Molecular entrapment by RNA: an emerging tool for disrupting protein–RNA interactions in vivo

ABSTRACT

mRNA function is controlled by RNA-binding proteins. The specificity of RNA-binding factors for their targets is critical in that it enables all subsequent regulation. Despite widespread recognition of the pervasive role RNA-binding proteins play in development and disease, they remain challenging to target with small molecules. A renaissance in RNA therapeutics has led to the identification of modifications that substantially increase RNA stability. When combined with information regarding specificity, a new class of oligonucleotide mimics has emerged as a means to competitively disrupt the regulation of endogenous substrates. These decoys have been used to inhibit RNA-binding proteins in living animals. Decoys will likely provide new insights into the expansive roles of RNA-binding proteins in biology and disease. Here, we describe examples where they have been used and discuss how they could be applied to new targets.

Introduction

RNA-binding proteins (RBPs) dictate virtually every aspect of mRNA function. The life of an mRNA begins with transcription in the nucleus followed by processing, export, localization, storage, translation, and ultimately decay. Each of these steps hinges on recognition of regulatory features situated in a given transcript by RBPs [1]. The interaction between cis-acting elements and their protein partners is fundamental to RNA control. Three parallel advances enable new tools for probing RBP function in vivo. These are systematic identification of RNA-binding proteins, tools to profile specificity, and chemical modifications that increase RNA stability in vivo. The use of rationally designed competitive inhibitors may enable major advances in our ability to achieve the coveted goal of understanding biological functions of liaisons between RBPs and their targets in vivo [2,3].

Genome-wide approaches to identify RNA-binding proteins in human cells have revealed over a thousand RBPs [4,5]. Several themes have emerged from unbiased assessments of the mRNA associated proteome [6]. First, many interactions between RBPs and RNA occur without the use of canonical domains (e.g. RRM, KH, etc.) [7]. Unstructured regions appear to serve prominent roles in RNA-recognition by proteins. Second, sites of RNA-binding are subject to post-translational modifications. This suggests that RNA binding can be fine-tuned within the cell. Finally, RBPs are deeply conserved between basal eukaryotes and metazoan [7]. This implies that RBPs contribute to organismal fitness and are maintained after millions of years of evolutionary divergence as a result. Beyond general principles elucidated from analyses of the growing compendium of RBPs, an equally challenging problem is to understand the specificity of these factors for targeted RNAs.

Numerous methods have emerged to characterize the specificity of RNA–protein interactions. Several enable the identification of RNA targets in cells (e.g. HITS-CLIP, iCLIP, eCLIP, sCLIP, irCLIP, PAPERCLIP, PAR-CLIP, CRAC, etc.) [8–17]. These approaches are tremendously powerful and yield transcriptome-wide views into RNA-binding preferences. Yet, extracting quantitative information is challenging as is discriminating functional sites from non-productive interactions. Additional complexity arises from combinatorial control of an RNA by protein complexes [18]. Protein partners can induce substantial changes inspecificity [19–22]. Thus, in vitro methods (e.g. SEQRS, RNA-bind and seq, RNA-compete, TGA, etc.) to assess the specificity of individual components and defined protein complexes are an invaluable source of information [23–27]. Despite tremendous advances in the area, specificity and targeting are generally insufficient to infer precise biological functions for the vast majority of RBPs.

Understanding the regulatory roles of RBPs requires strategies to perturb their interaction with targets – ideally in animal models that recapitulate key features of diseases (e.g. mice and rats). Genetic approaches remain the dominant approach and can be achieved in numerous ways. For example, conditional deletion enables cell-type specific knockout in adult animals [28]. This strategy is highly desirable for genes that are essential for development and thus cannot be examined in a whole-body knockout. Unfortunately, mouse models can require substantial investments of time and resources. Comparatively rapid approaches (e.g. antisense oligonucleotides, RNA interference, etc.) hinge on RNA-destabilization and/or translational repression to reduce protein abundance [29,30]. These methods are less robust than genetic deletion as they transiently diminish protein abundance. Nevertheless, several features of synthetic oligonucleotides for RBP depletion make the approach broadly appealing. First, the translational potential of ASOs is increasingly apparent given that more than a half dozen are FDA approved [31]. Because of the proven safety and efficacy of therapeutic oligonucleotides, this area of pharmacology is ripe for repurposing. Second, substantial improvements in the persistence and targeting of oligos can be achieved through covalent modifications to nucleotides and phosphodiester linkages. Third, the ability to mimic consensus binding elements is revolutionary given our lack of understanding of the biological importance for most RBPs. Because of the diversity of mechanisms employed by RNA-binding proteins, decoys enable manipulation of a broad range of mechanistic functions of RBPs that require direct binding to RNA and their physiological properties. Key regulatory events that could be accessed with this approach include stability, translation, processing and phase separation (Fig. 1). To date, translation and processing have been targeted with this strategy in vivo [2,3]. Yet, the diverse roles that RBPs play in disease biology suggest that potential applications are widespread.

