RNA pull-down confocal nanoscanning (RP-CONA) detects quercetin as pri-miR-7/HuR interaction inhibitor that decreases α-synuclein levels

Abstract RNA–protein interactions are central to all gene expression processes and contribute to a variety of human diseases. Therapeutic approaches targeting RNA–protein interactions have shown promising effects on some diseases that are previously regarded as ‘incurable’. Here, we developed a fluorescent on-bead screening platform, RNA Pull-Down COnfocal NAnoscanning (RP-CONA), to identify RNA–protein interaction modulators in eukaryotic cell extracts. Using RP-CONA, we identified small molecules that disrupt the interaction between HuR, an inhibitor of brain-enriched miR-7 biogenesis, and the conserved terminal loop of pri-miR-7–1. Importantly, miR-7′s primary target is an mRNA of α-synuclein, which contributes to the aetiology of Parkinson’s disease. Our method identified a natural product quercetin as a molecule able to upregulate cellular miR-7 levels and downregulate the expression of α-synuclein. This opens up new therapeutic avenues towards treatment of Parkinson’s disease as well as provides a novel methodology to search for modulators of RNA–protein interaction.


INTRODUCTION
RNA-protein interactions coordinate the whole of RNA metabolism, including transcription, RNA processing, modification, translation and turnover (1). Dysregulated RNA metabolism can result in serious diseases including cancer (2) and neuropathological conditions (3). In recent years, therapies targeting RNA-protein interactions have shown promising results and some of them are already in clinical trials (4)(5)(6). Among many types of RNAs, short non-coding microRNAs (miRNAs, miRs) that regulate gene expression by imperfect base-pairing to mRNA hold great potential both as biomarkers (7) and therapeutics (8) in multiple human pathologies. The hairpin containing transcripts, known as primary miRNAs (pri-miRNAs), are processed by the Microprocessor complex to become stemloop precursor miRNAs (pre-miRNAs) (9)(10)(11)(12)(13). The intermediates are further cut by an RNase III enzyme DICER, generating miRNA duplexes (14,15). Only the final product, the single-stranded mature miRNAs together with the Argonaute proteins are functional regulators of gene expression (16).
Parkinson's disease (PD) is an incurable neurodegenerative disease that affects all ages but is most prevalent in the elderly population, with over 1% of those over the age of 60 suffering from this disease (17). One of the main causes behind PD is overproduction and aggregation of a protein called ␣-synuclein (␣-Syn), expressed from the SNCA gene in the brain cells of affected individuals (18)(19)(20)(21). There is a large body of evidence that decreasing the levels of ␣-Syn should be beneficial for PD patients, and several clinical trials are now focusing on ␣-Syn clearance with small molecules, antibodies or vaccines (22). Notably, miR-7 has been shown to target ␣-Syn production (23) and is significantly downregulated in the substantia nigra pars compacta (SNpc) of PD patients (24). MiR-7 also targets other genes implicated in PD, including RelA, Sir2 or Nlrp3 (25). For these reasons, approaches for miR-7 replacement therapies have been put forward (26).
The biogenesis of miRNAs is tightly controlled through RNA-protein interactions (27)(28)(29)(30)(31). Our group has identified that an RNA-binding protein (RBP) HuR specifically binds to the Conserved Terminal Loop (CTL) of primary miR-7-1 (pri-miR-7-1). By recruiting another RBP MSI2, HuR increases the rigidity of the pri-miR-7-1 stem loop, therefore, blocking Microprocessor cleavage and preventing production of mature miR-7 (32). HuR-mediated inhibition of miR-7 biogenesis was independently reported by Lebedeva et al. (33). We have also found that the monounsaturated fatty acid-oleic acid (OA), previously found to bind to MSI2 (34), can dissociate the HuR/MSI2 complex from pri-miR-7-1 and facilitate the biogenesis of miR-7 (35). However, OA is not effective at micromolar concentrations, has poor bioavailability when tested in cells and is toxic in high concentrations (36). Thus, it is necessary to identify potent pri-miR-7-1/HuR inhibitors that will elevate miR-7 levels and lead to downregulation of its targets, including ␣-Syn, which could provide alternative solutions to PD therapy.
Here, we developed a fluorescent on-bead screening assay based on RNA Pull-Down Confocal Nanoscanning (RP-CONA) to identify small molecules that modulate the strength of RNA-protein interactions. Our method uses an ultra-sensitive RNA-protein pull-down assay in cell extracts from human cultured cells (37) detecting RNAprotein complex modulators by confocal microscopy (38)(39)(40)(41). We have demonstrated that the method works for various RNA-protein complexes and screening platforms. By employing RP-CONA, we identified quercetin, a natural flavonoid and known inhibitor of HuR/TNF-␣ mRNA interaction (42), as the most potent pri-miR-7-1/HuR interaction inhibitor. Quercetin induces miR-7 level and inhibits ␣-Syn expression in an HuR-dependent fashion. In summary, in this study we introduce RP-CONA as a new assay technique for the identification of small molecule modulators of RNA-protein interactions. We identified novel inhibitors of HuR/RNA complexes and have discovered a validated hit compound towards attenuation of ␣-Syn levels in PD.

Chemicals
The in-house 54-compound library was kindly provided by Professor Neil Carragher (The University of Edinburgh), including FDA-approved drugs and natural products of well-established anti-cancer mechanisms. The concentrations of library compounds were varied from 0.1 to 10 mM according to their optimal effects in previous cell studies in the Carragher's laboratory. Ro 08-2750 was purchased from R&D Systems. Oleic acid, quercetin, luteolin, genistein, dihydrotanshinone I (DHTS), CMLD-2, cetylpyridinium chloride (CPC) and gossypol were purchased from Sigma-Aldrich. All nine reagents were dissolved in DMSO to prepare 20 mM stock solutions.

