Prevention of dsRNA‐induced interferon signaling by AGO1x is linked to breast cancer cell proliferation

Abstract Translational readthrough, i.e., elongation of polypeptide chains beyond the stop codon, was initially reported for viral RNA, but later found also on eukaryotic transcripts, resulting in proteome diversification and protein‐level modulation. Here, we report that AGO1x, an evolutionarily conserved translational readthrough isoform of Argonaute 1, is generated in highly proliferative breast cancer cells, where it curbs accumulation of double‐stranded RNAs (dsRNAs) and consequent induction of interferon responses and apoptosis. In contrast to other mammalian Argonaute protein family members with primarily cytoplasmic functions, AGO1x exhibits nuclear localization in the vicinity of nucleoli. We identify AGO1x interaction with the polyribonucleotide nucleotidyltransferase 1 (PNPT1) and show that the depletion of this protein further augments dsRNA accumulation. Our study thus uncovers a novel function of an Argonaute protein in buffering the endogenous dsRNA‐induced interferon responses, different than the canonical function of AGO proteins in the miRNA effector pathway. As AGO1x expression is tightly linked to breast cancer cell proliferation, our study thus suggests a new direction for limiting tumor growth.

The aut hors have to thoroughly revise their experiment s and add appropriat e cont rols to test the st at ed hypot heses. I do not recommend this manuscript for publicat ion wit hout major revision.
Major comment s: In Figure 1. The aut hors show expression of AGO1x in two human cancer cell lines (MDA-MB-231 and HeLa) but not in HEK293 cells using a novel ant ibody generat ed to specifically recognize the read-t hrough product . The lack of AGO1x in Hek293 compared to Hela cells is surprising, because similar amount s of AGO1x have been det ect ed in bot h cell lines previously (Singh et al., and Eswarappa, EMBO J 2019, PMID: 31330067). Furt hermore, the aut hors observe localizat ion of AGO1x to the nucleus in cont rast to the previously report ed cyt oplasmic localizat ion. The aut hors should comment on these discrepancies and use knock-down or knock-out of AGO1 to assess the specificit y of their immunost aining. To assess specificit y of this signal, the aut hors could use tools present ed in lat er st ages of the manuscript , i.e. 'loss of AGO1x' cells. Because AGO1 is not an essent ial gene, a st raight -forward KO cell line would also be a great cont rol. The aut hors should also perform st aining using an est ablished AGO1 ant ibody that will enable det ect ion of bot h AGO1 and AGO1x. Figure 3. Here the aut hors report the use of CRISPR/Cas9 genome edit ing to eliminat e the AGO1 read-t hrough product (AGO1x). In panel A the aut hors forgot to indicat e the generat ed mut at ions (from their diagnostic genotyping). The authors should also add a methods section about the detailed characterization of these cell lines. The accompanying western blot seems confusing as the higher band corresponding to AGO1x is clearly visible in all three lanes using the pan_AGO1 antibody, but a signal is absent in the two clonal lines using the AGO1x antibody. The authors should use quantitative imaging technologies like Odyssey or other methods for this analysis. Cell growth assays should be performed with clonal control cell lines that arose from the same initial CRISPR/Cas9 transfection. Mutations by repair mistakes of a guided Cas9 cut generally result in cell clones with variable mutations and deletions that will have different impact on AGO1 read-through translation. The authors should characterize a set of cell lines that include functional AGO1 null, AGO1x null, AGO1 null and no effect on AGO1X expression mutations. This is particularly important as exact homozygous editing events are rare and MDA-MB-231 are in addition aneuploid (ATCC). Alternatively, the authors could use precise CRISPR editing (with designed donor constructs) to eliminate either only the canonical AGO1 STOP codon or only the AGO1x STOP codon. Multiple clonal cell lines have to be tested for both interference and controls. FLAG-tagged Rescue constructs expressing either FLAG-AGO1 or FLAG-AGO1x (as used in Fig. S8) could also be used to address the specificity of the observed effect.  For pull-down experiments, the authors should compare AGO1 pull downs with AGO1x pull downs to identify proteins and RNAs that specifically interact with AGO1x but not AGO1. Using IgG as a control does not allow discrimination of factors associated with AGO1 or AGO1x. Because the authors aim to claim a specific function for AGO1x and not a novel function for overall AGO1, a thorough differential analysis of both isoforms is essential. Why would PNPT1 knock down increase the abundance of dsRNA in both the parental and the presumable AGO1x mutant cell lines? Is this effect independent of AGO1x?
Minor points: The authors should revise their introduction to better represent relevant miRNA literature: Argonaute proteins predominantly localize to the cytoplasm and target mRNAs for degradation. Nuclear localization and potential function of human Argonaute proteins has long been a controversial topic in the small RNA field and is not a universally observed phenomenon (reviewed by David Bart el Cell 2018, PMID: 29570994 ).

Referee #2:
Ghosh and co-workers report evidence for the exist ence of AGO1x, an unexpect ed Argonaut e isoform, which is proposed to promot e proliferat ion of breast cancer cells by prevent ing dsRNA-induced int erferon signaling. Key findings include: • evidence for expression of AGO1x in human cell lines and breast cancer samples • preferential nuclear localization of AGO1x • AGO1x promotes cell proliferation • AGO1x depletion increases dsRNA accumulation, interferon response, and apoptosis • AGO1x interacts with PNPT1 This manuscript comes on the heels of Singh et al, 2019, which also demonstrated the existence of AGO1x and also characterized it's biogenesis and functional properties. Together, the two studies provide compelling evidence for the existence of AGO1x, and complementary biological insights that will have a substantial impact on the field.
