If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
To whom correspondence should be addressed. Laboratory of Cancer Genetics, Institute of Bioorganic Chemistry, Polish Academy of Sciences, Noskowskiego 12/14 St., 61-704, Poznan, Poland. Tel.: 48-61-8528503; Fax: 48-61-8520532;
* This work was supported by the State Committee for Scientific Research, Grant 2PO5A08826, PBZ/KBN/040/P04/12, and the Foundation for Polish Science, Grant 8/2000. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. The on-line version of this article (available at http://www.jbc.org) contains supplemental Figs. 1–5.
We have established the structures of 10 human microRNA (miRNA) precursors using biochemical methods. Eight of these structures turned out to be different from those that were computer-predicted. The differences localized in the terminal loop region and at the opposite side of the precursor hairpin stem. We have analyzed the features of these structures from the perspectives of miRNA biogenesis and active strand selection. We demonstrated the different thermodynamic stability profiles for pre-miRNA hairpins harboring miRNAs at their 5′- and 3′-sides and discussed their functional implications. Our results showed that miRNA prediction based on predicted precursor structures may give ambiguous results, and the success rate is significantly higher for the experimentally determined structures. On the other hand, the differences between the predicted and experimentally determined structures did not affect the stability of termini produced through “conceptual dicing.” This result confirms the value of thermodynamic analysis based on mfold as a predictor of strand section by RNAi-induced silencing complex (RISC).
) of Caenorhabditis elegans that trigger the translational inhibition of their target mRNAs by partial base-pairing within the 3′-untranslated region. According to the results of recent surveys performed using the experimental (
). It appears that designing siRNAs so that their properties resemble those of putative double-stranded miRNA intermediates, produces highly effective siRNAs. The strand whose 5′-end is less tightly paired to its complement is selected to enter into the RNAi-induced silencing complex, whereas the opposite strand is degraded. Thus, the structure of miRNA precursors and their short lived processing intermediates turned out to be the key for successful siRNA design (
). Other than the specific thermodynamic and reactivity signatures of individual precursors, characteristic profiles have been revealed corresponding to the groups of precursors containing miRNAs either at the 5′- or 3′-side of their hairpin stem.
MATERIALS AND METHODS
Preparation of DNA Templates for in Vitro Transcription—DNA oligomers were obtained by chemical synthesis and purified by polyacrylamide gel electrophoresis. Each oligomer contains a DNA sequence complementary to the microRNA precursor sequence and to the sequence of T7 RNA polymerase promoter at the 3′-end (see Table I). The double-stranded templates for in vitro transcription were prepared using the primer extension procedure (200 pmol of template oligomer and 1 mmol of T7GG primer 5′-taatacgactcactatagg, 200 μm each dNTPs, standard PCR buffer, and 0.5 units of TaqDNA polymerase/100 μl of reaction mixture) in two-step PCR, 50 cycles at 94 °C for 15 s and 45 °C for 15 s, and were purified using centrifugal filter devices Microcon YM30 (Millipore).
Table IOligodeoxynucleotides used for transcript synthesis
Residues added to the DNA templates to enable efficient in vitro transcription. These extra nucleotides when transcribed constitute the only differences between the proposed sequences of natural pre-miRNAs (3) and these investigated in this study
bResidues added to the DNA templates to enable efficient in vitro transcription. These extra nucleotides when transcribed constitute the only differences between the proposed sequences of natural pre-miRNAs (
Transcription in Vitro—The pre-microRNAs used in this study were prepared by in vitro transcription with T7 RNA polymerase. The transcription reaction carried out in a 50-μl volume contained 20 pmol of DNA template, 50 μm rNTPs, 3.3 mm guanosine, 40 units of ribonuclease inhibitor RNase Out (Invitrogen), and 400 units of T7 RNA polymerase (Ambion). The guanosine was added to the reaction mixture to place it at the 5′-end in a high proportion of transcripts, thus making the 5′-end dephosphorylation step unnecessary. The transcription reaction was carried out at 37 °C for 1 h, and transcripts were purified in denaturing 10% polyacrylamide gel, excised, eluted from the gel (0.3 m sodium acetate, pH 5.2, 0.5 mm EDTA, and 0.1% SDS), and precipitated. All transcripts were 5′-end-labeled with T4 polynucleotide kinase and [γ-32P]ATP (4500 Ci/mmol; ICN). The labeled RNAs were gel-purified and stored at -70 °C in water before use.
