Identifying proteins that bind to specific RNAs - focus on simple repeat expansion diseases.

RNA-protein complexes play a central role in the regulation of fundamental cellular processes, such as mRNA splicing, localization, translation and degradation. The misregulation of these interactions can cause a variety of human diseases, including cancer and neurodegenerative disorders. Recently, many strategies have been developed to comprehensively analyze these complex and highly dynamic RNA-protein networks. Extensive efforts have been made to purify in vivo-assembled RNA-protein complexes. In this review, we focused on commonly used RNA-centric approaches that involve mass spectrometry, which are powerful tools for identifying proteins bound to a given RNA. We present various RNA capture strategies that primarily depend on whether the RNA of interest is modified. Moreover, we briefly discuss the advantages and limitations of in vitro and in vivo approaches. Furthermore, we describe recent advances in quantitative proteomics as well as the methods that are most commonly used to validate robust mass spectrometry data. Finally, we present approaches that have successfully identified expanded repeat-binding proteins, which present abnormal RNA-protein interactions that result in the development of many neurological diseases.

INTRODUCTION

RNA–protein interactions play a key role in the regulation and coordination of gene expression in cells. Every stage of the RNA life cycle, including RNA synthesis, maturation, modification, transport, and degradation, is tightly controlled by a multitude of RNA-binding proteins (RBPs). RNA molecules and their interacting protein partners form distinct, highly dynamic ribonucleoprotein (RNP) particles, which comprise the basic unit underlying these posttranscriptional events (1–3). Any defects in RBP expression and function as well as mutations in target RNA molecules can disrupt protein-RNA networks and cause human diseases, such as cancer, autoimmune pathologies, metabolic and neurological diseases (4–7). Moreover, interactions between viral RNA and host cell proteins mediate various aspects of viral replication, leading to infectious disease development (8–10). Therefore, intensive efforts are being undertaken to explore protein-RNA interactions, not only to better understand the complex interplay between RNAs and their associated RBPs in the regulation of fundamental cellular processes but also to gain more insight into the pathogenesis of numerous diseases.

To date, various genetic, biochemical or microscopic in vitro and in vivo methods have been developed to study the RNA-binding proteome (11–16). Several of these methods allow the comprehensive identification of new RNA-associated factors, whereas others characterize known or suspected RNA–protein interactions in detail. Undoubtedly, recent developments in high-throughput technologies, such as RNA affinity purification combined with mass spectrometry (MS), protein microarrays and next-generation sequencing, have significantly contributed to deciphering the repertoire of RNP complexes (17–20). Among the widely used methods that explore the RNA–protein interactome, RNA-centric approaches employ MS to identify the protein partners associated with a specific RNA, which is used as bait. Using this RNA-based strategy, RNP complexes are formed in vitro or in vivo, and a given RNA obtained from in vitro synthesis or cell lysates is immobilized to a chromatographic matrix either covalently or non-covalently. Non-specifically binding proteins are removed by several extensive washing steps, and RNP complexes are eluted from the solid support for MS analysis. Despite the availability of various RNA affinity purification methods, the identification of relevant RNA–protein complexes remains technologically challenging. This difficulty is predominantly associated with the isolation of low-abundance proteins that specifically interact with an RNA of interest from complex protein mixtures containing highly abundant proteins that bind non-specifically to RNA. Another important issue that can impede RNA affinity purification is the fact that RNA–protein interactions in cells are highly dynamic and can undergo extensive remodeling. This transient nature of RNP complexes principally results from the fact that RNA is structurally very flexible and can adopt a large variety of tertiary structures (21,22). With regard to the above difficulties in identifying specific RBPs, many investigators are continuing to optimize existing strategies by stabilizing the aptamer structures that are used to tag RNA molecules of interest, using quantitative proteomics, or developing new approaches that ensure the specific elution or identification of RBPs that associate with RNA in living cells.

In this review, we first present RNA affinity purification approaches that are currently used for the identification of proteins that bind to a given RNA. We briefly discuss the advantages and limitations of purifying of in vitro- and in vivo-assembled RNP complexes, including the increasing demand for quantitative MS. In the second part, we focus on the methods used to determine proteins that bind to expanded RNA repeats, which are involved in aberrant RNA–protein interactions that lead to the development of many neurological diseases. This particular group of disorders is associated with the expansion of simple repetitive elements within specific single genes. Depending on the location of the mutation, various pathomechanisms might be involved, one of which is mutant RNA gain-of-function. According to this mechanism, the expression of transcripts harboring repeats of abnormal length leads to the formation of RNA foci that capture specific RNA-binding proteins, resulting in their altered function. It is likely that the proteins that have been identified thus far represent only a small fraction of mutant repeat-binding proteins and that most such proteins await discovery. Thus, there is a need to use more advanced protein capture approaches and to develop new, more efficient methods for identifying proteins that are associated with these unusual types of transcripts to gain better insight into the RNA-triggered mechanisms that contribute to these disorders.

DESIGN AND IMMOBILIZATION OF RNA BAIT

Covalent linking

Different protocols to immobilize the RNA bait on a solid support have been developed. This is achieved by chemically modifying bait RNAs by introducing RNA tags or by using antisense strategies (Supplementary Table S1). In the first case, in vitro transcribed RNA baits are covalently linked to the solid support, as is the case for oxidized RNAs linked to cyanogen-activated Sepharose beads (23) or for adipic acid dihydrazide agarose beads (24–27). Unfortunately, covalent attachment of bait RNAs does not permit elution of RNA-binding proteins in a highly specific manner. Therefore, elution of RBPs is performed primarily using highly denaturing buffers (24,28) or by digesting RNA with robust RNases (29). Moreover, covalent attachment of bait RNA has been used almost exclusively with shorter transcripts (<100 nt), such as short mRNA regulatory motifs (25,26), specific pri-miRNAs (27), pre-miRNAs (30) and miRNAs (31).

Biotin tagging

In an attempt to circumvent some of previously mentioned limitations, non-covalent methods to immobilize bait RNAs to a solid support have been used. Among the most common and efficient approaches is chemical modification of bait RNAs by incorporating specifically modified rNTPs during in vitro transcription (32–34). These modifications include biotin (vitamin H), desthiobiotin, and digoxigenin. Among these, biotinylation of the bait RNAs is the most widely used, as biotin shows unprecedented, high affinity to streptavidin/avidin - proteins isolated from Streptomyces avidinii (Kd ∼ 10−13 M) (Supplementary Table S1). This binding was shown to be very rapid, highly specific, and resistant to high salt concentration, extreme heat, pH and proteolysis. The caveat of this strategy is that incorporation of biotin internally into bait RNAs may cause changes in the RNA structure and formation of nonphysiological RNA–protein complexes. Therefore, biotin tagging of in vitro transcribed bait RNAs is carried out at their 5′ or 3′ ends using T4 polynucleotide kinase or T4 RNA ligase (35). This single modification is unlikely to perturb the natural function of the molecule due to the small size of biotin (MW = 244.31 g/mol). However, 3′ and 5′ end biotinylation reactions can be inefficient or require long reaction times, especially for large (e.g. telomerase RNA, 451 nt) (36) or highly structured bait RNAs (e.g. let-7 pre-miRNA) (37). Furthermore, the high affinity of biotin to streptavidin/avidin makes native elution, which is performed by adding excess biotin, inefficient; therefore, elution of biotinylated RNAs is generally performed with highly denaturing buffers or with RNases. To circumvent these problems, desthiobiotin, which has a much lower affinity for streptavidin (Kd ∼10−5 M) can be used instead of biotin (38). Tagging of RNA bait with biotin is principally restricted to the analysis of in vitro assembled RNP complexes due to the limited transfection efficiency of biotinylated RNAs and the high cost of obtaining a sufficient quantity of biotinylated transcripts for transfection.

MS2 aptamer tagging

The MS2 aptamer (23 nt long) is a commonly used, naturally occurring RNA stem–loop aptamer that is utilized for the in vitro or in vivo isolation of RNA-binding proteins (Supplementary Table S1). The MS2 RNA stem–loop structure can bind specifically and with a high affinity to a coat protein from an Escherichia coli bacteriophage, MS2cp (39–41). In this approach, repeats of the MS2-binding RNA stem–loop are incorporated (during in vitro or in vivo transcription) into an RNA of interest, and the tagged RNA complex is purified by coupling the MS2 protein to a solid support or resin. Generally, the 3′ end of the RNA is tagged with MS2, but occasionally the 5′ end is used. Two variants of MS2 aptamers are used: the wild-type MS2 and high-affinity C-loop MS2 with Kd values of 1–3 and 0.2–0.6 nM, respectively. Due to this high binding affinity between MS2cp and MS2 aptamer, the limiting step in the affinity purification of RNA-binding proteins is the elution of the purified complex under native conditions. MS2cp is often fused with another protein, such as maltose-binding peptide (MBP) or streptavidin-binding protein (SBP), to ensure that the elution is specific and to reduce the identification of background proteins. The MBP domain of the fusion protein is captured on amylose beads, and elution of intact protein-RNA complexes from the beads is facilitated by adding excess maltose (42,43). In the second case, SBP fusion allows for the purification of RNA–protein complexes using streptavidin-conjugated beads and a specific, native elution using biotin as a competitor (44). Alternatively protease cleavage site (TEV) can be inserted between MS2cp and another fused protein attached to the resin (45,46). The MS2 approach was successfully used to identify the RBPs associated with long non-coding RNAs, highly stable and abundant RNPs, such as the U1 small nuclear ribonucleoprotein particles (snRNP) (39), and less stable mRNA-binding proteins (44).

