The Core Microprocessor Component DiGeorge Syndrome Critical Region 8 (DGCR8) Is a Nonspecific RNA-binding Protein*

Background: Double-stranded RNA-binding domain-containing proteins play key roles in microRNA biogenesis. Results: The Microprocessor protein DGCR8 binds RNA targets nonspecifically. Conclusion: DGCR8 alone is not responsible for specific RNA target recognition by the Microprocessor. Significance: Faithful RNA processing is not causatively coupled to specific substrate binding properties of DGCR8. MicroRNA (miRNA) biogenesis follows a conserved succession of processing steps, beginning with the recognition and liberation of an miRNA-containing precursor miRNA hairpin from a large primary miRNA transcript (pri-miRNA) by the Microprocessor, which consists of the nuclear RNase III Drosha and the double-stranded RNA-binding domain protein DGCR8 (DiGeorge syndrome critical region protein 8). Current models suggest that specific recognition is driven by DGCR8 detection of single-stranded elements of the pri-miRNA stem-loop followed by Drosha recruitment and pri-miRNA cleavage. Because countless RNA transcripts feature single-stranded-dsRNA junctions and DGCR8 can bind hundreds of mRNAs, we explored correlations between RNA binding properties of DGCR8 and specific pri-miRNA substrate processing. We found that DGCR8 bound single-stranded, double-stranded, and random hairpin transcripts with similar affinity. Further investigation of DGCR8/pri-mir-16 interactions by NMR detected intermediate exchange regimes over a wide range of stoichiometric ratios. Diffusion analysis of DGCR8/pri-mir-16 interactions by pulsed field gradient NMR lent further support to dynamic complex formation involving free components in exchange with complexes of varying stoichiometry, although in vitro processing assays showed exclusive cleavage of pri-mir-16 variants bearing single-stranded flanking regions. Our results indicate that DGCR8 binds RNA nonspecifically. Therefore, a sequential model of DGCR8 recognition followed by Drosha recruitment is unlikely. Known RNA substrate requirements are broad and include 70-nucleotide hairpins with unpaired flanking regions. Thus, specific RNA processing is likely facilitated by preformed DGCR8-Drosha heterodimers that can discriminate between authentic substrates and other hairpins.

miRNAs 2 are a conserved class of ϳ22-nucleotide (nt) noncoding RNAs that regulate gene expression by base pairing with target mRNA sequences. They play key roles in modulating cellular and developmental processes (1,2) and have been implicated in a broad range of metabolic, immunological, psychiatric, and cell cycle disorders (3)(4)(5)(6). Mature miRNAs are generated through a conserved succession of processing steps. Typically, miRNA biogenesis begins in the nucleus with the generation of a long pri-miRNA transcript by RNA polymerase II. A hairpin harboring the pri-miRNA is selected for processing by the nuclear RNase III Drosha, which liberates the miRNA-containing pre-miRNA. Exportin-5 transports the pre-miRNA to the cytoplasm where it undergoes a second round of processing (7,8). Another RNase III, Dicer, trims the terminal loop to generate a 21-24-nt RNA duplex bearing 2-nt 3Ј overhangs (9).
Processing of pri-and pre-miRNAs by members of the RNase III family is facilitated by accessory proteins comprising multiple dsRBDs featuring characteristic ␣-␤-␤-␤-␣ motifs. In humans, DGCR8 is required for Drosha processing and contains tandem dsRBDs (10). DGCR8 interacts with ferric heme as a cofactor to increase processing activity (11)(12)(13), and although the cooperative formation of oligomeric DGCR8/pri-miRNA assemblies has been described as another positive modulator of pri-miRNA processing (14), lysine acetylation within the tandem dsRBDs decreases the affinity of DGCR8 for pri-miRNA thereby down-regulating processing (15).
Two modes of pri-miRNA recognition have been presented. In one model, the Microprocessor binds a large (Ͼ10-nt) terminal loop to position Drosha's catalytic center roughly two helical turns (ϳ22-nt) from the stem-loop junction, resulting in the liberation of the ϳ70-nt pre-miRNA (16). A more recent model suggests that the terminal loop is dispensable for Microprocessor recognition. Instead, DGCR8 recognizes the hairpin's basal ss-dsRNA junction and recruits Drosha for cleavage (17). However, the generality of this model is unclear because further experiments revealed that the length and symmetry of 5Ј-and 3Ј-nonstructured ssRNAs required for efficient processing are variable and pri-miRNA-dependent (18). Most Caenorhabditis elegans pri-miRNAs apparently lack determinants required for processing in human cells, yet one-fifth of all human pri-miRNAs lack primary sequence determinants such as downstream SRp20 binding, basal UG, and the apical GUG motifs, recently described by Bartel and co-workers (19). Collectively, this leaves 70-nt hairpins with unpaired flanking regions as the only feature and common denominator across Microprocessor substrates.
Although most dsRBDs have been reported to bind duplex RNA as the name implies, unpaired ssRNA in loops, bulges, and mismatched pairs can also be recognized (20 -23). This flexibility in substrate recognition presents a challenge in identifying both protein and RNA features required for specific protein-RNA interaction in the absence of structural data. A recent genome-wide view of DGCR8 function employing RNA crosslinking immunoprecipitation sequencing studies revealed a much broader RNA target base than anticipated (24). Specifically, these results showed that DGCR8 targets hundreds of mRNAs in addition to the anticipated interactions with pri-miRNA substrates. Implications for DGCR8 function ranged from roles in small nucleolar RNA and long noncoding RNA stabilization to regulation of alternatively spliced isoforms.
In this study, we investigated the RNA binding properties of the integral Microprocessor component DGCR8. Along with ssRNA and perfectly paired dsRNA substrates, we designed a comprehensive series of pri-mir-16-1 variants to elucidate the role of RNA secondary structure elements in DGCR8-RNA in general as well as the initiating event of pri-to pre-miRNA processing. These mutants were also used to investigate the link between recognition and processing of pri-miRNA substrates. Because detailed structural information on DGCR8⅐pri-miRNA complexes remains elusive, we utilized NMR spectroscopy to elucidate the interaction at single residue resolution and diffusion measurements to probe complex formation in solution. Although DGCR8 in isolation did not discriminate between the broad array of RNAs tested, functional processing assays confirm the specificity of Microprocessor assemblies. Thus, our results point to an intricate relationship between DGCR8 and Drosha in which both proteins are required for binding and processing, rather than a hierarchical DGCR8-mediated pri-miRNA recognition and subsequent Drosha recruitment model.

