Cpy(a/t)(q/w)e D-Hexapeptides Bind CUG Repeats and Rescue Phenotypes of Myotonic Dystrophy Myoblasts and A Drosophila Model of the Disease


 In Myotonic Dystrophy type 1 (DM1), a non-coding CTG repeats rare expansion disease; toxic double-stranded RNA hairpins sequester the RNA-binding proteins Muscleblind-like 1 and 2 (MBNL1 and 2) and trigger other DM1-related pathogenesis pathway defects. In this paper, we characterize four D-amino acid hexapeptides identified together with abp1, a peptide previously shown to stabilize CUG RNA in its single-stranded conformation. With the generalized sequence cpy(a/t)(q/w)e, these related peptides improved three MBNL-regulated exon inclusions in DM1-derived cells. Subsequent experiments showed that these compounds generally increased the relative expression of MBNL1 and its nuclear-cytoplasmic distribution, reduced hyperactivated autophagy, and increased the percentage of differentiated (Desmin-positive) cells in vitro. All peptides rescued atrophy of indirect flight muscles in a Drosophila model of the disease, and partially rescued muscle function according to climbing and flight tests. Investigation of their mechanism of action supports that all four compounds can bind to CUG repeats with slightly different constant affinities, but binding did not strongly influence the secondary structure of the toxic RNA in contrast to abp1. Finally, molecular modeling suggests a detailed view of the interactions of peptide-CUG RNA complexes useful in the chemical optimization of compounds.


Introduction
Myotonic dystrophy type 1 (DM1; OMIM #160900) is a rare autosomal dominant disease with symptoms that typically affect the musculoskeletal system, with degenerative muscle atrophy and myotonia (or muscle hyperexcitability), heart conduction defects, and cognitive involvement, which combined with several other multisystemic alterations severely affect the life expectancy and quality of life of patients 1 .
DM1 originates from an abnormal expansion of an unstable CTG trinucleotide repeat in the 3'untranslated region of the DM1 protein kinase (DMPK) gene 2 . Expanded CTG repeats are transcribed but not translated and get retained in the cell nucleus where mutant transcripts accumulate, forming foci that sequester proteins of the Muscleblind-like (MBNL) family of alternative splicing regulators, among other molecular consequences 3 . The CUG expansions fold into metastable hairpin structures that facilitate the binding and sequestration of nuclear factors, among which MBNL1 is the most relevant in this work.
Speci cally, it has been shown that compound loss of Mbnl1 and Mbnl2 in mice reproduce several DM1 symptoms 4 . At the same time, Mbnl1 overexpression rescues DM1 phenotypes in a mouse model that expresses CTG repeats throughout the skeletal muscles 5 6 . Human MBNL1 proteins are tissue-speci c CCCH zinc nger factors with crucial roles in the regulation of alternative splicing and alternative polyadenylation during development, in which it promotes a switch from fetal to adult patterns in a wide number of transcripts 7 8 9 . Although lack of MBNL1 function is one of the main molecular hallmarks of DM1 myopathy, many additional molecular contributors have been reported 10 . Hyperactivated GSKbeta and autophagy have been proposed to contribute to muscle atrophy in DM1 by stabilizing a repressive form of CELF1 alternative splicing regulator in the nucleus and downregulation of miR-7, a master regulator of autophagy, respectively 11 12,13 . Although most therapeutic strategies have focussed on degrading the expanded CUG RNA or preventing MBNL sequestration by the toxic RNA, with small molecules or oligonucleotide-based approaches 14,15 , direct upregulation of endogenous MBNL1 levels is becoming accepted as a complementary approach 16 17 18 . Drosophila is one of the experimentation animals used to model DM1 by expressing non-coding CTG repeat expansions to the insect muscles, brain, and heart to reproduce critical DM1 molecular defects and test candidate therapeutics 19 20 21 . We previously targeted the expression of 480 interrupted CTG repeats to the Drosophila mushroom bodies, which are a pair of brain structures in insects. This generates a semilethal phenotype at the pupal stage used to screen a positional scanning synthetic combinatorial library of D-amino-acid hexapeptides that identi ed 16 candidate peptides, of which Abp1 (ppyawe) was characterized in more detail 22 . Our current study addressed the characterization of the remaining 15 peptides in a secondary screen in a DM1-derived cell model of the disease and identi ed four closely related peptides that improved cell and Drosophila DM1 phenotypes by directly binding to the CUG RNA.