PP7 aptamer tagging

The PP7 stem–loop aptamer is a different, naturally occurring RNA aptamer that is similar to the MS2 system and is also used for in vitro and in vivo purifications of RNA-binding proteins (Supplementary Table S1) (41,47). This 25 nt long stem–loop aptamer can be fused to the 5′ or 3′ end of an RNA of interest and was shown to bind with high specificity to the Pseudomonas aeroginosa bacteriophage 7 coat protein PP7 (47). Although the PP7 coat protein maintains a high affinity to aptamer-tagged RNA molecules across a wide range of ionic strength and pH, the PP7 system is used less frequently than the MS2 system, possibly due to a higher dissociation constant (Kd ∼ 1 nM). Furthermore, the PP7 system was shown to efficiently isolate both stable RNA–protein complexes, i.e. 7SK RNPs (48), as well as more transient RBPs associated with nascent mRNAs (49).

S1 and D8 aptamer tagging

Another approach to purifying specific RNA–protein complexes is to use artificial RNA aptamers with high binding affinity to known proteins or to small molecule ligands. In contrast to the described bacteriophage systems, these aptamers do not require the synthesis of recombinant proteins. Using SELEX approaches, the artificial S1 aptamer (44-nt long) that binds to streptavidin and the D8 aptamer (33-nt long) that binds to Sephadex (polysaccharide dextran B512) were identified (Supplementary Table S1) (50,51). Because the binding affinity of the S1 aptamer to streptavidin (Kd ∼ 70 nM) is higher than that of D8-Sephadex, this artificial aptamer is preferentially used. Historically, the S1 aptamer has been used extensively for the in vitro and in vivo purification of RNA–protein complexes, such as those associated with RNase P ribonuclease (51,52), 28S rRNA (45,53) and mRNAs (54–56). Furthermore, the weaker affinity of the S1-streptavidin interaction compared with the biotin–streptavidin interaction is commonly used to enable the specific elution of the S1-tagged RNAs from the streptavidin resin using excess biotin (57). Moreover, by optimizing the S1 aptamer structure and repeat conformation (60 nt), a 15-fold increase in the purification of a specific RNA–protein complex was observed when compared to that obtained using the MS2 and PP7 systems (58). This modified 4xS1m aptamer was used to capture and identify (from cellular extracts; an in vitro strategy) both known and novel RBPs that bind to AU-rich element of tumor necrosis factor α (58).

Tobramycin and streptomycin aptamer tagging

Tobramycin aptamer (40 nt) and streptomycin aptamer (46 nt) are also artificial, stem–loop oligonucleotides that were identified using the SELEX approach and were shown to bind with high affinity to tobramycin (Kd ∼ 5 nM) and streptomycin (Kd ∼ 1 μM) matrices, respectively. Purification of RNA–protein complexes using these two aptamers is performed mainly in vitro (Supplementary Table S1). Using tobramycin aptamer-tagged pre-mRNA, structurally intact and catalytically active pre-spliceosomal complexes were purified under physiological conditions for the first time (59,60). The elution of native spliceosomes was performed using a competitive approach in the presence of excess tobramycin (61). One copy of this aptamer tag was generally attached to an RNA of interest at either the 5′ or 3′ end of the RNA or internally. In turn, streptomycin aptamer-tagged (StreptoTag) U1 snRNA was used to identify the complement of proteins that are associated with 5′ splice site selection during pre-mRNA splicing (62). The streptomycin-tagged RNA–protein complexes were bound with high affinity (Kd ∼ 1 μM) to an affinity column containing Sepharose-immobilized streptomycin. This association is highly dependent on the presence of Mg2+ ions (63). Following binding, the complexes were recovered from the affinity matrix by elution with excess streptomycin. One copy of the aptamer tag was attached to an RNA of interest, usually (but not always) at the 3′ end of the RNA. Furthermore, a StreptoTag was successfully used to purify yeast and phage RNA-binding proteins, as well as proteins that are associated with group II intron, viral and bacterial ncRNAs (noncoding RNAs). Furthermore, certain bacterial mRNPs that bind ncRNAs have also been identified (62,64,65).

The CRISPR/Csy4 system

Recently, a novel RNA affinity tag has been developed that utilizes a Pseudomonas aeroginosa CRISPR/Csy4 system for the highly efficient purification of RNA-binding proteins associated with specific RNAs (Supplementary Table S1) (66). In this system, in vitro generated RNA transcripts engineered with a short 16 nt hairpin (5 bp stem and 5 nt loop) were shown to irreversibly bind, with exceptionally high affinity (Kd = 50 pM), to an inactive, biotinylated form of Csy4 endoribonuclease. Upon immobilization of biotinylated Csy4 on a solid support, e.g. streptavidin-conjugated beads, Csy4 catalytic activity can be rescued by the addition of imidazole, to cleave the hairpin-tagged RNA, removing the tag and releasing the remaining RNA together with its associated proteins to the solution. This highly specific elution step was shown to generate fewer false positives compared to other elution methods. Although this system was used to in vitro purify RNA-binding proteins associated with relatively short RNAs (pre-miRNAs) tagged with this 16 nt hairpin at their 5′ends, longer transcripts, different positions of the hairpin within studied RNAs, and utilization of this system in an in vivo context should be considered (66).

Antisense oligomers for RNA capture

Antisense affinity capture approaches are another powerful tool for analyzing protein complexes bound to specific RNA. These approaches allow isolation of in vitro transcribed or, more importantly, endogenously expressed specific RNAs from cells together with associated proteins (Supplementary Table S1). In principle, antisense oligonucleotides, modified or unmodified, are first attached to a chromatographic support, e.g. by using the streptavidin–biotin interaction, and are used to isolate the RNA bait of interest by nucleic acid complementarity. These RNP complexes can then be eluted in their native form via excess competitor oligonucleotides or simply by using denaturing buffers or RNases. The DNA modifications that are commonly used to enhance the specificity and affinity of antisense oligonucleotides to target RNAs are 2′-O methylation and 2′-O-alkylation (67–69). Moreover, in a recently developed PAIR strategy (PNA-assisted identification of RBPs), described in detail later in the text, PNA oligomers were used to isolate a specific endogenously expressed ankyloses mRNP complex with satisfactory efficiency. PNAs are peptide nucleic acid analogues that bind RNA with high sequence specificity, forming highly stable RNA–PNA hybrids and increasing the stability and affinity of these duplexes by significantly increasing their melting temperatures (13). In contrast to previously described methods, antisense affinity capture does not require chemical or sequence modification of the RNA of interest. Moreover, these methods are suitable for cells that are difficult to transfect with exogenous expression constructs, theoretically enabling any endogenous RNA of interest to be isolated. The major caveat of these antisense methods is that due to RNA secondary structure, some mRNA sequences may not be accessible to antisense oligonucleotides, and therefore, the design of a working fishing probe is a challenging and time-consuming step. Moreover, these methods have been thus far applied to isolate RNP complexes that are extremely abundant in cells, and therefore, these strategies may not be suitable for purification of all RNPs. Several of the most prominent examples of antisense affinity capture methods are isolation of U4/U6 snRNPs from HeLa nuclear extracts (70), purification of telomerase complexes (71), and more recently, identification of protein components constituting the small nucleolar ribonucleoprotein (snoRNP) complex MBII-52 from mouse brains (72). Furthermore, a comprehensive analysis of an mRNA-bound proteome referred to as interactome capture, which used immobilized oligo(dT) probes, was performed to capture and analyze poly(A)-tailed RNAs and their interacting proteins (73,74).

