The cis-acting replication element of the Hepatitis C virus genome recruits host factors that influence viral replication and translation

The cis-acting replication element (CRE) of the hepatitis C virus (HCV) RNA genome is a region of conserved sequence and structure at the 3′ end of the open reading frame. It participates in a complex and dynamic RNA-RNA interaction network involving, among others, essential functional domains of the 3′ untranslated region and the internal ribosome entry site located at the 5′ terminus of the viral genome. A proper balance between all these contacts is critical for the control of viral replication and translation, and is likely dependent on host factors. Proteomic analyses identified a collection of proteins from a hepatoma cell line as CRE-interacting candidates. A large fraction of these were RNA-binding proteins sharing highly conserved RNA recognition motifs. The vast majority of these proteins were validated by bioinformatics tools that consider RNA-protein secondary structure. Further characterization of representative proteins indicated that hnRNPA1 and HMGB1 exerted negative effects on viral replication in a subgenomic HCV replication system. Furthermore DDX5 and PARP1 knockdown reduced the HCV IRES activity, suggesting an involvement of these proteins in HCV translation. The identification of all these host factors provides new clues regarding the function of the CRE during viral cycle progression.

. HCV CRE structure and pull-down strategy. (A) HCV genome sequence corresponding to the C + HV fragment containing the CRE structure and the downstream region for the hybridization of the oligonucleotide used in the pull-down experiments (red). (B) Proteomic method for identifying CRE-interacting candidates. Binding reactions between Huh-7 S10 lysates and CRE were incubated in the presence of streptavidin-coated magnetic beads (Dynabeads) and pulled-down. Some 1/3 of the sample was subjected to SDS-PAGE and the protein stained. Line 1: control samples corresponding to CRE linked to nonbiotinylated asHCV-9414; Line 2: positive samples corresponding to CRE linked to biotinylated asHCV-9414 (b-asHCV-9414). Some 2/3 of the samples were digested, analyzed by LC-MS/MS, and compared two by two as explained in Methods.
sequences. Additionally, in silico binding predictions suggested that subdomain 5BSL3.2 plays a role in the proteins recruitment. Via knockdown experiments it was confirmed that protein heterogeneous nuclear ribonucleoprotein A1 (hnRNPA1) and High Mobility Group Box 1 (HMGB1) exert biological effects on viral replication whereas DEAD box protein 5 (DDX5) and Poly [ADP ribose] polymerase 1 (PARP1) play a role in HCV translation. Given the involvement of these proteins in different pathways of cell metabolism, the present results support the idea that CRE plays a biological function in the viral cycle by interacting with the host factors here identified. This work provides the first evidence of association of a pool of host cell factors with HCV genome through CRE and the involvement of some of them in different stages of the HCV cycle.

Results
Identification of cellular proteins that associate with CRE. To identify eukaryotic proteins able to interact with the HCV CRE, pull-down assays were performed involving CRE and Huh-7 cell lysates. To this end, a genomic HCV RNA fragment comprising the CRE plus the downstream hypervariable region (9181-9414) was transcribed (Fig. 1A). This RNA fragment allows the stem loop conformation of the CRE subdomains, and can be annealed to a biotinylated oligonucleotide at its 3′ end, thus allowing fragment capture by affinity chromatography (Fig. 1A). As a control sample the same RNA molecule was annealed to a non-biotinylated oligonucleotide with the same sequence. The non-biotinylated oligonucleotyde control strategy has already been used by other authors for LC-MS/MS analysis of RNA-protein complexes in the HCV related Japanese Encephalitis Virus, confirming the suitability of this control 25 . S10 lysates were obtained as described in Methods and incubated with the RNA fragment in the presence of streptavidin-coupled electromagnetic beads (Dynabeads) that interact specifically with biotin-containing molecules. To avoid recovering proteins with low binding stability, reactions were performed at low ion concentrations and performing washes under stringent conditions. After washing and elution, some 1/3 of the samples were subjected to SDS-PAGE (Fig.1b). As expected, efficient protein enrichment (compared to the control) was seen for the samples with biotinylated oligonucleotides (Fig. 1B).