Figure 1.

Four classes of RNA-binding protein (RBP) decoys and their potential implications.

(A) RNA decoys promote mRNA destabilization. In this example, the RBP protects the mRNA from 3ʹ end degradation through inhibition of deadenylase activity. In the presence of the decoy, the RBP is unable to bind to the mRNA transcript due to its preferred binding affinity to the decoy. The anticipated outcome is reduced mRNA stability, as deadeynlation is more efficient resulting in reduced stability. (B) RNA decoys promote translation repression. The RBP in this example promotes recruitment of the eIF4F complex. When the decoy is introduced, translation is no longer enhanced resulting in diminished rates of initiation. (C) RNA decoys that influence exon skipping. An RBP binds to an intronic splice enhancer that promotes the inclusion of an alternative exon (exon 2). In the presence of an mRNA decoy, the RBP is no longer bound to the mRNA and exon 2 is skipped. (D) RNA decoys that antagonize phase separation events. RBPs that undergo phase transitions within a cell exist in equilibrium between phase-separated granules and soluble RBP-mRNA complexes. When a decoy is introduced in the cell, the solubility of an RBP could be increased resulting in a reduction in the proportion of the RBP in a phase-separated granule.

RNA–protein interactions are prominent in disease states [6,32,33]. Pain is exemplary. Translational control of mRNA in the peripheral nervous system has emerged as a critical mechanism that governs the transition from acute to chronic pain [34–37]. Phosphorylation of the mRNA cap-binding protein eIF4E is required for persistent pain in mice [38]. Similarly, multiple RBPs have been linked to pain (e.g. FMRP, CPEB, PABP, etc.) [2,39–43]. In the central nervous system, RBPs play equally critical roles. Mutations in multiple RBPs (e.g. TARDBP, FUS, hnRNAp A1, A2, B1, etc.) have been implicated in neurodegeneration [44]. Thus, neurons rely heavily on RNA control to accomplish a range of biological functions spanning plasticity and viability. General translation factors also play prominent roles in the growth of cancer cells [45,46]. Compromised cap-dependent translation impairs tumorigenesis in vivo [47,48]. Many sequence-specific RNA-binding proteins appear to contribute to oncogenesis. One hundred and thirty-nine RBPs are consistently mutated in cancer and 76 may contain driver mutations [49]. It is unclear how many are essential in vivo. Yet, for the vast majority of RBPs implicated in cancer, their biological functions are not fully understood. As the technology for studying RBPs in living animals improves, our appreciation of RBPs in human health is likely to expand substantially. Stabilized RNA decoys are one such tool that could be applied to study RBP function in pre-clinical models.

