|Home | About | Journals | Submit | Contact Us | Français|
Most ligands targeting RNA are identified through screening a therapeutic target for binding members of a ligand library. A potential alternative way to construct RNA binders is through rational design using information about the RNA motifs ligands prefer to bind. Herein, we describe such an approach to design modularly assembled ligands targeting the RNA that causes myotonic dystrophy type 2 (DM2), a currently untreatable disease. A previous study identified that 6′-N-5-hexynoate kanamycin A (1) prefers to bind 2×2 nucleotide, pyrimidine-rich RNA internal loops. Multiple copies of such loops were found in the RNA hairpin that causes DM2. The 1 ligand was then modularly displayed on a peptoid scaffold with varied number and spacing to target several internal loops simultaneously. Modularly assembled ligands were tested for binding to a series of RNAs and for inhibiting the formation of the toxic DM2 RNA-muscleblind protein (MBNL-1) interaction. The most potent ligand displays three 1 modules, each separated by four spacing submonomers, and inhibits the formation of the RNA-protein complex with an IC50 of 25 nM. This ligand is higher affinity and more specific for binding DM2 RNA than MBNL-1. It binds the DM2 RNA at least 20-times more tightly than related RNAs and 15-fold more tightly than MBNL-1. A related control peptoid displaying 6′-N-5-hexynoate neamine (2) is >100-fold less potent at inhibiting the RNA-protein interaction and binds to DM2 RNA >125-fold more weakly. Uptake studies into a mouse myoblast cell line also show that the most potent ligand is cell permeable.
RNA is an important yet underutilized target for small molecule intervention. Most RNA-targeting ligands have been identified via high throughput screening of a ligand library for binding a therapeutically important RNA.(1) Despite these studies, most RNA targets represent untapped potential for drug or probe design. This is in part due to the limited information on the RNA motif space that prefers to bind ligands and the chemical space in ligands that prefers to bind RNA. To better understand RNA-ligand interactions, we have been screening chemical and RNA spaces simultaneously to construct a database of RNA motif-small molecule partners.(2–4) The goal of the database is to enable the rational and modular design of ligands targeting RNA (Figure 1a–c).
Previously, several related RNA internal loops were found to bind 6′-N-5-hexynoate kanamycin A (Figure 1a, 1) by selecting members of an RNA internal loop library for binding 1.(2–4) In these studies, pyrimidine-rich, 2×2 nucleotide internal loops (5′UU/3′UC or 5′CU/3′UU) bound 1 with the highest affinity.(2, 4) We therefore searched the literature for disease-causing RNAs that display multiple copies of these or related motifs and found similarities with the rCCUG tetra-repeat that causes DM2. DM2 is caused when expanded repeats form a hairpin that displays multiple copies of a 2×2 5′CU/3′UC internal loop resulting in a toxic gain of function (Figure 2).(5) The repeating nature of the DM2 internal loops makes it an ideal system to test a modular assembly approach to target RNA. Modular assembly has been used previously as a strategy to increase the affinity of ligands for binding to a variety of targets.(6)
The rCCUG repeat expansion that causes DM2 resides in intron 1 of the zinc finger 9 pre-mRNA; its pathology is related to myotonic muscular dystrophy type 1 (DM1), which is caused by an expansion of rCUG in the 3′ untranslated region of the dystrophia myotonica protein kinase mRNA.(7) Both expanded repeats bind the RNA splicing regulator muscleblind (MBNL-1), resulting in the alternative splicing of the main muscle chloride channel and the insulin receptor.(8) Importantly, this disease model points to a therapeutic approach for treating DM—designing compounds that bind toxic RNA hairpins and inhibit MBNL-1 binding. In this report, we describe the first series of designed nanomolar inhibitors.