To identify the pool of proteins, pull-downs were subjected to trypsin digestion and LC-MS/MS. Biological replicates were compared two by two using PEAKS software, and the proteins enriched in at least three of the four replicates selected (see Methods). As shown in Table 1, more than 50 proteins pulled-down with CRE were identified. In most of the proteins, at least 80% of detected peptides were unique; LC-MS/MS therefore provided very good discrimination. Interestingly, some of these proteins, such as XRCC6, hnRNPA1 and DDX5, have already been proposed to interact with other HCV genome regions and/or proteins 12,24,26 . Functional classification of pulled-down proteins. The candidates that interact with CRE were classified using the Panther Classification System gene list analysis tool. Figure 2A shows the CRE-pulled-down protein classification based on "molecular function" criteria. The most represented gene ontology (GO) terms were  Table 1 "binding" (GO:0005488) (42.1%) and "catalytic activity" (34.2%). As expected, the majority of proteins classified as "binding" were "nucleic acid binding" proteins (GO:003676) (79.5% of the binding group); proteins capable of entering a stable union with CRE were therefore enriched. In addition, a large number of proteins with catalytic activities was seen, with transferases and helicases the most strongly represented. Thus, we can hypothesize a possible biological function for either group of proteins in the viral cycle, mediated by their association with the CRE. An additional classification tool with a broader and both functional and structural focus was used. DAVID bioinformatics resources can extract biological features/meanings associated with large gene lists by simultaneously comparing terms from different annotation categories 27 . The DAVID Clustering Annotation tool revealed eight significant protein clusters for the whole group of pulled-down host factors ( Table 2). The cluster with the highest enrichment score (15.9) was composed of terms related to RNA binding motifs and domains; all had very high significance values. Among the proteins clustered as RNA binding proteins were several heterogeneous nuclear ribonucleoproteins (hnRNPs), Ras-GTPase activating-binding proteins (G3BPs) and splicing factors.
All the proteins belonging to the RNA-binding cluster were checked against the RNA-Binding Protein (RBPDB) database (http://rbpdb.ccbr.utoronto.ca/) to identify the type and number of RNA binding domains involved in the protein-CRE interaction. Interestingly, all showed the presence of RRMs in their sequence (Fig. 2B). The presence of just one RRM seems sufficient to allow pulldown with CRE, as seen for G3BP1 G3BP2, RBM3 and RBM7. Several hnRNPs showed two RRMs in their sequence, but only the hnRNPM protein had more than two RRMs. To search for similarities between RRMs, all 29 protein sequences corresponding to these domains were compared, and two highly conserved regions found (Fig. 3). Together, these data suggest that there is a large group of the pulled-down proteins (29 out of the 55) that contain conserved RRMs.

In silico interaction of RNA-and DNA-binding proteins with CRE.
To further study the associations detected by LC-MS/MS, an in silico interaction study was performed, introducing the CRE sequence to the human RNA-and DNA-binding proteome. catRAPID tools estimate the binding propensity of a protein-RNA pair using an algorithm that considers protein and RNA secondary structure, hydrogen bonding, and contribution of van der Waals forces 28 . The CRE nucleotide sequence was compared against the Homo sapiens proteome using the catRAPID omics tool, searching for interacting RNA-and DNA-binding proteins. Interestingly, 71% (39/55) of the proteins identified in the pull-down experiments were predicted to bind as well to the CRE sequence (Fig. 4). In addition, among these 39 proteins, 30% fit within the upper quartile score, indicating a good reliability on the identification. The in silico binding prediction is in agreement with the pull-down results and reinforces these proteins as CRE-associated factors.
