Affecting RNA biology genome-wide by binding small molecules and chemically induced proximity

Abstract

The ENCODE and genome-wide association projects have shown that much of the genome is transcribed into RNA and much less is translated into protein. These and other functional studies suggest that the druggable transcriptome is much larger than the druggable proteome. This review highlights approaches to define druggable RNA targets and structure–activity relationships across genomic RNA. Binding compounds can be identified and optimized into structure-specific ligands by using sequence-based design with various modes of action, for example, inhibiting translation or directing pre-mRNA splicing outcomes. In addition, strategies to direct protein activity against an RNA of interest via chemically induced proximity is a burgeoning area that has been validated both in cells and in preclinical animal models, and we describe that it may allow rapid access to new avenues to affect RNA biology. These approaches and the unique modes of action suggest that more RNAs are potentially amenable to targeting than proteins.

Introduction

Although ~80% of the human genome is transcribed into RNA, only 1% is translated into protein [1], suggesting that there are more opportunities to affect biology at the transcriptional level (Fig. 1). Furthermore, RNA has diverse functions tied to its structure [2], including regulating pre-mRNA splicing outcomes [3], cellular localization [4], phase separation [5], and microRNA (miRNA) biogenesis [6,7]. Indeed, RNA-targeting oligonucleotide-based modalities have garnered Food and Drug Administration (FDA) approval, including those that target RNAs associated with central nervous system disorders such as spinal muscular atrophy (SMA) [8] and liver disease [9]. Although these medicines have transformed patients’ lives, their transcriptome-wide implementation has been limited to tissues readily amenable to oligonucleotide targeting, that is, direct delivery to the central nervous system, liver, or kidney.

Figure 1: Disparity in Drug Development Against Protein and RNA Targets.

The human genome, transcriptome, and proteome are shown to scale, with current clinical targets[1] highlighted as a small slice (bright red) of the druggable proteome. Current RNA therapeutics, including antisense oligonucleotides, currently account for ~1% FDA approved therapies, shown to scale as a purple pixel within all FDA-approved small molecules.

Highly structured regions in RNA are often functional [10], and the three-dimensional shapes they adopt could provide small-molecule binding pockets [11], occupied by compounds with complementarity of size, shape, stacking, and charge. Although there are challenges associated with small-molecule targeting of RNA, small molecules can be penetrant to a variety of tissues, and often, lead compounds can be medicinally optimized to enhance target engagement [12] and minimize suboptimal features [13]. Herein, we discuss lead drug and chemical probe discovery against RNA targets on a genome-wide scale to elicit effects on the proteome (Table 1): downregulation of ‘undruggable’ proteins by targeting RNA structures in untranslated regions, upregulation of proteins by targeting miRNAs that repress their expression, and altering the ratio of protein isoforms by influencing pre-mRNA splicing outcomes. Collectively, targeting RNA structures with small molecules can indeed be an effective way to study and control biology.

Table 1:

Compounds from Highlighted References

Compound Name {ref #}	Chemical Structure	RNA target {Disease}	Potency (EC50) or change {Cell line or Model}	Selectivity {off-target}	Structure	Development Stage
Risdiplam {8/83/85}		SMN2 Splicing {SMA}	4nm (EC1.5X) {SMA patient derived Fibroblasts}	16.75nm {F0XM1 Splicing}	Yes	FDA Approved
pyrido[2,3-d]pyrimidine {29}		r(CUG) {DM1}	** Foci Reduction {DM1 patient cells}	NR	MD Only	NA
DB1246 {49}		r(G4C2)exp {c9ALS/FTD}	** Foci Reduction {iPSC Cortical Neurons}	NR	No	NA
bPGN {70}		Pre-miR-21 {Colorectal cancer}	0.035 ± 0.003 μM (GI50) {HTC-116: Colorectal cancer cells }	>10μM {CRL-1790: Normal Colon}	No	NR
Compound 5 {95}		MALAT triple helix {Metastasis and Proliferation}	*** Branching Morphogenesis {MMTV-PyMT tumors}	No significant differences {Neat 1}	MD Only	NR
TGP-21-C1–3 {97}		Pre-miR-21 {Triple Negative Breast Cancer}	** Cell Invasion {MCF-10A + miR-21 transfection}	No significant differences {miRNA Profiling}	MD Only	NR