Nuclease Digestions and Metal Ions Induced Cleavages of RNA—Prior to structure probing, the 32P-labeled transcripts were subjected to a denaturation and renaturation procedure, in a solution containing 12 mm Tris-HCl, pH 7.2, 48 mm NaCl, and 1.2 mm MgCl2 by heating the sample at 90 °C for 1 min followed by a slow cooling to 37 °C. Limited digestion of RNA was performed at 37 °C in a solution containing 10 mm Tris-HCl, pH 7.2, 40 mm NaCl, and 1 mm MgCl2 (0.5 mm ZnCl2 was also present in reactions with nuclease S1) obtained by mixing 8 μl of the RNA solution described above (∼5 pmol of RNA) with 2 μl of a probe at different concentrations. Final concentrations of probes in the reactions were as follows: S1 nuclease, 0.3, 0.6, and 1.2 units/μl; T1 ribonuclease, 0.1, 0.2, and 0.3 units/μl; T2 ribonuclease, 0.03, 0.04, and 0.05 units/μl; V1 ribonuclease, 0.03, 0.04, and 0.05 units/μl; Pb2+ ions, 0.1, 0.2, and 0.4 mm; Mn2+ ions, 1.5, 5, and 7.5 mm; Mg2+ ions, 1.5, 5, and 7.5 mm; Ca2+ ions, 1.5, 5, and 7.5 mm. Reactions with Mn2+, Mg2+, and Ca2+ ions were performed at 37 °C in a solution containing 10 mm Tris-HCl, pH 8.5, or 9.0 and 40 mm NaCl. The reactions with nucleases and lead ions were stopped after 10 min, reactions with Mn2+ after 2 h, and reactions with Mg2+ and Ca2+ after 16 h by adding an equal volume of a stop solution containing 7.5 m urea, 20 mm EDTA, and dyes. The alkaline hydrolysis ladder was generated by the incubation of the labeled RNA in formamide containing 0.5 mm MgCl2 at 100 °C for 10 min. The partial digestion of RNAs (∼5 pmol) with T1 ribonuclease was performed under semidenaturing conditions (10 mm sodium citrate, pH 5.0, 3.5 m urea) for 12 min at 55 °C with 0.2 units/μl of the enzyme. Electrophoresis was in 15% polyacrylamide gel under denaturing conditions at 1500 V and was followed by autoradiography at -80 °C with an intensifying screen. The products of the structure-probing reactions were also visualized and analyzed by phosphorimaging (Typhoon, Molecular Dynamics). To minimize the possible contributions from secondary cleavages, the cuts generated in the transcripts that underwent the reaction in less than 10% were considered in the structure analysis.
Electrophoresis in Nondenaturing Conditions—The homogeneity of the RNA structure was analyzed for all investigated transcripts by the electrophoresis of radiolabeled samples in 10% nondenaturing polyacrylamide gel (dimensions 150 × 140 × 1 mm) (acrylamide/bisacrylamide, 29/1) buffered with 10 mm Tris-HCl, pH 7.2, and containing 40 mm NaCl and 1 mm MgCl2 at a controlled temperature of 37 °C. Prior to gel electrophoresis, the 32P-labeled transcripts (∼5 pmol each) were subjected to a denaturation and renaturation procedure as described in the preceding section but in a solution containing 10 mm Tris-HCl, pH 7.2, 40 mm NaCl and 1 mm MgCl2 and mixed with an equal volume of the same solution containing 7% sucrose and dyes. Electrophoresis was performed at 100 V with buffer circulation at 2 liter/h.
RNA Secondary Structure Prediction—RNA secondary structure prediction was performed using the mfold program version 3.1 (
). This program is designed to determine the optimal and suboptimal secondary structures of RNA and to count free energy contributions for various secondary structure motifs.
