Cis- and trans-regulations of pre-mRNA splicing by RNA editing enzymes influence cancer development

RNA editing and splicing are the two major processes that dynamically regulate human transcriptome diversity. Despite growing evidence of crosstalk between RNA editing enzymes (mainly ADAR1) and splicing machineries, detailed mechanistic explanations and their biological importance in diseases, such as cancer are still lacking. Herein, we identify approximately a hundred high-confidence splicing events altered by ADAR1 and/or ADAR2, and ADAR1 or ADAR2 protein can regulate cassette exons in both directions. We unravel a binding tendency of ADARs to dsRNAs that involves GA-rich sequences for editing and splicing regulation. ADAR1 edits an intronic splicing silencer, leading to recruitment of SRSF7 and repression of exon inclusion. We also present a mechanism through which ADAR2 binds to dsRNA formed between GA-rich sequences and polypyrimidine (Py)-tract and precludes access of U2AF65 to 3′ splice site. Furthermore, we find these ADARs-regulated splicing changes per se influence tumorigenesis, not merely byproducts of ADARs editing and binding.

1. The splicing analysis of high-throughput sequencing data uses a low cut-off of p-value 0.05. If the authors account for multiple testing in the large dataset and use a p-adjusted cut-off, how many transcripts exhibit splicing changes that are affected by ADAR1 and ADAR2? 2. In nearly every figure, the authors provide semi-quantitative RT-PCR analysis of transcripts. For most, representative images from one replicate are located above the quantitation of multiple replicates. However, in many cases, the representative image is difficult to ascertain changes, for example Figure 2b the effect of ADAR2 on CCDC15 in EC109 cells. For a few targets/experiments (such as those in Figure 2), can the authors provide all three representative images of the RT-PCR experiments and/or comment on the variability? This is particularly important as the final figure connecting splicing changes in Normal and Tumor tissues does not contain quantitation/only representative images.
In addition, for the significance of the semi-quantitivate RT-PCR results, it is unclear why the authors used paired t-tests as opposed to ANOVA or other multiple variable tests. While some effects may have minor significance (due to technical repeats as opposed to biological consequences), attention to those small effects detracts from the main conclusions of the paper. 3. The results regarding ADAR2 binding to the GA motif in RELL2 and inhibits U2AF65 binding are not convincing. The RNA binding experiments with ADAR2 and U2AF65 alone as well as together ( Figure  5g) seem to suggest that ADAR2 binding to the RELL2 Py3 RNA is enhanced in the presence of U2AF65. Can the authors provide relative affinities (using a titration curve) for ADAR2 binding to this sequence in the presence and absence of U2AF65? 2. Figure 2 has parallel experiments performed in HEK293T and EC109 cells, except for the dosedependent response. Is there a reason the authors did not include that cell line? Also, the statistical analysis of the dose-dependent response is lacking. 3. For the MEME analysis ( Figure 6), can the authors speculate on what it means that 3 of the 4 most enriched motifs in the ADAR1/2 cassette exons do not contain adenosine?
Minor points with text: 1. There are a number of typographical errors in the discussion (bottom page 13, page 14-omissions of the word "the") and introduction (bottom of page 3-majority of studies sentence).

