Cryo-EM Structures of Human Drosha and DGCR8 in Complex with Primary MicroRNA

Summary

Metazoan microRNAs require specific maturation steps initiated by Microprocessor, comprised of Drosha and DGCR8. Lack of structural information for the assembled complex has hindered understanding how Microprocessor recognizes primary microRNA transcripts (pri-miRNAs). Here we present a cryo-electron microscopy structure of human Microprocessor with a pri-miRNA docked in the active site, poised for cleavage. The basal junction is recognized by a four-way intramolecular junction in Drosha, triggered by the Belt and Wedge regions that clamp over the ssRNA. The belt is important for efficiency and accuracy of pri-miRNA processing. Two dsRBDs form a molecular ruler to measure the stem length between the two dsRNA-ssRNA junctions. The specific organization of the dsRBDs near the apical junction is independent of Drosha core domains, as observed in a second structure in partially docked state. Collectively, we derive a molecular model to explain how Microprocessor recognizes a pri-miRNA and accurately identifies the cleavage site.

Graphical Abstract

eTOC blurb:

Partin et al. present cryo-EM structures of the Microprocessor complex bound to a pri-miRNA, revealing how Drosha and DGCR8 recognize substrates and determine cleavage sites. Two dsRBDs form a molecular ruler to measure the stem length, and “Belt” and “Wedge” in Drosha play key roles in detecting the basal junction.

Introduction

MicroRNAs (miRNAs) constitute a major class of small non-coding RNAs that regulate gene expression throughout normal development and also in many pathological processes. To generate functional ~22 nt-long metazoan miRNAs, primary microRNA transcripts (pri-miRNAs) undergo specific cleavage events by Drosha to generate precursor miRNAs (pre-miRNAs), which are further processed by Dicer (Bernstein et al., 2001; Hutvagner et al., 2001; Lee et al., 2003). The initial step by Drosha requires another RNA binding protein, DGCR8 (Denli et al., 2004; Gregory et al., 2004; Han et al., 2004; Landthaler et al., 2004). Active Microprocessor is known to contain one copy of Drosha and two copies of DGCR8 (Nguyen et al., 2015; Partin et al., 2017). Both proteins contain domains that typically bind RNA, including RNase III domains (RIIIDs) and dsRNA binding domains (dsRBDs). DGCR8 also contains a heme-binding region that gains specificity and affinity for terminal loops of pri-miRNAs upon binding heme (Nguyen et al., 2018; Partin et al., 2017). Short RNA motifs have been observed to affect cleavage efficiency and/or location, and the length of the stem has also been shown to be an important determinant of a suitable pri-miRNA substrate (Auyeung et al., 2013; Fang and Bartel, 2015; Ma et al., 2013; Zeng and Cullen, 2005). Although partial structures of isolated Drosha and DGCR8 have been determined by X-ray crystallography (Kwon et al., 2016; Sohn et al., 2007), how the polypeptides work together to recognize and process pri-miRNAs accurately remains unclear.

Results

Cryo-EM structure of a Drosha/DGCR8/pri-miRNA complex

To derive a structural model for an active Microprocessor/pri-miRNA (MP/RNA) assembly, we aimed to prepare a homogenous protein/RNA complex. We purified recombinant protein complexes using the truncated constructs of Drosha and DGCR8 that are sufficient for accurately processing pri-miRNAs in vitro when fully saturated with heme (Figure 1A). We chose pri-miR-16–2 as the substrate RNA for structure determination, because it not only binds Drosha/DGCR8 tightly enough to copurify during size-exclusion chromatography (Figure S1A and S1B), but is processed correctly regardless of the heme presence (Figure S1C). Heme-independent processing suggests that pri-miR-16–2 has an optimal Drosha-binding site at the basal dsRNA-ssRNA junction (Partin et al., 2017). We included 20 nucleotides flanking the Drosha cleavage sites, as they were enough to maintain robust processing. Furthermore, to capture the pre-catalytic state while inhibiting RNA cleavage we replaced the magnesium ions required for RNase III activity with calcium (Figure S1D). Finally, because the complex dissociates readily during grid preparation, we treated the complex with a crosslinking agent to enable structure determination by cryo-electron microscopy (cryo-EM).

Figure 1. Cryo-EM structure of a Drosha/DGCR8/pri-miRNA complex.

(A) Domain organization of Drosha and DGCR8. CED, central domain of Drosha. RIIID, RNaseIII domain. dsRBD, double-stranded RNA-binding domain. HBR, heme-binding region. CTT, C-terminal tail. The sequences used for cryo-EM structure determination are marked underneath. (B) Cryo-EM maps of Drosha/DGCR8/pri-miR-16–2, segmented and colored by domain as shown in Figure 1A. The HBR regions of both DGCR8 monomers are in gray. (C) Model of the MP/RNA complex shown with three different views. Nucleotides on either side of cleavage sites are colored magenta. C-terminal tails (CTTs) of DGCR8 are labeled. (D) Secondary structure diagram of pri-miR-16–2 used for cryo-EM. Protein/RNA contacts (distance < 4.5 Å) are highlighted according to the protein domain or region, with the same color scheme used in Figures 1A–C. Wedge and Belt are subregions of the CED of Drosha. Nucleotides that were not modeled due to poor resolution or disorder are colored gray. Drosha cut sites are indicated with magenta arrowheads. Pri-miR-16–2 in the model is numbered from 1–105 from 5’ to 3’.

Using the MP/RNA sample as described above, we determined a cryo-EM structure of an active human Microprocessor in complex with pri-miR-16–2 (Figures 1B, 1C, S1E–S1G, and S2). The overall resolution is approximately 3.7 (Figure S2D and Table 1). The quality of the map varies for different regions, with higher resolution for Drosha and the majority of the pri-miRNA, and lower resolution for DGCR8 and the apical loop of the substrate RNA (Figures S2E and S2F). We used crystal structures of isolated Drosha and DGCR8 dsRBDs to aid with model building (Kwon et al., 2016; Sohn et al., 2007). The regions resolved in the crystal structure of Drosha in isolation aligns well with the cryo-EM density, except for the dsRBD, which swings toward the bound RNA (Figure S1H–I). In addition, we had to model many regions of Drosha (221 residues out of 908 included in our final model) de novo, because they previously could not be resolved in the isolated Drosha structure (Kwon et al., 2016). The cryo-EM map for the apical region, especially for the heme-binding region (HBR) of DGCR8, is resolved to lower resolution. However, using the crystal structures of the dsRBDs (residues 489–583 and 592–701 for dsRBD1 and dsRBD2, respectively) we could perform initial rigid-body docking followed by refinement to reveal their spatial disposition with respect to Drosha and the bound pri-miRNA. We were also able to visualize the density connectivity and thus model de novo the majority of pri-miR-16–2, except for the apical loop and a few nucleotides near the 5’ and 3’ ends (Figure 1D). Determining the register of the RNA was guided by the secondary structure calculated by mfold (Zuker, 2003), as the location of the mismatches are obvious in the cryo-EM map (Figure S1F).