Rise of the decoys in vivo

The initial proof of concept for the use of RBP decoys suitable for use in vivo targeted the Poly(A)-binding protein (PABP) [2]. To test the notion that PABP is specific for poly(A), PABP was first subjected to an unbiased selection and high-throughput sequencing analysis. Based on these data, a compact 12-base RNA termed a specificity-derived competitive inhibitor oligonucleotide or SPOT-ON was devised. A variety of modifications can increase RNA stability and have differing effects on the immune response. To enhance the stability of the SPOT-ON, 2ʹO-methyl linkages were introduced as well as terminal 5ʹ and 3ʹ phosphorothioates. The SPOT-ON RNA displayed a half-life on the order of 10 days as compared to 18 h for an unmodified poly(A) sequence. Importantly, the modifications did not significantly impair binding to the target. Introduction of the SPOT-ON to cells resulted in attenuation of nascent translation specifically at the initiation phase. In neurons, the SPOT-ON reduced translation both in the soma and at sites of local translation in axons. To demonstrate efficacy in vivo, the SPOT-ON was examined using standard mouse models of pain sensitization. In these animals, a specific behavioural response called paw withdrawal, is scored after application of a von Frey filament. Under baseline conditions, animals have a withdrawal threshold of ~1g of force. Injection of inflammatory mediators or noxious chemicals cause nociceptor activity to increase and promote sensitivity to mechanical stimulation. A decrease in withdrawal threshold to ~0.3g is interpreted as an enhanced pain response. In this model, pain sensitization can be blocked with a wide array of pharmacological agents that reduce cap-dependent or poly(A)-dependent translation[50]. Similarly, injection of modest amounts of the poly(A) SPOT-ON ca. 1µg blocks sensitization[2]. Given the complexity of the system, a logical question becomes if the underlying mechanism is neuronally mediated. To address this point, a second series of experiments was conducted with a stimulus that is highly specific for neurons, capsaicin. The SPOT-ON blocked sensitization following injection of capsaicin at a high dose (10 µg). This suggests that the relevant site of action of the compound is in the axons that innervate the paw. It is unclear how the oligo is taken up by neurons and should be noted that carriers were not used in this study (e.g. polymers).

A second example of RNA-decoys targeted RNA-processing factors implicated in cancer – specifically RBFOX, SRSF1, and PTBP1[3]. The oligos, termed splicing factor decoys, were designed based on consensus sequences identified from the literature and confirmed by pulldowns and splicing assays. As with SPOT-ONs, 2ʹOmethyl substitutions were added to all of the ribose groups. In all but one experiment, the backbone was unmodified. To improve the potency of the oligos, tandem copies of binding elements were introduced ranging in number from 1 to 4. Intriguingly, increasing the number of binding sites reduced equilibrium dissociation constants implying that cooperative effects promote synergistic binding to the decoys. Injection of the RBFOX oligo (8 pg) with a fully modified phosphorothioate backbone into zebrafish caused severe developmental defects of musculature in vivo. In a subsequent series of experiments, PTBP1 and SRSF1 decoys impaired growth of breast and glioblastoma cancer cells in vitro, respectively [3,51]. Intriguingly, glioblastoma tumours seeded into the brain were noticeably smaller 3 weeks after injection of a single dose of an SRSF1 decoy. These data suggest that RBP decoys targeted to RNA processing may provide a promising new type of therapeutic for cancer. As the oligonucleotides were injected without carriers, it is similarly unclear how they were taken up in vivo.

Decoys in vitro

A third example implies that RBP decoys could be used to attenuate viral transcription. Efficient transcription of HIV-1 and replication can be reduced by mimics of the TAR RNA[52]. The TAR RNA element promotes transcription by RNA-pol II through increasing phosphorylation of the C-terminal domain (CTD) via a mechanism that requires cyclin T1 and P-TEFb CTD kinase[53]. To generate a stable decoy of the TAR RNA, a circular RNA was generated[52]. The circular RNA was shown to persist for more than 12 h in HeLa nuclear extracts[52]. The RNA decoy blocks HIV-transcription in vitro and has yet to be demonstrated in vivo. Yet, the approach provides a valuable demonstration for how to decoy oligos can be used to modulate transcription. The use of circular RNAs provides an intriguing alternative to stabilize RNA-decoys. Circular RNAs are also less immunogenic than other types of RNA (e.g. mRNA) and can be tolerated at higher dosages in vivo[54].

A fourth example employs the use of RBP decoys as a means to prevent TDP-43 cytoplasmic inclusions[55]. TDP-43 is implicated in amyotrophic lateral sclerosis (ALS) and is characterized by aberrant phase transitions that lead to neurotoxic effects [55–57]. A modified oligonucleotide decoy of the one of the TDP-43 RNA-recognition motifs was made to determine its effects on optogenetically induced TDP-43 phase transitions[55]. HEK293 cells were treated with the TDP-43 decoy 4 h before optogenetic induction of TDP-43 assemblies. The bait oligonucleotide reduced TDP-43 dependent phase transition in a dose-dependent manner and thus caused a decrease in neurotoxicity[55]. The TDP-43 decoy prevents exaggerated phase transition and neurotoxicity in vitro. Although the effect of the decoy has not been characterized in vivo, these data indicate that RBP decoys can also be used to modify aberrant phase transitions and suppress accumulation of RBPs to prevent cytotoxicity.