PURIFICATION OF IN VITRO- AND IN VIVO-ASSEMBLED RNA–PROTEIN COMPLEXES

The methods used to isolate novel RNA-binding protein complexes can be grouped into two main classes: in vitro and in vivo purification strategies (14,15,75). To date, most of the known RNA–protein interactions have been identified using in vitro RNA pull-down assays (Figure 1). As shown in Supplementary Table S1, the in vitro reconstitution of RNP complexes employs various RNA-tagging strategies, such as covalent linking, biotin labeling, or the introduction of artificial or natural aptamers, which ensure that the RNA of interest is immobilized to a solid support. To increase the specificity of the identification of in vitro interactions, intensive efforts have been undertaken to improve existing purification strategies, which predominantly rely on aptamer structure stabilization and the use of specific elution strategies or quantitative proteomics (54,57,58,76). Although in vitro methods are the most widely used, controllable and efficient strategies, they depend on the formation of RNA–protein complexes from synthetic target RNAs and cell extracts. These complexes may not represent genuine RNA–protein complexes because immobilized RNAs may not fold properly and nonspecific RNA protein interactions can form during the purification process. These methods also do not allow the analysis of RNA–protein interactions that are formed in response to environmental conditions.

Figure 1.

In vitro RNA affinity capture approaches. Four strategies can be used to immobilize an in vitro transcribed RNA of interest on a solid support. (1) RNA can be covalently linked to a solid support. (2) RNA can be chemically tagged through incorporation during in vitro synthesis of biotin-containing ribonucleotides or after transcription by attachment of a biotin tag by T4 RNA ligase. In this case, immobilization of target RNA is possible due to interactions between biotin and streptavidin beads. (3) Various natural or artificial aptamers can be attached co-transcriptionally to the RNA of interest. Using this tagging strategy, the RNA of interest is bound to chromatographic support through aptamer-ligand interactions. (4) RNA baits can be also isolated with antisense oligonucleotides, which are coupled with various beads. RNAs (1, 2, 3, 4) are then used to assemble ribonucleoprotein complexes using cell lysates. After incubation with various cellular extracts, RNAs with associated proteins are pulled down and washed to remove non-specifically bound proteins. Then, RNA-binding proteins are released from RNA using various elution strategies depending on the used methods. Eluted proteins are usually separated by SDS-PAGE, and the protein composition is analyzed by mass spectrometry (MS). The authenticity of the MS data is subsequently confirmed using different validation methods.

To overcome the limitations that are associated with in vitro RNA purification methods, in vivo purification approaches that capture RNA–protein complexes that are present in living cells can be performed (Supplementary Table S1). These in vivo methods permit the RNA bait to form in a native, intracellular environment; therefore, protein complexes that are more physiologically relevant are formed (Figure 2). However, these experiments are more technically challenging, especially if the target RNA is of low abundance in the cell, and result in a relatively low recovery of the RNP complexes (usually not exceeding 20%) (46,49,51,72,76). Another advantage of in vivo strategies is the application of intracellular RNA–protein crosslinking, which can be performed to ‘freeze’ physiological RNA–protein complexes and to preserve transient protein-RNA interactions within the cell. Furthermore, crosslinking approaches enable specific RNA-binding protein complexes to be purified under fully denaturing conditions, thereby limiting the identification of false-positive contaminant interactions.

Figure 2.

Selected in vivo RNA-centric approaches to identify novel RNA-binding proteins. (A) The MS2 in vivo biotin-tagged RNA affinity purification (MS2-BioTRAP) strategy relies on the co-expression in living cells of a MS2-tagged RNA of interest and MS2 coat proteins fused to an HB tag, which contains a signal sequence for in vivo biotinylation. Generally, these recombinant MS2 coat proteins are stably expressed. After UV crosslinking, cells are lysed, and the associated proteins are captured by streptavidin-coupled beads. (B) Interactome capture allows identification of RBPs that specifically associate with mRNAs in living cells. This approach employs two strategies that differ in the type of in vivo UV crosslinking that covalently links RNAs with interacting RBPs: conventional crosslinking (cCL-254 nm) and photoactivatable-ribonucleoside enhanced crosslinking (PAR-CL-365 nm). After cell lysis, covalently bound RBPs are isolated using oligo(dT) magnetic beads. (C) Peptide nucleic acid (PNA)-assisted identification of RBP (PAIR) technology uses a specific mRNA-binding probe, PNA, containing the photoactivatable amino acid adduct p-benzophenylalanine (Bpa). PNA can cross the cell membrane of living cells due to coupling with a cell-penetrating peptide and hybridizes to complementary sequences of the endogenous RNA of interest. UV light induces covalent crosslinks between Bpa and the nearest RBP. After cell lysis and RNase treatment, PNA-RBP complexes are captured by hybridization of a biotinylated oligonucleotide antisense to PNA, coupled to streptavidin. (D) The CRISPR/RdCas9 system may represent a future RNA-based approach. Using this system, proteins bound to endogenous unmodified RNAs of interest could be captured using catalytically inactive biotinylated dCas9 tethered to streptavidin beads. Specific recruitment of dCas9-guide-RNA to a given RNA is possible using a protospacer adjacent motif (PAM) in trans as a separate DNA oligonucleotide. UV crosslinking of living cells before pull-down experiments might additionally increase the specificity of the identified proteins. (A, B, C, D) After stringent washing conditions to remove non-specific RNA–protein interactions, bound proteins are eluted from RNA and subjected to proteomic analyses. Obtained MS data are subsequently validated to confirm biologically relevant RNA–protein interactions. (*) in case of PAIR technology cell lysis is followed by RNase treatment.

Two main strategies are extensively used to purify in vivo-formed RNP complexes; these strategies differ in the presence or absence of target RNA sequence modification. Aptamer-tagged RNA affinity purification uses the first strategy, and antisense RNA capture uses the second. In methods using various RNA affinity tags, cells are transfected with plasmids expressing aptamer-tagged RNAs of interest, and depending on the strategy used, additional plasmid-expressed proteins that allow for the specific isolation of RNP complexes. Several improvements in aptamer strategies have been developed to significantly increase the purification of biologically relevant endogenous RNA–protein interactions (44,46,48,49).

The RAT system

One of them was the RNA affinity in tandem method (RAT), in which efficient isolation of in vivo assembled RNP complexes was achieved by two steps of purification due to the presence in the target non-coding RNAs of two affinity tags: PP7 and tobramycin (48,49). Using this method, endogenously formed RNP complexes were first recovered by recombinant PP7cp tagged with a TEV protease cleavage site. Following elution by TEV protease, tagged RNP complexes are bound to tobramycin resin and eluted by elevation of the buffer pH and denaturation. The high RNP yield and enrichment predominantly results from the strong association between recombinant PP7cp and its cognate binding site. Additionally, this optimized form of PP7cp, in addition to higher RNA binding affinity, had also negligible dimer aggregation properties. It appears that despite exogenous expression of RAT-tagged target RNAs, this method allows purification of physiologically relevant RNP complexes because of the similar expression levels of tagged and endogenous RNAs of interest (48,49).

The RaPID system

Another optimized in vivo system, RBP purification and identification (RaPID), uses a novel fusion protein, MS2cp-GFP-SBP, which enables both affinity purification of RNA–protein complexes using streptavidin-conjugated beads due to interaction with SBP and visualization of intracellular localization of target mRNAs bearing the MS2 aptamer using microscopy due to a fluorescence reporter (MS2cp-GFP) (44,77). Specific and strong binding of MS2cp to the MS2 loops and of SBP to streptavidin as well as formaldehyde crosslinking allows for stringent washing conditions, which prevent isolation of nonspecific RNA–protein interactions. Another advantage of RaPID analysis is simple and specific elution of RNP complexes by competition with biotin. Additionally, to avoid MS2cp aggregation following high expression levels, genes encoding fusion proteins were placed under the control of an inducible promoter, which allow tightly regulated stable expression in both yeast and mammalian cells (44,77).

The MS2-BioTRAP system

An in vivo biotin-tagged RNA affinity purification (MS2-BioTRAP) approach that integrates an RNA-tagging strategy with UV crosslinking and SILAC-based quantitative MS has also been developed for the isolation of in vivo assembled RNA–protein complexes under native or fully denaturing conditions (Figure 2A) (46). In this strategy, the target RNA is tagged with 4 copies of MS2 aptamers that are recognized by optimized MS2cp that is linked to a specific signal sequence for in vivo biotinylation (HB tag). The presence of this specific biotinylation tag allows for the rapid and efficient one-step purification of target RNA together with its associated protein complexes by endogenously biotinylated MS2cp-HB proteins (serving as a ‘fishing rod’) and streptavidin-coated beads. MS2-BioTRAP was successfully used to identify proteins that are associated with cellular IRES elements (46). Recently, this method was also applied to confirm in vivo MS data that were obtained from an in vitro analysis of RNA-binding proteins that interact with mouse Nanog mRNA (78).

