIQ for 6–12 weeks showed a reduction of myotonia phenotype from grade 3 to 1.5–2, demonstrating the therapeutic potential of this treatment. It should be noted that, despite being generally recognized as safe, quercetin possesses mutagenic activity, which may hamper its clinical use [187, 188].

Tris‐acridine azacryptand TrisA (Figure 31D) was identified by Mergny and coworkers by using FRET melting assay, in which thermal denaturation of fluorophore‐labeled oligonucleotides is detected by monitoring the fluorescence signal [189]. In this assay, several polyazamacrocyclic compounds strongly stabilized the hairpin form of d(CTG)₇; however, most compounds displayed limited selectivity, as their stabilizing effect vanished in the presence of a double‐stranded DNA competitor. TrisA appeared as the most selective ligand, as its interaction with d(CTG)₇ was only marginally affected in the presence of competitor DNA (ΔT _m = 17.5°C and 12.5°C in the absence and in the presence of 50 equiv. of unlabeled 26‐mer duplex). Moreover, TrisA stabilized the RNA counterpart, r(CUG)₈ (ΔT _m = 16.8°C), suggesting a potential for further development. Unfortunately, the capacity of this ligand to disrupt the r(CUG)^exp–MBNL1 interaction or its activity in DM1 models have not been investigated.

Quinazoline derivative VLT037 (Figure 31E) belongs to the series of 30 compounds identified in the high‐throughput in vivo screen performed in transgenic “spliceosensor” Drosophila flies, in which the expression of splicing variants of the human INSR gene was coupled to the expression of the luciferase reporter; the structures of other hits have not been disclosed [190]. VLT037 binds to r(CUG)₂₃, as demonstrated by fluorescence polarization experiments with a fluorescein‐labeled oligonucleotide; however, its affinity or the precise binding mode have not been determined. Despite its structural similarity to pyrido[2,3‐d]pyrimidin‐7‐(8H)‐one units in 119.1‐3118.1‐3 and 119.1a (cf. Figure 14E), the chemical structure of VLT037 does not allow efficient hydrogen bonding with uracil residues in r(CUG)^exp. Of note, several structurally close quinazoline derivatives have been described as ligands targeting r(CCUG)^exp RNA repeats in DM2 [191].

2.8. Peptides and Proteins

2.8.1. Dimeric Peptides

Miller and coworkers designed a resin‐bound dynamic combinatorial chemistry (RB‐DCC) approach, allowing DNA or RNA template‐guided selection of ligands from DCLs of disulfide‐linked dimeric tetrapeptides whose design was inspired by the structure of echinomycin (cf. Figure 28C) [174, 176]. Using an RNA oligonucleotide presenting 10 CUG repeats as a template, they identified four dimeric peptides (175.3‐3, 175.4‐4, 175.2‐4, and 175.3‐4, Figure 32A) from an 11,325‐member resin‐bound library [175]. All four dimers feature similarly high affinity to r(CUG)₁₀, as evidenced by the filter binding assay (FBA, K _d = 4.1–6.7 μM) and fluorescence titrations with fluorescein‐labeled oligonucleotides (K _d = 1.4–2.2 μM); the affinity to well‐paired RNA hairpins was several times lower. In line, all dimers inhibited the r(CUG)₁₀₉–MBNL1 interaction in the enzyme fragment complementation assay with a surface‐bound RNA, with K _I values in the same range as ligands’ affinity to the RNA target (cf. Figure 32A). The inhibitory activity was only slightly decreased in the presence of tRNA as a competitor, evidencing high selectivity of these ligands for binding to r(CUG)^exp.

FIGURE 32.

(A) Structures of disulfide‐linked peptide ligands of r(CUG)^exp identified using RB‐DCC approach by Miller and coworkers. (B) Structures of second‐generation dimeric peptides. The K _d values refer to the values determined by the filter binding assay (FBA) or surface plasmon resonance (SPR).

Since the disulfide bridge linking the two peptide moieties in the ligands presented in Figure 32A is incompatible with the reducing cellular environment, Miller and coworkers designed a second generation of dimeric peptide ligands, in which the quinoline residues were replaced with larger benzo[g]quinoline groups (potentially allowing better RNA binding) and the disulfide group with an isostere olefin (192.4 and 192.5) or a longer hydrocarbon linker (192.10 and 192.11, Figure 32B) [192]. Indeed, SPR measurements and fluorescence titrations demonstrated that these ligands were much stronger r(CUG)₁₀ binders, with 192.11 featuring the highest affinity (K _d = 22.5 nM, SPR) and selectivity. Of note, the monomeric analog of the dimeric peptides 192.4/192.5 did not bind to r(CUG)₁₀, demonstrating the importance of the dimeric scaffold. Likewise, the analogs of 192.4/192.5 devoid of benzo[g]quinoline groups showed a strongly reduced affinity and a lack of selectivity for r(CUG)₁₀, highlighting the critical contribution of aromatic residues to RNA binding. Compounds 192.4, 192.5, and 192.11 were cell‐permeable (as evidenced by the fluorescence of benzo[g]quinoline groups) and not toxic at concentrations up to 100 μM. In mouse myoblasts expressing 800 CUG repeats in the 3′‐UTR of luciferase reporter, treatment with 192.4, 192.10, and 192.11 (1 to 100 μM) resulted in a concentration‐dependent increase in luciferase activity, indicating the ligand‐induced disruption of r(CUG)^exp–MBNL1 foci, release, and translation of the luciferase transcript. Importantly, ligand 192.11 was the most active at low concentrations (1–10 μM), in agreement with its higher in vitro affinity to r(CUG)₁₀. Finally, in the murine HSA^LR DM1 model, 192.4 and 192.11 administered at a dose of 40 mg/kg via i.p. injections modestly, yet statistically significantly rescued the MBNL1‐dependent missplicing of Clca1 and Atp2a1, demonstrating in vivo efficacy of these dimeric peptides.