Plasmid construction
Human genomic DNA was isolated from HeLa cells using a GenElute™ Mammalian Genomic DNA Miniprep Kit (Sigma-Aldrich). Target genes were PCR amplified from genomic DNA using Phusion ® High-Fidelity DNA Polymerase (NEB) and propagated following the instructions of CloneJET PCR Cloning Kit (Thermo). The HuR open reading frame was inserted downstream from the mCherry gene between XhoI and EcoRI in a pJW99 plasmid. DNA segments encoding miRNA stem loop sequences were cloned into a pCG plasmid (32) between the XbaI and BamHI cleavage sites. The 3 -UTR of human ␣-Syn mRNA gene was inserted into a psiCHECK-2 plasmid between the XhoI and NotI sites downstream the Renilla luciferase (Rluc) gene. Sequence confirmation was carried out in Edinburgh Genomics or GENEWIZ.

Western blot analysis
Cultured cells were harvested and resuspended in Roeder D (200 mg/ml glycerol, 100 mM KCl, 0.2 mM EDTA, 100 mM Tris pH 8.0, 500 M DTT and 200 M PMSF). Cell lysis was carried out in a Bioruptor® Plus sonication device (Diagenode) for 10 min (low intensity settings, 30 s on/off). Sixty micrograms of proteins in cell lysates were separated on a NuPAGE™ 4-12% Bis-Tris Protein Gel (Invitrogen) and transferred onto a nitrocellulose membrane (GE) in a GENIE® blotter (Idea Scientific) at 12 V for 1 h. The membrane was blocked with 1:10 Western Blocking Reagent (Roche) in TBST (20 mM Tris pH 7.5, 137 mM NaCl and 0.1% (v/v) Tween 20). Proteins were detected with the following primary antibodies in TBST containing 1:20 Western Blocking Reagent, including rabbit polyclonal anti-HuR (Millipore), rabbit polyclonal DHX9 antibody (Proteintech), monoclonal anti-␣-tubulin antibody (Sigma-Aldrich) and purified mouse anti-␣-Synuclein (BD Biosciences). Following three washes in TBST, the membrane was incubated in the horseradish peroxidase (HRP) conjugated secondary anti-rabbit or anti-mouse IgG antibodies (Cell Signalling Technology) and developed with chemiluminescent substrate (Thermo #34580). Quantification of western blot bands were carried out in Image Studio Lite Ver 5.2.

RNA quantification
Total RNA was isolated from cells or cell extracts following manufacturer's instruction of TRI Reagent™ Solution (Invitrogen) or TRI Reagent ® LS (Sigma-Aldrich), respectively. To quantify miRNA levels, reverse transcription and quantitative real-time PCR (qRT-PCR) were performed with miScript II RT Kit and SYBR Green PCR Kit respectively (QIAGEN). To quantify ␣-Syn mRNA levels, the GoTaq® 1-Step RT-qPCR System (Promega) was used with the primers (F: 5 -gttgtggctgctgctgagaaa; R: 5 -tccctcct tggttttggagcctac). The qRT-PCR reactions were performed in a Roche LightCycler ® 96 System.

Dual luciferase assays
Luciferase reporter assays were performed according to the manufacturer's instruction of Dual-Luciferase ® Reporter Assay System (Promega). 1.2 × 10 4 of HeLa cells were plated in a well of 96-well plates. For each well, 30 ng of psiCHECK2-␣-Syn-3 UTR were co-transfected with 16.7 ng of pCG-pri-miRNA or 50 ng pCDNA-HuR plasmids in HeLa cells. Cells were lysed 48 hours after transfection by Passive Lysis Buffer (Promega). The luminescence levels of firefly and Renilla luciferases were recorded by a Po-larStar OPTIMA Multidetection Microplate Reader (BMG LABTECH). The mRNA levels of luciferases were determined by qRT-PCR with following primers (Renilla-F:ggaa tgggtaagtccggcaa; Renilla-R:ccaagcggtgaggtacttgt; firefly-F:gaacagctctgggtctaccg; firefly-R:gggatgatctggttgccgaa).