The manuscript may be improved by addressing the following:

Major Concerns
Conclusions drawn from data in Figs. 1-4 are exciting and compelling. In contrast, the evidence supporting the final model (as illustrated in Fig. 5i) has several distinct weaknesses: 1) AGO1x and PNPT1 are suggested to physically interact (Figs. 5i-j and S7a-b) and collaborate to prevent accumulation of 'dsRNA' (Fig. 5I), and yet AGO1x is shown be nuclear (Fig. 2) while PNPT1 is cytoplasmic (Fig. S7c). How do the authors reconcile this? Is the AGO1x/PNPT1 interaction direct or at least insensitive to RNase A?
2) dsRNA staining in Figs. 5d and 5k is shown in only a few cells. Considering these observations are used as a major piece evidence for cellular dsRNA up regulation upon Ago1x knockout, more rigorous analysis is suggested. i.e. what is the distribution of dsRNA staining levels when surveying >100 cells for each sample?
3) Purified Argonaute proteins bind RNA non-specifically. Is dsRNA binding (Fig. 5h) specific to AGO1x? An AGO1 IP could be included as a control to Fig. 5h-since AGO1 levels are >10-fold AGO1x (Fig. 1d) the AGO1 IP should be dominated by RNAs associated with AGO1 (and not AGO1x). 4) Evidence that dsRNA accumulation is responsible for the observed changes in IFN response, apoptosis, and cell proliferation upon AGO1x knockout is tenuous. Thus, there remains a substantial possibility that a different mechanism (including altered miRNA function, activation of the RP-MDM2-p53 pathway, or some other unknown pathway) is the true cause of these effects. In the absence of direct evidence linking the observed 'dsRNA' to the observed cellular responses, I strongly suggest downplaying this conclusion. 5) Related to the comment above, Fig. 5e shows increased rRNA levels in W1A and W5A cells. Are ribosomal protein or mature ribosome levels also altered in these cells? Considering the nucleoar enrichment of AGO1x in Fig. 2b it seems reasonable to suggest a role in ribosome biogenesis, which if perturbed can initiate nucleolar stress leading to anti-tumor responses.
Minor concerns/suggestions 6) The authors state: "Important for its characterization was the realization that in contrast to the ectopically-expressed AGO1x, which has cytoplasmic localization both in our and in the study of Singh et al. (Singh et al , 2019), endogenous AGO1x is found primarily in the nucleus." This statement does not discuss or acknowledge results in Fig. S4 of Singh et al., 2019, which appear to show cytoplasmic distribution of endogenous AGO1x. 7) Similarly, Singh et al , 2019 report that AGO1x makes up 40% of the endogenous AGO1 in HEK293 cells, and appears to be similarly expressed in HeLa cells. These results stand in contrast to the Western blot in Fig. 1d. This difference should be mentioned in some way. 8) Where is PNPT1 in Fig. S8a? 9) Why is the expression change of IFIH1 not shown in Fig. 5b? 10) Describing rRNA as 'dsRNA' is confusing/inaccurate because, although it does contain a substantial amount of base-paired helical structures, rRNA is not double stranded.
11) The authors state: "PNPT1, whose siRNA-mediated depletion replicates the AGO1x depletion phenotype and also leads to the accumulation of dsRNAs." This statement seems incongruent with data in Fig. 5k, which appears to show more dsRNA staining in the W6A sample than either siPNPT1 sample. Fig. S2a: the AGO1x band appears to run much slower than the AGO1 band (in comparison with Fig. S2b). How did the authors verify that this band corresponds to AGO1x?

12)
13) It would be reasonable to suggest that the TR extension reduces miRNA affinity (Fig. S8c) by moving the C-terminal carboxyl group, which interacts with the miRNA 5' phosphate via a lysine and ordered water molecule in crystal structures of canonical human Ago proteins.
Referee #3: In this manuscript Ghosh et al. report that a highly conserved TR isoform of AGO1, AGO1x, localizes exclusively to the nucleus, in the vicinity of nucleoli, where it prevents accumulation of double stranded RNAs by an unknown mechanism and thus, supposedly, mitigates the interferon response and apoptosis. A similar role has also been proposed for Polyribonucleotide Nucleotidyltransferase 1, a newly identified, specific binding partner of AGO1x. Finally, they also showed that AGO1x is overrepresented in highly proliferative breast cancer cells and speculate that targeting AGO1x could potentially represent a new direction for limiting tumor growth. This is a clearly written, technically well-executed, fairly interesting story supported by numerous experiments. However, it suffers from 1) insufficient detail (I've had an impression that it was originally written for some "short-format" journal) forcing the reader to spend quite some time trying to understand what this or that result means and/or how exactly this or that experiment was executed, and 2) the lack of any mechanical insight into the presented observations. Below I summarize all my concerns (in the order of their appearance in the text) that, in my opinion, should be addressed to improve the quality of this study before its prospective acceptance. 1) I suggest the authors to go through the entire manuscript, a result after result, a figure after figure (especially Sup. Figures but not only those), and explain in greater detail how each experiment was done and what can be seen in each figure (expected vs. unexpected and why; many proteins are listed with no description what they do in cells and why they are expected to change their expression in this or the other direction, etc.) with a non-expert reader in mind. It will improve the clarity and readability of your work by a great margin. Also, label the pages and lines, please.
