Action and function of helicases on RNA G-quadruplexes.

The nucleic acid structure called G-quadruplex (G4) is currently discussed to function in nucleic acid-based mechanisms that influence several cellular processes. They can modulate the cellular machinery either positively or negatively, both at the DNA and RNA level. The majority of what we know about G4 biology comes from DNA G4 (dG4) research. RNA G4s (rG4), on the other hand, are gaining interest as researchers become more aware of their role in several aspects of cellular homeostasis. In either case, the correct regulation of G4 structures within cells is essential and demands specialized proteins able to resolve them. Small changes in the formation and unfolding of G4 structures can have severe consequences for the cells that could even stimulate genome instability, apoptosis or proliferation. Helicases are the most relevant negative G4 regulators, which prevent and unfold G4 formation within cells during different pathways. Yet, and despite their importance only a handful of rG4 unwinding helicases have been identified and characterized thus far. This review addresses the current knowledge on rG4s-processing helicases with a focus on methodological approaches. An example of a non-helicase rG4s-unwinding protein is also briefly described.

Introduction

Nucleic acid secondary structures provide additional biological information beyond the primary sequence. The formation of nucleic acid structures is widely recognized for their importance during several cellular processes adding up layers of complexity to the web of cellular regulatory mechanisms. Among these are G-quadruplex structures (G4s), stable DNA or RNA structures formed within guanine-rich sequences, which can self-assemble via Hoogsteen hydrogen bonding into (at least two, most frequently three) stacking guanine tetrads stabilized by monovalent cations [1], [2], [3]. The relative orientation of each guanine in the tetrads defines the overall quadruplex topology, which can be summarized in the conventional parallel, antiparallel and hybrid scheme [4]. The inherent biophysical differences between DNA and RNA reflect onto the respective G4 structures [5], [6], because the extra 2′-OH moiety in RNA enables additional hydrogen bonding and affects molecular hydration thus making RNA G4s (rG4s) more stable than the DNA counterpart [1], [7], [8]. In contrast to DNA G4s (dG4s) that can explore a broader conformational space, the 2′-OH moiety in rG4s also introduces steric constriction that forces rG4s into the parallel topology [9], [10] with some remarkable exceptions [11], [12], [13]. In addition to the biophysical aspects the biology of the two classes of G4 diverges because of the diverse processes they operate in although through similar underlying principles.

Roughly 700.000 putative G4-forming sequences are spread throughout the human genome. They are enriched at functional loci, hinting at the role that G4s have in a vast network of cellular processes [14], [15], [16], [17]. Notwithstanding early skepticism there is cogent evidence for the in vivo formation and involvement of dG4s in transcription regulation [18], [19], telomeres stability [20], [21], epigenetics [22], [23], [24], DNA repair mechanisms [25], [26], replication [27], [28], [29], [30], genome stability [30], [31], [32] and recombination [33], [34], [35], which are all reviewed in detail elsewhere [36], [37], [38]. In particular the finding of dG4 formation within promoters of prominent oncogene e.g. KRAS [39], [40], [41], [42], VEGF [43], KIT [44], [45], BCL2 [46] and MYC [47], [48], [49], [50] and the potential to use these structure to control the gene expression, telomerase action and genome stability opened the possibility to use G4s as potential druggable anticancer targets [37], [51], [52], [53], [54]. To improve current therapeutic strategies, it is essential to understand the relevance of DNA and RNA G4s within cells and how they are regulated within cells by helicases.

Likewise dG4s, the actual folding of rG4s in living cells has been long debated given the emergence of conflicting findings [55]. Most rG4s-forming sequences are enriched at ncRNAs and 5′- and 3′-UTR non-coding regions [14] and were first visualized in human cells by ligand-based rG4-selective probing [56], [57] with roughly 13.000 putative rG4s mapped to the transcriptome by reverse transcriptase stop assay combined with next-generation sequencing [58]. Controversy arose upon the report of the coupled DMS/RT-stop analysis from Guo and Bartel [59] arguing for mostly unfolded rG4s in eukaryotes, at equilibrium. This finding is in contrast to other publications showing that although DMS can be used to measure G4-folding kinetics in vitro, DMS-seq can not be used to map G4 formation in living cells. Using a single molecular live cell imaging approach, a recent publication showed that G4 structure formation dynamically changes in the cell, which makes DMS ineffective, because only the unfolded state will be trapped by this method [60]. Furthermore, different methods in different model systems and cell types support the existence of rG4s. For example, with the fluorescent probe QUMA-1 [61] as well as with the biotinylated template-assembled synthetic G-quartets (Bio-TASQ) [62], [63] rG4 visualization, pull-down analysis and global mappings are possible. These and other experiments demonstrate the relevance and function of rG4s in a dense grid of cellular processes [55], [64], [65]. In detail, rG4 have been shown to modulate post-transcriptional events, particularly translation [66], [67], [68], [69], [70], [71], [72], localization [73], [74] and splicing [75], [76], [77], [78], [79], [80], [81], [82], [83], [84]. These findings advocate for a dynamic protein-regulated rG4 scenario with profound implications in living cells. The apparent incoherence between the work of Guo and Bartel with other experimental findings about rG4 is summarized and discussed in recent reviews [64], [85].

rG4s are now accepted to form dynamically in vivo as transient events in an ever-changing rG4 landscape influenced by exogenous and endogenous stimuli and under control of the cellular regulatory machinery. This model can consistently explain the seemingly contradictory findings from the aforementioned experimental approaches. Accordingly, only specific rG4 subsets exist at any given time as a function of cell type, cell-cycle progression [86], environmental stress [87], etc. The higher turnover of the rG4 pool compared to dG4 is also conceivable considering the inherent nature of RNA and its very dynamic processes which, unlike given for DNA, are intrinsically transient and relatively short-lived to ensure readily responsive regulation of cellular functions [88].