From previous studies, it was determined that 1 prefers 2×2 nucleotide, pyrimidine-rich internal loops.(2, 4) Similar loops were identified in the RNA that causes myotonic muscular dystrophy type 2 (DM2).(5) In order to study ligands designed to bind the DM2 RNA and inhibit the disease-causing RNA-MBNL-1 complex, a model system was constructed. We inserted one or 12 copies of the internal loops present in long rCCUG repeats (5′CCUG/3′GUCC) into a hairpin cassette (4, 9) to afford RNA2 and RNA7 (Figure 4). RNA7 contains enough tetra-repeats so that binding affinity to MBNL-1 is the same as binding to rCCUG47 (Supporting Information and (10)). RNAs displaying other internal loops (Figure 4) were also designed and synthesized in order to assess the specificity of the best ligand for inhibiting the RNA7-MBNL-1 complex. For example, RNA5 and RNA12 contain 1×1 nucleotide UU internal loops (the DM1 motif) while RNA6 and RNA13 contain 1×1 nucleotide AA internal loops (the polyQ motif). RNA3 and RNA10 contain 2×2 5′CC/3′CC internal loops; RNA4 and RNA11 contain 2×2 5′UU/3′UU internal loops. RNA8 and RNA9 are fully base paired RNA controls.
Modularly assembled ligands displaying 1 were synthesized by anchoring 1 onto azide-functionalized peptoids using a Huisgen dipolar cycloaddition reaction, a variant of “click chemistry” (HDCR, Figure 1b and c).(11) This approach is attractive because a HDCR was used to anchor ligands onto microarray surfaces to identify and study the RNA motif-small molecule partners in our database.(2–4) By using the same chemistry to anchor ligand modules onto a surface and to modularly assemble them, deleterious effects that conjugation could have on molecular recognition may be minimized. It should be noted that a HDCR was previously used to synthesize a library of 2-deoxystreptamine dimers as size-specific ligands for RNA hairpin loops.(12)
Peptoid synthesis allows for precise control of ligand valency and spacing.(11) Valency is controlled by the number of 3-azidopropylamine submonomers used to conjugate 1 while spacing is controlled by inserting 3-aminopropane submonomers between azides (Figure 1c). Each of the peptoids is named using the nomenclature described in Figure 1. The general format for peptoid nomenclature is as follows: nL-m where n is the ligand valency (a + c, Figure 1), L is the ligand module displayed on the peptoid, and m is the number of propylamine submonomers between ligand modules (b, Figure 1). The ligand modules that were conjugated to the peptoid are indicated by: K which indicates the kanamycin derivative 1; N which indicates the neamine derivative 2; or P which refers to propargylamine. Thus, 2K-4 describes a peptoid that displays two 1 modules separated by four spacing modules (a dimer) while 3K-4 describes a peptoid that displays three 1 modules each separated by four spacing modules (a trimer). The full structures of dimer 2K-4 and trimer 3K-4 are shown in Figure 3. We synthesized a library of dimers and trimers with varied spacing in order to identify the optimal distance between ligand modules such that two adjacent loops can be targeted simultaneously.
The library of ligands synthesized using this approach was initially studied for inhibiting formation of the RNA7-MBNL-1 complex (Figure 2 and Table 1). The IC50’s ranged from 100 μM for FITC-1 to 1.6 nM for trivalent 3K-4. (FITC-1 was synthesized previously (2) by conjugating 3-azidopropylamine to Boc-protected 1 via a HDCR. After reaction with fluorescein isothiocyanate, the compound was deprotected to afford FITC-1 (structure available in the Supporting Information). Thus, FITC-1 contains the triazole moiety that is also present in the multivalent ligands.) The range of IC50’s for the dimers was 0.089 to 10 μM while the range for the trimers was 0.0016 to 0.03 μM. The most potent dimer and trimer had four propylamine spacers separating the modules, or 2K-4 and 3K-4, and had IC50’s of 80 and 1.6 nM, respectively (Figure 3). These results indicate that four propylamine spacing units provide the appropriate spacing to target adjacent loops in the DM2 hairpin. Two control trimeric peptoids with the same spacing as 3K-4 were synthesized and tested to confirm that functionalization with 1 is required for potency. The peptoid that displays propargylamine modules (3P-4) was used to determine if functionalization with a triazole moiety is sufficient for potency. The second peptoid, 3N-4, displays 6′-N-5-hexynoate neamine (2, Figure 1) and determines whether any aminoglycoside with the same number of amino groups can be used to inhibit the RNA-protein interaction. Neither compound inhibited the formation of the RNA7-MBNL-1 complex whatsoever at concentrations < 1 μM, indicating that display of 1 modules is required for inhibition.(13)
To assess the enhancement in potency that modular assembly provides, the magnitude of the multivalent effects were calculated by normalizing the IC50’s for the number of RNA binding modules displayed on the peptoid and compared to the IC50 for FITC-1. When the ligands were allowed to pre-equilibrate with the RNA and then MBNL-1 added, multivalent effects ranged from 8.5 to 20,000 with the trivalent ligands giving the largest enhancements. When 2K-4 and 3K-4 and MBNL-1 were added to DM2 RNA simultaneously, multivalent effects were >420 and >6700, respectively; these values are a lower estimate because the IC50 for inhibition of the RNA7-MBNL-1 complex by FITC-1 was >500 μM under these conditions (Table 1).