Antisense affinity capture methods

In general, one of the major limitations of using the in vivo aptamer tagging of target RNAs is the requirement for cell transfection, and thus, analysis of exogenously assembled RNA–protein complexes. Furthermore, gene overexpression may result in the aggregation or mislocalization of interacting molecules, thus leading to the formation of non-physiological RNA–protein complexes. Additionally, aptamer-tagged RNAs cannot be used in cells that are difficult to transfect. Therefore, methods utilizing nucleic acid hybridization, in which there is no need for the modification of RNAs of interest, are also used for the purification of in vivo-assembled RNP complexes (Supplementary Table S1). Interactome capture has allowed the complete repertoire of predominantly mRNA interactome proteins in cultured cells (HeLa and HEK293 cells, embryonic stem cells, yeast) to be identified using antisense oligonucleotides that are complementary to polyA tails (Figure 2B) (73,74,79,80). An important feature of the interactome capture technique is the use of two conditions that promote only direct RNA–protein UV crosslinking: (i) conventional UV crosslinking (cCL), in which cells are irradiated with UV light at 254 nm to crosslink naturally photoreactive nucleotide bases with amino acids such as Phe, Trp, Tyr, Cys and Lys and (ii) photoactivatable-ribonucleotide-enhanced crosslinking (PAR-CL), which utilizes photoactivatable nucleotide 4-thiouridine incorporation into RNA in living cells and subsequent UV irradiation at 365 nm. This method can be applied to study RNP composition under different biological cues and environmental stimuli. More recently, modification of the interactome capture method has been developed, which allows identification of mRNA–protein interactions with subcellular resolution due to a multiple purification procedure (81). Using this strategy, called serial RNA interactome capture, the first human nuclear RNA interactome in myeloid leukemia cells was obtained (81). Despite the notable advantages of interactome capture over the above-mentioned purification techniques, this approach is limited to identification of only mRNA-binding proteins. It appears that the PAIR technology has the potential to isolate RBP complexes for any RNAs expressed in living cells (Figure 2C) (13,82,83). This method uses a cell membrane-penetrating peptide to efficiently deliver a linked PNA oligomer, which is complementary to target endogenous RNAs, into living cells. The presence of a photoactivatable amino acid, p-benzoylphenylalanine, in the PNA sequence promotes capture of the adjacent RBP by UV crosslinking. Endogenously formed RNA–proteins complexes can be isolated by using a biotinylated PNA oligomer coupled to streptavidin magnetic beads that is complementary to a specific RNA of interest. In general, PAIR technology, such as interactome capture, can be applied to analyze RBP dynamics in varied biological settings and can be compatible with quantitative proteomics. Using PAIR, it was possible to identify RBPs associated with ankyloses RNA (13).

The CRISPR/RdCas9 system

As has been shown recently, the CRISPR/Cas9 system that has been designed for genome editing can also be used for RNA-guided binding and/or for the cleavage of specific ssRNA sequences. The developed system (CRISPR/RdCas9) has great potential because it can be used to selectively and effectively purify endogenously expressed, untagged RNA–protein complexes from cells (Figure 2D) (84). This system comprises three crucial elements: nuclease-inactive Cas9 protein (dCas9), a dCas9-associated guide RNA (sgRNA) that matches the target ssRNA, and a short PAM motif-presenting DNA oligonucleotide (PAMmer). This PAMmer is presented in trans and sits upstream of the target ssRNA sequence. The system needed some modification to capture specific ssRNA sequences. First, the original Cas9 protein was mutated (D10A;H840A) to abolish its catalytic activity, thereby preventing cleavage of the target ssRNA. Second, a 5′ overhang covering part of the target sequence was added to PAMmer beyond the PAM motif to generate specific RNA recognition that is programmed by sgRNA. Third, the PAMmer sequence was mismatched at the PAM motif to achieve a specificity of the CRISPR/RdCas9 system for RNA rather than the corresponding DNA locus. Finally, to prevent the cellular RNase-H-mediated cleavage of target ssRNA during the pull-down step, the PAMmer DNA oligonucleotide was chemically stabilized using LNA, 2′ OMe, or 2′-F ribose modifications (84,85). Thus far, this CRISPR/RdCas9 approach has only been applied to the selectively capture of an endogenous, untagged, approximately 1500-nt long GAPDH transcript from HeLa cell extracts under physiological salt conditions. As a proof-of-concept, dCas9 was bound to solid resin by the site-specific biotin labeling of dCas9. However, the complement of proteins associated with GAPDH mRNA was not determined (84). More recently, the application of this CRISPR/RdCas9 system for in vivo RNA studies was demonstrated, not for RNA–protein complex purification, but for recognizing and visualizing specific endogenous mRNAs by live confocal microscopy imaging (85). To track the journey of mRNA from the nucleus to the cytoplasm, researchers have used a nuclear localization signal-tagged dCas9 fused to GFP, and this did not alter target mRNA abundance and localization or the amount of translated protein.

PROTEOMIC ANALYSIS OF ISOLATED RBPs

To comprehensively identify RNA-binding proteins by MS, two different approaches are being used: non-quantitative and quantitative proteomics. The non-quantitative analysis requires one or two dimensional gel-based separation of eluted proteins from the RNA bait and the control purifications. The gel is stained for total protein, and protein bands that are exclusively present in the RNA bait proteome and not in the control purification are then processed and identified by MS analysis. The major caveats of this non-quantitative proteomic approach are that it impedes identification of proteins that due to a low abundance are not visible on a gel, and it results in difficulties in confirming the binding specificity of the abundant cellular proteins that also bind, to a certain extent non-specifically, to the solid support (beads) (86). To overcome these issues, quantitative proteomics can be used. There in a single MS analysis quantification and comparison of the protein interactomes from control and RNA bait purifications can be simultaneously performed. Several different approaches have been developed to perform this type of analysis (Figure 3) (87). These strategies rely on the chemical (ICAT, iTRAQ, dimethyl, 18O labeling) or metabolic labeling (e.g. SILAC, 15N, 13C labeling) of either already isolated protein complexes or whole cells/organisms (Figure 3). Following isotopic labeling, differentially tagged protein pools are simultaneously analyzed by MS and compared to provide direct quantification. Among the quantitative proteomics strategies, the simplest, most reproducible and most frequently used approach to identify RNA-binding proteins is stable isotope labeling by amino acids in cell culture (SILAC) (Figure 3B) (88,89). In this method, by growing two cell populations in medium containing different amino acid isotopes (e.g. ‘light’ lysine or ‘heavy’ lysine), cells are metabolically labeled to generate differentially tagged protein pools for MS analysis. During these analyses, the isotopically labeled proteins from different cells can be compared to provide direct relative quantification. The main advantage of the SILAC approach is that it allows identification of true binding partners from non-specific interactors by comparing the ratios of peptides from the experimental and control samples. Moreover, the SILAC approach can monitor the changes between different interactomes in terms of protein composition (control versus RNA bait) and also enables quantification of various post-translational modifications (e.g. phosphorylation, acetylation, ubiquitination) (90). Furthermore, this technique was already successfully used to identify many novel RNA-binding proteins that interact with various classes of RNAs, such as pri- or pre- miRNAs (30), telomeric repeat-containing RNA (91), spliceosomal snRNAs (92,93), viral RNAs (61,76) and expanded GGGGCC hexanucleotide repeats (94).

Figure 3.

Major quantitative proteomic approaches for the identification and analysis of proteins associated with specific RNAs. The stage in each workflow when samples are isotopically labeled for quantitative MS analysis is indicated by blue (light – control sample) and red (heavy – RBPs of interest). (A) The exception is label-free quantitation, where the samples are separately collected, prepared, and analyzed by MS, after which the data from the control and studied sample are compared using multiple approaches (peak intensities and spectral counting). To account for any experimental variations, label-free quantification experiments should be more carefully controlled than stable isotope approaches. (B) In the case of metabolic labeling of cells in culture, e.g. SILAC, labeling of proteins is performed in vivo by growing cells in medium containing different isotope-labeled amino acids, with arginine (R) and lysine (K) being the most commonly used. The cells used for control purification (RNA aptamer tag only) and cells that express the aptamer-tagged RNA of interest are grown in light medium (R0K0) and heavy medium (R10K8), respectively. Then, the cells or protein extracts used as a source for RBP isolation are combined and processed together for the quantitative analysis. By combining these samples early in the labeling workflow, this strategy has the lowest risk of experimental bias. (C) If metabolic labeling is not possible (e.g. human tissues) or cost prohibitive (e.g. mouse model organisms), alternative approaches, such as chemical or enzymatic labeling of isolated RNA-binding proteins, are applied. This labeling can be performed by adding isotopic (ICAT, dimethyl labeling) or isobaric mass tags (iTRAQ, TMT) to already purified proteins or to peptides generated after proteolytic cleavage. The resulting differentially labeled peptides, from control and the sample of interest, are then pooled together to be analyzed by MS. Unlike isotopic labeling methods that use MS1 precursor ion spectrum for relative quantification, when isobaric mass tags of identical masses and chemical properties are chosen, relative quantification is obtained from MS2 spectra representing peptide fragment ions generated after collision-induced dissociation.