Regardless of the polymer nature highly stable G4 structures can act as kinetic traps and must ultimately be unfolded to ensure plasticity to the regulatory system and prevent detrimental effects on downstream processes. Several works probing the misregulation of G4s using stabilizing ligands [22], [89], [90] or targeted knockouts (KO) [91] have been conducted. Whilst the folding process is thermodynamically driven and can occur regardless of supporting proteins [26], [92], [93], [94], G4 unfolding requires proteins that either compete with the folding of the quadruplex (i.e. binding and ‘sequestering’ the unfolded or alternative species) [68], [95] or specialized helicases actively unfolding the G4 [19], [33], [96], [97], [98], [99], [100]. Although the information gathered from dG4s largely dominates the field the pervasive impact on the cell homeostasis of rG4-mediated processes and the identification of rG4-unwinding helicases and proteins increasingly shifted the interest to also address how rG4s interface with the cellular regulatory machinery [55], [64], [96], [62], [101], [102], [103], [104], [105], [106], [107], [108], [109], [110], [111], [112], [113].

Several recent reviews exhaustively document the generalities the biology of processes mediated by dG4-unwinding helicases [55], [64], [114], [115]. Here, we focus on rG4-unwinding helicases and the biological processes they serve (Fig. 1) with an emphasis on experimental approaches that were performed to investigate rG4-helicase action.

Fig. 1

Helicases intervene in a multitude of RNA processes through the unwinding of rG4s. Genome homeostasis is kept through the processing of telomeric rG4. The resolution of rG4 in 5′-UTRs of mRNAs is a regulatory layer in translation, whereas in 3′-UTRs rG4 processing is linked to stress response and transcripts stability. Helicases modulate the splicing pattern of transcripts with rG4s within CDSs. Also, the miRNA metabolism is subject to helicase modulation of rG4 that would otherwise outcompete the cleavage targets of DICER. At last, rG4-processing by helicase impacts the IgH class recombination.

Excurse: Methods to demonstrate G4 formation in vitro and in vivo

As described above G4 DNA and RNA structures can form within specific guanine-rich regions harboring a G4 motif. Originally this motif harbored four G-tracts with at least three guanines next to each other that were separated by variable regions (loop regions) with a length of 1–7 nucleotides (nt) (GGGN1-7GGGN1-7GGGN1-7GGG) [116], [117]. Recent studies demonstrated that dG4 as well as rG4 can also form with two guanines per G-tract as well as with loops that are longer [118], [119]. Note, the stability of the G4 structures positively correlates with the amounts of guanines per G-tract and decreases with increasing loop length [120]. Further, G4s are stabilized by cations that are centrally coordinated to the O6 of the guanines [117]. The dependency of G4 stability on salt might be of interest for studies using helicases, because also helicase action is salt dependent.

Studying G4 formation in vitro and in vivo is challenging due to different topologies and stabilities of the G4 (reviewed in [121]). G4 formation in vitro can be observed using X-ray crystallography, NMR spectroscopy, circular dichroism, ultraviolet melting and differences in electromobility behavior [122], [123]. Also different methods have been developed to map G4 in cells in vivo [124]. For example, G4 visualization using immunofluorescence can be achieved using either G4-specific antibodies (e.g. Sty49, 1H6, BG4) [125], [126], [127], [86], [128] or synthetic small molecules that recognize and stabilize G4s and are coupled to a fluorescence probe (e.g. PDS, PhenDC3, DAOTA_M2, QUMA) [129], [130], [131], [61]. Recently, also G4 quantification by flow cytometry has been demonstrated [132]. Different genome-wide sequencing approaches are established using G4-specific antibodies (BG4) [133], G4 forming peptides [134], G4-binding molecules [91], [135], [136], mapping of G4-binding proteins or by using the ability of G4s to stop polymerases [15].

RNA G4-unwinding helicases

Most RNA molecules exhibit a precisely defined three-dimensional arrangement. Still, the RNA folding landscape is jagged by many thermodynamically stable intermediates that kinetically trap long-lived misfolded species, particularly in vitro [137], [138]. The efficient RNA folding in vivo is secured by specialized proteins that guide RNAs to their native state and resolve kinetic folding traps [139], [140], [141]. Helicases represent the largest group of multifunctional, ubiquitous and pleiotropic DNA- and RNA-remodeling enzymes [142], [143]. They are classified as ATP-dependent enzymes and are sorted in six superfamilies (SF1-6) according to conserved sequence motifs and include dual-purpose DNA/RNA processing enzymes [144]. All the RNA helicases identified so far to intervene in rG4 metabolism (Fig. 1), belong to the superfamilies (SF) 1 and 2. SF1 and SF2 share two highly similar helicase domains resembling the recombination protein RecA [145] and include protein families with distinct sequence, structural and mechanistic features, such as the Pif1-like family in SF1 [146] or the DExD/H-box in SF2 [147], [148].

The SF1 helicase family is marked by the GDxxQ motif, although several variations are found in the group and shared with SF2. Based upon the primary structure the SF1 branches into three distinct families; UvrD/Rep-like, the Upf1-like and Pif1-like enzymes [149]. UvrD-like enzymes are predominantly SF1Aα 3′-5′ DNA helicases, whereas Upf1-like and Pif1 helicases are SF1Bα enzymes (with 5′-3′ processivity) and include examples of proteins that process both DNA and RNA. Slight variations are observed among the helicase motifs throughout the three families presumably correlating with their different motor properties. The founding member of the family, S. cerevisiae Pif1 (P07271), is a ubiquitous eukaryotic protein fulfilling an assortment of tasks in DNA maintenance including dG4 unwinding [98], telomerase inhibition at DNA breaks and Okazaki fragment maturation [150]. To date, the human helicase MOV10 and its testis paralog MOV10L are the sole SF1 members identified to process rG4s.