2.8.2. D‐Peptides

In 2011, Artero and coworkers established a Drosophila DM1 model for in vivo screening of molecules suppressing the DM1‐characteristic lethal phenotype. Screening of a synthetic combinatorial library of d‐hexapeptides (containing ∼2.5 million individual hexapeptides in 120 peptide mixtures), administered in the nutritive medium, led to the identification of d‐hexapeptide ABP1 (Ac‐d{PPYAWE}‐NH₂, Figure 33A), which reduced the DM1‐characteristic semi‐lethality in a dose‐dependent manner [194]. Alanine scanning confirmed that all amino acid residues of ABP1 were required for its activity. In line, transgenic expression of the natural l‐amino acid analog of ABP1 with a reversed sequence (mimicking the spatial arrangement of side chains in ABP1), but not the one with the direct ABP1 sequence (leading to a spatially different arrangement of side chains), also suppressed the DM1‐characteristic phenotype, confirming that the spatial arrangement of ABP1 side chains was crucial for its activity. Moreover, d‐amino acid building blocks of ABP1 confer resistance to proteases in vivo, enabling high stability and long‐lasting biological effects.

FIGURE 33.

(A) Structure of d‐hexapeptide ABP1 and its analogs 193.79–193.82; the common structural units are highlighted in gray. (B) Molecular model of d‐hexapeptide 193.79 bound to a fragment of r(CUG)₁₆ hairpin. Reproduced with permission [193]. Copyright 2021, A. Rapisadra et al.

Binding of ABP1 to r(CUG)^exp in vitro was evidenced by EMSA and fluorescence polarization experiments. The latter showed that ABP1 was unable to displace MBNL1 from its complex with FAM‐r(CUG)₂₃ in vitro, suggesting that the peptide and MBNL1 may have different binding modes with respect to r(CUG)^exp. Furthermore, CD spectroscopy and fluorescence measurements with 2‐aminopurine‐modified RNA suggested that high concentrations of ABP1 induce unfolding of the hairpin structure formed by r(CUG)^exp.

Studies in Drosophila DM1 flies demonstrated that treatment with ABP1 did not reduce the level of the r(CUG)^exp transcript, but strongly decreased the number of ribonuclear r(CUG)^exp foci and released the sequestered Mbl protein to the cytoplasm, indicating that, contrary to in vitro conditions, the peptide disrupts the r(CUG)^exp–Mbl interaction in vivo. In the murine HSA^LR DM1 model ABP1, administered at doses of 0.5 or 10 μg via a single intramuscular injection, showed a significant reduction of histopathological DM1 markers one month post‐injection. In addition, mice treated with 10 μg ABP1 revealed a significant rescue of splicing defects of Atp2a1 (Serca1, by 38% one month post‐injection) and Tnnt3, without affecting Mbnl1‐independent splicing [194].

In a follow‐up study, Bargiela, Artero, and coworkers investigated four other d‐peptides (193.79 to 193.82, Figure 33A), identified along with APB1 from the combinatorial library of d‐hexapeptides [193]. These peptides share with ABP1 three out of the six d‐amino acids and follow the consensus sequence d{CPY(A/T)(Q/W)E}, strongly suggesting a structure–function relationship. In vitro, these peptides had essentially no effect on the thermal stability of the r(CUG)₂₃ hairpin at concentrations up to 100 μM, as assessed by differential scanning fluorimetry. However, fluorescent indicator displacement (FID) titrations with Thiazole Orange probe indicated that the peptides bind to r(CUG)₂₃ with K _d values between 5.3 and 24 μM, with 193.79 being the strongest binder (Figure 33A). In all cases, the complete displacement of the probe could not be achieved, suggesting the formation of ternary complexes. Unfortunately, the parent peptide ABP1, for which a destabilization of the hairpin structure has been proposed (vide supra), has not been included in these assays. Molecular docking suggested that the peptides interact with the double‐stranded form of r(CUG)^exp by binding in the major groove; among the four peptides, 193.79 and 193.82 made contacts with both RNA strands, while 193.81 and 193.82 interacted with only one strand. Only 193.79 was able to make contacts with both mispaired uracil residues (Figure 33B). In all cases, aromatic residues of d‐tyrosine and d‐tryptophane were exposed to the solvent and did not participate in π‐stacking interactions that could lead to additional stabilization of the complex; however, this may represent a bias of the molecular docking approach that does not take into account the intrinsic flexibility of RNA.

