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

Integration of proteomic data with functional and imaging analysis revealed that Rab1b and ARF5, two ER-Golgi members, are RNA-binding proteins that colocalize with picornavirus IRES-RNA reporters in human cells.

The manuscript by Fernandez-Chamorro et al. describes the identification of novel proteins associated with domain 3 of the foot-and-mouth-disease virus IRES, and the initial functional characterization of two of these interactors Rab1b and ARF5. The proteomics data will be a useful resource for the community. It is very interesting that the authors find evidence of direct RNA binding activity of Rab1b and ARF5. This is a surprising result but the in vitro data are, overall, compelling. Experiments in cells support a functional role of Rab1b and ARF5 in regulating the distribution of IRES-containing RNAs. Although the paper has its merits, there are a number of substantive issues that need to be addressed before I could recommend publication.
Major points: 1. Although it appears that the distributions of Rab1b and IRES-luc RNA cells are related, the term co-localization is too strong. There is very little overlap, if any, of the two signals ( Figure 5B). A key part of the model in Figure 7 is the co-incidence of the RNA and Rab1b on the ER but there is no evidence presented to support this. Either more experiments need to be performed to determine the location of the IRES-luc clusters (e.g. with other GFP-tagged markers of the ER, Golgi or ER-Golgi intermediates) or the model needs to be revised substantially and conclusions toned down. I anticipate that additional experiments to refine/strengthen the model could be completed in the usual timeframe for revisions.
On a related point, it is not clear from the methods what criteria were used to accept or reject colocalization throughout the study. This information should be added including any measures taken to exclude subconscious bias in the analysis (e.g. blinding).
2. On two occasions the authors draw conclusions by comparing values that were seemingly obtained from different experiments.
"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 GFPDN-Rab1b protein (34 and 32%, respectively) ( Fig 6C). Hence, a significant decrease in GFP-DNRab1b 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 IRES driven RNA localization." "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".
If the data in question were derived from side-by-side experiments this should be made clear. If not, it is not appropriate to compare these values directly. This point can be illustrated by there (not surprisingly) being some differences in mean values between equivalent experiments performed separately (e.g. cap-luc data in Figure 6B and C).
3. I could not find any information on the nature of the control RNAs (sequence, identity, length, secondary structure) used for the aptamer-based pulldowns or EMSA so it is impossible to judge if they are suitable. This information is essential and should be included in the main text (with any additional details in the methods).
1. For non-specialists the authors should make it clear that FMDV is a picornavirus. 2. There are several grammatical, typographical and spelling errors that need correcting (e.g. autor (author), life spam (lifespan), punctuated (punctate), stent (extent), hypergeometric (hypergeometric); grammatical issues include but are not limited to: "Overrepresented networks associated with domain 3 unveil the ER-Golgi transport, besides RNA-related processes"; "Protein nodes functionally related"). 3. "Overall, the RNA-protein binding results match with the proteomic identification in about 75% of the four proteins analyzed". This sentence needs clarifying. What exactly do the authors mean? 4. The term GFP-ARF5-IRES colocalization is potentially misleading due to the use of a hyphen for GFP-ARF5 and to denote an interaction. The authors might want to use "colocalization of GFP-ARF5 with IRES". 5. For non-specialists, what SHAPE analysis reports on should be defined in the Discussion. 6. Do the authors know what the nucleotide status of the recombinant ARF5 and Rab1b is likely to be? It might be worth commenting that the influence of GTP or GDP on RNA binding activity should be investigated in a future study. 7. The methods section does not describe how RNA-protein complexes were eluted from the beads in the pulldown experiments. Was this through biotin or SDS buffer? This information should be provided in the methods. 8. The legends should indicate the number of repeats in all cases. For example, in Figure 3 the n number is only given for panel A. 9. Figure 7. Is the GTP status of Rab1b dependent on being membrane associated? Free Rab1b-GFP is shown in the model. Perhaps a comment needs to be added to the legend to clarify what the authors think is happening in this step. Might the RNA be associating with vesicles?
Reviewer #2 (Comments to the Authors (Required)): In this manuscript by Fernandez-Chamorro et al the authors use a streptavidin-based purification system to isolate proteins that bind to different region of the FMDV IRES. Rather surprisingly, they find enrichment of proteins involved in the Golgi-ER transport system and they show that the IRES is sufficient to cause colocalisation of reporter RNAs with one of these proteins, Rab1b. This is a very interesting finding and could provide new information about how the FMDV virus sustains infection in mammalian cells and in the longer term, provide new therapeutic avenues.
Major points. 1. Since the authors have only shown binding of ARF5 and Rab1b to the FMDV IRES they should name the virus in the title. In addition, there is no mention of the fact that it is the FMDV IRES that is being studied in the abstract and this needs to be amended.
Figures 1 and 2. These were both clear and the data are well described. It is stated in the discussion that there are no previous reports of ARF5 or rab1b binding to RNA. Have all the interactome capture data sets that are available been compared in this regard? In addition, there are several new manuscripts on bioRxiv which describe more extensive ways in which to identify RNA binding proteins using separation with trizaol based reagents (e.g. The Human RNA-Binding Proteome and Its Dynamics During Arsenite-Induced Translational Arrest Jakob Trendel, Thomas Schwarzl, Ananth Prakash, Alex Bateman, Matthias W Hentze, Jeroen Krijgsveld bioRxiv 329995; doi: https://doi.org/10.1101/329995).