Results
Four related peptides rescue MBNL-dependent missplicing events in DM1 myoblasts.
Immortalized human DM1 muscle cell lines display disease-associated molecular features such as nuclear RNA aggregates and alternative splicing defects 23 . DM1-related phenotypes can be used as readouts to screen candidate therapeutics in vitro for effects on RNA toxicity associated with the DM1 mutation. MBNL protein depletion explains most aberrant splicing patterns observed in DM1 24 25 26 . Thus, we established three splicing events typically altered in DM1 as screening criteria for the 15 Damino-acid hexapeptides previously identi ed (Supplementary Table 1) 22 : the inclusion of exon 5 of cardiac troponin T gene (cTNT; Entrez ID: 7139) and the exclusion of exon 78 of the dystrophin gene (DMD; Entrez ID: 1756), both MBNL1-dependent and the exclusion of exon 23 of spectrin alpha nonerythrocytic gene (SPTAN1; Entrez ID: 6709), which is MBNL2-dependent 27 . Immortalized control and DM1 broblasts were transdifferentiated into myoblasts for 48 h. After that, they were incubated two more days with 10 µM of each peptide dissolved in myoblast differentiation medium (MDM). Semiquantitative RT-PCR evaluated the activity of the peptides on the missplicing events. Despite most peptides being able to improve inclusion of at least one of the alternative exons, only peptides cpyaqe (79), cpyawe (80), cpytqw (81), and cpytwe (82) rescued all of them ( Fig. 1a-d). Notably, the four peptides shared 4 out of 6 amino acids, and only the fourth and the fth positions changed, generating the consensus sequence cpy(a/t)(q/w)e, which strongly suggests a structure-function relationship. Treatment with these peptides did not change the inclusion of exon 8 of the CAPZB gene, regulated by CELF1 28 , nor exon 19 of the DLG1 gene, which remains unchanged in DM1 patients 29 , suggesting a speci c effect on the regulatory factors MBNL1 and 2 in the disease ( Fig. 1e-g). Finally, we tested whether such activity was (CUG) exp -speci c or not; thus, we treated control cells with 10 µM of peptide 80 and quanti ed their activity on the alternative exons of cTNT and SPTAN. The peptide produced no signi cant change, suggesting that its activity depended on the presence of the mutation that causes DM1 (Fig. 1h-j).
Peptides 79, 80, 81, and 82 were selected for further evaluation, including toxicity assays in control myoblasts (Fig. 1k) and broblasts ( Supplementary Fig. 1). Peptides were not toxic even at 100 µM, which is ten times higher than the concentration at which they rescued missplicing.
Candidate peptides enhance MBNL expression and its normal distribution in the cell.
To understand what could trigger the rescue of the analyzed splicing events, we used cDNA from treated cells to perform quantitative PCR (qPCR) and detect any modi cation in the mRNA levels of MBNL1 and MBNL2. We found that peptides 79, 80, 81, and 82 doubled MBNL1 expression while peptides 79 and 81 slightly increased MBNL2 mRNA amounts (Fig. 2ab). This nding was encouraging as the increase in MBNL1 and 2 gene expression has been proposed as a valid strategy to improve the clinical outcome of DM1 5 6 30 16 18 19 . However, at the protein level, only cells treated with the peptides 80 and 81 showed a statistically signi cant increase in MBNL1 levels by western blot compared to untreated DM1 cells ( Fig. 2c, Supplementary Fig. 2). Finally, since the subcellular localization of MBNL1 is altered in DM1 19,30 , we evaluated this phenotype using an anti-MBNL1 antibody. Immuno uorescence images con rmed that the MBNL1 signal was increased in the cytoplasm of the treated cells compared to the untreated and approached normal intensity and subcellular distribution ( Fig. 2d-i). Taken together, these results indicate that the candidate peptides were able to target the upregulation of MBNL1. Therefore, we studied additional molecular phenotypes related to MBNL1 in the pathogenesis pathway.
A 48-h treatment is su cient to reduce the number of foci with no effect on the DMPK transcripts.