In cultured human DM1 myotubes, all four peptides, at a concentration of 10 μM, partially rescued the DM1‐characteristic splicing defects in the TNNT2, DMD and SPTAN1 genes, without inducing toxic effects at concentrations up to 100 μM. In addition, treatment with each of the peptides reduced the number of r(CUG)^exp foci in myotubes and increased the fraction of foci‐free cells, without altering the expression level of r(CUG)^exp‐containing DMPK transcripts. This confirms the putative mechanism of action of these peptides via binding to r(CUG)^exp and disruption of the r(CUG)^exp–MBNL1 interaction. Moreover, all four peptides doubled the MBNL1 expression level, although only 193.80 and 193.81 increased the amount of MBNL1 at the protein level. In a Drosophila DM1 model, flies fed with the peptides 193.80, 193.81, and 193.82 (administered at 10 μM in the nutritive medium) demonstrated a significant increase in climbing speed. All four peptides increased the percentage of flies capable of flying, without increasing the height of the landing distance indicative of flight capabilities. Taken together, these results indicate a partial recovery of the muscular function in DM1 flies.

2.8.3. r(CUG)^exp‐Binding Proteins

As an extension of the approach based on r(CUG)^exp‐binding peptides, an emerging strategy to target r(CUG)^exp and correct the DM1‐associated splicing defects relies on engineered RNA‐binding proteins that can be expressed directly in the cells of DM1 patients via gene therapy. These artificial proteins act as decoys by competing with endogenous MBNL1 for binding to r(CUG)^exp, thereby releasing functional MBNL1 and restoring its splicing activity.

In 2022, Furling and coworkers designed a decoy protein derived from the native human MBNL1 [43]. This synthetic protein, called MBNL1Δ, contains two tandem ZnF domains required for r(CUG)^exp binding (cf. Section 1.3), but lacks the C‐terminal domain responsible for the splicing activity and protein oligomerization (Figure 34A). MBNL1Δ binds r(CUG)^exp with only slightly lower affinity than MBNL1 (K _d = 1159 vs. 858 nM, EMSA) and competes with MBNL1 in an in vitro assay. Expression of MBNL1Δ in DM1 patient‐derived muscle cells using a lentiviral vector resulted in normalization of DM1‐associated missplicing events (with a mean level of splicing correction of 72%), consistent with the release of endogenous MBNL1 and restoration of its splicing function. In addition, expression of MBNLΔ led to a strong decrease in the number and intensity of r(CUG)^exp RNA foci; although the aggregates were still observed, they were characterized by higher dynamics due to the reduced self‐aggregation of MBNL1Δ. In the HSA^LR mouse DM1 model, treatment with a GFP‐tagged MBNL1Δ packaged into an adeno‐associated virus 9 (AAV9) vector and administered via intramuscular injections resulted in a long‐lasting (over 1 year), partial correction of splicing defects, including missplicing of Clcn1, Atp2a1, and Mbnl1, and an improvement of the muscle relaxation time as a DM1‐characteristic phenotype. The treatment was also effective upon systemic administration of MBNL1Δ‐AAV9 via retro‐orbital sinus injection, without inducing an adverse immune response. Of note, MBNL1Δ was expressed in muscles at a level similar to the endogenous MBNL1 (which has not been modified by the treatment), which was sufficient to release the endogenous protein from r(CUG)^exp foci and restore its splicing activity.

FIGURE 34.

(A) Schematic representation of MBNL1 and MBNLΔ proteins and their K _d values to (CUG)₄ RNA (EMSA). (B) Schematic representation of 18‐mer PPR proteins binding to a (CUG)₆ RNA and the amino acid sequences of 35‐AA PPR motifs recognizing C, U, and G residues. Adapted with permission [196]. Copyright 2025, T. Imai et al.