It may be worth examining the data in these papers as well since they identify many more proteins that have RNA binding capacity. Figure 3. The authors show that Ra1b1b binds to RNA in vitro therefore it would be useful for the authors to carry out experiments using competitor RNAs to calculate dissociation constants. This would enable the relative strength of the interaction can be determined relative to other proteins which interact with this IRES, such as Gemin 5. It is suggested that Rab1b interacts directly with the apical region of domain 3. It would be of interest to prove this by transcribing this region of RNA in vitro and showing that it acts as a competitive inhibitor in the EMSAs. Figure 4. The effects of knockdown on IRES activity are very small and look only just significant. It would seem that these proteins individually do not really have a major effect the function of the IRES in terms of mRNA translation. What happens if the proteins are knocked down in combination? Is there a synergistic effect on IRES function? Figures 5 and 6. The data show that the IRES is sufficient to localise the reporter RNA within groups in the cytoplasm and that there is colocalisation with Rab1b, although for technical reasons it was not possible to identify other components of the ER. These experiments are technically very challenging. However, there is a very interesting recent publication (which shows that HCV secretion is regulated by Rab1b (Constantin N. et al Differential regulation of lipoprotein and hepatitis C virus secretion by Rab1b Cell Rep. 2017 21(2): 431-441. doi: 10.1016/j.celrep.2017.09.053). Would it not be possible to inhibit Rab1b using some of the vectors shown in the paper to determine whether this effects the localisation of the FMDV IRES? These experiments would have the advantaged of linking the RNA observations with the viral infection. In any case this paper should be discussed in light of the data obtained in the current manuscript.
Reviewer #3 (Comments to the Authors (Required)): In the manuscript entitled 'Rab1b and ARF5 are novel RNA-binding proteins involved in IRES-driven RNA localization', Fernandez-Chamorro et al. describe identification of two novel IRES-binding proteins, Rab1b and ARF5, and propose that they regulate IRES-dependent translation and localization of IRES-containing mRNA to Golgi. Overall the manuscript is well written and the rationale is well structured and easy to follow. However, the authors do not provide sufficient evidence to support their conclusions and the presentation of the results in the manuscript should be improved. 1. The authors perform protein pull-down followed by mass spectrometry using D3-domain fragments of IRES-element. These fragments are contained in one another, such that: D3>SL123>SL3abc>SL3abc. Therefore, it is expected that majority of the proteins bound to smaller fragments should also be detected in the pull-downs with the larger domain pieces. However, the authors observe only a modest overlap between different domains' interactomes ( Figure 1) and many proteins detected with the short fragments do not bind larger fragments. For example, more than 30% (70 out of 214) proteins detected with SL123 are unique for this fragment, and not detected in D3-interactome, despite the fact that entire SL123 fragment is contained in D3 domain. These observations are concerning, as it suggests there is a large non-specific binding in the shorter domains when they are expressed alone. The authors should address this issue. Moreover, there is insufficient validation. In the pull-down experiments, the authors must perform a validation using immunoprecipitation followed by a western blot for Rab1b and ARF5. This is especially important given that Rab1b was not detected in all of the mass-spec replicates. 2. Figure 3: Only some EMSA gels are shown. For example, in Figure 3B SL123 with Rab1b and in Figure 3C SL3abc with ARF5. As the Figure shows quantitation for all of the possible RNAfragment/protein pairs, the authors should provide gel images for all of these experiments in the supplement. The legend should explain which RNA was used for negative control and how table in Figure 3A was generated, i.e. what '+' and '-' stand for. If '+' means interaction detected in EMSA, the authors should explain why their negative control interacts with PCBP2.Moreover, ARF5 concentration required to observe binding to IRES subdomains is ~200nM, a 100-times more than for other proteins. Such high protein concentration could result in non-specific binding. The authors should include a control protein to exclude the possibility that IRES domains bind to non-specific proteins at higher concentrations. 3. In Figure 4 the authors examine the effect of Rab1b and ARF5 silencing on IRES-mediated translation using a luciferase assay. The observed effects are modest and it is unclear if they would be physiologically relevant. Silencing of known IRES-translation regulator or pharmacological inhibition should be included to estimate the background. The authors use normalized Luc activity/amount of protein which does not reflect the effects of transfection efficiency. Dual firefly/renilla luciferase assay is a better normalization approach. Moreover the data are normalized to the extent where it is hard to judge how the original data look like. For example, is luciferase activity of IRES-luc and cap-luc in the presence on Rab1b-wt indeed the same or did the author used two different normalizers within the same plot? Two luciferase plot shown if Figure 4A should be merged. The biology of Rab1b-DN is not properly explained. If authors expect that it no longer interacts with IRES, they should provide additional experiments to support this conclusions. Moreover, the use of Rab1b-DN is not sufficient to support authors conclusions that Rab1b modulates IRES-dependent translation, as the use of dominant-negative construct frequently leads to artefacts. The authors should use double knockdowns of Rab1b and Rab1a to address their concern of functional overlap. Also, rescue experiments should be included to ensure specificity of the observed changes in IREStranslation. Figure 4B is also ambiguous: "disrupted Golgi phenotype" as visualized by GM130 staining seems to be present not only in GFP-Rab1b-DN-trasnfected cells, but also in few untransfected cells in the upper panel. 