Versatile Target-Guided Screen for Discovering Bidirectional Transcription Inhibitors of a Trinucleotide Repeat Disease

Abstract

Myotonic dystrophy type 1 originates from d(CTG·CAG) repeats that undergo aberrant expansion during normal processing because the d(CTG) repeat forms stable hairpin structures. Bidirectional transcription of d(CTG·CAG) yields two RNA transcripts that undergo repeat-associated non-ATG (RAN) translation to form homopolymeric proteins. Thus, both the r(CUG) transcript and the r(CAG) transcript are known to be toxic. We report a pairwise fragment-based, target-guided approach to screen for proximity-induced click dimers formed on the nucleic acid template. This screen uses an azide/alkyne clickable fragment library of nucleic acid-binding ligands incubated in parallel, pairwise reactions as an alternative to our previously reported one-pot screening method. MALDI-TOF mass spectroscopy was used to detect template assisted click products. Hit compounds inhibited the in vitro transcription of d(CTG·CAG)₉₀ bidirectionally with IC₅₀ values in the low micromolar range. This approach may be broadly applicable to other trinucleotide repeat diseases and in targeting other disease-associated nucleic acid sequences.

As our understanding of the complex interplay between DNA, RNA, and proteins has increased, particular attention has focused on the large amount of noncoding RNA produced and its role in disease.¹⁻⁴ Indeed, the role of RNA in a number of disease pathways has become clear and opened new opportunities for nucleic acid-targeted therapeutic approaches.^1,5 For example, in a number of cases it is desirable to downregulate certain RNAs. Thus, strategies have been developed to inhibit synthesis of RNA by targeting each of the major players involved in DNA transcription.⁶ Numerous natural products and small molecules target the protein-based transcriptional machinery. Compounds like α-amanitin have been shown to specifically bind to RNA polymerase II and slow the rate of RNA polymerization,⁷ whereas others such as triptolide covalently bind to transcription factors like transcription factor II H (TFIIH) to halt transcription.⁸

In developing transcription inhibitors, much less attention has been focused on targeting the DNA itself.⁹ Actinomycin D (ActD) is an FDA-approved natural product and anticancer agent that binds to B-form DNA nonspecifically to block RNA polymerase binding and inhibit RNA synthesis.^10,11 Transcription inhibition with small molecules that sequence selectively bind to DNA has the added advantage of significantly decreased toxicity. However, only two fully general strategies have emerged, namely Dervan’s polyamides¹² and PNAs that undergo strand invasion.¹³

Myotonic dystrophy type 1 (DM1) originates in an expanded trinucleotide repeat, d(CTG·CAG)^exp located in the 3′-untranslated region of the DMPK gene.¹⁴ The repeat is bidirectionally transcribed, forming r(CUG)^exp and r(CAG)^exp transcripts.¹⁵ The former long noncoding RNA (lncRNA) sequesters alternative splicing regulators, such as muscleblind like 1 (MBNL1) into foci within the nucleus.¹⁶ This sequestration of MBNL1 causes more than 100 pre-mRNAs to be mis-spliced through a toxic gain-of-function mechanism (Figure 1).¹⁷ One way to mitigate the effects of these toxic RNA transcripts would be to prevent their formation by halting transcription. Of note, it has recently been reported that R-loops consisting of the DNA template, the newly synthesized RNA, and the nontemplate DNA strand likely form in the transcription process and may lead to repeat instability in the form of either expansion or contraction.¹⁸ The d(CTG)^exp strand of the duplex may slip out spontaneously to form a hairpin secondary structure or a d(CTG)^exp hairpin may form during transcription.¹⁹ In either mechanism, the hairpin secondary structure offers a target for selective small molecule binding and transcription inhibition. For example, both ActD²⁰ and melamine-containing ligands²¹⁻²³ that we have developed selectively bind d(CTG)^exp hairpins and inhibit transcription in vitro to reduce toxic r(CUG)^exp loads in cellular studies and in vivo. Little attention has been paid to the r(CAG)^exp transcript in DM1 therapeutic discovery efforts, despite its ability to undergo repeat associated non-ATG (RAN) translation and form toxic homopolymeric protein aggregates.^24,25

Figure 1.