Another approach to engineer r(CUG)^exp‐binding proteins was explored more recently by Nakamori and colleagues, who harnessed the sequence‐specific RNA‐binding properties of pentatricopeptide repeat (PPR) proteins. PPR proteins, commonly found in plants, are composed of tandem 35‐aminoacid sequence motifs, with each motif specifically binding a given RNA nucleotide. The capacity of PPR proteins to sequence‐specifically bind single‐stranded RNA and thereby regulate gene expression makes them valuable tools for synthetic biology [196]. Nakamori and coworkers designed and produced four synthetic r(CUG)^exp‐binding PPR proteins (CUG‐PPRs), each containing 18 PPR motifs organized into six blocks of three motifs specific for cytosine, uracil, and guanine, respectively, and therefore enabling the recognition of the (CUG)₆ RNA sequence (Figure 34B) [195]. Among the four variants, CUG‐PPR1 and CUG‐PPR2 showed the highest affinity to r(CUG)₆ in vitro (K _d = 65 and 48 nm, respectively, ELISA), without significant binding to the d(CTG)₆ counterpart, r(CAG)₆, or double‐stranded DNA or RNA repeats. Both CUG‐PPRs also interacted with longer CUG repeats forming hairpin structures, even though the binding to single‐stranded substrates was stronger (as is also the case for MBNL1, cf. Section 1.3). The ability of CUG‐PPRs to bind to expanded CUG repeats was verified by immunoprecipitation experiments in HEK‐293T cells expressing CUG‐PPR1 and a DT960 plasmid containing 960 interrupted CTG repeats, demonstrating an almost 5‐fold higher level of CUG‐PPR1 bound to the DT960 transcript compared with normal endogenous DMPK mRNA. In the C2C12 murine myoblast DM1 model expressing 900 CUG repeats in the 3′‐UTR of DMPK, the transient expression of CUG‐PPR strongly decreased the number of cells presenting r(CUG)^exp foci and reduced nuclear sequestration of MBNL1. In parallel, a rescue of missplicing of Atp2a1 and a reduction of myotube formation defects were observed, indicating a reversion of the r(CUG)^exp‐related toxicity; similar effects were observed in DM1 patient‐derived myoblasts. For in vivo studies, CUG‐PPR1 and CUG‐PPR2 were embedded into the AAV6 vector and administered to HSA^LR DM1 mice via intramuscular injections. With this local treatment, both CUG‐PPR1 and CUG‐PPR2 corrected the splicing of Atp2a1 and Clcn1 and improved the myotonia score from 2 to 1 in a part of the treated mice. Systemic delivery of CUG‐PPR1 was performed using an AAV9 vector, administered via tail vein injection. The systemic treatment allowed the expression of CUG‐PPR1 not only in the muscles but also in other tissues, including the liver, spinal cord, kidney, lung, eye, brain, spleen, and testis, with minimal immune response and no signs of toxicity. Mice treated by systemic administration of AAV9‐CUG‐PPR1 demonstrated a global transcriptome modulation (including a rescue of Atp2a1 and Clcn1 missplicing) and an improvement of myotonia, with the effects lasting for at least 4 months post‐injection. These promising preclinical results demonstrate the potential of engineered CUG‐PPR proteins for future gene therapy‐based approaches to DM1 treatment.

2.9. r(CUG)^exp–Targeting Oligonucleotides

2.9.1. Antisense Oligonucleotides

Beyond small molecules, peptides and proteins, antisense oligonucleotides (ASOs) represent an important class of “ligands” suitable for targeting r(CUG)^exp with therapeutic purposes. ASOs are short (15–30 bases), single‐stranded, synthetic oligonucleotides complementary to the target RNA sequence, which are typically chemically modified to improve their stability and specificity. ASOs can be divided into two classes, depending on their mechanism of action. Blocking (or steric‐blocking) ASOs bind to their target RNA with high affinity and thereby interfere with RNA–protein interactions, such as MBNL1 sequestration by r(CUG)^exp in the DM1 context (Figure 35A). The mechanism of these ASOs is thus similar to that of small‐molecule ligands that stabilize the hairpin form of r(CUG)^exp characterized by reduced MBNL1 binding. In contrast, RNase H1‐dependent ASOs hybridize to their target RNA with the formation of a hybrid DNA–RNA duplex, which is recognized and cleaved by RNase H1 (Figure 35B). This mechanism is therefore reminiscent of r(CUG)^exp‐targeting “RNA degraders” (cf. Section 2.3.2). RNase H1‐dependent ASOs are also termed “gapmers” since they consist of a central DNA part complementary to the target RNA sequence, flanked on both sides by short runs of chemically modified nucleotides, which increase the affinity to the target and prevent degradation by nucleases. Both types of ASOs have been exploited for targeting of r(CUG)^exp, and their use in the DM1 context has been extensively reviewed [197, 198, 199]. Only the most significant milestones are described below, with an emphasis on the chemical design of r(CUG)^exp–targeting ASOs.

FIGURE 35.

(A,B) Mechanism of action of (A) blocking ASOs and (B) RNase H1‐dependent ASOs (gapmers) in the context of DM1 pathology. (C) Chemical structures of selected r(CUG)^exp‐blocking ASOs. (D) Chemical structure of the gapmer MOE‐CAG14.

Thornton and coworkers pioneered the use of blocking ASOs as a potential DM1 therapy back in 2009 [200]. A 25‐mer phosphorodiamidate morpholino oligonucleotide (PMO) CAG25, complementary to r(CUG)^exp (Figure 35C), was shown to invade r(CUG)^exp and form a stable RNA–PMO heteroduplex in vitro, preventing the formation of r(CUG)^exp–MBNL1 complexes. In parallel, Wansink and coworkers, upon screening of 13 ASOs complementary to various regions of the mutated CUG‐expanded DMPK transcript, identified the 21‐mer 2′‐OMe phosphorothioate (PS) RNA oligonucleotide PS58 containing seven CAG repeats as an alternative blocking ASO (Figure 35C) [201]. CAG25 and PS58 were studied in murine DM1 models, using a delivery via intramuscular injection followed by in vivo electroporation. Both ASOs reduced nuclear r(CUG)^exp–MBNL1 foci in the muscular tissue, normalized the DM1‐affected missplicing, and reduced myotonia in mice, with the effects, in the case of CAG25, appearing at 3 weeks and lasting up to 14 weeks post‐injection [200]. In addition, both ASOs strongly reduced the level of toxic r(CUG)^exp (by ∼50% for CAG25 and ∼80% for PS58), presumably by accelerating its degradation, despite the fact that neither PMOs nor 2′‐OMe PS ASOs induce RNase H cleavage of their target RNA. In both cases, the effects were specific for the mutant r(CUG)^exp‐DMPK, likely due to the higher ASO‐binding capacity of expanded CUG repeats.