Deeper mapping of the binding sites of the CRE-interacting proteins was performed using nine representative factors detected in both in silico and the pull-down experiments. The selected proteins were analyzed using the catRAPID graphics tool, comparing each candidate protein's whole sequence against the C + HV sequence transcript. Figure 5 shows the aminoacid-nucleotide interaction heat-map for each protein against the CRE. Remarkably, all the proteins examined showed very similar CRE interaction patterns. The majority of the proteins most strongly interacted with the region comprising nucleotides 9280-9330. This region includes the 5BSL3.2 apical loop and the 5′ end of 5BSL3.3. In contrast, the nucleotides downstream of the CRE have no apparent role in these interactions; their catRAPID scores were low. This indicates that, in the C + HV fragment, nucleotides mainly belonging to the CRE show protein binding functionality. It should be noted that, for the RNA-binding proteins, the interacting aminoacids clearly belonged to their RRMs. For instance, hnRNPA1 (RRMs at 14-97 and 105-184) showed its strongest interaction within the first 200 aminoacids. In addition, hnRNPM showed strong overlapping between its RRMs and the interacting regions, especially at its third RRM (653-729). This was the case too for the second RRM of MSI1 (109-186) and both the RRMs of the hnRNPDL (148-230 and 233-312). Together, these findings suggest a clear involvement of CRE 5BSL3.2 and 5BSL3.3 in CRE-protein association, mediated by RRMs in the case of the RNA-binding proteins.

Effect of gene modulation on HCV replication.
To study the potential effect of these proteins in the viral replication, gene silencing assays were performed. Six proteins were selected based on previously suggested HCV interactions or a possible role in the viral cycle, and knockdown experiments performed on them. To this end, a cell line carrying a stable HCV subgenomic replicon (Huh-7 NS3-3′ ) was transfected with siRNAs designed against the mRNA of these proteins. For this purpose those siRNA sequences that have already been proved to be effective were used [29][30][31] . The specific effect of each siRNA was confirmed by immunoblotting (Fig. 6A). As can be seen, all the siRNAs reduced their target protein levels (from 40% to 70%) without significantly affecting any of the other assayed proteins. RNA was extracted from cells treated with the different siRNAs and HCV replicon levels quantified by qRT-PCR (Fig. 6B). Interestingly, hnRNPA1 gene knockdown resulted in a two-fold increase in HCV replicon levels compared to those associated with a non-targeting siRNA. The same tendency, but with only a 1.5-fold increase, was observed in cells transfected with siHMGB1. No significant effect was detected for G3BP1, DDX5 and PARP1 knockdown compared to results obtained with non-targeting (NT) siRNA.
To confirm the knockdown results we selected hnRNPA1 and HMGB1 and cloned their respective coding genes into the pcDNA3 eukaryotic expression vector. Huh-7 cells were transfected with pcDNA3 (as a control) or the corresponding hnRNPA1-or HMGB1-pcDNA3 constructs. Interestingly, 48 h after transfection a significant inhibition of the replicon levels was detected in both cases (Fig. 6C), reinforcing the results obtained in the positive samples by tryptic peptides (%); Ratio, average of peak area ratios from peptides in positive samples; compared to control samples; HCV gen., other reported HCV genome domains which associate with the protein; Ref., references for the named interactions.
knockdown experiments for these two proteins. These results suggest that hnRNPA1 and HMGB1 negatively regulate HCV replication.
Moreover, direct interaction of hnRNPA1 with CRE was assayed through an RNA-immunoprecipitation protocol (RIP). Results shown in Fig. 6D shows that CRE RNA fragment, but not a control RNA 667, was amplified in immunoprecipitated samples after RIP using an anti-hnRNPA1 antibody. This experiment confirms a direct interaction between hnRNPA1 and HCV CRE. DDX5 and PARP1 knockdown reduce viral translation efficiency. Since CRE was been shown to long-range interact with the IRES and inhibit its translation function 22 , an RNA construct containing a luciferase reporter gene was generated (ICU), to perform translation assays in knock down cells. This molecule contains the HCV IRES linked to the CRE and 3′ UTR through the Firefly luciferase gene 22 (Fig. 7A). After protein knockdown with the specific siRNAs, Huh-7 cells were transfected with the RNA construct, (Fig. 7A), and luciferase activity measured. The luciferase sequence of ICU includes its own stop codon, ensuring the correct translation of the reporter protein. A non-related transcript with a cap structure, containing Renilla luciferase reporter (cap-RLuc) was used for normalization of translation data. Translation efficiency at 4 h post-transfection is shown in Fig. 7B. Knockdown of DDX5 and PARP1 showed nearly of 40% of inhibition of HCV IRES activity. This effect was present, but patently lower, in siHMGB1 treated cells. All these data indicate that DDX5 and PARP1 are required for full HCV IRES translation.