2) Fig. S5a is placed out of order in the main text; in between S2 and S3.
3) "Endogenous AGO1x localizes ...." Fig. 1b should be 2b. Fig.3a; the AGO1x strip is smudgy compared to others shown in the work. Since it is a fairly important result I suggest repeating it aiming for a nicer western blot image. . We found that their expression was indeed increased, both at the mRNA and at the protein levels ( Fig.  5a, b ). In addition, using a previously published method (Post et al , 2017), we uncovered evidence for protein phosphorylation events downstream of interferon signaling, in EIF2AK2 and ADAR ( Fig.  5c )." This is a typical example of the "go figure yourself" approach mentioned above. 9) The fact that the source of dsRNA lies predominantly, if not solely (based on Fig. 5e), in rRNA is rather unexpected. The authors propose that: "...the increased demand for ribosome biogenesis in rapidly dividing cells requires the readthrough AGO1x isoform for resolving dsRNA structures in rRNAs and other molecules."; also rather unexpected, but why not. Since no other explanation has been offered, either for the AGO1x role in dsRNA metabolism in the nucleus, or for its contact with PNPT1, I suggest the authors to explore this idea in greater detail. If correct, ribosome biogenesis should be affected in AGO1x null (as well as PNPT1 knocked-down) cells, which can be easily tested with a bunch of well-established approaches.

4)
10) What could be reason for the observed differences between endogenously vs. ectopically expressed AGO1x? Any idea? Major comments: 26 27 In Figure 1. The authors show expression of AGO1x in two human cancer cell lines (MDA-MB-28 231 and HeLa) but not in HEK293 cells using a novel antibody generated to specifically 29 recognize the read-through product. Furthermore, in HEK293 cells the immunofluorescence with the AGO1 antibody ( Fig. S4) did not 52 recapitulate the WB shown in Fig. 1e, where AGO1 and AGO1x had comparable intensities. In 53 addition, some degree of AGO1x nuclear localization can be inferred from Fig. S4 of Singh et al. 54 as well. Finally, assuming the mechanism proposed by Singh et al., in which the let-7 miRNA 55 guides the TR of AGO1, we note that the level of the let-7a miRNA is, in fact, lowest in our 56 HEK293T cells, followed by HeLa and MDA-MB-231 (Appendix Fig S2C). This is also consistent 57 with our estimates of AGO1x expression in these cells. 58 59 Figure 2. The authors perform immunofluorescence analysis using their new AGO1x antibody in 60 MDA-MB-231 cells and detect a speckled nuclear staining with some co-localization with 61 Nucleolin. To assess specificity of this signal, the authors could use tools presented in later 62 stages of the manuscript, i.e. 'loss of AGO1x' cells. Because AGO1 is not an essential gene, a 63 straight-forward KO cell line would also be a great control. The authors should also perform 64 staining using an established AGO1 antibody that will enable detection of both AGO1 and 65 AGO1x. 66 67 We have indeed addressed the specificity of the antibody for immunofluorescence analysis as 68 the reviewer suggested in Fig the AGO1x staining signal reflected the levels of AGO1x generated from these constructs. We 75 also included AGO1 staining to illustrate that commercially available AGO1 antibody detects 76 both isoforms, and that the signal that we obtain with the AGO1x antibody overlaps with the 77 signal from the canonical AGO1 antibody. Using a pool of siRNAs directed against the AGO1 78 transcript (which also encodes AGO1x) we found that the signals from both the AGO1x and 79 AGO1 were strongly depleted. The results we obtained in immunofluorescence were consistent 80 with those we obtained by WB (previous Fig S2 c,d,g and current Fig EV1E,G). 81 82 Figure 3. Here the authors report the use of CRISPR/Cas9 genome editing to eliminate the 83 AGO1 read-through product (AGO1x). In panel A the authors forgot to indicate the generated 84 mutations (from their diagnostic genotyping). The authors should also add a methods section 85 about the detailed characterization of these cell lines. The accompanying western blot seems 86 confusing as the higher band corresponding to AGO1x is clearly visible in all three lanes using 87 the pan_AGO1 antibody, but a signal is absent in the two clonal lines using the AGO1x 88  Fig. S8) could also be used to address the specificity of the 100 observed effect. 101 102 In the methods section and associated tables we have described our CRISPR/Cas9 editing 103 experiments, including the selection of sgRNAs, the selection and characterization of the cell 104 lines. We have made this description more verbose in the revised manuscript. We added the 105 genotyping in Appendix Fig S4, as requested. We have tested a whole number of cell lines, all 106 of which are clonal but the mutation is not homozygous. As the results we obtained with 107 different sets of sgRNAs and on two distinct parental lines were very consistent, and different 108 from those obtained with control lines, in which an unrelated sgRNA set was used (directed 109 against GFP), we think that our results are not due to spurious effects of transfection or of 110 CRISPR editing. We could not use the FLAG-tagged constructs to investigate the functional 111 effects of AGO1x expression because the localization of the encoded protein did not fully 112 recapitulate the localization of the endogenous form (as discussed in the main text). While we 113 did attempt to replace the AGO1 stop codon, we did not succeed in doing so. However, we think 114 that our results with the lines described above address the issue of specificity, and we did not 115 pursue more complex additional experiments. We point the reviewer to the supplementary material (initial Fig 3, Fig. S4 and S5, current Fig 3  133 and Appendix Fig S4) that contains the requested characterization. We also explained our 134 control line in more detail in the text. 135 136 For pull-down experiments, the authors should compare AGO1 pull downs with AGO1x pull 137 downs to identify proteins and RNAs that specifically interact with AGO1x but not AGO1. Using 138 IgG as a control does not allow discrimination of factors associated with AGO1 or AGO1x. 139 Because the authors aim to claim a specific function for AGO1x and not a novel function for 140 overall AGO1, a thorough differential analysis of both isoforms is essential. 