Dynamic Protein-RNA Recognition in Primary MicroRNA Processing

Abstract

MicroRNAs are prevalent regulators of gene expression, controlling most of the proteome in multicellular organisms. To generate the functional small RNAs, precise processing steps are required. In animals, microRNA biogenesis is initiated by Microprocessor that minimally consists of the Drosha enzyme and its partner, DGCR8. This first step is critical for selecting primary microRNAs, and many RNA-binding proteins and regulatory pathways target both the accuracy and efficiency of microRNA maturation. Structures of Drosha and DGCR8 in complex with primary microRNAs elucidate how RNA structural features rather than sequence provide the framework for substrate recognition. Comparing multiple states of Microprocessor and the closely related Dicer homologs shed light on the dynamic protein-RNA complex assembly and disassembly required to recognize RNAs with diverse sequences via common structural features.

Introduction: MicroRNAs need precise maturation steps

MicroRNAs are small (~22nt) RNAs that regulate most protein-coding genes [1]. The human genome encodes at least hundreds, perhaps thousands, of microRNAs. MicroRNAs influence almost all biological processes, from individual cell fate to organ development. In multicellular organisms, microRNAs are especially important to support the specialization of distinct cell types [2]. The vast majority of microRNAs require specific processing steps to become the mature fragment capable of modulating gene expression (Figure 1). The 5’ region—second through seventh nucleotides—of a mature microRNA constitutes the minimum identifying sequence (“seed”) that largely determines the targeted mRNAs. Even a single-nucleotide variation at the 5’ end of a microRNA (isomiR) can change the target mRNA [3].

Figure 1:

Schematic representation of microRNA endonucleolytic processing steps. For the pri-miRNA, sequences motifs (CNNC, GHG, UG and UGU) and flipped nucleotides are shown enclosed in grey boxes. Cut sites for Drosha and Dicer are indicated with blue and green arrows respectively. Relative nucleotide position from Drosha (blue or orange) and Dicer (purple) cleavage sites is indicated with numbers.

Primary transcripts (pri-miRNAs) are first cropped by the Microprocessor complex that minimally consists of an RNase III—Drosha—and the obligate partner protein, DGCR8 [4–6]. Microprocessor typically generates a signature 3’ dinucleotide overhang on the precursor microRNA (pre-miRNA) to provide a landmark for the next processing enzyme; Dicer anchors at the Drosha cut site and measures the proper distance to locate where to produce another double-stranded RNA (dsRNA) break near the apical end. Thus, regardless of which strand is eventually chosen to become the mature microRNA, Drosha processing needs to be precise. Furthermore, Drosha processing efficiency has a major impact on mature microRNA levels, highlighting how processing steps are used to regulate microRNAs [7,8].

To initiate the microRNA processing cascade properly, the Drosha-DGCR8 complex faces the challenge of identifying the exact cut site in hundreds to thousands of pri-miRNAs. Despite the primary sequence diversity, most pri-miRNAs contain a large stem-loop structure (Figure 1). Due to the nature of stem loops, there are two ssRNA-dsRNA junctions, at the apical and basal ends. Multiple short sequence motifs have been discovered, including basal UG, apical UGU, 3’ CNNC, and stem GHG [9,10]. Although most microRNAs do not contain all of them, the short motifs are thought to contribute additively to support recognition by Microprocessor. Nonetheless, the lack of a definitive signature motif poses a challenge in predicting the suitability of a stem-loop as a substrate for Drosha-DGCR8. Most likely, the tertiary structures of pri-miRNAs provide the critical framework for Drosha-DGCR8 to recognize them and cleave accurately.

Structures Reveal Microprocessor in Action

Since Drosha and DGCR8 were discovered ~20 years ago, structural studies have provided valuable insights to understanding the mechanisms underlying the first cleavage. A crystal structure of the Drosha core domains helped determine its orientation on a pri-miRNA, including how the DGCR8 C-terminal peptides stabilize folded Drosha RNase III domains (RIIIDs) [11,12]. Each DGCR8 polypeptide contains a heme-binding region (HBR) and two dsRNA binding domains (dsRBDs), and partial structures are also available at high resolution [13,14]. More recently, cryogenic electron microscopy (cryo-EM) structures helped reveal how Drosha and DGCR8 coordinate to recognize pri-miRNAs. Remarkably, RNA structural features are recognized by many dedicated protein modules, and multiple states have been captured using cryo-EM [15,16]. Comparing the structures of two different pri-miRNAs reveals how common structural features are recognized by Microprocessor, providing the groundwork to investigate more microRNA variabilities in the future. Overall, the protein-RNA interactions can be divided into 3 sections: I) Drosha interactions with the basal ssRNA-dsRNA junction, II) Drosha-DGCR8 contacts with the stem, and III) DGCR8 recognition of the apical junction (Figure 2).

