Structural basis for terminal loop recognition and processing of pri-miRNA-18a by hnRNP A1

Post-transcriptional mechanisms play a predominant role in the control of microRNA (miRNA) production. Recognition of the terminal loop of precursor miRNAs by RNA-binding proteins (RBPs) influences their processing; however, the mechanistic and structural basis for how levels of individual or subsets of miRNAs are regulated is mostly unexplored. We previously described a role for hnRNP A1, an RBP implicated in many aspects of RNA processing, as an auxiliary factor that promotes the Microprocessor-mediated processing of pri-mir-18a. Here, we reveal the mechanistic basis for this stimulatory role of hnRNP A1 by combining integrative structural biology with biochemical and functional assays. We demonstrate that hnRNP A1 forms a 1:1 complex with pri-mir-18a that involves binding of both RNA recognition motifs (RRMs) to cognate RNA sequence motifs in the conserved terminal loop of pri-mir-18a. Terminal loop binding induces an allosteric destabilization of base-pairing in the pri-mir-18a stem that promotes its down-stream processing. Our results highlight terminal loop RNA recognition by RNA-binding proteins as a general principle of miRNA biogenesis and regulation.

Post-transcriptional mechanisms play a predominant role in the control of microRNA (miRNA) production. Recognition of the terminal loop of precursor miRNAs by RNA-binding proteins (RBPs) influences their processing; however, the mechanistic and structural basis for how levels of individual or subsets of miRNAs are regulated is mostly unexplored. We previously described a role for hnRNP A1, an RBP implicated in many aspects of RNA processing, as an auxiliary factor that promotes the Microprocessor-mediated processing of pri-mir-18a. Here, we reveal the mechanistic basis for this stimulatory role of hnRNP A1 by combining integrative structural biology with biochemical and functional assays. We demonstrate that hnRNP A1 forms a 1:1 complex with pri-mir-18a that involves binding of both RNA recognition motifs (RRMs) to cognate RNA sequence motifs in the conserved terminal loop of pri-mir-18a.
Terminal loop binding induces an allosteric destabilization of base-pairing in the pri-mir-18a stem that promotes its down-stream processing. Our results highlight terminal loop RNA recognition by RNA-binding proteins as a general principle of miRNA biogenesis and regulation.
Keywords: microRNA biogenesis; miR-18a, miR-17-92 cluster; hnRNP A1; nuclear magnetic resonance (NMR); X-ray crystallography; small angle X-ray/neutron scattering (SAXS/SANS); RNA recognition motif (RRM) MicroRNAs (miRNAs, miRs) are a class of highly conserved small non-coding RNAs that play a crucial role in the regulation of gene expression. They are involved in a variety of biological processes including cell growth, proliferation and differentiation 1 . Mature miRNAs are generated by two RNA cleavage steps involving nuclear and cytoplasmic RNase III enzymes (Drosha and Dicer, respectively). Primary miRNA (pri-miRNA or pri-mir) transcripts are cropped by the Microprocessor complex (comprising Drosha and DGCR8) in the nucleus forming ~70 nucleotide (nt) stem-loop precursor miRNAs (pre-miRNAs or pre-mir), which, following export to the cytoplasm, are further processed by Dicer into mature miRNAs (reviewed in 2 ). Many miRNA genes in higher organisms are transcribed together as a cluster 3 .
A prototypical example is the miR-17-92 cluster that is encoded as an intronic polycistron on chromosome 13 in humans. This cluster encodes six individual miRNAs that are highly conserved in vertebrates (miR-17, miR-18a, miR-19a, miR-20a, miR-19b-1, miR-92a-1, reviewed in 4 ). The miR-17-92 cluster is frequently amplified and overexpressed in human cancers; hence, it is also referred to as OncomiR-1. Its oncogenic role was confirmed in a mouse model of B cell lymphoma 5 . Furthermore, its targeted deletion is associated with developmental defects in mouse model systems 6 .
The biogenesis of miRNAs is tightly regulated and results in tissue-and developmentalspecific expression patterns of miRNAs 7 . A number of specific RNA-binding proteins (RBPs) have recently emerged as important post-transcriptional regulators of miRNA processing. However, very little is known about their mechanism of action. Previously, we identified heterogenous nuclear ribonucleoprotein A1 (hnRNP A1) as a factor, which positively regulates the processing of miRNA-18a primary transcript (pri-mir-18a) by making specific contacts to the terminal loop of the RNA 8,9 (Fig. 1a,b). HnRNP A1 is a highly abundant RBP that has been implicated in diverse cellular functions related to RNA processing, including alternative splicing regulation [10][11][12] , mRNA export 13,14 , IRES (internal ribosome entry site)-mediated translation 15,16 , mRNA stability 17,16 and telomere maintenance 18,19 .
Here, we have combined an integrative structural biology approach with biochemical and functional assays to provide mechanistic insights into the role of hnRNP A1 in stimulating pri-4 mir-18a processing. We show that hnRNP A1 forms a 1:1 complex with pri-mir-18a in solution, with the recognition of two UAG motifs by the tandem RRM domains of hnRNP A1 revealed by a high-resolution crystal structure. NMR and biophysical data show that high-affinity binding involves recognition of two UAG motifs in the pri-mir-18a terminal loop and the proximal stem region. Notably, binding to the terminal loop induces an allosteric destabilization of basepairing in the pri-mir-18a stem that promote its processing. These findings may serve as a paradigm for the regulation of miRNA processing by the recognition of the terminal loop by RBPs.