(A) DM1 pathobiology with bidirectional transcription producing r(CUG)^exp and r(CAG)^exp. (B) Therapeutic intervention strategy involving small molecule ligands to selectively target d(CTG)^exp hairpin.

We sought a more comprehensive therapeutic strategy wherein a single agent might inhibit the formation of both the toxic sense and antisense transcripts by stabilizing d(CTG)^exp hairpins. Such an agent could reduce the MBNL1 sequestration by r(CUG)^exp and reduce the formation of toxic RAN translation homopeptides from both expanded transcripts. To this end, we developed a target-guided screen using a small library of azide- and alkyne-containing ligands known to bind d(CTG) hairpin structures and performed pairwise template-assisted click selections in the presence or absence of a d(CTG)₁₆ target.

We previously described a one pot selection method using target-guided synthesis for discovering multitarget agents that selectively bind both d(CTG)^exp and r(CUG)^exp.²⁶ In this method, several RNA- and DNA-binding ligands with an azide or alkyne group were mixed and incubated with r(CUG)₁₆ or d(CTG)₁₆ to produce dimeric and trimeric ligands via template-assisted [3 + 2] cyclization. This selection method was effective in discovering selective inhibitors of DM1 RNA and DNA but was limited in the structural diversity of the dimers and trimers that were detected in the reaction. Indeed, all of the multitarget ligands discovered in the one pot selection assay featured the same activated alkyne binding partner (I) and five of the six partners featured acridine-melamine conjugate (II) (Figure 2). This observed selection bias may reflect the tighter binding of the melamine-acridine conjugate to the templates.

Figure 2.

(Top) Alkyne I and azide II previously found to be click partners to form triazole III in an in situ template-assisted selection method.²⁶ (Bottom) Small library of compounds (“fragments”) used in this target-guided, template-assisted click assay.

To explore the full scope of possible reactions and potentially find additional inhibitors with optimized potency and other desirable properties, we developed a target-guided screen that is performed in a pairwise, parallel fashion. This new method should allow for the detection of hit compounds from less competitive binders that nonetheless lead to strong binding dimers. Because the focus here was in screening for bidirectional transcription inhibitors, the template-assisted click reactions were performed in the presence or absence of a d(CTG)₁₆ template and with one azide and one alkyne ligand in parallel runs. Hit compounds were further studied for their ability to inhibit the bidirectional transcription of d(CTG·CAG)₉₀.

The previously developed clickable fragment library for target-guided screening²⁶ contained compounds 1–5 (Figure 2). Click partner 1 contains a bisamidinium groove binder and 2 features an acridine intercalator, both linked to melamine recognition units. Our previous reports showed that the melamine unit enables selective binding to the UU or TT mismatches found in d(CTG)^exp and r(CUG)^exp hairpins.^27,28 Acridines 3 and 4 and bisamidinium 5 are generic DNA and RNA intercalators and groove binders, respectively. The fragment library was further expanded and diversified to include bisamidinium 6, the azide analogue of 1. Aminoglycoside 7 and bisamidinium 8 were included because analogous compounds have been shown to bind to the major groove of A-form RNA hairpins.^29,30 Likewise, doxorubicin intercalates nonspecifically into duplex DNA,³¹ so azide analogue 9 was prepared. The heterocyclic nucleus in 10 was found in a high-throughput screen to selectively bind to d(CTG)^exp.³² Hargrove has demonstrated the importance of amiloride and its analogues as RNA binding scaffolds,³³ and Nakatani has used naphthyridine derivatives for recognition of GG mismatches.³⁴ Thus, 11 and 12 were also included to explore new potential binding roles for these ligands. As can be seen in Figure 2, the binding scaffolds were modified with either one or two azide or alkyne groups for potential proximity-induced, template-assisted click (see the Supporting Information for synthesis and characterization). Preliminary studies suggested that activated alkynes were necessary to afford triazole products at an appreciable rate.²⁶

As an initial screen, we used the four activated alkyne-containing and eight azide-containing compounds in Figure 2 to perform 32 parallel reactions, with the pairwise combinations run in the presence or absence of the target d(CTG)₁₆ or a random duplex. The short DNA target d(CTG)₁₆ was chosen for its ability to form a hairpin with a length that would limit click products to dimers and trimers. Reactions with random duplex DNA and in buffer without template eliminate nonselective ligands or those that have an alternative off-template mechanism for undergoing click coupling.

