Rab1b and ARF5 are novel RNA-binding proteins involved in IRES-driven RNA localization

Internal ribosome entry site (IRES) elements are organized in domains that guide internal initiation of translation. Here we have combined proteomic and imaging analysis to study novel IRES interactors recognizing specific RNA structural subdomains. Besides known IRES-binding proteins, we identified novel factors belonging to networks involved in RNA and protein transport. Among those, Rab1b and ARF5, two components of the ER-Golgi, revealed direct binding to IRES transcripts. However, these proteins exert different effects on translation. While a dominant-negative mutant of Rab1b decreased IRES function, ARF5 silencing stimulated IRES activity. RNA FISH studies revealed novel features of the IRES element. First, IRES-RNA formed clusters within the cell cytoplasm, whereas cap-RNA displayed disperse punctuated distribution. Second, the IRES-driven RNA colocalized with ARF5 and Rab1b, but not with the dominant-negative of Rab1b. Thus, our data suggest a role for domain 3 of the IRES in RNA localization around ER-Golgi, a ribosome-rich cellular compartment.


INTRODUCTION
Internal ribosome entry site (IRES) elements promote internal initiation of translation using capindependent mechanisms (Yamamoto et al, 2017). Despite performing the same function, IRES elements, which were first identified in the RNA genome of picornavirus, are characterized by a high diversity of sequences, secondary structures, and requirement of factors to assemble translation competent complexes, which led to their classification into different types. RNA structure organization of IRES elements plays a critical role for IRES function. For instance, type II IRES elements such as the encephalomyocarditis (EMCV) and foot-and-mouth disease virus (FMDV) differ in 50% of their primary sequence, yet they fold into similar secondary structures 3 . Domain 3 is a self-folding cruciform structure (Fernandez et al, 2011). The basal region of this domain consists of a long stem interrupted with bulges that include several non-canonical base pairs and a helical structure essential for IRES activity. The apical region harbors conserved motifs essential for IRES activity, which mediate tertiary interactions (Fernandez-Miragall & Martinez-Salas, 2003;Jung & Schlick, 2013;Lozano et al, 2016).
However, the transacting factors interacting with this domain and their potential functions remain poorly studied and need to be investigated.
Beyond internal initiation of translation, a few evidences for the involvement of the IRES in other steps of the viral cycle have been reported. A role for the poliovirus IRES in RNA encapsidation was reported based on the different genome stability and encapsidation efficiency of RNA replicons carrying chimeric IRES elements (Johansen & Morrow, 2000). Similarly, interaction of the core protein of hepatitis C virus (HCV) with the IRES region was involved in nucleocapsid assembly (Shimoike et al, 1999). RNAs harboring IRES elements have been reported to locate around the endoplasmic reticulum (ER) in picornavirus infected cells (Lerner & Nicchitta, 2006). This is consistent with the view that translation-active ribosomes show different subcellular distributions, with enriched ER-localization under cell stress (Reid & Nicchitta, 2015). However, the specific domains of the IRES controlling RNA localization on the ER remain elusive.
To gain a better understanding of the role of the FMDV IRES subdomains in cellular events linked to cap-independent translation, we conducted a systematic proteomic approach using streptavidinaptamer tagged transcripts encompassing stable domain 3 subdomains. Besides proteins previously reported to interact with this IRES region, we identified factors belonging to functional networks involved in transport. In particular, we focused on two small GTPases, the Ras-related protein Rab1b and the class II ADP-ribosylation factor 5 (ARF5). While Rab1b is a regulator of coat complex protein I (COPI) and COPII ER-Golgi transport pathway depending upon its GTP-binding state (Monetta et al, 2007;Segev, 2011;Slavin et al, 2011), ARF5 is located on the trans-Golgi independently of its GTP-binding state (Jackson & Bouvet, 2014). It is well established that the anterograde transport pathway participates in the life cycle of various RNA viruses (Belov et al, 4 2008;Gazina et al, 2002;Martin-Acebes et al, 2008;Midgley et al, 2013). Yet, the pathways affecting distinct RNA viruses are currently under intense investigation (Reid et al, 2018;van der Schaar et al, 2016).
Beyond the identification of RNA-binding proteins by proteomic approaches, we have found that the IRES transcripts bind directly with purified Rab1b and ARF5, revealing a previously unknown RNA-binding capacity of these proteins. RNA FISH studies showed that mRNA carrying the IRES element displayed a cluster arrangement relative to mRNA lacking the IRES. Remarkably, IREScontaining RNAs colocalized with Rab1b and ARF5 to a higher stent than cap-RNA. However, in support of the different role in IRES-dependent translation, a dominant negative form of Rab1b decreased IRES function, while ARF5 silencing stimulated IRES activity. In sum, our data show that both ARF5 and Rab1b exhibit RNA-binding capacity, and suggest a role for domain 3 of the IRES in RNA localization into specific cellular compartments.