Discussion
Many processes are run by viral genomes through the use of structurally conserved functional RNA domains which form a dynamic, complex network of direct RNA-RNA interactions that may be regulated or stabilized by the recruitment of host or viral factors. In the case of HCV, the CRE region plays an active role in the organization of an RNA-RNA contact network involved in the regulation of viral translation and replication 20    To our knowledge, very few attempts to study CRE-protein association have been made. The present work identifies a pool of 55 host proteins that are pulled-down by an RNA transcript corresponding to the HCV CRE. The majority of these proteins are proposed as HCV-interacting proteins for the first time. Using MALDI-MS, Oakland et al. 24 identified six CRE-interacting proteins in 293T cells. Interestingly, two of these proteins were detected in the present study as well (hnRNPM and XRCC5), and another two (DDX17 and DDX3) are closely related to the reported RNA-helicase DDX5. In the present work, the resolving power of the LC-MS/MS system used not only allowed the identification of many more new proteins, it detected, with high coverage, a large number of unique peptides (Table 1). Moreover, label-free quantification using PEAKS software allowed the positive samples to be compared with the controls and discard any factors that were recovered in a non-specific manner, increasing the reliability of the results. The use of a non-biotinylated oligonucleotide as a control has been successfully used previously 25 and it constitutes from our belief a good approach to discard unspecific targets.
As would be expected for an RNA-protein pull-down method, most of the proteins identified in this work were classified as nucleic acid binding proteins. This suggests direct association between them and the CRE. However, protein-mediated interactions cannot be ruled out since 6% of the factors were initially termed as protein-binding proteins. Using bioinformatics resources such as DAVID searches, which do not rely exclusively on GO terms, protein pool was clustered into eight different groups; that with highest score corresponded to RNA-binding proteins (Table 2). Interestingly, all this cluster′ s proteins shared the presence of one or more RRM, also termed RNA binding domains, within their sequences. The comparison of all RRMs showed two clear, highly conversed regions that matched consensus sequences previously defined as ribonucleoprotein domain RNP1 and RNP2 (see consensus sequence in Fig. 3) 34 . The high degree of conservation in these sequences suggests they are involved in the association with the CRE. One example of this is the La protein, that interact with HCV IRES through its RRM and promote linkage between the 5′ and 3′ ends of the viral genome 35 .
The catRAPID algorithm, which studies RNA-protein interactions based on their sequences and secondary structure, strengthened the reliability of identification of the pulled-down proteins, even though some are termed DNA-binding proteins. HMGB1 provides an example of these proteins reported to bind with high affinity to branched RNA structures 36 . Another interesting feature of a group of proteins is their ability to bind AU-rich elements (AREs), The present work identified several proteins reported by other authors to act as ARE binding proteins (ARE-BPs): hnRNPA1 37 DAZAP1 38 , hnRNPA0 39 , hnRNPA3 39,40 and RBM3 40 . It is noteworthy that CRE has exposed AU-rich elements, i.e., the ARE consensus sequence (-AUUUA-) located in the apical loop of 5BSL3.1, and a non-canonical sequence in the apical loop of 5BSL3.2 (5′ -AUAUAU-3′ ) (Fig. 1A). This would explain the abundance of recovered peptides corresponding to these proteins after LC-MS/MS.