Affinities and stoichiometries of RNA-ligand complexes were measured for a series of RNAs (RNA1–13 and bulk yeast tRNA, Figure 4 and Table 2) to FITC-1,(2) 2K-4, 3K-4, and 3N-4 using a fluorescence emission assay. These RNAs were selected to probe the effect of the number of repeating motifs in the RNAs and the identity of the repeating unit. Bulk yeast tRNA was used because it is a mimic of a common bystander RNA present in cells. It should be noted that RNAs with regularly repeating base paired motifs are commonly found in genomic RNAs; RNAs with repeats of either DM1- or DM2-like 1×1 or 2×2 nucleotide internal loops, however, are not (14) (see http://www.rna.ccbb.utexas.edu/ for the Comparative RNA Web Site Project).
Results show that FITC-1, 2K-4, and 3K-4 bind more weakly to bulk yeast tRNA and control RNAs that do not display the DM2 motif than RNA2 or RNA7. Each of the three ligands form 1:1 complexes with RNA2, with Kd’s ranging from 100–450 nM (Table 2). Increasing ligand valency increases affinity to RNA7 more; FITC-1, 2K-4, and 3K-4 bind with Kd’s of 400, 50, and 8 nM, respectively (Table 2). The 6.25- to 8-fold increase in affinity that each RNA binding module provides is similar to enhancements observed when dimerized neamines were studied for binding a bacterial rRNA A-site mimic.(15) Stoichiometries suggest that each ligand module displayed interacts with a single DM2 motif in the target RNA. For example, FITC-1 forms a 12:1 complex with RNA7 while 2K-4 and 3K-4 form approximately 6:1 and 4:1 complexes, respectively.
In order to assess binding specificity, 3K-4 was tested for binding to RNA8–13, which each displays 12 copies of a repeating motif (Table 2). Results show that this ligand binds with at least a 20-fold specificity for RNA7. Specificity is the lowest for other closely related loops: RNA10 which has 2×2 5′CC/3′CC loops (30-fold) and RNA12 which displays the 1×1 UU DM1 motif (20-fold). Specificities for the other RNAs range from 110-fold to >200-fold. Additionally, 3N-4 was tested for binding to RNA7 and RNA8 (fully GC paired stem). It was found to bind with Kd’s of >1000 and 190 nM, respectively. The specificity of 3N-4 to RNAs containing a continuous stretch of GC pairs is not unexpected because 2 was previously identified to bind RNA hairpins and duplexes with continuous stretches of 5′GC and 5′CG steps; the results described herein suggest that 2 could be a useful module to target stretches of GC pairs in RNA.(16) Thus, 3K-4 is highly specific for binding RNA7.
To compare the affinity and specificity of 3K-4 to the in vivo ligand, MBNL-1, Kd’s for the binding of MBNL-1 to a subset of the RNAs shown in Figure 4 were determined (Table 2). Results are in good agreement with previously published values.(10) MBNL-1 binds most tightly to the RNA containing the DM2 motif, RNA7, with a Kd of 120 nM; the second highest affinity binder is RNA12, which contains 12 copies of the DM1 motif (250 nM). RNA13 contains 12 copies of the polyQ motif (5′CAG/3′GAC) and has been recently implicated in binding MBNL-1 and causing spinocerebellar ataxia type 3 (17). MBNL-1 binds RNA13 with a Kd of 630 nM. Weaker interactions are observed with RNAs that do not contain repeats implicated in binding MBNL-1 in vivo, including RNA8 and RNA9 which are fully paired and RNA10 and RNA11 which contain 2×2 nucleotide 5′CC/3′CC and 5′UU/3′UU internal loops, respectively. These RNAs have Kd’s that range from 920 to >2500 nM (Table 2).