Generation of gene knockout cells
To knockout HuR in HEK293T and HeLa cells, a pair of guide RNAs (5 -cgaagucuguucagcagcau and 5 -cuuggguc auguucugaggg) targeting the exon2 of human HuR gene were designed. The Alt-R® CRISPR-Cas9 crRNAs and tracRNA were synthesized by IDT. 100 M of each cr-RNA was mixed with 100 M of tracRNA in 100 l duplex buffer (IDT) at 95 • C for 5 min to form two crRNA-tracRNA duplexes. HEK293T or HeLa cells were seeded in a 24-well plate. 1.5 l of each duplex were co-transfected with 1 l of GeneArt™ CRISPR Nuclease mRNA (Invitrogen). The cells were diluted and aliquoted to 96-well plates to make <1 cell count per well and lysed by 30 l of Passive Lysis Buffer (Promega) when confluent. 2 l of the cell extracts were loaded onto nitrocellulose membranes and tested against HuR and DHX9 antibodies. Cells showing HuR negative and DHX9 positive were selected and confirmed using western blot analysis. The genomic DNA of the knockouts were extracted and fragments covering the expected HuR knockout sites were PCR amplified using the forward primer (5 -gccctggacagtacactcgcc) and reverse primer (5 -ccacatggccgaagactgca). After ligating into the cloning vector using the CloneJET PCR Cloning Kit (Thermo), the DNA fragments were sequenced, and the mutations were identified.
To knockout pri-miR-7-1 in HeLa cells, a pair of guide RNAs (5 -acauucaauacuaaucuugc and 5 -accaaucauuuguc cuguag) was designed flanking the stem loop sequence of human pri-miR-7-1 gene. Transfection of the CRISPR-Cas9 system was carried out as described before. Crude DNA was extracted by dissolving cells in solution 1 (25 mM NaOH, 0.2 mM EDTA) at 98 • C for 1 h and terminated by equal volume of solution 2 (40 mM Tris-HCl, pH 5.5). PCR was performed with primers flanking the targeted region (F: 5 -ctgcagaacaggtcagtttaagtt, R: 5 -tgcagaac acctatgaagcaga). The PCR products were visualized on an agarose gel and cells generating band shifts were selected. The miR-7 levels were tested by qRT-PCR. Sequences were determined using PCR products amplified from purified genomic DNAs of putative pri-miR-7-1 knockouts.
HuR KO-HEK293T cells were plated in a p150 dish and transfected with 40 g of pJW99-HuR plasmids. The cell lysates were obtained 24 h after transfection by sonication. The concentration of mCherry-HuR in the lysates was quantified following the instruction of an mCherry quantification kit (BioVision). The lysates containing 300 nM of mCherry-HuR were diluted in 20 l of no glycerol-Roeder D (100 mM KCl, 0.2 mM EDTA, 100 mM Tris pH 8.0, 0.5 mM DTT and 0.2 mM PMSF) and mixed with 30 l of pulldown solution (1.5 mM MgCl 2 , 25 mM creatine-phosphate, 0.5 mM ATP, 0.25 l Ribolock RNase Inhibitor (Invitrogen)) before loaded to a well of a black 384-well plate (SWISSCI or Greiner). 0.5 l of compounds (dissolved in DMSO at 100 times of the final concentration) were added to the cell lysates and mixed vigorously at room temperature, 1500 rpm for 20 min.

RNA pull-down assay
RNA pull-down assay was carried out to detect binding of RNA-binding proteins, from whole cell extracts, to RNA immobilized on the beads. 500 pmol of pri-miR-7-1-CTL (5 -uguuguuuuuagauaacuaaaucgacaacaaa) was treated with 100 mM NaAc and 5 mM sodium (meta)periodate in 200 l of water and rotated for 1 h at room temperature in the dark. The RNA was precipitated by adding 600 l of 100% ethanol and 15 l of 3 M NaAc in dry ice for 20 min, followed by centrifugation at 13 000 rpm, 4 • C for 10 min. The RNA pellet was washed with 70% ethanol and resuspended in 500 l of 100 mM NaAc pH 5. 200 l of adipic acid dihydrazide-agarose (Sigma-Aldrich) was washed with 100 mM NaAc and mixed with 500 l of the periodate oxidized pri-miR-7-1-CTL overnight at 4 • C in the dark. The pri-miR-7-1-CTL-beads were washed by 2M KCl, Buffer G (20 mM Tris-HCl pH 7.5, 137 mM NaCl, 1 mM EDTA, 1% Triton X-100, 10% glycerol, 1.5 mM MgCl 2 , 1 mM DTT and 200 M PMSF) and Roeder D, respectively.
250 l of HeLa cell lysates containing 1 mg of total protein were pre-incubated with 100 M of test compounds and 400 l of pulldown solution at 37 • C, 700 rpm for 20 min. The pri-miR-7-1-CTL-beads were incubated with the treated cell lysates, at 37 • C, 1200 rpm for 30 min. After washing with Buffer G, the beads were mixed with 6 l of NuPAGE™ Sample Reducing Agent, 15 l of LDS Sample Buffer and 39 l of water. Proteins captured by RNA were denatured at 70 • C, 1000 rpm for 10 min. 30 l of the supernatant was loaded onto an SDS-PAGE gel and western blot was performed to detect the level of proteins.

RNA immunoprecipitation (RIP) assay
HuR-bound pri-miR-7-1 and ␣-Syn mRNA were immunoprecipitated following a method developed from a previously published protocol (45). In brief, 1.4 × 10 6 HuR-KO HeLa cells were plated in p100 dishes, treated with DMSO or 20 M quercetin, and transfected with 500 ng of pCDNA-HuR. RNA/proteins were cross-linked using 1% formaldehyde and incubated at room temperature for 10 min with rocking. Crosslinking reactions were stopped by the addition of 0.25 M glycine (pH 7.0) followed by incubation at room temperature for 5 min. The cells were resuspended in 500 l of RIPA buffer (50 mM Tris-Cl, pH 7.5, 1% Nonidet P-40 (NP-40), 0.5% sodium deoxycholate, 0.05% SDS, 1 mM EDTA, 150 mM NaCl), sonicated and centrifuged to obtain cell lysates as described above. The protein levels were determined using the Pierce® BCA Protein Assay Kit (Thermo). 50 l of Dynabeads™ Protein A for Immunoprecipitation (Invitrogen) were coupled with 3 l of HuR antibody in BWB (0.02% Tween-20 in PBS) for 30 min at room temperature with rocking, then washed with RIPA buffer. The extracts containing 100 g of protein were diluted with 500 l of RIPA buffer, mixed with the antibody-coated beads at room temperature for 30 min. The beads were collected at 6000 g and the supernatant was kept for RNA extraction as loading control. The beads were washed five times with high-stringency RIPA buffer (50 mM Tris-Cl, pH 7.5, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS, 1 mM EDTA, 1 M NaCl, 4 M urea and 0.2 mM PMSF). The beads containing the immunoprecipitated samples were collected and resuspended in 100 l of 50 mM Tris-Cl, pH 7.0, 5 mM EDTA, 10 mM dithiothreitol (DTT) and 1% SDS. The beads were then incubated at 70 • C for 45 min to reverse the crosslinking. The RNA was extracted from these samples using TRI Reagent® LS according to the manufacturers protocol. qRT-PCR was performed to detect pri-miR-7-1 and ␣-Syn mRNA levels. mRNA stability assay About 1.44 × 10 5 WT or HuR KO HeLa cells were plated in 12-well plates and treated with DMSO or 20 M of quercetin, together with 10 g/ml actinomycin D (Sigma-Aldrich). The cells were collected at 0, 6, or 12 h after actinomycin D treatment. Total RNA was isolated using TRI Reagent™. ␣-Syn mRNA and 18S levels were quantified using qRT-PCR.

