hDicer binds RNA G-quadruplexes
The 12-nt guanine-rich RNA, found to bind to hDicer, was perfectly complementary to the apical loop of pre-mir-210, and consequently, it was named “AL-210” . We hypothesized that the structure adopted by AL-210 determined its ability to bind to hDicer and to inhibit its cleavage activity. Accordingly, we used Fold and bifold algorithms provided by the RNA structure web server  to predict the lowest free energy secondary structures for AL-210 and three other 12-mers used in our previous studies: AL-16-1, AL-21, AL-33a (i.e., the oligomers designed to target the apical loops of pre-mir-16-1, pre-mir-21, pre-mir-33a, respectively). The results indicated that AL-33a and AL-210 can form stable homodimers (Supplementary Fig. S1). AL-16-1 and AL-21 were predicted to be monomeric. Since the structure adopted by RNA may depend on the RNA concentration, each 12-nt RNA was assayed by polyacrylamide gel electrophoresis (PAGE), under nondenaturing conditions, at four different concentrations (0.01, 0.1, 1, and 10 µM). Under the applied conditions AL-16-1 and AL-21 always migrated in the gel as a single conformer, whereas AL-33a and AL-210, depending on the concentration, migrated as faster- or slower-moving conformers. The faster-moving conformers seemed to represent monomers (single-stranded RNA, ssRNA) (Fig. 1a). At concentrations equal to or greater than 0.1 µM, AL-33a migrated as the slower-moving conformer, presumably corresponding to a homodimer (dsRNA) form. Interestingly, at a concentration equal to or greater than 1 µM, AL-210 migrated even slower than a putative dsRNA form (Fig. 1a and Supplementary Fig. S1).
When analyzing the primary structure of the 12-mers (Fig. 1b), we noticed that a striking feature of AL-210 is a stretch of four guanines at the 5′ end. Such a sequence motif is commonly found in nucleic acids adopting G-quadruplex structures. To test the possibility that the slower-migrating conformer of AL-210 represents a G-quadruplex form, we performed native PAGE followed by G-quadruplex-specific staining with the N-methyl mesoporphyrin IX (NMM) fluorescent probe . To promote and stabilize a potential G-quadruplex structure, oligomers were incubated in buffer containing 100 mM KCl before the electrophoresis. As a positive control, we used a 14-nt RNA (called “QU14”) known to adopt a G-quadruplex architecture . Under the applied conditions, QU14, AL-16-1, AL-21, and AL-33a each migrated as a single conformer (Fig. 1c, left panel), whereas AL-210 migrated as faster- and slower-moving conformers, as demonstrated above (Fig. 1a). Staining with NMM showed that the G-quadruplex-specific dye was bound only by QU14 and the slower-moving AL-210 conformer, but not by AL-16-1, AL-21, AL-33a, or fast-moving AL-210 conformers (Fig. 1c, right panel). Together, these results suggest that AL-210 adopts a G-quadruplex structure and, presumably, in such a form, it could bind to hDicer.
We hypothesized that hDicer could also bind other RNAs containing G-quadruplex structures, including QU14. We performed electrophoretic mobility shift assays (EMSAs) involving hDicer and 5′-32P-labeled QU14 or a control 14-nt RNA (called “LIN14”) that did not adopt a G-quadruplex structure. QU14 was bound efficiently by hDicer, while only residual binding of LIN14 was detected (Fig. 1d). Weak binding of LIN14 agrees with our previous observations that hDicer does not interact efficiently with RNAs shorter than 20-nt . Next, we tested whether hDicer would also bind other short RNAs (< 20-nt) with G-quadruplex structures. Accordingly, we performed EMSAs with hDicer and 5′-32P-labeled oligonucleotides with sequences corresponding to the well-characterized telomeric repeat-containing RNAs (TERRAs) of human  or ciliate Oxytricha nova  origin, i.e., r(GGGUUAGGGU) called “TER10” and r(GGGGUUUUGGGG) called “G4U4G4”, respectively (detailed information about guanine-rich oligomers used in the studies is presented in Table S1). Both TER10 and G4U4G4 were bound by hDicer. A control binding reaction contained the canonical hDicer substrate, 58-nt pre-miRNA (Fig. 1e). These results confirm that hDicer is able to bind RNAs adopting a G-quadruplex structure.