4. The interpretation of Figure 6 is inaccurate as IRES-luc RNA (red) and GFP-Rab1b (green) do not colocalize (there is no yellow) and are only adjacently localized. Therefore, it is unclear how the authors performed colocalization quantitation in Figure 6 (also corresponding EV figure). Furthermore, GFP-Rab1b does not always co-localize with Golgi marker GM130 ( Fig EV4). Therefore, the authors cannot conclude that IRES drives mRNA to localize to ER-Golgi. It would be useful to use Golgi marker together with the Rab1b or ARF5 and the RNA-FISH to prove that these mRNAs are Golgi localized. Moreover, here distribution of GFP-Rab1b (dispersed or in foci) seems to be dependent on reporter transfected (IRES vs no-IRES reporter). 5. It is difficult to follow which data is used in analysis for particular figures. For example, the authors state that 'Only factors identified in both replicates with more than 2 peptides (FDR <0.01) were considered for computational studies (R2 {greater than or equal to} 0.81)'. However, in Table 1 Rab1b, was not detected in both replicates for SL123 pull-down but it is still considered as SL123 interactor in Figure 3A. This discrepancy is especially troubling given that Rab1b is one of the proteins that authors choose for further investigation. 6. Some of the presented figures are redundant and can be combined or moved to the Supplement. For example, it is unclear what additional information Figure 1D provides. The Figures are not very well described in the main text and the legends. For example, the importance and meaning of Figure  2 is poorly reported in the text and the neither the significance level of white nodes nor the scale for circle sizes are reported in the figure legend. Error bars are missing in Figure 5A. The statistical test used and exact p-value should be described for each symbol used in the Figures. The manuscript by Fernandez-Chamorro et al. describes the identification of novel proteins associated with domain 3 of the foot-and-mouth-disease virus IRES, and the initial functional characterization of two of these interactors Rab1b and ARF5. The proteomics data will be a useful resource for the community. It is very interesting that the authors find evidence of direct RNA binding activity of Rab1b and ARF5. This is a surprising result but the in vitro data are, overall, compelling. Experiments in cells support a functional role of Rab1b and ARF5 in regulating the distribution of IRES-containing RNAs. Although the paper has its merits, there are a number of substantive issues that need to be addressed before I could recommend publication.
Major points: 1. Although it appears that the distributions of Rab1b and IRES-luc RNA cells are related, the term co-localization is too strong. There is very little overlap, if any, of the two signals ( Figure 5B). Manders' overlap coefficient (M1) is defined as the ratio of the "summed intensities of pixels from the green image for which the intensity in the red channel is above zero to the total intensity in the green channel" and M2 is defined conversely for red. Therefore, M1 (or M2) is an indicator of the proportion of the green signal coincident with a signal in the red channel over its total intensity, which may even apply if the intensities in both channels are really different from one another.

RESPONSE -To
Since this quantitative analysis determines the coefficient of colocalization for different signals, including RNA and proteins (Cerase et al, 2014, PNAS 111: 2235-2240), we would like to maintain the term colocalization for our data, in spite of the fact that we agree with the reviewer that it is difficult to distinguish between overlap and colocalization. Importantly, the revised quantitation data are similar, though not identical, to the data shown in original Fig 6B (not Fig 5B). Moreover, there is statistically significant differences between the colocalization of Rab1b-GFP and IRES-luc RNA compared to the data obtained with cap-luc RNA (P = 0.0003). Same results apply to colocalization of ARF5-GFP with IRES-luc RNA (Fig. S6A). Please, see changes in p. 10 We would like to stress that we are measuring the overlap of the GFP-tagged protein signal and the red signal derived from the RNA FISH using Stellaris probes. The latter is observed as a concentrated spot (Kochan et al, 2015, Biotechniques 59: 209-212), while GFP-ARF5 and GFP-Rab1b signals mark the ER-Golgi, therefore a much larger cytoplasmic area. Since the red signal occupies a tiny spot compared to the diffuse green GFP signal, the probability to see yellow signals is reduced. Attempts to carry out RNA FISH using other procedures (Shaffer et al, 2013, PLoS One 8: e75120; Raj et al, 2008, Nat Methods 5: 877-879) failed to improve the RNA signal detected.
A key part of the model in Figure 7 is the co-incidence of the RNA and Rab1b on the ER but there is no evidence presented to support this. Either more experiments need to be performed to determine the location of the IRES-luc clusters (e.g. with other GFP-tagged markers of the ER, Golgi or ER-Golgi intermediates) or the model needs to be revised substantially and conclusions toned down. I anticipate that additional experiments to refine/strengthen the model could be completed in the usual timeframe for revisions. Fig 7 to better summarize the results of our work in relation to published data, as well as to soften some conclusions. Briefly, Rab1b-GTP is now both on the ER and cis-Golgi, and the IRES-luc reporter RNA is close to but separated from the ER. To support our model we have included data for Rab1b-ER colocalization in the revised text (Martinez et al Furthermore, to address the point raised by the reviewer we have modified Fig 4B,C. We show better images for the colocalization of the GM130 Golgi marker (endogenous protein) with Rab1b and ARF5 (GFP-tagged proteins expressed from transfected constructs). White arrows in the image denote colocalization of both proteins in transfected cells, while white asterisks mark red signals corresponding to the Golgi images in non-transfected cells. In addition, we have done experiments in order to measure colocalization of other ER-Golgi markers (such as GM130, and ERGIC53, and calnexin-CT) using antibody-guided protein staining) with little success, presumably due to the lack of detection of the RNA signal under these conditions. This lack of RNA signal detection (p. 14) has been noted in previous works (Kochan et al, 2015, Biotechniques 59: 209-212).