More recently, Sobczak and coworkers introduced much shorter, 8‐ to 10‐mer locked nucleic acid (LNA)‐based ASOs such as LNA‐CAG‐10 (Figure 35C) for the same purpose [202]. This oligonucleotide binds to r(CUG)₁₀₀ with a K _d value of 7 nM in vitro (EMSA) and inhibits the r(CUG)₁₀₀–MBNL1 interaction with K _I ≈ 10 nM (FBA). In DM1 patient‐derived fibroblasts, LNA‐CAG‐10 rescued the missplicing of MBNL1 and MBNL2 pre‐mRNA, with full correction to the normal level observed at 125 nM. Notably, a shorter 8‐mer analog, LNA‐CAG‐8, had even higher affinity invitro but was less active with respect to splicing correction in the cellular assay. In contrast to CAG25 and PS58 described above, LNA‐CAG‐10 only slightly decreased the level of r(CUG)^exp, but strongly reduced the number and the size of nuclear r(CUG)^exp foci. In the murine HSA^LR DM1 model LNA‐CAG‐10, delivered via intramuscular injection followed by electroporation, improved the DM1‐characteristic missplicing pattern and myotonia grade (from 3 to <1), with the effects lasting for at least 14 days.

In contrast to blocking ASOs that need to be delivered to the cells at relatively high doses to ensure efficient coverage of r(CUG)^exp, RNase H1‐dependent ASOs that induce RNA cleavage without their own degradation can be used in catalytic amounts. Cooper and coworkers designed a series of gapmers complementary to r(CUG)^exp, in which the central 8‐nt DNA‐PS sequence was flanked with two LNA‐PS or 2′‐MOE‐PS triplets, such as MOE‐CAG14 (Figure 35D) [203]. In COSM6 cells expressing DMPK mRNA with 960 CUG repeats, MOE‐CAG14, at a concentration as low as 50 nM, induced an ∼80% reduction of the DMPK transcript level, with the LNA‐PS gapmer being slightly less efficient. Moreover, the effect was specific for the long CUG repeats, as the transcripts containing only 12 CUG repeats were not affected. In the murine EpA960/HAS‐Cre DM1 model, treatment with 2 μg MOE‐CAG14, delivered via intramuscular injection and in vivo electroporation, decreased the level of the r(CUG)‐containing transcript (by ∼50%) and the number of r(CUG)^exp foci, and partially rescued the MBNL1‐dependent missplicing. Unfortunately, this treatment led to inflammatory muscular damage, regardless of the gapmer sequence. Intriguingly, a combination of MOE‐CAG14 and the blocking morpholino ASO CAG25 showed a synergy with respect to the reduction of the r(CUG)^exp level, suggesting that the two ASO strategies can be combined for an optimal efficacy [203]. Another study found that LNA‐ and analogous bridged nucleic acid (BNA^NC)‐based gapmers targeting the DMPK transcript outside the r(CUG)^exp repeats were more efficient in terms of mutant DMPK knockdown and splicing correction than gapmers directly targeting the CUG repeats, presumably due to steric hindrance caused by the MBNL1 binding and/or formation of secondary structures within the r(CUG)^exp repeats [204]. However, a direct comparison of blocking and RNase H1‐recruiting ASOs sharing the same sequence and oligonucleotide chemistry (2′‐OMe PS, except for the central DNA part in the gapmer) in DM1 patient‐derived myoblasts demonstrated that, despite similar levels of DMPK knockdown, the r(CUG)^exp‐targeted blocking ASO was more efficient than either r(CUG)^exp‐targeted or non‐r(CUG)^exp DMPK‐targeted gapmers in terms of dispersion of MBNL1 from ribonuclear foci and correction of DM1‐associated splicing defects [205]. In addition, while the r(CUG)^exp‐targeted gapmer led to downregulation of not only DMPK but also other CUG repeat‐containing transcripts, the blocking, r(CUG)^exp‐targeted ASO induced less off‐target effects on the transcriptome level. Thus, blocking ASOs appear to be safer and more suitable for pre‐clinical development in the DM1 context.

Despite the promising results obtained in preclinical models, the major hurdle for the use of ASO‐based therapies is their poor cellular uptake, which hampers systemic administration and requires harsh delivery methods such as in vivo electroporation. The conjugates of ASOs with molecular transporters have been developed to overcome this challenge [198, 199]. Thus, a conjugate of the blocking ASO CAG25 with the arginine‐rich cell‐penetrating peptide K was shown to be active upon systemic (intravenous) delivery at 30 mg/kg in the murine HSA^LR DM1 model, resulting in a reduction of DM1 biomarkers and a full relief of myotonia [206]. Likewise, a conjugate of the (CAG)₇ PMO with the cationic cell‐penetrating peptide Pip6a was active upon intravenous administration at 12.5 mg/kg in DM1 mice, inducing long‐lasting (up to 6 months) correction of molecular and functional phenotypes, while the corresponding “naked” ASO was inactive even at a 16‐fold higher dose [207]. These examples strongly support further clinical development of peptide‐conjugated, r(CUG)^exp‐targeted blocking ASOs.