NR = Not reported

NA = Not applicable

Drug discovery on a genome-wide scale

Defining small-molecule binding landscapes for RNA structures

Targeting RNA with small molecules is not without its challenges, particularly garnering sufficient selectivity and potency to elicit a biological outcome. Selectivity is a composite of (i) the selectivity of the small molecule for the desired RNA structure, (ii) the presence of that structure elsewhere in the transcriptome, (iii) the relative expression levels of all targets with said structure, and (iv) the structure’s functionality. We developed a method, two-dimensional combinatorial screening (2DCS), that studies the binding capacity and selectivity of small molecules for RNA structures, defining binding landscapes and structure–activity relationships on a transcriptome-wide scale. Coupling 2DCS with the accurate modeling of the RNA’s structure from sequence [14, 15, 16, 17] has enabled the development of small molecules against RNA targets on a genome-wide scale.

2DCS probes the binding potential of small molecules against a library of discrete RNA structures, wherein compounds from a chemical library are immobilized onto functionalized agarose microarrays. RNA structures that bind a small molecule are identified by high-throughput RNA sequencing (RNA-seq). The resulting RNA-seq data are analyzed to calculate the statistical confidence in the enrichment of RNA structures from a 2DCS selection vs. the starting RNA pool (library) [18]; that is, the analysis accounts for biases that may arise during in vitro transcription to generate the RNA library, reverse transcription, polymerase chain reaction amplification, and sequencing. Statistical confidence is a metric of binding fitness, including both affinity and selectivity. Logos [19] and DiffLogos [20] can then be used to visualize privileged sequences and relate structural similarities and differences between molecules, elucidating structure-activity relationships (SAR).

SAR on a transcriptome-wide scale

Fortuitously, RNA secondary structure can be readily and accurately modeled from sequence. RNA’s limited number of building blocks and the large contribution of base pairing energy to its stability enable rapid secondary structure prediction. This ability to model RNA structure facilitates mapping of potential small-molecule binding pockets across the transcriptome and insight into compound selectivity, a function of the number of structures that a compound binds, its relative affinity for each, and the presence of these structural elements across the transcriptome. Notably, the folding landscape for RNA can be dynamic, wherein interconversion between states is necessary for function in some cases [10]. Although labor-intensive, RNA conformational dynamics can be modeled and used to identify lead molecules, as demonstrated with a dynamic ensemble of HIV trans-activation response element (TAR) RNA [11]. Likewise, Hong et al [21, 22] and Dohno and Nakatani [21, 22] have exploited the base pairing propensity of unpaired and dynamic RNA structures by developing ligands to induce higher order interactions within target RNAs.

To integrate RNA structure prediction and small-molecule design on a transcriptome-wide scale, our laboratory has developed a pipeline named Inforna [23,24]. This pipeline starts with computational prediction of the target RNA’s secondary structure based on its genomic sequence alone, which can be further refined by analyzing sequence conservation [25] and by experiment [26]. The structures within an RNA target are then compared with the RNA structure–small molecule interactions identified by 2DCS or other methods. In the following section, we describe how a fundamental understanding of the molecular recognition of RNA structures by small molecules enables the development of compounds with various modes of action (MOAs) that influence the content of the proteome (Table 1).

Affecting RNA biology: binding small molecules

As coding and noncoding RNAs adopt 3D folds that influence their biological roles, there are likely many ways to modulate their function. This section describes how simple binding compounds can alter cellular protein content by various mechanisms (Fig. 2).

Figure 2: RNA Structure-Function Relationships.