RESPONSE -We have revised the model presented in
On a related point, it is not clear from the methods what criteria were used to accept or reject colocalization throughout the study. This information should be added including any measures taken to exclude subconscious bias in the analysis (e.g. blinding).
RESPONSE -As explained above, we have performed a blind analysis using the plug-in Coloc 2 to determine the Manders colocalization coefficient. This is indicated in the revised text, p. 19 and revised Figs 4B, 6B,C and S6A. Please see also answer above to point 1.
2. On two occasions the authors draw conclusions by comparing values that were seemingly obtained from different experiments.
"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 GFPDN-Rab1b protein (34 and 32%, respectively) ( Fig 6C). Hence, a significant decrease in GFP-DNRab1b colocalization with IRESluc (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 IRES driven RNA localization." "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".
If the data in question were derived from side-by-side experiments this should be made clear. If not, it is not appropriate to compare these values directly. This point can be illustrated by there (not surprisingly) being some differences in mean values between equivalent experiments performed separately (e.g. cap-luc data in Figure 6B and C). 3. I could not find any information on the nature of the control RNAs (sequence, identity, length, secondary structure) used for the aptamer-based pulldowns or EMSA so it is impossible to judge if they are suitable. This information is essential and should be included in the main text (with any additional details in the methods).

RESPONSE -Transfections and hybridization assays shown in
RESPONSE -The control RNA used in pull-down is indicated in p. 5 (Ponchon et al, 2009, Nat Protoc. 4: 947-59). The control RNA used in EMSA is described in p. 17. This RNA was obtained by T7 RNA polymerase transcription of the polylinker sequence of pGEMT (Promega). The predicted secondary structure of this 94 nt RNA, GC content 68%, predicted to fold into a stemloop interrupted by internal bulges, G= -33.4 Kcal/mol.
Minor points: 1. For non-specialists the authors should make it clear that FMDV is a picornavirus.
RESPONSE -We appreciate these corrections. We have revised the text to avoid typographical errors.
3. "Overall, the RNA-protein binding results match with the proteomic identification in about 75% of the four proteins analyzed". This sentence needs clarifying. What exactly do the authors mean?
RESPONSE -To address this point we decided to delete old Fig 3A and instead cite Table 1 that provides the score obtained in each biological replicate of selected proteins interacting with the IRES subdomains (see changes in p. 8, and revised Fig 3) 4. The term GFP-ARF5-IRES colocalization is potentially misleading due to the use of a hyphen for GFP-ARF5 and to denote an interaction. The authors might want to use "colocalization of GFP-ARF5 with IRES".

RESPONSE -Revised.
5. For non-specialists, what SHAPE analysis reports on should be defined in the Discussion.
RESPONSE -To address this point we have inserted a sentence to better explain the results of SHAPE reactivity upon incubation of the IRES with purified ribosomes, and the potential implication of RNA structure on protein binding (p. 12).
6. Do the authors know what the nucleotide status of the recombinant ARF5 and Rab1b is likely to be? It might be worth commenting that the influence of GTP or GDP on RNA binding activity should be investigated in a future study.
RESPONSE -We agree with the reviewer that it will be wise to carry out assays using GTP-or GDP-charged proteins in future studies. Currently we do not know the status of the Rab1b or ARF5 proteins purified from bacteria used in our band-shift assays. Data from other authors suggest that a high proportion of bacterial purified GTPases are bound to GDP after the long purification steps because of their intrinsic GTPase rate (Smith & Rittinger, 2002, Methods Mol Biol. 189: 13-24).
7. The methods section does not describe how RNA-protein complexes were eluted from the beads in the pulldown experiments. Was this through biotin or SDS buffer? This information should be provided in the methods.
RESPONSE -Elution of proteins from beads was carried out using SDS buffer, 95º 2 min (p. 16). 9. Figure 7. Is the GTP status of Rab1b dependent on being membrane associated? Free Rab1b-GFP is shown in the model. Perhaps a comment needs to be added to the legend to clarify what the authors think is happening in this step. Might the RNA be associating with vesicles? RESPONSE -Published data (Alvarez et al, 2003, Mol Biol Cell 14: 2116-2127Hutagalung & Novick, 2011, Physiol Rev 91: 119-149) showed that the GTP status of Rab1b is critical to determine its association to ER membranes, as stated in the revised manuscript (p. 13). This idea was incorporated on the model presented in Figure 7. RNA association with vesicles has not been analyzed.
Reviewer #2 (Comments to the Authors (Required)): In this manuscript by Fernandez-Chamorro et al the authors use a streptavidin-based purification system to isolate proteins that bind to different region of the FMDV IRES. Rather surprisingly, they find enrichment of proteins involved in the Golgi-ER transport system and they show that the IRES is sufficient to cause colocalisation of reporter RNAs with one of these proteins, Rab1b. This is a very interesting finding and could provide new information about how the FMDV virus sustains infection in mammalian cells and in the longer term, provide new therapeutic avenues.
Major points.
1. Since the authors have only shown binding of ARF5 and Rab1b to the FMDV IRES they should name the virus in the title. In addition, there is no mention of the fact that it is the FMDV IRES that is being studied in the abstract and this needs to be amended. RESPONSE -Done as requested (p.1, 2).