The PPC cassette of hDicer binds RNA and DNA G-quadruplexes
According to the current knowledge about the mechanism of Dicer action, the initial recognition and anchoring of the substrate occur within the region spanning the Platform–PAZ–Connector helix cassette, called “PPC” . The PPC contains two adjacent pockets, a 2-nt 3′-overhang-binding pocket (3′-pocket) within the PAZ domain and a phosphate-binding pocket (5′-pocket) within the Platform and PAZ domains (Fig. 2a) [3, 23]. To test whether PPC can also bind RNA G-quadruplexes, we produced PPC in a bacterial expression system (Supplementary Fig. S2A) and then carried out EMSAs with PPC and either 5′-32P-labeled r[AGGG(UUAGGG)3] called “TER22”, which is an extended version of TER10, or G4U4G4 (Table S1). PPC formed stable complexes with both RNA oligomers, with a Kd value of ~7 nM for the PPC and TER22 complex (Fig. 2b), and ~10 nM for the PPC and G4U4G4 complex (Fig. 2c). For comparison, we also determined binding affinities for PPC and 5′-32P-labeled 21-nt RNA (called “LIN21”), 58-nt pre-mir-21, and 19-bp RNA duplexes having 2-nt 3′ overhanging ends (called “dsRNA_OV”) (Supplementary Fig. S2B). The Kd values calculated for PPC and RNA substrates were ~46 nM for LIN21, ~205 nM for pre-mir-21 and ~50 nM for dsRNA_OV (Supplementary Fig. S2C). The results indicate that the PPC cassette of hDicer binds RNA G-quadruplexes with higher affinity than it binds a pre-miRNA substrate, a miRNA-size RNA, or a miRNA-like duplex.
Given the fact that telomeric repeat-containing sequences are present not only in telomeric RNA but also within telomeric DNA at the chromosome ends, we asked whether PPC could interact with DNA G-quadruplexes as well. Accordingly, we used 5′-32P-labeled DNA oligomers with sequences identical to TER22 and G4U4G4, i.e., “TEL22” and “G4T4G4”, respectively (Fig. 2d, e). The Kd value was ~318 nM for the PPC and TEL22 complex and ~369 nM for the PPC and G4T4G4 complex. These data indicated that PPC bound DNA G-quadruplexes much weaker than their RNA G-quadruplex counterparts. Additionally, TER22 and TEL22 were bound by PPC slightly better (~1.4 and ~1.2 times) than G4U4G4 and G4T4G4, respectively. Together, these results indicate that both the type and length of nucleic acid adopting a G-quadruplex structure determine its binding affinity to the PPC cassette of hDicer.
Both the 3′-pocket and the 5′-pocket of hDicer PPC cassette are involved in the binding of G-quadruplexes
To test whether oligonucleotides adopting G-quadruplex structures could, like canonical Dicer substrates, be bound in the 3′-pocket and/or 5′-pocket of the PPC cassette, we performed molecular docking and modeling. For this we used the structure of hDicer PPC cassette (PDB entry 4NGF)  and the G-quadruplex structures of the 10-nt human TERRA (TER10) (PDB entry 2M18)  or O. nova G4T4G4 (PDB entry 1JPQ) . TER10 formed a dimer of bimolecular G-quadruplexes , whereas G4T4G4 appeared as a bimolecular G-quadruplex .
We inspected docked poses of both TER10 and G4T4G4 with hDicer protein for clashes with pre-miRNA binding (for detailed information see “Materials and methods” section). We selected eight best models (docked poses) for the TER10-PPC complex and six best models for the G4T4G4-PPC complex. In seven out of the eight models selected for the TER10-PPC complex, TER10 was positioned within the 3′-pocket of the PPC cassette. In all these models, the 3′ end of one out of the four RNA strands was anchored in the 3′-pocket (Fig. 3a). In one of the eight models, the 3′ end of TER10 was located in the vicinity of amino acid residues lining up the 5′-pocket (Fig. 3b). Three docked poses obtained for the G4T4G4-PCC complex showed that the G4T4G4 quadruplex bound within the 3′-pocket (Fig. 3c), and the other three poses revealed G4T4G4 located within the 5′-pocket of the PPC cassette (Fig. 3d).