EXPERIMENTAL PROCEDURES
RNA Transcription and Purification-Pri-miRNA-containing plasmids were purified with the Omega Plasmid Giga kit, linearized with the appropriate 3Ј restriction endonuclease, phenol/chloroform-extracted, and ethanol-precipitated. The minimal pri-mir-16-1 construct includes 18 nt upstream of the 5Ј-and 16 nt downstream of the 3Ј-cleavage site. Although two additional Gs were added to the 5Ј overhang for efficient transcription, an additional AUUU sequence was added to the 3Ј overhang after linearization employing SwaI (New England Biolabs). In vitro reactions were performed in transcription buffer (40 mM Tris, pH 8.0, 1 mM spermidine, 10 mM dithiothreitol (DTT), 0.01% Triton X-100) containing 0.05 g/ml plasmid template, 30 mM MgCl 2 , 1 M T7 RNA polymerase and 6.5 mM each ATP, CTP, GTP, and UTP. Transcription reactions were incubated at 37°C for 4 h. Transcripts were purified as described previously (25). Briefly, the RNA was exchanged to buffer containing 10 mM NaPO 4 , pH 6.5, 25 mM KCl, 0.5 mM EDTA, and 25 M NaN 3 . Full-length transcripts were separated from the protein, unincorporated nucleotides, and abortive transcripts by anion exchange chromatography (GE HiTrapQ HP).
Protein Expression and Purification-N-terminal, tobacco etch virus protease-cleavable His 6 -GB1/DGCR8 ⌬275 , DGCR8 PC , DGCR8 core , and DGCR8 RBD1 (residues 509 -582) sequences were transformed into Escherichia coli BL21-CodonPlus (DE3)-RIPL competent cells, and expression was induced with 0.5 mM isopropyl ␤-D-thio-galactoside for 2-4 h at 37°C (DGCR8 ⌬275 cultures were supplemented with 1 mM ␦-aminolevulinic acid) (13) and purified with a HisTrapFF affinity column per the manufacturer's instructions. Partially purified fusion proteins were cleaved with tobacco etch virus protease overnight at room temperature and reapplied to the HisTrap column. The purified protein was then exchanged into buffer appropriate for NMR analysis or biochemical assays using a PD-10 desalting column. Cultures of E. coli BL21-CodonPlus (DE3)-RIPL cells expressing DGCR8 core and DGCR8 RBD1 (residues 509 -582) were grown from a single colony in 2 ml of LB/Amp 100 at 37°C overnight. These cultures were then adapted to growth in 500 ml of 100% D 2 O-15 N, 13 C-enriched M9 minimal media (26) with a succession of small volume cultures grown at 37°C with 200 rpm agitation. Expression was induced with 1 mM isopropyl ␤-D-thio-galactoside for 4 h at 37°C and purified as described previously (27). Identical batches of DGCR8 protein variants were used in successive binding and processing assays.
Microfiltration Double-filter Binding Assays-Filter binding was performed essentially as described in Arraiano et al. (28). Briefly, in vitro transcribed, purified RNA was end-labeled with ␥-32 P and then re-purified by urea-PAGE and gel extraction. The labeled RNA stocks were quantified and diluted to 30 M. [ 32 P]RNA was diluted to 200 pM (2ϫ stock) with microfiltration buffer (20 mM Tris, pH 7.5, 50 mM NaCl, 1 mM DTT). The RNA was annealed by incubating at 65°C for 5 min and then ice for 3 min. Purified protein for filter binding was prepared as 2ϫ stocks by serial dilution in 1ϫ microfiltration buffer. Equal volumes of 2ϫ RNA and protein stocks were combined to yield 50-l binding reactions. The reactions were incubated at room temperature for 1 h.
The binding reactions were filtered through nitrocellulose and nylon membranes sequentially using the BioDot microfiltration apparatus. Sample wells were washed with 2ϫ 100 l of microfiltration buffer before and after sample loading. Membranes were air-dried, and the intensity of the spots was quantified by a phosphorimager. The data were fit using a nonlinear least sum of squares analysis with Equation 1,

DGCR8 Binds RNA Nonspecifically
where FB is the fraction of RNA bound to protein; FB min is the minimum fitted value of FB; FB max is the maximum fitted value of FB; [DGCR8] is the molar concentration of protein, and n is the fitted value of the slope of the curve. Electrophoretic Mobility Shift Assay (EMSA)-Purified pri-miRNA transcripts were ␥-32 P-end-labeled and purified as described previously (29). A 20ϫ pri-miRNA stock was heated to 65°C for 5 min and annealed on ice for 3 min. DGCR8 variants were diluted to the appropriate concentration as 20ϫ stocks with EMSA reaction buffer. EMSA samples were prepared in 20 l in reaction buffer containing 50 mM Tris-HCl, pH 7.5, 200 mM NaCl, 5% glycerol, 0.1% Triton X-100, and 0.2 nM pri-miRNA. Reactions were incubated for 1 h at 37°C and resolved on a 6% nondenaturing polyacrylamide gel in 1ϫ TBE.
RNA Secondary Structure Determination by NMR-All NMR spectra were recorded at 288 K or 298 K on Bruker Avance 800 or 850 MHz spectrometer equipped with a TCI cryoprobe. NMR experiments were performed on samples of 500-l volume containing 0.1-1.6 mM of pri-mir-16 RNA variants. Data were processed using nmrPipe (30) and analyzed using Sparky (31). One-dimensional imino proton spectra were acquired using a jump-return echo sequence (32). Imino proton resonances were assigned sequence-specifically from water flipback two-dimensional NOESY spectra ( mix ϭ 200 ms) (33). Assignments were confirmed using a jump-return (32) 15 N-HMQC (34). Elucidation of base pairing and secondary structure was verified from the 2 J NN coupling through hydrogen bonds in the TROSY-based HNN-COSY data (35,36).