Figures 1 and 2. These were both clear and the data are well described. It is stated in the discussion that there are no previous reports of ARF5 or Rab1b binding to RNA. Have all the interactome capture data sets that are available been compared in this regard? RESPONSE -We have compared several published proteome obtained by RNA interactome capture looking for Rab1b and ARF5 with little success. However, it is worth mentioning that the yeast GTPase Ypt1, the homologue of Rab1, associates in vivo with unspliced HAC1 RNA. This association is impaired during unfolding protein response (Tsvetanova et al, 2012, PLoS Genet 8: e1002862).
In addition, there are several new manuscripts on bioRxiv which describe more extensive ways in which to identify RNA binding proteins using separation with trizaol based reagents (e.g. The Human RNA-Binding Proteome and Its Dynamics During Arsenite-Induced Translational Arrest Jakob Trendel, Thomas Schwarzl, Ananth Prakash, Alex Bateman, Matthias W Hentze, Jeroen Krijgsveld bioRxiv 329995; doi: https://doi.org/10.1101/329995). It may be worth examining the data in these papers as well since they identify many more proteins that have RNA binding capacity.
RESPONSE -Thanks for this comment. Rab1b, but not ARF5, is on the list of this study (bioRxiv 329995; doi: https://doi.org/10.1101/329995). However, none of them are included in the work by Castello et al (Cell 149: 1393(Cell 149: -1406. Thus, to emphasize the newly found IRES-binding feature of Rab1b and ARF5, we inserted a comment and the references suggested by the reviewer in the revised text (p. 14). Figure 3. The authors show that Ra1b1b binds to RNA in vitro therefore it would be useful for the authors to carry out experiments using competitor RNAs to calculate dissociation constants.

RESPONSE -As requested we have performed new experiments using three different competitor
RNAs (SL123, SL3abc, and the control RNA) for Rab1b and ARF5, using the protein concentration that produced the highest % of retarded probe (4.5 nM for Rab1b, and 100 nM for ARF5). The results are shown in the revised Fig 3B,D. The data show that unlabelled RNA SL123 (0 to 30 nM, probe to competitor RNA ratio 1:1 to 1:200) effectively competes out Rab1b binding (Kd ~ 4.91e-005 nM), while higher concentration of RNA SL3abc was required to reach similar level of competition (Kd ~ 2.16 nM). Likewise, binding of ARF5 to probe SL123 was competed by unlabelled SL123 RNA (0 to 75 nM, ratio of probe to competitor RNA 1:1 to 1:500). In this case, the competition of SL3abc was very similar to SL123 (Kd 10.61 nM and 20.59 nM). As expected, competition assays carried out with control RNA fail to compete the interaction of either Rab1b or ARF5 with SL123. These results reinforce the RNA binding capacity of these proteins. Please, see changes in p. 7.
This would enable the relative strength of the interaction can be determined relative to other proteins which interact with this IRES, such as Gemin 5.  RESPONSE -We appreciate this comment. However, these proteins do not form part of the same complex, and to our knowledge it is not known whether Rab1b and ARF5 interact with each other. Therefore, results from the experiment would be difficult to interpret. Studies of ARF4 and ARF5 proteins conducted with Dengue virus (Kudelko et al, 2012, J Biol Chem 287: 767-777), a non-IRES containing RNA virus, suggested their requirement at an early pre-Golgi step for virus secretion. However, the study of the IRES-containing HCV (Farhat et al, 2016, Cell Microbiol 18: 1121-1133 suggested that ARF4 and ARF5 are required for viral RNA replication, together with a modest effect on translation in siRNA ARF4 plus ARF5 depleted cells, but not for secretion. In contrast to these studies, GEF activity and ARF activation are not required in poliovirus infection (Belov et al., 2010Cell Microbiol 12: 1463-1479. RESPONSE -Thanks for raising this point, which is discussed in p. 12-13: "The finding that Rab1b and ARF5 proteins interact directly with the IRES was unprecedented. No report of their RNA-binding capacity were available despite anterograde transport pathway participates in the life cycle of various RNA viruses, including picornaviruses and flaviviruses (Kudelko et al, 2012;Midgley et al, 2013). In the case of Rab1b, different results have been reported for hepatitis C (HCV). While a work using viral replicons (Farhat et al, 2016) involved this protein on viral RNA replication and translation, a recent study inactivating the endogeneous Rab1b via expression of the legionella pneumonia DrrA protein as well as using DN mCherry-Rab1b constructs caused intracellular accumulation of HCV RNA, suggesting that inhibition of Rab1b function inhibits virus particles release (Takacs et al, 2017). Our study does not deal with infectious RNA. Hence, analysis of the secretion step is beyond the scope of our work. "

RESPONSE -
Reviewer #3 (Comments to the Authors (Required)): In the manuscript entitled 'Rab1b and ARF5 are novel RNA-binding proteins involved in IRESdriven RNA localization', Fernandez-Chamorro et al. describe identification of two novel IRESbinding proteins, Rab1b and ARF5, and propose that they regulate IRES-dependent translation and localization of IRES-containing mRNA to Golgi. Overall the manuscript is well written and the rationale is well structured and easy to follow. However, the authors do not provide sufficient evidence to support their conclusions and the presentation of the results in the manuscript should be improved.