Figure 2:

Cryo-EM structures of Drosha-DGCR8 complexes. (a) Domain organization of human Drosha and DGCR8. (b) Ribbon representation of the Microprocessor structures at different stages of binding a pri-miRNA. Unmodelled cryo-EM density map corresponding to the HBR is shown in grey. Domain colors match the diagram in (a). (c) Ribbon representation of the 4-way intersection where the helical Belt, Wedge, dsRBD, and C-terminus of Drosha become organized around the basal junction RNA. (d) Superposition of pri-miRNA-16-2 (orange) and pri-miRNA-16-1 (grey) basal junctions from two Microprocessor-RNA complex structures with proper basal junction engagement. Three nucleotides corresponding to the CNNC motif are depicted in yellow. (e) Top view of the apical junction-DGCR8 interactions showing different interdomain orientations in the two copies of DGCR8.

I. Basal Junction - Drosha Interactions

In both structures of Drosha without bound RNA (apo), a significant portion (~22-26%) is disordered [12,16]; binding pri-miRNA rigidifies multiple RNA-detection modules of Drosha (Figure 2b). As Drosha folds around the RNA, four polypeptide regions are brought together to create a 4-way intersection near the basal junction: helical belt (hairpin), helical wedge, dsRBD, and the C-terminus (Figure 2c). The Drosha dsRBD uses an extra-long loop [17] to contact the ssRNA-binding modules, to ensure that there is a dsRNA-ssRNA junction. The helical belt—termed Belt due to its ability to cross over the basal junction only after the RNA is seated in its binding groove—is connected to a fold reminiscent of PAZ domains [18] but the module is structurally distinct in Drosha. More importantly, Belt works with the helical Wedge—also called mobile helices—on the opposite side of the ssRNA to ensure that the RNA strands are unpaired. Wedge also packs into the groove formed as the RNA stem unwinds. Consequently, coordinating multiple protein modules enables Drosha to recognize a complex RNA structure, similar to multiple fingers that wrap around a grip to recognize its pattern. Orchestration of the different binding modules is superimposable in both cryo-EM structures with fully engaged basal junctions, although different sequence motifs are present in the two pri-miRNAs [15,16].

Despite the limited resolution of the RNA-bound cryo-EM structures of Microprocessor, there is little evidence for extensive nucleobase contacts to read the RNA sequence, which would explain why only short motifs have been identified to affect processing modestly and only in certain contexts. The GHG motif was shown to affect the efficiency and accuracy of Drosha [10]. Unlike some bacterial RNase IIIs that can sense sequence information through a dsRBD [19], the nucleobase specificity of Drosha is not as obvious. More recent biochemical interrogation of the GHG motif showed that the sequence requirement may be less strictly defined [20]. The middle mismatch, however, seems to be sensed indirectly through the changed backbone structure (Figure 2c). Wedge detects the mismatch through the protruding phosphate backbone, which might explain why the location of a bulge can regulate Microprocessor activity [21]. Indeed, treating GHG as a structural motif has yielded algorithms that can more reliably predict processing efficiency and precision [22]. Thus, Wedge helps to correctly register Drosha at the basal junction by detecting the mismatch at the 4th nucleotide away from the 3’ Drosha cut site. More studies are needed to determine variable contributions of Belt and Wedge for recognizing more diverse pri-miRNAs, and the recent structural insights will help guide such investigations.