The pri-mir-16 lower RNA tertiary structure was predicted using the MC-sym software (37). All information required to generate and edit the scripts are found on line on the MC-pipeline website. Briefly, all-atom models for the 65-nucleotide primir-16 lower satisfying the NMR-derived secondary structure were generated. 132 structures were clustered using MC-Sym, and five representative structures were selected. AMBER energy minimization was employed to refine the final five atomistic three-dimensional structures as implemented in MC-sym.
DGCR8 NMR Analysis-Backbone 1 H N , 15 N, 13 CO, 13 C␣, and 13 C␤ chemical shift assignments for DGCR8 core have been obtained using a strategy based on three-dimensional triple resonance experiments employing transverse relaxation-optimized spectroscopy (TROSY), and chemical shifts have been deposited in the BioMagResBank under accession number 17773 (27). Standard three-dimensional NMR techniques (38) were employed to obtain and confirm backbone assignments for DGCR8 RBD1 (residues 509 -582) (39). Line width and crosspeak intensities of 1 H, 15 N-TROSY experiments at varying protein/RNA ratios were determined using Sparky, assuming Gaussian line shapes (31).
Measurement of DGCR8-pri-miRNA Diffusion Constants by PFG NMR-All PFG-NMR experiments were recorded at 298 K on a Bruker Avance III spectrometer operating at 600 MHz and equipped with a 5-mm QXI quadruple resonance ( 1 H, 19 F-13 C, 31 P) room temperature probe with triple-axis shielded gradients. The bipolar pulse pair longitudinal eddy current delay (BPPLED) sequence (40,41) was used, and one-dimensional spectra were recorded after two-axis gradient encoding and decoding separated by a diffusion delay of 100 ms. Dephasing of the NMR signal in the BPPLED experiment was analyzed according to Equation 2 (40,42), where A(g) is the measured intensity of the NMR signal as a function of the fractional gradient strength g; A(g 0 ) is the peak intensity of the reference spectrum (g 0 ϭ 2%); D is the translational diffusion constant; ␥ H is the gyromagnetic ratio of a proton (2.675197 ϫ 10 4 G Ϫ1 ⅐s Ϫ1 ); ␦ is the gradient duration; G max is the combined maximum power (87.8 G⅐cm Ϫ1 at g ϭ 100%) offered by the x and z gradients; ⌬ is the time between gradients, and the recovery time. By taking the natural logarithms of both sides of Equation 2, a linear equation results, and D can be calculated from least squares linear fits. To exclude complications with any exchangeable protons, the diffusion analysis of deuterated protein samples was performed by integrating over the 0.35-1.05 ppm section of the methyl region of the spectrum (primarily CH 2 D and CHD 2 isotopomers). RNA samples were analyzed by integrating over the 7.3-8.0 ppm section. The diffusion rate of a molecule depends on its size and shape, its concentration, the temperature, and solvent viscosity.
The Stokes-Einstein equation shows that a sphere's diffusion coefficient (D) is inversely related to the hydrodynamic radius (R H ) and solvent viscosity () as shown in Equation 3, where k B is the Boltzmann constant, and T is the temperature (298 K).
Obtaining the hydrodynamic radius R H for a spherical particle using Equation 3 requires an accurate measure of the solvent viscosity . All samples were dissolved in identical aqueous (90: 10% H 2 O/D 2 O) buffer solutions containing 150 mM KCl. Using viscosities of 1.097 and 0.8929 kg⅐cm Ϫ1 ⅐s Ϫ1 for a 100% D 2 O and a 100% H 2 O solution (43), the viscosity of a mixture of light and heavy water can be represented by a linear function of concentration (44) shown in Equation 4, which yields 0.91331 kg cm Ϫ1 s Ϫ1 at 298 K. A salt effect correction on viscosity was performed according to the Jones-Dole Equation 5 (45), where c is the molar salt concentration; 0 is the uncorrected and ϭ 0.9018 kg cm Ϫ1 s Ϫ1 the corrected solvent viscosity with A ϭ 0.0057 and B ϭ Ϫ0.0147 for KCl (46).
Fractional populations of monomer and dimer states can be estimated using Equation 6, where D M and D D are the respective diffusion coefficients for monomer and dimer and f D is the molar fraction of the dimeric species.
Size Exclusion Chromatography (SEC)-SEC characterization of 150 M DGCR8 core was performed in NMR buffer (20 mM NaPO 4 , pH 7.0, 150 mM KCl, 5 mM tris(2-carboxyethyl) phosphine) at 12°C with a Superdex75 10/300 column at a flow rate of 0.5 ml/min. Conalbumin, ovalbumin, carbonic anhydrase, GB1, and aprotinin were used to generate the standard curve from which the apparent DGCR8 core molecular weight was estimated.
pri-miRNA Processing Employing an Immunoprecipitated Microprocessor-In vitro processing assays were performed essentially as described by Lee and Kim (29) with the following modifications: c-Myc-Drosha and FLAG/HA-DGCR8 (47) were co-transfected (Lipofectamine, Invitrogen) in the mammalian cell line AD-293 (Stratagene), and Drosha/DGCR8 was co-immunoprecipitated with either EZview anti-c-Myc affinity gel (Sigma) or agarose-conjugated anti-HA monoclonal antibody (Sigma). Immunoprecipitated Microprocessor was incubated with 32 P-pri-miRNA, in the presence of Mg 2ϩ and RNaseOUT (Invitrogen) for 90 min at 37°C. RNA was purified by phenol/chloroform and resolved on a 12% denaturing PAGE. Gels were dried and analyzed by phosphorimaging.
For Microprocessor reconstitution assays, the Drosha immunoprecipitate was washed with a basic, high salt buffer to remove co-purified proteins. Stringent washes were performed with 20 mM Tris, pH 8.0, 2.5 M NaCl, 5 mM EDTA, 1% Triton X-100, 0.1% SDS, 0.1% deoxycholic acid, and 1 M urea. A final wash with 20 mM Tris, pH 8.0, 100 mM NaCl was performed. For the processing assays, c-Myc-Drosha was preincubated with recombinant DGCR8 ⌬275 in the presence of Mg 2ϩ and RNase OUT (Invitrogen) for 10 min at room temperature. After addition of [ 32 P]RNA, samples were incubated at 37°C for 90 min. RNA was purified, and samples were processed as described above.
