To establish whether isolated CUG repeats and other trinucleotide repeats (TNRs) adopt higher-order RNA structures, two comparative studies were performed (Sobczak et al., 2003, 2010). First, using chemical (Pb²⁺ ions) and enzymatic (S1 nuclease, T1, T2 and V1 ribonucleases) cleavages, the structures of CCUG, AAG and all CNG repeat motifs (N = A, C, G or U) in solution were analyzed (Sobczak et al., 2003). In that study, a CUG motif repeated 17 times was shown to form hairpin structures composed of a stem with periodically occurring standard C-G and G-C base pairs and a single periodic U-U base pair whose nature was further examined by X-ray crystallography (Figure 1A; Mooers et al., 2005; Kiliszek et al., 2009). The terminal loop of this hairpin was composed of four nucleotides. Moreover, these CUG repeats formed several alternative, “in register” alignments, i.e., “slippery hairpins”. These hairpin variants differed in the lengths of their protruding 3′ ends. By using CUG repeat RNAs with end sequences that form stable GC-clamps, the “slippage effect” could be eliminated to produce a single CUG hairpin alignment. When an even number of the CUG repeats is clamped, a 4-nt terminal loop forms; however, 3-nt loops are present with an odd number of repeats, thus illustrating the influence of the sequences flanking the CUG repeats (and other TNRs) on the structural features and biological properties of these motifs.

More recently, a comprehensive structural study of a complete set of 20 TNRs that were repeated 17 times was carried out using a set of chemical (Pb²⁺ ions) and enzymatic (S1, Cl3, Mung bean nucleases; T1 and V1 ribonucleases) structure probing and biophysical methods (UV melting spectra, circular dichroism (CD) spectra and gel mobility analysis). As a result, TNRs have been grouped into four different structural classes: (1) unstructured RNAs; (2) semistable hairpins; (3) fairly stable hairpins; and (4) very stable G-quadruplexes. In agreement with previously described studies, CUG repeat motifs (together with CAA, CGU and other three CNG motifs) form fairly stable hairpins (Sobczak et al., 2010). In the same work, the thermodynamic stability of CNG repeats was further assessed by UV-monitored structure melting experiments (Sobczak et al., 2010). Among all TNRs which are implicated in Triplet Repeat Expansion Diseases (TREDs), the CUG motif has been shown to be the least thermodynamically stable, regardless of whether the measurements were performed in the presence of Na⁺ or K⁺ ions. The order of stability, starting from the most stable repeat, is as follows: CGG, CAG, CUG and CCG (in 100 mM NaCl) or CAG, CGG, CUG and CCG (in 100 mM KCl; Broda et al., 2005). Other calorimetric and structural (UV melting and/or CD spectroscopy) studies that also included CUG repeats have been performed (Pinheiro et al., 2002). Moreover, the hairpin structure formed by isolated, expanded CUG repeats (CUG₁₃₆) has been visualized using electron microscopy (Yuan et al., 2007).

Thus far, only one study has provided structural insights into the CUG repeat region from the 3′UTR of the DMPK transcript, which is involved in DM1 pathogenesis (Figure 2A; Napierala and Krzyzosiak, 1997). This study was the first of a series of experiments that began to probe CNG repeat structures in a wider transcript context. The analysis was performed using chemical (Pb²⁺ ions) and enzymatic (S1 and T1 nucleases) structure probing of in vitro-transcribed RNAs containing increasing lengths of CUG repeats (5, 11, 21 and 49) together with flanking sequences (30 and 35 nucleotides at the 5′ side and 3′ side of the CUG repeat, respectively). The analysis demonstrated that five repeats, which is the most common, non-pathogenic number of repeats in the population, do not form any secondary structures. Upon increasing the length of the CUG repeats, the stability of the formed structures increased: 11 repeats formed unstable hairpins, 21 repeats formed semi-stable hairpins, and the expanded 49 CUG repeats formed fairly stable hairpins. Moreover, as the sequences flanking the repeats did not “freeze” the repeat alignment, thus resulting in alternative structures, the CUG repeat hairpins are referred to as “slippery” (Figure 2A). Furthermore, increasing the CUG repeat length from, e.g., 21–49 repeats caused the increases in the length and stiffness of the repeat hairpin stem and appeared to enlarge the hairpin terminal loop (Napierala and Krzyzosiak, 1997).

By using X-ray crystallography, a number of studies have revealed the detailed structures of short CUG-containing oligoribonucleotides (Mooers et al., 2005; Kiliszek et al., 2009; Kumar et al., 2011; Coonrod et al., 2012; Tamjar et al., 2012). When crystalized, these oligomers (up to six CUG repeats) pile onto each other to form an intermolecular duplex that presumably represents the stem part of the hairpin, which is formed by long CUG repeats (Figure 1A). Furthermore, these studies showed that RNA is double helical and has general features of the A form, in which two standard C-G and G-C base pairs are interrupted by a non-canonical U-U pair. While standard Watson-Crick pairing between C-G and G-C stabilizes duplex formation, the occurrence of the periodic U-U pair contributes to the unique geometry of the helix. Unlike in most of the previously observed U-U pairs that had two hydrogen bonds, the U-U pairs in the CUG repeat context appear to be “stretched” (“stretched U-U wobble”) and to form only a single direct hydrogen bond between the N3 imino group of one uracil residue and the carbonyl O4 atom of the other (Figure 1A). In this CUG repeat duplex, the two opposing uracil residues do not create any accommodation problem because they remain well separated from each other; thus, the disruption of the overall helix structure is low (Kiliszek et al., 2009).