T he protein interactome of I R ES subdomains reveals distinct recruitment of cellular factors
The large size of the picornavirus IRES region (450 nt) compared to other IRES elements prompted us to investigate whether this RNA region harbors motifs involved in additional RNA life steps, overlapping with internal initiation of translation. The FMDV IRES element is organized in domains, designated 1-2, 3, 4, and 5 . The central domain (designated D3 therein) is organized in a long basal stem interrupted by several bulges, and the apical region encompassing stem-loops SL1, SL2, and SL3abc ( Fig 1A). To understand potential implications of D3 on the RNA life spam, we have undertaken a systematic study of host factors interacting with structural motifs present in D3. To this end, we prepared four transcripts SL3a, SL3abc, SL123, and D3 (Fig 1A), encompassing stem-loops previously defined by mutational studies and RNA probing (Fernandez-Miragall et al, 2006;Lozano et al, 2014). In principle, this strategy could allow us to identify specific factors recognizing individual IRES subdomains.

5
To obtain transcripts with stabilized secondary structure, cDNAs were inserted into pBSMrnaStrep vector (Ponchon et al, 2009), which allows streptavidin-aptamer tagged RNA purification. Purified D3 RNAs and a control RNA ( Fig EV1) were used in RNA-pull down assays using HeLa cells soluble cytoplasmic extract as the source of proteins. Following streptavidin-affinity purification, proteins co-purifying with the individual RNA subdomains were visualized on silver stained SDS-PAGE ( Fig 1B). A distinctive pattern of bands was readily detected relative to the control RNA, as shown for SL3abc and SL3a RNAs, suggesting specific binding of factors to each D3 subdomain.
Next, the factors associated with each transcript were identified by LC/MS-MS in two independent biological replicates (Dataset EV1). Only factors identified in both replicates with more than 2 peptides (FDR <0.01) were considered for computational studies (R 2 Fig EV2A). revealed that the number of proteins specific for each subdomain was higher than those shared among transcripts ( Fig 1C). Furthermore, the apical subdomains SL3a and SL3abc shared similar factors, while those copurifying with SL123 were similar to D3. These results suggest that RNAprotein interaction is, at least in part, consistent with the structural organization of each subdomain.
Functional group analysis of the filtered proteins remaining on these transcripts indicated that > 30% belong to nucleic acids binding , irrespectively of the subdomain used to capture them ( Fig EV2B). Moreover, the best represented were annotated RBPs (including RNA processing, hnRNPs and RNA helicases), ribosomal proteins, followed by organelle and transport, signaling, translation factors, and metabolism ( Fig 1D). Of note, the ribosomal proteins were more abundant within SL123 and D3 RNAs. Then, representation of the log 10 score of proteins bound to the distinct subdomains showed that the correlation for factors interacting with the apical domains 6 ( Fig EV3A), or the basal domains ( Fig EV3B) was better than comparing apical domains with D3 ( Fig EV3C,D). Of interest, the apical subdomains SL3a and SL3abc shared annotated RBPs and organelle members. In summary, these data revealed a preferential association of factors belonging to different functional groups to the distinct subdomains.
O ver represented networ ks associated with domain 3 unveil the E R-Golgi transport, besides