The template-assisted click reactions were run with ligands at 100–150 μM and DNA at 10 μM, if present, the mixture incubated for up to 7 d at 37 °C and qualitatively assayed for product formation using MALDI-TOF mass spectrometry. Of the 32 reactions tested, 19 showed dimeric click products after a one-day incubation with d(CTG)₁₆ at physiological pH and 4 trimeric products were also observed in these conditions (Figure 3A). Indeed, the structures of the dimeric and trimeric products could be readily assigned based on the observed m/z value (Figure 3B), making this a convenient method for hit determination. Only 1 reaction pair (3+9) showed click product both with random DNA and with a buffer only control. The latter result suggests that the acridine can stack on the aromatic system of doxorubicin off-template to induce azide–alkyne proximity.

Figure 3.

(A) Target-guided pairwise screen of 4 alkynes and 8 azides yielded a total of 19 dimeric and 4 trimeric template-assisted click products as detected by MALDI-TOF MS. The MS response was calculated directly from smoothed average data. (B) MALDI-TOF mass spectra of representative reactions 1 + 2, 1 + 9, and 1 + 11 after 24 h showing masses corresponding to monomers as well as heterodimers 1 + 2, 1 + 9, and 1 + 11 and heterotrimer 2 + 1 + 2. (C) Representative HPLC traces at 24 h. See the Supporting Information for additional details.

HPLC analysis of the pairwise click reactions over time allowed for more direct observation of product formation as well as validation of the formation of dimeric and trimeric hits observed in the rapid MALDI MS-based screen. As seen in the representative reactions 1 + 2, 1 + 9, and 1 + 11, the peaks for monomers as well as dimeric (e.g., D1 + 2), and trimeric (e.g., T2 + 1 + 2) products were well-resolved and prominent after 1 day (Figure 3C), whereas MALDI-MS nonhits such as 1 + 5 did not show product peaks (data not shown). Given the >10-fold higher concentration of ligands compared to the template, there is the potential for d(CTG)₁₆ to act as a catalyst. In fact, although additional product can be observed over a 7 day period, its growth is slow, indicating product inhibition upon formation of dimers or trimers that bind the template more tightly than the monomeric starting compounds. All of the reactions involving monomer 1 showed a progressive decrease in alkyne 1 over the 7 day incubation (Figure S4). Although potentially attributable to monomer 1 binding competing with product formation or 1 reacting via a Michael addition with the template, a control reaction with d(CTG)₁₆ but no azide partner did not show the progressive loss of 1 (Figure S4). The process by which alkyne 1 disappears has not yet been determined.

The high level of hits found in the MALDI-MS screen (i.e., 19 of 32 combinations) contrasts with the narrow structure type found in the previously reported in situ selection method in which all of the azides and alkynes compete in the same solution. Thus, the preferential click reaction between I and II likely originates in the faster proximity-induced [3 + 2] cycloaddition that results from their tighter binding to the template and higher residence times. Although tighter monomer binding may lead to tighter heterodimer binding, the greater diversity of products observed in this parallel screen compared to the previously reported selection allows a broader and more diverse set of hit compounds to be detected.

We hypothesized that d(CTG)^exp hairpins formed as slip-outs from duplex d(CTG·CAG)^exp before or during transcription could be targeted by some of the heterodimeric or trimeric compounds discovered in the above selection. To determine whether such binding might cause the polymerase to stall in both directions during transcription, we screened for activity in a bidirectional transcription inhibition assay. The assay used linearized pSP72-CTG90 and pTRI-Xef plasmids. The former contains T7 and SP6 promoter sites on either end of a d(CTG·CAG)₉₀ insert to test bidirectional transcription to form r(CUG)₉₀ and r(CAG)₉₀. The latter is a random sequence control with SP6- and T7-mediated transcription producing 1.92 of 1.89 kb transcripts, respectively (see the Supporting Information for details).