Table 1.

Cryo-EM data collection, processing, and model validation.

Cryo-EM Map	Active state	Inactive state
Data collection and processing
Microscope	Titan Krios	Titan Krios
Voltage (kV)	300	300
Camera	Gatan K2 Summit	Gatan K2 Summit
Magnification	130,000x	130,000x
Pixel size (Å)	1.06	1.06
Total exposure (e-/Å2)	46.8	46.8
Exposure time (s)	6	6
Number of frames per exposure	30	30
Energy filter slit width (keV)	20	20
Data collection software	EPU	EPU
Defocus range (μm)	−1.3 – −3	−1.5 – −3.6
Number of micrographs	12,681	6,070
Number of initial particles	1,385,678	1,063,710
Number of particles for 3D analyses	937,668	842,037
Symmetry	C1	C1
Number of final particles	505,640	381,468
Resolution (0.143 gold standard FSC, Å) masked/unmasked	3.7/4.6	4.4/5.8
Local resolution range (Å)	3 – 7	4 – 8
Atomic model refinement
Software	Phenix	Phenix
Number of protein residues	1341	1336
Number of nucleotide residues	78	66
Number of atoms	12593	12254
Geometric parameters (r.m.s.d.)
Bond length (Å) (# > 4σ)	0.007	0.006
Bond angle (°)(# > 4σ)	1.061	1.113
Ramachandran statistics
Favored (%)	88.35	86.48
Allowed (%)	11.50	13.52
Disallowed (%)	0.15	0
Validation
MolProbity score	2.09	2.25
Clashscore	9.48	11.32
Rotamer outliers (%)	0.33	0.75
Crosscorrelation	0.83	0.81
B factor (Protein/Nucleotide)	132.85/127.69	184.94/200.75
C-beta deviation	0	0
EMRinger score	1.52	0.69
CaBLAM outliers (%)	6.77	8.36
Nucleic acid geometry
Probably wrong sugar puckers	0	0
Bad bonds	0	0
Bad angles	0	0

FSC, Fourier shell correlation; r.m.s.d., root-mean-square deviation.

The location of the RIIID active sites coincides with the known Drosha cleavage sites on pri-miR-16–2, according to miRBase (Griffiths-Jones, 2004). We used the Q-score to evaluate the density resolvability and the quality of the model per residue, for both Drosha and RNA, and it is consistent with the expected scores for the reported resolution (Pintilie et al., 2019) (Figures S2H–S2L). Analysis of the MP/RNA complex structure allows us to visualize the relative locations of various domains of Drosha and DGCR8 and thus to interpret their functional roles in recognition of a pri-miRNA, as the direct contacts can be mapped for almost the entire length of the stem (Figure 1D). Furthermore, we observe extra density with high resolvability Q score (Pintilie et al., 2019) near the active sites, especially in the RIIIDb domain, most likely due to ordered calcium ions (Figure S2G) (Court et al., 2013; Nicholson, 2013). The relatively high metal density near RIIIDb in comparison to RIIIDa may be because calcium binds the RIIIDs differently than the natural ligand, magnesium. RIIIDb may also be innately better than RIIIDa at binding divalent cations. Interestingly, RIIIDa contains a less common asparagine (N1042) at a position occupied by aspartate in other RNaseIII domains. The latter hypothesis is intriguing since E1147--one of the key coordinating residues in RIIIDb--is a mutation hotspot for Wilms tumor patients, while RIIIDa mutations are much less frequently found in these patients (Lin and Gregory, 2015). Together, we present a three-dimensional model for how Drosha and DGCR8 assemble on a pri-miRNA substrate, poised for cleavage.

Multiple rearrangements in Drosha enable recognition of the basal dsRNA-ssRNA junction

The cryo-EM structure reveals the detailed interactions between Drosha and the bound pri-miRNA, and various regions are likely to undergo RNA-induced folding, as they could not be modeled in the isolated crystal structure (Figures 2A and 2B). Most of the RNA stem docks against the globular core of Drosha composed of the RIIIDs and the Central domain (CED). On the opposite side of the core domains, various parts of Drosha rearrange or become folded to wrap around the RNA, consequently increasing contact between protein and RNA. The dsRBD of Drosha swings inward to bind the basal stem region of the pri-miRNA (Figure S1H–I). Previously unresolved regions of CED, such as a long segment that includes two antiparallel helices (757–848, green) and a stretch with two shorter helices (930–957, light green) are visible in our cryo-EM complex structure (Figures 2A and 2B), likely because of the intimate contact with RNA to rigidify them. Moreover, newly visualized regions such as the basal tip of the CED and the C-terminal tail of Drosha likely depend on the polypeptide regions that become ordered with RNA for folding. Together, our cryo-EM structure of Microprocessor with fully docked pri-miRNA elucidates how several distinct polypeptide regions are involved in coordinating RNA binding to ensure stable and specific interactions.

Figure 2. Multiple rearrangements in Drosha enable recognition of the basal dsRNA-ssRNA junction.

(A-B) Views of Drosha and pri-miRNA with front (A) and side (B) views. Colored regions highlight major conformational changes of Drosha when compared to the isolated Drosha structure, except for RIIIDs (dark gray) which are colored for orientation. Nucleotides on either side of cleavage sites are colored magenta. (C) Cryo-EM density for Helix-1 of the Belt with model. (D-E) Cartoon illustration of MP/RNA showing the RNA-induced 4-way junction (D), and rotated to show the interaction with RIIIDb (E). Same color scheme as Figure 2A. 4-way junction is indicated by a red circle. (F) Close-up view of the interactions of the Belt with the rest of Drosha and RNA. Same color scheme as in Figure 2A. (G) Top-scoring intra-Drosha crosslinks in MP/RNA complex (top) and MP alone (bottom). Red lines represent hits identified in 3 out of 3 replicates. Gray lines represent hits found in 2 of 3 replicates. (H) Residues involved in red crosslinks from MP/RNA in Figure 2G are shown with distances.