Nuclear localized UP1 is necessary and sufficient for stimulating pri-mir-18a processing
We have previously shown that hnRNP A1 acts as an auxiliary factor for miRNA biogenesis, by binding to pri-mir-18a and inducing a relaxation at its lower stem creating a more favorable cleavage site for Drosha 9 . However, the underlying molecular mechanisms are unknown.
HnRNP A1 has two RNA recognition motif (RRM) domains, each harboring conserved RNP-1 and RNP-2 submotifs, and a C-terminal flexible glycine-rich tail, which includes the M9 sequence, responsible for nuclear import and export 20,13 . The RNA-binding region of hnRNP A1, comprising the tandem RRM1-RRM2 domains, is referred to as UP1 (Unwinding Protein 1) 21 (Fig. 1a). To identify which regions in hnRNP A1 are required for stimulating pri-mir-18a processing in living cells we used an in vivo processing assay. For this, several N-terminal T7tagged hnRNP A1 constructs were transiently overexpressed in HeLa cells and the level of mature miR-18a was analyzed by qRT-PCR. We found that overexpression of full-length hnRNP A1 results in a ~2-fold increase in the levels of mature miRNA-18a, whereas UP1, comprising both RRMs but lacking the M9 sequence, has no effectmost likely due to its cytoplasmic localization ( Fig. 1c; Supplementary Fig. 1a Table 1) and, importantly, displayed similar activity as full-length hnRNP A1 in stimulating miR-18a production in vivo (Fig. 1c). By contrast, RRM1-M9 and RRM2-M9 that partially localize to the nucleus (Supplementary Fig. 1b i.e. comprising both RRMs of hnRNP A1, is required for function (Fig. 1c).

UP1 specifically recognizes the loop region of pri-mir-18a
We next wanted to determine the regions of pri-mir-18a that are recognized by hnRNP A1. To this end, we performed electro-mobility shift assays (EMSA) with RNA variants corresponding to the terminal loop and the stem of pri-mir-18a with UP1 (Fig. 1b), which has been shown to recapitulate most of the functions of full-length hnRNP A1 in vitro 21 (Fig. 1d). We observed that UP1 specifically binds to the terminal loop RNA, whereas no binding to the stem RNA was detected even at higher protein-to-RNA ratios. Single RRM1 and RRM2 domains do not show any detectable RNA-binding activity in this assay ( Supplementary Fig. 1c). Altogether, these data show that i) UP1 binds specifically to the terminal loop region of pri-mir-18a, and ii) both RRM domains of UP1 are required for high affinity RNA binding, indicating that they bind cooperatively.

Identification of a minimal UP1-binding sequence in the pri-mir-18a terminal loop
To provide a quantitative analysis of binding affinities of UP1 with pri-mir-18a, we performed isothermal titration calorimetry (ITC) experiments. Pri-mir-18a 71-mer RNA binds to UP1 with a dissociation constant (KD) of 147 nM (Fig. 2a) 1b) shows more than 200-fold higher affinity (KD = 15.5 nM with a 1:1 stoichiometry) to UP1 (Fig. 2a, middle panel). The 12-mer RNA harbors two UAG motifs suggesting that each of these can be recognized by one of the RRMs in a cooperative manner. This is evident from 7 the very large increase in binding affinity compared to the binding of the 7-mer to UP1. It is remarkable that binding of UP1 to this single-stranded 12-mer RNA is at least 10-fold stronger than binding to full-length pri-mir-18a (KD = 15.5 nM vs. 147 nM, respectively) (Fig. 2a, middle and right panel, respectively). This may be related to the observation that in the pri-mir-18a stem-loop the second UAG motif is predicted to be base-paired in the loop-proximal stem and thus not freely accessible (Fig. 1b). Binding of UP1 to pri-mir-18a may thus require opening (melting) of these base-pairs, whereas in the single-stranded 12-mer RNA both UAG binding sites are readily available for interaction with UP1 (see below), thus the higher affinity. We conclude that the 12-mer RNA is a high-affinity UP1-binding sequence where both RRM domains of UP1 recognize UAG motifs. The RNA recognition features in the 12-mer are expected to represent the interaction of UP1 with the pri-mir-18a. A summary of thermodynamic parameters of the hnRNPA1-RNA interactions is given in Table 1.