Limitations

Despite the efficacy of RBP decoys in a limited number of in vivo studies, there are several areas where they can be substantially improved. The specificity of the decoy oligo for the target RBP is crucial. There are at least four general strategies that could be employed to characterize the specificity of existing decoys and potentially improve targeting. First, numerous modifications to ASOs improve their targeting to mRNA (e.g. LNAs)[58]. It is unclear how similar modifications would impact binding to proteins (and is likely idiosyncratic depending on the RBP) but nevertheless these could improve specificity. Second, many RBPs bind to short compact sequences [59,60]. As a result, a decoy could be recognized by multiple factors. Small molecule inhibitors provide a useful parallel, perhaps modified RNA bases could be used to design RBP decoys that are rejected by some proteins due to steric clashes but that retain binding to their preferred target. The disadvantage of this approach is the need for structural information and the requirement for sophisticated medicinal chemistry. Third, validation of RBP targeting would ideally be conducted in an unbiased way. For example, CLIP with modified oligos followed by mass spectrometry would provide a means to identify off-target RBPs that bind to a decoy. Fourth and finally, validation of on-target specificity is critical. Ideally, the combined use of genetic models lacking an RBP of interest can be used to confirm the biological consequences of a given decoy. A priori, the oligo should be inert in a system devoid of the targeted factor.

Precise delivery of oligonucleotides to target cells is a coveted goal that extends to decoy oligos. Conjugation of ASOs to cell-penetrating peptides has been demonstrated to increase delivery of oligos relative to naked oligos[61]. Similar additions could improve cellular uptake of decoys. A related problem is targeting of oligos to specific cell types. Conjugation of small molecules to the RNA could be used to impart cellular specificity. For example, the addition of folate to miRNAs substantially improves targeting to breast cancer cells and tumours[62]. Finally, not all tissue types are equally accessible via injection. Tremendous progress in polymer and lipid-based carriers has been shown to aid in the delivery of antisense oligonucleotides[63]. Their use for decoys could enable delivery to difficult tissues to access such as the lung[64].

Opportunities

The advent of decoy oligonucleotides that compete with endogenous RNA substrates enables tremendous tools for discerning the biological functions of RBPs. Given the widespread involvement of RNA-binding proteins in disease, specificity is well established for many potential targets (Table 1). We describe potential targets that have been validated in vivo in three biological contexts, neurodegeneration, cancer, and pain. While the general approach should be applicable to many disease states, these models are particularly well suited given that multiple RBPs are integral to each process.

Table 1.

Potential RNA-binding proteins as targets for decoys implicated in disease.