The largest consecutive region of Drosha that we modeled de novo includes a two-stranded coiled coil (753–848) that we refer to as the “Belt” (Figures 2A–2C). Previously this region of Drosha was proposed to fold into a “PAZ-like” domain similar to Dicer (Kwon et al., 2016); instead, it forms a two-helix segment that binds the basal junction. The Belt acts like a seatbelt that straps across the RNA and buckles into a 4-way junction consisting of the Belt, the dsRBD, the C-terminal peptide, and another previously unresolved region we have named the “Wedge” (Drosha residues 930–957) (Figures 2D and 2E). The Belt mostly interacts with the ssRNA immediately next to the dsRNA, to clamp down the separated RNA strands. When the Belt crosses over the unwound RNA immediately adjacent to the basal dsRNA-ssRNA branch point, the tip of the helices packs tightly against the 4-way junction. Van der waals interactions are observed between P808 and P1335, and I942 bridges between H802 and T1271 (Figure 2F). In this conformation, the ɑ-helices of the Wedge and the C-terminal tail bundle with the last helical segment of the RIIIDb (Figure 2E). The linker between the dsRBD and the RIIIDb is flexible prior to substrate binding (Figure S1H–I), but RNA induces the connection between the dsRBD and the RIIIDs to rigidify. Thus, when the Belt locks into the 4-way junction, the RNA-binding modules become more rigidly coupled, strengthening the interactions with bound RNA. Consequently, the formation of the multi-domain buckle provides specificity for the ssRNA/dsRNA junction by positive reinforcement. To validate our structural model, we used crosslinking mass spectrometry to probe the Microprocessor conformation in solution (Leitner et al., 2014). Few crosslinks are observed for the MP/RNA complex, but two of the most robust are between the Belt and the dsRBD, which come in close contact in the conformation observed in the cryo-EM structure (Figures 2G and 2H, and Table S1). The same interdomain linkages are not observed when the crosslinking reagents are added to MP without RNA, suggesting that the Belt/dsRBD interaction as observed in the structure is specific to the state in which the pri-miRNA is properly docked.

The Helical Belt is important for pri-miRNA processing

The cryo-EM structure of the MP/RNA complex provides a physical framework to understand how various features of the pri-miRNA might affect processing. Most of the dsRNA is well-resolved in our cryo-EM structure, and we have modeled a total of 33 base-pairs (bps): 21 out of 23 total bps on the apical side and 12 bps on the basal side of the 5’ cut site (Figures 3A and 3B). The required lengths of dsRNA on both sides of the cleavage sites determined by previous studies are thus explained by the RNA-binding protein domains that span the entire distance (Auyeung et al., 2013; Fang and Bartel, 2015; Han et al., 2006; Ma et al., 2013; Nguyen et al., 2015). To measure the stem length consistently, both ends are marked as dsRNA-ssRNA junctions. On the apical end, the terminal loop is capped by the HBR and the four dsRBDs surround the apical end of the stem. The basal side is clamped down via the Belt, which locks into the 4-way junction. Between these clearly marked ends, the dsRBDs from Drosha and DGCR8–1 interact to form a contiguous, elongated protein mass that measures the entire length of a stem of about 35 bps, acting as a “double-dsRBD ruler” (Figure 3A–3B). If we model a longer dsRNA protruding basally, the double-helix structure is likely to clash with the Belt when the stem is only 2 bp longer (Figure 3C). For a typical pri-miRNA with a dsRNA-ssRNA junction, the Belt keeps the 5’ flanking arm away from the 3’ strand. Previously a “Bump helix” (residues 910–919) was proposed to block unbranched dsRNA from binding Drosha (Kwon et al., 2016). Although Arg914 of the Bump helix makes contact with the phosphate backbone of the 3’ ssRNA (Figure 3C), it is directed away from the RNA. Thus, while the Bump helix may support the unwound conformation by stabilizing the hold on the 3’ ssRNA arm, it does not appear to present a “block” against dsRNA as previously suggested. In summary, the cryo-EM structure of the MP/RNA complex highlights the roles of the double-dsRBD ruler and the Belt-triggered 4-way junction in recognizing the correct stem-loop for processing.

Figure 3. The Helical Belt is important for pri-miRNA processing.

(A-B) Cartoon and surface representation of MP/RNA showing distances from cut site to apical and basal branch points with front (A) and back (B) views. CED and RIIIDs are shown in blue, Drosha dsRBD in cyan, Belt in green, DGCR8–1 in yellow-orange, and DGCR8–2 in purple. (C) Pri-miRNA structure compared to a model with an elongated stem. The Belt (green) clashes with the modeled dsRNA (right). Arg914, a residue within the “Bump Helix”, is labeled. (D) In vitro processing assays of pri-miR-16–2 and RNAs containing stem insertions. [MP] is 32 nM. (E) Quantified in vitro pri-miRNA processing assays with a series of Belt mutants. Data are from three replicates, and error bars represent ± S.D. [MP] is 8 nM. *p < 0.001 (unpaired t-test for WT v. each mutant). (F) In vitro processing assays using wild type Microprocessor or the ΔBelt mutant on various pri-miRNAs. [MP] is 65 nM. Solid or dashed lines indicate grouping of separate gel regions.

When we insert bps into the stem of pri-miR-16–2 near the basal junction, we observe a precipitous decline in processing efficiency when more than 1 bp is inserted (Figures 3D, S3A and S3B). The endonucleolytic cleavage that occurs for the mutant pri-miRNAs is at the same location as the wild type pri-miRNA, whether the flanking regions are short (20 nt) or long (60–70 nt), suggesting that additional mechanisms are in place for choosing the cut site in pri-miR-16–2 (Figure S3C). When the stem is shortened by a mere 1 bp, the enzymatic activity is also dramatically reduced (Figure S3D). Shorter stems would result in the double-dsRBD ruler hanging over the edge of the dsRNA or deforming, thereby preventing the formation of a proper 4-way multi-domain junction (Figure 2D). Together, these results suggest that forming the 4-way junction acts as a positive checkpoint; in the presence of a dsRNA-ssRNA junction 12–13 bps away from the 5’ cut site, the Belt clamps down and buckles into the 4-way junction to stabilize the tightly bound conformation of Drosha on pri-miRNA.

Given its critical contact with the basal dsRNA-ssRNA junction, we investigated whether the Belt affects the processing of pri-miR-16–2. When the two helical segments are deleted (ΔBelt), the processing efficiency is dramatically reduced (Figures 3E and S3E–S3G). The first helix (H1) interacts with the groove formed by the most basal portion of the stem, and the basic side chains (K792, K795, and K799) of H1 face the phosphate backbone to support favorable interactions (Figure 2F). Mutating all three to glutamates (K3E3) also reduces the Microprocessor activity (Figure 3E). Moreover, mutating the side chains of N806 and Q1336 to Asp and Glu, respectively, is detrimental to pri-miRNA processing, likely due to charge repulsion in the Belt/C-terminal peptide contact area. The mutant Microprocessor complexes were purified over multiple chromatography steps and have similar K_D values for pri-miR-16–2 as the wild type Microprocessor, except for the MP^K3E3, which results in ~ 2 fold decrease in RNA affinity (Figure S3H). When the enzyme concentration is increased by 2 fold, MP^K3E3 shows similar activity as MP^WT, but the other mutant complexes still exhibit largely impaired enzymatic activity (Figure S3I). Moreover, when the substrate concentration is increased by 2 orders of magnitude, MP^ΔBelt remains less efficient than MP^WT (Figures S3J–S3K). Therefore, missing the Belt is likely to debilitate MP activity on pri-miR-16–2 in a way that affects more than the overall affinity for pri-miRNA. We also investigated the possibility of the Belt affecting stem length specificity. Inserting or deleting bps in the stem did not improve the activity of MP^ΔBelt over MP^WT, over a wide range of substrate concentrations (Figures S3L–S3M). Thus, the Belt is critical for detecting the basal RNA structure, and strengthening the 4-way junction (Figure 2D) upon binding appropriate substrates, thereby enhancing processing.