DGCR8 and Drosha Co-immunoprecipitation-Whole cell extracts of AD293 mock-transfected or transfected with c-Myc-Drosha/FLAG-DGCR8 were obtained as described above. Supernatant from lysed cells was incubated with either RNase inhibitor or RNase A/T1 mix (Ambion). After incubation for 60 min at 4°C, EZ view c-Myc beads (Sigma) were added and incubation proceeded for another 16 h at 4°C. Beads were washed with lysis buffer and resuspended in SDS-loading buffer. Proteins were subjected to SDS-PAGE, and Western blot analysis was performed with anti-c-Myc or anti-FLAG antibodies (Sigma).

RESULTS
pri-miRNA Recognition by DGCR8-To investigate the secondary structure elements required for pri-miRNA recognition by DGCR8, we performed quantitative binding studies using a series of pri-mir-16-1 variants. Along with the 104-nt hairpin that includes single-stranded regions at both the base and terminal loop, we generated a terminal loop mutant (pri-mir-16⌬loop) by replacing 14 nucleotides of the loop with a G:C-C:G clamp and UUCG tetraloop. We also created a pri-mir-16 mutant lacking single-stranded 5Ј and 3Ј overhangs (pri-mir-16⌬ss) and a pri-mir-16⌬⌬ mutant that incorporates both modifications (Fig. 1A). Each pri-miRNA secondary structure was confirmed by NMR spectroscopy monitoring imino proton resonances (Fig. 2).
To complement the pri-mir-16 constructs, we produced three recombinant DGCR8 variants to be tested in our binding studies, DGCR8 ⌬275 , DGCR8 PC , and DGCR8 core (Fig. 1B). DGCR8 ⌬275 lacks the first 275 amino acids but retains WW and heme-binding domains, two dsRBDs and the Drosha interaction/trimerization domain and is competent for both dimerization and pri-miRNA processing (13,14,48). DGCR8 PC includes residues 484 -750 and represents the processing-competent (PC) core (48). DGCR8 core (residues 493-706) lacks the Drosha interaction domain but retains the tandem dsRBDs and is therefore unable to participate in pri-miRNA processing but is RNA binding-competent (14,49).
To test whether pri-mir-16 mutants contain enough ssRNA in internal loops or bulges to attract DGCR8, we have modified the pri-mir-16⌬⌬ construct to produce a fully paired hairpin (pri-mir-16fp) by altering unpaired and bulged nucleotides to effectively eliminate ssRNA and maximize duplex stability (Fig.  1A). DGCR8 ⌬275 was still able to bind pri-mir-16fp with affinity and cooperativity equivalent to that of pri-mir-16-1, ⌬loop, ⌬ss, and ⌬⌬ (Fig. 3F). To rule out the possibility that DGCR8 ⌬275 recognition of pri-mir-16-1 is driven by a sequence determinant common to all of our RNA substrates, we tested pri-mir-16x (Fig. 1A), a Quikfold (50)-predicted 82-nt hairpin co-transcribed with mir-16-1 and mir-15a on the DLEU2 lincRNA (long intergenic noncoding RNA) (51,52). Located 397 nt upstream of mir-16-1, pri-mir-16x is overlooked by the Microprocessor. As shown in Fig. 3G, filter binding analysis revealed that DGCR8 ⌬275 binds pri-mir-16x with no discernable loss of affinity or cooperativity relative to the other RNAs tested. To test whether the lack of significant variation in our results could be explained by DGCR8 ⌬275 recognition of predominantly dsRNA elements, we produced an 87-nt ssRNA principally composed of nine tandem CCCCUAAA repeats (Fig. 1A). NMR analysis of exchangeable imino protons confirmed that no sta-ble secondary structure was formed in this transcript (data not shown). Finally, we also generated a pri-let-7b transcript to serve as a second bona fide pri-miRNA substrate. Again, DGCR8 ⌬275 bound both constructs with K d and Hill values consistent with other RNA substrates (Fig. 3, H and I).
Similarly, binding affinities and cooperativity measurements remained consistent when we analyzed DGCR8 core and a subset of the pri-miRNA variants, even though the affinities were reduced ϳ10-fold for DGCR8 core , compared with DGCR8 ⌬275 (Table 1). Collectively, our filter binding experiments demonstrate that DGCR8 recognizes the tested RNA variants with high affinity and Hill coefficients ranging from 0.9 to 1.8. None of the DGCR8 constructs tested in filter binding demonstrated a pronounced tendency to bind cooperatively beyond a dimer model.
To directly detect oligomeric DGCR8-RNA complex formation, we performed electrophoretic mobility shift assays (EMSA) with DGCR8 ⌬275 , DGCR8 PC , or DGCR8 core and the pri-mir-16 transcript. As seen in Fig. 4A, progressively more truncated variants of DGCR8 all populate similar heterogeneous complexes with pri-mir-16-1. Additionally, DGCR8/pri-miRNA interactions reveal a "smearing" effect rather then discrete band formation, which could indicate that DGCR8 binds pri-miRNA transiently and with variable stoichiometry, sampling different regions of the substrate until excess protein effectively saturates all potential binding sites on the RNA at any given time. EMSA titrations of DGCR8 core into radiolabeled pri-mir-16-1 in the presence of varying amounts of nonspecific competitor yeast tRNA Phe revealed that increasing amounts of tRNA increase the apparent K d values for the DGCR8core/pri-mir-16-1 complex (Fig. 4B). Thus, on the basis of our EMSA results, DGCR8's mode of binding resembles other dsRBDs (such as protein kinase R (PKR) (53)) that interact with RNA through transient and/or noncooperative interactions featuring multiple, potentially overlapping RNA-binding sites with comparable affinities. Taken together, our binding data imply that DGCR8 alone is not able to distinguish miRNA-containing hairpins from other RNA species but rather retains high, indiscriminate affinity for RNA within its dsRBDs.