A clearer demarcation of the dsRNA-ssRNA border would help Drosha to locate where the basal junction starts more readily. The preference for ssRNA in flanking arms has been shown biochemically [23,24]. However, most pri-miRNA arms extend significantly beyond the footprint of Drosha, and the structure of pri-miR-16-1 provides an example of how the flanking regions tend to anneal together [16]. To help maintain ssRNA for a stretch of (>4) nucleotides, Drosha provides pockets to stabilize an unpaired base on each strand, to compensate for the lost ring-stacking interactions. In the available cryo-EM structures, the 14th nucleotide away from the cut site on each strand is flipped out and buried in protein pockets—formed at the Belt/Wedge interface— ~30 Å apart from each other (Figure 1, 2d). Even though the 5’ pocket binds U of the UG motif, it is likely flexible enough to accommodate other bases, similar to the 3’ pocket that can accommodate A or U in the two available structures. This unwinding mechanism is distinct from the previously proposed model that depends on a “Bump helix” to act as an obstacle between the two strands [12], because the Bump helix is located peripherally to the fork of the RNA. Rather, a nucleotide on each strand is pinched out and pinned down by Belt and Wedge of Drosha to keep them apart.

Although Drosha can support unwinding, a pri-miRNA with highly structured flanking arms would pose a challenge to recognize as ssRNA near the basal junction. CNNC is a more common motif on the 3’ arm, and RNA-binding factors such as DDX17 and SRSF3 interact with ssRNA containing CNNC to enhance processing, likely by helping to stabilize the unpaired RNA structure close to the Belt-Wedge clasp [9,25–27]. The CNNC motif (~18nt from 3’ cut site) is located adjacent to where the Wedge footprint ends (~16nt), and ssRNA-binding proteins may keep the RNA strands separated to promote the basal junction conformation needed to bind Drosha productively. Therefore, RNA elements beyond the minimally required for recognition by Drosha and DGCR8 can modulate processing, and many factors may act through similar regions depending on their specificity [28,29].

II). Stem interactions with Drosha and DGCR8

Drosha and DGCR8 together engage with the entire length of the dsRNA in pri-miRNAs. Most of the stem surface is coated with protein on the back and one side. The pri-miRNA rests against a seatback formed by the RNase III domains of Drosha. Meanwhile, the Drosha catalytic domains require the C-terminal tails of two copies of DGCR8 to maintain stable folds, making the stem interactions dependent on all three polypeptides. Closer to the apical end, the RNA stem is surrounded by three out of four total dsRBDs from the two copies of DGCR8, breaking the symmetry of the homodimeric DGCR8. As a result, the relative domain orientations are different for the two copies of DGCR8 (Figure 2e). One dsRBD stacks head-to-head against the dsRBD of Drosha, forming a contiguous, chimeric, double dsRBD spacer to measure the ~35 base pairs between the basal and apical junctions. The architecture explains the narrow stem length tolerance and why alterations of either junction can impact the cleavage site and the stem length that has been previously observed [10,30]. Irregularity to the dsRNA, such as mismatches, wobble pairs, and modified bases may affect the proper recognition of the stem that is cradled snuggly, leading to changes in efficiency and precision of Drosha cleavage [31–34].

To capture the wild-type Microprocessor in action, Ca²⁺ was used instead of Mg²⁺, to support RNA binding but not hydrolysis [15]. The density for Ca²⁺ was more prominent for RIIIDb than RIIIDa, suggesting that RIIIDb contains a more robust catalytic site with higher affinity for the required metal ion. Indeed, while RIIIDb contains all the key conserved residues near the cleavage site, RIIIDa is more divergent [35]. Mutations of RIIIDb are found more frequently in cancer [36], and recent reports suggest that the occasional single cleavage events of Microprocessor tend to occur on the 5’ strand involving RIIIDb [27,37,38]. Structural asymmetry between the two catalytic sites may impact the maturation of certain target microRNAs.

III). Apical junction - DGCR8 interactions

Apical and basal junctions of wild-type pri-miRNAs share enough structural similarities that Drosha confuses them easily [27,37,39]. Such ambiguity is resolved by DGCR8 only when it is equipped with heme. Heme binding induces a conformational change that also activates DGCR8 to recognize the terminal loops of pri-miRNAs without affecting the dimerization state [39]. Despite the limited resolution, cryo-EM structures shed light on how DGCR8 interacts with an apical junction. The HBRs cap the top of the stem, poised to interact extensively with the terminal loop (Figure 2b). The HBRs require a large unstructured RNA [40] and may detect the apical UGU motif [9,11,41]. However, the sequence preference may be more flexible when more pri-miRNAs are considered [42]. The HBRs likely anchor the DGCR8 dsRBDs at the apical end of the stem, as the dsRBDs are only ordered in the cryo-EM structure that also includes the HBRs [15,16]. Higher-resolution structure are required to reveal how specific interdomain and protein-RNA interactions enable homodimeric DGCR8 to recognize the structural features of an apical junction.