A prime mechanism to enhance functional MBNL1 levels in cells is to prevent its sequestering into ribonuclear foci by either blocking its binding to CUG repeats or changing the secondary structure of the toxic RNA so it is less prone to bind to the protein 15,22,31 . We quanti ed the number of foci per nucleus to shed light on this problem, using uorescence in situ hybridization with an RNA probe (Cy3-(CAG)7-Cy3) and an IN Cell Analyzer High-Content Cellular Analysis System to acquire images ( Fig. 3a-h). The treatment with each of the peptides produced a signi cant change in the number of cells without foci, which increased, and in the percentage of foci per cell, which was signi cantly reduced ( Fig. 3i,j). These changes occurred without altering the relative expression levels of DMPK transcripts (Fig. 3k), half of which carry the expanded CUG triplets, suggesting the possibility of a direct interaction of the peptides with the RNA, with or without a subsequent in uence on its secondary structure.
Treated DM1 myoblasts reduce the differentiation delay.
Symptoms of myotonia, muscle weakness, and muscular atrophy are the main features of DM1 1 , and the molecular contributions to these symptoms are numerous 10 . One of them is a delay in the process of muscle differentiation, which can be quanti ed as a reduction in the fusion index after the induction of the broblasts to myoblasts transdifferentiation 32 33 . After incubating with MDM both the control and DM1 broblasts for four and seven days and treatment with the peptides for 48 h, we carried out immuno uorescence with an anti-Desmin antibody and quanti ed Desmin-positive (differentiated) cells.
While the percentage of terminally differentiated cells remained unchanged after four days in MDM medium, after seven days the percentage of Desmin-positive DM1 cells increased signi cantly upon treatment with peptides 79 and 81 and remained unchanged in the presence of a scrambled control peptide ( Fig. 4a-i). The fusion index, however, did not signi cantly increase ( Supplementary Fig. 3).
Another molecular mechanism contributing to muscle atrophy in DM1 is the activation of autophagy 12,13,34 . To check the autophagy status in DM1 cells after peptide treatments, we used the lysotracker reagent, which stains acidic lysosomes 35 . First, we con rmed that the level of autophagy of the diseased cells was considerably higher than that of healthy cells (Fig. 4j,k). Furthermore, we veri ed that the treatment with the peptide 80 caused a robust reduction in the signal associated with the lysosomal vesicles, indicating a recovery of the normal autophagy levels. Along the same lines, peptides 81 and 82 increased the number of cells devoid of autophagic vesicles around the nucleus (Fig. 4j-o). Thus, these observations indicate a general reduction in the autophagy pathway, which has previously been shown to contribute to muscle atrophy in Drosophila and human cells in vitro 12,13,34 .
Candidate peptides rescue muscle atrophy of a Drosophila model of the disease.
Rescue of two molecular phenotypes related to muscle atrophy in the cell model, namely delayed differentiation and hyperactivated autophagy, prompted us to verify if these peptides were also active in vivo in a muscle phenotype in Drosophila. In this model, the expression of toxic CUG repeats is controlled by the myosin heavy chain promoter and reproduces muscle phenotypes observed in humans 13,36,37 .
After feeding the DM1 ies with 10 µM of each peptide mixed with the food, we embedded the thorax of the ies to obtain cross-sections of indirect ight muscles. The quanti cation of the muscle area from these images showed a marked improvement in the atrophic phenotype in peptide-treated ies, bringing the muscle area to values very close to those observed in control ies ( Fig. 5a-i)). Concomitant to muscle atrophy, model ies have reduced locomotor abilities, which in ies can be assessed through climbing, taking advantage of Drosophila's negative geotropism, and ight assays. First, we used 30 male ies for the climbing experiment to measure the height climbed by the ies in a given time. The results revealed a signi cant increase in the speed of the ies treated with the four peptides compared with DM1 ies fed with DMSO supplemented food or scrambled peptide controls (Fig. 5j). In ight tests, while there were no signi cant increases in the height of the landing distance (indicative of better ight capabilities), we found an increase in the percentage of ies capable of ying, especially in the case of peptide 82 treatment, where it reached statistical signi cance and almost doubled the value observed in controltreated DM1 ies (p = 0.0044, Fisher´s exact test; Fig. 5k). In conclusion, the increase in the number of ies showing the ability to y indicates a partial rescue of the Drosophila muscle function consistent with the increase in the IFM muscle area.