NMR analysis of UP1-RNA interactions
We next characterized the binding interface of UP1 with various RNA ligands derived from primir-18a using NMR titration experiments. Addition of the 7-mer RNA harboring one UAG motif causes extensive chemical shift perturbations (CSPs) in RRM1 and RRM2 constructs, thus demonstrating that each of the RRMs can interact with the 7-mer RNA (Fig. 2b). The CSP pattern obtained upon titration of the tandem RRM domains in UP1 with the 7-mer RNA is very similar to the one obtained for individual RRM domains (Fig. 2b, c). This shows that the recognition of the RNA in the isolated RRM domains and in the context of the UP1 construct is very similar. Similarly, the 12-mer RNA harboring two UAG motifs induces large CSPs in both RRM domains of UP1 (Fig. 2d) that are comparable to those obtained at saturating levels of the 7-mer. The CSPs map to the canonical RNA binding surface on the -sheets of the two RRM domains (Fig. 2e). In addition, strong CSPs are observed for residues in the C-terminal region of RRM2. This region is flexible in free UP1 but upon RNA binding forms an additional helix (α3), which is not present in the free protein 23 (Supplementary Fig. 3a, b), suggesting that helix α3 is induced and stabilized upon RNA binding. Interestingly, a number of NMR 8 signals that correspond to residues in RRM2 and the RRM1-RRM2 linker are severely broadened in the RNA-bound spectrum, suggesting dynamics on the µs-ms time-scale. The affected residues map to the interface between the RRM1 and RRM2 domains (Fig. 2e), suggesting that some conformational dynamics and adaptation of this domain interface is associated with RNA binding.

Structural basis for the recognition of the pri-mir-18a terminal loop by UP1
To gain insight into the molecular details of pri-mir-18a recognition by UP1, we determined the crystal structure of the UP1/12-mer RNA complex at 2.5 Å resolution (Table 2; Fig. 3a; Supplementary Fig. 4a). Surprisingly, the crystal structure of the UP1/12-mer RNA complex exhibits two molecules of UP1 and two RNA chains in the asymmetric unit in a 2:2 stoichiometry ( Supplementary Fig. 4a), similar to a previously reported structure of UP1 with single-stranded telomeric DNA 24 . As this peculiar 2:2 stoichiometry most likely does not represent the UP1:RNA complex in solution, we analyzed UP1, RNA and the protein-RNA complexes by static light scattering (SLS) (Fig. 3b). Both UP1 and pri-mir-18a alone behave as single species with a molecular weight corresponding to respective monomeric conformations. Importantly, the molecular weight obtained for the UP1/12-mer complex (22.4 kDa) indicates a 1:1 complex and demonstrates that the 2:2 stoichiometry observed in the crystal structure is an artifact induced by the crystal environment. Notably, the molecular weight obtained for the UP1/pri-mir-18a complex (45 kDa) is fully consistent with the formation of a 1:1 complex (Fig. 3a, b).
The crystal structure reveals that each RRM domain in UP1 specifically recognizes one UAG motif in the RNA. Although the stoichiometry does not reflect the solution conformation, the RNA contacts are expected to be conserved, consistent with the NMR titrations. Each UAG motif is recognized by contacts mainly through conserved RNP motif residues on the β-sheets ( Fig. 3c, d), which resembles the recognition of TAG in the UP1-telomeric DNA complex 24 and recently reported structures with RNA 25,26 . Two conserved aromatic residues in RRM1, Phe17 (RNP-2 motif residue located on β1) and Phe59 (RNP-1 motif residue located on β3), are 9 involved in stacking interactions with the bases of A4 and G5, respectively (Fig. 3d). A third aromatic residue, Phe57 (RNP-1 motif residue located on β3), interacts with the ribose rings of A4 and G5. Similarly, in RRM2 Phe108 (RNP-2 motif residue located on β1) and Phe150 (RNP-1 motif residue located on β3) stack with the bases of A9 and G10, respectively, whereas Phe148 (RNP-1 motif residue located on β3) makes contacts with the sugar rings of A9 and G10. In addition to the stacking interactions with RNP residues (vide infra), the central adenosine in the UAG motifs is specifically recognized by hydrogen bonds of its exocyclic NH2 group with the main chain carbonyl oxygen of residues Arg88 and Lys179 in RRM1 and RRM2, respectively. A positively charged residue in each domain, Arg55 in RRM1 and Arg146 in RRM2, makes electrostatic interactions with the phosphate backbone of the AG dinucleotide.
Two charged residues in each domain, Glu85 and Lys87 in RRM1 and Glu176 and Arg178 in RRM2, make specific contacts to the uridines in the UAG motifs (U3 and U8), while another charged residue, Lys15 in RRM1 and Lys106 in RRM2, interacts with G5 and G10, respectively (Fig. 3d), thereby specifying the U and G residues in the UAG motif. The mode of RNA recognition by RRM1 and RRM2 is very similar; in each domain an AG dinucleotide is sandwiched between the -sheet surface and a C-terminal helix (Fig. 3d).
The two RRM domains of hnRNP A1 are connected by an approximately 17-residue linker, which is evolutionary conserved both in terms of sequence and length 27  To confirm the domain arrangement of RNA-bound UP1 in solution, we measured NMR paramagnetic relaxation enhancements (PRE) for UP1 spin-labeled at position 66 (UP1 Glu66Cys mutant). The PRE data provide long-range (up to ~20 Å) distance information and 10 can thus report on domain/domain arrangements [28][29][30] (Supplementary Fig. 4b). The PRE profiles of free and RNA-bound form of UP1 are similar, suggesting that the domain arrangement does not change significantly in the presence of the 12-mer RNA.
To obtain additional restraints to determine a structural model of the 1:1 UP1/12-mer RNA complex in solution we measured residual dipolar couplings (RDCs) and small angle X-ray scattering (SAXS) on the UP1/12-mer RNA complex (Fig. 3e,