VALIDATION OF MS DATA

Generally, MS analysis of RNA pull-down assays identifies tens of proteins, among which only some directly bind to a given RNA. Moreover, many of the identified proteins can interact with RNA nonspecifically/non-physiologically or indirectly through other proteins. Therefore, one of the required steps in RNA affinity purification is validation of MS-identified RNA–protein interactions. A variety of in vitro and in vivo methods allow not only verification of the authenticity of MS data but also provide a detailed analysis of RNA–protein interactions. Using in vivo methods is particularly important because they indicate which RNA–protein interactions are truly biologically relevant. Interestingly, these methods have also been successfully applied to identify putative RNA–protein interactions. In this case, proteins that potentially bind to the RNA of interest are selected based on the sequence, structure or functional similarities.

Among in vitro methods that are commonly used in validation of MS data are filter binding assay (FBA) and electrophoretic mobility shift assay (EMSA). These methods are also mainstays for determination of direct RNA–protein interactions. FBA is based on the premise that proteins, but not RNA molecules, can bind to nitrocellulose membrane filters; therefore, only RNAs associated with proteins can be retained on a membrane filter and assayed, while unbound RNA will pass through (95,96). EMSA relies on the fact that the electrophoretic mobility of RNA–protein complexes is usually lower than that of the free RNA (11,97–99) The typical FBA or EMSA experiment is performed with a constant trace amount of 32P-labeled RNA that is titrated with increasing concentrations of purified protein, generally recombinant protein. Because radiolabeled RNA is used, FBA and EMSA are extremely sensitive assays. Additionally, EMSA, in contrast to FBA, allows not only the determination of binding affinities but also the post-electrophoretic visualization of the RNP complexes. Despite the many advantages of FBA and EMSA, these methods suffer from the limitation that the binding reaction is not measured in free, aqueous solution, and therefore, they do not provide real-time kinetic data. When true equilibrium conditions are necessary, particularly with highly dynamic RNA–protein interactions, fluorescence anisotropy (FPA) can be employed. This solution-based technique measures the rate of depolarization of a fluorophore during its lifetime, which is often in the low nanosecond range and depends on the molecular volume and/or flexibility of the labeled RNA molecules (100–102). In general, RNA–protein complexes, due to their larger size, rotationally diffuse more slowly and retain more emission polarization than RNA alone, which rapidly rotates and more effectively depolarizes the emission. Thus, changes in FPA reflect the dynamics of protein binding to a fluorescent RNA substrate.

UV crosslinking assays and footprinting are other in vitro methods that can be used to confirm RNA–protein interactions identified by MS. The first method utilizes UV irradiation to covalently attach any proteins that are directly bound to the RNA of interest (103,104). Briefly, radiolabeled RNA probes and purified recombinant proteins or protein extracts are incubated to form RNP complexes spontaneously. The binding reaction is then exposed to UV light, followed by RNase digestion to remove RNA not covalently bound to the protein. The RNP complexes are analyzed by SDS-PAGE, and the signals visualized by phosphoimaging (103,104). The second assay, also known as a protection assay, is based on the ability of ligands (e.g. proteins) to protect RNA from cleavage at its binding site. Therefore, in addition to identification of an interaction between RNA and protein, footprinting also allows a detailed analysis of the sequence recognized by a given protein (105,106). In a classical footprinting assay, the RNA fragment, which is usually radiolabeled at one end, is cut by a chemical or enzymatic cleavage agent (sequence- and structure-specific RNases) in the presence or absence of the protein of interest. After cleavage, the resulting ladders are resolved on a denaturing polyacrylamide gel and visualized by autoradiography. The gaps in the array indicate RNA binding sites protected by the protein (105,106).

The yeast three-hybrid (Y3H) system is a useful tool to validate RNA–protein interactions in an in vivo context. This powerful genetic method involves the expression in yeast cells of three chimerical molecules, which assemble to activate the reporter genes, HIS3 and LacZ (16,107,108). The first hybrid protein consists of a DNA-binding protein linked to an RNA-binding protein. The second chimeric protein is a fusion protein of a transcription-activating domain and the RNA-binding domain of interest. The third element of Y3H is a hybrid RNA molecule that promotes the interaction of the two hybrid proteins by providing the two specific RNA targets for the RNA-binding proteins. However, this approach requires a careful interpretation because the most common problem of Y3H analysis is the detection of a large number of false positive clones that may prohibit isolation of genuine RNA–protein partners. Therefore, it is necessary to perform additional control steps to confirm the biological relevance of the identified interaction (16,107,108).

Antibody-based RNA-immunoprecipitation (RIP) is a very common method for studying in vivo-formed RNA–protein complexes. Using a high-quality antibody, an RNA-binding protein is immunoprecipitated together with its associated RNA, which can then be analyzed using RT-PCR, qPCR or next-generation sequencing (109–111). Furthermore, using different lysis buffers, it is possible to isolate either weakly or strongly associated RBPs. However, due to the post-lysis reorganization of co-immunoprecipitated complexes (112) and the inability to distinguish direct from indirect RNA–protein interactions, several improvements in RIP analysis, such as CLIP (113,114), HiTS CLIP (115,116), PAR-CLIP (18,117), iCLIP (118,119) and eCLIP (120,121), have been developed. These approaches not only confirm the direct binding of proteins to the RNA of interest but also determine the exact binding site of the RNA.

A powerful tool for validation of MS data in vivo is various microscopic analyses. RNA fluorescence in situ hybridization combined with immunofluorescence (FISH-IF) allows detection and localization of RNA–protein complexes at the cellular level in fixed cells or tissues (12,122,123). FISH-IF utilizes fluorescently labeled nucleic acid probes complementary to the desired RNA and antibodies detecting the protein of interest. Recent advances in RNA FISH, such as single-molecule FISH (smFISH), allow for the analysis of proteins co-localizing with individual RNA molecules in single cells (123). However, because of the limitations of optical resolution, the FISH-IF technique does not provide evidence for the direct interaction between proteins and RNA. Recently, various methods based on adaptation of proximity ligation assay (PLA) have been successfully used for more precise determination of RNA–protein interactions in situ (124–126). PLA uses proximity probes—oligonucleotides attached to antibodies against two epitopes—to guide the formation of circular DNA strands when bound in close proximity (<40 nm). This newly created DNA molecule can serve as template for localized rolling-circle amplification (RCA), which results in coiled-single stranded DNA. These PLA products can be easily detected by hybridizing complementary fluorescently labeled oligonucleotides (127–129). Among PLA-based microscopic analyses that allow visualization and quantification of RNA–protein interactions in situ in single cells with single-interaction sensitivity is a method that combines peptide-modified, multiply-labeled tetravalent RNA imaging probes (MTRIPs) with proximity ligation and RCA (124,130). Fluorescence resonance energy transfer (FRET) analysis enables direct visualization of RNA–protein complexes in live and fixed cells is. FRET is a photophysical phenomenon in which energy transfer between two adjacent fluorophores, donor and acceptor, are strongly dependent on the distance between these molecules (102,131). Therefore, FRET is observed only when the donor- and acceptor-labeled molecules are in close proximity (typically, 2–10 nm) and can directly indicate the interaction. Typically, RNA is stained with SytoxOrange or labeled with MS2, while the RNA-binding protein is tagged with a fluorescence protein (130,132,133). Using FRET, it is possible to analyze temporal and spatial association of proteins with RNA inside cells. Recent advances in developing the CRISPR/RdCas9 system have resulted in the ability to track the dynamics of RNA–protein interactions in live cells (85). An indisputable power of this method is the recognition of unmodified, endogenous RNAs. Most techniques that are used for live-cell RNA imaging require the incorporation of exogenous tags that might affect RNA folding, localization or stability (134). The CRISPR/RdCas9 system has the potential to be highly selective towards its target transcripts, as suggested by smFISH-IF results (135).