R N A-related processes
Gene ontology classification of the filtered proteins captured with each subdomain in functional categories using BiNGO showed a distribution in statistically significant nodes (Maere et al, 2005).
As shown in Fig 2, nodes overrepresented on these transcripts relative to a whole human proteome belong to functional networks. The networks biosynthetic processes, translation factors and RNA processing were identified in all transcripts. In particular, translation factors and biosynthetic processes have high statistical significance in SL123 and D3 (ranging from P = 7x10 -30 to 1x10 -12 ) ( Fig 2). Conversely, networks differentially associated to distinct subdomains were RNA transport network with SL3a, immunity with SL3abc, and ribosomal proteins with D3, while cell cycle and proteolysis were exclusive of SL123 and D3. We noticed an increase in the number and the significance level of nodes, and also in the number of functional networks, in correlation with the number of subdomains present in the transcript used to capture proteins. Of interest, overrepresentation of the ER-Golgi transport network was statistically significant in all transcripts, ranging from P = 3x10 -4 to 5x10 -5 (Fig 2).
Collectively, these results reinforce the hypothesis that specific IRES subdomains could be involved in the assembly of ribonucleoprotein complexes participating in distinct biological processes, such us ER-Golgi trafficking.
M embers of the E R-Golgi transport networ k display R N A-binding capacity As expected from the established function of the IRES element, our study identified a high number of annotated RBPs (Dataset EV1). Beyond known IRES-binding factors, ribosomal proteins and translation factors, we also identified ER-Golgi transport factors (Table 1). Among the identified 7 members of the ER-Golgi network we focused on two factors, which were not previously reported as RNA-binding proteins, Rab1b and ARF5 ( Fig 3A). While Rab1b is a regulatory protein involved in both COPI and COPII transport (Monetta et al, 2007;Slavin et al, 2011), ARF5 is an integral member of Golgi (Jackson & Bouvet, 2014).
To rule out that the factors identified in the proteomic analysis were derived from secondary interactions, we set up to determine whether individual RNA subdomains were involved in the recognition of factors trafficking between organelles. Thus, to assess their direct RNA-binding capacity, we performed gel-shift assays with purified proteins. Increasing amounts of His-Rab1b yielded positive interactions with transcripts D3, SL123, and SL3abc, but not with SL3a ( Fig 3B).
In contrast, His-ARF5 showed interaction with all transcripts encompassing the apical region (SL3a, SL3abc, and SL123) ( Fig 3C), although its RNA-binding affinity was lower than that of His-Rab1b (compare Fig 3C to Fig 3B). The interaction of ARF5 with D3 was weaker, requiring high protein concentration. To further analyze the RNA-binding specificity, we used a probe differing in sequence and predicted secondary structure as a control. None of these proteins yielded a band-shift at the same protein concentration ( Fig 3D). Collectively, we conclude that both Rab1b and ARF5 are bona-fide IRES-binding proteins, although the later show lower RNA-binding affinity.
Next, we wished to compare the interactions of these factors to PCBP2 and Ebp1, two proteins known to interact with domain 3 (Monie et al, 2007;Pacheco et al, 2008;Walter et al, 1999;Yu et al, 2011), which were also identified in our proteomic approach ( Fig 3A). As shown in Fig 3E, PCBP2 induced the formation of a complex with transcripts D3 and SL123 in a dose-dependent manner, fully compatible with the presence of the C-rich motif on these RNAs ( Fig 1A). Gel-shift assays performed in parallel with labeled SL3a or SL3abc probes, lacking the C-rich motif, failed to form complexes (Fig 3E), confirming the recognition of specific motif by PCBP2 under our conditions. Similar assays conducted with Ebp1 yielded a complex only with D3 ( Fig 3F), suggesting that the Ebp1-binding site is primarily located on the basal stem of this domain. Thus, 8 concerning RNA complex formation, Rab1b resembled PCBP2, while ARF5 was dissimilar from both PCBP2 and Ebp1.
Overall, the RNA-protein binding results match with the proteomic identification in about 75% of the four proteins analyzed (see Fig 3). According to the results derived from these independent approaches, it is tempting to suggest that Rab1b interacts directly with the apical region of domain 3 (SL3abc subdomain), while ARF5 recognizes the SL3a subdomain.