By using both enzymatic and chemical RNA structure probing experiments, it was initially shown that isolated CAG repeats (CAG₁₇), similar to the previously described CUG repeats, form hairpin structures in solution, with a stem composed of periodically occurring standard C-G and G-C base pairs that are divided by a single, periodic non-canonical A-A pair (Sobczak et al., 2003). These CAG repeat hairpins were shown to form several alternative alignments (“slipped hairpins”) that differ in the lengths of their protruding 3′ ends. When GC-clamps were added at both ends of the CAG repeat sequence, the slippage effect was abolished. Moreover, in this clamped configuration, a CAG repeat hairpin composed of an even number of repeats (CAG₁₆) forms a 4-nt apical loop. However, when an odd-numbered hairpin was studied (CAG₁₇), a larger 7-nt loop appeared (Sobczak et al., 2003). In another more robust study (a complete set of TNRs was assayed), it was shown that the CAG repeats (CAG_{17 or 20}) form fairly stable hairpins (3rd structural class, similar to CNG, CAA and CGU motifs; Sobczak et al., 2010). Moreover, UV-monitored structure melting experiments have shown that among all TREDs-related repeats, the CAG repeat motif is the most thermodynamically stable under K⁺ ion conditions (second to the CGG motif under Na⁺ ion conditions; Broda et al., 2005; Sobczak et al., 2010).

In recent years, using X-ray crystallographic methods, the structures of several oligoribonucleotides composed of CAG repeats have been reported (Kiliszek et al., 2010; Yildirim et al., 2013). These short RNA fragments stack onto each other, forming intermolecular duplexes that represent the stem portions of CAG repeat hairpins (Figure 1B). Furthermore, these duplexes acquire the general characteristics of RNA-A helices, where the non-canonical, periodic A-A base pairs are settled between the canonical C-G and G-C pairs that play a stabilizing role. As the positioning of two bulky adenine rings opposite each other within the helical structure seems to be “sterically challenging”, local, moderate disruption of the helix geometry, resulting in local unwinding of the helical structure and subsequent widening of the major groove, was observed (Kiliszek et al., 2010). In this CAG repeat duplex, both of the opposite adenine residues are in the anti conformation and are shifted out of the helical axis towards the major grove to avoid collision. The adenine residue, which plays a role as a H-bond donor, is shifted more, thus resulting in a “thumbs up” conformation. Furthermore, the A-A pairs in the CAG repeat context form only a single, unusual, weak hydrogen bond between the carbon atom C2-H2 of one adenine residue and the nitrogen N1 atom of the other (C2-H2⋅⋅⋅N1 hydrogen bond) (Figure 1B; Kiliszek et al., 2010). This type of A-A wobble pairing has not been previously reported. In a more recent study, by using X-ray crystallography, nuclear magnetic resonance (NMR) spectroscopy and molecular dynamics (MD) simulation analysis, different nature of the non-canonical A-A base pairs has been shown (Tawani and Kumar, 2015). In the analyzed model RNA duplex containing three CAG repeats with additional flanking sequences, conformational dynamics were suggested due to the specific hydrogen bonding pattern and stacking interactions of the non-canonical A-A base pairs. Unlike the previous reports, one of the closing A-A base pairs showed syn-anti conformation with one hydrogen bond between the exo-amino group of A(syn) and N1 atom of the A(anti) (N6H⋅⋅⋅N1 hydrogen bond). Moreover, the second closing A-A base pair, as well as, the A-A base pair located in the center of the duplex all had anti-anti conformations with no hydrogen bonds (Tawani and Kumar, 2015). As poor and ambiguous electron density maps were analyzed in this study it is still a matter of a debate whether different, dynamic A-A base pair conformations truly exist within CAG repeats.

Until now, the structures formed by CAG repeats in their native, transcript context are the most extensively studied TNRs. These studies focused on assessing whether specific sequences flanking the CAG repeats contribute to the overall characteristics of the CAG hairpin structures, i.e., the formation of the multiple alignment hairpins (“slipped” hairpin) or a single alignment hairpin (“frozen” hairpin; Michlewski and Krzyzosiak, 2004; Sobczak and Krzyzosiak, 2004a, 2005; de Mezer et al., 2011). Moreover, the influence of naturally occurring repeat interruptions on the structures formed by CAG repeats was also investigated. These structural studies were mainly performed using both chemical (Pb²⁺) and enzymatic RNA structure probing experiments with the use of a battery of probing reagents, e.g., S1, Mung Bean nuclease and RNases T1, T2, V1 and H.