The secondary structure of (CUG) 23 RNA remains unchanged after candidate peptides binding.
We used a Differential Scanning Fluorimetry (DSF) assay to monitor CUG RNA thermodynamics in the presence of increasing concentrations of candidate peptides. DSF is technique used to study the effect of compounds on RNA stability as RNA undertakes structural conversions upon thermal unfolding 38 . When RNA changes its structure the single stranded form increases the available binding sites for RiboGreen dye. We represented the rst derivatives of normalized uorescence of RiboGreen with an RNA probe containing 23 repeats of CUG versus temperature in the presence of concentrations of each hexapeptide ranging from 1 to 100 µM (Fig. 6a,d,g,j). The titration with increasing concentrations of peptides 79, 80, and 81 only changed the height of the peak indicating interference in the intrinsic folding properties of the RNA probe rather than stabilization or destabilization of the RNA hairpins. Peptide 82, however, did slightly shift the curve peak towards lower temperatures, which meant that the interaction between the hexapeptide and the probe does not stabilize the single strain RNA conformation, which is in striking contrast with the proposed destabilization of CUG RNA by abp1 22 . Taken together, the DSF experiments strongly support that the candidate peptides, at least 79, 80, and 81, do not signi cantly modify the secondary structure of the CUG RNA.
Candidate peptides interact with CUG RNA with similar a nities.
A uorescent indicator displacement (FID) assay was used to investigate the nature of the interaction between the candidate peptides and the toxic RNA. Thiazole orange (TO) is an asymmetric cyanine intercalator with little uorescence when free in an aqueous solution but strong emission when forming complexes of different nature with nucleic acids. These characteristics can be exploited to study changes in the interaction between the dye and the nucleic acid of interest in response to external factors 39,40 . Speci cally, the uorescent reporter interacted with the (CUG) 23 RNA probe resulting in uorescent emission. By adding the peptide, it was possible to displace the TO, causing its uorescence to decrease.
In this way, by analyzing the uorescence emission at different RNA-peptide ratios, the value of the peptide's a nity constant for the CUG sequence could be indirectly calculated. For all peptides, a progressive reduction in uorescence was observed in response to increasing concentrations of each peptide (Fig. 6b,e,h,k); in the case of peptide 79 (k log = 5.273 ± 0.007), the greatest decrease in uorescence was observed, followed by peptide 80 (k log = 5.04 ± 0.01) and the smallest change was observed with peptides 82 and 81 (respectively k log = 4.74 ± 0.02 and k log = 4.65 ± 0.02) (Fig. 6c,f,i,l).
However, it should be noted that in none of the cases did the level of uorescence observed reach the characteristic values of a single-stranded RNA. This indicates that, although there was a clear interaction between the peptides and the RNA molecule, peptides did not interfere with the stability of the secondary structure of CUG RNA, which was consistent with the data generated by the DSF assay.
In parallel with the above experiments, we used the peptide showing the highest a nity for CUG repeats to investigate a potential direct interaction with MBNL1 proteins as an alternative mechanism of action since it might similarly prevent sequestration by the repeats. Double immunostaining with biotin-labeled peptide 79 revealed it accumulated in the cytoplasm of DM1 myoblasts, mainly in the perinuclear area, but no signi cant overlap was found with the MBNL1 protein signal (Supplementary Fig. 4). Thus, candidate peptides do not seem to interact physically with MBNL proteins, at least peptide 79.
Study of the interaction mechanism by molecular docking.