Four heterodimers of 2 and the corresponding monomers were evaluated for their ability to decrease the amount of r(CUG)₉₀ and r(CAG)₉₀ transcripts. Azide 2 was selected because its click products exhibited the most intense hits in the MALDI-MS screen. As seen in Figure 4A, three of the four dimers showed ca. 100% bidirectional transcription inhibition at 10 μM whereas dimer 2 + 12 was much less effective. Most of the corresponding monomeric compounds showed weak inhibition. Indeed, heterodimer 1 + 2 showed higher transcription inhibition levels at lower concentrations compared to each monomeric unit. For example, monomer 1 at 50 μM and dimer 1 + 2 at 1 μM achieved a comparable level of inhibition of r(CAG)₉₀ (Figure 4B).

Figure 4.

In vitro bidirectional transcription inhibition of d(CTG·CAG)₉₀. (A) DNA sequence and promoters used. (B) Inhibition by select heterodimers containing reactive partner azide 2 and constituent monomers. Compounds at 10 μM. (C) Dose dependence of bidirectional transcription inhibitor dimer 1 + 2 and constituent monomers 1 and 2. (D) IC₅₀ curves and calculated IC₅₀ values for hit dimers 1 + 2, 3 + 2, and 8 + 2 for inhibition of both r(CUG)₉₀ and r(CAG)₉₀ formation. Error is standard error of the mean, n = 3.

To further study the potency of the hit dimers in inhibiting transcription bidirectionally, IC₅₀ values for the formation of both r(CUG)₉₀ and r(CAG)₉₀ were determined for dimeric compounds 1 + 2, 3 + 2, and 8 + 2 using the in vitro transcription inhibition assay (Figure 4C). Interestingly, some of the inhibition curves are very steep, particularly for compound 8 + 2, consistent with the inhibition observed for actinomycin D.²⁰ We hypothesize that the cause of the sharp decline in transcription over a narrow concentration window could be the result of either (i) cooperative binding of the compound to hairpin in the presence of T7 or SP6 polymerase or (ii) covalent attachment of the reactive alkyne compounds to the template.

For all three of these dimers, the observed IC₅₀ value for r(CAG)₉₀ formation was lower compared to that for the formation of r(CUG)₉₀. This result can be explained by the compounds targeting d(CTG), thereby more effectively stalling polymerase on the sense strand. Of note, transcription of d(CAG)₉₀ (formation of r(CUG)₉₀) is still significantly inhibited, supporting our hypothesis that compound binding to one strand of the d(CTG·CAG)₉₀ structure could inhibit the transcription of both strands through a stalling of the transcription complex. The inhibition of formation of both r(CUG)₉₀ and r(CAG)₉₀ at low micromolar concentrations suggests that these dimeric compounds may be effective against DM1.

We have outlined a method for the discovery of new ligands that bind to d(CTG)^exp and inhibit the production of the expanded transcripts, r(CUG)^exp and r(CAG)^exp, through a parallel, pairwise screen that can detect strong, slow binding hit compounds. These hits were found to inhibit transcription of d(CTG·CAG)₉₀ bidirectionally. Bidirectional transcription inhibition is an important therapeutic strategy for DM1 and other TREDs because it not only eliminates toxic gain-of-function interactions between RNA and splicing proteins but also mitigates toxic protein aggregates formed by RAN translation. Future efforts will examine whether these multivalent inhibitors are effective at reducing toxic RNA levels in cells. This target-guided, pairwise screening platform and the general DNA- or RNA-targeting nature of the molecules in the clickable fragment library allow for the discovery of new inhibitors for DM1 and the method may be extended to other diseases that have bidirectional RNA expression as a part of their pathogenesis including ALS, FTD, and Huntington’s disease.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.1c00064.

• General experimental procedures and detailed synthetic procedures and characterization data for small molecules and methods describing the template-assisted click reactions, MS and HPLC analysis, and transcription inhibition experiments (PDF)

The authors declare no competing financial interest.