In another study, the unique, complex architectures of the hairpins formed by CAG repeats present in HTT and androgen receptor (AR) transcripts, which are affected by both the specific flanking sequences and another type of neighboring, triplet repeats, have been established (de Mezer et al., 2011). As shown in Figure 2B, in the HTT transcripts expansion-prone CAG repeats [(CAG)_n] have polymorphic CCG repeat tract [(CCG)_{7 or 10 nt}] at their 3′ side that is separated by a 12-nucleotide specific sequence. This specific neighborhood causes the HTT CAG repeat variants to have a tripartite modular structure composed of the base, which is the most stable part of the stem and where the 5′ part of the CAG repeats are engaged in base pairing with CCGs; the central module, which is formed by the partially base pairing CAG repeats with 12-nucleotide specific sequence; and the terminal section, which is part of the hairpin structure and which is composed exclusively of CAG repeats. The difference between normal and mutant HTT transcripts is only restricted to the terminal section, which in the mutant transcript gets elongated, while the other structural modules remain the same. As the stability of the base module consisting of base-paired CAG and CCG repeats is much higher than other hairpin modules, the latter may form alternative structural variants resulting from the slippage effect of the CAG repeats (de Mezer et al., 2011). More recently, the structures of CAG repeats of normal and pathogenic length (HTT transcript), in the context of the entire first exon (harboring the repeats) and the neighboring 5′UTR sequence, have been investigated by in vitro selective 2′-hydroxyl acetylation analyzed by primer extension (SHAPE) analysis (Busan and Weeks, 2013). In addition to the observation made by de Mezer et al. (2011), additional base pairing between a few CAG repeats from the 5′ side with nucleotides in the 5′UTR sequence was observed. Moreover, it was shown that when a normal number of CAG repeats was studied, the CAG hairpin was either absent (CAG₁₇) or short (CAG₂₃; Busan and Weeks, 2013).

In the case of AR transcript, which is involved in spinal and bulbar muscular atrophy (SBMA), not one but two different types of neighboring, triplet repeats, together with specific flanking sequences, contribute to the formation of a composite CAG repeat hairpin structure (de Mezer et al., 2011). As shown in Figure 2C, the AR transcript contains three monomorphic CUG repeats [(CUG)₃] directly upstream of a polymorphic, expansion-prone CAG repeat tract [(CAG)_n]. Moreover, six monomorphic CAG repeats [(CAG)₆] are located downstream of the CAG repeat tract and are separated by an 18-nucleotide specific sequence. As presented in Figure 2C, the monomorphic (CUG)₃ repeats fully base pair with the last three CAG repeats of the (CAG)_n tract to form a strong 12-bp stabilizing clamp, which is responsible for the presence of both normal and mutant AR transcripts in only the “frozen” type of hairpin form with a single RNA alignment. In addition, further stabilization of the repeat hairpin is conferred by the base-pairing system of the nearest specific CAG flanking sequences. Moreover, the monomorphic (CAG)₆ repeat tract, which is located 18 nt downstream from the polymorphic, expansion-prone CAG repeat tract [(CAG)_n], is not involved in the formation of the long repeat hairpin structure, but rather, forms an autonomous short hairpin. The only difference between the normal and mutant AR transcript is the length of the CAG repeat hairpin, which is formed solely by polymorphic expansion-prone CAG repeats and which contains either a 4- or 7-nt terminal loop, depending on the CAG repeat number according to the rule observed for isolated, clamped, CAG repeat structures (Sobczak et al., 2003).

Structural studies of the CAG repeat regions from the ATXN1 (Sobczak and Krzyzosiak, 2004a) and the calcium voltage-gated channel subunit alpha1 A (CACNA1A) transcripts (Michlewski and Krzyzosiak, 2004), which are implicated in SCA1 and SCA6, respectively, have also revealed the importance of the specific flanking sequences in stabilizing CAG repeat hairpin structures (Figures 2D,F). In both transcripts, there is a strong, naturally occurring clamp that is formed by the specific flanking sequences at the base of the repeat hairpin. For ATXN1, the clamp consists of a perfectly matching 6-bp long fragment. These clamps cause the CAG repeats to “freeze” in a single alignment and to form stable hairpins with terminal loops of different sizes depending on the repeat number (4-nt or 7-nt loops). The stem structure and loop size follow the pattern observed for model of isolated CAG repeats containing a GC clamp (Sobczak et al., 2003).

Using similar structure probing approaches, several “slipped” hairpin variants have been observed for CAG repeats of the atrophin (ATN1) (Michlewski and Krzyzosiak, 2004), ATXN2 (Sobczak and Krzyzosiak, 2005), and ATXN3 (Michlewski and Krzyzosiak, 2004) transcripts, which are implicated in dentatorubral-pallidoluysian atrophy (DRPLA), SCA2 and SCA3, respectively. In ATN1 and ATXN2 transcripts (Figures 2E,G), their specific repeat flanking sequences do not stabilize the CAG repeat structure, as no clamping of the hairpin by flanking sequences is observed. Furthermore, the “slipped” ATN1 hairpins have the repeats moved towards their 3′ end and contain a 4-nt terminal loop, as observed for unclamped CAG model transcripts (Figure 2E; Sobczak et al., 2003). In contrast, in the ATXN2 transcript, the specific flanking sequences located at the 3′ side of the CAG repeat tract interact with 3′ terminal CAG repeats, which results in a reduction of the CAG repeat stem length. This interaction, however, does not force single alignment of the uninterrupted CAG repeats in the ATXN2 mutant transcript because several “slippery” hairpin variants are observed (Figure 2G). For the ATXN3 transcript, the CAG repeat region has a particular architecture that is greatly influenced by both 3′- and 5′-specific flanking sequences (Figure 2H). First, several 3′ terminal CAG repeats are involved in a quasi-stable interaction with the 18-nt pseudo-repeat sequence that flanks the CAG repeats on its 5′ side. Moreover, the (CAG)C/G single-nucleotide polymorphism (SNP) that is located between CAG repeat tract and its specific 3′ flanking sequence does influence the structure of both of these sequences. Depending on which SNP variant is present, the size of the CAG hairpin terminal loop and, to a minor extent, the structure formed by the 3′ flanking sequence (M2 module) change. Furthermore, the CAG repeat hairpin forms several alternatively aligned variants that are “slipped” towards the 5′ end, in contrast to the unclamped CAG repeats in model transcripts due to the presence of the 18-nt pseudo-repeat sequence.