RNA Thermodynamic Profiling—The internal hairpin stability values were calculated according to the nearest neighbor method using thermodynamic parameters determined at 37 °C, for all stacking free energy values (expressed in kcal/mole), taking into consideration all the different destabilizing elements such as internal loops and bulges, using the 3.1 version of the mfold program (
). Two different kinds of RNAs, pre-miRNAs and duplex intermediates, were analyzed using procedures that differed in some aspects as follows. The analysis of precursor structures was performed using trinucleotide subsequences to define their central position. To adjust the stability profiles to the corresponding secondary structures the ΔG values characteristic for the base pair closing the single-stranded region were assigned also for other structural elements of this region (supplemental Fig. 5A). On the other hand, in the analysis of duplex intermediates, instead of using the scanning window, the thermodynamic parameters were calculated for each interacting base pair and each structure-destabilizing element (supplemental Fig. 5B).
Selected miRNA Precursors and Probes Used for Structure Analysis—We intended to experimentally analyze the secondary structure of representative human pre-miRNAs. We selected the following 10 pre-miRNAs of which five, pre-let-7c, pre-let-7f-2, pre-miR-15a, pre-miR-16-1, and pre-miR-18, have the mature miRNA at the 5′-side, four, pre-miR-19a, pre-miR-25, pre-miR-29a, and pre-miR-30a, have the mature miRNA on the 3′-side, and one, pre-miR-17, on both sides. When these RNAs were subjected to structure prediction using the mfold program (
) the resulting lowest energy precursor hairpins differed in their terminal loop size as well as in the number, size, and location of various internal loops and/or bulges. For half of the analyzed precursors two or more alternative structures were proposed within the 10% range of the suboptimality parameter.
For the experimental analysis, we chemically synthesized DNA templates (Table I) that allowed the in vitro transcription of pre-miRNAs. The extra nucleotides added to the precursor sequences to allow their efficient in vitro transcription did not change their mfold predicted structures. The transcripts were 5′-end-labeled and gel-purified. Their structural homogeneity was confirmed by electrophoresis in nondenaturing polyacrylamide gel (supplemental Fig. 1A). Next, we subjected the RNAs to structure probing as shown in supplemental Fig. 1, B and C for the pre-let-7f-2 and pre-miR-29a. The chemical probes used to analyze each transcript were Mg2+ (
). The metal ions differentiate between rigid and flexible regions of the RNA structure. The former are resistant and latter susceptible to cleavage in agreement with the mechanism proposed for phosphodiester bond cleavage by lead (
). In brief, the reaction begins with the activation of the ribose 2′-OH group by the metal ion hydroxide, and the attack of the 2′-O-nucleophile on the adjacent phosphorus atom, which requires conformational flexibility of the sugar phosphate backbone.
The Majority of Experimentally Determined Structures of miRNA Precursors Differ from Those Predicted—When we subjected the labeled transcripts to single-strand-specific probes the centrally located nucleotides of all miRNA precursor sequences were found to be the most reactive regions. This suggested hairpin structures in which the regions of enhanced reactivity corresponded to their terminal loops. The Mg2+ and Ca2+ ions turned out to be the probes distinguishing most precisely between paired and unpaired nucleotides in the stem regions of the precursor hairpins (Fig. 1). The reactive phosphodiester bonds mapped by these two probes often overlap each other, and the Ca2+ ions cut the great majority of internucleotide bonds present in numerous bulges and internal loops. For example, among eight stem-structure-distorting motifs proposed to exist in pre-miR-18, as many as seven are recognized by the Ca2+ ions and five by Mg2+ ions. The only motif undetected by Ca2+ ions in this structure is the single nucleotide A-bulge. This pattern of reactivity is in line with the flexible geometry and variable coordination number of the Ca2+ aquacation in solution (
The sizes of terminal loops of the precursor hairpins are well mapped by the single-strand-specific nucleases (Fig. 2 and supplemental Fig. 2). On the other hand, the small internal loops, bulges, and mismatches are poorly mapped by the enzymatic probes. For example, neither the symmetrical 6-nucleotide nor asymmetrical 3-nucleotide internal loops present in pre-miR29a were cleaved by the nucleases. Among the detected motifs are bulges mapped by the T2 ribonuclease in pre-let-7c, miR-15a, miR-16-1, miR-19a, and miR-30a precursors. Their detection is in agreement with the earlier observations that nucleases recognize well larger single-stranded regions. The failure of nucleases, and in some cases also metal ions, to detect small symmetrical internal loops and mismatches when they are surrounded by stable double-helical regions may indicate that these motifs form noncanonical base pairs (
), which do not distort the duplex structure significantly. The advantageous feature of metal ions in RNA structure probing is that their hydrates are much smaller than nucleases, penetrate folded RNAs easier, and reveal more details of the analyzed structures (
Altogether, aside from hairpin terminal loops, 47 other structure motifs such as internal loops, bulges, and mismatches other than G-U wobble pairs are proposed to occur in the experimentally determined secondary structures of 10 analyzed precursors (Figs. 1 and 2, supplemental Figs. 2 and 3). The 12 of the 13 internal loops are mapped by the Ca2+ ions at both strands (Fig. 1). The Mg2+ ions failed to detect a single internal loop, and the presence of two of such loops was revealed by cleavages generated in one strand only. All bulges larger than a single-nucleotide were mapped by both types of metal ions and two of the eight single-nucleotide bulges escaped detection by both ions. Of the 14 mismatches present in the proposed precursor structures, as many as 13 were detected by the Ca2+ ions and 10 by Mg2+ (Fig. 1). In sum, 44 of the 47 distinct structure motifs were detected by both metal ions. This compares favorably with only 12 of the 47 such motifs detected by nucleases.