MiR-7 is a major inhibitor of ␣-Syn expression
MiR-7 and other miRs, such as miR-133b, miR-153, miR-34b or miR-34c, which also have the potential to target ␣-Syn, have been shown to be downregulated in PD, thus allowing ␣-Syn overproduction and accumulation (23,25,(46)(47)(48). We collated all validated and predicted miRs targeting the 3 -UTR of ␣-Syn mRNA and cloned their pri-miR sequences into the pCG expression vector ( Figure 1A). We used a dual luciferase reporter assay in HeLa cells with overexpression of individual miRs targeting ␣-Syn mRNA 3 -UTR coupled with Renilla luciferase mRNA. This assay showed that only miR-7 exhibited significant inhibition  (47) or provided by miRTarBase (48). Sites highlighted by the red arrows were previously reported. (B). MiR-7 inhibited luminescence of a dual luciferase reporter bearing the gene of ␣-Syn mRNA 3 -UTR downstream the Renilla luciferase gene. Equal amount of each pCG-pri-miR plasmid was co-transfected with the luciferase reporter. PCG-pri-miR-9 was tested as a negative control. Luciferase levels were recorded 48 h after transfection. Mean Renilla/firefly values and SEM from six independent repeats are shown. (C). Single-nucleotide mutation of miR-7 binding site inactivated the inhibitive effects of miR-7. The nucleotide on the third position of miR-7 targeted seed regions on ␣-Syn mRNA 3 -UTR gene was mutated individually. The mutants were numbered according to the binding sites from 5 to 3 of ␣-Syn mRNA 3 -UTR. Luciferase levels were measured 48 h after the co-transfection of pCG-pri-miR-7-1 plasmids with reporters bearing wild-type or mutated ␣-Syn mRNA 3 -UTR gene. The luciferase levels were relative to co-transfection of pCG-pri-miR-9 with wild-type reporter. Mean Renilla/Firefly values and SEM from three independent repeats are shown. (D and E). ␣-Syn expression was inhibited by upregulated miR-7. 1: Mock HeLa cells without DNA transfected. 2-6: An increasing amount of pCG-pri-miR-7-1 was transfected into HeLa cells. The expression of ␣-Syn and ␣-tubulin were detected by western blot 48 h after transfection. Relative ␣-Syn/␣-tubulin levels were normalized to mock. Mean values and SEM from three independent repeats are shown. Statistically significant differences compared to mock were interpreted by SPSS one-way ANOVA, with post hoc LSD test, *P < 0.05, **P < 0.01, ***P < 0.001.
(>50%) of Renilla luciferase compared to other miRs, when equal amounts of pri-miR plasmids were transfected (Figure 1B). The upregulation of mature miR-7 was equal or less when compared with other miRs ( Supplementary Figure S1A). Moreover, a single-nucleotide mutation in the previously identified miR-7 binding site (miR-7 s1) could desensitize ␣-Syn mRNA 3 -UTR to miR-7 overexpression, while other miR-7 binding sites (miR-7 s2 and miR-7 s3) did not present significant differences between the wildtype and mutated reporters ( Figure 1C). Interestingly, mutations of miR-133b s1 and miR-153 s1 sites exerted significant upregulation of expression from the ␣-Syn mRNA 3 -UTR Renilla luciferase construct, proving that miR-133b and miR-153 are involved in regulating ␣-Syn levels, albeit to lesser extent and with more complex regulatory networks (Supplementary Figure S1B). Importantly, a gradient of miR-7 overexpression exerted gradual and significant reduction of the expression of endogenous ␣-Syn in HeLa cells ( Figure 1D, E), which is consistent with a previous publication (23). HeLa cells were chosen because according to Human Protein Atlas these cells have high basal levels of ␣-Syn transcripts when compared with most other cultured cells (49). Other miRs, exemplified by miR-153 and miR-133b, exerted no significant inhibitive effects on endogenous ␣-Syn levels, in spite of their previously reported function as ␣-Syn inhibitors (Supplementary Figure S1C) (50,51). In summary, we conclude that miR-7 is the most potent ␣-Syn suppressor and uses a well-defined target site on the ␣-Syn mRNA 3 -UTR.