In molecular modeling, docking is a method that predicts the preferred orientation of one molecule to a second when they bind together to form a stable complex. Multiple docking studies were performed using Autodock VINA and Molecular Operating Environment (MOE) software to assess the preferred binding mechanism between hexapeptides and CUG repeats. Next, the results obtained with blind docking and guided docking techniques were compared; in the latter case, tests were carried out keeping the RNA rigid or admitting certain exibility. The nal docking protocol was validated by correlating the binding a nities predicted by docking (score) and FID results. According to the analysis, peptides 79 and 80 showed the most remarkable tendency to interact in the exposed part of the RNA (Fig. 7a,b). More in detail, peptide 79 was the only D-hexapeptide capable of recognizing two uracils of the two RNA chains by means of the two terminal amino acid residues. This observation could explain that it had the highest RNA binding constant a nity. Hexapeptide 80 would interact with the two strands of RNA but showing only the recognition of one uracil. According to the results, Trp would not be favorably available for interaction with the RNA backbone. This result is in agreement with the experimental data in which the presence of Trp did not contribute to the increase in the RNA binding constant (cpyaqe > cpyawe, cpytqe cpytwe). Although this conclusion could be due to a limitation of the simulation protocol used, which only contemplates slight exibility of the RNA, it can be inferred that Trp would not interact by stacking (attractive and non-covalent interactions between aromatic rings). As for the two peptides with the lowest experimental interaction energy, it was surprising that the binding mechanism obtained, both for peptides 81 and 82, only showed interaction with one strand of RNA (Fig. 7c,d). Of the two, 82 could interact with three consecutive nucleotides, providing a slight additional stabilization.

Discussion
In this paper, we report identifying four related peptides, with the sequence cpy(a/t)(q/w)e, that improved splicing and differentiation phenotypes in DM1-derived cells and reduced ribonuclear foci number through the direct binding to the repeats. Muscle area and functional defects in a Drosophila model also improved upon oral administration to the ies. We show that the candidate peptides bind CUG repeats with similar a nity but do not impinge on the secondary structure of the toxic RNA and propose a detailed mechanism of potential interactions between the peptides and the repeats. Although the peptides behaved similarly in many experimental conditions, they also showed differences. The explanation for these differences could lie in the sequence variations, which are responsible for molecular structural changes. Minimal structural changes can signi cantly interfere in the molecular interaction between peptides and their targets 41 . Despite some unspeci c effects by the scrambled peptide in climbing assays, the relevance of the sequence is clearly demonstrated in Drosophila muscle area determinations and in cell experiments in which SC peptide did not affect cross-sectional muscle area (Fig. 5a), the number of foci per cell, percentage of cells without foci (Fig. 3) or cell differentiation (Fig. 4).
Peptide speci cities are best illustrated as a spider graph, in which nine DM1-related parameters are simultaneously represented for each of the peptides using a semiquantitative scale that ranks peptides in each of these parameters (Fig. 8a). Thus, for example, while in terms of foci reduction peptide 81 shows the most activity, followed by 80, 82, and 79, for cell differentiation the rank was peptide 81 > 79 > 82 > 80. Indeed, since the higher the score, the better the rescue, we can integrate the area of the polynomial de ned by each peptide to rank the overall therapeutic potential of peptides as 81 > 82 > 80 > 79 (Fig. 8b). These images also help illustrate that peptide 81 exhibited the best corrective capacity in DM1 patient cells, while peptide 82 was more effective in the Drosophila disease model.
Thanks to their intrinsic property of high speci city for a target biomolecule and the use of strategies to improve stability and uptake, and to reduce toxicity, several non-natural peptides have already reached medical use 42 . Indeed, once identi ed, peptides with promising anti-DM1 potential can be further improved using various strategies to develop a lead compound suitable for clinical studies 42 43 44 .
Among the four candidate peptides, 81 is regarded as the most promising since it had the highest human cell activity and signi cant positive effects in Drosophila. The identi ed peptides are also very short, which in general terms can be expected to enhance distribution among biological tissues. In the hit-tolead process, some well-established strategies could improve the pharmacological characteristics of the peptides. For example, various studies have shown that the bioactive peptide conjugation with a cellpenetrating peptide considerably enhanced its activity 45 . This is particularly important for a peptide that must reach the muscles. Therefore, it would be of interest to conjugate our candidate peptides with the TAT fragment ( 48 GRKKRRQRRR 57 ) or with a poly-Arginine peptide (Arg 5 − 8 ), either directly or through a spacer linker 46 . Other successful strategies are the conversion of the peptide with the desired biological activity into a peptoid, in which the side chain is connected to the nitrogen of the peptide backbone, instead of the α-carbon as in peptides, and other types of peptidomimetics, to improve stability and cellular uptake 47 .