Finally, aside from ASO‐based approaches, recognition of r(CUG)^exp by a complementary RNA strand has been exploited in other strategies aimed at the degradation of r(CUG)^exp and the release of sequestered MBNL1. Along these lines, the siRNA approach is based on the use of (CAG)·(CUG) (‘siCAG’) RNA duplex, in which the 21‐nt CAG strand hybridizes to r(CUG)^exp once delivered into cells, causing its cleavage by the Argonaute (AGO2) protein [208]. In the HSA^LR mouse DM1 model, treatment with siCAG decreased the level of the repeat‐expanded transcript by >75% and restored the MBNL1‐regulated splicing at a 10‐fold lower dose than PMO or 2′‐OMe ASOs; however, the effect waned after 14 days. Furthermore, recognition of r(CUG)^exp by complementary single guide RNA has been exploited in repurposed CRISPR systems, in which the toxic r(CUG)^exp is eliminated by RNA‐targeting Cas9 (RCas9) [209, 210] or Cas13 nucleases [211]. These gene therapy approaches are beyond the scope of this review.

2.9.2. Peptide Nucleic Acid (PNA)‐Based Probes

Peptide nucleic acid (PNA) probes, which mimic nucleic acids with a charge‐neutral peptide backbone, offer a promising avenue for targeting r(CUG)^exp by combining several features of small molecules, peptide ligands, and antisense oligonucleotides. PNAs hybridize to their target RNA and DNA sequences with very high affinity and sequence specificity, enabling the use of rather short PNA sequences; however, their applications are hampered by poor solubility. A structure of the 8‐mer RNA strand containing two CUG repeats and hybridized to the complementary PNA strand has been solved by X‐ray crystallography, demonstrating the formation of a heteroduplex resembling the A‐form of RNA, but with a lower twist (26° vs. 33° in A‐RNA) and rise (2.4 vs. 2.6–3.3 Å in A‐RNA) [212].

Along these lines, Ly and coworkers designed an approach for targeting r(CUG)^exp using relatively short (three to six bases) miniPEG‐γ peptide nucleic acid (MPγPNA) probes, containing diethylene glycol residues at the γ‐positions of the PNA backbone to improve aqueous solubility. The bis‐pyrene hexanucleotide MPγPNA probe 213.P4 (Figure 36A) was developed for binding and fluorimetric detection of r(CUG)^exp [213]. Upon hybridization to r(CUG) _n (n = 12 or 16), the terminal pyrene groups of the probe form excimers, whose red‐shifted fluorescence can be detected by the naked eye (Figure 36B). In addition, 213.P4 disrupts the r(CUG)₉₆–MBNL1 interaction in vitro at low‐micromolar concentrations, with the formation of r(CUG)₉₆–213.P4 heteroduplex detected by gel electrophoresis. The MPγPNA motif has also been used in the r(CUG)^exp template‐guided oligomerization approach exploiting native chemical ligation (NCL) [214]. In this approach, the trinucleotide MPγPNA probe 214.P2 (Figure 36C) was endowed with reactive thioester and cysteamine residues. After deprotection of the cysteamine residue of 214.P2, hybridization of the probe with the r(CUG)₁₂ RNA template enables the spontaneous NCL reaction of thioester and cysteamine groups leading to the formation of peptide oligomers, detected by MALDI mass spectrometry and strongly stabilizing r(CUG)₁₂ in thermal denaturation experiments (Figure 36D). These examples highlight the potential of short PNA‐based probes for detection and therapeutic targeting of r(CUG)^exp.

FIGURE 36.

(A) Chemical structure and (B) working principle of the bis‐pyrene MPγPNA probe 213.P4. (C) Chemical structure and (D) schematic depiction of the template‐guided NCL oligomerization of the triplet MPγPNA probe 214.P2 on the r(CUG)^exp template.

3. Conclusions and Perspectives

With its molecular pathogenesis now firmly established for more than two decades, DM1 became a playground for the design of diverse therapeutic strategies, predominantly based on DNA and RNA ligands targeting expanded CTG and CUG repeats, respectively. In this review, we exposed ‐‐various molecular design approaches toward d(CTG)^exp and r(CUG)^exp ligands rooted in medicinal chemistry, supramolecular chemistry, biochemistry, chemical and molecular biology, as well as their successes and limitations in the DM1 context. Despite the large number of potential drug candidates, many common patterns may be identified with regard to their biomolecular targets, mechanisms of action and molecular design features, as schematically shown in Figure 37 and summarized in Table 6.

FIGURE 37.

Schematic representation of therapeutic intervention points using ligands targeting CTG and CUG repeatsin the DM1 context. Note that d(CAG)^exp ligands are not covered in this review.

TABLE 6.