Potential target RBP	Biological function	Disorder
RBPs in neurodegeneration and developmental disorders
Poly(A) binding protein nuclear 1 (PABPN1)	PABPN1 binds to the poly(A) tail of the mRNA, promoted stability and export[65,67].	Oculopharyngeal muscular dystrophy (OPMD)
Fused in sarcoma RNA-binding protein (FUS)	FUS facilitates regulates alternative splicing of mRNA and RNA export from the nucleus[68,69].	Amyotrophic lateral sclerosis (ALS)
Matrin 3 (MATR3)	MATR3 is involved in RNA decay and splicing[77–80].	Amyotrophic lateral sclerosis (ALS)
RhoA-specific guanine nucleotide exchange factor (p190RhoGEF)	p190RhoGEF binds to the light neurofilament (NF-L) mRNA[81,82].	Motor neuron degeneration
DEAD-box helicase 3 (DDX3X)	DDX3X is a member of the DEAD-box helicase family and binds to the Asp-Glu-Ala-Asp (DEAD) sequence to stimulate ATP-dependent RNA helicase activity[83].	Fragile X Syndrome (FXS)
RBPs in cancer
Epithelial splicing regulatory protein 1 (ESRP1)	ESRP1 is involved in mRNA splicing of specific epithelial cell proteins by binding to GU-rich motifs on the mRNA[85].	Colorectal cancer
HuR or ELAV-like RNA-binding protein 1 (Elavl1)	HuR binds AU-rich motifs on the mRNA, regulates mRNA stability and decay[86].	Meningioma
Heterogeneous nucear ribonucleoproteins (hnRNPA1 and hnRNPM)	hNRNP proteins are a class of nuclear RNA-binding proteins that are known to be involved in stability, alternative splicing, transcription, and translation[88].	Colon and gastric cancers
KH-type splicing regulatory protein (KHSRP)	KHSRP is an RNA-binding protein implicated in splicing[92].	Lung cancer
Musashi RNA-binding proteins (MSIs)	MSI1 controls mRNA on a post-transcriptional basis[93].	Osteosarcoma
LIN28	LIN28 binds promotes translation of targets. Also implicated in miRNA control[95,104].	Tumorigenesis of many cancer
RBPs in pain
HuD or ELAV-like RNA-binding protein 4 (Elavl4)	HuD is neuron specific, binds cis-acting AU-rich elements to coordinate mRNA stability[100].	Anti-retroviral induced neuropathy
Eukaryotic translation initiation factor 4E (eIF4E)	eIF4E is a key component of the eIF4F complex to promote translation after binding to the 5ʹ cap on the target mRNA[105].	Chronic pain
Cytoplasmic polyadenylation element binding protein (CPEB)	CPEB facilitates polyadenylation and elongation of the poly(A) tail on the mRNA[106].	Chronic pain

RBP targets in neurodegeneration and developmental dissorders

Genetic lesions in the PABPN1 gene that cause protein misfolding are linked to the onset of oculopharyngeal muscular dystrophy (OPMD)[65]. PABPN1 is one of 6-PABP proteins but is restricted to the nucleus[66]. In this compartment, it regulates the length of the Poly(A) tail and promotes export and stability. Knockdown of PABN1 with viral vectors in murine models reduces muscle fibrosis and restores muscle strength in mice with OPMD[67]. Decoys could be used to target PABPN1 as a means to resolve muscular dystrophy onset without the use of virus-based therapies.

Gain-of-function mutations in the RNA-binding protein FUS cause amyotrophic lateral sclerosis (ALS) [68,69]. FUS plays a role in regulating RNA polymerase II and has been implicated in regulating alternative splicing [70–73]. FUS is primarily located in the nucleus, but C-terminus mutations can induce phase separation of FUS resulting in cytoplasmic inclusions [74–76]. These, in turn, disrupt RNA metabolism. Decoys that bind to FUS could increase FUS solubility and decrease its propensity for aggregation. The prior example of decoys for TDP-43 establishes a valuable proof of concept for this approach[55].

Similarly, RNA- and DNA-binding protein Matrin 3 (MATR3) has been implicated in ALS[77]. MATR3 is involved in the regulation of alternative splicing and regulation of mRNA nuclear export [78–80]. Deletion mutants of an RRM promote aggregation of MATR3 in the nucleus[77]. MATR3 is neurotoxic when RNA-binding activity is removed [77,78]. Given that pathogenic mutations in MATR3 reduce its solubility, one way to modulate MATR3 function would be through the use of RNA decoys.

We propose a similar mechanism of action for an RNA decoy against p190RhoGEF, a protein involved in motor neuron degeneration. p190RhoGEF binds the NF-L mRNA and plays a role in NF-L protein aggregation[81]. NF-L aggregation promotes neuron degeneration [81,82]. siRNA knockdown of p190RhoGEF causes reversal of NF-L protein aggregation in this context[82]. RNA decoys tailored to p190RhoGEF could prevent its association with the NF-L mRNA and might attenuate motor neuron degeneration.

DDX3X is a DEAD-box helicase that has recently been implicated as a modifier of RAN (non AUG) translation, specifically in the context of Fragile X syndrome (FXS) [83,84]. Knockdown of DDX3X in vivo and in cell lines reduces FMR1-associated RAN translation and decreased neurotoxicity in those models[83]. Decoys against DDX3X could prevent stimulation of FMR1-associated RAN translation and resolve core FXS behavioural deficits.

RBP targets in cancer

ESRP1 promotes colorectal cancer[85]. It regulates splicing in epithelial cells[85]. ESRP1 overexpression promotes AKT activation and upregulation of fibroblast growth factor receptor[85]. The use of RNA decoys could clarify the role of ESRP1 as a tumour promoting RBP.