DGCR8/pri-miRNA Binding Monitored by NMR-In the interest of reducing overall molecular weight for NMR studies, we used a 65-nt version of pri-mir-16-1 (pri-mir-16 lower , Fig.  1A) that is correctly processed in vitro (data not shown) (17). Using NMR spectroscopy, we have determined that the secondary structure of pri-mir-16 lower includes 5Ј-and 3Ј-ssRNA overhangs, a 21-nt stem featuring tandem G-A mismatches and a C-A mismatch, and a stable 8-nt loop (Fig. 5).
Our recent NMR analysis confirmed that our purified DGCR8 core contains all of the secondary structure features of the reported crystal structure (Fig. 7A) (27, 49). Furthermore, we observed pronounced chemical shift differences between the backbone amide 1 H, 15 N-correlations of an isolated RBD1 domain and those of the RBD1-RBD2-containing DGCR8 core construct (Fig. 6). This demonstrates that the two tandem RBD domains interact with each other in solution confirming earlier reports based on the crystal structure (49) and MD simulations (54).

DGCR8 Binds RNA Nonspecifically
amounts of RNA. pri-mir-16 lower was titrated into 250 M 2 H, 15 N-DGCR8 core to produce protein/RNA solutions ranging from 48:1 to 1:2 molar ratios. We observed differential broadening of backbone amide 1 H, 15 N correlations evident even at low concentrations of pri-mir-16 lower (Fig. 7). Because binding and dissociation between DGCR8 core and pri-mir-16 lower occurred on the intermediate time scale, 1 H, 15 N correlation experiments do not reveal discrete resonances for RNAbound and free states commonly observed for systems in fast or slow exchange. Thus, our protein-RNA interface analysis focused on differential broadening at a molar ratio where most protein resonances were still observed (24:1 ratio of DGCR8 core /pri-mir-16 lower ).
Differences in amide proton LW (⌬LW) upon RNA binding were calculated by subtracting the LW bound of a given residue in the presence of sub-stoichiometric RNA from the LW free observed in the absence of RNA (Fig. 8A). A value of zero indicates no change in LW. The ⌬LW values for infinitely broadened resonances (Fig. 8A, gray bars) were calculated assuming a conservative LW bound of 50 Hz, which corresponds to the largest 1 H-TROSY LW bound observable at 800 MHz. The residues with the largest change in line width (most negative ⌬LW) values map to the first and second helices of the each dsRBD motif (RBD1:␣1 and ␣2, RBD2:␣5 and ␣6) (Figs. 7 and 8B). This pat-tern of RNA-dependent line broadening correlates with previously characterized canonical RBD interactions in which approximately one helical turn of a successive minor and major groove dsRNA is occupied by the ␣-helical regions of the ␣-␤-␤-␤-␣ dsRNA-binding domain (20,55). However, no significant broadening is detected for either the ␤2-␤3 loop located in RBD1 or the ␤5-␤6 loop in RBD2, which one would predict to contact an adjacent minor groove in a canonical dsRBD-dsRNA interaction (Figs. 7 and 8B). Instead, noticeable broadening occurs within the ␤4, ␤7, and ␣7 regions. For its part, ␣7 has been shown to stabilize the DGCR8 core through extensive intramolecular contacts with both dsRBDs (49,54). Therefore, line broadening in this region is not likely attributed to direct association with pri-mir-16 lower but is probably reporting on conformational exchange on an intermediate time scale between bound and free DGCR8 core forms. Ultimately, line broadening analysis of our DGCR8 core /pri-mir-16 lower titration suggests some variation in archetypal dsRNA binding for both RBD subunits of the DGCR8 core where three structural elements of each dsRBD, the N-terminal ␣-helix, the loop between ␤1 and ␤2, and the C-terminal second helix, engage adjacent minor grooves and the intervening groove of dsRNA (56).
Finally, we tested whether protein resonances, which are broadened beyond detection at substoichiometric RNA ratios,  would sharpen as a result of changes in exchange regime brought about by the formation of specific, stable interactions in the presence of excess pri-mirR16 lower . Over the course of the titration and at the end point ratio 1:2 DGCR8 core /pri-mir-16 lower , the vast majority of assigned backbone amide resonances was broadened beyond detection. The remaining 30 residues experienced negligible chemical shift perturbations and map primarily to the flexible linker between RBD1 and RBD2 (residues Asp-579 to Glu-594), the N terminus, including strand ␤1 (residues Gln-495 to Ile-505), and the C terminus of the protein (residues Ser-702 to Val-706) (Figs. 7 and 8C). Overall, our findings indicate that DGCR8 core exchanges between and among pri-mir-16 lower -binding sites on the intermediate chemical shift time scale leading to differential line broadening. Such a model was further corroborated by a titration revealing global imino proton line broadening of pri-mir-16 lower in the presence of increasing amounts of DGCR8 core (Fig. 9A). Divalent cations often stabilize important RNA tertiary structures. Therefore, titrations were repeated in the presence of 5 mM MgCl 2 . Imino proton chemical shift changes observed for pri-mir-16 lower and larger RNA variants (Fig. 1A) in the presence of Mg 2ϩ -cations were invariably small and did not exceed 0.04 ppm (data not shown). Moreover, intermediate exchange kinetics prevailed in the presence of Mg 2ϩ cations (Fig. 9B). The NMR titration data also suggest that the structure of DGCR8 core may not be significantly perturbed by the pri-mir-16 lower interaction because chemical shifts of resonances observable during the titration are mostly unchanged.
DGCR8/pri-miRNA Binding Monitored by PFG-NMR-A distinct advantage of PFG-NMR methods is that translational self-diffusion coefficients can be determined under experimen-tal conditions (protein and RNA concentration, temperature, and viscosity) identical to those used for collecting the NMR titration data described above. Because diffusion measurements are sensitive to changes in molecular size and shape, these experiments provide a supplementary characterization of DGCR8/pri-mir-16-1 association to complement our line broadening analysis. The results of PFG-NMR experiments are reported in Table 2. A single (average) diffusion coefficient was observed in all experimental measurements of free DGCR8 protein variants or pri-mir-16 lower , as evident from linear correlations with R values Ն0.996 (Fig. 10A).