The binding data for MBNL-1-RNA and multivalent ligand-RNA complexes were compared to quantify if 2K-4 and 3K-4 (Figure 3) are higher affinity and more specific for binding to RNA7 than MBNL-1. Both 2K-4 (2.3 kDa) and 3K-4 (3.4 kDa) are higher affinity for RNA7 than MBNL-1 (40 kDa), by 2-fold and 15-fold, respectively (Table 2). The trimer, 3K-4, is also more specific for binding to RNA7 than MBNL-1. For example, 3K-4 is 20-fold and >100-fold specific for binding RNA7 over RNA12 and RNA13, respectively, while MBNL-1 is only 2.1- and 5.3-fold specific (Table 2). Thus, despite 3K-4 having an ~12-fold lower molecular weight than MBNL-1, it is ~10-fold more specific and 15-fold higher affinity for binding RNA7.
The enhancement in binding affinity afforded by modularly assembled ligands (2K-4 and 3K-4) over the monomer (FITC-1) were then compared to multivalent effects observed for inhibition of the DM2 RNA7-MBNL-1 interaction. This comparison is only made under equilibrium conditions when the ligand and MBNL-1 were added to RNA7 simultaneously (Table 1). Different incubation times (1, 2, 4, and 8 h) were used in order to determine when the system reached equilibrium so that multivalent effects could be correlated with binding affinities. The IC50’s are the same when 2, 4, and 8 h incubation times are used, suggesting that a 2 h incubation is sufficient for the system to reach equilibrium. Results show that multivalency increases the inhibitory potency of 2K-4 and 3K-4 by >420 and >6700-fold, respectively. Affinity comparisons show that 2K-4 binds 8 times more tightly to RNA7 than FITC-1, and 3K-4 binds 50 times more tightly (Tables 1 and and2).2). Thus, increases in affinity do not account for the enhanced potency of the ligands. These results suggest that surface area effects may play a significant role in enhancing potency. Sequestration of surface area on the target RNA by modularly assembled ligands is enhanced because the peptoid linker used to conjugate modules together can also act as a steric block for MBNL-1 to bind to the target RNA.
Modular assembly or multivalency has been used as a strategy to design inhibitors for a variety of targets.(6, 18–20) For example, STARTFISH ligands displaying multiple copies of a weakly binding trisaccharide (millimolar Kd) have been developed to inhibit the Shiga-like toxins.(21) Multivalent enhancements can be as high as 1,000,000-fold when 10 copies of the sugar are displayed on the scaffold. Such values of multivalent enhancements are some of the highest reported. Structural analysis of the ligand-Shiga-like toxin complex revealed that the STARTFISH scaffold pre-organized the multivalent ligands for binding to the target protein. More typical multivalent enhancements for display of ligands on random polymers are ~1,000-fold.(6) Thus, the multivalent enhancement of >6700 for 3K-4 trimer observed herein are higher than those observed with randomly displayed polymers and most likely not as high as the designed and pre-organized STARFISH displayed ligands.
Fragment-based assembly typically uses monomers that are higher affinity binders (micromolar Kd’s) to their target than the sugar monomers described above for STARFISH ligands. These assembly methods have been used in SAR (structure-activity relationships) by NMR, (20) in screening of small chemical libraries, (22, 23) and in the modular design of polyamides targeting the DNA minor groove.(24) In these cases, smaller numbers of ligands, dimers for example, are most commonly used. For all of these studies it has also been shown that pre-organization of the ligand modules for binding to the target has also been a critical consideration in the design of potent ligands. The peptoid scaffold used to display the RNA binding modules in this study is known to have little, if any, defined structure; peptoids can be designed to have helical structures if chiral and/or bulky side chains are present in the molecule but these functionalities are not part of our spacing modules.(25) Future studies will therefore be completed to optimize the display of ligand modules in order to pre-organize them to bind the DM2 RNA. Such rational design of spacing modules awaits result of high resolution structures of these ligands bound to RNA.