Many protein or RNA features have been shown to affect processing of distinct miRNAs differently (Nguyen et al., 2018; Partin et al., 2017). Therefore, we tested the importance of the Belt in processing of other miRNAs by using the Microprocessor lacking the two helices (ΔBelt). For most miRNAs, Drosha cleavage is less efficient in the absence of the Belt, though to varying degrees (Figure 3F). For miR-107 and miR-106b, we also observe a shift in cut site location (Figure S3N). Interestingly, some pri-miRNAs (pri-miR-30a and pri-miR-125a) are processed similarly with or without Belt deletion, and this independence from the Belt is observed even at lower enzyme concentrations (Figures S3O–S3P). For certain pri-miRNAs that contain other features or motifs that make them optimal substrates (Auyeung et al., 2013; Fang and Bartel, 2015), loss of Belt may not result in a significant processing defect. We have also determined that pri-miR-30a and pri-miR-125a contain optimal basal junctions because their processing is heme-independent (Partin et al., 2017). Therefore, the Belt helices of Drosha generally contribute to both efficiency and accuracy of Microprocessor, and the specific effect on each miRNA may vary.

The Wedge and Belt of Drosha form a narrow tunnel for the ssRNA at the basal junction

When the basal junction is clamped down by the Belt, it forms a narrow tunnel with another region of CED which we call the Wedge (928–957) (Figures 4A and 4B). The hollow is lined with basic residues to mediate interactions with the acidic phosphate backbone (Figure 4A). The Wedge is disordered in the isolated Drosha structure, but it is induced to fold in our complex as it wedges into the crevice between the stem and the unwound ssRNA near the 5’ end (Figures 4B and 4C). Moreover, the Wedge also contacts the “GHG” motif (GUG in pri-miR-16–2) that has been proposed to affect cleavage efficiency and accuracy, although the interactions are through the RNA backbone (Figure 4C) (Fang and Bartel, 2015; Kwon et al., 2019). The mismatched U may be recognized via a unique bulge in the helical backbone (Fig S1F). The GUG sequence is located near the basal end of the dsRNA, and the dsRBD of Drosha is situated proximally, as predicted from modeling the typical mode of binding for dsRBDs (Kwon et al., 2019; Masliah et al., 2013). However, the distance between the GUG nucleobases and the dsRBD is large, and there is little evidence for sequence-specific contact. The only interaction we observe is a weak contact between Q1266 of Drosha and the Uracil base in the most permissive “H” position. Therefore, although the Wedge serves to recognize the mismatch in the middle of the GHG motif via the structural change on the backbone, we only observe weak structural evidence for sequence preference at the GHG site. Our structural findings suggest that the most critical RNA feature at the GHG position is the mismatch, thus supporting the idea that a certain degree of local deformability is preferred in the stem region basal to the cut sites (Quarles et al., 2013).

Figure 4. The Wedge and Belt form a narrow tunnel for the ssRNA at the basal junction.

(A) Vacuum electrostatic potential surface depicting basal side of the tunnel formed by the Belt and Wedge. RNA is shown in orange. (B) Cartoon representation of MP/RNA showing pathway of the basal junction between the Belt and Wedge. (C) Same as Figure 4B, but rotated 30° to show interactions between the Wedge and GHG motif (black). Nucleotides of the GHG motif (G88, U89, and G90) are indicated by white circles. (D) Overall view of Wedge and single-stranded nucleotides in flipped conformation near the basal junction. (E-F) Close-up views of interactions between Drosha and single-stranded nucleotides A99 (E) and U7 (F). (G-H) Quantification of in vitro pri-miRNA processing results on pri-miR-16–2 variants containing shortened 5’ (G) or 3’ (H) flanking segments. [MP] is 8 nM. Data are from three individual replicates, and error bars represent ± S.D. (I) RNA secondary structure diagram showing the essential ssRNA nucleotides in orange and the nonessential ones in gray, with the same number system as in Figure 1D.

The Wedge, together with the Belt, forms intimate contact with the first few nucleotides in the ssRNA region (Figures 4D–4F). On each ssRNA strand, one nucleotide (Ura7 on 5’ and Ade99 on 3’) is best modeled in a flipped out conformation (Figure 4D). Drosha stabilizes the two nucleobases in pockets where pi-pi and pi-cation interactions are observed between the base and the side chains, and these pockets are formed at the interface between the Belt and the Wedge (Figures 4E and 4F). Furthermore, as a narrow groove is formed between the dsRNA and the 5’ ssRNA arm, the concentrated negative charge from the phosphates is mitigated by the basic side chains from the Wedge (R938, K939, and K940) (Figure 4F). When mutations are introduced to disrupt a nucleobase binding pocket or the electrostatic interactions with the backbone, there is a markedly reduced processing without affecting the overall affinity for pri-miR-16–2 (Figures S4A–S4F). Therefore, forming a narrow tunnel is likely to ensure the presence of ssRNA flanking regions, to help Drosha to distinguish pri-miRNAs from other dsRNAs or dsRNA-containing RNAs.

Given that the length of the tunnel is not long, we investigated how many single-stranded flanking nucleotides are needed for processing of pri-miR-16–2. For both of the flanking arms, we determined that 13 nucleotides (counting from the Drosha cut site) are necessary for robust processing (Figures 4G, 4H, and S4G). This means that only 1 or 3 single-stranded nucleotides are critical on 5’ and 3’ ends, respectively (Figure 4I). Although we observe such short essential arms for pri-miR-16–2, longer arms may be necessary for other miRNAs, especially if the basal junction is not a strong binding site for Drosha. Interestingly, the flipped nucleotides mark where the seemingly less critical regions of ssRNA begin (Figure 4D). Since pri-miRNAs normally have longer arms in vivo, the ability of the Belt and the Wedge to stabilize the flipped nucleobases may be important to maintain separation of the ssRNA strands, thus making them easier to clamp.