To validate these models, we generated two variants of the hDicer PPC cassette, one containing two substitutions in the 3′-pocket (Y926F/R927A variant), and the other containing six substitutions in the 5′-pocket (R778A/R780A/R811A/H982A/R986A/R993A variant) (Supplementary Fig. S3A). These changes significantly affect the binding of small interfering RNA (siRNA) by hDicer PPC cassette , and the corresponding full-length hDicer variants have been well characterized biochemically . Using EMSAs, we examined the ability of PPC variants to bind 5′-32P-labeled TER22, G4U4G4, TEL22, and G4T4G4 oligomers. Changes in both the 3′-pocket and the 5′-pocket reduced PPC binding to all G-quadruplexes analyzed (Fig. 3e). Mutations in the PPC pockets had a greater impact on binding to the DNA G-quadruplexes than to their RNA counterparts. However, this can be explained by the fact that, under the applied reaction conditions, the wild-type protein displayed much lower maximum binding capacity for DNA G-quadruplexes than for the corresponding RNA G-quadruplexes (Fig. 2b–e), and mutations in the PPC pockets only proportionally decreased binding. The lower binding capacity of PPC to DNA G-quadruplexes than to RNA G-quadruplexes may be due to the participation of ribose 2′-hydroxyl groups in hydrogen bonding with amino acid residues of the PPC cassette. Taken together, binding assays conducted for the wild-type protein and the 3′-pocket and the 5′-pocket variants (Fig. 3e) corresponded well with the models of PPC and G-quadruplex complexes (Fig. 3a–d), collectively indicating that both pockets of the hDicer PPC cassette are important for the binding of G-quadruplexes.
To investigate whether the free ends of the RNA adopting a G-quadruplex structure are necessary for its binding to the hDicer PPC cassette, we circularized 5′-32P-labeled TER22 using T4 RNA ligase, as described previously  (Supplementary Fig. S3B, C). In adequate protein concentrations, PPC bound this “circTER22” with similar efficiency as the TER22 RNA with free ends (Fig. 3f, g), indicating that hDicer PPC cassette binds G-quadruplexes no matter whether their 3′- or 5′ ends are available.
RNA G-quadruplexes bound by the PPC cassette retain their structure
To test whether RNA G-quadruplexes retain their structures upon binding the hDicer PPC cassette, we conducted a binding assay involving TER22 with a derivative of 3,6-bis(1-methyl-2-vinyl-pyridinium) carbazole diiodide (o-BMVC) covalently attached to its 5′-end (Supplementary Fig. S3D). o-BMVC is a fluorescent light-up probe that selectively binds to G-quadruplex structures . As a control, we used 32-nt RNA (called “LIN32”) that does not contain a G-quadruplex motif, labeled with o-BMVC in the same way as TER22. Each oligomer was incubated with PPC, and the reaction mixtures were analyzed in a polyacrylamide gel under native conditions. After electrophoresis, we exposed the gel first to 532 nm light to detect bands corresponding to the RNA species adopting G-quadruplex structures (Fig. 4a). Subsequently, we stained the gel with SYBR Gold solution and exposed it to 473 nm light to visualize the total RNA pool (Fig. 4b). The results showed that RNA G-quadruplexes retained their structure upon binding the PPC cassette.
RNA and DNA G-quadruplexes inhibit pre-miRNA processing by hDicer
In vitro, RNA oligonucleotides bound to hDicer inhibit the cleavage of pre-miRNAs by this enzyme [15, 16]. To investigate whether oligonucleotides containing G-quadruplex motifs can also influence the cleavage activity of hDicer, we used RNA oligomers representing human and O. nova TERRA (TER10, TER22, G4U4G4) and DNA oligomers representing the corresponding telomeric repeats (TEL22, G4T4G4). As well as those mentioned above, we used three RNAs of 12 to 20-nt: TER12, TER18, TER18-2A. All of the chosen oligomers are known to adopt various G-quadruplex architectures (Table S1). To assess the effect of RNA and DNA G-quadruplexes on hDicer cleavage activity, we performed a set of assays involving hDicer, 5′-32P-labeled pre-mir-21 or pre-mir-33a, and a respective oligomer. The particular pre-miRNAs were chosen as they did not interact with tested oligomers (Supplementary Fig. S4A), which ensures that the effects observed in the experiment does not stem from direct RNA–DNA or RNA–RNA binding, but rather, from protein–DNA or RNA interaction. Moreover, pre-mir-21 and pre-mir-33a represent structurally distinct substrates (Supplementary Fig. S5E).