Hubs of RNA structure are essential for many cellular processes[10] and provide targetable sites for small molecule binding. Shown are a subset of important RNA structural motifs and their associated biological processes, both in health and disease. RNA-binding molecules, shown in dark blue, have been developed to target a number of these structures and alter their function. Primary and precursor microRNA structures are shown on the bottom right, which are recognized by the microRNA biogenesis machinery by their specific structural motifs and the distances between them.

RNA repeat expansions

Microsatellite disorders are a class of more than 40 inherited diseases [27] caused by RNA repeat expansions, perhaps ideal targets for small molecules. In general, repeat length is correlated with disease onset and/or severity, and disease mechanism is dependent on the location of the repeat within the transcript. Once the repeat expansion reaches a pathogenic length, which is unique to each disease, the RNA folds into a periodic array of loops that form small-molecule binding pockets. These pockets are absent in transcripts lacking the repeat expansion, and the structural differences between pathogenic and short repeats afford the possibility of developing allele-specific small molecules. Furthermore, the repeating nature of these 3D structures enables the design of dimers that bind two adjacent loops simultaneously, which increases affinity, specificity, and potency [28,29]. Notably, repeat expansions, owing to the highly thermodynamically stable structures that they form, are not amenable to antisense oligonucleotide (ASO) targeting, which are not allele specific as they recognize the repeat sequence present in both mutant and wild-type alleles as well as in other transcripts.

The loops formed by RNA repeat expansions can be toxic via gain-of-function mechanisms, such as binding and sequestration of RNA-binding proteins. Such is the case in myotonic dystrophy 1 [DM1; r(CUG)^exp] and myotonic dystrophy type 2 [r(CCUG)^exp], the two most common forms of adult-onset muscular dystrophy. Indeed, many laboratories have developed ligands that bind r(CUG)^exp, although some have mixed MOAs. In early work, the Miller lab used a dynamic combinatorial screening method to identify compounds that bind a model of r(CUG)^exp [30], while Nakatani [31] and Zimmerman [32,33] laboratories synthesized dimeric compounds to bind r(CUG)^exp and inhibit the sequestration of MBNL1. A peptide has also been developed to bind r(CUG)^exp and alleviate defects in a Drosophila model of DM1 [34].

A dimer that binds two adjacent loops in r(CUG)^exp developed by our laboratory is allele selective, occupying the target in DM1-affected cells while not affecting mRNAs that contained these shorter repeats [35]. This selectivity can be traced to the lack of structure adopted by short repeats in their corresponding transcripts [29]. The dimer was later appended with bleomycin [36,37], affording a chimeric molecule that cleaved the mutant allele selectively over short repeats found in the wild-type allele and other transcripts in patient-derived muscle cells and in a mouse model with no detectable off-targets [38].

This notion that expanded repeats can be selectively targeted by small molecules over short, nonpathogenic repeats has been explored for other repeat expansions [39], for example, r(CAG)^exp, which causes Huntington disease [40, 41, 42, 43]; r(CGG)^exp, associated with fragile X syndrome and fragile X-associated tremor/ataxia syndrome [44, 45*, 46, 47]; and r(G₄C₂)^exp, which causes C9orf72 frontotemporal dementia and amyotrophic lateral sclerosis (c9ALS/FTD). The latter repeat has been shown to adopt various structures [48, 49, 50*] (an array of internal loops and G-quadruplexes) that have been targeted with small molecules for therapeutic benefit [45,50*, 51, 52*].

MicroRNAs

Small, but highly structured, miRNAs suppress translation of complementary mRNAs, acting as genetic ‘dimmer switches.’ Expression levels of a subset of proteins can therefore be tuned by modifying miRNA abundance, that is, miRNA biogenesis. Primary and precursor miRNAs form well-defined structures that are recognized by processing enzymes, affording the mature miRNA. Small molecules that bind structures located in processing sites and thus inhibit miRNA biogenesis have been discovered through various methods, including microarrays [53] and a click chemistry approach that assays for Dicer inhibition [54]. The Duca lab developed aminoglycoside conjugates and polyamines to show specific interaction with miR-372 [55,56], while many studies have focused on inhibiting the biogenesis of miR-21 [57*, 58, 59, 60, 61*, 62**, 63] as its upregulation has been linked to cardiovascular disease, fibrosis, and many cancers.