Both 2K-4 and 3K-4 were tested for uptake into a mouse myoblast cell line (C2C12) when simply added to the culture medium (Figure 5). After washing, cells were imaged using fluorescence microscopy, and images show that 2K-4 and 3K-4 enter mouse myoblasts without the aid of a transfection agent or vehicle. Compound 2K-4 is nucleolar and cytoplasmic while 3K-4 is perinuclear with some cytoplasmic and nuclear staining. Propidium iodide staining was excluded from cells that are fluorescent due to uptake of either ligand, indicating that they are non-toxic. Previous studies have shown that guanidinylated aminoglycosides enter mammalian cells and are nontoxic.(26)
One issue with using a modular assembly approach is that molecular weight increases as ligand valency increases. High molecular weight ligands, including oligonucleotides and oligonucleotide derivatives such as PNAs, are often not cell permeable.(27, 28) The positive uptake results observed with these ligands may suggest that they, or other related ligands, could have activity in cell culture or mouse models of DM.(5, 7, 29) Furthermore, cellular uptake and localization properties could be tuned by changing the submonomer used to space ligand modules, as appropriately functionalized peptoids (30) and aminoglycosides (26) have been used to transport cargo into cells.
During the course of this work, DM1-MBNL-1 inhibitors were disclosed.(31) Inhibitors had Ki’s of ~4 μM, although no ligand inhibited more than 50% of the r(CUG)109-MBNL-1 at the highest concentration tested (~20 μM). In related work herein, 3K-4 binds DM1-containing RNA12 with a Kd of 210 nM (Table 2) and inhibits the r(CUG)109-MBNL-1 interaction with an IC50 of 280 nM. This ligand binds with a similar affinity to RNA10 (displays 2×2 5′CC/3′CC loops; Kd of 270 nM) but more weakly to RNA8, RNA 9, RNA11 which displays 2×2 5′UU/3′UU loops, and RNA13, which displays 1×1 AA loops (Figure 4 and Table 2). These results are encouraging because they demonstrate that modular assembly provides nanomolar inhibitors for both DM1 and DM2 systems. 3K-4 may not be an optimal DM1 binder because four propylamine spacers between ligand modules may not be ideal for targeting DM1 repeats since the distance between loops is likely to be shorter. Further investigation of DM1 and other systems will allow the development of rules for spacing ligand modules to bind motifs separated by different distances. Results herein have already provided insight into how to space ligand modules to target simultaneously two 2×2 nucleotide internal loops separated by two base pairs. These results are likely applicable to other RNAs.
It remains to be seen if a ligand that binds DM repeats and displaces MBNL-1 would be effective at curing myotonic dystrophies. For example, it should be noted that DM1 mutant RNAs are retained in the nucleus and not as effectively translated in vivo.(32) Another report has shown, however, that many of the pre-mRNA splicing defects associated with DM1 are corrected when MBNL-1 is overexpressed in DM1-affected mouse models of the disease.(33) Thus, increasing the free amount of MBNL-1 can allow for correction of splicing defects. The latter report suggests that displacement of MBNL-1 from DM repeats by a ligand might lead to correction of splicing defects associated with DM. The former study suggests that a ligand that targets DM RNAs in the nucleus may inhibit the ligand from being exported to the cytoplasm and translated. Only cellular studies of ligands will address these issues. Taken together, cellular uptake and localization properties of DM targeting ligands are likely critical in designing effective therapies. Modular assembly may allow for ligand uptake and localization to be tuned by using different spacing modules.(30, 34, 35) It may be possible to identify spacing modules that localize ligands to the nucleus and other spacing modules that allow ligands to move freely between the nucleus and cytoplasm. If these ligands can be identified, then the effects of ligand localization on correction of splicing and translational defects could be more thoroughly understood.
In summary, specific and potent nanomolar inhibitors targeting the toxic RNA hairpin that causes DM2 have been described. Initial results also show that this strategy can be applied to DM1. Presently the rational design of ligands targeting RNA is difficult, and most RNA targeting approaches rely on screening RNA targets for binding to chemical libraries. A more efficient approach to design ligands targeting RNA might be the one described here, which is related to modular polyamide targeting of DNA (24). Such an approach would not require subjecting each new RNA target to high throughput screening assays to identify lead ligands.
Please see the Supporting Information for all details related to the synthesis of the compounds used in these studies.