141 142 As AGO1 and AGO1x differ in their localization, we used the nuclear fraction to increase the 143 specificity of identification of AGO1x interactors. Furthermore, to discriminate protein interactors 144 17th Mar 2020 1st Authors' Response to Reviewers 5 specific for AGO1x or AGO1 we have indeed performed precisely the assay suggested by the 145 reviewer (former Fig. S7b, current Fig 6E). In the revised manuscript we have carried out 146 AGO1-IP from the nuclear and cytoplasmic fractions, and found that some putative rRNA 147 targets of AGO1x are also present in the cytoplasmic AGO1-IP (Appendix Fig S10). what we see ( Fig 6G). Furthermore, the IP experiments that we carried out in response to a 165 question by reviewer #2 indicate that the interaction of PNPT1 with AGO1x is direct, and not 166 strictly dependent on RNAs (Appendix Fig S8C). 167 168 Minor points: 169 170 The authors should revise their introduction to better represent relevant miRNA literature: 171 Argonaute proteins predominantly localize to the cytoplasm and target mRNAs for degradation. 172 Nuclear localization and potential function of human Argonaute proteins has long been a 173 controversial topic in the small RNA field and is not a universally observed phenomenon 174 (reviewed by David Bartel Cell 2018, PMID: 29570994). 175 176 We have mentioned the controversy in the introduction of the revised manuscript. 177 178 Referee #2: 179 180 Ghosh and co-workers report evidence for the existence of AGO1x, an unexpected Argonaute 181 isoform, which is proposed to promote proliferation of breast cancer cells by preventing dsRNA-182 induced interferon signaling. Key findings include: 183 184 • evidence for expression of AGO1x in human cell lines and breast cancer samples 185 • preferential nuclear localization of AGO1x 186 • AGO1x promotes cell proliferation 187 • AGO1x depletion increases dsRNA accumulation, interferon response, and apoptosis 188 to prevent accumulation of 'dsRNA' (Fig. 5I), and yet AGO1x is shown be nuclear (Fig. 2) while 204 PNPT1 is cytoplasmic (Fig. S7c). How do the authors reconcile this? Is the AGO1x/PNPT1 205 interaction direct or at least insensitive to RNase A? 206 207 We thank the reviewer for this suggestion. Although we could not do the co-staining with the 208 antibodies for immunofluorescence, as they were both raised in rabbit, we carried out PNPT1 209 immunoprecipitation from nuclear fractions treated or not with RNase A. The WB of these IPs 210 indicated that the AGO1x interaction was slightly impaired, but not abrogated by the RNase A 211 treatment of the lysate prior to the pulldown (Appendix Fig. S8C). This indicates that PNPT1 212 interacts with AGO1x directly, at least in part. Furthermore the identification of PNPT1 in these 213 samples suggests that PNPT1 is also present in the nucleus, as also indicated by our initial 214 imaging analysis (Appendix Fig S8A-B). 215 216 2) dsRNA staining in Figs. 5d and 5k is shown in only a few cells. Considering these 217 observations are used as a major piece evidence for cellular dsRNA up regulation upon Ago1x 218 knockout, more rigorous analysis is suggested. i.e. what is the distribution of dsRNA staining 219 levels when surveying >100 cells for each sample? 220 221 We appreciate the reviewer for pointing to this observation, which is indeed a major piece of 222 evidence in our study. We have provided additional quantifications and expanded the 223 description of these results as suggested Specifically, we counted the numbers of dsRNA foci in 224 Z stacks of each cell type and provided a new Fig 5F (right panel) showing the statistics in the 225 revised manuscript. We have further measured the amount of dsRNAs immunoprecipitated from 226 the different cell types (Fig 5H). Both of these assays show that the total amount of dsRNAs is 227 ~10-fold higher in the mutant lines compared to the control. 228 229 3) Purified Argonaute proteins bind RNA non-specifically. Is dsRNA binding (Fig. 5h) specific to 230 AGO1x? An AGO1 IP could be included as a control to Fig. 5h-since AGO1 levels are >10-fold 231 17th Mar 2020 1st Authors' Response to Reviewers 7 AGO1x (Fig. 1d) the AGO1 IP should be dominated by RNAs associated with AGO1 (and not 232 AGO1x). 233 234 The reviewer correctly points out that AGO proteins bind RNA non-specifically. We were aware 235 of this issue, which we tried to circumvent by purifying the AGO1x protein by IP from the nuclear 236 fraction before carrying out the qPCR. We reasoned that the presence of AGO1 in the nucleus 237 is much less pronounced than in the cytoplasm, and thus the signal that we obtained, further 238 filtered through the AGO1x-IP, should correspond to AGO1x-bound RNAs. To better address 239 the reviewer's question, in the revised manuscript we also provide qPCRs of the J2 antibody 240 targets (identified by sequencing) in AGO1-IP of nuclear and cytoplasmic fractions (Appendix 241 Fig S10). These experiments show that both AGO1 and AGO1x associate with rRNAs that were 242 identified by sequencing RNAs enriched by the J2 antibody. A precise quantification is difficult, 243 because the antibody recognizes both AGO1 and AGO1x isoforms (e.g. Fig EV1) increased activity (Fig 4D-E, Appendix Fig S7B,C). By inhibiting JAKs, which are downstream of 258 IFN signalling, with ruxolitinib, we could rescue the growth of mutant lines ( Fig 4G). Finally, 259 consistent with PKR's inhibition of protein synthesis via eIF2ɑ phosphorylation (Meurs et al,  260 1992), we found the levels of eIF2ɑ to be increased in the AGO1x mutant lines (Fig 5D). 