Both, Rab1b and A R F5 are involved in I R ES-dependent translation
To analyze the functional implications of these factors on IRES activity, we relied on siRNAmediated approaches to reduce the level of these proteins. Silencing of Rab1b did not alter IRES activity relative to a control siRNA ( Fig 4A). However, since the siRNA targeting Rab1b does not deplete Rab1a (Tisdale et al., 1992), which was also identified in the proteomic approach (Table 1), it may occur that Rab1a functionally substitutes for Rab1b. In contrast, ARF5 silencing stimulated IRES activity (Fig 4A), suggesting that an ARF5 related pathway could affect internal initiation of translation.
Given that the result of Rab1b silencing could be explained by functional redundancy with Rab1a, we generated a dominant negative mutant of Rab1b replacing Serine 22 by Asparagine, which inactivates both Rab1b and Rab1a and disrupts the Golgi (Alvarez et al, 2003). Expression of the dominant negative GFP-Rab1b-DN protein disrupted the Golgi (Fig 4B), and decreased IRESdependent translation of luciferase, whereas cap-dependent mRNA translation was not significantly affected ( Fig 4C). Thus, we conclude that altering the GTP-binding affinity of Rab1b (Alvarez et al, 2003), hence destabilizing the ER-Golgi, decreases IRES-dependent translation.

T he I R ES element mediates R N A ar rangement in clusters within the cell cytoplasm
Taken into consideration the factors related to ER-Golgi transport associated with domain 3 we sought to investigate the involvement of this region on mRNA localization. To this end, we compared two mRNAs, designated cap-luc and IRES-luc, which only differ in the presence of the IRES element on the 5´UTR (Fig 5A). Cells transfected with constructs expressing cap-luc or IRES-luc mRNA were first used to determine the expression of the reporter protein. As expected, both RNAs produced luciferase activity although to different extent ( Fig 5A) (Lozano et al, 2018).
Then, we conducted RNA-FISH experiments using probes targeting the luciferase-coding region.
As shown in Fig 5B, we observed bright spots corresponding to IRES-luc and cap-luc RNAs in each case. No signals were observed in cells transfected with a control plasmid lacking the CMV promoter (pluc), demonstrating lack of DNA detection with these probes. Importantly, quantitative analysis of RNA spots in cells expressing the IRES-luc RNA showed an enhanced cluster arrangement spots/cluster), while spots observed in cells expressing cap-luc RNA were dispersed along the cell cytoplasm (P = 3.7x10 -18 ) ( Fig 5C). This result showed a different distribution of RNA signals within the cellular cytoplasm depending upon the presence of an IRES element in the mRNA.