By using comprehensive population genotyping surveys, specific interruptions CAT and CAA were found within CAG repeat tracts of ATXN1 and ATXN2 alleles, respectively (Sobczak and Krzyzosiak, 2004b; Rozanska et al., 2007). Interestingly, these types of repeat interruptions (usually 1–3) were observed in normal, but not in the expanded, mutant CAG repeat tracts and therefore are believed to function on the genomic level as protective elements, preventing further repeat expansion during maternal transmission and development (Pearson et al., 1998). Additionally, the effects of CAU and CAA interruptions on the CAG hairpin structures in ATXN1 (Sobczak and Krzyzosiak, 2004a) and ATXN2 (Sobczak and Krzyzosiak, 2005) transcripts have been shown. As presented in Figure 3A, both the number and localization of the interruptions determine the structure of this region in both the transcripts. Most commonly, in the ATXN1 and ATXN2 transcripts, one or two CAU triplets (approximately 90% in populations) and two CAA triplets (78%–95% in populations), respectively, were shown to destabilize the stem of the single, long CAG repeat hairpin. If two interruptions break the regularity of the CAG repeat tracts, they are always separated by one CAG repeat in the ATXN1 transcripts and by four CAG repeats in the ATXN2 transcripts. Depending on how the CAU and CAA interruptions are organized in transcripts, they can widen the existing terminal loop, nucleate out additional loops, split the sequence into two separate hairpins (mainly in ATXN1 transcripts), or form specific branched structures with the interruptions localized in terminal loop/s (mainly in ATXN1 transcripts; Figure 3A). As these effects ultimately lead to the shortening of the single long CAG hairpin structure, it is hypothesized (Sobczak and Krzyzosiak, 2004a, 2005) that both CAU and CAA triplets can delay disease onset or severity (Matsuyama et al., 1999; Tian et al., 2000; Peel et al., 2001; Hussey et al., 2002). These observations might be the consequence of the reduced, diseased-causing sequestration of RNA-binding proteins by interrupted CAG repeat tracts.

Similar to the other CNG repeats, the higher-order RNA structures that were formed in solution by isolated CGG repeat motifs (CGG₁₇) were found to form hairpins with the stem composed of periodically occurring C-G and G-C base pairs that are separated by a single non-canonical G-G pair (Sobczak et al., 2003). In contrast to the other CNG repeats, the structure of the CGG hairpin is more rigid and does not form “in register” conformations, i.e., “slippery hairpins” (Sobczak et al., 2003). When the CGG repeat hairpin was GC-clamped and had an even number of repeats (CGG₁₆), a 4-nt apical loop was formed. In the case of odd-numbered CGG repeat hairpin (CGG₁₇) a tighter 3-nt terminal loop was observed (Sobczak et al., 2003). In a more recent, comprehensive, RNA structure probing study, it was shown that isolated CGG motifs (CGG_{17 or 20}), together with other three CNG, CAA and CGU motifs, form fairly stable hairpins (3rd structural class; Sobczak et al., 2010). Moreover, calorimetric studies revealed that among all CNG repeats, CGG repeat motif is the most thermodynamically stable in Na⁺ ions conditions (second to CAG motif in K⁺ ions conditions; Broda et al., 2005; Sobczak et al., 2010). Other biophysical (NMR and UV spectroscopy—Zumwalt et al., 2007) and biochemical (gel mobility analysis—Khateb et al., 2004; Ofer et al., 2009) studies on CGG repeat motifs have also been performed.

In recent years, the X-ray crystal structures of CGG-containing oligoribonucleotides have been published (Kiliszek et al., 2011; Kumar et al., 2011). The analyzed CGG oligomers [G(CGG)₂C] formed intermolecular duplexes representing the stem portions of the CGG repeat hairpin (Figure 1C). Furthermore, the identified helices retained an A form, where the non-canonical G-G pairs were flanked by canonical stabilizing C-G and G-C pairs. The steric hindrance caused by two bulky guanine residues opposite each other within the helical structure is resolved by having one guanosine in the syn conformation and the other in the anti conformation (Figure 1C). This syn-anti arrangement causes the local, moderate, unwinding of the helix structure and widening of the major groove. In all G-G pairs, two direct hydrogen bonds are formed between the O6, N1 atoms from the G(syn) and N7, N2H atoms from the G(anti), respectively (O6⋅⋅⋅N1H and N7⋅⋅⋅N2H) (Kiliszek et al., 2011). This kind of interaction between a wobble G-G pair has been commonly observed in other NMR and crystallographic RNA structures. The observed strong hydrogen bonding system could explain why the model CGG repeat transcripts do not show the slippage effect, as opposed to the less stable CAG and CUG repeats (Sobczak et al., 2003).