261 Altogether, these experiments provide strong evidence for the dsRNA-IFN axis being 262 responsible for the observed phenotypes. To determine whether the observed IFN pathway 263 activation has a functional impact, for the revised manuscript we have also carried out infections 264 of the HeLa lines with the Sindbis virus and found less accumulation of virus upon infection of 265 mutant lines compared to the control line, consistent with their higher interferon pathway activity 266 (Fig EV4). We do not exclude an involvement of the p53 pathway, as crosstalk between PKR 267 and P53 pathways has been reported (Cuddihy et al, 1999) and indeed, GSEA analysis of our 268 transcriptomics data showed some enrichment of this pathway. However, as the JAK inhibitor 269 rescues the phenotype, we think that the growth limitation is imposed further upstream, and 270 hence cannot be solely p53-dependent (Fig 5A). We have discussed these aspects in more 271 detail in the revised manuscript and also mentioned that we cannot exclude some effect of 272 miRNAs. 273 274 275 17th Mar 2020 1st Authors' Response to Reviewers 8 5) Related to the comment above, Fig. 5e shows increased rRNA levels in W1A and W5A cells. 276 Are ribosomal protein or mature ribosome levels also altered in these cells? Considering the 277 nucleoar enrichment of AGO1x in Fig. 2b it seems reasonable to suggest a role in ribosome 278 biogenesis, which if perturbed can initiate nucleolar stress leading to anti-tumor responses. 279 280 We thank the reviewer for this comment. Fig. 5e (currently 5L) showed the abundance of rRNA 281 fragments in the J2 antibody pulldown. We have carried out the suggested evaluation of 282 ribosome biogenesis, in particular measuring pre-and mature rRNA levels (by qRT-PCR, 283 relative to GAPDH), and we found them to be reduced in mutant lines (Fig Appendix S11A,B). 284 Also reduced was the 40S/60S ratio (Appendix Fig S11 C,D). Consistently, the overall 285 translation was reduced in the AGO1x mutant lines (Appendix S11E,F). The perinucleolar 286 localization of AGO1x and its binding to rRNAs do suggest a role in ribosome biogenesis, which 287 could thereby be perturbed in the mutant lines, leading to some degree of nucleolar stress, 288 consistent with the p53 pathway being identified in our GSEA analysis (Fig 4B and Fig EV3G). 289 Whether these changes impact anti-tumor responses are indeed an interesting topic for future 290 work. We have added these points to the discussion. 291 292 Minor concerns/suggestions 293 294 6) The authors state: "Important for its characterization was the realization that in contrast to the 295 ectopically-expressed AGO1x, which has cytoplasmic localization both in our and in the study of 296 Singh et al. (Singh et al , 2019), endogenous AGO1x is found primarily in the nucleus." This 297 statement does not discuss or acknowledge results in Fig. S4  HEK293 cells, and appears to be similarly expressed in HeLa cells. These results stand in 318 contrast to the Western blot in Fig. 1d. This difference should be mentioned in some way. 319 17th Mar 2020 1st Authors' Response to Reviewers 9 320 At the reviewer's request, we have mentioned this in our discussion of the results obtained in 321 the Singh et al. study. We have also added a few lines about prior estimates of endogenous 322 readthrough frequency in mammalian cells, which have been reported to be at a maximum of 323 ~7% (Loughran et al, 2018). Even for a very prominent target of TR, the gag/pol transcript of 324 retroviruses, the probability of readthrough of the gag UAG stop codon to form the gag-pol 325 polyprotein necessary for virion assembly is ~5% (Honigman et al, 1991)). Therefore, we think 326 that our estimates, based on the results shown in Fig 1D are  We tried to correct phrasing accordingly in the revised manuscript. 344 345 11) The authors state: "PNPT1, whose siRNA-mediated depletion replicates the AGO1x 346 depletion phenotype and also leads to the accumulation of dsRNAs." This statement seems 347 incongruent with data in Fig. 5k, which appears to show more dsRNA staining in the W6A 348 sample than either siPNPT1 sample. 349 350 We agree and have rewritten this statement in the revised manuscript. 351 352 12) Fig. S2a: the AGO1x band appears to run much slower than the AGO1 band (in comparison 353 with Fig. S2b). How did the authors verify that this band corresponds to AGO1x? 354 355 The experiment in initial Fig. S2a gave us a hint of AGO1x expression, but we did not use it to 356 directly confirm the presence of AGO1x, for which we carried out additional analyses (Fig. S2,  357 current Fig EV1), perturbing the levels of AGO1x and verifying the response of the band 358 intensity with the specific AGO1x antibody. Regarding the direct comparison of the gels in Figs. 359 S2a and S2b (current Fig EV1A and B), we note that the gels were run for different times which 360 could account for the different resolution of the AGO/AGO1x region. 361 362 17th Mar 2020 1st Authors' Response to Reviewers 13) It would be reasonable to suggest that the TR extension reduces miRNA affinity (Fig. S8c)  363 by moving the C-terminal carboxyl group, which interacts with the miRNA 5' phosphate via a 364 lysine and ordered water molecule in crystal structures of canonical human Ago proteins. 365 366 We thank the reviewer for pointing this out. We have added a paragraph of discussion on the 367 AGO1 structure and possible consequences of a C-terminal extension for binding to miRNAs 368 and GW182 proteins. The AGO1/AGO1x protein interactomes further support the notion that 369 AGO1x-GW182 interaction is perturbed. 