Importantly, in 8 of the 10 investigated transcripts the experimentally determined structures of miRNA precursors turned out to be different from those predicted to be the most favorable. In seven transcripts (pre-let-7c, pre-let-7f-2, pre-miR-15a, pre-miR-16-1, pre-miR-18, pre-miR-25, and pre-miR-29a) the hairpin terminal loop regions were different (Fig. 2 and supplemental Fig. 2). The differences included the size and localization of terminal loops and accompanying changes in the neighboring stem regions. For example, in the pre-miR-18 the nucleotides G30-C54 form a 5-nucleotide terminal loop, a 4-nucleotide symmetric internal loop, and a 3-nucleotide bulge at the 3′-side of the stem. In the predicted structure there is a 6-nucleotide terminal loop, a 6-nucleotide asymmetric internal loop, and a 1-nucleotide bulge at the 5′-side of the stem. It is also worth noting, that in the case of eight precursors none of the experimentally determined structures were computer-predicted within the 10% range of the suboptimality parameter. Thus, only the structures of pre-miR-29a and pre-miR-30a predicted by mfold to be the most stable were confirmed by the experiment.
Duplex Portions of MicroRNA Precursors Show Polarity in Their Reactivity and Thermodynamic Stability—Having established the secondary structures of the investigated precursors we intended to gain an insight into the features of their helical portions. We applied for this purpose the ribonuclease V1, which cleavages were expected to be sensitive to distortions in helical conformation. We also used the Pb2+ and Mn2+ ions, which are more reactive than Ca2+ and Mg2+ and in addition the single-stranded regions should also cleave the flexible portions of duplexes. Finally, we used the thermodynamic profiling of the precursor structures to show how the structure probing and structure stability data corresponded to each other.
The ribonuclease V1 cleavages occurred only at the 5′-side of the hairpin stem in 4 of 10 analyzed precursors, pre-let-7c, pre-let-7f-2, pre-miR-29a, and pre-miR-30a. In the remaining six precursors, several much weaker 3′-side cleavages also occurred (not shown). When the sites of V1 cuts were compared with the experimentally determined secondary structures of the precursors, the highly variable patterns emerged (Fig. 3 and supplemental Fig. 4). The number of reactive bonds was generally higher than that of the resistant ones, and several very strong cuts were typically observed in each transcript. Both duplexes composed of standard W-C base pairs and G-U wobble pairs only (e.g. in pre-miR-17 and pre-mir-15a) as well as duplexes containing some internal mismatches (e.g. in pre-let-7c and pre-miR-25) were digested. This may suggest that the distortions of helical structures in the latter are rather modest. The strongest V1 cuts frequently occur in investigated transcripts at the sites of increased sequence regularity such as mono-(GGG, UUU, UUUU, AAA), di-, or trinucleotide repeats (UGUG, UAUA, CACA, CUCU, AAUAAU), which are likely to be also the sites of more regular structure. The V1-resistant regions, other than the 4–6 internucleotide bonds from the 5′ terminus, typically occur in the neighborhood of duplex structure distortions such as some bulges, internal loops, and the U-C mismatches. With regard to the barely known sequence specificity of ribonuclease V1 it is interesting to note that the enzyme prefers to cleave the phosphodiester bonds at weak A-U, U-A, and wobble U-G pairs. Of the 12 strong cleavages taking place after the A residue, the GC or GU sequences are always located 5′ to the cleavage site.