Non‐exhaustive classification of CTG and CUG ligands with respect to their biomolecular target(s) and proposed mechanism of action in DM1 models.

Biomolecular target	Molecular mechanism of ligand action	Examples
d(CTG)·d(CAG) duplex	Transcription inhibition; reduced synthesis of r(CUG)exp	Pentamidine and other aromatic diamidines; furamidine; ChroA3 a; CWG‐cPIP; quercetin;
d(CTG)exp hairpins	Transcription inhibition; reduced synthesis of r(CUG)exp	ActD, doxorubicin; DAP; PQA‐19; iQN; Z1; 88.3; A4D3; 82.1; 96.5; 100.9; D‐III; CWG‐cPIP; TrisA
	Oligomerization or self‐assembly on d(CTG)exp hairpins	110.2 + 110.3; A5/N16
	DNA damage; contraction of CTG repeats	Doxorubicina; 82.1
r(CUG)exp hairpins	Stabilization of the hairpin form; MBNL1 release	Daunorubicin; (di)lomofungin; DAP; DDAP; JM642; 119.1–3; Z1; 89.1; 90.9; PLG 50 ‐1/2; 94.3; 96.5; 97.2a; 98.4; 100.9; D‐III; D(8+II); 2H‐4 and other bis‐benzimidazole oligomers; 2H‐K4NMeS; 131.22; 2H‐2C2; H1; 136.2b; 140.1; 140.16–17; 142.p2; DEL1; pentamidine and other aromatic diamidinesa; furamidine; P1; 151.1–151.3; 3K‐4, 4K‐2 and other kanamycin oligomers; 4K‐2‐DR9; erythromycin; 178.2; 179.1a; 119.2–5; 142.p7; VLT037; 175.2‐4; 192.11; 193.79 and analogsa
	Oligomerization or self‐assembly on r(CUG)exp hairpins	2H‐K4NMeS‐Aak + N 3 ‐2H‐K4NMeS; N 3 ‐2H‐K4NMeS‐Aak (129.6); 139.7
	r(CUG)exp cleavage	100.9; 2H‐2‐CA;2H‐K4NMeS‐CA‐Biotin; Cugamycin; DeglycoCugamycin; 2H‐K2‐Pro‐Bleo; 131.22‐Bleo; DEL‐Bleo; 179.1b
	r(CUG)exp photooxidation or photocleavage	2H‐4‐HPT; 2H‐4‐Ru
Single‐stranded r(CUG)exp	MBNL1 release	ABP1 a;
		r(CUG)exp‐binding proteins: MBNL1Δ; CUG‐PPR1/2; ASOs: CAG25; PS58; LNA‐CAG‐10; 213.P4
	Oligomerization on single‐stranded r(CUG)exp	214.P2
	r(CUG)exp cleavage	ASO: MOE‐CAG14

^a Presumed or not firmly established molecular mechanism.

A distinctive feature of DM1 molecular pathogenesis, despite its apparent simplicity, is the variety of biomolecular targets that may be exploited as therapeutic intervention points by targeting them with small‐molecule ligands. These include not only the well‐established hairpin form of r(CUG)^exp RNA, but also the single‐stranded form of r(CUG)^exp, slipped d(CTG)^exp and d(CAG)^exp hairpins that are—at least transiently—formed at the CTG expansion loci, as well as d(CTG)·d(CAG) repeats in duplex DNA. Therefore, DM1 drug candidates covered in this review can be principally grouped as ligands acting at the DNA level and those acting at the RNA level (Figure 37). While the former typically reduce the expression level of toxic r(CUG)^exp via transcription inhibition at CTG repeats, ligands acting on the RNA level reduce MBNL1 sequestration either by directly competing with MBNL1 for binding to r(CUG)^exp (as in the case of r(CUG)^exp‐binding proteins and blocking ASOs), or—which is the case of most r(CUG)^exp ligands—by stabilizing the hairpin form, less prone to MBNL1 binding. However, considering the structural similarity of d(CTG)^exp and r(CUG)^exp hairpins (cf. Section 1.2), it is not unexpected that many ligands actually interact with both of these targets (Table 6). Although the selectivity of ligands toward one or another target can be tuned via careful molecular design, as was nicely illustrated within the series of triaminotriazine–acridine conjugates (Section 2.1.1) and bis‐benzimidazole oligomers (Section 2.3.1), this layered mechanism of action of such multitargeting ligands can actually be beneficial in the context of DM1 therapy due to the synergy between the reduction of the r(CUG)^exp transcript level and MBNL1 release.