SF2 is the largest group and comprises most G4-processing helicases identified thus far. Its members share high structural similarities in the catalytic core and intervene in every aspect of RNA [142] and DNA metabolism [145], [149]. The catalytic core is characterized by a flexible linker that allows the correct arrangement of two RecA-like domains reported to be required for sequence-unspecific ssRNA binding and ATP loading [145]. Both RecA-like domains unspecifically contact ssRNAs at the phosphate backbone, whereas auxiliary flanking domains provide specific functions or interaction and recruitment of partner proteins [151], [152], [153]. A remarkable exception is the eukaryotic Initiation Factor 4A (eIF4A) that represents the archetype of a minimal core helicase [154]. The most prominent SF2 are the DEAD-box (from the Asp-Glu-Ala-Asp motif, also referred to as DExD-box due to motif variations, alias DDX with 40 human members) and the DEAH-box family (alias DHX, 15 human members) [142]. Only a limited set of DExD/H helicases has directional processivity with most exhibiting local, chaperone-like RNA unwinding. They are still subject to substrate secondary structure features or availability of flanking overhangs [155], [156].

DHX36

The ATP-dependent 3′-5′ DEAH-box helicase DHX36 (Q9H2U1), also known as RHAU (RNA helicase associated with AU-rich element) due to its regulation of mRNA stability through the binding of AU-rich elements (AREs) [157], is a ∼ 114 kDa SF2 helicase. It has two isoforms, which at least in HEK293T cells operate in different cellular compartments, the nucleus (isoform 1) or the cytoplasm (isoform 2) [158]. DHX36 is a multi-functional helicase that can act on different substrates [157], [159], [160] by efficiently recognizing and processing both dG4s (K ∼77 pM) and rG4s (K ∼39 pM) [161], [162], [163]. DHX36 function is linked to human health e.g. heart development [164], hematopoiesis [165] and spermatogonia differentiation [166]. Among other functions, at the DNA level DHX36 intervenes to resolve dG4s in gene promoters and thereby activating transcription [159], [161], [167], [168], [169], [170]. The detailed discussion of DHX36 function within cells is reviewed elsewhere [171]. Like other DExD/H members [172], [173], DHX36 bears two tandem RecA-like domains making up the core helicase and ATPase functions, a winged-helix DNA/RNA-binding domain, a “ratchet” like domain (RL) and the DExD/H-specific β-barrel oligonucleotide- oligosaccharide-binding (OB) domain [174] (Fig. 2). The N-terminal, DHX36 specific motif (DSM) confers the high affinity for G4s [175], [176] by contacting the outer G4-tetrad (Fig. 2) through hydrophobic interactions due to its amphipathic α-helix, as seen by solution NMR (PDB 2N21) [177], X-ray crystallography (PDB 6Q6R) [178] and molecular dynamics [179]. The crystallographic structures of the highly-conserved orthologues from B. taurus in complex with the dG4 from the MYC promoter (Q05B79 - PDB 5VHE) [180] and D. melanogaster with ssRNAs (Q8SWT2 - PDB 5N98) [181] elucidate the DSM-dG4 recognition in the context of the full-length protein and the role of the helicase core, the OB and ratchet domains in creating a positively-charged cage to complement the interaction, whereas the helicase core binds to 3’single-stranded substrates longer than 5 nt. Single-molecule fluorescence resonance energy transfer (FRET) experiments [180], [182] support the current model of a partial, ATP-independent G4 destabilization by DHX36 consequent to the binding, followed by the complete G4 resolution by an ATP-driven, translocation-based mechanism threading the single-stranded nucleotides through the helicase core channel [183], similar to double-stranded substrate processing seen in other DExD/H helicases [184], [185], [186]. The functions of the DHX36 auxiliary domains in processing G4s were further clarified for the M. musculus orthologue (Q8VHK9 – PDB 6UP4) [187]. Pre-steady-state reaction RNA remodeling experiments, i.e. under conditions allowing repeated cycles of enzyme-substrate interaction and easier readout as compared to steady-state reactions [188], allowed to assess the stringent ATP-dependence of mDHX36 for the complete remodeling of both dsRNA and rG4 in the presence of a complementary trap sequence preventing the re-annealing [187]. Substrate affinity and ATPase assays clarified the essential function of the OB domain in coupling the ATPase moiety to nucleic acid binding and thus for the ATP-driven remodeling of quadruplex and duplex structures alike. mDHX36ΔOB mutants in fact exhibit a reduced binding affinity for all substrates and basal ATPase activity comparable to mDHX36. Unlike the wild-type, the mDHX36ΔOB ATPase activity was not stimulated in the presence of the substrate pointing to the loss of coupling function from the OB domain. Similarly, the conserved N-terminal β-barrel domain interacting with the OB domain proved to be essential for contacting and remodeling both the duplex and the G-quadruplex [187].

Fig. 2

Structure and domains of Bos taurus Dhx36 in complex with the parallel quadruplex from the MYC promoter (PDB 5VHE). The DSM domain (dark red) is primarily responsible for the G4 (orange) recognition by contacting the exposed G4-tetrad through its hydrophobic surface. The interaction is complemented by the OB domain (purple) and RecA1 (light blue) engaging the G4 backbone by creating a positively-charge “cage”, which also contributes to the proposed unwinding mechanism. The N-terminal linker (NL) is depicted in white. Rotations refer to the upper-left model. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