Considering the involvement of CRE in HCV translation is interesting to remark that two proteins related to the regulation of translation were identified: RNA-helicase DDX5 and eukaryotic translation initiation factor I subunit I (eIF3I). In fact, other subunits such as eIF3E, eIF3F and eIF3H were also detected in one or two experimental replicates, and other proteins related to translation (such as eIF1A, eIF2A and the 40S/60S ribosomal  proteins) were detected in previous pull-down experiments performed under less stringent conditions (data not shown). This is in good agreement with the HCV IRES translation regulation exerted by CRE 22 . The translation efficiency assays performed in this work show a clear involvement of DDX5 in HCV translation. DDX5 knockdown up to a 40% was able to cause a specific decrease of IRES activity of 40% (Fig. 6B). However, this effect was not caused by other siRNAs assays that displayed even higher knockdown effects (e.g. sihnRNPA1 or siG3BP1). Remarkably, other proteins from the DDX5 family (DEAD box) have been shown to promote translation through their helicase activity 41,42 . In addition, siPARP1 decreases translation efficiency; however an indirect effect of PARylation inhibition cannot be ruled out because of the high amount of PARP1 substrates that take part in a variety of cell processes 43 . The detection of translation-related proteins in the pulled-down assays plus the described modulation of translation activity of HCV CRE-containing RNA constructs, suggests an association between the translational machinery and the CRE when the cellular scenario is suitable. Further, a group of nine proteins (with hnRNPs, RNA-binding Musashi homologues, RNA-binding proteins and DAZ-associated protein among them) have been classified as structural constituents of ribosomes (GO:0003735), and to belong to the structural molecule activity group (GO:0005198). This reinforces that CRE is able to interact with ribosome components and modulate viral translation through association with host translation factors. Experiments designed to see whether this is the case are currently underway.
Finally, some proteins were selected to check whether they had any important involvement in HCV replication. In this work, replication experiments were performed on a hepatoma cell line containing a subgenomic HCV replicon defective for the viral protease (NS2/NS3). As an advantage, these assays reproduce quite accurately the HCV genome replication, being a good biological model. On the other hand one cannot exclude additional interactions of the assayed proteins with other regions of the HCV genome contained in the replicon, more than the CRE element. In fact some of the proteins tested for their implication in replication have been suggested to interact with other regions of the genome of the HCV (see references in table 1). Regarding the studied candidates, HMGB1 is an antiviral factor that changes its cell location and is secreted to the extracellular medium after HCV infection 44 . In agreement with a previous report showing that blocking of HMGB1 with specific antibodies increases HCV infectivity 44 , the results derived from the knockdown and overexpression experiments suggest that HMGB1 negatively modulates HCV replication in the replicon system. On the other hand, in the present work no significant effect of G3BP1 on HCV replication was detected, although other authors report enhancing effects of G3BP1 on this process 12,45,46 . These discrepancies, however, could be due to differences in the replicon model used or the amount Standard deviation is displayed for each siRNA. Significant differences between siRNAs and non-targeting siRNA are indicated above the corresponding bar (p < 0.05 in a two-tailed Student's t-test).
Scientific RepoRts | 6:25729 | DOI: 10.1038/srep25729 or strategy of interference RNA employed. G3BP1 has been described to bind HCV RNA polymerase NS5B 47 suggesting that it possesses both RNA-and protein-interacting sites, and that it may be able to regulate replication. It should be noted that G3BP1 has been reported to associate with viral genome 3′ elements that regulate replication in Dengue virus, a member of the family Flaviviridae 45 . All this together suggests G3BP1 to be a cellular host factor that can be hijacked by the genome of RNA viruses, with a consequent effect on viral replication. Finally, a protein with high antireplicative activity was found: hnRNPA1. Here we report the direct association of hnRNPA1 to the HCV CRE by the pulldown and the immunoprecipitation experiments. This ribonucleoprotein has previously been reported to bind as well to the viral polymerase NS5B 48 . In the latter work, the binding of hnRNPA1 to the 5′ and 3′ UTR (downstream of CRE) was also shown. Since CRE plays a role in both NS5B binding 49 and in viral 5′ -3′ contacts 21,22 , one might hypothesize that the binding of hnRNPA1 to CRE would have an effect on HCV replication. In the present work, hnRNPA1 clearly downregulates HCV replication, one explanation could be that this protein competes with NS5B for CRE binding, thus reducing the efficiency of the viral polymerase. However, due to the limitation of the replicon system, we cannot exclude the possibility of a partial effect of hnRNPA1 knockdown on HCV replication by direct interaction with NS5B or even synergic effects of both actions. Additionally, hnRNPA1 has been reported to act as a splicing factor in the maturation of IFR3, which is involved in the interferon response 50 . The role of hnRNPA1 in HCV resistance to IFN response would certainly be interesting to explore.