Along these lines, a notable emerging trend is the rise of RNA‐ and DNA‐“degraders,” that is, ligands that not only bind, but also induce cleavage or chemical modification of r(CUG)^exp and d(CTG)^exp, respectively. At the RNA level, degradation of r(CUG)^exp, typically induced by bleomycin‐conjugated ligands (of which Cugamycin represents the most prominent example, cf. Section 2.3.2) and gapmers (cf. Section 2.9.1), allows irreversible disruption of the r(CUG)^exp–MBNL1 foci and prevents repeated MBNL1 sequestration. This is illustrated by the systematically higher biological activity of bleomycin derivatives compared with the parent r(CUG)^exp ligands (e.g., DEL1‐Bleo vs. DEL1, Figure 23). The biological activity of d(CTG)^exp degraders is actually even more intriguing. Although only two examples of such ligands have been described so far, namely doxorubicin and 82.1 [82, 177], in both cases, a prolonged treatment was shown to induce contraction of CTG repeats in at least a subset of cultured DM1 cells, allowing a reversion to the normal genotype. Of note, similar repeat contraction has been described for ligands binding to d(CAG)^exp hairpins [215, 216]. Further development and investigation of d(CTG)^exp degraders will be needed to demonstrate the feasibility of this approach in vivo.

Regarding molecular design strategies, an obvious feature of r(CUG)^exp is its highly repetitive pattern, propitious for the use of dimeric, oligomeric, or polymeric ligands containing multiple, regularly spaced RNA‐targeting units. This multimerization strategy has been most extensively exploited by the Disney lab, who applied it to the ligands based on bis‐benzimidazole derivatives (Section 2.3.1) and kanamycin oligomers (Section 2.5), resulting in ligands with impressive efficacy results in terms of both r(CUG)^exp affinity (e.g., 2H‐K4NMeS, K _d = 12 nM; 4K‐2, K _d = 4 nM) [129, 159] and biological activity, typically observed at sub‐micromolar or low‐micromolar concentrations. The dimerization approach has also been used in the design of several 1,3,5‐triaminotriazine‐based ligands (Section 2.1) as well as with small ligands that specifically recognize uracil residues in r(CUG)^exp, such as DDAP, 119.1‐3, and 114.1a (Section 2.2). The multimerization effect is also at play in r(CUG)^exp‐targeting ASOs that comprise multiple (3 to 8) CAG repeats (Section 2.9.1). However, no attempts to design ligands combining more than two small uracil‐recognizing units in one molecule have been described so far. Of note, this approach could enhance not only the affinity, but also the r(CUG)^exp selectivity of resulting oligomers, as in the case of bis‐benzimidazole oligomers where the multimerization drastically reduced the DNA affinity of oligomers compared to the monomer. Also, the multimerization approach has not yet been applied to d(CTG)^exp ligands designed to operate chiefly on the DNA level.

A disadvantage of the multimerization approach is the high molecular weight of the multimers, leading to their poor solubility, drug‐likeness, and cellular uptake. Two possible solutions to this issue have been showcased with r(CUG)^exp ligands. The first approach, which has been explored within the series of 1,3,5‐triaminotriazine derivatives, resides in the incorporation of r(CUG)^exp ligands into the structure of cationic, cell‐permeable polymers, as in the case of PLG ₅₀ ‐1/2 [91] and 98.4 [98]. Another approach, explored both by Zimmerman with 1,3,5‐triaminotriazine‐based ligands and by Disney with bis‐benzimidazole derivatives, relies on the self‐assembly of high‐affinity oligomers from small cell‐permeable monomers in DM1 cells directly on the r(CUG)^exp template. In the latter case, several reactions have been exploited for on‐target self‐assembly, including imine condensation [110, 111], Huisgen alkyne–azide cycloaddition [108, 129] and tetrazine ligation [139], with the latter being the most promising due to its high rate and biorthogonality (Sections 2.1.3 and 2.3.3). A related approach harnesses the NCL self‐assembly of the PNA probe 214.P2, even though this reaction has not been studied in DM1 cells so far [214]. Remarkably, the on‐target self‐assembly has not yet been explored with small, high‐affinity r(CUG)^exp ligands such as 178.2 or 179.1a (Section 2.7.1).

Last but not least, DM1 represents only one of about 30 hereditary repeat expansion diseases (REDs) [45, 217]. These include myotonic dystrophy type 2 (DM2) caused by the CCTG repeat expansion in the CNBP gene and whose molecular pathogenesis involves MBNL1 sequestration by r(CCUG)^exp repeats; Huntington's disease, associated with the expansion of CAG repeats in the HTT gene; fragile X syndrome, linked to the expansion of CGG repeats in the FMR1 gene; Friedrich's ataxia resulting from the expansion of GAA repeats in the FXN gene; many types of spinocerebellar ataxia linked to the expansion of CAG repeats in different genomic loci; and others. Due to its well‐characterized molecular mechanism, easily accessible biomarkers and the availability of robust cellular and in vivo disease models, DM1 attracted a massive research effort from the medicinal chemistry community. Therefore, the principles of ligand design and molecular mechanisms of potential drug candidates developed in the DM1 context, summarized in this review, may provide a framework and an inspiration for addressing other REDs with similar molecular pathogenesis.

Author Contributions

Camille Richagneux: formal analysis (supporting), visualization (supporting), writing – original draft (supporting), and writing – review and editing (supporting). Anton Granzhan: formal analysis (lead), investigation (lead), and project administration (lead), visualization (lead), writing – original draft (lead), and writing – review and editing (lead).

Funding

This work was supported by Agence Nationale de la Recherche (ANR‐22‐CE44‐0039).

Conflicts of Interest

The authors declare no conflicts of interest.