Possible explanations for a slower diffusion rate (8.7 Ϯ 0.01 ϫ 10 Ϫ11 m 2 ⅐s Ϫ1 ) of the 214-residue DGCR8 core (larger apparent R H ) in comparison with calculated values (9.8 ϫ 10 Ϫ11 m 2 ⅐s Ϫ1 ) include differences between the solution structure and the crystal structure used for the prediction, intermolecular oligomerization, and/or the presence of unfolded protein regions. In the case of intermolecular oligomerization, one would expect the slower diffusion to be caused by partial dimerization. Assuming exchange averaging between a monomer and a dimer on the diffusion time scale, we calculated a 47% dimer population of the DGCR8 core . In good agreement with results previously published (49), our SEC analysis confirmed an estimated molecular mass of 31 kDa for the DGCR8 core , significantly larger than the theoretical molecular mass of 24.2 kDa (data not shown). The overall agreement of experimental and calculated translational diffusion coefficients and hydrodynamic radii for the DGCR8 variants and pri-mir-16 lower in isolation confirms that physically meaningful data could be obtained for macromolecular DGCR8⅐pri-mir-16 complexes. FIGURE 5. NMR imino assignments and secondary structure determination of pri-mir-16 lower . A, sequential imino-imino proton NOE assignment paths are shown by different colors for A-form helical stems I (black), II (magenta), and III (cyan). The spectrum is labeled with assignment information. B, NMR-based secondary structure of the lower stem of pri-mir-16 lower . The secondary structure representation was generated using PseudoViewer (74); pri-mir-16 lower numbering is consistent with full-length pri-mir-16-1 (Fig. 1).
For the DGCR8 core ⅐pri-mir-16 lower complex, diffusion rates of the protein in the presence of excess RNA (1:2 molar ratio) revealed a deviation from linearity (R ϭ 0.990) (Fig. 10B), suggesting a polydisperse system described by Equation 2. Linear correlations obtained after subtracting the contributions from the slowly diffusing complex (i.e. elimination of last five points) or the faster diffusing component (i.e. elimination of first five points) yielded upper and lower boundaries of D ϭ 10.1 Ϯ 1.02 ϫ 10 Ϫ11 m 2 ⅐s Ϫ1 with (R ϭ 0.999) and D ϭ 6.4 Ϯ 0.15 ϫ 10 Ϫ11 m 2 ⅐s Ϫ1 (R ϭ 0.999). Extrapolated to infinite dilution, Stokes radii of R H ϭ 24.1 Ϯ 2.6 Å and 38.1 Ϯ 0.9 Å can thus be estimated for the smallest and largest diffusing spherical particles for the DGCR8 core ⅐pri-mir-16 lower complex exchanging on the diffusion time scale. The smallest diffusing molecule is likely unbound DGCR8 core given the close agreement with the predicted R H calc value of 24.7 Å. Recombinant DGCR8 ⌬275 Is an Active Microprocessor Component-We performed in vitro Microprocessor assays that mimic the first step in miRNA biogenesis to examine the relationship between observed DGCR8/pri-miRNA binding patterns and processing. When incubated with immunopre-cipitated Microprocessor, the pri-mir-16-1 and 16⌬loops were efficiently processed to release respective 65-and 59-nt pre-miRNAs as well as the 20-nt 5Ј and 3Ј reaction by-products (Fig.  11A). However, pri-mir-16 variants lacking ss-flanking RNA (16⌬ss and 16⌬⌬) were not cleaved by the Microprocessor, in agreement with previously published work (14,17,18). In addition, no cleavage products were observed when pri-mir-16x was used in the processing assay. These data demonstrate that only a distinct subgroup, including pri-mir-16-1 and 16⌬loop, is able to productively engage the Microprocessor, even though all RNA constructs display similar binding by DGCR8.
To address whether DGCR8's RNA specificity determinant could reside in the truncated portion of the DGCR8 ⌬275 constructs, we have performed in vitro processing with a DGCR8 ⌬275 -reconstituted Microprocessor. c-Myc-Drosha alone did not efficiently process pri-mir-16-1, 16⌬ss, or 16⌬loop (Fig. 11B). However, when incubated with purified recombinant DGCR8 ⌬275 , 5Ј cleavage products were observed for pri-mir-16-1 and 16⌬loop but not 16⌬ss, confirming that recombinant DGCR8 ⌬275 is a properly folded, active component of a highly specific in vitro pri-miRNA processing complex.  15 N chemical shift differences of DGCR8 RBD1 (residues 509 -582) and DGCR8 core (residues 493-706). Backbone ribbon color of the DGCR8 crystal structure (PDB code 2YT4) varies between white (⌬␦ ϭ 0 ppm) and red (⌬␦ Ն0.2 ppm). Residues Val-581 and Lys-582 (⌬␦ Ն 0.6 ppm) at the very C-terminal end of the RBD1 construct have been omitted for clarity. Side chains of residues with ⌬␦ Ն0.2 ppm are shown. The location of a putative salt bridge involving Asp-549 and Lys-659 (italic) inferred from the distances observed in the x-ray structure is indicated using arrows. B, backbone amide 1 H N and 15 N chemical shift differences of DGCR8 RBD1 (residues 509 -582) in comparison with DGCR8 core (residues 493-706) as a function of residue number. Secondary structure elements are numbered and highlighted using gray boxes. SEPTEMBER 13, 2013 • VOLUME 288 • NUMBER 37

JOURNAL OF BIOLOGICAL CHEMISTRY 26793
Taken together, our binding and processing data strongly suggest that DGCR8 alone is not the specificity determinant for recognition of bona fide miRNA-containing hairpins. Instead, DGCR8 works in concert with Drosha to promote specific recognition and processing of pri-miRNA transcripts.
DGCR8-Drosha Co-immunoprecipitation-In the assumption that the DGCR8-Drosha complex provides the specificity for pri-miRNA recognition, it is necessary that DGCR8 and Drosha can form heterodimers prior to pri-miRNA binding. To probe for the prerequisite heterodimer, a co-immunoprecipitation assay was performed in the absence or presence of RNases. Western blot analysis of Drosha-immunoprecipitated complexes revealed the presence of DGCR8 in extracts treated with either RNase inhibitor or RNase A/T1 mix (data not shown). This observation confirms earlier reports by Han et al. (10) and suggests that association of Drosha and DGCR8 does not depend on the availability of RNA substrate molecules.