The exact structural properties of transcripts containing CGG repeats have been a matter of debate. While strong biophysical and biochemical evidence for CGG duplex and hairpin structures has been presented (Sobczak et al., 2003, 2010; Napierala et al., 2005; Zumwalt et al., 2007; Kiliszek et al., 2011; Kumar et al., 2011), other higher-order structures, i.e., quadruplexes, have also been observed (Handa et al., 2003; Khateb et al., 2004, 2007; Ofer et al., 2009; Malgowska et al., 2014; Gudanis et al., 2016). Recently, by using a battery of biophysical methods, such as UV-visible, CD and NMR spectroscopies, electrospray ionization mass spectrometry (ESI-MS), small-angle X-ray scattering (SAXS) and native gel electrophoresis analysis, two different quadruplex structures that short CGG-containing oligoribonucleotides formed in solution were identified (Malgowska et al., 2014; Gudanis et al., 2016). Figure 4A shows the specific G-quadruplex structure that is formed by short CGG-containing oligoribonucleotides [G(CGG)₂C] in the presence of K⁺ ions (Malgowska et al., 2014). As demonstrated, this symmetrical, tetramolecular G-quadruplex structure is composed of two mixed, antiparallel G:C:G:C tetrads divided by parallel G:G:G:G tetrad (G-tetrad) planes (Figures 4B,C). However, the structures formed by both G(CGG)₂C and G(CGG)₄C oligomers in solution are highly polymorphic, and other type of structures can be adopted depending on the presence of different cations (K⁺, Na⁺ or NH₄⁺), the length of the repeats and the concentration of RNA used. For example, when the G(CGG)₂C oligomer is studied in Na⁺ solutions, it maintains a balance between G-quadruplex/duplex structures, but when it is studied in HeLa cellular extracts where K⁺ ions are prevalent, the G(CGG)₂C oligomer almost exclusively forms duplexes. Recently, a novel quadruplex architecture formed by the 8-bromoguanosine-modified molecule GC^BrGGCGGC was characterized (Gudanis et al., 2016). This unnatural modification locks the ^BrG:G pairs exclusively in favorable ^BrG(syn)-G(anti) conformations, thus increasing the thermodynamic stability and the homogeneity of the RNA structures that formed. As the folding rules of both the quadruplexes formed by CGG repeats are yet unclear and may depend on other factors, it is of great interest to study these structures further.

Thus far, two studies have been carried out to provide structural insights into the CGG repeat region from the 5′UTR of the FMR1 transcript, which is associated with a range of different clinical phenotypes (e.g., FXTAS and premature ovarian failure (POF)) of increased severity depending on the extent of CGG tract expansion (Handa et al., 2003; Napierala et al., 2005). As seen in Figure 2I, uninterrupted CGG repeats form a hairpin that is composed of non-canonical G-G pairs flanked by stabilizing C-G and G-C pairs (with a 3- or 6-nt terminal loop) and that is further stabilized by base-paired 3′- and 5′-specific flanking sequences. Moreover, the 3′ end of the CGG repeats was shown to be engaged in base pairing with a section of the 3′ flanking sequence (Napierala et al., 2005) The majority of normal and premutation size FMR1 alleles contain specific AGG interruptions (1–4, usually 2) in the CGG repeat tract. These AGG triplets were shown to destabilize the single long CGG hairpin structure in different ways (Figure 3B). Depending on their localization within the repeats, they can enlarge the existing terminal loop, nucleate out additional autonomous hairpin loops or form a Y-shaped structure. Regarding the latter, the polymorphic A lies in one of the double stranded arm and base pairs with the U residue located in the 3′ flanking sequence. Moreover, a pH-dependent mechanism that causes the Y-shaped structure to adopt different alternative conformers (S and F conformers) was identified (Figure 3B; Napierala et al., 2005). These AGG interruptions have been shown to function as protective elements that prevent CGG repeat expansion at the DNA level during maternal transmission (Eichler et al., 1994; Pearson et al., 1998; Dombrowski et al., 2002). Additionally, depending on the number and localization of the interruptions, it is hypothesized that they can protect some premutation carriers from FXTAS and POF by shortening the pathogenic length of hairpins composed of pure CGG repeats.