The rG4 biology of DHX36 embraces the whole of post-transcriptional processing, including miRNA and pre-mRNA maturation, RNA degradation, telomeres maintenance and translational regulation. DHX36 intervenes in the rG4-mediated translational regulation through diverse mechanisms, yet not completely rationalized. There appears to be a twofold regulation fate for mRNAs based on the localization of rG4s and DHX36-binding at either the 5′- or 3′-UTR. A system-wide, nucleotide-resolution photoactivatable ribonucleoside enhanced crosslinking and immunoprecipitation (PAR-CLIP) [189] analysis in human cells mapped the preferential binding of DHX36 to rG4-prone mRNA sequences [158]. RNA-seq showed that a CRISP/Cas9 DHX36-KO resulted in a significant increase of target mRNAs, yet ribosome foot-printing and stable isotope labeling in cells (SILAC) coupled with mass spectrometry proved the concomitant reduction in ribosome occupancy and protein translation, respectively. The KO also induced the formation of stress granules and protein kinase R (PKR/EIF2AK2) phosphorylation which, along with the finding that mRNAs targets of DHX36 that harbor rG4s preferentially localize in stress granules, suggests that DHX36 functions in resolving stress responses to re-establish cellular homeostasis [158]. This regulatory dichotomy can be explained in the light of the regulation of NKX2-5 by DHX36 in M. musculus [164]. DHX36 -/- cells had significantly higher mRNA levels as measured by microarray and RT-qPCR, yet reduced protein at western blot. RNA pull-down showed a strong DHX36 association at 3′-ARE (AU-rich Elements), which was abolished by using mutant with depleted 3′-ARE, proving that the binding of DHX36 to ARE in the 3′-UTR labels the mRNA for degradation. This is supported by a similar experimental design that proved the negative regulation operated by DHX36 on the translation of the transcription factor PITX1 through processing 3′-UTR rG4s in its mRNA [190]. Conversely, either mutating the NKX2 rG4 upstream of a GFP expression vector or restoring DHX36 expression significantly increased the protein levels, proving the role of DHX36 in regulating translation by resolving the 5′-UTR rG4s [164].

A label-free, liquid chromatography-tandem mass spectrometry (LC-ESI-MS/MS) global proteomic analysis upon DHX36 RNAi silencing identified a small set of significantly downregulated proteins enriched in 5′-UTR rG4s [191]. BCL2 and NRAS bear well-characterized examples of 5-UTR rG4-mediated translation regulation. Nonetheless, neither BCL2 [192] nor NRAS [193] are altered upon DHX36 silencing [191], perhaps due to functional redundancy of other helicases, e.g. of the paralog DHX9 [194]. The two paralogs indeed appear to regulate the translation efficiency, measured by transcriptome-wide ribosomal profiling, on a common subset of transcripts in HeLa cells [194]. Noteworthily, the multifunctional ATP-dependent RNA helicase DDX3X (O00571), whose rG4-unwinding activity still lacks evidence, was identified to interact with the NRAS 5′-UTR rG4 along with two known rG4 helicases, DDX5 and DDX17 [101].

The rG4 biology of DHX36 spans over the modulation of genome stability by regulating telomere maintenance and the expression of the gatekeeper p53. Telomeres, the terminal structures protecting most eukaryotic chromosomes from degradation, end-to-end joining or recognition as double-strand breaks (DSBs) [195], [196], are maintained by telomerase, a reverse transcriptase harboring an internal RNA template for the synthesis of telomeric repeats. The template RNA 5′-end controls the transcription termination through a stem-loop structure (P1) [197], [198] which, due to its high guanine content, is challenged by an rG4 shown to disrupt the P1 helix [199], [200]. DHX36 directly binds this rG4 as demonstrated by independent non-crosslink co-immunoprecipitation/RT-qPCR [201], [202] and non-crosslink RNA immunoprecipitation microarray analyses [203]. Knockdown (KD) of DHX36 by siRNAs leads to decreased telomerase efficiency and shorter telomeres [201], as assessed by monochrome multiplex quantitative PCR [204].

DHX36 also indirectly controls genomic stability by processing a 3′-rG4 in the pre-mRNA of p53 [205], the “guardian of genome stability” [206], through a seemingly different mechanism. The direct interaction between DHX36 and the p53 rG4, proved by RNA immunoprecipitation, is necessary to upkeep the p53 pre-mRNA 3′-end processing upon UV-induced damage. The pre-mRNA 3’-end processing efficiency, as assessed by the ratio between unprocessed RNA and total nuclear RNA by RT-qPCR, was significantly diminished upon DHX36 siRNA KD, resulting in reduced p53 response to the UV-induced damage. Similar effects were obtained using G4-stabilizing ligands [207].

The DHX36-mediated resolution of rG4s is also required for correct miRNA maturation, a stream of coordinated processes pervading many cellular functions [208]. Interestingly, potential rG4s map to miRNA regions [187] and several rG4s are proved to compete with the hairpin structures necessary for pre-miRNA cleavage by Dicer [209], [210], [211], thus prospecting vast implications of rG4s for miRNA-controlled pathways. DHX36 was recently proved to intervene with the maturation of miR26-a [212], a miRNA involved in multiple processes and regarded as a tumor suppressor that is found downregulated in cancers [213], [214], [215]. The miR26-a precursor harbors a stable rG4, as seen by CD and EMSA, which impairs the post-transcriptional maturation with detrimental effects both in vitro and in vivo [212]. MiR26-a expression was higher in cells transfected with a plasmid bearing a mutant, not rG4-forming sequence than with the wild-type sequence, which was also sensitive to the G4-stabilizer PDS. Also, DHX36 overexpression significantly enhanced miR26-a levels only in the wild-type, as measured by RT-qPCR [212].

DHX9 (DDX9)

The 3′-5′ ATP-dependent DEAH-box helicase DHX9 (Q08211), also known as nuclear DNA helicase II (NDH II) or RNA helicase A (RHA), is a paralog to DHX36 and also a member of SF2. DHX9 bears an RGG-box but lacks the N-terminal G4-binding domain present in DHX36 and intervenes in several cellular processes, both at DNA and RNA level. These include replication [216], transcription [217], [218], translation [219] and maintenance of genomic stability [220]. The complex biology of DHX9 was recently reviewed [221].