METHODS FOR IDENTIFYING PROTEINS THAT BIND TO EXPANDED RNA REPEATS

Most of the previously described RNA affinity purification approaches are used to study various types of RNA–protein interactions that are essential for the normal regulation of RNA biogenesis, RNA processing and stability, or mRNA translation (Supplementary Table S1). However, in some diseases, abnormal RNA–protein interactions occur, which disrupt the physiological function of bound proteins. Such diseases include dominantly inherited disorders that are associated with the presence of long repeat expansions in the non-coding or coding regions of individual genes (Figure 4). In this case, the interaction between proteins and mutant RNA often results in the immobilization of these proteins in specific structures that are termed RNA foci, which are the pathogenic hallmarks of this type of disease (136–139). The identification of proteins that are bound to CUG repeats in myotonic dystrophy type 1 (DM1) led to the development of the RNA gain-of-function model, which posits that expanded repeats sequester RNA-binding proteins from their normal function (140–143). Since then, proteins that are trapped not only by CUG repeats but also by CAG, CGG, CCUG, AUUCU and GGGGCC repeats have been the subject of many studies, the results of which have broadened our understanding of the pathomechanisms of these disorders (Figure 4) (Table 1 and Supplementary Table S2). These abnormal RNA–protein interactions affect the alternative splicing of specific pre-mRNAs (144), alter the use of alternative polyadenylation sites of a number of mRNAs (145), change nuclear transport and export (146,147), affect translation (148), induce nucleolar stress (94), and dysregulate miRNA processing (149). What makes these expanded repeats so attractive for cellular proteins? The structures that are formed by the expanded repeat RNAs likely trigger protein recruitment. Depending on the type of expansion involved, mutant transcripts can adopt in vitro stable hairpin or G-quadruplex structures (94,150,151). Despite intensive studies on RNA-mediated toxicity in these disorders, many questions regarding the molecular disease-causing mechanisms remain unanswered. Determination of the complete protein composition of these specific RNP complexes will help to elucidate these RNA gain-of-function mechanisms. Proteins that are recruited into mutant RNA foci may be attractive therapeutic targets for these as-yet incurable disorders. In the following chapter, we briefly describe approaches that have been used thus far to identify proteins that bind to various types of expanded repeats.

Figure 4.

Schematic representation of the RNA gain-of-function mechanism in the selected repeat expansion disorders. Upper: Localization of expanded simple repeats in disease genes. Middle: Transcripts containing repeat expansions function as pathogenic agents via the sequestration of specific RNA-binding proteins, resulting in their impaired cellular function. Lower: Methods applied to capture and identify proteins that are associated with different expanded repeats.

Table 1.

Proteins interacting with expanded repeat RNAs identified by RNA pull-down assays combined with WB or MS

Type of repeats	RNA affinity capture	Extract source	Identified protein	Validation method	Reference
CAG
(CAG)15, 128 Biotin-tagged RNA	RNA pull-down and WB	Cytoplasmic extract from human brain	PKR	IHC	(172)
(CAG)27, 78 S1 aptamer-tagged RNA	RNA pull-down and WB	Transgenic flies	NCL	in vitro RIP	(173)
(CAG)20, 51 Biotin-tagged RNA	RNA pull-down and WB	Whole-cell lysate from HeLa cells	MID1 complex (MID1, S6K, PP2Ac)	Pull-down followed by WB	(148)
CGG
(CGG)105 Biotin-tagged RNA	RNA pull-down and MS	Cytoplasmic extract from mouse cerebellum	hnRNP A2/B1, Pur α	Pull-down followed by WB, EMSA, RIP	(181)
(CGG)60 Biotin-tagged RNA	RNA pull-down and MS	Nuclear extract from mouse brain and COS7 cells	Sam68 and other 37 proteins identified	FISH-IF	(183)
(CGG)20, 60, 100 Biotin-tagged RNA	RNA pull-down and MS	Nuclear extract from mouse brain	DROSHA, DGCR8 and other 30 proteins identified	Pull-down followed by WB, FISH-IF, EMSA, UV crosslinking	(149)
CUG
(CUG)85 Biotin-tagged RNA	RNA pull-down and MS	Nuclear extracts from HeLa cells	hnRNP H	FISH-IF, UV crosslinking	(103)
(CUG)95 Biotin-tagged RNA	RNA pull-down and MS	Nuclear extracts from HeLa cells and myoblasts and myotubes from C2C12	p68/DDX5 and other 100 proteins identified	FISH-IF, EMSA	(104)
AUUCU
(AUUCU)500 Biotin-tagged RNA	RNA pull-down and MS	Nuclear extract from mouse brain	hnRNP K	Pull-down followed by WB, RIP, FISH-IF	(185)
GGGGCC
(GGGGCC)23 Biotin-tagged RNA	RNA pull-down and MS	Nuclear extract from HEK293 cells	hnRNP A3,hnRNP A1 and other 20 proteins identified	Pull-down followed by WB	(194)
(GGGGCC)10 Biotin-tagged RNA	RNA pull-down and MS	Whole-cell lysate from mouse spinal cord	Pur α, Pur β, Pur γ	Pull-down followed by WB, FBA, RIP	(195)
(GGGGCC)30 Biotin-tagged RNA	RNA pull-down and MS	Nuclear extract from mouse brain	hnRNP H1, hnRNP H2 and other 30 proteins identified	FISH-IF	(196)
(GGGGCC)72, 48 Biotin- and S1 aptamer-tagged RNA	RNA pull-down and WB	Nuclear extract of SH-SY5Y cells and rat brain cortex	hnRNP-H	FISH- IF	(190)
(GGGGCC)4 Biotin-tagged RNA	RNA pull-down and MS	HEK293T (SILAC)	NCL, hnRNP U and other 81 proteins identified	Pull-down followed by WB, FISH-IF	(94)
(GGGGCC)5 Biotin-tagged RNA	RNA pull-down and MS	Whole-cell lysate and nuclear extracts from SH-SY5Y cells, total extracts from human cerebellum	ALYREF, SRSF1, SRSF2, hnRNP A1, hnRNP H1/F and other 103 proteins identified	FISH-IF, UV crosslinking	(197)
(GGGGCC)6.5 5′Cy5-labeled RNA	Proteome array	in vitro	ADARB2 and other 19 proteins identified	FISH-IF, RIP, EMSA	(199)
(GGGGCC)31 Biotin-tagged RNA	RNA pull-down and MS	Whole-cell lysate from mouse brain and spinal cord	hnRNP H, eIF2α, eIF2β, RAX, ILF3	Pull-down followed by WB, FISH-IF	(198)

IHC, immunohistochemistry; WB, western blotting; FISH, fluorescence in situ hybridization, IF, immunofluorescence; RIP, RNA immunoprecipitation; EMSA, electrophoretic mobility shift assay; FBA, filter binding assay.

CUG/CCUG-binding proteins

CTG repeat expansions (50 to >3500 repeats) in the 3′ UTR of the DMPK gene and CCTG repeat expansions (75 to ∼11 000 repeats) in the first intron of the ZNF9 gene are causative agents of myotonic dystrophy type 1 (DM1) and myotonic dystrophy type 2, respectively (152–154). Proteins that bind to CUG/CCUG repeat expansions were initially identified using in vitro biochemical assays, such as EMSA and the UV crosslinking assay, followed by immunoblotting. These analyses relied on the assumption that proteins that can bind dsRNA or proteins that localize to the nucleus potentially interact with expanded CUG/CCUG repeats. The potential interactors were subsequently assayed for their co-localization with expanded RNAs using FISH-IF (Supplementary Table S2). In the case of EMSA analysis, in vitro-transcribed and radiolabeled expanded CUG/CCUG RNA was incubated with whole cell protein extracts or with a set of pre-defined, purified recombinant RBPs. The first approach identified the alternative splicing regulator CUG-BP (CUG triplet repeat, RNA-binding protein 1) as well as two CCUG-interacting multiprotein complexes: the 20S catalytic core complex of the proteasome and the CUG-BP1–eIF2 complex, which is involved in translational regulation (140,141,155). Using the second approach, CUG-binding properties were determined for mETR3 (muscarinic acetylcholine receptor M3) and PKR (interferon-induced, double-stranded RNA-activated protein kinase) (156,157). The UV crosslinking of whole cell extracts with radiolabeled RNA is also commonly used to detect protein-RNA interactions. Using this method, alternative splicing regulator MBNL1 (muscleblind-like protein 1) was identified as a protein that binds to CUG repeats (142). Structural studies showed that at least 20 CUG repeats are needed for the RNA to acquire the RNA hairpin secondary structure that functions as a sequestration trigger and is necessary for MBNL1 binding (143,156,158). FISH-IF not only confirmed MBNL1 co-localization with expanded CUG repeats but was also the first approach to show that MBNL1 possibly interacts with expanded CCUG RNAs (159). Interestingly, FISH co-localization of MBNL1 with nuclear CUG RNA inclusions was also demonstrated for the rarely occurring CUG repeat-related disorders spinocerebellar ataxia type 8 (SCA8) (160) and Huntington's Disease-like 2 (HDL2) (161). Using additional in vivo and in vitro approaches, it was shown that sequestration of MBNL1 by CUG/CCUG repeats leads to abnormalities in alternative splicing (144,162), mRNA localization and transport (163), stability (164), and microRNA biogenesis (165); these abnormalities which affect many cellular functions, resulting in disease. RIP has also been used to identify proteins that interact with expanded CUG repeats. Using RIP, interactions between expanded CUG repeats containing transcript and Staufen 1 (which is involved in mRNA transport, stability, and translation) (166), DDX6 (ATP-dependent RNA helicase) (167), and two transcription factors, Sp1 and RARγ (168), were examined. Only two studies attempted to comprehensively examine the complement of proteins that bind to expanded CUG repeats (Table 1). In both studies, RNA affinity chromatography with in vitro transcribed and biotinylated CUG RNAs combined with MS analysis was performed to identify novel factors that bind these repeats (103,104). Among the newly identified proteins were two helicases, p68 (DDX5) and p72 (DDX17), which play a role in remodeling of RNA and RNA-complexes, and a key regulator of mRNA metabolism, hnRNP H. Some of these proteins were further shown to affect nuclear retention of mutant RNA or to favor MBNL1 binding to the CUG/CCUG repeats (103,104).