RP-CONA: a novel, lysate-based scanning microscopy onbead screening platform for small molecules that modulate RNA-protein complexes
Due to the supremacy of miR-7 in inhibiting the expression of ␣-Syn, it will be crucial to pursue therapeutic approaches for PD through the miR-7/␣-Syn pathway. We hypothesize that as the biogenesis of pri-miR-7-1 is inhibited by HuR through an interaction with the Conserved Terminal Loop (CTL), small molecules that dissociate HuR from miR-7 CTL will facilitate the production of mature miR-7, which in turn will contribute to a reduction of ␣-Syn levels (32).
To identify pri-miR-7/HuR inhibitors, we developed a screening platform that combines techniques of RNA Pull-Down (RP) from eukaryotic cell extracts, and on-bead scanning confocal microscopy (CONA), given the name RP-CONA ( Figure 2). First, we attached 5 FITC and 3 biotin-tagged pri-miR-7-1-CTL to 6 × His-Streptavidin Ni-NTA agarose beads. Then we used HuR KO HEK293T cells, generated with CRISPR-Cas9 targeting, to overexpress mCherry-HuR (Supplementary Figure S2A-C). The HEK293T cells were chosen as they are known to support highly efficient recombinant protein production (52). Moreover, the HuR KO cells were used to avoid signal dilution from an untagged, endogenous HuR. Next, we performed RNA pull-down and imaged the fluorescently labelled RNA and mCherry-HuR on a confocal imaging system. Fluorescent rings/halos on the periphery of microbeads indicate binding events in separate detection channels. The amounts of bound RNA or protein were detected from fluorescence emission intensities of the rings/halos. Pri-miR-7-1-CTL/mCherryHuR inhibitors are expected to attenuate mCherry ring intensities without affecting FITC levels.
We first tested the RP-CONA assay using the Opera HCS instrument. The preliminary results showed that RP-CONA generated FITC rings in the presence of the fluorescently tagged RNA, and mCherry rings only when mCherry-HuR was pulled down by pri-miR-7-1-CTL (Figure 3A, B). No overlapping signals were observed in different detection channels. The addition of an increasing concentration of untagged pri-miR-7-1-CTL competitively decreased mCherry signals ( Figure 3C, D). Oleic acid, the known pri-miR-7-1/HuR complex inhibitor reduced mCherry ring intensities at high millimolar concentrations, as reported before ( Figure 3E, F) (35). To show that RP-CONA can be run on various scanning instruments, we then switched to another confocal image platform, the ImageX-press, which gave similar quality of images compared to Opera HCS (Supplementary Figure S3A, B). Moreover, the relative mCherry/FITC value was proportional to the concentration of mCherry-HuR (Supplementary Figure S3C-E). Importantly, if we replaced the CTL of pri-miR-7-1 with the non-conserved Terminal Loop (TL) of pri-miR-30a that does not bind HuR ( Figure 4A, B), or overexpressed mCherry instead of mCherry-HuR to make the cell extracts ( Figure 4C), the mCherry rings were not observed. Moreover, the addition of polyclonal anti-HuR antibody to FITC-pri-miR-7-1-CTL/mCherry-HuR complex significantly decreased the mCherry ring intensities, compared with the control anti-Lin28a antibody ( Figure 4D, E). This evidence proves that RP-CONA is a sensitive and specific method to probe pri-miR-7-1-CTL/HuR interactions.
These results demonstrate that RP-CONA is a robust technique to screen for RNA-protein interaction modulators and that it could be used with various confocal image scanning platforms.

Pri-miR-7/HuR inhibitors identified by RP-CONA screening of a focused library
MSI2 and HuR inhibit miR-7 biogenesis and the level of DROSHA cleavage (32). A range of compounds have been recognized to inhibit HuR or MSI2 binding to their target mRNAs (6). By collecting the commercially available HuR/MSI2 inhibitors, we built up a focused library and tested them using RP-CONA at 100 M concentration ( Figure 6A). The Z' factor of the screen is 0.93, and the coefficients of variation (CVs) of the negative control (DMSO) and positive control (untagged pri-miR-7-1-CTL) equal to 1.75% and 0.96%, respectively ( Figure 6B). As reported previously (35), OA did not show significant disruptive effects on pri-miR-7-1/HuR complex at 100 M. Importantly, we identified quercetin, luteolin and gossypol as pri-miR-7-1/HuR inhibitors, which generated >50% inhibition at 100 M concentration according to the relative mCherry/FITC signals compared to the DMSO control ( Figure 6C). In summary, RP-CONA identified new potential inhibitors of pri-miR-7-1-CTL/HuR complexes.

A small-scale prototype RP-CONA screen of pri-miR-7-1/HuR inhibitors
In order to analyse RP-CONA's ability to scale up and to establish the sensitivity of the method, we performed a smallscale prototype screen using an in-house library containing 54 FDA-approved drugs or natural products, together with eight compounds from the previous focused screen ( Figure  7A). Here, we applied quercetin as a positive control and DMSO as a negative control. The CVs of the negative and positive controls were 6.70% and 2.18%, respectively. Most compounds did not show significant stabilization or destabilization of the pri-miR-7-1-CTL/HuR complex. Additionally, we identified genistein as a new hit, albeit not as effective as quercetin or luteolin. To validate the active compounds identified in the primary screen, we tested them in a standard pull-down assay where pri-miR-7-1-CTL was covalently linked to the beads (56). Importantly, this assay, unlike RP-CONA, only gives semi-quantitative readout. The western blot analysis of the HuR pull-down confirmed the strong disruptive effects of quercetin and luteolin ( Figure  7B). Similar effects from DHTS, Ro and genistein were also observed. Meanwhile, gossypol largely reduced pulldown of the RNA helicase DHX9, implying the effects of this compound are unspecific. Our most effective inhibitors quercetin and luteolin are close analogues ( Figure 7C). We tested quercetin and luteolin in RP-CONA at a range of concentrations. The RP-CONA signals were reduced in a dose-dependent manner for both quercetin and luteolin, showing the half maximal inhibitory concentrations (IC 50 s) at 2.15 ± 0.16 M and 2.03 ± 0.25 M, respectively ( Figure  7D). These observations indicate that RP-CONA primary hits were successfully validated using a classic biochemical method and further confirm quercetin and luteolin as the most promising inhibitors of the pri-miR-7-1/HuR complex.