Proteins Rab1b and A R F5 enable I R ES-driven R N A localization
Next, considering the role of Rab1b in ER-Golgi transport (Monetta et al, 2007), cells transfected with GFP-Rab1b and either pCAP-luc or pIRES-luc constructs were processed for RNA-FISH ( Fig   6A). The GFP-Rab1b protein showed ER-Golgi localization, as shown by its colocalization with the Golgi marker GM130 (Fig EV4). Interestingly, the IRES-luc mRNA exhibited a cellular colocalization with GFP-Rab1b in 52% of Rab1b-transfected cells (Fig 6B), while colocalization of cap-luc mRNA with GFP-Rab1b accounted for 19% of Rab1b-transfected cells. These results revealed 2.7-fold increase (P = 3.1x10 -39 ) in the percentage of colocalization of IRES-luc mRNA with Rab1b compared to cap-luc mRNA.
These data prompted us to analyze RNA colocalization with GFP-DN-Rab1b protein, which yielded a disrupted Golgi (Fig 4C). In contrast to the results observed with the wild type Rab1b, the IRES-luc RNA and the cap-luc RNA showed a lower, very similar colocalization with the GFP-DN-Rab1b protein (34 and 32%, respectively) ( Fig 6C). Hence, a significant decrease in GFP-DN-Rab1b colocalization with IRES-luc (34%) was noticed in comparison to the wild type GFP-Rab1b (52%) (Fig 6B and 6C). These data strongly suggests the biological relevance of Rab1b for IRESdriven RNA localization.
Then, since Rab1b is located at the ER and cis-Golgi, we analyzed the colocalization of IRES-luc RNA with ARF5, an integral member of trans-Golgi (Fig EV4). Cells expressing GFP-ARF5 showed a higher frequency of colocalization of the IRES-luc mRNA with ARF5 than cap-luc (44 and 18%, respectively) (Fig EV5), reinforcing the role of IRES-driven location of mRNA within the ER-Golgi. However, the mean values obtained for Rab1b and ARF5 were statistically significant different (P = 2.6x10 -5 ). Therefore, we suggest that the IRES-containing RNA is preferentially located on the ER-cisGolgi compartment.
Taken together, we conclude that both, Rab-1b and ARF5 are involved on the IRES-driven RNA localization on the ER-Golgi area of the cell cytoplasm, although they exert opposite effects. Rab1b stimulates translation, while ARF5 diminishes IRES-dependent translation. In both cases, this property resides on their capacity to interact with domain 3 of the IRES element.