In conclusion, the current work reports a number of new HCV RNA-host protein associations. Further work on these factors will be needed to better comprehend their function in the viral cycle.

Methods
DNA templates and RNA synthesis. DNA coding for the CRE transcript was obtained by PCR amplification, ensuring the presence of the precise 3′ end. Briefly, T7pC + HV was amplified from plasmid pU3′ HCV9181 using the primers T7pHCV-9181 33 and asHCV9414 (5′ -AACAGGATGGCCTATTGGCCTG-3′ ) to obtain the template for CRE plus the downstream hypervariable region (9181-9414). A shorter amplicon, CRE (9181-9384), was amplified using T7pHCV-9181 and as HCV9384 (AGCTCCCCGTTCATCGGTT) primers. The template for RNA 667 was derived from the pcDNA3 vector (Invitrogen) linearized with DraIII 51 . RNA synthesis was performed using the TranscriptAid T7 High Yield Kit (Thermo Fisher Scientific), following the manufacturer's instructions. The resulting transcript was purified as previously described 52 . The RNA concentration was determined by UV spectrophotometry (A 260 ) and the degree of protein and carbohydrate/phenolic contamination assessed from the A 260 /A 280 and A 260 /A 230 ratios respectively. The integrity of the RNA was confirmed by denaturing agarose-formaldehyde gel electrophoresis.
RNA construct (ICU) containing a luciferase reporter gene flanked by the HCV genomic ends was obtained by amplification of pGLICU plasmid with 5′ pT7HCV and 3′ HCV primers 22 , in vitro transcription and purification as explained above. The generation of a capped RLuc RNA was generated as previously described 51 . Cell culture. The cell lines used in this study were a gift from Dr. R. Aldabe (University of Navarra, Spain).
The human hepatoma Huh-7 cells were maintained in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% heat-inactivated fetal bovine serum and 1 mM sodium pyruvate, at 37 °C in a 5% CO 2 atmosphere. The knockdown assays were performed in the named cells and in a human hepatocarcinoma cell line harboring an HCV subgenomic replicon system (Huh-7 NS3-3′ ) 53,54 . Replicon cells were maintained in DMEM supplemented with 20% heat-inactivated fetal bovine serum and 0.5 mg/ml G-418 under the same conditions as described for the Huh-7 cells.
Cell lysate preparation. S10 fractions of Huh-7 cell lysates for use in pull-down assays were essentially prepared as previously described 55 , with some modifications. Briefly, around 5 × 10 7 Huh-7 cells were grown as described above to reach 100% confluent monolayers. They were then treated with trypsin, washed with 10 volumes of isotonic buffer (35 mM HEPES/KOH, pH 7.6, 150 mM NaCl and 11 mM glucose) and pelleted by centrifugation at 1,000 × g for 5 min. Cellular lysis was performed by adding 1.5 volumes of cold hypotonic buffer (20 mM HEPES/KOH, pH 7.6, 10 mM KCl, 1.5 mM MgAc, 1 mM DTT and protease inhibitors) and incubation at 4 °C for 20 min. Cell lysate was homogenized in a glass Dounce homogenizer (25 strokes) and equilibrated with 0.2 volumes of S10 buffer (100 mM HEPES/KOH pH 7.6, 0.6 M KAc, 20 mM MgAc, 25 mM DTT and protease inhibitors). Cellular debris was removed by centrifugation at 10,000 × g for 10 min. The resulting cleared lysates contained around 30 A 280 units ml −1 .