Cryo-EM structure of Drosha/DGCR8 with partially docked pri-miRNA

Most of the Drosha/RNA contacts are between the pri-miRNA stem and the two RIIIDs, and divalent cations are known to be important for mediating RIIID interactions with RNA (Court et al., 2013). However, even in the presence of EDTA or mutations of key metal-coordinating residues (E1045Q/E1222Q), the overall affinity for pri-miRNA is not reduced significantly (Figures S5A–S5E). Drosha RIIID residues involved in magnesium coordination have also been found to be frequently mutated in Wilms tumor patients (Lin and Gregory, 2015). To gain insight into the architecture of the mutant MP/RNA complex, we determined a cryo-EM structure of Microprocessor containing Drosha with a point mutation in each of the catalytic sites (E1045Q/E1222Q), at ~ 4.4 Å-resolution (Figures 5A, 5B, S5F, and S6, and Table 1). Altering the acidic side chains abrogates the ability to coordinate divalent cations that are needed to mediate interactions with the pri-miRNA stem and thus facilitate hydrolysis. The two cryo-EM structures are best compared by superimposing the dsRBDs (2 from each DGCR8 and 1 from Drosha); the five dsRBDs exhibit similar relative orientations in the two structures, and aligning the domains also results in good agreement between the two RNA conformations (Figure 5C). Therefore, the arrangement of all of the dsRBDs on pri-miR-16–2 is preserved in both structures. However, the core domains of Drosha--the RIIIDs and CED--move away from the RNA, reducing direct contact with the substrate. When the two structures are aligned by the Drosha core domains, the RIIIDs and the CED align well, indicating that the catalytic mutations do not directly induce a major conformational change. However, as the RIIIDs release the stem in the absence of divalent cations, there are two major conformational changes: the Belt swings inward to where the RNA usually docks, and the dsRBD also moves away from the core, likely to retain its contact with the dsRNA (Figure 5D and Movie S1). Thus, without a properly docked pri-miRNA, the helical Belt can no longer interact with the 4-way junction across the RNA; instead, it leans against the core of Drosha on the other side of RNA (Figure 5D; green arrow). Comparing the two structures provides a model for how the Belt buckles into the 4-way junction upon engaging productively with the pri-miRNA (Movie S2). We tested if the observed change in Belt conformation is also evident in solution, using crosslinking mass spectrometry. When we use catalytically inactive Drosha to form MP/RNA complexes, we do not detect the Belt-dsRBD crosslinks observed in the active state (Figure 5E and Table S1). Therefore, even when RNA is associated with MP, the RIIIDs require properly coordinated divalent ions to support the activated conformation where the Belt reaches across the ssRNA to interact with the rest of Drosha.

Figure 5. Cryo-EM structure of Drosha/DGCR8 with partially docked pri-miRNA.

(A-B) Cryo-EM map (A) and model (B) of the partially docked Drosha/DGCR8/pri-miR-16–2 complex. The HBR regions of both DGCR8 monomers are gray in the map and not modeled. Colors are according to key in panel B. (C) Superimposition of partially docked (gray) and fully docked (colored according to Figure 3A) structures, showing shift in Drosha orientation. (D) Superimposition of inactive (gray) and active (marine, cyan and green) Drosha, showing the conformational rearrangements of the belt (green) and dsRBD (cyan) upon binding RNA. (E) Crosslink map of top-scoring intra-Drosha crosslinks for inactive (top) and active (bottom) MP/RNA conformations. Same color scheme as for Figure 2A. (F) Cartoon representation of apical RNA-binding modules observed in both partially and fully docked structures. Fully docked structure is shown, with same color scheme as Figure 5C. (G) Surface representation of RNA from fully docked structure. Colored nucleotides participate in interactions conserved between both MP structures, with cyan, yellow-orange and purple indicating interactions with Drosha, DGCR8–1, and DGCR8–2, respectively. (H-I) Electrophoretic mobility shift assay (EMSA) results showing the binding of DGCR8 (H) and Drosha^+CTT (I) to pri-miR-16–2. Protein concentrations are (left to right): 0, 0.13, 0.26, 0.52, 10.4, and 2.08 μM. Estimated dissociation constants (K_d) are listed below each gel. Spaces indicate grouping of separate gel regions.

Despite the conformational changes in Drosha, DGCR8 and the dsRBD of Drosha maintain remarkably similar conformations in both cryo-EM structures (Figures 5C and 5F). The footprints of the dsRBDs on pri-miR-16–2 (conserved in both structures) span nearly the entire length of the stem region (Figures 5G, and S5G). Consistent with such extensive contact, the affinity of DGCR8 for pri-miR-16–2 is high (Figures 5H and S5H). In contrast, when we test a Microprocessor complex lacking any RNA-binding regions of DGCR8, Drosha^+CTT (containing only the C-terminal tails of DGCR8 to increase Drosha solubility (Nguyen et al., 2015)), we observe no detectable affinity for the RNA substrate without added divalent ions (Figures 5I and S5I). Therefore, in the partially docked (calcium-free) structure, what we observe is likely a state in which both specificity and affinity for RNA are driven by the HBRs and the dsRBDs, mostly from the DGCR8 polypeptides. After we crosslink the wild type MP/RNA complex in the absence of calcium (similar to the mutant MP/RNA complex), adding magnesium can still activate its ability to cleave the pri-miRNA similar to the uncrosslinked complex (Figure S5J). Thus, the partially docked conformation is likely capable of transitioning to the active conformation (Figure 1). To drive the complex to the fully docked conformation, the coordinated divalent cations are necessary to switch on the specificity of Drosha for dsRNA (Figure S5K). It is noteworthy that there is little difference in overall RNA-binding affinity among MP^WT with calcium, MP^WT without calcium, and MP^{E1045Q/E1222Q} (Figures S5C–S5E). The presence of numerous RNA-binding domains makes dissecting individual affinity contributions challenging. Thus, the partially docked structure provides an opportunity to examine the RNA/protein interactions at the basal and apical regions separately. Therefore, by comparing the affinities of the individual components (Figure 5H) and analyzing conserved contacts in both structures (Figure 5F), we conclude that the HBRs and the dsRBDs have specificity and affinity for pri-miRNAs independent of the conformational state of Drosha. Furthermore, comparing the two structures also reveals the extent of flexibility in interdomain linkers (eg. to dsRBD and Belt of Drosha) that permit large domain movements within a MP/RNA complex.

Discussion

Our results allow us to propose a physical model for how a pri-miRNA interacts with the Drosha/DGCR8 complex (Figure 6). The robust specificity of Microprocessor for the apical junction is driven by DGCR8; the HBRs bind the terminal loop and the dsRBDs of DGCR8 surround the apical half of RNA stem. The dsRBD of Drosha recognizes the dsRNA on the basal half, and in tandem with the first dsRBD of DGCR8–1, forms a “double-dsRBD” molecular ruler to measure ~35 bp of RNA stem. Similar RNA binding modes are observed for the HBRs and the dsRBDs in both the fully docked and partially docked structures, and they are sufficient for nanomolar binding affinity. In the presence of divalent cations such as magnesium or calcium, the catalytic domains RIIIDa and RIIIDb also engage with the pri-miRNA stem. Proper docking of the pri-miRNA in the active site positions the basal dsRNA-ssRNA junction near the two-helix Belt, which can then trigger coordinated interactions with other regions of Drosha (dsRBD, C-terminal tail, and Wedge) to form the 4-way junction (Figure 2D). The observed Belt rearrangement stabilizes Drosha in a conformation that favorably contacts the dsRNA-ssRNA junction, thus reinforcing substrate RNA specificity. The specific and tight contacts via the HBRs and the dsRBDs near the apical junction, combined with the molecular clamp imposed by the helical Belt and the Wedge at the basal end, together ensure that the bound RNA has two junctions separated by a stem of particular length (35 ± 1 bp (Fang and Bartel, 2015)). Therefore, our physical model reveals many features of the Microprocessor that are important to ensure that a proper substrate is loaded and poised for hydrolysis at the specific nucleotide linkages.