The structures embedded in miRNA processing sites are not always unique in the transcriptome. In the following paragraphs, we describe the lead optimization of a small molecule that binds a structural element found in the Dicer processing sites of two different miRNAs, miR-421 and miR-377, to create a miR-377–selective inhibitor (Fig. 3) [64]. The lead molecule for the two Dicer sites was generated from the RNA-binding preferences of AstraZeneca’s small-molecule library, as determined by 2DCS. As expected, the lead molecule occupies both miRNAs in cells similarly and hence inhibits their biogeneses to similar extents. Target occupancy was studied by Chemical Cross-Linking and Isolation by Pull-down (Chem-CLIP), in which the RNA-binding small molecule is functionalized with a cross-linking module. RNAs cross-linked to the Chem-CLIP probe are isolated by cells and analyzed by reverse transcription-quantitative polymerase chain reaction (RT-qPCR) or RNA-seq.

Figure 3: 2D Combinatorial Screening and Fragment Assembly of RNA-Binding Small Molecules.

RNA sequence can be combined with folding simulations to reveal important structural motifs across the transcriptome. To demonstrate that binding preferences can inform design of a selective small molecule inhibitor, a degenerate Dicer site in pre-miR-377 and pre-miR-421 was used as a test case. A structural feature unique to miR-377 was exploited, affording a dimer that binds the degenerate and unique site simultaneously, abrogating binding and activity against pre-miR-421.

Thus, the question at hand was if the lead molecule could be optimized in facile fashion to selectively inhibit miR-377. Careful inspection of pre-miR-377 and pre-miR-421 revealed a structural feature unique to pre-miR-377, which fortuitously was nearby the Dicer site and could be exploited to enhance selectivity. Inforna identified a compound that binds the unique site, and tethering the two binding fragments together (a heterodimer) afforded selective inhibition of miR-377 biogenesis, as determined by miRNome-wide profiling and proteomics studies. Importantly, competitive Chem-CLIP studies, in which cells are cotreated with the Chem-CLIP probe and parent compound, showed that the pre-miR-377 target, but not pre-miR-421, was occupied.

Collectively, these studies show that degeneracy in RNA structural motifs in the genome does not preclude the development of selective small molecules that induce a biological effect. The combination of RNA structure modeling and small-molecule binding preferences, in the form of binding landscapes, can inform lead molecules and identify degenerate sites in the transcriptome. Furthermore, this degeneracy can be alleviated by exploiting structural differences in the RNAs with the degenerate site. Although these chimeric molecules lie outside traditional oral bioavailability guidelines, such as the Lipinski Rule of Five [65], there is growing evidence that therapeutic small molecules can indeed deviate from these standards.

Targeting pre-mRNA splicing with RNA-binding compounds, molecular glues that stabilize RNA–protein complexes

Small molecules can also affect the protein isoform generated from an mRNA by affecting alternative pre-mRNA splicing, with important proof-of-concept studies demonstrated with oligonucleotides [66]. The most extensive efforts have centered on alleviation of SMA, culminating in therapies such as risdiplam [8]. In SMA, the survival motor neuron 1 (SMN1) gene is defective, resulting in loss of SMN1 protein. Fortuitously, humans have a second SMN gene, SMN2, that varies by a single nucleotide, or single nucleotide polymorphism (SNP). This SNP changes SMN2’s splicing pattern, causing exon exclusion and producing a protein isoform with a reduced half-life. Risdiplam directs the splicing of the SMN2 gene to include SMN2 exon 7, affording the stable protein isoform that can compensate for loss of SMN1 [67].