First, we carried out cleavage assays with RNA G-quadruplexes. The efficiency of pre-miRNA (~5 nM) cleavage in reactions with individual oligomers applied at one of three concentrations (0.1, 0.5, or 2 µM) was normalized to the cleavage efficiency in a control reaction with no oligomer added. Another set of control reactions involved in 12-nt RNA (called “LIN12”), which does not bind to hDicer, nor adopts a G-quadruplex structure. In all cases, upon the addition of an RNA G-quadruplex, we observed a dose-dependent inhibition of the cleavage of pre-mir-21 (Fig. 5a and Supplementary Fig. S4B) and pre-mir-33a (Fig. 5b and Supplementary Fig. S4C). At the lowest concentration of any of the tested RNA G-quadruplexes, the level of miRNA was reduced by at least 50% in comparison to the control reaction with no oligomer added. When the highest concentration of any of the tested RNA G-quadruplexes was applied, the cleavage of pre-miRNA was abolished by at least 80%. No decrease in the miRNA level was observed in control reactions with LIN12. We did not find any apparent correlation between the length or architecture of the RNA G-quadruplexes and the degree of inhibition they exerted.
Next, we investigated whether the cleavage of pre-miRNA (~5 nM) by hDicer can be affected by DNA counterparts of G4U4G4 and TER22, i.e., G4T4G4 and TEL22 (Fig. 5c, d and Supplementary Fig. S4D, E). At the lowest concentration (0.1 µM), G4T4G4 inhibited cleavage of pre-mir-21 and pre-mir-33a by ~90%, while at the highest concentration (2 µM), the inhibition reached ~96% for pre-mir-21 and 100% for pre-mir-33a. We observed a similar tendency for TEL22; it inhibited the cleavage of pre-mir-21 and pre-mir-33a by ~70% at the lowest concentration (0.1 µM), and by ~95% at the highest concentration (2 µM). The results are comparable with those obtained for G4U4G4 (~90% and ~100% inhibition at 0.1 and 2 µM concentration, respectively) and TER22 (~75–80% and ~95% inhibition at 0.1 and 2 µM concentration, respectively), which indicates no differences in the inhibition potency between RNA G quadruplexes and their DNA G quadruplex counterparts under the applied reaction conditions (Fig. 5a–d and Supplementary Fig. S4B–E).
Subsequently, we performed a time course assay to measure the level of miR-21 or miR-33a produced in reaction with either no inhibitor, a selected G-quadruplex (G4T4G4, G4U4G4, TEL22, TER22) or pre-mir-16-1 added as a competitor. The reactions were performed under the low-turnover conditions, i.e., a twofold molar excess of hDicer (10 nM) over a substrate was used. In addition, the G-quadruplexes and the competitor were in 50-fold molar excess to hDicer. As expected, in control reactions with either pre-mir-21 or pre-mir-33a and hDicer, but no other oligomer added, we observed a hyperbolic relation between the yield of miRNA and the incubation time (Fig. 5e, f and Supplementary Fig. S5A–D). After 12 h, ~90% of either of the substrates was processed by the enzyme.
In reactions with pre-mir-21 as a substrate and pre-mir-16-1 as a competitor, we observed a significant inhibition of pre-mir-21 cleavage at all analyzed time points; even after 12 h only ~22% of pre-mir-21 was processed (Fig. 5e and Supplementary Fig. S5A, B). We also observed stable inhibition for the tested G-quadruplexes; the effect exerted by them was even more prominent than in the case of pre-mir-16-1: after 12 h only ~5% of pre-mir-21 was processed in reactions with G4T4G4 or G4U4G4, and ~10% in reactions with TEL22 or TER22 (Fig. 5e and Supplementary Fig. S5A, B).
In reactions with pre-mir-33a as a substrate, and pre-mir-16-1 as a competitor, initially the levels of miR-33a were much lower than in the control reactions without the inhibitor (e.g., ~ 10% vs ~ 50% after the first 30 min of the reaction) (Fig. 5f and Supplementary Fig. S5C, D) but with time the inhibition was gradually abolished; after 12 h the amount of pre-mir-33a processed in the reaction with the competitor and in the control reaction reached ~80% and ~95%, respectively. In reactions with the G-quadruplexes, we also observed low initial levels of miR-33a as in the analogous reactions with pre-mir-16-1 and the reduction of the inhibition with time (Fig. 5f and Supplementary Fig. S5C, D). The amount of pre-mir-33a cut by hDicer increased from ~5% after 30 min to ~10% after 12 h in reactions with G4T4G4 or G4U4G4, and from ~10% to ~25% in reactions with TEL22 or TER22. Despite the observed accumulation of miRNA over time, and in contrast to pre-mir-16-1, the inhibitory effect exerted by G-quadruplexes remained high even after 12 h of incubation (~75% decrease in miRNA production in comparison to the control reaction in the case of G-quadruplexes, vs ~15% in the case of pre-mir-16-1).