To date, higher-order RNA structures formed by pentanucleotide AUUCU repeats, which are present in intron 9 of the ATXN10 transcript implicated in SCA10, have been a subject of two structural studies (Handa et al., 2005; Park et al., 2015). By using a combination of enzymatic (S1 nuclease, RNase V1) RNA structure probing and biophysical (CD and NMR) approaches, it was demonstrated that as few as nine AUUCU repeats [(AUUCU)_{9/11/14 or 17}] form an unusual RNA hairpin structure (Handa et al., 2005). The stem of this hairpin has characteristics of A-form geometry and, as revealed by NMR analysis, contains a mixture of A-U and U-U base pairing in a 1:1 ratio. This ratio suggests that the stem of the AUUCU repeat hairpin harbors ⁵′UCU³′/³’UCU⁵′ internal loops with two noncanonical U-U pairs and one noncanonical C-C pair, which are closed by two canonical A-U pairs (⁵′AU³′/³′UA⁵′ loop closing pairs) (Figure 1D; Handa et al., 2005). Recently, the structural characteristics of AUUCU repeat-containing RNAs have been confirmed, and further insights were gained using X-ray crystallography followed by a robust computational analysis of the structure via MD simulations (Park et al., 2015). In that study, a crystallization-promoting tetraloop/tetraloop receptor motif was utilized to aid crystallization of a model RNA containing two copies of ⁵′AUUCU³′/³′UCUUA⁵′ motifs (refined to 2.8 Å; Figure 1D). This analysis showed standard Watson-Crick base pairing in the ⁵′AU³′/³′UA⁵′ loop closing pairs, thereby stabilizing the AUUCU repeat duplex/hairpin structure. For the hydrogen bonding status of the internal ⁵′UCU³′/³’UCU⁵′ loops, a more complex and dynamic transitions were suggested. According to the X-ray data, the opposite uracil residues constituting both noncanonical U-U pairs are in the anti conformation and possess two hydrogen bond geometries, where the carbonyl O4 atom and the N3 amino group of the first U form two hydrogen bonds with the N3 amino group and carbonyl O2 atom of the second U, respectively (O4⋅⋅⋅N3H; N3H⋅⋅⋅O2) (Figure 1D). In the case of the central noncanonical C-C base pair, two different conformations were suggested: (1) a frequent, one hydrogen bond geometry, where both cytosine residues are in the anti conformation and form one very weak hydrogen bond between the exo-amino group of one C residue and the N3 atom of the other C residue (N4H⋅⋅⋅N3); and (2) a stable zero hydrogen bond conformation, which is stabilized by a specific hydration pattern, as suggested by in silico computational analysis. Moreover, it is hypothesized that these extremely weak and dynamic C-C non-canonical pairing interactions as well as the disrupted base stacking of the C-C pair with the neighboring U-U pairs cause the internal ⁵′UCU³′/³’UCU⁵′ loops to be the least thermodynamically stable elements of the AUUCU repeat hairpin structure. Furthermore, as proposed by the MD simulation analysis, this dynamic feature of the internal loops affects the overall AUUCU A-form helix stability, causing its rearrangement into single-stranded conformations (quasistable hairpin structures). This ⁵′UCU³′/³’UCU⁵′ loop is thought to be a site where RNA unwinding starts.

In recent years, numerous structural and functional studies have been carried out to understand RNA toxicity in newly described hexanucleotide repeat expansion which is implicated in C9ALS/FTD. In these diseases, the expanded repeats are transcribed bidirectionally, generating noncoding sense (GGGGCC)_n and antisense (CCCCGG)_n transcripts that both form different higher-order RNA structures.

Initially, Fratta et al. (2012) used biophysical approaches (NMR and CD spectroscopies) to confirm the results of an in silico computational analysis performed with a G-quadruplex prediction tool (QGRS mapper) and showed that an isolated (GGGGCC)₃GGGGC RNA oligomer (C9Gru) is minimally required for the formation of a specific G-quadruplex structure (Fratta et al., 2012). As shown in Figure 4D, this highly stable G-quadruplex structure adopts a parallel topology (typical for most other RNA G-quadruplexes) and consists of four stacks of parallel G:G:G:G tetrad (G-tetrad) planes. In each G-tetrad plane, guanine bases are arranged in a square cyclic pattern connected by eight Hoogsteen hydrogen bonds and are arranged in a planar configuration around a central monovalent metal cation (interacting with guanine O6 atoms) that significantly affects G-quadruplex stability and topology (Figure 4B). The four stacked G-tetrads, which are bound with metal ions positioned centrally and phosphate backbones positioned laterally, are connected through a propeller loop arrangement composed of two cytosines, which ensures parallel topology.

Some deviations from the above G-quadruplex structure have also been observed. As suggested by Reddy et al. (2013) and Haeusler et al. (2014), the G-quadruplex structure consists of three, not four, G-tetrad planes (Reddy et al., 2013; Haeusler et al., 2014). By using an RNase protection assay (RNase T1, which cleaves ssRNA at the 3′ end of guanine residues), it was shown that a model (GGGGCC)₄ RNA in 100 mM KCl formed a symmetrical, parallel G-quadruplex structure almost exclusively, and the digestion pattern revealed a three-stacked G-tetrad plane topology, with guanine and two cytosines in the single-stranded loop region (not involved in the formation of the G-quadruplex structure) that connects laterally G-quadruplex phosphate backbones (Figure 4D).

Whether the formation of G-quadruplexes results from the association of GGGGCC repeats of the same C9orf72 transcript (intramolecular G-quadruplexes) or from the interaction of GGGGCC repeats from different C9orf72 molecules (intermolecular G-quadruplexes) is still a matter of a debate (Fratta et al., 2012; Reddy et al., 2013). It was suggested by CD spectroscopy that an isolated, (very short) model (GGGGCC)₃GGGGC transcript (minimal C9Gru) can form stable G-quadruplex structures only through an intramolecular association (Fratta et al., 2012). It was suggested by CD spectroscopy and native PAGE analysis that an isolated, model (GGGGCC)₄ transcript can form an extremely stable intramolecular G-quadruplex structure (Reddy et al., 2013) Additionally, by native PAGE, it was shown that the heterogeneity of the formed G-quadruplexes increases as the repeat length or RNA concentration increases. In contrast to (GGGGCC)₄ RNA, additional slower migrating species were observed for (GGGGCC)_{6 or 8}, which is consistent with the formation of additional intermolecular G-quadruplexes (multimeric). Moreover, such an increase in heterogeneity was observed, as both additional intramolecular and intermolecular G-quadruplexes were formed when a native 5′ flanking sequence (15-nt long) from C9orf72 RNA was present upstream of the GGGGCC repeat tract. Hairpin formation may compete with or contribute to the formation of G-quadruplexes or other structures (Reddy et al., 2013). By using a battery of biochemical approaches (enzymatic structure probing (RNase A, T1, A/T1), UV crosslinking, and native and denaturing PAGE), it was demonstrated that an isolated, model (GGGGCC)₁₀ RNA (oligomer) forms stable multimeric G-quadruplex structures in U87 nuclear extracts (in vitro, Conlon et al., 2016). These analyses showed two major, distinct G-quadruplex states resulting from 10 GGGGCC repeat transcripts that were differentially folded: one in which four GGGGCC repeats form one G-quadruplex and the other major conformation in which eight consecutive GGGGCC repeats form two G-quadruplex structures.