Validation of the UP1/12-mer RNA structural model
The structural model of the 1:1 UP1/12-mer RNA complex was confirmed by mutational analysis of protein and RNA. ITC data with 12-mer RNAs where the first or second UAG motif has been replaced by UUU, 12-mer-mut1: AGUUUAUUAGCA and 12-mer-mut2: AGUAGAUUUUCA show 10-fold and 20-fold (KD = 154 nM and 330 nM) reduced binding affinity, respectively, compared to the wildtype sequence (Table 1; Supplementary Fig. 2b).
This demonstrates that both motifs are recognized by the protein. Additional RNA variants with an AGUU mutation or lacking the initial AG dinucleotide have the same binding affinity as the wildtype 12-mer RNA (Table 1; Supplementary Fig. 2b). This shows that UP1 has a preference for the recognition of two neighboring UAG motifs.
Further, the domain interface in UP1 was probed by mutations that are expected to disrupt the two salt-bridges Arg75-Asp155 and Arg88-Asp157 in the RRM1-RRM2 interface, which have been observed in all reported structures of free and nucleic-acid bound forms of UP1 24,23,25,26 .
Introducing charge clashes (UP1-Arg75Glu/Arg88Glu) in this interface, which is remote from the RNA binding surface, decreases the binding affinity to the 12-mer RNA by ≈3-fold (KD ~15.5 nM vs. ~40 nM) ( Supplementary Fig. 2c). This suggests that the salt bridges play an indirect role for RNA binding by stabilizing the arrangement of the two RRM domains.
To assess the effect of RNA-binding in the functional activity of UP1, we mutated conserved Phe residues within RNP-1 motifs that directly contact the RNA and are required for RNAbinding of hnRNP A1 21 . Substitution of Phe with Asp or Ala within individual or combined RRM1 and RRM2 domains was sufficient to abolish the activity of UP1-M9 in our in vivo primir-18a processing assay without affecting the nuclear localization of the protein constructs ( Supplementary Fig. 1a, b). This indicates that the RNA-binding activity of hnRNP A1 is essential for its stimulating activity of miRNA-18a biogenesis. The in vivo functional data further confirm that the stimulatory function of hnRNP A1 in processing of pri-mir-18a requires both RRM domains, as mutations that affect RNA binding in one domain (or deletion of one domain) abolish the activity of hnRNP A1.
Collectively, the structural model of the 1:1 UP1/12-mer RNA complex is fully consistent with our biochemical and functional data regarding the requirement for two RRM domains and two UAG motifs for high-affinity interaction.