Our molecular modeling study offers a detailed view of how peptides 79-81 might be interacting with CUG repeats. One observation is that, given the known exibility of the RNA under study, it is not surprising that the observed interactions may show uctuations. Even so, the results indicate that the two interactions established by peptide 81 with C and G are more labile than the three presented by peptide 82, a fact that could explain its slightly higher binding constant. It is also observed that peptide 80 maintains its interaction with RNA in a signi cant way, although Trp is not capable of establishing a permanent interaction. Similarly, the interactions initially de ned by peptide 79 with consecutive CUG uracils are labile but are recovered during the simulation. It is nally interesting to remark that peptides 79-81 start with a cysteine (c) and end with a glutamic acid (e), which is a structural feature conserved within the zinc ngers 2 and 4 of MBNL1, 2 and 3, and zinc nger 1 of MBNL2 and also show other coincidences with conserved amino acids within zinc ngers of these proteins in positions 4 (a) and 5 (q) (Supplementary Fig. 5). Interestingly, molecular modeling predicts the terminal residues of peptide 79 to interact with two U in the double-stranded CUG RNA (Fig. 7a).
In brief, we propose that 79-81 peptides bind CUG repeats and prevent MBNL1 depletion with little or no change in the secondary structure of the CUG RNA. An increase in MBNL1 expression, perhaps as a consequence of the improved MBNL1 function itself, rescues MBNL1-dependent activities in both the cell nucleus and the cytoplasm. Enhanced MBNL1 levels and close-to-normal distribution in the cell brings about cell model rescues at the molecular and cytological levels. Despite the close similarity of 79-81 peptides with the previously identi ed abp1 22 , cpy(a/t)(q/w)e and ppyawe, respectively, the mechanism abp1 used to prevent sequestration was the stabilization of CUG RNA in its single-stranded conformation compared to an apparent sterical hindrance by 79-81. Together with abp1, these peptides offer relevant substrates for more focused medicinal chemistry studies towards developing therapies for myotonic dystrophy.

Methods
Hexapeptides used D-amino acid hexapeptides used in this work (Supplementary Table 1) were purchased from GenScript (purity >98%) and were dissolved in 100% DMSO as stock solutions and stored at room temperature until use.

Drosophila methods
For the evaluation of compounds, we used a previously described recombinant Drosophila line in which a UAS-(iCTG) 480 transgene (480 interrupted CTG repeats) was combined with the Myosin heavy chain (Mhc)-Gal4 driver for continuous expression of the toxic RNA in the y muscles, hereafter referred to as Rec-2 36 . For the crosses, 20 w females and 10 Rec2 males were crossed and 12 females of the offspring were hand-collected and transferred to a tube containing 3 ml of regular Drosophila medium supplemented with 10 μM of each compound, or 0.01% of DMSO as a control. Flies were moved to a tube with fresh food (supplemented with compounds as described above) every other day for seven days when they were processed for muscle area determinations. All y lines were maintained at 25° C on a standard day-night cycle. Climbing and ight assays were as described in 19 .

Cross-sectional muscle area determination
Analysis of the IFM area in Drosophila thoraces was performed as previously described 48 . Brie y, six thoraces of seven-day-old females were embedded in Epon. Semi-thin 1.5 µm-sections were obtained using an ultramicrotome (Ultracut E, Reichert-Jung and Leica). Images were taken at 100× magni cation with a Leica DM2500 microscope (Leica Microsystems, Wetzlar, Germany). Five images containing IFMs per y were converted into binary images using the ImageJ software, and the percentage of black pixels (corresponding to muscle) to the total number of pixels within a xed-size frame was calculated.
For toxicity assays, cells were seeded at 10 4 cells/ml in 96-well plates differentiated for 96 h and treated with different concentrations of peptides (10 μM, 50 μM, 75 μM, and 100 μM); 48 h post-incubation, cell proliferation was measured using the CellTiter 96® AQueous Non-Radioactive Cell Proliferation Assay (Promega) following the manufacturer's instructions. The IC 10 and dose-response inhibition curves were calculated using non-linear least-squares regression, and absorbance levels were determined using a Tecan In nite M200 PRO plate reader (Life Sciences).
The fusion index was de ned as the percentage of nuclei within myotubes (>2 myonuclei) regarding the total number of nuclei in each condition. The average number of total nuclei per myotube was determined by counting over 250 nuclei from randomly chosen Desmin-positive cells (5-7 micrographs).