Quercetin inhibits cellular ␣-Syn and upregulates miR-7
Subsequently, we tested quercetin and luteolin in HeLa cells at 20 M concentration. Quercetin treated cells had a significantly reduced level of ␣-Syn protein, with a 1.5-fold upregulation of mature miR-7 level ( Figure 8A-C). We found that quercetin induced pre-miR-7-1 but not pri-miR-7-1 levels >2-fold (Supplementary Figure S5A, B). The expression of DGCR8 and DICER were not significantly changed after quercetin treatment, albeit DROSHA levels were slightly upregulated (Supplementary Figure S5C). This suggests the amount of Microprocessor and DICER remained largely unaffected and that quercetin rescues the inhibitive effect of HuR that blocks the pri-miR-7-1 processing. Importantly, the quercetin-mediated dissociation of HuR/MSI2 with pri-miR-7-1 was confirmed by RP-SMS (RNA Pull-down SILAC Mass Spectrometry (37)) (Supplementary Figure  S5D). Interestingly, luteolin had no significant effect on ␣-Syn or miR-7, suggesting different bioavailability, dynamics or cellular metabolism when compared with quercetin (Figure 8A-C). Beads images of blank beads; blank beads incubated with cell lysates containing mCherry-HuR; FITC-pri-miR-7-1-CTL-beads incubated in lysates-free buffer; and FITC-pri-miR-7-1-CTL-beads after mCherry-HuR pulldown. (C and D) mCherry/FITC signals were reduced by untagged pri-miR-7-1-CTL. Cell lysates containing mCherry-HuR were treated with an increased concentration of untagged pri-miR-7-1-CTL before pull-down. (E and F) mCherry/FITC signals were reduced by a pri-miR-7-1/HuR inhibitor. Cell lysates containing mCherry-HuR were treated with DMSO or an increased concentration of OA before pull-down. All values were obtained from three technical repeats. Mean mCherry/FITC ring intensities and SD between triplicates are shown. Statistically significant differences compared to 0 M untagged pri-miR-7-1 or DMSO were interpreted by SPSS independent sample t-test, ****P < 0.0001.
Finally, quercetin reduced luciferase levels in our dual luciferase reporter assay with ␣-Syn mRNA 3 -UTR coupled with Renilla luciferase ( Figure 9E). The overexpression of HuR upregulated Renilla luciferase level by 2-fold, which was reduced to the mock control levels upon quercetin treatment. Notably, the relative levels of the Renilla luciferase mRNA were not enhanced by HuR ( Figure 9F), implying HuR induces the translation of ␣-Syn through the 3 -UTR, and this can be reversed by quercetin. In the stability assay, quercetin treatment only slightly destabilized the ␣-Syn mRNA in WT HeLa cells after actinomycin D treatment for 6 h, while exerted no effects in the HuR KO cells ( Figure 9G). In conclusion, quercetin interrupts the ␣-Syn-ARE/HuR interaction, suppressing the HuR-engaged ␣-Syn expression mainly at the translational level with some contribution from regulating mRNA levels. However, the contribution of quercetin-mediated induction of miR-7 towards destabilization of ␣-Syn mRNA could be still important in other cellular systems such as mature neurones.