Helicase and competition assays proved DHX9 to bind and resolve G4s with stronger kinetics toward rG4 and strict ATP-dependence in vitro, as confirmed by the complete loss of G4-unwinding function upon ATPase inhibition by N-ethylmaleimide poisoning [222]. More recently, DHX9 was confirmed to directly intervene in rG4 biology in vivo as a translational modulator of mRNAs harboring rG4s in their 5′-UTR [194]. Polysome fractioning coupled with mass spectrometry demonstrated that DHX9 is bound to fully-assembled polysome, thus actively translating mRNAs. A further indication of an rG4-specific translational modulation role was provided by cell proliferation and eIF2 phosphorylation (a marker of the global translation efficiency) assays upon DHX9 siRNA KD: neither DHX9-KD nor DHX36-KO had a relevant impact on the overall translation, advocating for functional redundancy. However, the translational efficiency of a subset of mRNA with 5′-UTR rG4s was significantly impacted. Individual-nucleotide resolution UV crosslinking and immunoprecipitation (iCLIP) [223] followed by deep sequencing experiments proved the correlation with 5′-UTR rG4s sequences. DHX9 peaks and rG4s centered ∼ 40 nt apart on average possibly reflecting the helicase loading downstream of the rG4 in a pre-processing stage [194] in agreement with the reported need for a 3′single-stranded tail for the DHX9 helicase function [224]. The same work confirms the DHX9-dependent (and DHX36 alike) translational modulation by GFP-reporter assay [225].

DDX5/DDX17/Dbp2 subfamily

The human DEAD-box paralogs DDX5 (P17844) and DDX17 (Q92841) and the S. cerevisiae orthologue Dbp2 (P24783) [226] belong to the SF2 and share high sequence similarity among each other [227]. The two human paralogs mostly differ in the N-terminal domain likely responsible for yet uncharacterized specialized functions. The C-terminal proline-rich element likely reported to protein–protein interactions [228] distinguishes both from non-mammalian orthologs. All three helicases are predominantly nuclear [229], [230]. There is evidence that DDX5 is a shuttling protein [231] whose localization varies in different cell lines [232] and as a function of the cell cycle [233], [234], [235]. This possibly reflects the changing G4 landscape throughout the cell cycle. DDX5 lacks sequence stringency in vitro with conflicting evidence regarding the helicase directionality in processing dsRNA [126], [134], [135], [136]. Accordingly, the DDX5/Dbp2 subfamily functions in virtually every step of the post‐transcriptional gene regulation [236], [237], [238], [239], [240], [241] and correlates with cancer onset and progression [242], [243], [244], [245], [246], HIV [247], RVFV (Rift Valley Fever Virus) [248] and general innate antiviral response [249]. KD experiments also prove the functional redundancy of the human paralogs [250]. The vast non-G4 biology of the DDX5/Dbp2 subfamily is detailed in a recent review [227].

The first indirect evidence for the involvement of DDX5 and DDX17 in rG4 biology came from investigating the co-dependent functionality with hnRNP H/F, a splicing factor previously reported to bind G4-prone sequences [251] and recently shown to intervene in rG4-mediated translational regulation [252]. The silencing of either helicase by siRNA significantly impacts splicing patterns prompting the authors to postulate that DDX5/17 could regulate mRNA splicing by binding of hnRNP H/F to G-rich target motifs otherwise embedded within folded G4s. Further, the transient expression of either DDX5/17, but not helicase-impaired mutants, can rescue the phenotype, bolstering the functional redundancy and backing the central role of the helicase function. This model is supported by the observation that the treatment with the G4-stabilizer TMPyP4 [253] had the same effects on splicing as seen with siRNA silencing. It has to be noted that the ligand is unspecific and the effects cannot be univocally ascribed to the stabilization or rG4s. Proximity ligation assays [254] confirmed the close interaction of the two helicases with hnRNP H/F in vivo. However, these were not definitive evidence that the function was actually due to the unwinding of rG4.

Direct observation of the DDX5 and DDX17 interaction with rG4s has been obtained using an rG4-bait pulldown coupled with LC-mass spectrometry which, among other proteins, also identified hnRNP H/F [101]. The same experimental approach in S. cerevisiae [255] identified the homologous Dbp2 and three other DEAD-box RNA helicases, Ded1 (P06634) [256], Dbp1 (P24784) [257] and Mss116 (P15424) [258]. Surprisingly, fluorescence anisotropy and experiments using a trap single-strand complementary sequence sequestering the rG4-forming sequence upon unfolding [259] showed that Ddp2 destabilizes rG4s regardless of ATP antipodal to previous findings with dsRNA [226] and jarring with in vivo findings using the ATPase-deficient mutants [251] whilst confirming a threefold binding affinity enhancement by 3′-overhangs.

Finally, Wu and colleagues demonstrated DDX5-binding and unwinding activity on both dG4 and rG4 using biotin-dG4s immobilized on streptavidin-coated plates and ELISA readout [97]. They also provided proof of DDX5-dG4-mediated modulation of the oncogene MYC transcription in vivo. The unwinding activity was tracked by FRET using a dG4 from the MYC promoter labeled with 6-fluorescein (6-FAM) at the 3′-end and black hole-1 quencher (BHQ-1) at the 5′-end. FRET with either shorter or no flanking regions returned similar results indicating that DDX5 does not need 3′-overhangs for its activity. Competition assays on the other hand support the influence of siding regions on binding affinity, at least with dG4s, because the dG4 “MYCpu28” binding appears to be an order of magnitude higher. Surprisingly, both K144N (unable to bind ATP) and D248N (unable to hydrolyze ATP) mutants [260] were able to destabilize dG4s, as also confirmed in glucose/hexokinase ATP-free experiments (ATP-dependent phosphorylation to 6-p-glucose) [97]. Nonetheless, whether ATP is required to resolve the more stable rG4s remains unclear, because only data from dG4 was reported. Whilst not producing effects on the unwinding activity, ATP significantly reduced the binding of DDX5 to immobilized dG4, suggesting it might be required for unloading the unfolded oligonucleotide from the protein, as also backed by DMS protection assays [97].