Isolation and identification of CRE-interacting proteins. Transcript C + HV (500 pmol) was annealed to 1 nmol of the 5′ biotinylated oligonucleotide (b-asHCV-9414) (Eurofins MWG Operon) or to the respective unlabeled asHCV-9414 for control reactions (sequence detailed in DNA template and RNA synthesis). It was then incubated with Huh-7 S10 cell extracts (~10 mg of proteins) pretreated with 50 μ g of tRNA and CL buffer (100 mM HEPES pH 7.6, 50 mM KAc, 1 mM MgAc and 1 mM DTT). Reactions were performed at 25 °C for 20 min to allow for CRE-protein interactions to take place. For the affinity chromatography step, 5 mg of streptavidin-coupled Dynabeads ® (Invitrogen) were washed and equilibrated following the manufacturer's instructions. Ribonucleoprotein complexes were captured by incubation of the Dynabeads in the binding reaction in a rotation wheel for 30 min at room temperature. Unbound factors were removed after washing three times with buffer B (0.1 M NaCl and 0.1% Tween ® 20) pre-warmed at 37 °C. CRE-protein complexes were eluted from the beads by heating for 2 min at 95 °C in the presence of 95% formamide and 10 mM EDTA. The recovered polypeptides were precipitated in methanol/chloroform. Some 2/3 of the sample volume was aliquoted for LC-MS/MS analysis and the rest used in SDS-PAGE. For the latter, samples were resuspended in Laemmli buffer 1X (10 mM Tris-HCl pH 6.8, 5% SDS, 2 mM EDTA, 5% Glycerol, 1% β -mercaptoethanol, 0.1 mg/ml Bromophenol blue) and loaded onto a polyacrylamide gel (5% stacking, 12% resolving). Proteins were stained by Sypro-Ruby (BioRad) Scientific RepoRts | 6:25729 | DOI: 10.1038/srep25729 and scanned in a Typhoon 9400 scanner (GE Healthcare). Both biotinylated and non-biotinylated reactions were conducted in duplicate (biological samples = 4).
Mass spectrometry analysis. The rest of the samples were resuspended, digested and run in LC-MS/MS following a protocol similar to that used by others 56,57 with minor modifications 58 . The specifics of this protocol are depicted in Supplementary Note S1. Each replicate was run twice, independently (two technical replicates per sample). Protein enrichment was established by PEAKS 6 software as protein ratio ≥1.00:0.50 in biotinylated samples versus the controls. Only proteins with > 15% coverage in positive samples (biotinylated) were finally selected. The number of recovered peptides, coverage and ratios of positive samples were calculated as average of the four corresponding technical replicates. The list of the proteins enriched for each comparison is shown in Supplementary Table 1.
Bioinformatics resources. The Panther gene list analysis tool was used to classify the pulled-down proteins 59 . The list for analysis included the Uniprot IDs shown in Table 1 (or analogue IDs in the very few necessary cases). Proteins were classified in gene ontology (GO) terms related to "molecular function", and their proportions represented in a pie chart. Successive classification was performed for the two most represented groups.
The same protein list was examined using the Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.7 bioinformatics resources (http://david.abcc.ncifcrf.gov/home.jsp). Functional annotation clustering of the protein list was performed under the highest classification stringency. A total of 13 annotation categories were selected (by default) for the search. As indicated by Huang and co-workers 27 , only clusters with an enrichment score of ≥ 1.3 and a p value of < 0.05 were taken into account for further analysis.
The catRAPID webserver was used for predicting RNA-protein interaction (http://service.tartaglialab.com/ page/catrapid_group). The CRE sequence was run against the Homo sapiens nucleotide-binding proteome using the catRAPID omics tool. A search was performed for RNA-or DNA-binding full-length proteins. Proteins were ordered based on their Rating Star Score, which considers the protein-RNA Interaction Propensity score (i.e. the interaction probability) and the presence of RNA/DNA binding domains and motifs. Additionally, nine different FASTA protein sequences, taken from the Uniprot database, were run (using the catRAPID graphic tool) against the complete CRE sequence, and nucleotide-aminoacid interaction heat-maps obtained.

Protein knockdown mediated by RNA interference and subgenomic HCV RNA quantification.
Huh-7 NS3-3′ cells were cultured (300,000 cells/well) in 6 well-plates to obtain a confluence of at least 80% after 24 h. Cells harbouring a subgenomic HCV replicon were transfected with small interference RNAs (siRNAs) designed against the mRNAs of the proteins of interest. For hnRNPA1, DDX5, G3BP1 and HMGB1, a pool of four targeting siRNAs for each protein was designed (Dharmacon). For PARP1 inhibition, a specific siRNA (Sigma) was used. As a negative control, a single non-targeting (NT) siRNA (Dharmacon) was selected. siRNAs were mixed with Transfectin TM solution in Opti-MEM ® medium. The transfection mixture was added to the Huh-7 NS3-3′ cells in an antibiotic-deprived medium to achieve a final siRNA concentration of 50 nM. After 48 h, the cells were collected for RNA or protein extraction.