Next, we used the Pb2+ and Mn2+ ions to reveal in which parts of the pre-miRNA stem structures the phosphodiester bonds are more flexible than in others. We wanted to find out how the distribution of reactive regions in duplexes corresponds to the location of miRNA ends when they are still embedded in the precursor hairpin structure? As expected, the Pb2+ and Mn2+ ions mapped also the phosphodiester bonds of several paired nucleotides located in the neighborhood of bulges or internal loops (Fig. 3 and supplemental Fig. 4). In several precursors the region of the miRNA 5′-end is more susceptible to cleavages than that of their 3′-end. For example, in the pre-miR-19a the nucleotides U50-G51 of the miRNA 5′-end are localized in the 3-nt bulge, but a longer stretch of nucleotides U49-C54 shows an increased susceptibility to cleavage. On the other hand, the region of the miR-19a 3′-end is poorly cleaved by the Mn2+ and Pb2+ ions. What is noteworthy is that the regions of enhanced reactivity occur also in the central portions of sequences corresponding to the mature miRNAs. Although these regions correlate with the localization of RNA structure-destabilizing base mismatches, bulges, and internal loops in the miRNA precursors, the Mn2+ and Pb2+ cuts usually span more internucleotide bonds of increased flexibility, as the C24-U29 region of pre-miR-15a (supplemental Fig. 4). Moreover, in the pre-let-7f-2, in which the region of mature miRNA is completely paired except for its single 5′-nucleotide, the reactivity of centrally located miRNA sequences is also enhanced (Fig. 3).
For each of the 10 experimentally determined miRNA precursor structures (Fig. 2A) the stability profiles were generated by calculating the ΔG values for each base pair or unpaired base, surrounded by its neighbors, 1 nucleotide from each side, along the entire pre-miRNA hairpin stem. It appears (Fig. 3 and supplemental Fig. 4) that the average stem stability for each single stacking parameter differs considerably between the precursors and is the highest for the pre-miR-30 (-4.2 kcal/mole) and the lowest for pre-miR-18 (-2.6 kcal/mole). Interestingly, the miRNA ends are located in 16 of 20 cases in the region of stability lower than the average, and with the exception of the 3′-ends of miR-19a and miR-29a as well as the 5′-ends of miR-29a and miR-30a, their first nucleotide or first two nucleotides are located either in the mismatches, bulges, or internal loops. There is a good correlation between the shapes of thermodynamic profiles and patterns of the Pb2+ and Mn2+ ion-promoted cleavages (Fig. 3 and supplemental Fig. 4). Nearly all thermodynamically most stable regions are poorly cleaved by these ions and vice versa. For example, in the central region of the long 17-bp-containing pre-miR-19a helix, four consecutive A-U base pairs both are cleaved by Pb2+ ions and cause an increase in the ΔG value. On the other hand, the tandem G15-U76 and U16-G75 wobble pairs in pre-let-7f, which do not decrease duplex stability significantly, are well mapped by the Mn2+ and Pb2+ ions. It is also noteworthy that in 28 of 30 of cases the strongest V1 cuts occur in regions that are thermodynamically more stable than the average.
Thermodynamic Features of miRNA Precursors Relevant to miRNA Biogenesis and Active Strand Selection—In the next step of our analysis we intended to gain a deeper insight into the features of the precursor structures potentially important for miRNA biogenesis. We wanted to do that by comparing thermodynamic profiles obtained for precursors containing miRNAs at their 5′- and 3′-sides. We also wanted to compare our data with the results of a recent study (
) in which the averaged thermodynamic profiles of many precursors of both types taken together were discussed in relation to miRNA strand selection by the RNAi-induced silencing complex. For that, we generated the averaged profiles of the precursors having experimentally determined structures, separately for those containing miRNAs at the 5′- and 3′-sides (Fig. 4).