UP1 binding destabilizes the dynamic pri-mir-18a RNA terminal loop
The structure of UP1 bound to the 12-mer RNA in a 1:1 stoichiometry implies that the loopproximal region of the pri-mir-18a stem should be destabilized to enable recognition of two UAG motifs, one accessible in the terminal loop and the second one as part of the pri-mir-18a duplex. To study this further, we first analyzed the structure of pri-mir-18a alone. For this, a model of the RNA was prepared using the MC-Fold/MC-Sym server 31 (Fig. 4a) and assessed with experimental SAXS and NMR data. The predicted secondary structure and elongated shape of pri-mir-18a 71-mer is supported by SAXS data of the free RNA (Fig. 4b). We then used NMR to analyze the base-pairing in the pri-mir-18a 71-mer using imino NOESY spectra ( Fig. 4c), as imino proton NMR signals probe the presence and stability of base pairs. We could unambiguously assign imino-imino cross-peaks corresponding to the stem region of the 71-mer RNA. However, no imino correlation was observed for the upper part of the stem loop  Fig. 3d), which binds with a KD of ~3 µM, i.e. corresponding to a 20-200-fold reduced binding affinity compared to the full-length pri-mir-18a and the single-stranded 12-mer RNA (Table 1; Supplementary   Fig. 2d). Thus, the availability of only one single-stranded UAG motif for binding to UP1 yields micromolar affinity comparable to the 7-mer RNA. The fact that pri-mir-18a exhibits one UAG motif in the terminal loop and a second one in the weak and dynamic upper stem region suggests that binding of UP1 will require melting of the stem region flanking the terminal loop to enable recognition of this partially hidden UAG motif and high affinity (low nanomolar KD) RNA binding.
Next, we compared the 1 H, 15 N imino correlations of the pri-mir-18a 71-mer RNA in the free form and bound to UP1 (Fig. 4b). Notably, signals for the U10:G52 base pair in the middle of the stem region are not detectable in the complex but readily observed in the free form, indicative of destabilization and partial melting of this part of the duplex. Also, other residues especially next to mismatches and other less stable regions of the RNA exhibit reduced intensity or chemical shift perturbation such as G23, U49 and U59 (Fig 4c, green nucleotides). This is consistent with allosteric effects that lead to destabilization of the complete pri-mir-18a stem-loop induced by binding of UP1 to the terminal loop.

Structural model of the UP1/pri-mir-18a complex and biochemical validation
To derive a structural model of UP1 bound to pri-mir-18a, we performed molecular dynamics simulations restrained by the structural information obtained for the UP1/12-mer RNA complex and experimental SAXS data of the UP1/pri-mir-18a complex (  Table 2). The structural model shows very good agreement with all experimental data and is also consistent with ab initio SAXS derived models of the protein-RNA complex ( Supplementary Fig. 4e, f). The UP1/pri-mir-18a complex shows the recognition of two UAG motifs in the terminal loop and a partially melted upper stem of primir-18a (Fig. 5a).
To validate the structural model described and to assess changes in accessibility of the primir-18a induced by UP1 binding, we performed foot-printing analysis of the complete pri-mir-18a RNA in the absence and presence of UP1. This revealed that in the free RNA the terminal loop and flanking stem region comprising the two UAG motifs are accessible and dynamic ( Fig. 5b), which is consistent with the NMR data (Fig. 4c, d). Binding of UP1 protects this region in the RNA, in full agreement with the structural model of the UP1/pri-mir-18a complex.
The significant reduction in accessibility observed for residues in the terminal loop is consistent with the simultaneous binding of both RRM domains to the RNA (Figs. 3c, 5a). Interestingly, residues at the bottom part of the stem of pri-mir-18a become more accessible for nuclease cleavage upon binding of UP1 to the terminal loop (Fig. 5b, green nucleotides). This is consistent with the NMR analysis of pri-mir-18a free and bound to UP1 that shows UP1/primir-18a interactions (Fig. 4) and the proposed destabilization of the RNA stem induced by UP1 binding (Fig. 5c).