DISCUSSION
Accessory dsRNA-binding proteins involved in small RNA pathways have been shown to be required for efficient and accurate modulation of specific RNA species, when part of mul-tiprotein complexes (10,57,58). Despite the importance and characterized functions of endonuclease-associated dsRBDcontaining proteins in miRNA biogenesis, few reports have described the detailed mechanistic and structural nature of their interactions with RNase IIIs or specific RNA substrates (59).
Previous results from filter binding experiments form the basis for a recognition model in which highly cooperative binding by DGCR8 is the distinguishing characteristic of specific pri-miRNA recognition (14). Although our K d values are in agreement with those previously reported for filter binding (13,49), we have been unable to demonstrate conclusive deviations in binding cooperativity among various RNA substrates. Our direct, EMSA-based observation of higher order complex formation involving DGCR8 core , which lacks a previously identified trimerization domain, also questions a generalized model of pri-miRNA recognition involving stepwise, cooperative binding mediated by the DGCR8 trimerization domain (14).
We employed concentrations of 100 pM of end-labeled RNA and a low salt buffer (50 mM NaCl) in filter binding assays and observed tighter affinity and lower Hill coefficients for the heme-bound DGCR8 ⌬275 /pri-mir-16-1 interaction when com- pared with previously published studies by Faller et al. (13,14) using unspecified concentrations of body-labeled pri-mir21 and pri-mir30a in complex with DGCR8 ⌬275 in a reaction buffer containing 85 mM NaCl. We observed reduced affinity for the DGCR8 core /pri-mir-16-1 interaction using EMSA in a high salt buffer (150 mM KCl, data not shown). Thus, the apparent K d values for DGCR8 binding to pri-mir-16-1 generally decrease with increasing salt concentration.
Hill coefficients often do not provide accurate estimates of the number of binding sites (60), particularly when multisite complex formations are involved. In general, nitrocellulose filter binding describes only macroscopic association parameters that are composite averages of individual binding sites and cooperativity constants. Here, one must assume a relationship between the number of ligands bound and complex retention by the nitrocellulose membrane to accurately describe the interaction. Senear et al. (61) have concluded that it is impossi-ble to correctly resolve microscopic binding and cooperativity constants of a multisite complex by filter binding. The apparent steepness of dose-response curves and the corresponding Hill coefficients (n) depend on invariant 32 P-labeled RNA concentration below the K d . When RNA concentrations approach the K d value, the resulting best fits to the Hill equation no longer provide accurate information on either K d or n (60,62). Dimerization at higher DGCR8 concentration can potentially alter binding characteristics that would further complicate reliable curve fitting. Using SEC and NMR diffusion measurements, we and others (49) have detected a dimerization tendency for DGCR8 variants. Finally, because we investigated DGCR8-pri-mir-16-1 and pri-let-7b complex formation, we cannot rule out the possibility that the highly cooperative association previously observed could be a specific feature of DGCR8 binding to pri-mir30a and pri-mir21 studied by Faller et al. (13,14). Although processed by the Microprocessor, alter- ⌬LW values were calculated by subtracting the LW bound of a given residue from the LW free . ⌬LW for infinitely broadened resonances (gray bars) were calculated assuming a conservative LW bound of 50 Hz corresponding to the largest 1 H-TROSY LW bound observable at 800 MHz. Secondary structure elements (␣1-7 and ␤1-7) as determined by chemical shift index using TALOSϩ (75) are identified below the residue number. B, color-coded worm representation of DGCR8 core /pri-mir-16 lower titration at a 24:1 protein/RNA ratio. Secondary structure elements are numbered and indicated. Worm thickness of the DGCR8 core crystal structure (PDB code 2YT4) varies between 2.00 (⌬LW ϭ Ϫ38.3 Hz, red) and 0.25 (⌬LW ϭ 9.2 Hz, blue). Missing crystal residues were modeled and energy-minimized in Chimera (76). C, DGCR8 core /pri-mir-16 lower titration at a 1:2 protein/RNA ratio. Individual RBDs are numbered; N and C termini are shown and indicated. The 39 visible TROSY peaks remaining in the presence of excess RNA predominantly map to the N and C termini, and flexible linker regions of the DGCR8 core crystal structure (PDB code 2YT4) and are mapped and highlighted in blue on the protein sequence (A). native biogenesis pathways for pri-mir21 independent of DGCR8 have been suggested (63).
UV cross-linking experiments in the presence of varying amounts of competitors have previously been employed to establish relative affinities of DGCR8 to pri-miRNA (17). Although DGCR8 could be cross-linked with decreasing efficiency to diverse targets such as pri-mir-16-1, a small interfering (si)RNA duplex and ssRNA, the ability to compete for DGCR8 binding revealed some advantage for ss-dsRNA junctions. Furthermore, a length dependence where shorter RNA constructs competed gradually less efficiently with the longer pri-miRNA was observed. Our DGCR8 core /pri-mir-16-1 EMSA and increasing apparent K d values in the presence of varying amounts of nonspecific competitor yeast tRNA Phe suggest that DGCR8 core /RNA interactions are nonspecific. Consistent with a nonspecific binding mode, the results obtained from competition experiments may have uncovered a correlation between efficient competition and the number of nonspecific binding sites provided by the various competitors.
To overcome the binding assay limitations described above and to obtain independent information about the DGCR8 core ⅐pri-mir-16 complex equilibrium in solution, we conducted NMR titration and diffusion measurements. Results from NMR titrations strongly suggest that the DGCR8 core ⅐primir-16 lower complex represents a nonspecific interaction. The observed line broadening in the absence of chemical shift perturbations indicates the formation of transient, nonspecific complexes (64). Results obtained from diffusion measurements further corroborated nonspecific DGCR8 core /pri-mir-16 lower binding models as we observed interconverting free DGCR8 core and complexes with hydrodynamic radii of ϳ38.1 Å.