RNA1-RNA13 were transcribed from double stranded DNA templates that were generated from PCR amplification using a primer with a T7 promoter. DNA templates and PCR primers were purchased from Integrated DNA Technologies. The r(CUG)109 (36) and r(CCUG)47, RNAs were transcribed from plasmids encoding the repeats; the r(CCUG)47 plasmid was cloned into pUC57 by GenScript Corporation while the plasmid encoding r(CUG)109 was a generous gift from Dr. Charles Thornton (Department of Neurology, University of Rochester). The plasmids were linearized with BamHI to afford RNAs suitable for affinity measurements or XbaI to afford RNAs suitable for MBNL-1 displacement assays. RNAs shown in Figure 4 that were used in MBNL-1 displacement assays contained an additional single stranded region (3′ tail) that is complementary to a biotinylated DNA capture probe. RNAs used to determine binding affinities did not contain the 3′ tail. All oligonucleotide sequences and PCR cycling conditions used are available in the Supporting Information.
Recombinant MBNL-1 protein which is fused to a 25 amino acid sequence encoding the LacZα peptide was expressed and purified as previously described.(31) The MBNL-1 displacements assays were completed using a modified procedure based on reference 22. The average amount of RNA immobilized in the well was determined using SYBR Green II (Invitrogen) and known concentrations of RNA. On average, when 2.5 pmoles of RNA were delivered to a well, 0.5 pmoles were immobilized. Please see the Supporting Information for detailed procedures for the MBNL-1 displacement assays and the determination of the number of moles of RNA immobilized.
The resulting data were then fit to a four parameter logistic curve to determine the IC50’s when the percentage of MBNL-1 bound ranged from 0–100% using the following equation:
where y is the percentage of MBNL-1 bound, x is the concentration of ligand, D is the minimum response plateau, and A is the maximum response plateau. A and D are typically 0% and 100%, respectively. Each IC50 was the average of at least two measurements, and the error is the standard deviation in those measurements. The values for the multivalent effects were computed using the two equations below:
The affinity of MBNL-1 for a subset of the RNA constructs shown in Figure 4 was completed by gel shift.(10) Please see the Supporting Information for experimental details and representative autoradiograms.
For affinity measurements of ligands and RNAs, fluorescence intensity was monitored as a function of RNA concentration. Plots of [RNA]/[Ligand] vs. change in fluorescence were used to determine stoichiometries. Scatchard analysis was used to determine dissociation constants.(37) For RNA-ligand interactions with 12 repeats, statistical effects had to be taken into account; therefore, the interaction was treated as a large ligand binding to a lattice-like chain as described.(37, 38) As such, the resulting curves were fit to Equation 4:
where v is the moles of ligand per moles of RNA lattice, [L] is the concentration of ligand, N is the number of repeating units on the RNA, l is the number of consecutive lattice units occupied by the ligand, and k is the microscopic dissociation constant.
The C2C12 (mouse myoblast) cell line was maintained as a monolayer in 1X DMEM supplemented with 10% (v/v) FBS and 0.5% penicillin/streptomycin. Cells were trypsinized and added to a well of a 6-well plate containing a sterile glass cover slip onto which the cells adhered. After 24 h, the medium was replaced with fresh medium containing 5 μM of either 2K-4 or 3K-4 and incubated for 14 h. After washing with 1X DPBS, the cover slip was removed and placed face down in 5 μL of 1X DPBS + 50% (v/v) glycerol. The slide was sealed, and cells imaged using a Zeiss photomicroscope equipped with a Princeton Micromax CCD and Scanalytics IPLab software. Procedures for propidium iodide staining are available in the Supporting Information.
We thank J. L. Childs-Disney for invaluable help with RNA binding assays and for critical review of the manuscript, K. Sobczak for help with the MBNL-1 inhibition assays, and A. J. Siegel and the University at Buffalo’s Microscopic Imaging Facility, Department of Biological Sciences, for obtaining all microscopy images. Financial support by A NYSTAR JD Watson Award, The Research Corporation by a Cottrell Scholar Award, The New York State Center of Excellence in Bioinformatics and Life Sciences, and the National Institutes of Health (R01-GM079235) is gratefully acknowledged.
Supporting Information Available: This material is available free of charge via the Internet.