DDX21

The DEAD-box RNA helicase DDX21 (Q9NR30), alias nucleolar RNA helicase 2, was initially proposed to possess a 5′-3′ ATP-dependent core helicase domain [261] and an ATP-independent C-terminal “foldase” function [262], [263], [264], i.e. able to remodel secondary structure in a seemingly single-stranded RNA. However, a recent FRET continuous helicase assay [265] proved its dsRNA helicase activity to be stimulated by, but not to depend on, ATP [266]. DDX21 intervenes in fundamental aspects of RNA metabolism. It is essential for ribosomal RNA biogenesis [267], [268], [269] in tandem with RNA Pol I-mediated transcription [270] and for pre-ribosomal complexes assembly [271]. Functions of DDX21 have also been reported in the immune response by double-stranded RNA sensing [272], regulation of viral RNA synthesis [273], HIV replication [274], HCMV (human β-herpesvirus 5) replication [275] and activation of the innate antiviral response [276], [277]. DDX21 acts as an epigenetic modificatory of gene expression [278] and correlates with cancer [279]. DDX21 is abnormally expressed in colorectal [280], [281], [282] and breast cancers [283]. It intervenes in R-loops regulation [275], [284], [285] and acts as a sensor and mediator of transcription during the nucleotide stress response [286].

DDX21 was identified to interact with rG4s by a streptavidin pull-down coupled with LC-mass spectrometry from HEK293T cellular lysates using the rG4 from the PITX1 3′-UTR as a bait [287]. Transient expression and pull-down of different DDX21 truncations indicated that the C-terminal DDX21(574–783) domain is necessary and sufficient for binding rG4s. However, the binding affinities determined by microscale thermophoresis (MST) [288], [289] demonstrated that the recognition of rG4s also involves other domains of the protein, as seen with DHX36. MST [289] and isothermal titration calorimetry [290] confirmed the direct interaction of DDX21(574–783) with the human telomeric rG4 TERRA [291] in the nanomolar range with a remarkable higher affinity compared to dG4 [292]. The bulk interaction is driven by electrostatic interactions as observed by a reduction in affinity at higher ionic strength. Bidimensional saturation transfer difference NMR allowed mapping the contact with the 2′-OH of loop nucleotides, which might explain the rG4 selectivity [292].

CD measurements and thioflavin T-based assays proved the G-rich RNA sequence used to define the foldase activity [261], [262], [263], [264] to fold into an rG4 [287] and led to question whether such observations could rather be the result of rG4 unwinding activity, rather than a dsRNA re-modeling. In facts, digestion with either RNase A or T1, enzymes that cut at specific single-stranded regions (T1 cuts after guanines, RNase A after cytosines and uracils [293]) showed significant differences only for the T1 cutting pattern in the presence of DDX21, indicating a change in the accessibility of guanine residues but not in cytosine or uracil as a result of DDX21 rG4 unwinding [287]. Altogether, DDX21 appears to bind and process rG4 primarily through its conserved (F/PRGQR) C-terminal repeat, yet, with the determinant support from other domains [292]. This was recently confirmed by RNase T1-based helicase assays using different DDX21 mutants [294].

Gene reporter assays upon transient DDX21 silencing by siRNA provided in vivo evidence of DDX21 to suppress genes with 3′-UTR rG4 as a consequence of its rG4-unwinding activity, regardless of ATP [287]. DDX21 was also shown to regulate the translation of MAGED2 (Q9UNF1), a transcriptional repressor of tumor necrosis factor-related apoptosis-inducing ligand (TRAIL) death receptor 2 (TRAIL-R2 - O14763) mRNA [295], [296] through the processing of its 5′-UTR rG4 [297]. MAGED2 was identified as a candidate DDX21-regulated protein by differential proteomic on HEK293T cells expressing either the wild-type or a helicase impaired DDX21 mutant. This latter was unable to express MAGED2 at a luciferase assay [297].

DDX2 (eIF4)

The ATP-dependent DEAD-box DDX2 helicase family [298], commonly known as the eukaryotic initiation factor 4 (eIF4), comprises three paralogs: DDX2A (P60842) and DDX2B (Q14240) (eIF4A1 and A2, respectively) are highly similar cytosolic helicases that as part of the eIF4F complex exhibit the critical ATP-dependent [299], [300], [301] unwinding of 5′-UTR secondary structures for the recruitment of the 40S complex in cap-dependent initiation of translation [302], [303], [304], [305]. The two are largely redundant and will be interchangeably referred to as DDX2, although there is emerging evidence of different functions in vivo [306], [307]. A specific role of DDX2B in esophageal squamous cell carcinoma was recently reported [308]. DDX2 upregulation drives translation reprogramming in pancreatic [309], [310], breast [311], [312] and other cancers [302], [313], [314], [315]. It is sequestered in trophoblasts to re-modulate the global protein production through a recently discovered mechanism [316]. DDX2 is also hijacked to support viral infections through different mechanisms [317]. The third paralog, DDX48 (P38919 - eIF4A3), is localized in the nucleus and is part of the exon junction complex [318], [319], [320].

Although it is evident that DDX2 is necessary for modulating the translational efficiency of mRNAs by processing secondary structures in their UTRs the extent this can be ascribed to G-quadruplexes is subject to conflicting findings. The first evidence for the intervention of DDX2 in rG4-mediated processes was reported by Wolfe et al. [321]. Ribosomal foot-printing and deep sequencing with and without the specific DDX2 inhibitor silvesterol showed reduced translational efficiency for a set of transcripts with enriched putative rG4-forming motifs in their 5′-UTR. Notably, many were transcription factors and oncogenes, in agreement with other findings [312], [322]. In firefly luciferase assays the translational efficiency of constructs with 5′-UTR rG4s was significantly reduced over scrambled-sequence controls, in the presence of silvesterol. This could be reversed by overexpressing DDX2 or using the silvesterol-insensitive P159Q DDX2 mutant [323] leading to conclude that DDX2 helicase activity is a limiting factor for the translation of mRNAs harboring 5′-UTR rG4s, effectively adding up an extra regulatory layer to such set of transcripts.