By using CD spectroscopy (Fratta et al., 2012; Reddy et al., 2013; Haeusler et al., 2014) and native PAGE (Reddy et al., 2013; Haeusler et al., 2014) analysis, it was shown that similar to other RNA G-quadruplexes, the folding of isolated model (GGGGCC)_n repeat RNAs (n = ~4) into a stable G-quadruplex structure is strongly influenced by the presence of monovalent cations (K⁺ promotes stable folding over Na⁺ and Li⁺ ions). Furthermore, at a physiological K⁺ ion concentration (pH), these structures were shown to be extremely thermodynamically stable (up to 95°C; Fratta et al., 2012; Reddy et al., 2013). Additionally, by using CD spectroscopy and RNase protection assay (RNase T1) analyses, it was demonstrated that an isolated model (GGGGCC)₄ RNA oligomer can adopt either a G-quadruplex structure (almost exclusively) or a structure consistent with single-stranded bulges and hairpin conformations, depending on the presence or absence of K⁺ ions, respectively (Haeusler et al., 2014).

The G-quadruplex structures formed by GGGGCC repeats have only recently been observed in cells (Conlon et al., 2016). By using immunofluorescence (IF) microscopy and a G-quadruplex recognizing antibody (BG-4), it was shown that G-quadruplexes formed by the expanded GGGGCC repeats in mutant C9orf72 RNAs exist and are components of pathogenic RNA foci observed in fibroblasts and astrocytes derived from C9ALS/FTD patients; they are also major components of insoluble protein/RNA aggregates that have been isolated from disease-relevant regions of post-mortem c9ALS brains. However, the nature of these G-quadruplexes could not be established, as BG4 does not differentiate between any particular G-quadruplex conformation (parallel, anti-parallel, mixed parallel/anti-parallel, intermolecular or intramolecular; Conlon et al., 2016). Recently, the formation of G-quadruplex structures in mammalian transcriptome was analyzed with a modified, high-throughput RNA chemical probing techniques—DMS-seq and SHAPE-seq (Guo and Bartel, 2016). By applying these methods to human transcriptome the authors were able to identify thousands (>10,000) of novel G-quadruplex structures in vitro. However, when studied in cells (in vivo), these G-quadruplex forming regions were shown to be globally unfolded, presumably by a robust and effective machinery consisting of unknown RNA helicases and other ssRNA-binding proteins, that await to be fully characterized. Therefore it is predicted that in C9ALS/FTD patients this machinery could be either “switched off” or inefficient in the certain cell types, states or subcellular compartments thereby allowing G-quadruplex structures to form over time and exert significant pathogenic effects.

In contrast to the sense GGGGCC repeats, the antisense (CCCCGG)_n RNAs do not seem to form stable G-quadruplex structures; instead, hairpin structures are formed, regardless of the presence of K⁺ ions, as assessed by CD spectroscopy and native PAGE analysis (Reddy et al., 2013; Haeusler et al., 2014). Recently, more detailed solution studies of the formation of A-form-like helical conformations were conducted by using CD spectroscopy and differential scanning calorimetry (DSC; Dodd et al., 2016). The authors showed the dependence of the CCCCGG repeat number on the formation of stable helical structures: (CCCCGG)₂, slipped intermolecular duplexes; (CCCCGG)₄, intermolecular multimers; (CCCCGG)_10, intramolecular A-form like helixes. In the same study, the X-ray crystal structure of a model (CCCCGG)₃(CCCC) RNA refined to 1.47 Å resolution was determined (Dodd et al., 2016). The identified intermolecular helix structure, likely representing the stem of the CCCCGG hairpin, had an A-form-like geometry and consisted of repeating units of four canonical Watson-Crick G-C/C-G base pairs separated by two regularly spaced tandem noncanonical C-C pairs (Figure 1E). The cytosine bases of each C-C pair were in the anti conformation and created one of two hydrogen bond interactions that presumably interconvert in solution: the N4 exo-amino group of one cytosine residue (H-bond donor) interacted with either the carbonyl O2 atom (N4H⋅⋅⋅O2 hydrogen bond) or with the N3 atom of the other cytosine residue (N4H⋅⋅⋅N3 hydrogen bond) (Figure 1E; Dodd et al., 2016). Moreover, perturbed electrostatic surface potential and the smaller width of the minor groove at the tandem C-C pairs (helix contraction) were observed compared to the typical A-form helix parameters. These changes are thought to be a consequence of the accommodation of the hydrogen bonding distances between the smaller size pyrimidine residues (between tandem C-C pairs).

The most extensively investigated simple RNA repeat-protein interaction is the one between Muscleblind-like 1 (MBNL1) protein and CUG repeat-containing RNA. It is commonly known that abnormal MBNL1 sequestration by RNA foci formed by a mutant DMPK transcript interferes with the proper functioning of MBNL1 in the cell, which is mainly alternative splicing regulation.