DISCUSSION
The data presented herein represents the first instance of the characterization of IRES interactions with ER-Golgi factors, reinforcing the importance of exploring novel RNA-protein interactions to understand host-pathogen interface. Here we describe a robust RNA-protein interaction approach, which allows detecting ribonucleoprotein complexes associated with specific subdomains of the IRES element. In this manner, a number of RBPs were selected, including known IRES interacting factors (Lee et al, 2017;Martinez-Salas et al, 2015), validating the approach used in our study.
Notwithstanding, we noticed that the IRES element not only recruited translation factors and IREStransacting factors, but also proteins involved in ER-Golgi transport, as exemplified in Table 1.
Following uncoating, the first intracellular step of picornavirus life cycle requires translation of the viral genome, which is governed by the IRES element. We hypothesize that interplay between host factors and viral RNA motifs could be an integral part of the regulation of viral RNA function, and as such, can be studied in the absence of infection. In accordance with this hypothesis, our data show that the IRES-containing mRNA exhibited a cluster arrangement, while the cap-luc RNA 11 showed disperse punctuated cytoplasmic location (Fig 5), suggesting that the IRES element was specifically involved in mediating RNA localization. The IRES-driven RNA clustering is in agreement with long-range RNA-RNA interactions involving domain 3 (Diaz-Toledano et al, 2017; Ramos & Martinez-Salas, 1999), which could contribute to hold IRES-containing transcripts in specific subcellular location. Furthermore, our data are also in accordance with a recent report showing that IRES-containing mRNAs are enriched in ribosomal subunits purified from cell lysates relative to cap-mRNAs (Lozano et al, 2018). Interestingly, purified ribosomes induced SHAPE reactivity changes within domains 2 and 3 of the IRES, including the apical SL3abc subdomain. In support of the relevance of the IRES element for RNA localization, IRES-dependent translation is compartmentalized to the ER in picornavirus infected cells (Lerner & Nicchitta, 2006), consistent with visualization of poliovirus RNA complexes on the anterograde membrane pathway to the Golgi (Egger & Bienz, 2005).
Concerning the implication of IRES subdomains in directing RNA to specific subcellular locations, we selected two factors involved in ER-Golgi trafficking, Rab1b and ARF5, for further characterization. Rab1b is a key regulatory protein involved in COPI and COPII transport (Monetta et al, 2007), whereas ARF5 is an integral member of the Golgi (Jackson & Bouvet, 2014). We show here that both ARF5 and Rab1b interact with domain 3 in vitro in the absence of other factors. However, they exhibit distinct features. While Rab1b preferentially interacts with all transcripts with the exception of SL3a, ARF5 shows a preferential binding to the apical subdomains ( Fig 3B,C). These features are compatible with the recognition of distinct IRES stemloops. While ARF5-RNA interaction from proteomic data suggests recognition of SL3a, the recruitment of Rab1b was observed with SL3abc ( Fig 3A), which is also present within SL123 and D3. In the context of the total cell extract, the lack of detection in MS/MS approaches can be due to interference with other RBPs interacting with this IRES region.
The finding that Rab1b and ARF5 proteins interact directly with the IRES was not anticipated, since no reports of their RNA-binding capacity were available. We hypothesize that, beyond governing internal initiation of translation, the interaction of the IRES element with proteins such 12 as Rab1b and ARF5 mediate the localization of IRES-driven RNA at the ER-Golgi, in a rich ribosome environment (Fig 7). This pathway may occur concomitantly to eIFs-and IREStransacting factors mediated translation (Lee et al, 2017;Martinez-Salas et al, 2015). Several experimental evidences support this model. First, we have found direct interactions of purified Rab1b, and also ARF5, proteins with the IRES transcripts in the absence of other factors (Fig 3B   and 3C). Second, in comparison to mRNA lacking the IRES element, we observed colocalization of the IRES-luc RNA and the protein Rab1b-GFP ( Fig 6B) and GFP-ARF5 (Fig EV5) in living cells. Third, the study of the GFP-Rab1b-DN revealed a significant decrease of IRES-dependent translation, concomitant to ER-Golgi disruption and RNA localization impairment (Fig 4C and Fig   6C). Given that the ER-Golgi is disorganized in cells expressing the negative dominant mutant of Rab1b (Fig 4B), we are tempted to speculate that disruption of the ER-Golgi compartment induced by GFP-Rab1b-DN, and hence the ER-associated ribosomes, could impair IRES activity but not global cap-dependent protein synthesis. In contrast, the results of silencing ARF5 in conjunction with the GFP-ARF5-IRES colocalization allow us to suggest that interaction of the IRES with ARF5 may sequester the mRNA on the trans-Golgi, hence interfering IRES-driven translation.
Further supporting the notion that specific members of the anterograde transport pathway mediate IRES recognition, as shown here by Rab1b, several members of the anterograde and retrograde transport were identified in the proteomic approach (Table 1), although their validation remains for future studies. We attempted to study IRES-driven RNA colocalization with other ER-Golgi components (GM130, ERGIC53, and calnexin-CT) using antibody-guided protein staining with little success, presumably due to the degradation of the probe and/or the RNA.
Here we also focused on ARF5 aiming to unveil its functional implication on IRES-dependent expression. Recent studies have shown that ARF4 and ARF5 are involved in distinct steps of the infection cycle of RNA viruses, demonstrating different functions for class II ARF proteins. While Dengue virus secretion was affected at an early pre-Golgi step (Kudelko et al, 2012), these factors were involved in HCV replication (Farhat et al, 2016). Our data show that purified ARF5 form RNA-protein complexes with the IRES subdomains in vitro, supporting the possibility that the 13 colocalization observed in transfected cells is biologically relevant. Moreover, since depletion of ARF5 stimulated IRES-dependent translation, we hypothesize that a fraction of the IRES-RNA can be sequestered by ARF5 on the trans-Golgi (Fig 7). As a result, Rab1b-mediated location of the IRES-RNA on the ER could be diminished, removing at least part of the IRES-containing RNA from the pool of actively translated mRNAs.
In summary, our data suggest a role for domain 3 of the IRES in RNA localization at the ER-Golgi, a ribosome-rich cellular compartment. We have identified two novel factors, Rab1b and ARF5, interacting with IRES transcripts, reflecting additional functions of this RNA regulatory region apart from its involvement in internal initiation of translation. Furthermore, we have found that both proteins, ARF5 and Rab1b exhibit RNA-binding capacity, although they promote different effects on IRES-dependent translation. We propose that the IRES element provides a link between RNA localization and selective translation. Whether this hypothesis could also apply to protein synthesis guided by different RNA regulatory elements awaits further investigations. pIRES-luc and pCAP-luc, tagged with MS2 hairpins, were described (Lozano et al, 2018). Plasmid peGFP-C1-Rab1b was generated by PCR using primers C1-GFPRabs, C1-GFPRabas, and template pPB-N-His-Rab1b. The PCR product was inserted into peGFP-C1 via XhoI-BamH1. peGFP-C1-Rab1bDN was obtained by QuikChange mutagenesis using primers Rab1bS22Ns and