This generalized model is, however, questioned by later reports from Waldorn and co-workers [324], [325]. Reverse transcriptase stop and in vitro translation assays reported only three-layered rG4s (rG4(3)) to form in full transcripts and to exhibit sensitivity to the DDX2 inhibitor hippuristanol [324], [326]. Two-layered rG4s (rG4(2)) could not form in full transcripts possibly due to competing hairpin structures. Only under G4-stabilizing conditions, i.e., in the presence of the G4-stabilizing ligand PDS, rG4(2) could reduce the translation efficiency. This stands in contrast to finding of Wolfe et al. [321]. Experiments with constructs bearing 7-deazaguanine (c7G), which impairs Hoogsteen but not Watson-Crick pairing [327], confirmed a greater translational gain for c7rG4(3) over the controls. However, c7rG4s(3) were still significantly sensitive to hippuristanol, suggesting the intervention of alternative, non-rG4 secondary structures. Two constructs bearing either a GGAx or GGCx motif (the former likely forming rG4s but less enriched in the foot-printing analysis, the latter able to form stable non-G4 secondary structures) were compared with and without either PDS or hippuristanol. PDS had a significant effect only on the rG4(2) sensitivity to hippuristanol suggesting that the DDX2 translational regulation is more reliant on canonical hairpin structures than on rG4s, notwithstanding translational repression operated by rG4(3) [324].

A combination of polysome profiling, the DMS-based Structure-seq2 method [328] and DDX2 inhibition by hippuristanol was later used to investigate the structural composition of translationally DDX2-dependent 5′-UTRs [325]. A set of DDX2-dependent transcripts with longer than average 5′-UTRs was enriched in Cs, but not Gs. Computational prediction of rG4 using G4RNA [329] did not show any significant prevalence compared to a set of DDX2-independent mRNAs. Similarly, DMS Structure-seq2 under DDX2-inhibition did not point out any significant difference between the two sets of transcripts, further advocating that the enrichment in (GGC)4 motif in the DDX2-dependent set is due to canonical secondary structures outcompeting the more common rG4s(2) [325].

However, indirect evidence for the rG4 role comes from the oligonucleotide inhibitors of the 48S-PIC complex ES6S region, which constitutes the extended binding channel for the DDX2-mediated mRNA unwinding and scanning [330]. The ES6S region consists of four RNA helices at the bottom of the 40S ribosomal subunit. Polysome mapping upon ES6S inhibition showed significant enrichment in rG4-forming sequences in the set of transcripts with reduced translational efficiency. To note, the authors used the rG4-search algorithm QuadBase2 [331]. The translational efficiency dropped consistently with the 5′-UTR rG4s stability by comparison with constructs bearing more stable rG4 motifs from BCL2. Furthermore, ribosome profiling after treatment with the novel DDX2 inhibitor eFT226 also yielded enriched rG4-forming sequences (the applied search criteria are unclear) [332]. The translational dependence on the original 5′-UTR was proved by comparison with constructs unable to form rG4s. Neither work offers definite evidence for the degree to which the DDX2-dependent translational control relies on rG4s or outcompeting canonical secondary structures, because there is no direct proof for the rG4 folding. The KRAS-controlled, DDX2-mediated translation of the 5′-UTR rG4-containing ARF6 led to similar conclusions [333].

DDX1

The DEAD-box RNA helicase DDX1 (Q92499) has three human splicing isoforms that localize both in the nucleus and the cytoplasm and exhibit a 5′single-stranded RNase activity and participate in several processes, including translation activation [334], splicing regulation [335], fatty-acid dependent insulin regulation [336], double-strand break repair [337] and microRNA maturation [338]. Loss of DDX1 results in embryonic lethality [339] and severe gametogenesis defects [340]. It also has a multifaced role in either inhibiting or facilitating viral infections, including HIV [341], tumorigenesis [342] and cancer progression [343].

DDX1 was shown to bind and process lncRNA rG4s, facilitating R-loop formation and thus assisting in the recruitment of the activation-induced cytidine deaminase (Q9GZX7) to initiate IgH class recombination [344]. DDX1 was pulled down using the S-region containing rG4-forming sequences as a bait, but not with scrambled controls or under Li+ rG4-destabilizing conditions. DDX1 was also extracted from CRISPR/Cas9 AID-KO cells demonstrating that its interaction is AID-independent, as verified by native EMSA assays. Furthermore, DDX1-K52A mutants with impaired ATP-binding capability tended to form a more stable complex but it is uncertain whether this is due to decreased ATP-dependent unwinding or decreased ATP-dependent substrate turnover, i.e., substrate release. Interestingly, the interaction is RNA-specific as no interaction was detected with dG4s. Mouse lymphoma CH12 cells (CVCL_6818) [345] transiently transfected with DDX1-K52A showed reduced class switch recombination and R-loops. In vitro assay with 32P-rG4 and complementary trap DNA strands proved DDX1 to directly convert rG4s into hybrid RNA:DNA R-loops. This depends on ATP because DDX1-K52A was significantly less efficient in the conversion. To date, this is the only direct example of the intervention of DDX1 in rG4 biology [344].

MOV10/L

The mammalian ATP-dependent 5′-3′ SF1 RNA-helicases MOV10 (Q9HCE1) unspecifically loads on single-stranded RNA proximal to the stop codon and translocates along 3′-UTRs intervening in nonsense-mediated decay [346] and suppression of retrotransposition [347], [348]. Its testis-specific paralog MOV10L1 (Q9BXT6) is critical for the control of PIWI-interacting RNA (piRNA) biogenesis. piRNAs are important for retrotransposon silencing and protect the genome integrity of post-meiotic germ cells [349]. In fact, impaired ATP-binding and ATP-hydrolysis motifs cause aberrant piRNA biogenesis and male sterility [350], [351]. Both paralogs preferentially bind G-rich sequences, as proved by CLIP-seq experiments [351], [352].