These findings are consistent with reports of other dsRBD-containing proteins involved in small RNA pathways. For example, dsRBDs of both HYL1 (HYPONASTIC LEAVES 1) and PKR that exhibit noncanonical binding and extensive line broadening have been shown to bind RNA without apparent specificity (55,65,66). Similarly, we have been unable to detect by NMR nucleotide-specific changes to RNA targets complexed with DGCR8 core . Analogous to the protein spectra, pri-mir-16 imino-proton peaks were broadened globally and severely. This result suggests that DGCR8 dsRBDs simultaneously bind overlapping interfaces of the RNA in a manner resembling nonspecific PKR⅐dsRNA complex formation (66). Finally, the Dicer accessory dsRBD-containing protein-transactivating response RNA-binding protein and two homologs, including PKR, were most recently shown to diffuse along dsRNA substrates, which is clearly an activity conceivably adopted by DGCR8 to anchor Drosha-DGCR8 heterodimers (67).
Although our nonspecific DGCR8/pri-miRNA interaction model differs from the proposed cooperative recognition model (14), several common observations have been described. Two studies have identified that helix 2 in each ␣-␤-␤-␤-␣ dsRNA-binding domain of DGCR8 is important for pri-miRNA FIGURE 9. pri-mir-16 lower ⅐DGCR8 core complex formation monitored by NMR. A, 850 MHz imino one-dimensional jump-return echo experiment of pri-mir-16 lower collected at 298 K (black) labeled with assignment information; numbering is consistent with full-length pri-mir-16-1 (Fig. 1). RNA/protein molar ratios ranging from 48:1 (orange) to 1:1 (gray) were achieved by adding dilute aliquots of DGCR8 core to a 250 M pri-mir lower sample in NMR buffer. B, same as above with 5 mM MgCl 2 added to the NMR buffer. recognition (15,49). In good agreement, our NMR titration analysis recognizes the first and second helices of each dsRBD motif (RBD1:␣1 and ␣2, RBD2:␣5 and ␣6) as primary pri-mir-16-1 interaction interfaces. The binding affinity of DGCR8 for the 160-nt P4 -P6 domain of the Tetrahymena ribozyme is similar to the ones observed for bona fide pri-miRNA substrates (14). Likewise, we cannot distinguish ssRNA, fully paired (primir-16fp) or random hairpin (pri-mir-16x) interactions involving DGCR8 from pri-miRNA-derived substrates based on K d values. Previous studies have highlighted the importance of oligomeric (DGCR8) n ⅐pri-miRNA complexes using filter binding assays, SEC, and low resolution electron tomography (14). Our NMR titration and diffusion analysis directly probe solution equilibria and offer further support for oligomeric, albeit nonspecific, binding of DGCR8 to RNA substrates. Further-more, we were able to visualize DGCR8⅐pri-miRNA complexes with variable stoichiometry using EMSA. DGCR8 alone shows no preference for a particular RNA substrate, yet we and others have demonstrated through in vitro processing assays that the Microprocessor is highly selective when cleaving its RNA targets (17,18). Thus, we conclude that DGCR8-RNA substrate binding is not the specificity determinant of pri-to pre-miRNA processing, and a model of stepwise pri-miRNA recognition and Drosha recruitment by DGCR8 seems unlikely. Because known requirements common to all Microprocessor substrates remain broad and include 70-nt hairpins with unpaired flanking regions (19), the functional implication of our finding is that the concerted action of a DGCR8⅐Drosha complex is required for effective recognition of miRNA-containing transcripts. Consistently, recent studies FIGURE 10. Pulsed field gradient NMR and DGCR8-pri-mir-16-1 complex models. A, natural log scale of signal integrals of BPPLED-PFG 1 H NMR experiments as a function of gradient amplitude squared and linear data fits. Diffusion of individual protein and RNA components is as follows: DGCR8 RBD1 (red), DGCR8 core (blue), pri-mir-16 lower (green), and DGCR8 PC (black). B, diffusion of 1:2 molar ratio DGCR8 core ⅐pri-mir-16 lower complexes as follows: average (dashed red line), lower limit (solid blue line), and upper limit (dashed black line). The inset (G 2 -G 0 2 ϭ 0.00 to 0.05, ln(1/I 0 ) ϭ 0.0 to Ϫ0.5) in the upper right corner shows the best fit for the upper complex limit (dashed black line). FIGURE 11. In vitro pri-miRNA processing assays. A, processing of uniformly labeled pri-miRNAs in the absence (Input) or in the presence (c-myc-IP) of immunoprecipitated Microprocessor. Assays were performed as described under "Experimental Procedures." Sizes of oligonucleotide DNA markers are shown on the left. 5Ј, 3Ј, and pre-miRNA processing products are indicated on the right. B, DGCR8 ⌬275 -dependent pri-miRNA processing. Recombinant DGCR8 ⌬275 was incubated with stringently washed IP-Drosha. 32 P-5Ј-End-labeled pri-mir-16-1 variants were incubated in the absence (Ϫ) or presence (ϩ) of Microprocessor components. Images in A and B are representative of at least two independent experiments. revealed that association with the HIV-1 LTR requires the presence of both Microprocessor components, Drosha and DGCR8 (68). An alternative scenario, where Drosha can distinguish its target among nonspecific DGCR8⅐pri-miRNA complexes cannot be strictly ruled out based on our results. Importantly, stable association of DGCR8 and Drosha occurs independently of RNA (10). Presumably, this protein/protein interaction leads to enhanced RNA binding specificity by way of conformational modulation of DGCR8's dsRBDs, by steric limitation of binding opportunities to constrained regions of the RNA hairpin, or a combination of both scenarios. The DGCR8/ RNA interaction and other examples (55,67,69,70) suggest that dsRBDs associated with miRNA biogenesis are generally nonspecific binders. Although this form of association might seem less than satisfying, nonspecific and/or loose binding can be an important characteristic for an accessory protein. Thus, the biological implication is that nonspecific associations can greatly enhance the RNA binding flexibility of a protein complex, which is in good agreement with reports demonstrating DGCR8's involvement in controlling the fate of diverse classes of RNA such as mRNAs, small nucleolar RNAs, long noncoding RNAs, and pri-miRNAs (24). For the Microprocessor in particular, the ability to bind a broad range of loosely related hairpin structures could be critical to carry out its universal function.