SHAPE analysis
Next, to assess the RNA structure of pri-mir-18a and potential effects from the presence of flanking regions in the context of the pri-mir-17-19 cluster, we performed structural analysis by selective 2'-hydroxyl acylation analyzed by primer extension (SHAPE, Supplementary Fig.   5a, b) 33 . To this end, in vitro transcribed RNAs comprising either pri-mir-18a or the pri-mir-17-19 cluster were incubated with increasing amounts of purified UP1 or full-length hnRNP A1 proteins, prior to treatment with N-methylisatoic anhydride (NMIA) that reacts with the 2´hydroxyl group of flexible nucleotides. The SHAPE reactivity reflects the intensity of the NMIAtreated RNA primer extension products, normalized to the corresponding untreated RNA, in the presence or absence of UP1/hnRNP A1 proteins.
Importantly, upon addition of UP1 and hnRNP A1, SHAPE differences relative to the free RNA indicate that residues preferentially protected from NMIA attack include the terminal loop and flanking stem region (nts 170, 185-186) ( Supplementary Fig. 5a). The SHAPE reactivity values observed with pri-mir-17-19 free RNA were used to compute differences in SHAPE reactivity  Fig. 5b). Similar SHAPE results were observed upon incubation with UP1. Although the protection was less intense than that observed with the fulllength protein, the protected residues mapped to the same RNA region ( Supplementary Fig.   5). Importantly, these results confirm that the UP1 fragment of hnRNP A1 is sufficient to induce this protection, which is in agreement with the EMSA and functional assays (Fig. 1c, d).
Most of the highly reactive residues in pri-mir-18a display a similar behavior as the pri-mir-17-19 transcript upon addition of UP1/A1 proteins ( Supplementary Fig. 5). Despite the overall similar reactivity pattern, differences observed at nts 210-220 are presumably induced by the presence of the miR-17 and miR-19a in the whole transcript, which may stabilize the structure of the basal region of miR-18a. In summary, relative to the SHAPE reactivity observed with free RNA incubation of RNA with either hnRNP A1 or UP1 leads to a decreased SHAPE reactivity around the terminal loop of miR-18a, indicating that this region is the major binding site for the RRMs of hnRNP A1 and that this binding impacts the overall structure of pri-mir-18a in isolation or as part of the pri-mir-17-19 cluster.

Mechanism of hnRNP A1 stimulation of pri-mir-18a processing
Finally, we attempted to address the mechanism by which hnRNP A1 activates the processing of pri-mir-18a. One possible scenario is that binding of hnRNP A1 (or UP1) leads to partial opening/melting of the terminal loop, which can lead to destabilization of the stem region and thus render it more accessible for processing by Drosha as proposed before 9 . Indeed, footprinting and site-directed mutagenesis of pri-mir-18a suggested that hnRNP A1 alters the local conformation of the stem in the vicinity of Drosha cleavage sites 9 , although the molecular mechanism was unclear. Indeed, UP1 (unwinding protein 1), as the name suggests, can unwind secondary and higher order structures of DNA and RNA 34,35,19 . To examine this possibility, we constructed a series of mutants in the terminal loop region of pri-mir-18a. These include single and double nucleotide mutants, where an A residue within the UAG motif in the terminal loop or within the second UAG motif, was mutated to C (UCG within pri-mir-18a[A30C] and pri-mir-18a[A30C/A35C], respectively), and a triple mutant (pri-mir- , where all UAG motifs were mutated to AAG. We also designed a primir-18a mutant, in which the terminal loop was stabilized by five G:C base pairs (pri-mir-18a[5GC]). As expected the wildtype sequence was efficiently processed. The single (pri-mir-18a [A30C]) and double nucleotide (pri-mir-18a[A30C/A35C]) terminal loop mutant RNAs retained hnRNP A1 binding, although with lower affinity, and were accordingly efficiently processed (Fig. 6a). The triple mutant (pri-mir-18a[U21A/U29A/U34A]) showed binding to hnRNP A1, although with reduced affinity, and retained efficient processing (Fig. 6b). The primir-18a[5GC] mutant does not bind to hnRNP A1 in the RNA pull-down assay and consequently is not processed (Fig. 6a). This lack of processing could result from disruption of UP1 binding and/or conformational changes that inhibit Microprocessor activity, for example, by stabilizing the loop-proximal stem region. Based on these data we conclude that hnRNP A1 binding is essential for pri-mir-18a processing. The experiments also show that even very low hnRNP A1 levels are sufficient to stimulate processing activity. despite retaining full binding to hnRNP A1 in the RNA pull-down assay (Fig. 6b). To rule out the possibility that the internal 5GC affects Drosha processing irrespective of the requirement for hnRNP A1 as an auxiliary factor, we examined the processing of pri-mir-16 with and without the 5GC clamp. Both wildtype pri-mir-16 and pri-mir- 16[5GC_internal] were processed by Drosha, indicating that the internal 5GC clamp is not sufficient to impair Drosha processing (Fig. 6c). These data strongly support our hypothesis that unwinding of pri-mir-18a by hnRNP A1 can spread from the terminal loop towards the stem and is essential for stimulation of miR-18a biogenesis.