Immuno uorescence methods
For immuno uorescence detections, myoblasts were xed with 4% PFA for 15 min at room temperature For the detection of lysosomes, CNT and DM1 myoblasts were treated as described above for immuno uorescence but incubated with 100 nM LysoTracker RED-DND99 and 5 µg/ml Hoechst 33258 (Invitrogen and Sigma-Aldrich, respectively) at 37°C for 30 min, and were mounted using uorescence mounting medium (Dako, Glostrup, Denmark). Images were taken at 400× magni cation using a uorescence microscope Leica DM4000 B LED.
For uorescent in situ hybridization, broblasts were seeded into 96 well Cell Culture Plate (1×10 5 cells per well) and treated with peptides. In situ detection was performed as previously described 49 . Images were taken and analyzed using an IN Cell Analyzer 2200 Imaging System (GE Healthcare).
RNA extraction, RT-PCR, and quantitative real-time PCR (qRT-PCR) Total RNA from human myoblasts was isolated using TRIreagent (Sigma). One microgram of RNA was digested with DNase I (Invitrogen) and reverse-transcribed with SuperScript II (Invitrogen) using random hexanucleotides; cDNA was used in a standard PCR reaction with GoTaq polymerase (Promega). Speci c primers were used to analyze the alternative splicing of DMD, cTNT, SPTAN1, CAPZB, and DLG1 in control and DM1 human myoblasts 12 . GAPDH was used as endogenous controls. PCR products were quanti ed using ImageJ software (NIH  7.4). The solution was incubated at 25 C for 10 min before measuring the initial uorescence spectrum. Then, aliquots of the tested compound (peptide 0.125 mM solutions in DMSO) were subsequently added up to saturation. After each addition, the cuvette was rigorously homogenized and let to stand for ve min prior to measure the emission spectrum. All the experiments were performed at least in duplicate to ensure the reproducibility of the data. The values calculated for Ka and their associated errors come from averaging. The equipment used was a modular PTI uorescence instrument (slit widths of 5 nm and power of 750 mV).
The measurements were carried out using 1 ml quartz cuvettes with a path-length of 1 cm. The Thiazole Orange was excited at 495 nm and the emission spectrum was registered between 505 and 650 nm. The data was analyzed with the software HypSpec 50 . Once having established an initial equilibrium model for the interaction between the CUG, the TO and the peptide, the software applies an iterative algorithm in order to t the experimental data to the proposed model, enabling the determination of the a nity constants 39,51 .

Molecular modeling methods
The molecular structures of 79-82 peptides were created and prepared (capped with ACE and NME blocking groups and chirality modi ed) using the AMBER tleap module 52 . Molecular dockings were conducted using MOE 2019.01 software (Chemical Computing Group, Montreal, QC). The molecular structure of r(CUG)16 hairpin was previously modeled by homology 53 using ModeRNA software 54 , and its stabilization was consequently studied through molecular dynamics simulations. Peptide structures were modeled in MOE and confronted with RNA, following the available induced t protocol to consider both ligand and RNA as exible structures. Triangle matcher algorithm was de ned for placement, and binding energies were quantitatively estimated by GBVI/WSA dG rescoring function. The dynamic behavior of peptide-RNA complexes was assessed using molecular dynamics simulations using AMBER16 (University of California, San Francisco, CA). Molecular systems were prepared and solvated in TIP3P water solvent using tleap module. After solvent relaxation, constraining the RNA-peptide structure with a force constant of 2.0 kcal·mol-1·Å-2, the system was slowly heated at 300 K in 1 ns. A density equilibration stage under NPT ensemble (P = 1 bar) preceded the production stage, which was conducted at NVT conditions at 300 K during 10 ns. All simulations were performed under periodic boundary conditions, using the Particle Mesh Ewald (PME) to describe electrostatic interactions and SHAKE algorithm. The time step was xed to 2 fs. Trajectory analysis, including the calculation of helical parameters, were conducted using cpptraj and considering the last 5 ns of the simulation 55 .

Statistical methods
We assumed in all our experiments that parameters follow a normal distribution. In the molecular, functional, and histological analyses, pairs of samples were compared using two-tailed t-tests (α = 0.05), applying Welch's correction when necessary. The statistical signi cance of the differences for all data reported can be found in Supplementary Table 2. The sample size is stated in each gure. In the functional analysis of ight ability of the ies, Fisher's exact test was applied to compare percentages of ies able to y with those unable.