370 371 Referee #3: 372 373 In this manuscript Ghosh et al. report that a highly conserved TR isoform of AGO1, AGO1x, 374 localizes exclusively to the nucleus, in the vicinity of nucleoli, where it prevents accumulation of 375 double stranded RNAs by an unknown mechanism and thus, supposedly, mitigates the 376 interferon response and apoptosis. A similar role has also been proposed for Polyribonucleotide 377 Nucleotidyltransferase 1, a newly identified, specific binding partner of AGO1x. Finally, they also 378 showed that AGO1x is overrepresented in highly proliferative breast cancer cells and speculate 379 that targeting AGO1x could potentially represent a new direction for limiting tumor growth. 380 381 This is a clearly written, technically well-executed, fairly interesting story supported by numerous 382 experiments. However, it suffers from 1) insufficient detail (I've had an impression that it was 383 originally written for some "short-format" journal) forcing the reader to spend quite some time 384 trying to understand what this or that result means and/or how exactly this or that experiment 385 was executed, and 2) the lack of any mechanical insight into the presented observations. Below 386 I summarize all my concerns (in the order of their appearance in the text) that, in my opinion, 387 should be addressed to improve the quality of this study before its prospective acceptance. We apologize for the 'short-format' writing of our manuscript. We have thoroughly revised it to 398 provide more verbose descriptions of all results, and we hope that the reviewer finds this 399 version much more understandable. We have also incorporated line and page numbers. 400 401 2) Fig. S5a is placed out of order in the main text; in between S2 and S3. 402 403 We have now reordered our figures in the revised manuscript to accommodate this change. 404 405 3) "Endogenous AGO1x localizes ...." Fig. 1b  We have rectified the discrepancy in labelling. 408 409 4) Fig.3a; the AGO1x strip is smudgy compared to others shown in the work. Since it is a fairly 410 important result I suggest repeating it aiming for a nicer western blot image. 411 412 We thank the reviewer for the suggestion. The smudginess probably stems from the image 413 conversion from RGB to grayscale. We have adjusted the export parameters to make it more 414 clear. We found that their expression was indeed increased, both at the mRNA and at the protein 447 levels (Fig. 5a, b ). In addition, using a previously published method (Post et al , 2017), we 448 uncovered evidence for protein phosphorylation events downstream of interferon signaling, in 449 EIF2AK2 and ADAR ( Fig. 5c )." This is a typical example of the "go figure yourself" approach 450 We thank the reviewer for the suggestion to Include a sketch of IFN signaling pathway, which 455 we now show in Fig 5A. We have expanded the description of the mentioned data and clarified 456 that some of the proteins were not detected in the proteomic analysis, which generally has less 457 coverage than an RNA-seq based analysis of the transcriptome. 458 459 8) Figure 5d, k -signal quantifications would help. 460 461 We have done the quantifications, as requested by both reviewer #2 and #3. PNPT1 KD did not 462 change the overall relative distribution of the number of foci, only the size of some of the dsRNA 463 foci. We have therefore plotted the relative distribution of dsRNA foci in the edited and control 464 cell lines (Fig 5F). 465 466 9) The fact that the source of dsRNA lies predominantly, if not solely (based on Fig. 5e), in 467 rRNA is rather unexpected. The authors propose that: "...the increased demand for ribosome 468 biogenesis in rapidly dividing cells requires the readthrough AGO1x isoform for resolving dsRNA 469 structures in rRNAs and other molecules."; also rather unexpected, but why not. Since no other 470 explanation has been offered, either for the AGO1x role in dsRNA metabolism in the nucleus, or 471 for its contact with PNPT1, I suggest the authors to explore this idea in greater detail. If correct, 472 ribosome biogenesis should be affected in AGO1x null (as well as PNPT1 knocked-down) cells, 473 which can be easily tested with a bunch of well-established approaches. 474 475 We have carried out additional experiments, as suggested also by reviewer #2. Specifically, we 476 have quantified rRNA levels, the relative abundance of 40 and 60S ribosomal subunits and the 477 overall translation, all of which were reduced in the mutant lines (Appendix Fig S11). We think 478 that the results support a role of AGO1x on ribosome biogenesis. In addition, we showed that 479 AGO1x is co-IPed with PNPT1 in control cells (Fig 6). The statement to which the reviewer 480 refers is in the discussion of our manuscript, where we thought we can offer some speculation. 481 482 10) What could be reason for the observed differences between endogenously vs. ectopically 483 expressed AGO1x? Any idea? 484 485 We were puzzled by these differences as well. We have started to investigate a few 486 possibilities, in particular the effect of the amino acid that is incorporated at the stop codon, but 487 at this point we cannot provide a conclusive answer. Thank you for submitting your revised manuscript for our consideration. It has now been seen once more by the three original referees (see comments below). As you will see, the referees acknowledge that you have performed additional experiments and added clarifications, but are also not yet convinced that the conclusions are sufficiently supported by the data. Given the overall interest in the study, we would like to give you the opportunity to address the remaining issues in an exceptional second round of revision.