RNA extraction was performed using Trizol reagent 54 . Primer extension was achieved using the High Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific). Briefly, RNA extracted (40 ng) was annealed to random primers in a 10 μ l volume reaction, at 95 °C for 2 min, and at 16 °C for 15 min. cDNA was retrotranscribed in a total volumen of 20 μ l at 37 °C for 30 min following manufacturer instructions. Enzyme inactivation was performed at 85 °C for 5 min. Intracellular HCV replicon RNA levels were measured as previously described and normalized with those obtained for the internal mRNA encoding GAPDH 54 . The HCV fragment was amplified by PCR over 40 cycles (5 s at 95 °C and 5 s at 58 °C) in a CFX96 Thermal Cycler (BioRad).
For protein extraction, cells were scraped after adding 500 μ l of cold PBS, and centrifuged at 2500 rpm. The resulting pellet was resuspended in 200 μ l of protein Lysis Buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1% Triton X-100, supplemented with protease inhibitor cocktail tablets) for 30 min at 4 °C. Lysates were centrifuged at 12,500 rpm and the supernatant stored at −20 °C. The protein concentration was measured using BioRad Protein Assay reagent. Proteins were separated by SDS-PAGE and transferred to nitrocellulose membranes in a SemiDry Trans Blot apparatus (BioRad). Membranes were blocked in 5% dry milk in TBS-Tween and incubated with specific primary antibodies against hnRNPA1, DDX5, HMGB1 (Cell Signaling Technology) G3BP1 (Santa Cruz Technology) or PARP1 (a gift from Dr. Oliver, Institute of Parasitology and Biomedicine López-Neyra, Granada, Spain) in 1% BSA-TBS/Tween solution at 4 °C overnight. Secondary HRP-linked antibodies were added for 1 h. Chemiluminescence was then detected using the Clarity TM Western ECL Substrate Kit (BioRad).
Overexpression of hnRNPA1 and HMGB1 coding genes. Total RNA from Huh-7 cells was extracted using Trizol reagent. Primer extension was performed as described above using instead an oligodT and annealing at 25 °C. cDNA was amplified by PCR using 5′ HMGB1-EcoRI (GCGTATGAATTCATGGGCAAAGGAGATCCTAAGAA) and asHMGB1 (GCGTATCTCGAGTTCA TCATCATCATCTTCTTCT) primers for HMGB1 and 5′ A1 (GCGTATGGATCCATGTCTAA GTCAGAGTCTCCTAAA) and asA1 (GCGTATCTCGAGTTAAAATCTTCTGCCACTGCCATA) for hnRNPA1. PCR amplicons were digested with EcoRI and XhoI for HMGB1 sequence and BamHI and XhoI for hnRNPA1 and inserted in pcDNA3 vector digested with the corresponding enzymes. Plasmid isolation was perfomed using Qiagen ® Plasmid Maxi Kit (Qiagen). Huh-7 NS3-3′ cells were cultured in 6 well plates and allowed to grow up to 80% confluence, then cells were transfected with 2 μ g/well of pcDNA3 (as a control) or pcDNA3-hnRNPA1 or pcDNA3-HMGB1, and total RNA was extracted 24 h or 48 h after transfection using Trizol reagent. Quantification of HCV RNA levels was performed by RT-qPCR as described above. Relative amounts of hnRNPA1 and HMGB1 mRNAs to GAPDH mRNA in the total RNA mix was analyzed by RT-PCR Luciferase measurement in HCV CRE-containing RNA constructs. Huh-7 cells were seeded in 24-well plates and allowed to grow until reach 80% confluence. Cells were transfected with siRNAs against the mRNAs of selected proteins as indicated above. After 48 h cells were re-transfected with RNA constructs containing Firefly or Renilla luciferase reporter genes and translation efficiency measured after 4 h using Dual Luciferase Reported Assay System (Promega). Translation efficiency was determined by the ratio FLuc/RLuc (Firefly luciferase /Renilla luciferase.