HuR has been identified as a potential target in human meningiomas[86]. HuR regulates many aspects of mRNA function including splicing and stability [86,87]. Overexpression of HuR has been linked to increased tumour expression and poor prognosis[86]. siRNA knockdown of HuR in vivo has been shown to reduce tumour progression[87]. Decoys targeted to HuR have the potential to attenuate meningioma growth.

The hnRNP family of RNA-binding proteins function as key regulators in mRNA stability[88]. Overexpression and aberrant activity of hnRNPA1 and hnRNPM has been linked to gastric and colon cancers, respectively, [89,90]. Specific reduction of these proteins via siRNA and viral vector knockdown indicate significant decreases in tumour formation and cancer progression [89,90]. Decoys could reduce aberrant hnRNPA1 and hnRNPM activity and cancer progression.

KHSRP has been identified as a key regulatory protein involved in non-small cell lung cancer[91]. KHSRP functions as a regulatory protein in the context of alternative splicing, cell proliferation, and cell differentiation[92]. Viral knockdown of KHSRP in vivo results in a decrease in cancer cell proliferation and migration[91]. The use of RNA decoys against KHSRP in this context might reduce the proliferation of lung cancer cells.

MSI1 is a key player in the formation and progression of many cancers[93]. This family of proteins function redundantly and MSI1 plays a functional role in post-transcriptional regulation [93,94]. In the context of osteosarcoma, knockdown of MSI1 in vivo via shRNA has been shown to reduce cancer cell proliferation[94]. A decoy against MSI1 has the potential to suppress aberrant cell proliferation.

LIN28 has been implicated in tumour formation and cancer cell proliferation [95,96]. LIN28 has been characterized as an oncogene and is overexpressed in most cancers [95,97]. It regulates the let-7 miRNA [95,97,98]. Knockdown of LIN28 with siRNAs in murine models decreases tumour growth and formation significantly[97]. A LIN28 decoy can sponge the RBP away from the let-7 miRNA to allow for functional let-7 activity and resolve aberrant cancer cell proliferation.

RBP targets in pain

Protein translation and pain are intimately linked [34,37]. This relationship coupled with evidence that RBPs are highly expressed in neurons, makes RBPs a compelling target for decoys[36]. HuD is an RNA-binding protein that is involved in regulating mRNA stability and expression in neurons[99]. HuD has emerged as a possible driver of nucleoside reverse transcriptase inhibitor (NRTI) induced neuropathy [100,101]. Protein kinase C (PKC) inhibitors prevent overexpression of HuD after NRTI treatment and block neuropathy in vivo[100]. RNA decoys against HuD in the context of NRTI-induced neuropathy are desirable because of the broad targeting of PKC.

Similarly, CPEB regulates cytoplasmic polyadenylation and translation[102]. It is implicated in chronic pain and promotes the persistence of a ‘pain memory’[42]. Activation of CPEB in neurons coincides with the generation of a pain memory[42]. In murine models, ASOs injected intrathecally against CPEB prevent hyperalgesia and formation of a pain memory, but do not reverse priming[42]. Decoys against CPEB have the potential to serve as a therapeutic to prevent chronic pain. Small molecule inhibitors of translation are potentially useful as analgesics [38,103]. RNA decoys specifically directed against eIF4E and possibly other components of the general translation apparatus could prevent pain-associated behaviours. These decoys have the potential to reduce eIF4E and CPEB expression and resolve the downstream consequences of their overexpression in the context of chronic pain.

Future directions

We have described examples where decoys have been used in vivo and highlighted potential opportunities to improve the design, specificity formulation, delivery, and validation. Given the increasing acceptance of ASOs in the clinic, repurposing stabilized oligos directed against RBPs provides a new pharmacological paradigm with incredible translational potential. The targets we describe represent a very small subset of RBPs. As our understanding of their specificity grows, decoys could be used to leverage specificity as a means to perturb biological function in an in vivo context. This approach should be broadly applicable to a range of pre-clinical models.

Funding Statement

This work was supported by NIH R01NS100788 to ZTC.

Disclosure statement

No potential conflict of interest was reported by the authors.