It appears from this analysis, that there are both similarities and differences between the profiles of precursors containing the miRNAs at opposite sides. Considering the similarities first, we may notice that in both types of precursors the 5′-end of miRNA is less stable than that of miRNA* (Fig. 4, yellow bars) and that central nucleotides 10–13 reside in a region of lower stability. Likewise, the similar stabilities of 3′-ends of the miRNA/miRNA* duplexes still embedded in the precursor structure may also be noticed (Fig. 4, compare red bars between the upper and lower panels). The presence of two regions of higher stability at both sides of the nucleotides 10–13 is also apparent. In one of these regions, the nucleotides 4–8 counting from the 5′-ends of miRNA (Fig. 4, upper panel) and miRNA* (lower panel) belong to the most stable. The stability of this region is also least variant. These trends in structure stability distribution corroborate those observed earlier for the combined profiles of both types of precursors (
Two important differences may also be noticed in going from the base of the precursor hairpin stem to its terminal loop. First, is the region of lower stability at nucleotide positions from -2 to +2, which is considerably less stable in precursors having 5′-miRNAs than in those harboring 3′-miRNAs (Fig. 4). This is the site of the weakest structure along the miRNA sequence in the precursors containing miRNAs at their 5′-side. The relatively high structure stability at that site may give rise to the 3′-miRNAs, that have their 5′-ends located closer to the hairpin terminal loop and vice versa. The second difference is the size of the stable region that is closer to the base of the hairpin stem. It is narrower and comprises nucleotides 4–8 in precursors containing 5′-miRNAs and wider (nucleotides 1–8) in precursors having miRNAs in their 3′-portion.
In the final step of our study we performed a thermodynamic profiling of the miRNA/miRNA* duplexes, which are putative products of Dicer and substrates for the protein components of the RNAi-induced silencing complex (
). We have focused on these structures as the conceptual dicing of miRNA precursors and the determined stability differences between the 5′-end regions of miRNAs and miRNAs* in their duplexes were shown to be instructive for the rational siRNA design (
). We have tested the performance of this thermodynamic approach on miRNA/miRNA* duplexes conceptually diced from the precursors analyzed in detail in this study as well as on other 30 mfold-predicted precursor structures (Fig. 5A). It turned out from the stability analysis of terminal 5 nucleotides in 10 precursors, that in 7 of 10 cases the 5′-end of miRNA is less stable than that of miRNA*. The corresponding scores were 6 of 10, 8 of 10, and 9 of 10 when terminal 4, 3, and 2 nucleotides were taken into consideration. Thus, the analysis of terminal 2 nucleotides has the highest discriminatory power in distinguishing between the miRNAs and miRNAs*. In 8 of the 10 analyzed duplexes the ΔG value calculated for terminal 2 nucleotides is at least two times higher for the 5′-end of miRNA (Fig. 5A). The exceptions are miR-30a, for which miRNA* was also detected in cells (
), and as expected the miR-17. The widening of the analyzed terminal sequence window gives worse results for the majority of duplex intermediates (Fig. 5, marked by stars). The same trend is also observed for both the individual (not shown) and averaged values obtained for 30 other duplexes diced from human miRNA precursors (Fig. 5A). There are also sequence determinants and antideterminants at the ends of miRNA sequences. The nucleotide sequence analysis of 140 human miRNAs shows a high occupancy of U (63%) at position 1 of miRNA resembling that observed in C. elegans (
) and reveals a high discrimination against U (6%) at the 5′-nucleotide of miRNA* (Fig. 5B). These results may have considerable implications for a rational design of the effective siRNAs and shRNAs.