1.The authors perform protein pull-down followed by mass spectrometry using D3-domain fragments of IRES-element. These fragments are contained in one another, such that: D3>SL123>SL3abc>SL3abc. Therefore, it is expected that majority of the proteins bound to smaller fragments should also be detected in the pull-downs with the larger domain pieces.
However, the authors observe only a modest overlap between different domains' interactomes ( Figure 1) and many proteins detected with the short fragments do not bind larger fragments. For example, more than 30% (70 out of 214) proteins detected with SL123 are unique for this fragment, and not detected in D3-interactome, despite the fact that entire SL123 fragment is contained in D3 domain. These observations are concerning, as it suggests there is a large nonspecific binding in the shorter domains when they are expressed alone. The authors should address this issue. Fig 1C. However, although the subdomains are contained within domain 3, they do not have the same RNA conformation. As explained in the introduction, a key point of this study was to understand the implication of RNA conformation on protein binding. Importantly, while these transcripts are subdomains of domain 3 contained in one another, there are tertiary interactions connecting the stem-loop 3a (SL3a) with the C-rich loop of transcript SL123, which in principle, would modify RNA accessibility affecting protein binding (p. 3): "Domain 3 is a selffolding 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 factors interacting with this domain and their potential functions remain poorly studied and need to be investigated". To better explain this point, we have tried to highlight the relevance of RNA structure for protein interactions in the revised version (p. 3-4, and p. 12).

RESPONSE -The overlapping between the interactome associated to the different subdomains is shown in
The increased number of proteins interacting with SL123 suggests a different accessibility to ligands compared to domain 3. Differences in protein binding between D3 and SL123 are noticed in band-shift assays, not only for Rab1b and ARF5, but also for PCBP2 and Ebp1. Furthermore, the competition assays shown in revised Fig 3 shows that while both Rab1b and ARF5 bind efficiently to SL123 and SL3abc, addition of SL123 competes the retarded complex formation more strongly for Rab1b than ARF5. Furthermore, we would like to point out that the proteomic approach is not quantitative, and does not distinguish direct binding from secondary interactions. This is one of the reasons that lead us to validate the RNA-binding capacity of Rab1b and ARF5 using EMSA with each subdomain and purified proteins. This rationale is indicated in p. 7: "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 interaction with factors trafficking between organelles." Moreover, there is insufficient validation. In the pull-down experiments, the authors must perform a validation using immunoprecipitation followed by a western blot for Rab1b and ARF5. This is especially important given that Rab1b was not detected in all of the mass-spec replicates.
RESPONSE -As requested, we have performed western blot assays of the pull-down samples to see whether Rab1b and ARF5 proteins were differentially recruited with the transcripts. Rab1b was readily detected with a specific antibody (see Figure below). In contrast, the antibody against ARF5 weakly detected the protein in the input sample. Regarding the differences in mass-spec identification of the replicates, we would like to stress that interaction of the proteins with the IRES subdomains was validated using two different complementary assays: i) RNA-protein interaction in vitro using purified proteins and in vitro synthesized transcripts (Fig 3), and ii) colocalization of RNA with the proteins expressed in double transfected cells (Fig 6B). Therefore, given the differences in protein identification between these methodologies, mass spectrometry identification of proteins and immunodetection of proteins using antibodies following biochemical purification of proteins present in total lysates, we do not think that the lack of immunodetection of ARF5 in the pull-down samples invalidates the results shown in our study.  Fig 3 (Fig S4).
The legend should explain which RNA was used for negative control and how table in Figure 3A was generated, i.e. what '+' and '-' stand for. If '+' means interaction detected in EMSA, the authors should explain why their negative control interacts with PCBP2. Fig 3A, we decided to delete this panel and instead cite Table 1 that provides the score obtained in each biological replicate of selected proteins (see changes in p. 8, and revised Fig 3).

RESPONSE -Given the confusion generated with
Moreover, ARF5 concentration required to observe binding to IRES subdomains is ~200nM, a 100-times more than for other proteins. Such high protein concentration could result in nonspecific binding. The authors should include a control protein to exclude the possibility that IRES domains bind to non-specific proteins at higher concentrations. RESPONSE -We respectfully disagree with this comment. As shown in Fig 3, a control RNA does not produce retarded complex using the same Rab1b or ARF5 protein concentration. The differences in RNA-binding affinity of well-characterized proteins are related to the distinct RNAbinding domains of each protein, as well as to the type of RNA sequence/structure recognized (as shown for PCBP2 and Ebp1, concentration range between 0.1 to 10 nM, or 1 to 100 nM, respectively). Other examples can be seen in Fernandez-Chamorro et al, 2014, Nucleic Acids Res 42: 5742-5754. Moreover, low affinity does not imply lack of specificity as reviewed by (Helder et al, 2016, Curr Opin Struct Biol 38: 83-91).
3. In Figure 4 the authors examine the effect of Rab1b and ARF5 silencing on IRES-mediated translation using a luciferase assay. The observed effects are modest and it is unclear if they would be physiologically relevant. Silencing of known IRES-translation regulator or pharmacological inhibition should be included to estimate the background. The authors use normalized Luc activity/amount of protein which does not reflect the effects of transfection efficiency. Dual firefly/renilla luciferase assay is a better normalization approach. RESPONSE -We have performed experiments using bicistronic constructs were the IRES-driven activity was normalized for the first cistron (revised Fig 4A,Mat and Meth,p. 17). However, we would like to point out that the results are similar to the data obtained using monocistronic RNAs (moved to Fig S5).