The human homolog of Drosophila MBNL1 was first identified as a protein that binds the CUG repeat in a repeat length-dependent manner (Miller et al., 2000). Experiments based on incubation of the in vitro-transcribed phospho-labeled DMPK RNAs with cellular nuclear extracts followed by UV cross-linking to capture proteins associated with these transcripts and by MS, resulted in the identification of MBNL1 as the predominantly bound protein. To verify whether the CUG repeats represent the sequence recognized by MBNL1, the investigated transcripts were limited to only CUG tracts of different lengths. It appeared that 20 and more CUG repeats cross-linked to MBNL1 with efficiency proportional to the CUG repeat length (Miller et al., 2000). Later, chemical and enzymatic structure probing of labeled RNA was employed to demonstrate that MBNL1 binds to stem portion in hairpins formed by CUG repeat tracts (Figure 5A; Yuan et al., 2007).

To determine whether MBNL1 recognizes CUG tracts in vivo, an yeast three-hybrid (Y3H) assay was used, which confirmed the interaction between protein and RNA through detection of the reporter gene activities (Kino et al., 2004). To confirm the three-hybrid system results, the authors also performed gel retardation assays and concluded that MBNL1 is a repeat-binding protein with a strong preference for long CUG repeat tracts. Fluorescence in situ hybridization (FISH) alone and in combination with IF (FISH/IF) techniques were also employed to investigate CUG repeats targeting by MBNL, which demonstrated the accumulation of long CUG tracts in nuclear foci where MBNL1 co-localized (Mankodi et al., 2001; Sznajder et al., 2016).

MBNL1 is composed of four conserved CCCH zinc-finger domains (ZnF1-4), wherein ZnF1 with ZnF2 and ZnF3 with ZnF4 are paired to form two tandems connected by a linker region (Figure 5B; Teplova and Patel, 2008; Grammatikakis et al., 2011; Konieczny et al., 2014; Sznajder et al., 2016). There are multiple MBNL1 splicing variants present in the cell, and not every isoform contains all four zinc-finger motifs. Y3H assay used for MBNL1 truncated versions showed that their ability to bind CUG triplet repeat RNA varies (Kino et al., 2004). It was postulated that only two ZnF domains and a GC dinucleotide interaction are sufficient for high-affinity MBNL1 binding to RNA (Cass et al., 2011; Purcell et al., 2012).

Electron microscopy demonstrated that MBNL1 binds to long CUG tracts as a ring-shaped structure (Yuan et al., 2007). It was postulated that MBNL1 recognizes GC-rich hairpins with pyrimidine mismatches in both physiologic and pathogenic transcripts, as previously demonstrated by others (Warf and Berglund, 2007). This finding is in agreement with previous results from Ishiura’s group, showing that MBNL1 target repeat sequences can be summarized as CHG and CHHG, where H stands for A, U or C (Kino et al., 2004). The same authors also suggested that MBNL1 prefers double-stranded RNA-containing bulges (Kino et al., 2004). Berglund’s group provided evidence that MBNL1 recognizes GC motifs embedded in pyrimidines that are present in both CUG long tracts and the natural RNA targets of the protein (Goers et al., 2010). Physiological MBNL1 targets were investigated transcriptome-wide using cross-linking and immunoprecipitation with a sequencing approach (CLIP-Seq), and as predicted, UGC- and GCU-containing four-mers were defined as MBNL1-binding sites (Wang et al., 2012).

Crystallography study revealed that MBNL1 ZnF3/4 binds to a CGCUGU single-stranded ribohexamer; thus, MBNL1 prefers single-stranded target RNAs (Teplova and Patel, 2008). To assess the structural changes in the target RNA upon MBNL1 binding, CD spectroscopy was utilized, which showed that the stability of the CUG repeat tract structure changed towards single-stranded when MBNL1 bound to the RNA (Fu et al., 2012). The authors concluded that less stable RNA structures can form more stable complexes with MBNL1. In agreement with this conclusion are results showing that MBNL1 binding is blocked when CUG repeats structure undergoes stabilization in double-stranded helical conformation by introducing pseudouridine or 2′-O-methyl modifications. Such structure stabilization inhibits MBNL1 sequestration, decreases foci size and in consequence reduces toxicity of CUG repeat containing RNA in human cells and in zebrafish (deLorimier et al., 2014). This may suggest that in contrast to in vitro system, RNA CUG repeats do not form double-stranded A-form structure in vivo.

Both the sequence and structure of the long CUG repeat RNA seem to be important for interactions with MBNL1. Although the studies mentioned above provided some insight into this interaction, the structure formed by expanded CUG repeats in RNA in vivo remains elusive, and many putative protein structure modulators present in the cell may play crucial roles in pathogenic transcript structure formation and in providing access to MBNL1.

In addition to the MBNL1 sequestration by long CUG repeats, many other proteins that interact with neurological disease-relevant simple repeats in RNA have been identified. These proteins were mainly searched for using RNA-pulldown combined with MS or biochemical assays as reviewed in Jazurek et al. (2016). Many unique proteins belonging to various protein families and containing diverse binding domains were identified (Jazurek et al., 2016). However, research on the nature of their interaction with mutant RNAs, particularly their structural requirements, is much less advanced. In many cases, the biological consequences of such abnormal RNA-protein interactions have not been validated. Thus, there is still a need to precisely explore the mutant RNA-binding properties of these proteins. Below, we present selected examples of the most studied interactions between mutant RNA and proteins other than MBNL1.