Referee #1 continues to have major concerns regarding the heterozygous cell lines and the levels of AGO1x. While the other referees agree that homozygous cell lines would be a more rigorous approach, they find that decisively showing a substantial depletion of AGO1x in the cell lines currently used would significantly contribute to addressing referee #1's concerns and supporting the conclusions drawn in Figures 4-6. Both referees are not convinced by the Western blot provided in Figure 3A, in particular by the reformatting of the blot rather than its repetition (ref #3 ad 4; ref#2 cross-commenting). To consider this study further for publication, this issue must be adequately addressed and convincing evidence for the reduced levels of AGO1x provided, as this affects all main conclusions of the following figures. Referee #2 suggests doing so by performing replicates (at least 3, preferably more) of the Western blots and quantitating AGO1x and AGO1 levels relative to the GAPDH control to provide clean and quantitative documentation of significant depletion of AGO1x (without effects on AGO1). Moreover, both referee #2 and #3 are not fully convinced of the small differences between wild type and the mutant cell lines in figure S11. The specific questions referee #3 raises should thus be addressed and the text carefully revised to more accurately interpret the results.
As indicated in the previous letter, it is normally EMBO Journal's policy to allow only one round of major revision, such that it is now crucial that you address the major concerns regarding Figure 3 fully and adequately respond to the other remaining issues raised. If you have any questions regarding this revision or would like to discuss how to proceed in more detail, please feel free to contact me. main concern: Fig. 3: the aut hors claim to have generat ed a genet ic mut at ion in the AGO1 gene that eliminat es the read-t hrough product (AGO1x) wit hout showing sufficient evidence that their cell lines indeed have specifically eliminat ed AGO1x. First , the aut hors add a supplement ary figure (Fig S4) that shows that the mut at ions affect different regions in the 3'UTR/read through region wit hout any clear indicat ion how the new read-t hrough prot ein looks like. Second, their seq track suggest and they also direct ly st at e that their mut at ions are het erozygous, meaning that the second AGO1 allele is st ill wild type and can produce the read-t hrough product AGO1X. Third, the aut hors st at e in the rebut tal that they at tempted the rescue of a pot ent ial AGO1x null phenot ype but failed to do so. Their own dat a and words raise addit ional concerns not only about the experiment s themselves but also about their general int erpret at ion of dat a. Their model remains unsupport ed by the present ed dat a. Their dat a remain insufficient ly cont rolled and explained. Their observat ions do not support their int erpret at ions. I do not recommend this manuscript for publicat ion.
Referee #2: The aut hors have addressed my previous comment s wit h thought and care. I congrat ulat e them on an excit ing and compelling st udy.

Referee #3:
This manuscript has great ly improved, no doubt about that. However, I think that a few more issues shown below have to be resolved before it can be published.
Ad 4) I am a bit worried by this response -a different image processing showing now what should be seen? It is relat ively very easy to repeat this experiment to convince the reader that everyt hing has been done properly, isn't it ? I do not doubt it , I am just somewhat surprised.
Ad 7) Great ly improved! Here, however, Fig. 5D is technically wrongly execut ed. In addit ion to what you show now, you must also show the eIF2alpha prot ein levels and normalize your "alpha-P" signal to them to be 100% sure about your claims. 9) Thank you for your effort but I am not convinced by these dat a at all. S11A -B: typically, a defect in ribo biogenesis result s in increased levels of precursors and decreased levels of mat ure forms of rRNAs. In your case everyt hing is down, which would suggest that the biogenesis pre se is okay, but the rDNA transcription is significant ly mit igat ed. Can you test it ? S11C -D: Perfect ly fit s wit h the above; visually, I see no significant differences what soever -how many times did you repeat e this experiment (I see no error bars in the panel D)? At least three times would be required for such a marginal difference, if any. Also, the wt (blue) tracing looks weird and prevents you from making any conclusions, in my opinion. Finally, no CHX should be added to the buffer and no Mg ions as well (this is much better than supplying the buffer with excessive amounts of EDTA). S11F: Same as above, multiple repetitions should be carried out and the Polysome to Monosome (P/M) ratios with SD should be calculated before any conclusions can be drawn. There could be a slight P/M decrease in your mutants, I agree -perhaps due to an overall reduction in the production of ribosomes but not a defect in their biogenesis, you just have to convincingly demonstrate it.