DISCUSSION
␣-Syn has become a popular target for investigational PD therapies, with RNA interference (RNAi) strategies focusing on ␣-Syn repression presently at preclinical stages (58). An shRNA therapeutic achieved a 35% knockdown of SNpc ␣-Syn, and showed protective effects in a PD rat model without notable toxicity (59). However, neuronal toxicity has been found in animal brains when another siRNAinduced ␣-Syn knockdown reached 90% (60,61). Therefore, the knockdown efficiency of ␣-Syn-targeted RNAi needs to be tightly controlled within an appropriate range in order to reach a compromise of both safety and efficacy (61,62). MiRNAs, known as fine tuners of gene expression, buffer the expression network against environmental or genetic stress (63). Here we provide evidence that miR-7 is the most effective suppressor of ␣-Syn expression, among all potential ␣-Syn targeting miRs that are downregulated in PD ( Figure 1). Therefore, fine-tuning ␣-Syn expression by elevating miR-7 level could provide a novel avenue for the development of PD therapy.
Previously we have shown that miR-7 biogenesis is inhibited through an interaction between the pri-miR-7-1-CTL and HuR (32). The best elucidated miR posttranscriptional regulation is between let-7 and Lin28 proteins. A range of screens targeting let-7/Lin28 have been carried out using fluorescence resonance energy transfer (FRET) or fluorescence polarization (FP) methods, and some hits have shown promising anti-cancer properties by dissociating the RNA-protein interaction and upregulating let-7 levels (6). However, due to a lack of understanding of the interaction domains where pri-miR-7 binds HuR, it seems challenging to identify pri-miR-7-1/HuR inhibitors using similar approaches. To tackle this problem, we developed the RP-CONA technique that combines the RNA pull-down assay in human cell extracts with the CONA screening platform (37)(38)(39).
The RNA pull-down assay captures proteins by on-bead RNA fragments from eukaryotic cell lysates (37). The application of cell extracts instead of purified protein allows the screening to take place in a more physiological environment and might help avoiding false positives that are not functional in vivo. Moreover, this technique removes the need of protein purification, making it an alternative solution to study RNA/protein binding events for those proteins with difficulty in purification. CONA is a previously established, sensitive and quantifiable on-bead imaging platform (64,65). The only known ligand targeting the RNA recognition motif-3 (RRM3) of HuR was identified by CONA from one-bead-one-compound libraries (66). Notably, using the RRM3 binder and CONA, an ATP-binding pocket was discovered in RRM3, which was not detected using conventional binding assays (66,67). CONA is also highly flexible, and it can be modified to monitor complex biological processes in real time, such as ubiquitination-related enzymatic activities and aggregation of ␣-Syn (38,40). A recent publication successfully identified protein binders from natural product extracts with CONA (41). Here, for the first time we applied CONA to study RNA-protein interactions by using a lysate-based screening strategy. We believe that this combination has a high potential for further exploitation in miniaturized drug screening. RP-CONA is a versatile technique that is not limited to monitor the pri-miR-7/HuR activity. We have shown that RP-CONA is applicable in different RNA/protein complexes such as pre-let-7a-1/Lin28a ( Figure 5), ␣Syn-ARE/HuR ( Figure 9C) and TNF␣-ARE/HuR (Supplementary Figure S7). This pro- ␣-Syn and GAPDH mRNA levels were determined by qRT-PCR. Mean ␣-Syn/GAPDH and SEM from three independent repeats are shown. Statistically significant differences compared to DMSO or between quercetin treated cells were interpreted by SPSS independent sample t-test, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. vides an alternative method to look for RNA-protein modulators in the environment of the mammalian cell extracts.
In order to look for pri-miR-7-1/HuR inhibitors, we started with a focused prototype screen. The library contained eight compounds that were known to interfere with the RNA binding activities of HuR or MSI2, as both of the proteins are essential during the blockage of miR-7 biogenesis (6). We found half of the compounds resulting in >50% inhibition of RP-CONA signals ( Figure 6). By integrating the focused library with a diverse in-house 54-compound library, we have shown that the former 4 compounds were still effective, and only one additional hit was discovered from the new library. The disruptive effects of these hits were confirmed in RP-CONA with a gradient of concentrations, and the IC 50 s of the most effective hits quercetin and luteolin were around 2 M. We subsequently validated the primary hits using the standard RNA pull-down assay followed by western blot and managed to exclude the unspecific disruptor gossypol, which also interrupted the binding of the reference protein DHX9 (Figure 7). Previously, quercetin was recognized as a pre-let-7g/Lin28a inhibitor with <2.5 M IC50s in EMSA and a fluorescence intensity-based binding assay (68). However, in our hands 100 M of quercetin only mildly inhibited Lin28a binding to pre-let-7a-1-CTL in RP-CONA (Supplementary Figure S8), suggesting that quercetin is more effective towards pri-miR-7-1/HuR in cell extracts.
Quercetin has been shown to inhibit the interaction between HuR and TNF-␣ mRNA, with an IC 50 of 1.4 M in RNA electrophoretic mobility gel shift assay (EMSA) (42). Luteolin binds the RRM1 of MSI1, the paralogue of MSI2, with a dissociation constant (K d ) of around 3 M (69). Interestingly, luteolin is a close analogue of quercetin. In an FP assay, both luteolin and quercetin interrupted MSI1-RRM1 binding with a short RNA motif at micromolar and millimolar IC 50 s, respectively (69). These findings imply that quercetin is likely to be a dual-inhibitor of both HuR and MSI2, marking it a strong candidate as an miR-7 enhancer. As expected, we identified a 1.5-fold upregulation of mature miR-7 in HeLa cells after 20 M of quercetin treatment ( Figure 8C). This is consistent with what we saw when HuR or MSI2 or both were knocked down (32). At the same time, a significant downregulation of ␣-Syn expression was observed ( Figure 8A, B). It has been reported that quercetin exerts neuroprotective effects in different cell or animal models of PD (70)(71)(72). Moreover, oxidized quercetin can prevent ␣-Syn fibrillization in vitro (73). Interestingly, in neuron-like PC12 cells ␣-Syn expression was induced by 50-500 M of quercetin, albeit reduced when quercetin reached 1 mM (74). These discrepancies may be attributed to the selection of cell lines. We also used quercetin to treat human neuroblastoma SH-SY5Y and neuron-like mouse embryonic carcinoma P19 cell lines. The reduction of ␣-Syn expression was notable but was less significant than that observed in HeLa cells (Supplementary Figure S5E). As mentioned above, HeLa cells were selected due to its abundant expression of ␣-Syn, so the reduced ␣-Syn level could have been easily detected. Additionally, luteolin did not show similar effects of miR-7 induction or ␣-Syn inhibition as its analogue quercetin ( Figure 8A-C). This suggests quercetin and luteolin may present different bioavailability in living HuR bound RNAs from cell lysates were