Figure 6. Model for assembly of Microprocessor/pri-miRNA complex.

Proposed model for pri-miRNA recognition by Drosha. Different protein modules recognize specific structural features of pri-miRNAs, with the HBR recognizing the terminal loop, dsRBDs recognizing the stem, and the Belt and Wedge recognizing the basal junction. Flexibility between the modules enables recognition of diverse substrates that meet the structural requirements. The dsRBDs, the Belt and the Wedge, together form a molecular ruler that selects stem-loops with a length of approximately 35 bp. DGCR8 CTTs are removed for clarity.

In addition to providing a detailed view of an active MP/RNA complex, the fully docked structure can be compared to the partially docked structure to provide insight into how the macromolecules assemble. Furthermore, examining the isolated Drosha crystal structure also shows the flexible nature of the important RNA-contacting regions (eg. dsRBD, Wedge, Belt), most of which are disordered or conformationally distinct from our MP/RNA structures. The ability of the dsRBDs to maintain the same contact with the pri-miRNA, even in the absence of Drosha RIIIDs engaging with the RNA, suggests that the dsRBDs (together with the HBRs) have robust affinity and independent specificity. Thus, the dsRBDs and the HBR domains may drive the assembly of the MP/RNA complex, or release the processed product last, especially in the case of miR-16–2.

Since the discovery of Drosha, many models have been proposed to describe how pri-miRNAs are distinguished from other RNAs. The characteristic features of pri-miRNAs have been investigated in various ways. First, the basal (lower) and apical (upper) stem lengths have both been suggested to affect substrate selection and cut site determination (Auyeung et al., 2013; Fang and Bartel, 2015; Han et al., 2006; Kwon et al., 2019; Ma et al., 2013; Nguyen et al., 2015). As shown in the current study with MP^ΔBelt, a lack of proper basal junction detection can manifest as a loss in efficiency, shift in cut site, or both (Figure 3F). Although the resultant processing defect may vary for each pri-miRNA, the biochemical data converge on the importance of the stem length. Our cryo-EM structures provide a framework to explain why the length of the dsRNA between the junctions is physically measured by the size of the polypeptide domains and their arrangement along the RNA.

Short sequence motifs in pri-miRNAs have also been shown to impact processing efficiency, although deriving a universal model for their importance was difficult (Auyeung et al., 2013; Fang and Bartel, 2015). A “GHG” motif has been highlighted for its effect on processing efficiency, and our structural observations show that the GHG motif is at the center of the 4-domain junction, the “buckle”. Although we do not observe base-specific contacts, the unique RNA backbone geometry near the motif suggests that it contributes to the RNA structure that is important to support the formation of the 4-way junction. A “UG” motif in the 5’ arm has been proposed to play a role in determining processing efficiency. The unpaired Ura in this motif is in the same position as the 5’ flipped Ura (Ura7) in our active model. The binding pocket for this uracil base may favor Ura over other nucleotides due to its small size (Figure 4F). However, the pockets for both of the flipped bases rely mostly on pi-interactions and may not be highly discriminating for the nucleobase identity. In the 3’ arm, the “CNNC” motif has been proposed to contribute to the processing of some pri-miRNAs, and DEAD-box helicases may act through this motif to remodel the RNA to make the cleavage event more efficient (Auyeung et al., 2013; Mori et al., 2014; Ngo, 2019). For certain pri-miRNAs, the additional help in unwinding the flanking ssRNAs may be necessary for Drosha to then clasp the flexible RNA in the tunnel formed by the Belt and the Wedge. Variations in stem length, primary sequence motifs, and structural features in the flanking arms are some of the elements that could influence the processing efficiency and accuracy of a pri-miRNA, as well as the degree to which it is sensitive to deletion of the Belt. The built-in plasticity of the Microprocessor complex shown through our cryo-EM structures is likely critical for its ability to accommodate and utilize the diversity among the pri-miRNAs, while simultaneously distinguishing them from other dsRNAs.

STAR★Methods

(1). Lead Contact and Materials Availability

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Yunsun Nam (yunsun.nam@utsouthwestern.edu). Reagents generated in this study are available from the Lead Contact with a completed Materials Transfer Agreement.

(2). Experimental Model and Subject Details

Bacterial strains: Expression of recombinant human DGCR8 was performed using Rosetta 2 (BL21-DE3) cells (Novagen), with genotype: F⁻ ompT hsdS_B(r_B⁻ m_B⁻) gal dcm pRARE2 (Cam^R). Cell Lines: Sf21 cells (Expression Systems) (cultured in Grace’s Supplemented Media (Expression Systems) with 10% v/v fetal bovine serum (Sigma) at 27 °C for expression of Baculovirus. HighFive cells (ThermoFisher) were cultured in ESF 921 insect cell culture medium (Expression Systems) at 27 °C for expression of Drosha/DGCR8 complexes using the baculovirus system.

(3). Method Details

Protein expression and purification

Drosha/DGCR8 complexes were expressed using the Bac-to-Bac system (Thermo Fisher Scientific) in HighFive cells. Human Drosha (353–1372, wild type sequence for active and with E1045Q/E1222Q mutations for the partially docked structure) and human DGCR8 (223–751 for fully docked, and 175–751 for partially docked structures) fragments were cloned into pFastbacDual, with a hexahistidine tag and FLAG tag, respectively. Drosha/DGCR8 complexes were purified by Ni-NTA affinity chromatography. Lysis and washing were performed using a buffer containing 20 mM Tris (pH 8.0), 1 M NaCl, 10% glycerol, and 1 mM DTT. Cation exchange chromatography was performed using a 100–800 mM NaCl gradient buffered with 20mM Bis-Tris (pH 7.0), followed by size-exclusion chromatography. Human DGCR8 (223–751) containing an N-terminal hexahistidine tag was cloned into pET21 and expressed in BL21 (DE3) Rosetta cells. DGCR8 was purified using the same strategy as for Drosha/DGCR8 complexes.

RNA transcription and purification

RNA templates were cloned into pRZ vectors containing self-cleaving ribozymes on either end to produce homogenous ends (5’ hammerhead, 3’ hepatitis delta virus), and in vitro transcribed RNA fragments were purified from denaturing PAGE, as described previously(Walker et al., 2003).

Preparation of cryo-EM samples

RNA and proteins were mixed at 2:1 ratio and dialyzed overnight at 4 °C against a dialysis buffer containing 20 mM HEPES (pH 7.1), 75 mM NaCl, 5 mM DTT, and 2 mM CaCl₂ (CaCl₂ was added for the active state only). The RNA-protein complex was purified away from extra RNA, using a Superdex200 size-exclusion column in dialysis buffer. Fractions from the RNA-protein peak were pooled and concentrated to obtain a measurement of protein concentration using A_450nm. The RNA/protein complex was diluted to 1 μM in dialysis buffer. The complex was then crosslinked by adding 3 mM DSG dissolved in DMSO and then quenched with 75 mM glycine. The samples were then purified by an additional round of SEC in dialysis buffer, and concentrated for grid preparation.