Discussion
Here, we have used a multi-disciplinary approach to reveal the molecular mechanism by which hnRNP A1 binds to pri-mir-18a and facilitates its processing. Our results establish that hnRNP A1 specifically binds to pri-mir-18a through interactions involving both RRM domains in UP1  Fig. 4d). Nevertheless, some adaption and fine-tuning of the domain arrangement occurs upon RNA binding. This is also indicated by line-broadening of amide signals in the RRM1/RRM2 interface upon binding to the 12-mer RNA (Fig. 2e) (Fig. 5c). In support of this model, processing of a mutant pri-mir-18a, in which the terminal loop was clamped by 5 G:C base pairs (5GC_internal), was abolished, despite strong hnRNP A1 binding (Fig. 6b). This strongly argues that the effect of hnRNP A1 binding at the terminal loop is somehow propagated and leads to a stimulatory effect of Drosha processing.
We had previously shown that pri-mir-18b, which is part of the homologous primary cluster miR106a~18b~20b located on chromosome X, does not require hnRNP A1 for efficient processing. Mechanistically, this can be explained by the fact that the conformation of the stem in pri-mir-18b, resembles the more open stem structure comprising a bulge in the stem (UCGU), which is only observed in pri-mir-18a, upon binding of hnRNP A1 8,9 . This is critical for more efficient Drosha processing as shown by the fact that simply introducing this bulge in the pri-miR-18a stem (UCGU) made its processing more efficient and completely independent of the presence of hnRNP A1 8,9 .
Importantly, our use of integrative structural biology combined with biochemical and functional assays allowed us to extend these previous observations and conclude that the main effect of hnRNP A1 binding to the terminal loop of pri-mir-18a is to promote the destabilization of the lower stem, which leads to increased Drosha cleavage, via a mechanism that is not fully understood. Recent biochemical and structural analyses have shown that the Microprocessor recognizes two regions at either end of the miRNA precursor 41 . We found that strengthening the upper part of pri-mir-18a stem by GC base-pairs blocks miRNA processing, whereas disruption of the base-pairs enhances Microprocessor cleavage efficiency. It is noteworthy, that the partial unwinding of the apical RNA helix by binding of hnRNP A1 induces an asymmetry in the stem region that may define in which orientation the pri-mir-18a is recognized by the Microprocessor.
The processing efficiency of pri-mir-18a is context-dependent, suggesting that the sequence and/or structure of 18a as part of the miR-17-92 cluster is not optimal for Drosha processing 8 .
Interestingly, several studies have recently shown that the miR-17-92 cluster adopts a compact tertiary structure, in which individual miRNAs have different expression levels depending on whether they are located on the surface or buried inside the core [42][43][44] . Notably, a recent SHAPE analysis of the miR-17-92 cluster revealed that the terminal loop of pri-mir-18a in the cluster is solvent inaccessible 42 . As the terminal loop corresponds to the sequence that we have identified as the main hnRNP A1-binding site in pri-mir-18a, it is tempting to speculate that binding of hnRNP A1 to the miR-17-92 cluster is associated with a conformational change in the RNA, which can facilitate Drosha cleavage. We propose that the tertiary structure of pri-mir-18a in the context of the miR-17-92 cluster, as well as sequences in the stem and loop region are important determinants of miRNA processing by Drosha. This 20 process is regulated by the trans-acting factor hnRNP A1, which primarily interacts with the conserved terminal loop of pri-mir-18a. The recognition of the pri-mir-18a stem-loop by hnRNP A1 thus adds an additional layer for the regulation of pri-miRNA processing by an RBP, beyond features that have been recently identified [45][46][47] .
In conclusion, our data demonstrate that recognition of a conserved terminal loop RNA sequence in pri-miRNAs by an RBP can strongly modulate miRNA biogenesis by conformational changes and dynamic destabilization induced by RNA binding. Together with few recent reports, this suggests that recognition of pri-miRNAs by RBPs is a general paradigm for context-dependent regulation of miRNA biogenesis and function.

Protein expression and purification
The

Expression vectors
The plasmid pCGT7 hnRNP A1 and pCG T7 UP1 have been previously described 48
Mouse monoclonal anti-tubulin (1:10000, T6199, RRID -AB_477583, Sigma-Aldrich) was used as loading control. miRNA qRT-PCR analysis was performed using the miScript qRT-PCR kit (Qiagen) on total RNA isolated with TRIzol reagent (Life Technologies), and each sample was run in duplicate. To assess the levels of the corresponding microRNAs, values were normalized to 5S RNA. For each measurement, three independent experiments were performed. Reactions were incubated at 4°C for 1 h followed by electrophoresis on a 6% (w/v) non-denaturing gel. The signal was registered with radiographic film or was exposed to a phosphoimaging screen and scanned on a FLA-5100 scanner (Fujifilm).