immunoprecipitated by anti-HuR antibody-coated beads and quantified by qRT-PCR. RNA levels were normalized to DMSO treated mock samples. Mean relative RNA levels and SEM from three independent repeats are shown. Statistical tests were interpreted by SPSS independent sample t-test, ** P<0.01. (C and D). Quercetin inhibits ␣-Syn-ARE/HuR binding in RP-CONA. A diagram illustrates the interaction between FITC-␣-Syn-ARE with mCherry-HuR. Cell extracts containing 50 nM of mCherry-HuR were pre-incubated with an increased concentration of quercetin, prior pull-down by FITC-␣-Syn-ARE-beads. Mean mCherry/FITC ring intensities and SD between triplicates are shown. Statistically significant differences compared to DMSO were interpreted by SPSS independent sample t-test, * P < 0.05, **P < 0.01. (E and F). Quercetin inhibits the HuR-induced expression of luciferase bearing the 3 UTR of ␣-Syn mRNA. Mock or HuR overexpressed HeLa cells were treated with DMSO or 20 M of quercetin, and transfected with the dual luciferase reporter carrying ␣-Syn-3 UTR downstream the Renilla luciferase gene. Mean luminescence levels (E) and mRNA levels (F) of Renilla/firefly luciferases, and SEM from 3 independent repeats are shown. Statistically significant differences between groups, or compared to mock DMSO were interpreted by SPSS independent sample t-test, *P < 0.05, **P < 0.01, ***P < 0.001. (G). The effects of quercetin on ␣-Syn mRNA decay. Wild-type and HuR KO HeLa cells were treated with DMSO or 20 M of quercetin, supplemented with 10 g/ml of actinomycin D for 0, 6 and 12 h, respectively. ␣-Syn mRNA and 18S levels were determined by qRT-PCR. Mean ␣-Syn/18S and SEM from three independent repeats are shown. cells, despite their similar performance in RP-CONA. This is reminiscent of the evidence that quercetin was readily incorporated into cells, while another close flavonoid analogue myrincetin had poor cellular uptake (75). Collectively, we present the first report of quercetin as a miR-7 enhancer and ␣-Syn inhibitor at low concentrations.
We hypothesized that quercetin inhibits ␣-Syn through the pri-miR-7-1/HuR pathway. However, we found similar inhibitive effects of quercetin in miR-7 KO HeLa cells, which means the upregulated miR-7 is not the major contributor of ␣-Syn suppression ( Figure 8D). Indeed, quercetin has reduced ␣-Syn expression by around 60% in wild-type HeLa cells, and it would require >1000-fold miR-7 upregulation to achieve a similar extent of ␣-Syn inhibition in the same cell line (Figure 1, Supplementary Figure  S1A). Crucially, we found that ␣-Syn remained unchanged upon quercetin treatment when HuR was absent ( Figure  8E, F). Nevertheless, the upregulation of miR-7 was still observed ( Figure 8G). These results strongly imply that HuR is an essential part in the quercetin-mediated inhibition of ␣-Syn levels, although it seems not connected to the regulation of miR-7 biogenesis in our cellular system. Moreover, the biogenesis of miR-7 is also regulated by some other protein complexes, including SF2/ASF, NF45-NF90 and QKI (76)(77)(78). Therefore, the removal of HuR may allow alternative miR-7 regulation pathways to take over. This may explain why luteolin did not affect miR-7 level in wild-type HeLa, but significantly induced miR-7 in HuR KO HeLa cells.
HuR is known to stabilize target mRNAs through the AREs in their 3 -UTRs (57). A recent study indicates that HuR interacts with the 3 -UTR of ␣-Syn mRNA and increases its stability in an mRNA decay assay in HeLa cells, although the knockdown of HuR only shows mild inhibition on ␣-Syn expression. Importantly, the stabilization seems to be independent of miR-7 (79). In our study, quercetin decreased ␣-Syn mRNA, and it was HuRdependent. At the same time, in HeLa cells, the pri-miR-7-1/HuR pathway is not a major contributor of quercetinmediated ␣-Syn inhibition ( Figure 8H). Using an immunoprecipitation assay, we found that quercetin inhibited HuR from interacting with pri-miR-7-1 in cells, and the effect towards ␣-Syn mRNA was milder ( Figure 9A). Consistently, in RP-CONA, quercetin was effective at 20 M on inhibiting pri-miR-7-1-CTL/HuR ( Figure 7D), but it required 200 M of quercetin to interrupt ␣-Syn-ARE/HuR ( Figure 9D), as well as TNF␣-ARE/HuR (Supplementary Figure S7). This suggests higher affinity between the ␣-Syn-ARE/HuR interaction than pri-miR-7-1-CTL/HuR, or different acting mechanisms of quercetin. Our luciferase assays showed that the overexpressed HuR enhanced the expression of luciferase with ␣-Syn mRNA 3 -UTR at translational level, which could be reversed by quercetin (Figure 9E, F). In an actinomycin D-based stability assay, endogenous HuR only slightly enhanced the stability of ␣-Syn mRNA, which could be eliminated by quercetin (Figure 9G). However, this assay may not be the best model to study the effects of quercetin on ␣-Syn mRNA stability, and the treatment time of 12 h (limited by actinomycin D toxicity) could be too short to induce significant effects. As there is clear evidence of quercetin reducing ␣-Syn mRNA lev-els after 48-h treatment ( Figure 8H), we cannot rule out the possibility that quercetin reduces steady state levels of ␣-Syn mRNA by regulating other post-transcriptional processes including splicing, export or localization. These assumptions are supported by some evidences with HuR regulating mRNA alternative splicing (80), as well as alternative polyadenylation (81). Based on our results we conclude that quercetin interrupts HuR binding with both pri-miR-7-1 and ␣-Syn mRNA, promoting miR-7 biogenesis, inhibiting ␣-Syn translation and reducing ␣-Syn transcripts levels. All of these together contribute to the strong repression on ␣-Syn expression by quercetin.
To summarize, we developed an on-bead screening platform, RP-CONA, for the identification of RNA-protein interaction modulators. We employed RP-CONA to look for inhibitors of pri-miR-7-1/HuR interactions. The most potent hit quercetin was proven to be a miR-7 enhancer as well as an ␣-Syn inhibitor, implying a potential as PD therapeutic. The mechanism of quercetin-mediated inhibition of ␣-Syn is HuR-dependent. Additionally, quercetin displayed positive effects on miR-7 levels, but the achieved upregulation was too small to directly affect ␣-Syn. Further work needs to be focused on the evaluation of quercetin in PD models, especially in the context of mature neurones as well as ␣-Syn overproduction and accumulation. In the meantime, the RP-CONA screening method can be used to identify more potent and specific miR-7 enhancers and ␣-Syn inhibitors. In summary, our work delivers a new platform for the identification of RNA-protein interaction modulators, and highlights quercetin as a potential hit compound towards treatment of PD.

DATA AVALIABILITY
All data generated or analysed during this study are included in this published article (and its additional files). The mass spectrometry proteomics data have been deposited in the ProteomeXchange Consortium via the PRIDE (82) partner repository with the dataset identifier PXD025617. Requests for material should be made to the corresponding author.

SUPPLEMENTARY DATA
Supplementary Data are available at NAR Online.

AKNOWLEDGEMENTS
We would like to express our acknowledgement to Prof. Neil Carragher and Ashraff Makda for kindly providing us with the in-house compound library. We also thank James Longden for helping us with the ImageXpress service.