Apical junctions are binding sites of many loop-binding proteins [28,29,43]. The cryo-EM structures suggest that most RNA-binding proteins and the HBRs would compete for the terminal loop to regulate microRNA maturation [44]. Since the HBR is skewed to one side of the stem, however, an unusually large terminal loop may make it possible for the HBR and another RNA-binding protein to bind to the same loop. More studies are required to build a thorough mechanistic model to explain how proteins binding the apical loop can impact processing differently. A single nucleotide substitution in the terminal loop has been suggested to cause enough structural changes to modulate apical junction-DGCR8 interactions [45], and post-translational modification of the HBR may also affect RNA binding [46]. How heme availability is used to regulate microRNA biogenesis is yet unclear. Alternative transcription initiation of DGCR8 was recently shown to cause overexpression and irreversible aggregation, which may be linked to the necessity of heme for proper folding of DGCR8 [47]. Thus, multiple pathways may target the apical junction-DGCR8 interactions to modulate mature microRNA levels.

Dynamic Assembly and Disassembly of Microprocessor and Implications for Regulation of microRNA Biogenesis

A partially-docked state of Microprocessor on pri-miRNA was revealed by cryo-EM, highlighting that DGCR8-apical junction interactions can play an important role in making assembly or disassembly of Microprocessor on pri-miRNAs dynamic (Figure 2b, 3a) [15]. The half-assembled Microprocessor reveals that the RNA binding domains of DGCR8 are properly localized to engage the apical junction when the mutant Drosha—E1045Q/E1222Q that cannot coordinate divalent cations—is insufficient to fully bind the basal junction. Thus, DGCR8 is capable of recognizing the apical junction with enough specificity and affinity, regardless of Drosha engagement. In another structure for a different pri-miRNA, Drosha fully engages with the basal junction without specific help from the RNA-binding domains of DGCR8 [16]. Therefore, depending on the microRNA, the assembly of the processing complex may be driven by either DGCR8 or Drosha. For the disassembly post-catalysis, however, cleaved RNA near the basal junction would no longer support favorable interactions with Drosha. Thus, after cropping the pri-miRNA, Drosha is more likely to disassemble first to result in an “apical junction loaded” state, if the DGCR8-apical junction interactions are robust enough for the particular pri-miRNA.

Figure 3:

Proposed models of dynamic Microprocessor-pri-miRNA complexes (a) Detailed illustration of the model for how Drosha and DGCR8 assemble and disassemble for a single pri-miRNA. DGCR8 HBRs (grey) and dsRBDs (purple and brown) and Drosha Belt/Wedge (dark blue) area are flexible when RNA is not bound. Partial assembly can occur depending on many factors including the suitability of each RNA junction, heme binding, regulator presence, and protein/RNA modifications. For disassembly after cleaving the pri-miRNA, lack of a basal junction would only allow an apical-junction loaded state if the microRNA supports it. (b) Potential model for processive Microprocessor activity for clustered miRNA. DGCR8 can maintain stable contact with the apical junction without Drosha. After cleaving the recruiter hairpin, Drosha can shift over to a neighbor pri-miRNA while being tethered to DGCR8 via the C-terminal tails (with help from accessory factors). (c) Potential model for how dimerization enables cluster processing. Dimerization (via accessory factors) would help recruit another copy of Microprocessor for the suboptimal pri-miRNA. Stable apical junction-DGCR8 interactions may help maintain the dimeric configuration to let the second cleavage to occur.