CAG-binding proteins

Expanded CAG repeats that are localized in the coding sequences of nine functionally unrelated genes are molecular triggers for a group of neurodegenerative disorders termed polyglutamine (polyQ) diseases. There are at least nine polyQ disorders, including Huntington's disease (HD), spinocerebellar ataxia (SCA) types 1, 2, 3, 6, 7 and 17, dentatorubral-pallidoluysian atrophy (DRPLA) and spinal and bulbar muscular atrophy (SBMA). These diseases take their name from mutant proteins containing long polyQ tracts. In the most common disorders (HD and SCA3), the expression of at least 36 CAG repeats that are localized in the first exon of the HTT gene and at least 60 CAG repeats that are localized in the 10th exon of the ATXN3 gene is sufficient to cause the pathogenic effects, respectively (169). Expanded CAG-repeat binding proteins were identified using similar approaches to those used in studies on CUG/CCUG-interacting proteins (Table 1 and Supplementary Table S2), which test the binding of extended CAG-repeat RNA to known RNA-binding proteins (142,156,170). The first strategy that attempted to identify proteins with CAG-binding properties was the use of an EMSA and a UV-crosslinking assay with cytoplasmic protein extracts from various human tissues (171). In this study, two CAG-repeat binding proteins (63 and 49 kDa) were identified. However, no further analysis allowing the precise identification of these proteins was performed. Another frequently used in vitro approach to identify CAG-binding proteins is the use of RNA pull-down assays followed by immunoblotting with antibodies that are specific for known RBPs (Table 1). The resulting in vitro transcribed RNAs containing different-length CAG repeats (normal or mutant) were immobilized on agarose beads either covalently or using biotin or S1 aptamer, followed by incubation with protein extracts obtained from different sources. The use of this method confirmed that PKR (protein kinase R) (172), the MID1 complex (which regulates translation) (148), and nucleolin (which regulates rRNA transcription) (173) are all CAG-repeat binding proteins. Further experiments showed that the interaction between nucleolin and expanded CAG RNA triggers the perturbation of rRNA transcription and, consequently, nucleolar stress and apoptosis (173,174). Moreover, the binding of the MID1 complex to the expanded CAG RNA results in the upregulated translation of mutant HTT transcripts, thereby leading to the overproduction of aberrant proteins (148). CAG-binding proteins were also identified using Y3H assays. Using this method, the CAG RNA-binding properties of two proteins, MBNL1 and PKR, were shown (170). What is the role of CAG repeat protein interactions in the pathogenesis of polyQ diseases? Co-localization of MBNL1 with RNA containing expanded CAG repeats results in aberrant alternative splicing similar to CUG/CCUG repeats (175). Proteins that may interact with expanded CAG repeats were also identified by FISH-IF. This method, however, was only used to identify potential interactions of CAG with MBNL1 (176). Another approach used to detect proteins with CAG-binding properties was RIP. In this case, interactions between candidate proteins and transcripts containing expanded repeats were studied. Using this approach, U2AF65, which regulates nuclear export of RNA (146), and SRSF6, which is involved in regulation of splicing and translation (177), were identified. Interestingly, SRSF6 has been selected to the analysis on the basis of bioinformatics predictions, which revealed binding site of SRSF6 with expanded CAG repeats (177). For U2AF65 and nucleolin, immunoprecipitation using GST-tagged fusion proteins was performed to identify direct interactions with expanded CAG RNA (146,173).

CGG-binding proteins

The pathological expansion of CGG repeats in the 5′ UTR region of the FMR1 gene causes one of to two different disorders, depending on the length of the repeats: Fragile X-associated tremor/ataxia syndrome (FXTAS) is caused by a premutation in which the expansion is between 55 and 200 CGG, whereas Fragile X syndrome (FXS) occurs when the number of CGG repeats exceed 200 triplets (178,179). The first attempt to identify potential CGG-binding proteins started with the purification of ubiquitin-positive FMR1-mRNA inclusions (the neuropathological hallmark of FXTAS), followed by a detailed protein characterization using MS analysis (Supplementary Table S2). This indirect analysis showed that at least two RNA-binding proteins, heterogeneous nuclear ribonucleoprotein A2 (hnRNP-A2) and MBNL1, were present in these inclusions (180). To identify proteins that bind specifically to CGG repeats, in vitro transcribed, biotinylated, expanded CGG-repeat RNA was used as bait (181). Putative CGG repeat-binding proteins were purified by binding to streptavidin magnetic beads, and protein identification was performed using MS analysis. Among the proteins selected for further studies were the known transcription and RNA transport regulators, Purα and hnRNP A2/B1 (Table 1). The MS identification of these proteins as potential CGG-repeat-binding proteins was further confirmed using other in vitro biochemical assays (EMSA and immunoprecipitation with (CGG)90-eGFP assays). Moreover, Purα was shown to be present in the ubiquitin-positive inclusions of FXTAS patient brains and in a Drosophila model of FXTAS (181). Furthermore, a medium-throughput Drosophila genetic screen was also used to identify potential CGG-binding proteins (182). Independent of MS analysis, hnRNP A2/B1 and CUG-BP were also shown to interact with expanded CGG repeats, as their overexpression suppresses the CGG-mediated eye toxicity in Drosophila (182). Other reports using in vitro transcribed and biotinylated RNA as bait to identify CGG-binding proteins by MS have also been published (Table 1). In 2010, Sellier et al. used mouse brains or COS7 cell protein extracts to identify over 20 proteins that specifically interacted with expanded CGG repeat RNA (183). In addition, FISH-IF studies revealed that only MBNL1 and hnRNP-G co-localize with mutant RNA foci in COS7 cells expressing these expanded repeats. Surprisingly, further studies demonstrated that recruitment of MBNL1 and hnRNP-G to the CGG foci required the presence of an alternative splicing regulator, Sam68, which itself does not bind directly to CGG repeats (183). Moreover, using normal (CGG)20 and mutant (CGG)60 RNA as baits followed by MS analysis, the double-stranded RNA-binding protein DGCR8 that exclusively binds to mutant RNA was identified (Table 1) (149). Further detailed studies showed that DGCR8 co-localizes to aggregates formed by mutant (CGG)60, resulting in the partial sequestration of DGCR8 and its partner DROSHA, which interestingly was not identified in MS analysis. As a consequence, the processing of miRNAs is reduced, resulting in decreased levels of mature miRNAs in neuronal cells expressing expanded CGG repeats and in brain tissues from patients with FXTAS (149).

AUUCU-binding proteins

The ATTCT pentanucleotide repeat expansion in intron 9 of the ATX10 gene has been associated with spinocerebellar ataxia type 10 (SCA10). Up to 15 repeats occur in healthy individuals, but extended repeats of 800–4500 AUUCU are found in patients suffering from SCA10 (184). To date, only one study identified proteins that bound specifically to expanded AUUCU repeats (185). In this report, purification of proteins bound to in vitro transcribed and biotinylated expanded RNA followed by MS analysis identified hnRNP K as a potential interacting protein (Table 1). Further analysis using RIP confirmed the presence of intron 9 of ATX10 after hnRNP K pull-down. Additionally, hnRNP K showed co-localization with AUUCU expanded foci in SCA10 fibroblasts and transgenic mouse brain tissues (185,186). Furthermore, it was demonstrated that sequestration of hnRNP induces alternative splicing defects in certain genes and induces caspase-mediated apoptosis in SCA10 fibroblasts (185,186).