Cryo-EM data collection

Three microliters of the Drosha/DGCR8/pri-miR-16–2 complex (WT at 4.2 mg/ml, mutant at 3.7 mg/ml) were applied onto 200-mesh R3.5/1 Quantifoil grids, which were glow discharged for 35 seconds using PELCO easiGlow (TED PELLA, INC.) at a plasma current of 15 mA. The blotting paper was standard Vitrobot filter paper Ø55/20mm, Grade 595 (TED PELLA, INC.). To improve particle distribution, 0.02% Nonidet P-40 substitute and 0.05% octyl-Beta glucoside were added prior to freezing. The grids were blotted for 4 s and rapidly cryocooled in liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific) at 4 and 100% humidity. The samples were screened using a Talos Arctica cryo-electron microscope (Thermo Fisher Scientific) operated at 200 kV and then imaged in a Titan Krios cryo-electron microscope (Thermo Fisher Scientific) with GIF energy filter (Gatan) at a magnification of 130,000x (corresponding to a calibrated sampling of 1.06 Å per pixel) using Stanford-SLAC Cryo-EM Facilities. Micrographs were recorded using EPU software (Thermo Fisher Scientific) with a Gatan K2 Summit direct electron detector, where each image was composed of 30 individual frames with an exposure time of 6 s and a dose rate of 7.8 electrons per second per Ų. A total of 12,681 movie stacks for active state and 6,070 movie stacks for partially docked state were collected, with defocus ranges of −1.3 to −3 μm and −1.5 to −3.6 μm, respectively.

Image processing

All micrographs were motion-corrected using MotionCor2 (Zheng et al., 2017) and the contrast transfer function (CTF) was determined using CTFFIND4 (Rohou and Grigorieff, 2015). All particles were autopicked using the NeuralNet option in EMAN2 (Tang et al., 2007) and manually checked, yielding 1,385,678 particles from selected 12,455 micrographs for active state and 1,063,710 particles from selected 5,994 micrographs for partially docked state. The particle coordinates were imported into Relion (Scheres, 2012), where multiple rounds of 2D classification were performed to remove poor quality 2D class averages. The initial model was built using the “ab-initio 3D” program in cryoSPARC (Punjani et al., 2017). A total of 937,668 particles for active state and 842,037 particles for partially docked state were used for 3D classification in Relion to remove the poor 3D classes. Next, 505,640 particles for active state and 381,468 particles for partially docked state were imported into cryoSPARC, where the final refinements were performed using the “Non-uniform refinement” program, achieving the 3.7-Å map with mask (4.6 Å without mask) for active state and 4.4-Å map with mask (5.8 Å without mask) for inactive state. The resolution for the final maps was estimated by the 0.143 criterion of the FSC curve (See more information in Figures S2 and S6, and Table 1).

Model Building, Refinement and Validation

The crystal structures of Drosha (PDB entry 5B16) and the DGCR8 core (PDB entry 2YT4) were fitted into the density of the active state structure by rigid body fitting using Chimera (Pettersen et al., 2004). The dsRBDs of Drosha and DGCR8–2 were rearranged to fit the density. The de novo model building of RNA and Drosha (residues 459–462, 501–521, 668–674, 712–849, 930–957, and 1334–1356) was performed using Coot (Emsley et al., 2010). Additionally, the RNA building was guided by mfold secondary structure predictions (Zuker, 2003). Real-space refinement was then performed using PHENIX, with secondary structure restraints for protein as well as RNA. Model geometries were assessed using Molprobity (Chen et al., 2010). Maps and structures shown in the figures were generated with PyMOL, Chimera and Coot. After docking the Drosha crystal structure, the density near the active site was closely analyzed to identify the scissile phosphate and calcium ions. In particular, the RIIIDb active site contains larger additional density in the region typically associated with magnesium binding. This density was modelled with a calcium ion, and the final models including the calcium ion were evaluated by Q-score (Pintilie et al., 2019). To compare the RNA-protein interactions in fully docked and partially docked MP/RNA structures, the script github/pdbfairy was used. For partially docked state, the model from the active complex was rigid-body fitted into the map of the inactive complex using Chimera, further optimized in Coot, and followed by real space refinement in Phenix with geometry restraints.

In vitro pri-miRNA processing assays

5’ end-labeling was performed using T4 Polynucleotide Kinase (NEB) and γ−³²P-ATP. Unless otherwise indicated, pri-miRNA processing assays were performed in 15 μL reactions containing approximately 1 nM end-labeled RNA substrate, 30 mM Tris (pH 7.5), 67 mM NaCl, 5% glycerol, 10 mM MgCl₂, 5 mM DTT, 8 U RNase inhibitor, and 1.5 μg yeast tRNA. Unless otherwise indicated, reactions were incubated for 10 minutes (except for assays comparing arm truncations, which were incubated for 15 minutes) at room temperature, stopped with 0.7% SDS and 24mM EDTA, and treated with 0.38mg/mL proteinase K at 50 °C for 30 minutes. Reactions were analyzed by denaturing PAGE using either a phosphorimager (for end-labeled) or SYBR Green II (Thermo Fisher Scientific)(for unlabeled). For quantified reactions, the assays were performed in triplicate, and the gels were analyzed using ImageLab software (BioRad) to quantify percent processed by densitometry. Quantified results were plotted and statistically analyzed using Prism 8 (Graphpad). For sequencing gel analysis, the reactions were run on an 8% denaturing sequencing PAGE for approximately 3.5 hours at a constant current of 30 mA. For activity assays on pre-assembled RNA-protein complexes, a reaction containing 1.7 μM MP-RNA complex and MgCl₂ (concentration as indicated) was incubated at 37 °C for 20 minutes. The reactions were then processed similar to the end-labeled reactions, and visualized using SYBR Green II.

Gel-shift assay (Electrophoretic Mobility Shift Assay, EMSA)

5’ end-labeled RNA (approximately 1 nM) was incubated with a dilution series of protein samples in 20 mM Tris (pH 7.5), 67 mM NaCl, 10 mM DTT, 1000 μg/mL yeast tRNA, and 10% glycerol. The mixture was analyzed by native PAGE and visualized with a phosphorimager.

Crosslinking Mass Spectrometry (XLMS)

RNA/protein complexes were prepared as described above. RNA/protein complexes were diluted to 1 μM by measuring absorbance at 450 nm. Samples without RNA were prepared similarly but in a buffer containing 20 mM HEPES (pH 7.1), 1 M NaCl, and 5 mM DTT. 1:1 mixtures of DSS and deuterated DSS (DSS-d4, Proteochem) were prepared in DMSO. Crosslinking reactions were performed at room temperature for 10 minutes, and quenched with 75 mM glycine. The crosslinked samples were concentrated, run on SDS PAGE, and stained with Coomassie Blue. Crosslinked bands were cut and submitted for mass spectrometry analysis by the UTSW Proteomics core facility. The samples were analyzed on a Fusion Lumos (Thermo Fisher Scientific) mass spectrometer coupled to a Dionex UltiMate 3000 RSLCNano LC system. The data were analyzed using the xQuest/xProphet pipeline (Rinner et al., 2008). The results were then sorted by ID-score and false discovery rate (FDR). Links shown in the crosslink maps consist of all inter-peptide Drosha-Drosha hits with an FDR ≤ 0.05, and Id-Score ≥ 20.