In vitro processing assays
Pri-miRNA substrates were obtained by in vitro transcription with [alpha- 32

RNA pull-down
RNA pull-down was performed as previously described 49 . In summary, protein extracts from HeLa cells were incubated with in vitro-transcribed RNAs chemically coupled to agarose beads. The incubation was followed by three washes with buffer G (20 mM Tris pH 7.5, 135 mM NaCl, 1.5 mM MgCl2, 10% (v/v) glycerol, 1 mM EDTA, 1 mM DTT and 0.2 mM PMSF).

Footprinting assays
The assays were performed as described earlier 50  RDCs were best-fitted to the structure by singular value decomposition (SVD) using PALES 56 .

X-ray crystallography
Screening representative intensities between 0.08 Å -1 and 0.22 Å -1 were calculated and used as restraints as described above. An approximate scaling factor relating calculated and measured SAXS intensities was estimated by the average ratio between the experimental SAXS intensities as taken from the fit and the first round of calculated SAXS intensities. SAXS intensities were calculated using the Debye formula and standard atomistic structure factors corrected for the effect of the solvent 65 .
Metainference 66 with a Gaussian likelihood per data point on the representative SAXS intensities was applied every 10th step. It is of notice that metainference in the approximation of the absence of dynamics as used here (without replicas and a standard error of the mean set to zero) is equivalent to the Inferential Structure Determination approach 67 . The attributes of the uncertainty parameter were initially set to large values to allow a slow increase of the restrain force. Additionally, an additional scaling factor between the experimental data was sampled using a flat prior between 0.9 and 1.1.
The protein-RNA interface was restrained by harmonic upper-wall potentials centred 3.5 Å applied on the distances between the centres of the respective rings (Phe17 -A4, Phe59 -G5, Phe108 -A9 and Phe150 -G10) with force constants of 1000 kJ/mol. Furthermore, two crystallographic salt bridges (Arg75 -Asp155 and Arg88 -Asp157) between the two RRM domains were restrained with similar potentials centred at a distance of 4 Å applied on the distances between their charged groups. Secondary structures identified by STRIDE 68 from the crystal structure were restrained using an upper wall potential on the rmsd of the backbone atoms of residues involved with a force constant of 10000 kJ/mol, centred at 0 Å.
For the UP1/12-mer complex RDCs restraints were applied using the θ-method 69 . To take into account for the multiple possible alignments of the molecule with the phage, RDCs were calculated as averages over two replicas and a linear restraint with a slope of -20000 kJ/mol was applied on the correlation between the average and experimental RDCs. Each replica was independently restrained with all the formerly introduced restraints.
After 300 preliminary annealing cycles, the refined structure was chosen as the one with the lowest metainference energy among those sampled at 300K in the latest 30 cycles. The quality 27 of the structure was then further assessed using ProCheck 70

and the SAXS/SANS profiles and
RDCs were independently confirmed using Crysol 71 /Cryson 72 and SVD, respectively.

Isothermal titration calorimetry (ITC)
ITC measurements were carried out at 25°C using an iTC200 calorimeter (GE Healthcare). After correction for heat of dilution, data were fitted to a one-site binding model using the Microcal Origin 7.0 software. Each measurement was repeated at least three times.

Static light scattering (SLS)
Static light scattering experiments were performed on a S75 10/300 size-exclusion column

Author Information
Atomic coordinates and structure files for the UP1/12-mer RNA crystal structure have been deposited in the Protein Data Bank (http://www.pdb.org/) with accession code XXX.
The authors declare no competing financial interests.  The dataset is from a single crystal. Values in parentheses are for highest-resolution shell.  Residues corresponding to amide signals that are exchange-broadened in the RNA-bound spectra are colored red.   Ribonuclease T1 at 1.5 U/μL. F and T identify nucleotide residues subjected to partial digest with formamide (every nucleotide) or ribonuclease T1 (G-specific cleavage), respectively. The cleavages intensities generated by Ribonuclease T1 are indicated on the pri-mir-18a secondary structure. The region of the major UP1 footprints is indicated by a blue oval shape.

Figure Legends
(c) A schematic model of the mechanism by which hnRNP A1 facilitates pri-mir-18a processing  cluster, wildtype and mutants), whereas pri-mir-18a mutant with a 5GC clamp does not bind hnRNP A1 (RNA pull-down assay is shown on the right) and is not processed by Drosha. (b) Pri-mir-18a with a 5GC_internal clamp and wildtype terminal loop binds hnRNP A1 but is not processed by Drosha. Pri-mir-18a with triple mutations [U21A/U29A/U34A] binds hnRNP A1 with lower affinity than the wildtype pri-mir-18a but is still efficiently processed by Drosha. (c) In vitro processing assay of pri-mir-16 with 5GC_internal clamp shows efficient processing by Drosha, similar to pri-mir-16 wildtype.  Figure 1 G 3'