In vitro helicase assays using complementary nucleotide trap sequences (to prevent rG4 refolding) showed both paralogs of MOV10 unwind rG4 depending on ATP. Neither helicase could process dG4s. The ATP dependence was confirmed using the non-functional K778A and DE888AA mutants and the non-hydrolyzable ATP-analogues AMP-PNP or ATPγS [353]. MOV10L, which requires the N- and C-terminal domains to unwind rG4s also proved higher unwinding efficiency than MOV10 [353]. The MOV10 N-terminal domain is necessary and sufficient for the rG4-unwinding function, as seen by mutagenesis and luciferase assays [354].

MOV10 was showed to intervene in the regulation of miRNAs by exposing the 3′-UTR miRNA recognition elements (MRE) to FMRP (Q06787), the fragile X mental retardation RNA binding protein, possibly through unwinding rG4s [354].

CNBP, a non-helicase rG4-resolving protein

Helicases are defined as motor proteins that move directionally along the phosphodiester backbone resolving secondary or tertiary nucleic acid structures using energy from ATP hydrolysis [355]. Although largely responsible for the G4 processing in the cell such classification is increasingly failing to embrace proteins still able to untie and impact the biology of G4s.

One example is the cellular nucleic acid binding protein (CNBP – P62633), a conserved eukaryotic zinc finger protein featuring seven CCHC-type ZnF domains and an arginine-glycine (RG/RGG)-rich motif. CNBP is ubiquitous and highly expressed in adult tissues, essential for embryonal development in mice and was reported to relate with myotonic dystrophy type 2 [356]. It binds to both G-rich single-stranded DNA [357] and RNA [358] in vitro and is demonstrated to promote the translation of specific mRNAs by associating to their 5′ oligopyrimidine sequences [359], [360]. PAR-CLIP analysis [189] mapped CNBP onto mRNA G-rich regions proximal to start codons previously reported to form rG4s or other secondary structures [59]. The translational regulatory role of CNBP could be defined by ruling out a significant effect on transcription given only minor, yet statistically significant, changes in target mRNA abundance in CRISPR/Cas9 CNBP-KO cells, as determined by transcriptome-wide RNA-seq. Noteworthy, this sufficed in reducing long-term cell viability (15 days), suggesting a specialized function of CNBP in supporting cell survival. Conversely, Ribo-seq and mass spectrometry highlighted a sizeable reduction of CNBP targets at the protein level pointing out the regulatory role of CNBP onto the translation of such mRNAs by de-structuring rG4s within the CDS, as supported by in vitro SHAPE and CD experiments [95] and consistent with previous findings [59]. Substantial biophysical and biochemical evidence for the ATP-independent CNBP unfolding of G4s, although limited to dG4s, was independently gathered by a combination of CD, NMR, polymerase stop assay, EMSA and unwinding PAGE assays [19]. Accordingly, CNBP might disrupt the G4 folding-unfolding equilibrium by preferentially binding to the unfolded species.

Concluding remarks

G-quadruplexes, which are somehow still considered non-canonical, are becoming more accepted as a regulating tool within cells that can alter gene expression and/or maintenance of genome stability. Accumulating evidence about their complex biology moved G4s from being considered “mere” targets for small molecules in the anticancer perspective toward a deeper understanding of their functions in a variety of cellular processes, most often at the DNA level. The knowledge about the RNA counterpart has long lagged owing to peculiar experimental difficulties in assessing their existence in vivo because of the inherent ephemeral RNA nature and has eventually been called into question. The major claim against rG4s is now clarified as resulting from an inherent weakness in the DMS/RT-stop profiling assay, which is better suited to determining G4 equilibrium shifts rather than capturing snapshots of folded rG4s [60]. Besides, rG4s are also supported by interesting biological evidence gathered by independent novel approaches, all of which point to a complex protein-regulated rG4 scenario with far-reaching implications in living cells. In vivo, rG4s can now be thought of as fast-paced, transient events shaping a swinging rG4 environment determined by exogenous or endogenous stimuli and regulated by the cellular regulatory machinery, primarily helicases. Anything from splicing, mRNA stability, miRNA and piRNA maturation, telomeres maintenance, IgH class recombination and transcriptional regulation is now clarified to be controlled by the helicase intervention on rG4s. However, despite recent advances in the structural and biochemical understanding of helicase-mediated rG4s processing a consistent and conclusive mechanistic model appears elusive.

The rG4s biology increasingly appears to reach far beyond individual cellular processes. It rather embraces systemic aspects of the cellular functioning with specific rG4s subset coordinated through the cell cycle, possibly idiosyncratic across cell types and tissues. However, the knowledge of the rG4 biology and more so its helicase/non-helicase regulation is limited to a fairly small set of cell types. Besides, along with several examples in the context of dG4s the paradigm of helicases as the sole source of negative G4 control in the cell is further called into question by the identification of CNBP as the only example (as yet) of non-helicase rG4-destabilizer. It is then utterly relevant to clarify the regulatory network beyond helicases with likely many more direct or indirect rG4 modulators to be identified. Indeed, the network of processes controlled by rG4 appears far too extended and sensitive to the cell to be sufficiently fine regulated by the relatively small number of proteins identified so far. These considerations lead to many more pending questions: how are the regulators regulated themselves? What are the cellular signals and intermediate modulators driving the helicase-mediated and non-helicase response to rG4 after a given stimulus? At last, it is worth noting the knowledge of the rG4 biology and more so its helicase/non-helicase regulation is limited to a fairly small set of mostly transformed cell lines.

CRediT authorship contribution statement

Marco Caterino: Data curation, Investigation, Writing – original draft. Katrin Paeschke: Supervision, Funding acquisition, Conceptualization, Writing – original draft, Writing – review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.