Referee #1: 1 2 main concern: Fig. 3: the authors claim to have generated a genetic mutation in the AGO1 gene that 3 eliminates the read-through product (AGO1x) without showing sufficient evidence that their cell lines 4 indeed have specifically eliminated AGO1x. First, the authors add a supplementary figure (Fig S4) that 5 shows that the mutations affect different regions in the 3'UTR/read through region without any clear 6 indication how the new read-through protein looks like. Second, their seq track suggest and they also 7 directly state that their mutations are heterozygous, meaning that the second AGO1 allele is still wild type 8 and can produce the read-through product AGO1X. Third, the authors state in the rebuttal that they 9 attempted the rescue of a potential AGO1x null phenotype but failed to do so. Their own data and words 10 raise additional concerns not only about the experiments themselves but also about their general 11 interpretation of data. Their model remains unsupported by the presented data. Their data remain 12 insufficiently controlled and explained. Their observations do not support their interpretations. I do not 13 recommend this manuscript for publication. 14 15 We did not claim that we have *eliminated* AGO1x, but rather that we depleted it. both by sequencing the DNA loci as well as by aligning the RNA-seq reads from these lines to the 20 corresponding locus (Appendix Fig. S4A,C). The depletion of the protein was apparent both in the western 21 blots (Fig. 3A lower panel) and in imaging data (Fig. 5G). Nevertheless, to hopefully dispel any remaining 22 concerns, we have carried out another round of western blotting experiments, with three replicates for 23 each of the control, W1A and W6A mutant MDA-MB-231 lines. The results re-confirm the depletion of 24 AGO1x in the mutant lines with respect to control, while the total AGO1 levels were not changed 25 substantially, likely because the abundance of AGO1 is higher than that of AGO1x. These results are shown 26 in revised Fig. 3A, lower panel. We have further repeated the western blots in longer running gels (where 27 lower MW proteins run out of the gel), as done by (Singh et al, 2019). Here we found that the faint, higher 28 MW band identified by the AGO1 antibody in control cells is reduced in intensity in the mutant lines. From 29 these better resolved gels we have again quantified the canonical AGO1 band and by probing the gels 30 with the AGO1x antibody we specifically quantified AGO1x. The results confirm the depletion of AGO1x 31 but not AGO1 in the mutant lines relative to control (Appendix Fig. S4E, quantification in S4F).

33
Regarding the rescue, we noted that the reason for not pursuing assays with the construct for ectopically-34 expressing the tagged protein was that this protein had a different localization than the endogenous 35 AGO1x.

37
We did not further attempt to determine whether the mutant lines express yet another version of a 38 readthrough protein. It is relatively very easy to repeat this experiment to convince the reader that everything has been done 60 properly, isn't it? I do not doubt it, I am just somewhat surprised. 61 62 In our initial revision, we responded to the reviewer's comment about the smudginess of the band and 63 did not realize that the reviewer expected us to redo the experiment and replace the figure. Indeed, this 64 is not a problem. We have now carried out western blot analysis of AGO1 and AGO1x in triplicate, using 65 the respective antibodies. We have incorporated these new results in our revised Fig. 3A. As the AGO1 66 antibody identified a single band in these blots, we also repeated the experiment running the gel longer 67 (as shown earlier in our manuscript and also done in the study of (Singh et al, 2019); this caused low MW 68 proteins to run out of the gel). We probed these gels first with AGO1 antibody, which revealed not only 69 the band corresponding to the canonical AGO1, but also the faint, higher MW band, present in the control 70 line and reduced in intensity in the mutant lines. By further probing the blot with AGO1x antibody, we 71 confirmed the AGO1x depletion. We have inserted these results in the Appendix Fig S4E-F. We hope that 72 all of these blots dispel any doubt that AGO1x is depleted in the mutant lines.

74
Ad 7) Greatly improved! Here, however, Fig. 5D is technically wrongly executed. In addition to what you 75 show now, you must also show the eIF2alpha protein levels and normalize your "alpha-P" signal to them 76 to be 100% sure about your claims. 77 78 Our proteomics data showed that eIF2α did not change significantly in the mutant lines, and if anything, 79 the small (< 20%) change was in the direction of decrease in mutant compared to control lines (log2 ratio 80 to control was -0.28 in the W1A line and -0.16 in the W6A line). In contrast, the WBs of p-eIF2α indicated 81 that this form was *increased* in the mutant lines relative to GAPDH control. However, to address the 82 reviewer's comment, we have augmented Fig. 5D with the results of probing the same cell lysates for eIF2α. 83 8th Jun 2020 2nd Authors' Response to Reviewers

A-Figures 1. Data
The data shown in figures should satisfy the following conditions: the data were obtained and processed according to the field's best practice and are presented to reflect the results of the experiments in an accurate and unbiased manner. figure panels include only data points, measurements or observations that can be compared to each other in a scientifically meaningful way. graphs include clearly labeled error bars for independent experiments and sample sizes. Unless justified, error bars should not be shown for technical replicates. if n< 5, the individual data points from each experiment should be plotted and any statistical test employed should be justified Source Data should be included to report the data underlying graphs. Please follow the guidelines set out in the author ship guidelines on Data Presentation.

Captions
Each figure caption should contain the following information, for each panel where they are relevant: a specification of the experimental system investigated (eg cell line, species name). the assay(s) and method(s) used to carry out the reported observations and measurements an explicit mention of the biological and chemical entity(ies) that are being measured. an explicit mention of the biological and chemical entity(ies) that are altered/varied/perturbed in a controlled manner. the exact sample size (n) for each experimental group/condition, given as a number, not a range; a description of the sample collection allowing the reader to understand whether the samples represent technical or biological replicates (including how many animals, litters, cultures, etc.). a statement of how many times the experiment shown was independently replicated in the laboratory. definitions of statistical methods and measures: Any descriptions too long for the figure legend should be included in the methods section and/or with the source data.
In the pink boxes below, please ensure that the answers to the following questions are reported in the manuscript itself. Every question should be answered. If the question is not relevant to your research, please write NA (non applicable). We encourage you to include a specific subsection in the methods section for statistics, reagents, animal models and human subjects.

B-Statistics and general methods
Please fill out these boxes ê (Do not worry if you cannot see all your text once you press return) No method was used to determine sample size.

N.A
For the cell growth and migration assays performed on the xCELLigence system up to 2 biological replicates were discarded when diverging significantly from the other more consistent replicates (at least 3).