Small molecules that influence splicing have been most commonly discovered by phenotypic screens and generally act by stabilizing RNA-protein interfaces [68]. More recently, drug/probe discovery efforts have focused on RNA-specific screening, as was the case for the discovery of the ‘5’ splice site bulge repair’ compound that directs the inclusion of SMN2 exon 7 [69].

Small molecules can also influence the protein isoforms generated from a transcript by effecting exon exclusion, as demonstrated for microtubule-associated protein tau (MAPT) [70, 71, 72]. In frontotemporal dementia with Parkinsonism-17, a splice site mutation enhances the interaction of a regulator element in MAPT with U1 snRNA, causing inclusion of exon 11 and an aggregation-prone protein isoform [73]. Small molecules that stabilize the mutated structure facilitate exon exclusion, shifting splicing toward the protein isoform that does not aggregate [70].

Targeting internal ribosome entry sites and G-quadruplexes

Small molecules can also alter the proteome by binding and occluding important structures in the encoding transcript that regulate protein expression. In particular, targeting internal ribosome entry sites sallows for gene-specific regulation of translation, for example, the internal ribosome entry site in c-MYC [74]. Likewise, inhibition of translation can be achieved by stabilization of other structures in the 5’ untranslated region (UTR), preventing cap-dependent translation. Of these UTR targets, G-quadruplex (G4) structures are often targeted as they are incredibly stable and are found in oncogenes KRAS [75], metalloproteinase 10 [76], NRAS [77], telomeric repeat-containing RNA [78,79], and others.

Small molecules that target metastasis-associated lung adenocarcinoma transcript 1, a long noncoding RNA

Long noncoding (lnc) RNAs are an emerging class of targets with a diverse array of functions, including epigenetic regulation, polycomb formation, nuclear localization, and others. A notable lnc RNA target, metastasis-associated lung adenocarcinoma transcript 1 (MALAT1), causes tumorigenesis by forming a highly stable triple helix structure that enables nuclear localization and results in constitutive function. Indeed, small molecules that bind the triple helix reduce oncogenic potential [80].

Genome-wide design: fully functionalized fragments

Fragment-based drug discovery has been a key technology in the protein-targeting field. Unlike traditional screening methods of large small-molecule libraries (~10⁶+), fragment-based approaches use a much smaller number (~10³) of low-molecular-weight compounds. Indeed, a recent study combined fragment-based ligand discovery with a cross-linking and mass spectrometry strategy that uses fully functionalized fragments (FFFs) equipped with an alkyne and a photoreactive diazirine [81]. The strategy enabled study of target specificity for each fragment in living cells and allowed for optimization of highly homologous proteins while reducing the resources required for fruitful screening efforts.

This FFF screening strategy was applied to RNA targets using pre-miR-21 as a test case. In brief, the library was screened in the presence of a previously developed Dicer site–binding molecule to identify FFFs that bind other sites on the target RNA [82]. One FFF bound cooperatively with the RNA-binding module, the tethering of which afforded a low-molecular-weight dimer with improved affinity and specificity over its individual counterparts, validated by experiments that measured RNA target occupancy in cells and RNA-seq. Collectively, these studies provide a facile route to lead molecule optimization without substantially increasing the molecule’s molecular weight and hence maintaining druglikeness.

Moving forward, the cellular screening of FFFs against RNA could enable rapid targeting of functional sites in the RNA’s structure, simplifying the pipeline for transcriptome-wide targeting of RNA; that is, functionalized fragment screening of RNA targets strengthens structure-based targeting platforms by directly mapping small-molecule binding.