GGGGCC-binding proteins

The expanded GGGGCC repeats located in the gene C9ORF72 have been associated with amyotrophic lateral sclerosis (c9ALS) (187,188). Less than 30 repeats are found in healthy people, while expanded repeats in the range of 500 to thousands are found in people affected with c9ALS, with an increased number of repeats associated with a more severe phenotype (188–190). The first approaches focused on the identification of proteins interacting with GGGGCC repeats were based on bioinformatics prediction followed by biochemical verification methods (Supplementary Table S2). The assumption that ASF2/SF2 and hnRNPA1 interact with GGGGCC repeats was confirmed by EMSA (191,192). Additionally, FISH-IF has been used to show co-localization of SC35, SF2 and GGGGCC repeats (190,193). As RNA gain-of-function model has been implicated in c9ALS quite recently (188), the well-established global approaches were predominantly used to identify GGGGCC repeats-binding proteins among which was a pull-down method with biotinylated RNA and MS analysis (Table 1) (94,190,194–198). Alternatively, instead of biotin, the S1 aptamer has been added in a template DNA prior to in vitro transcription (190). Because GGGGCC repeats are difficult to synthesize due to high GC content, nonpathological repeat tracts of GGGGCC were predominantly used as bait in the pull-down analyses. Shorter repeats are believed to have a similar ability and specificity to bind selective proteins as longer ones due to the existence of the same repetitive sequence motif and formation of similar tertiary structure (194). In nearly all studies analyzing the GGGGCC-interacting proteome, MS was used to identify RNA-binding proteins. A significant improvement in this technology was the introduction of the SILAC labeling technique, which allows for a comparative quantification of proteins that are bound to different RNAs (94). Proteome arrays have also been used to identify the proteins that interact with GGGGCC RNA expansions (Table 1) (199). The authors used nearly two-thirds of the annotated human proteome as yeast-expressed, 16,368 full-length ORFs with N-terminal GST-His×6 fusion proteins on a chip to which an RNA probe was hybridized. The disadvantages of this method are the limited number of proteins that can be bound on the chip and the limitations relating to already identified, known proteins. Moreover, only direct interactions with single proteins can be identified using the proteome array method because no protein complexes exist in that experimental system. Using this approach, ADARB2 (adenosine deaminase RNA-specific B2, a known RNA-binding protein and a member of the ADAR family, members of which are involved in RNA editing) was shown to bind to GGGGCC repeats (199). The hypothesis regarding the sequestration of proteins by expanded repeats was extensively studied for sense transcripts containing GGGGCC repeats. However, c9ALS is an example of a disease in which bidirectional transcription occurs and results in the production of both expanded sense GGGGCC tracts and antisense CCCCGG tracts. For antisense transcripts, it was demonstrated that hnRNP-K and PCBP2 (poly(rC)-binding protein) preferentially bind antisense CCCCGG4 over sense GGGGCC4 transcripts (94), and the association of these proteins with c9ALS pathomechanisms was further evaluated (197). In another study, antisense RNA foci were demonstrated to co-localize with SRSF2, hnRNP A1, hnRNP H/F, ALYREF and hnRNP K (200). Moreover, the spectrum of proteins that bind to GGGGCC repeats in a structure conformation-dependent manner was also analyzed. NCL (nucleolin) and hnRNP-U preferentially bind a G-quadruplex, whereas hnRNP-F and ribosomal protein RPL7 do not distinguish between hairpin and quadruplex structures. As a result, nucleolar stress is observed in c9ALS patients (94). Global analyses of proteins that bind to GGGGCC repeats showed that most of these proteins possess an RNA recognition motif (RRM) within the RNA-binding domain (RBD) or prion-like domain (PrLD). This RRM allows for RNA–protein interactions, whereas the PrLD allows for protein aggregation. Gene ontology classifications revealed that different classes of proteins recognize GGGGCC repeats: pre-mRNA splicing factors (FUS, EWSR1, hnRNP A1, hnRNP A2B1, SRSF1, SRSF2, SRSF3 and TAF15), processing, stability (SAFB2) (194,197,198), mRNA export adaptors (197), transcription activator proteins (Purα, Purβ and Purγ) (195,198), heterogeneous nuclear ribonucleoproteins (190,194,198), helicases (DDX21, DHX15, and DHX30) (194) and interleukins (ILF2) (194) (Table 1). The sequestration of proteins by RNA reduces the amount of sequestered protein in its native pathway and results in the dysregulation of cell functioning.

CONCLUSIONS AND FUTURE PERSPECTIVES

Although many different approaches have been developed to immobilize specific RNAs to affinity matrices and to identify, confirm and further examine RNA–protein interactions, serious challenges remain to be addressed. Although significant advances in mass spectrometry and peptide separation methods have enabled the reliable identification of these interactions, the capture and purification of proteins bound to specific RNAs remain great challenges. Therefore, the RNA affinity capture method used should be carefully chosen depending on the biological question that needs to be answered.

RNA-binding proteins are highly versatile factors that play crucial roles in a variety of cellular processes. However, the complement of proteins that associate with any given RNA is largely unknown, and this knowledge gap needs to be filled. Only recently, using the interactome capture strategy combined with a robust MS analysis, a comprehensive identification of RBPs associated with a total pool of mRNAs has been reported (73,74,79,80). However, these studies were performed using a limited number of cell types or tissues. Surprisingly, it was shown that over 50% of the identified RBPs lack typical RNA-binding domains, which causes difficulties in identifying these proteins using computational methods. Thus, it is very likely that numerous other RNA-binding proteins await identification in other biological sources. Furthermore, researchers know very little regarding the repertoire of proteins with which the more recently discovered lncRNAs and circular RNAs may interact (201–204). Some of these proteins, which have already been shown to interact with lncRNAs, do not resemble known RNA-binding proteins. Thus it is difficult to anticipate the structure and function of these RNA–protein complexes. Issues regarding the functioning of noncoding RNAs are likely to remain the focus of research in coming years.

The complement of proteins that are bound to a specific RNA that control its ‘life’ inside cells changes in a temporal and spatial manner. These changes in protein composition, stoichiometry and modification drive RNA molecules throughout different stages of the RNA cycle - transcription, various co-transcriptional maturation events, transport and nuclear export, translation, and finally, RNA degradation. Moreover, because these events are highly regulated by both intracellular and environmental signals, research in the field of RNA biology needs to focus on elucidating the spatial and temporal dynamics of RNA–protein networks. Thus far, predominantly static analyses of the physical association between RBP and specific RNAs have been conducted. To characterize the dynamic RNA-binding proteome, recent advances in RNA-centric, high-throughput quantitative proteomics strategies can be applied (46,89,205). The application of these methods will allow researchers to ‘freeze’, purify and identify RBP binding to specific RNAs under different conditions; e.g. during different stages of the RNA cycle, during cell cycle progression, and in cells that have been treated with different extracellular stimuli (stress conditions). Moreover, because many RNA-binding proteins interact either weakly or transiently with their target RNA throughout its ‘life’ cycle, it is crucial to streamline existing methods or to develop novel more accurate, high-throughput technologies that exploit crosslinking-MS (206,207) and next-generation sequencing (18,121,208). Thus far, among the major caveats of RNA-centric methods is their inability to explore RNA-binding proteins that associate with low abundance transcripts inside living cells (in vivo purification strategies). Therefore, the development of novel approaches is needed to isolate endogenously expressed, unmodified RNAs with increased specificity, together with their interacting proteins. This goal would likely be achieved by further developing recently described methods that rely on CRISPR/RdCas9 technology (84,85). Unlike the most commonly used methods for RNA–protein complex isolation, CRISPR/RdCas9 technology offers an incomparable means to selectively capture and identify endogenously expressed specific RNAs together with their interacting proteins without interfering with its RNA ‘life’ cycle. With all its advantages, this technology has still some challenges that need to be addressed such as efficient delivery of the system components to the cells, improvement of PAMmer and gRNA design and most importantly the decrease of CRISPR/RdCas9-mediated off-target effects. Some disadvantages of the CRISPR/RdCas9 technology might be likely circumvented by using the recently described CRISPR/C2c2 system from the bacterium Leptotrichia shahii, which, targets RNA in its native form (209).

In this review, we also highlighted the importance of studying abnormal RNA–protein complexes that accompany a class of neurodegenerative diseases; e.g. DM1, DM2, c9ALS, FXTAS, HD and a number of SCAs. As current studies show, specific RNAs containing an expanded simple repeat tract can function in pathogenesis as trapping agents for various RNA-binding proteins. Once captured, these proteins cannot properly fulfill their normal cellular roles in processes such as alternative splicing, RNA transport and nuclear export, translation, and microRNA biogenesis. The described expanded repeat-binding proteins might be only the ‘tip of the iceberg’. Most studies to date have been performed using outdated in vitro purification approaches due to the problems associated with capture and identification. Therefore, there is a great need to apply recently developed cutting-edge strategies to identify novel proteins that are trapped by these expanded RNA repeat tracts to better understand the mechanisms of pathogenesis and to develop treatments for these disorders.