Recent findings suggest that clustering microRNAs regulate Drosha processing [48–54]. Accessory factors—ERH and SAFB2—bind DGCR8 and Drosha, respectively, and are necessary for the observed cluster-dependent enhanced processing [48,50,55](Figure 2a). A structure of ERH bound to the N-terminal disordered region of DGCR8 has been determined [55], but how ERH and/or SAFB affect Microprocessor structure or activity on clustered pri-miRNAs is yet unclear. An intermediate state with a stable DGCR8-apical junction complex makes certain scenarios feasible (Figure 3b, c). In a processive model, the “apical junction loaded” state may enable Drosha to work on a neighboring microRNA while DGCR8 is docked on a more optimal substrate. Even for a model that requires dimerization of Microprocessor, ordering of DGCR8 around an apical junction that is not readily dissolved after a catalytic cycle may contribute to stabilizing a dimeric state that aids in the processing of the suboptimal neighbor. UROD, a heme biosynthesis factor, was another major hit in a CRISPR screen to find factors necessary for cluster assistance [48]. Thus, the ability of DGCR8 to robustly recognize the apical junction is likely critical to “cluster assistance”. While more work is necessary to determine the mechanism underlying clusters, structural insights into how Microprocessor assembles and disassembles on a single pri-miRNA provide a useful framework to decipher additional regulatory mechanisms that govern microRNA biogenesis.

Comparison of Drosha to Dicer and Dicer-like enzymes

Drosha and Dicer belong to discrete classes of RNase III enzymes with distinct domain organizations. Due to substrate differences (Figure 1), Dicer contains distinct RNA-binding modules: DExD/H-box helicase, DUF283, and PAZ domains (Figure 4a). In plants, both pri-miRNAs and pre-miRNAs are cleaved by Dicer-like 1 (DCL1), fulfilling the roles of Drosha and Dicer in animals, and the processing can be bidirectional (Figure 4b) [56]. Comparing the pri-miRNA-bound states of DCL1 and Drosha-DGCR8 reveals distinct RNA-protein interactions involved in plants and animals (Figure 4c). DCL1 recognizes the basal junction via a single PAZ domain that detects an internal loop on one strand, rather than the more elaborate clamping mechanism in animals that requires both strands to be unpaired. The helicase domains replace the role of DGCR8 by gripping the dsRNA at the apical end, and the size of the apical stem seems more flexible in plants due to the lack of a cap-like role that the HBR plays. The distinct checkpoints involved in recognizing pri-miRNAs in plants may explain how some transcripts are processed from either end and that additional factors such as HYL1 and SERRATE are required for optimal precision and efficiency [57].

Figure 4:

Structure comparison of RNase IIIs in microRNA biogenesis (a) Domain organization of Dicer with partners and Dicer-like proteins. (b) Schematic of pri-miRNAs in animals and plants. The numbers in circles indicate processing order. (c) Pipes and planks representation of the fully loaded Drosha-DGCR8 complex and Arabidopsis thaliana DCL1 with pri-miRNA (6V5B and 7ELD). Catalytic residues (red spheres) and cleavage sites (red arrows) are indicated. Similar colors as in Figure 2 except for the simplified shading. (d) Pipes and planks representation of Human Dicer with pre-miRNA and TRBP2 (5ZAL), Arabidopsis thaliana DCL1 with bound pre-miRNA (7ELD), DCL3 with dsRNA (7VG2), Drosophila melanogaster Dicer-2 with Loq (7W0E and 7W0F) and dsRNA substrates, and Dicer-2 with R2D2 and miRNA (7V6C). Catalytic residues and cut sites are similarly represented as in (b).

Multiple structures of Dicer homologs with dsRNA substrates have been determined by cryo-EM in the recent years (Figure 4d) [58–63]. When Drosha and Dicer homologs are aligned by the catalytic domains, the “apical” RNA is bound by the helicase and DUF283 domains while the “basal” end—containing 3’ overhangs—is recognized by the PAZ domain. Structures of multiple states of Dicer homologs unveiled how various interdomain movements contribute to binding RNA substrates in multiple ways, which is accentuated by the helicase-dependent RNA translocation that enables Dicer to be processive. Therefore, a dynamic assembly of multiple RNA-binding modules allows both Drosha and Dicer to act on diverse substrates.

Conclusions

Recent structures of Drosha or Dicer in complex with substrate RNAs elucidate how various RNA features are recognized structurally to generate small regulatory RNAs. For microRNAs in particular, precise processing is crucial for accurate target gene identification. Many RNA-detecting modules in Drosha and DGCR8 together accomplish fidelity in substrate recognition and pri-miRNA processing. Comparing multiple structures also reveals how the multimodal binding makes a more dynamic assembly feasible, opening doors to untangle regulatory mechanisms of small RNA biogenesis.