Precursor Structure Prediction, Experimental Probing, and Thermodynamic Profiling—Numerous miRNAs regulate gene expression in a wide variety of eukaryotic organisms, but the way in which this process is orchestrated in cells is poorly understood. The structure of miRNA precursors is among the factors that may influence the rate at which specific miRNAs emerge in cells. In this study we have used a battery of chemical and enzymatic probes to gain an insight into the structural features of human miRNA precursors. The experimental approach to the structure analysis turned out to be fully justified as 8 of the 10 structures differed from those predicted to be the most stable. The differences localized in the terminal loop region in seven precursors and close to the base of the hairpin stem in three precursors. Although the percentage of precursors with different predicted and determined structures seems high, this is so because no single differences were tolerated in the comparisons of irregular hairpin structures of these transcripts ranging in length from 68 to 91 nt. With regard to the percentage of correctly predicted base pairs in the lowest free energy structures, which ranges from 71.4% in pre-mir-16-1 to 100% in three precursors (average for 10 precursors is 88.4%), these results compare favorably with those reported earlier for other RNAs. For example, for numerous tested tRNAs, 5 S rRNAs, and signal recognition particle RNAs the accuracy of secondary structure prediction by mfold was shown to be 83, 77.7, and 73%, respectively (
The results of direct structure probing experiments with the use of Pb2+ and Mn2+ ions and V1 nuclease were also compared with those of in silico thermodynamic profiling of the miRNA precursors and a good correlation was found. Both methods reveal structure instability elements along the pre-miRNA hairpin stem. The regions of decreased thermodynamic stability usually overlap with those of enhanced reactivity with metal ions (Fig. 3 and supplemental Fig. 4). Thus, our experimental results showing the same trends in structure instability distribution supported and validated the results of thermodynamic profiling. To see whether this conclusion may have a more general significance, we performed thermodynamic profiling for several other types of RNAs having well characterized structures, for which the relevant chemical and biochemical structure probing data were available. It turned out that a similar good correlation was obtained for all of the following RNAs: trans-activation response element HIV1 (
The important question that arises from our study is what kind of changes in the thermodynamic stability profiles of miRNA precursors are produced by the differences between the predicted and experimentally determined structures? The relevant question is what is the significance of such changes? And do these structural differences influence also the outcome of conceptual dicing and predicted stabilities of miRNA/miRNA* ends? To answer the first question the thermodynamic profiles of the predicted and experimentally determined structures were compared (Fig. 6). In the case of pre-let-7c, the structural difference, which is located close to the terminal loop, does not influence the stability of the 3′-end of let-7c and the 5′-end of let-7c*. On the contrary, in the case of the pre-miR-29a and pre-miR-16-1 having miRNAs at their 3′- and 5′-sides, respectively, the structural differences do influence the stability of the 5′-ends of their miRNAs. The local stability difference is large, 4 kcal/mole for miR-16-1 and 6 kcal/mole for miR-29a. In the predicted structures of these precursors the 5′-ends of miRNAs and miRNAs* have similar thermodynamic stability (Fig. 6). This example shows that miRNA prediction based on thermodynamic profiling solely may give ambiguous results for some predicted precursor structures. The success rate is higher for the experimentally determined structures as in 9 of the 10 cases the 5′-ends of miRNAs show lower stabilities than the 5′-ends of miRNAs* (pre-miR-18 is the exception). On the other hand, the structural differences between the predicted and experimentally determined structures did not influence considerably the stability of the end regions of the miRNA/miRNA* duplexes obtained by conceptual dicing.
Relevance of Precursor Structures to miRNA Biogenesis—The picture that emerges from our structure analysis and thermodynamic profiling is that the miRNA precursor hairpin may be considered a mosaic of more and less stable regions that occur at certain sequence intervals and have both a structural and functional meaning. The precursor processing site, which is located closer to the base of its hairpin stem, seems to have both structure and sequence determinants. The analysis of thermodynamic profiles of two groups of the microRNA precursors harboring miRNAs at either the 5′- or 3′-side revealed the conservation of two high stability regions (positions 4–8 and 15, 16). This may suggest that these regions are involved in determining either the binding or cleavage specificity of the processing enzymes. Notably, within this region the nucleotide at position 6 is the most conserved in both its stability and sequence. The analysis of 140 miRNA precursor sequences showed that this position is occupied by U in 50% and by pyrimidine in 70% of cases. Furthermore, the region between positions 3 and 7 almost invariably forms an undistorted duplex, and its irregularities are usually the G-G, U-U, G-A, and A-G mismatches, that may destabilize the duplex structure only moderately. The position -1 is occupied by the C/G in 70%, and position 1 by A/U in 72% of cases (A predominates in precursors having miRNA at their 3′-side, and U in those with miRNA at the 5′-side). At the cleavage site located near the hairpin terminal loop such clear sequence preferences are not observed. Only is C selected against (7%) at position 20, and G and U predominate at positions 19–22 contributing 64–74% to all nucleotides occupying these positions.
In conclusion, we have shown the likely structures of selected miRNA precursors, which properties fit well their thermodynamic profiles generated by the improved method. These results provide a structural basis for further studies of the precursor processing by Drosha and Dicer and for successful siRNA/shRNA design.