Moreover the data are normalized to the extent where it is hard to judge how the original data look like. For example, is luciferase activity of IRES-luc and cap-luc in the presence on Rab1b-wt indeed the same or did the author used two different normalizers within the same plot? Two luciferase plot shown if Figure 4A should be merged.
RESPONSE -These are experiments conducted separately for siRab1b, or siARF5, including the sicontrol RNA each time. Therefore, sicontrol RNA data is used to normalize each assay.
The biology of Rab1b-DN is not properly explained. If authors expect that it no longer interacts with IRES, they should provide additional experiments to support this conclusions.
RESPONSE -We tried this experiment, however the protein was not purified in sufficient quality to be used in EMSA, and therefore we decided not to pursue this experiment.
Moreover, the use of Rab1b-DN is not sufficient to support authors conclusions that Rab1b modulates IRES-dependent translation, as the use of dominant-negative construct frequently leads to artefacts. RESPONSE -We agree with the reviewer that in some cases the use of dominant negative constructs may lead to artefacts. However, in the particular case of Rab1b, the dominant negative mutant used in our study is well characterized in previous studies (Alvarez et al, 2003, Mol Biol Cell 14: 2116-2127Midgley et al, 2013, J Gen Virol 94: 2636-2646, and it is important to point out that this construct inactivates both Rab1b and Rab1a, which are functionally redundant (Tisdale et al, 1992, J Cell Biol 119: 749-761). In addition, a DN-library was useful to study Rab1b regulation of HCV secretion from the ER to Golgi (Takacs et al, 2017, Cell Rep 21: 431-441). A comment to this study has been included in the Discussion (p. 12-13). Examples of DN constructs are eIF4A-DN constructs, which helped to demonstrate the requirement of the helicase eIF4A for translation initiation in many studies (de Breyne et al, 2008, RNA 14: 367-380;Pause et al, 1994, EMBO J 13: 1205-1215Svitkin et al, 2001, RNA 7: 382-394).
The authors should use double knockdowns of Rab1b and Rab1a to address their concern of functional overlap. Also, rescue experiments should be included to ensure specificity of the observed changes in IRES-translation.
RESPONSE -Rab1a and Rab1b share 92% amino acid homology (p. 9). Therefore, attempts to generate double knockouts have being unsuccessful. This caveat was solved in previous works using the Rab1b DN construct that inactivates both forms (Alvarez et al, 2003, Mol Biol Cell 14: 2116-2127Midgley et al, 2013, J Gen Virol 94: 2636-2646 Figure 4B is also ambiguous: "disrupted Golgi phenotype" as visualized by GM130 staining seems to be present not only in GFP-Rab1b-DN-trasnfected cells, but also in few untransfected cells in the upper panel. RESPONSE -We have modified the figure to show representative images of the Golgi phenotype in transfected cells. In addition, we are providing Manders coefficients for the co-localization of the endogenous GM130 Golgi marker protein with the GFP-tagged proteins expressed in transfected cells. Figure 6 is inaccurate as IRES-luc RNA (red) and GFP-Rab1b (green) do not colocalize (there is no yellow) and are only adjacently localized.

The interpretation of
RESPONSE -To address this point we have measured the colocalization of Rab1b-GFP and the RNA detected by FISH using the plug-in Coloc 2 (Schindelin et al, 2012)) and Manders coefficient (see p. 19,Figs 6B,C and S6B). Specifically, Manders´1 (M1) coefficient denotes the colocalization of red dots RNA with green GFP-protein signals in double transfected cells. We would like to stress that we are measuring the overlap of the GFP-tagged protein signal and the red signal derived from the RNA FISH using Stellaris probes. The latter is observed as a tiny spot (Kochan et al, 2015, Biotechniques 59: 209-212), while GFP-ARF5 and GFP-Rab1b signals mark the ER-Golgi, therefore a much larger cytoplasmic area. Since the red dot occupies a tiny spot compared to the green GFP area, the probability to detect yellow signals is reduced. Please, see also answer to comment 1 of reviewer #1.
Since this quantitative analysis determines the coefficient of colocalization for different signals, including RNA and proteins (Cerase et al, 2014, PNAS 111: 2235-2240), we would like to maintain the term colocalization for our data, in spite of the fact that it is difficult to distinguish between proximity and colocalization. As shown in Fig 6B, there is a statistically significant difference between the colocalization of Rab1b-GFP and IRES-luc RNA compared to the data obtained with cap-luc RNA.
Therefore, it is unclear how the authors performed colocalization quantitation in Figure 6 (also corresponding EV figure). Furthermore, GFP-Rab1b does not always co-localize with Golgi marker GM130 (Fig EV4). Material and Methods (p. 19). We have revised the images shown in Fig 6B,C, and also Fig S6. Concerning Fig S4 (now Fig 4B), we agree with the referee that there are some instances where the red signal of GM130 is not colocalizing with GFP, as expected since GM130 denotes the endogenous signal present in all cells. To avoid further confusion we are using white arrows to mark GM130 in transfected cells, while asterisks denote GM130 staining in untransfected cells.

RESPONSE -Quantitation is explained in
Therefore, the authors cannot conclude that IRES drives mRNA to localize to ER-Golgi. It would be useful to use Golgi marker together with the Rab1b or ARF5 and the RNA-FISH to prove that these mRNAs are Golgi localized.