Altogether, these findings indicate that RNA and DNA adopting a G-quadruplex structure can affect the cleavage of pre-miRNA by hDicer. Under the low-turnover conditions (excess enzyme to substrate), RNA and DNA G-quadruplexes exerted a similar inhibition effect on the hDicer cleavage of both pre-miRNAs used (Fig. 5e, f). G4U4G4 and G4T4G4 were slightly better inhibitors of pre-mir-21 and pre-mir-33a cleavage than TER22 and TEL22 (95% vs 90% inhibition of pre-mir-21 cleavage after 12 h, and 90% vs 75% inhibition of pre-mir-33a cleavage after 12 h, respectively) (Fig. 5e, f, and Supplementary Fig. S5A–D).
Subsequently, we performed a time course assay under the high-turnover conditions using a 50-fold molar excess of a substrate to hDicer (0.5 nM), and a 100-fold molar excess of the G-quadruplexes or the competitor to the enzyme. We found that G4U4G4, G4T4G4 and TER22 retained their inhibitory potential after 12 h incubation (Supplementary Fig. S6), whereas the inhibition of either miR-21 or miR-33a production was overcome in reactions with TEL22 (Supplementary Fig. S6). Similar results as for TEL22 were obtained for pre-mir-16-1 competitor, therefore, we conclude that under the applied reaction conditions (1:2 molar ratio of a substrate and an inhibitor), TEL22 can act as a competitive inhibitor. This conclusion is supported by the Kd values calculated for the PPC and TEL22 complex (Kd ~ 318 nM), and the PPC and pre-miRNA complex (Kd ~ 205 nM). However, considering the binding affinities, we cannot explain why G4T4G4 was not outcompeted by the substrate over the incubation time. TEL22 represents the human telomeric sequence, while G4T4G4 corresponds to ciliate O. nova telomeric DNA. Given that the human enzyme was used in the studies, a species-specific regulatory mechanism for Dicer binding to telomeric DNA might be responsible for the effects we observed.
Under the low-turnover conditions, we did observe a difference in the effect of pre-mir-16-1 competitor on pre-mir-21 and pre-mir-33a cleavage by hDicer (Fig. 5e, f). Despite the addition of 100-fold molar excess of pre-mir-16-1 with respect to the other pre-miRNA, after 12 h incubation, ~80% of pre-mir-33a (Fig. 5f) and only ~22% of pre-mir-21 were processed (Fig. 5e). These results can be explained by the differences among the structures of the three pre-miRNAs. Pre-mir-21 adopts the most compact structure, with the smallest terminal loop, whereas structures of pre-mir-33a and pre-mir-16-1 contain large internal loops and bulges. In addition, pre-mir-16-1 has the most relaxed terminal loop region (Supplementary Fig. S5E). It has been reported that the pre-miRNA structure influences the efficiency of miRNA processing by Dicer . In addition, the results of our previous studies have indicated that pre-miRNAs can compete for binding to hDicer . Based on these data and the models generated for PPC and individual G-quadruplexes (Fig. 3a–d), we propose that a pre-miRNA and a G-quadruplex compete for binding to substrate-anchoring domains of hDicer, i.e., PAZ and Platform. Since the interactions between pre-miRNA and Dicer encompass not only the PPC region of Dicer, but other Dicer domains as well [4, 6] (Fig. 6a), we hypothesize that the competition between two pre-miRNAs for binding to Dicer is more complex, compared with the case involving a pre-miRNA and a G-quadruplex. Consequently, we deduce that under the low-turnover conditions, even despite the high excess of the inhibitor, pre-mir-33a can outcompete pre-mir-16-1 from binding to hDicer, which was not observed for pre-mir-21. However, under the high-turnover conditions, when the excess of a substrate to the enzyme was applied, and the substrate to inhibitor molar ratio was low (1:2), the degree of the miRNA production inhibition caused by pre-mir-16-1 was similar in the case of both substrates (Supplementary Fig. S6E, F).