Affecting RNA biology: induced proximity

In addition to inhibiting RNA function through structure-specific targeting of tandem binding pockets, fragment assembly has enormous potential to enable design of small molecules with MOAs outside simple binding; that is, a structure-specific RNA-binding fragment can be coupled to a fragment that recruits enzymes or proteins that act on RNA, referred to as induced proximity. Originally developed for protein-targeted ligands by recruiting ubiquitin transferase [83] to target proteins for degradation, ‘protein targeting chimeric molecules,’ or PROTACs [84], were adopted for targeted RNA-induced degradation and similarly named ‘ribonuclease targeting chimeras’ or RIBOTACs [62,85,86]. These chimeric molecules comprise an RNA-targeting molecule attached to a molecule that selectively recruits ribonuclease (RNase) L to the desired RNA target. A miR-21–targeting RIBOTAC had enhanced activity and selectivity as compared with the parent binding molecule and selectively altered the proteome, short-circuiting oncogenic molecular pathways in triple-negative breast cancer cells and a mouse model of triple-negative breast cancer. The observed enhancement in selectivity was likely the combined effect of the RNA binder and the inherent substrate preferences of the RNase. A recent study showed that selectivity could be engineered into a promiscuous protein-binding small molecule by its conversion into a PROTAC [87]. Overall, these proof-of-concept studies of induced proximity directed at RNA targets illustrate the ability to recruit a cellular factor with desired function to a specific RNA using a small molecule. Importantly, induced proximity eliminates the requirement of RNA-targeted small molecules to bind functional sites. Although not induced proximity, a recent study showed that a bifunctional small molecule comprising an RNA binder and an inhibitor of Dicer blocked miR-21 biogenesis [88], an interesting approach that could be applied to other miRNAs.

Induced proximity has immense potential as other cellular factors could be recruited to specific RNA structures with a chimeric small molecule, expanding possible compound MOAs (Fig. 4). For example, induced proximity could be used to introduce site-specific modifications by recruiting adenosine deaminases that act on RNA (ADARs), supported by studies in which a guide RNA, delivered by an adeno-associated virus (AAV) vector, specifically edited an RNA target [89]. RNA lifetime could be altered by recruiting the polyadenylation machinery, deadenylating enzymes, terminal U transferase (TUTase) which targets an RNA for degradation, decapping enzymes, and so on. Induced proximity could unveil exciting new functions for RNA and further our understanding of RNA biology for therapeutic benefit.

Figure 4: Chemical Induced Proximity Expands the Mode of Action of RNA-targeted Small Molecules.

Using computational prediction and a fully functionalized fragment (FFF) screening strategy, structure-specific dimeric molecules can be used to reveal RNA function and downstream networks (red binding sites), for example the inhibition of miR-377 biogenesis[64]. In addition, induced proximity can be used to alter RNA lifetime (degradation by RNase L or TUTase[62,85] or change its sequence using editing enzymes such as ADAR (green), which could change the sequence of the encoded protein or alter splice site selection. Collectively, induced proximity has the potential to impart new modes of action for RNA-targeting small molecules.

Summary and outlook

There is vast potential to alter the proteome by targeting RNA, not only for RNA-related diseases but also for ‘undruggable’ proteins. RNA’s limited number of building blocks and diverse folding landscapes come with unique and challenging problems. However, its hierarchical folding and the accurate modeling thereof can be exploited to design lead molecules. With additional methodology such as Chem-CLIP and Chem-CLIP Fragment Mapping (Chem-CLIP-Frag-Map), small-molecule specificity can be directly measured in cells, validating the target and informing off-target effects. These methodologies can be synergistically combined into a platform for drugging the RNA transcriptome by sequence alone.

As genome-wide association studies provide a stream of RNA targets, many questions remain, such as how relatively small changes in sequence or repeat length cause large changes in structure [90], the uniqueness of RNA structures throughout the transcriptome, if small molecules can lock dynamic structures in a cellular RNA into a single conformation, and so on. Induced proximity could be an invaluable tool for future RNA research. While initial work demonstrated that removing a specific RNA through the recruitment of an RNase is indeed feasbile, induced proximity could provide a more extensive understanding of RNA localization, modification, lifetime, and epigenetic effects. Indeed, recent studies on RNA methylation [91] and sequence modification [89] illustrate its biological diversity. As we anticipate the discovery of new and important RNA functions, we believe that RNA-targeted therapies will become a pillar of small-molecule development.