(4). Quantification and Statistical Analysis

Cryo-EM validation was performed and is shown in Figures S2 and S6. Gold standard FSC plots (calculated in CryoSPARC), Euler angle distributions (calculated in CryoSPARC), Q-scores for individual residues of Drosha and RNA, and per-residue cross-correlation coefficient (calculated in Phenix) are shown for each structure.

Crosslinking mass spectrometry was performed using xQuest/xProphet, which utilizes the target-decoy strategy (Elias and Gygi, 2010) to determine the False Discovery Rate (FDR) for each hit. Only hits with FDR scores < 0.05 were considered, and three individual replicates of each experiment was performed.

Quantified in vitro processing assays were performed in triplicate. To compare the processing of each Drosha mutant compared to the wild type control, unpaired t-tests were performed.

(5). Data and Code Availability

The cryo-EM reconstructions of the Drosha/DGCR8/pri-miRNA complexes were deposited to EMDB (EMD-21051 and EMD-21052) and the Protein Data Bank (accession numbers 6V5B and 6V5C).

Supplementary Material

1. Movie S1. Transition between partially- and fully docked states, Related to Figure 5.

Cartoon representation of transition between partially- and fully docked states with RNA and CTTs of DGCR8 removed for clarity. CED is shown in blue, Belt in dark green, Wedge in light green, Drosha dsRBD in cyan, Drosha C-terminal tail in dark blue, RIIIDa/b in gray, DGCR8–1 in yellow-orange, and DGCR8–2 in deep purple.

2. Movie S2. Alternative view of transition between partially- and fully docked states, Related to Figure 5.

Cartoon representation of transition between partially- and fully docked states with back view. CTTs of DGCR8 have been removed for clarity. RNA is shown in orange, CED in blue, Belt in dark green, Wedge in light green, Drosha dsRBD in cyan, Drosha C-terminal tail in dark blue, RIIIDa/b in gray, DGCR8–1 in yellow-orange, and DGCR8–2 in deep purple.

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and Virus Strains
E. coli DH5alpha	ThermoFisher Scientific	Cat# 18258012
E. coli DH10Bac	ThermoFisher Scientific	Cat# 10361–012
Rosetta 2 (BL21-DE3)	Novagen	Cat# 70956
Chemicals, Peptides, and Recombinant Proteins
DSG	ThermoFisher Scientific	Cat# A35392
Nonidet P-40 substitute	ThermoFisher Scientific	Cat# 9002–93-1
Octyl-beta Glucoside	ThermoFisher Scientific	Cat# 28310
SYBR Green II RNA gel stain	ThermoFisher Scientific	Cat# S7564
DSS	ThermoFisher Scientific	Cat# 21655
Deuterated DSS (DSS-d4)	Proteochem	Cat# h5102 – 10mg
gamma-32P-ATP	Perkin Elmer	Cat# NEG035C001MC
Ethane	Airgas	Cat# ET R200
T4 Polynucleotide Kinase	NEB	Cat# M0201S
Ribolock RNase Inhibitor	ThermoFisher Scientific	Cat# E00381
Deposited Data
MP/RNA complex – fully docked state	This work	EMD-21051
MP/RNA complex – fully docked state	This work	PDB 6V5B
MP/RNA complex - partially docked state	This work	EMD-21052
MP/RNA complex - partially docked state	This work	PDB 6V5C
Experimental Models: Cell Lines
High Five insect cells	ThermoFisher Scientific	Cat# B85502
Sf21 insect cells	Expression Systems	Cat# 94–003F
Recombinant DNA
pET21 constructs for expression of DGCR8	This study	N/A
pET21 plasmid	Novagen (EMD Millipore)	Cat# 69770–3
pRZ constructs for in vitro RNA transcription templates	This study	N/A
pRZ plasmid	(Walker et al., 2003)	N/A
pFastBacDual constructs for Drosha/DGCR8 complex co-expression in insect cells	This study	N/A
pFastBacDual plasmid	ThermoFisher	Cat#10712024
pFastBacDual construct for coexpression of Drosha and DGCR8 in insect cells	This study	N/A
Oligonucleotides
RNA sequences (See Table S2)	This study	N/A
Software and Algorithms
PRISM 8.0 software	GraphPad	https://www.graphpad.com/scientific-software/prism/
Phenix	(Adams et al., 2010)	https://www.phenix-online.org/
Coot 0.8.9.1	(Emsley et al., 2010)	https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot
Chimera 1.13.1	(Pettersen et al., 2004)	https://www.cgl.ucsf.edu/chimera
Relion	(Scheres, 2012)	https://www3.mrc-lmb.cam.ac.uk/relion
EPU software	ThermoFisher Scientific	https://www.fei.com/software/epu-automated-single-particles-software-for-life-sciences/
MotionCor2	(Zheng et al., 2017)	https://msg.ucsf.edu/software
CTFFIND4	(Rohou and Grigorieff, 2015)	https://grigoriefflab.umassmed.edu/software_download
EMAN2	(Tang et al., 2007)	https://blake.bcm.edu/emanwiki/EMAN2
CryoSPARC	(Punjani et al., 2017)	https://cryosparc.com/
MolProbity	(Chen et al., 2010)	http://molprobity.biochem.duke.edu/
PyMOL	Schrodinger	https://pymol.org/2/
xQuest/xProphet	(Rinner et al., 2008)	http://prottools.ethz.ch/orinner/public/htdocs/xquest/
Q-score	(Pintilie et al., 2019)	https://cryoem.slac.stanford.edu/ncmi/resources/software/mapa
Image Lab	Bio-Rad	N/A
Other
200-mesh R3.5/1 Quantifoil grids	EMS	Cat# Q250CR-35
PELCO easiGlow Glow Discharge machine	Ted Pella	Cat# 91000
Vitrobot filter paper ∅55/20mm, Grade 595	Ted Pella	Cat# 47000–100
Vitrobot Mark IV	ThermoFisher Scientific	N/A
Talos Arctica G2	ThermoFisher Scientific	N/A
Titan Krios G3	ThermoFisher Scientific	N/A
GIF Energy Filter	Gatan	N/A

Highlights:

• Drosha and DGCR8 together form a “double-dsRBD ruler” to measure the stem length

• Drosha Belt and Wedge clamp the ssRNA to ensure efficient and accurate processing

• pri-miRNA GHG motif is recognized through the impact on the RNA structure via Wedge

• HBRs and dsRBDs drive complex assembly with high affinity and intrinsic specificity