Constructs and T ranscripts
Rab1bS22Nas. Oligonucleotides used for PCR and the restriction enzyme sites used for cloning are described in Table EV1. All plasmids were confirmed by DNA sequencing (Macrogen).
14 In vitro transcription was performed as described (Fernandez et al, 2011). When needed, IRES transcripts were uniformly labeled using 32 P-CTP (500 Ci/mmol). RNA integrity was examined in 6% acrylamide 7 M urea denaturing gel electrophoresis. RNAs SL3a, SL3abc, SL123, D3, and the control RNA were isolated from fresh bacterial cell lysates, as described (Ponchon et al, 2009). The integrity of purified RNA was analyzed in denaturing gels (Fig EV1).

R N A-protein pull-down
Streptavidin aptamer-tagged RNAs coupled to streptavidin-coated magnetic beads were used to purify proteins interacting with the IRES transcripts ( Fig 1A). Briefly, RNA-binding to beads (100 µl) was carried out in 500 µl binding buffer (

M ass Spectrometry identification
Mass spectrometry (LC/MS-MS) was performed as described (Francisco-Velilla et al, 2016). Two independent biological replicates were analyzed for all samples. Factors with score below 10% of the maximum within the functional group were discarded for further analysis, and only factors identified in both replicates with more than 2 peptides (FDR <0.01) were considered for computational studies. Finally, to eliminate false positives, the factors associated with the control RNA were subtracted from those identified with SL3a, SL3abc, SL123 and D3 transcripts. Proteins were classified by gene ontology using PANTHER (Mi et al, 2017).
The Biological Networks Gene Ontology application (BiNGO) was used to assess the overrepresentation of proteins associated with SL3a, SL3abc, SL123, of D3 transcripts, and to determine the statistical significance of overrepresented proteins relative to a complete human proteome (Maere et al, 2005). The results were visualized on the Cytoscape platform (Shannon et al, 2003). The biological processes nodes were classified according to a hipergeometric test in the default mode, FDR <0.01. P values for the overrepresented nodes were used to compute the average statistical significance of the network.

R N A gel-shift assays
RNA-protein binding reactions were carried out as described .
Electrophoresis was performed in native polyacrylamide gels. The intensity of the retarded complex was normalized to the free probe.
siR N A interference, immunodetection, and L uciferase activity Dye. Each Z-slice was exposed for 20-50 ms, except for Quasar 570, which required 2 s. After deconvolution from about 60 z-sections, 0.3 µm spacing, images were analyzed by local background subtraction and thresholding using Huygens Software (Scientific Volume Imaging).
Each Z-series was collapsed and rendered as a single max-intensity projected image using ImageJ v1.51u.
Cell borders were defined and spots associated with distinct cells were determined. RNA clusters ( spots) show unimodal distributions of RNA fluorescent signals. Three independent experiments were performed for each condition. For RNA-protein colocalization, double-transfected cells were analyzed from 3 independent experiments. In all cases, data represent mean ± SD.

Statistical analyses
We computed P values for different distribution between two samples with the unpaired two-tailed t-test. Differences were considered significant when P <0.05. The resulting P values were graphically illustrated in figures with asterisks.   The graphs represent the adjusted curves obtained from the quantifications of the retarded complex relative to the free probe (mean ± SD) from two independent assays for each probe.  F igure E V4. GFP-ARF5 and GFP-Rab1b colocalize with the Golgi marker GM130.
Representative images of Hela cells transfected with GFP-ARF5 or GFP-Rab1b, fixed 30 h posttransfection and permeabilized. Immunostaining of the Golgi was carried out using anti-GM130 antibody (Bar = 10 µm).