RESPONSE -In this study we show data for the colocalization of Rab1b and ARF5 with the Golgi marker GM130 (Fig 4B). Next we show that there is colocalization of RNA carrying the IRES element on its 5´UTR and Rab1b (Manders M1 coefficient 0.90) (Fig 6B). Although to a lesser extent (M1 0.66), there is also colocalization of IRES-luc RNA with ARF5 ( Fig S6A). Statistically significant differences were obtained for the colocalization of the same proteins with the cap-luc RNA lacking the IRES element (please see data and P values in p. 10). From these data we conclude that Rab1b and ARF5 are involved in the localization of IRES-reporter RNAs on the ER-Golgi.
In addition to these proteins, we have done experiments to measure colocalization with other Golgi markers, which unfortunately failed. The lack of RNA signal detection has been also noted in other works (Kochan et al, 2015, Biotechniques 59: 209-212), as mentioned in p. 14: "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." Moreover, here distribution of GFP-Rab1b (dispersed or in foci) seems to be dependent on reporter transfected (IRES vs no-IRES reporter). RESPONSE -Thanks for raising this point. There is not co-transfection with reporter IRES or no-IRES in Fig 4B (old Fig EV4 is now moved to Fig 4B). Therefore, the different location of the GFP-Rab1b depends on the expression of the wild type protein or the mutant construct Rab1b DN, and not the IRES reporter. For instance, the distribution of green signal of GFP-Rab1b in transfected cells (marked with a white arrow) is similar to the red signal of GM130 in all cells, irrespectively of whether they are transfected or not. Please see that we have marked with white arrows the Golgi in transfected cells, and white asterisks non-transfected cells in all panels of the revised version of Fig 4B. The different distribution of RNA reporters is shown in Fig 5, where there is no coexpression of Rab1b or ARF5 GFP-constructs. This figure shows the detection of the IRES-RNA signals in close proximity relative to cap-luc RNA.
5. It is difficult to follow which data is used in analysis for particular figures. For example, the authors state that 'Only factors identified in both replicates with more than 2 peptides (FDR <0.01) were considered for computational studies (R2 [greater than or equal to] 0.81)'. However, in Table  1 Rab1b, was not detected in both replicates for SL123 pull-down but it is still considered as SL123 interactor in Figure 3A. This discrepancy is especially troubling given that Rab1b is one of the proteins that authors choose for further investigation.
RESPONSE -The referee is right in that the criteria used to select candidates for computational analysis from the proteomic interactome was based in the identification of more than 2 peptides, FDR <0.01. Then, the riboproteomic data together with other observations was used as the first step to investigate how cellular proteins can contribute to IRES-driven translation. Hence, the computational studies reinforced the hypothesis that several members of the ER-Golgi network were present in the IRES-interactome (see Fig 2). Other members of the ER-Golgi network were detected in the proteomic interactome (for instance CopA, Sec31a, Sar1a shown in Table 1 among others listed in dataset 1). These proteins have been analysed in relation to the secretory pathway of picornaviruses (Belov et al, 2008, PLoS Pathog 4: e1000216;Midgley et al, 2013, J Gen Virol 94: 2636-2646van der Schaar et al, 2016, Trends Microbiol 24: 535-546) but their interaction with RNA remains unknown. Further studies will be needed to determine whether their identification in our IRES-pull down was mediated by secondary interaction.
Additionally, although Rab1b appeared only in one biological replicate with SL123 and D3, the score was high (18 and 31.65, respectively, see Table 1), even higher than the score of both replicates with SL3a and SL3abc. Therefore, for completeness, the selected proteins Rab1b and ARF5 were subjected to RNA-binding with all transcripts. The same reasoning lead us to perform the RNA-binding assays with the known IRES-binding factors PCBP2 and Ebp1. We reasoned that this result could provide a broader view of the RNA-binding capacity of Rab1b and ARF5. Thank you for submitting your revised manuscript entitled "Rab1b and ARF5 are novel RNA-binding proteins involved in FMDV IRES-driven RNA localization". We have just received the last report on your work, please excuse again the delay in getting back to you.
As you will see, reviewer #2 and #3 appreciate the introduced changes and reviewer #1 now also supports publication, pending further minor revision. We would thus be happy to publish your paper in Life Science Alliance pending final revisions to address reviewer #1's comments. We think it may be good to talk about 'juxtapositioning' of the signals instead of 'co-localization' and to indeed provide further imaging data in the supplement to allow others to recapitulate the different signals observed for IRES-luc versus cap-luc and Rab1/ARF5 signals.
To upload the final version of your manuscript, please log in to your account: https://lsa.msubmit.net/cgi-bin/main.plex You will be guided to complete the submission of your revised manuscript and to fill in all necessary information.
To avoid unnecessary delays in the acceptance and publication of your paper, please read the following information carefully.

A. FINAL FILES:
These items are required for acceptance.
--An editable version of the final text (.DOC or .DOCX) is needed for copyediting (no PDFs).
--High-resolution figure, supplementary figure and video files uploaded as individual files: See our detailed guidelines for preparing your production-ready images, http://life-sciencealliance.org/authorguide --Summary blurb (enter in submission system): A short text summarizing in a single sentence the study (max. 200 characters including spaces). This text is used in conjunction with the titles of papers, hence should be informative and complementary to the title. It should describe the context and significance of the findings for a general readership; it should be written in the present tense and refer to the work in the third person. Author names should not be mentioned.