Reviewer #2 (Remarks to the Author):
This study addresses the interplay between RNA editing and splicing in the context of cancer. Authors started by modulating the expression of ADAR enzymes in an esophageal squamous carcinoma cell line. Overexpression or silencing ADAR 1 and ADAR2 had global effects on both A-to-I editing and alternative splicing (Fig. 1). Next, authors focused on RNA processing of transcripts from the CCDC15 and RELL2 genes for further analysis. Making use of ADAR mutants devoid of either enzymatic activity or dsRNA-binding capability, the results presented in Fig. 2 elegantly show that both editingdependent and -independent mechanisms are involved in ADAR-mediated splicing regulation. Potential editing sites were identified in CCDC15 exon 9 and flanking introns and a minigene was mutated to mimic 100% editing at each site; these mutations altered exon 9 inclusion by a mechanism that likely involves binding of SRSF7 to the edited sequence (Fig. 3). Deletion and RNA binding analysis further suggest binding of ADAR enzymes to an intronic dsRNA spaning sequences in intron 8 and intron 9 (Fig.4). In the case of RELL2, ADAR2 binds to a dsRNA that forms at the Pytract, thereby competing for the binding of U2AF65, which is essential for spliceosome assembly (Fig. 5). Based on the results obtained with CCDC15 and RELL2 reporter genes, authors apply bioinformatics tools to predict ADAR-binding motifs involved in ADAR-mediated splicing regulation (Fig. 6). Finally, the potential biological consequences of ADAR-mediated splicing regulation are investigated, particularly in the context of cancer (Figs. 7 and 8). Overall, this is a timely and sound study; the manuscript is very well presented, and conclusions are supported by convincing experimental results. I have no criticisms. I have, however, some suggestions that authors may find useful to address on a revised version. In this manuscript, Tang et al. investigated the global effects of ADAR1/2 on RNA splicing, and discovered more than one hundred high-confident splicing events regulated by ADAR1 and/or ADAR2. They then chose two affected splicing events in CCDC15 and RELL2 as examples for further mechanistic and functional studies. The authors revealed that ADAR1/2 tend to bind to GA-rich dsRNAs for RNA editing and splicing regulation. In particular, their bindings to dsRNAs affect the recruitment of splicing factors such as SRSF7 and U2AF65, leading to repression of exon inclusion. Finally, they also showed that ADAR-affected splicing events in CCDC15 and RELL2 have significant impact on in vitro colony-formation of tumor cells and also in vivo tumorigenesis. Overall, this is an interesting and timely study, which provide compelling evidence showing that ADAR affect RNA splicing through their RNA editing dependent and independent activities, which in turn affect tumorigenesis. This manuscript is well written and ease to understand.
Nonetheless, I have a few questions/comments below that should be addressed by the authors: 1) There are discrepancies in the number of bands and sizes of ADAR1 in WB data. Figure 1A shows two bands for ADAR1, with one band ~130kD and another below 100kD. However, there is a single band for ADAR1 at ~130kD in Figure S2A and a single band at 100kD in Figure S2B. Could this be due to cell type differences and why? Similarly, ADAR2 overexpression sometimes produces one band ( Fig  1A and Fig S2A for EC109) or two bands (Fig S2A 293T, Fig S2B). The authors should also provide uncropped WB's in the supplementary figures.
2) In Figure 3D, the authors showed that mutation of site 2 from A to G facilitates exon 9 exclusion, and even more so when SRSF7 is overexpressed. This implied that SRSF7 binding is affected by ADAR1 editing. Can the authors validate whether RNA binding of ADAR1/2 affects SRSF7 binding? Do the mutations introduced in the intron 9 region (Figure 4), particularly del5, affect SRSF7 binding to RNA?
3) Since both ADAR1 and ADAR2 influence CCDC15 splicing in the same direction, can the effects of ADAR1 depletion be compensated by ADAR2 and vice versa? Does double knockdown of both ADARs have an additive effect? 4) From the text, 23 ADAR1/2 high and 10 ADAR1/2 normal/low patient samples were tested. However in Figures 8a andb, data is presented from only 6 samples each for ADAR1/2 high and ADAR1/2 normal/low patient samples. An additional graph summarizing the data from all the patient samples would be helpful (e.g. a bar graph of PSI of CCDC15 and RELL2 comparing ADAR1/2 high and normal/low groups). This way, statistical analyses comparing the PSI of different patient groups can be done, although the small sample size might limit this type of analysis.

5) WB
showing CCDC15 exon9 knockdown should be done to show that the effect of knockdown is specific for that particular isoform at the protein level, and that other isoforms are unaffected. In addition, these experiments (foci, soft agar, mouse experiments) should also include overexpression of CCDC15 containing exon9. 6) Although unrelated to the main focus of the manuscript, it would be interesting to discuss what makes CCDC15 exon 9 isoform oncogenic in esophageal cancer. A brief discussion in the Discussion section would be helpful.

Editorial Board:
Your manuscript entitled "Cis-and trans-regulation of pre-mRNA splicing by RNA editing enzymes: influencing cancer development" has now been seen by 3 referees. You will see from their comments below that while they find your work of interest, some important points are raised. We are interested in the possibility of publishing your study in Nature Communications, but would like to consider your response to these concerns in the form of a revised manuscript before we make a final decision on publication. We consider it particularly important that the revised manuscript provide all three images of the RT-PCR experiments (Figs. [2][3][4][5][6][7] and uncropped western blots (Fig.  1, Supplementary fig 2) in a source data file.
-We have included all required images and uncropped blots in the revised source data file. Please also see our point-by-point response to the reviewers' comments and suggestions below.

Reviewer #1:
This manuscript seeks to dissect ADAR effects on splicing and connect those changes to cancer development. As the author point out, effects of ADAR on splicing have been documented in many systems, but mechanistic details of how ADARs influence splicing are relatively unknown. Herein, the authors provide extensive dissection of the impact of ADAR1 and ADAR2 RNA binding and editing on splicing of two transcripts, CCDC15 and RELL2. In addition, they provide evidence for these two transcripts influencing oncogenic properties of cells. The manuscript needs improved in regard to discussion of the technical details of the experiments performed (for example, it is only mentioned in the methods that the RIP experiment in Figure 4f is done on minigenes and not endogenous transcripts), the results from each experiment and avoidance/explanation of specific splicing terms (ISS).
-We thank the reviewer for pointing this out. We have carefully checked through all the figure legends and provided more details for the experiments performed and added explanations of ISS.
Points that would support the conclusions in the manuscript are listed below: 1. The splicing analysis of high-throughput sequencing data uses a low cut-off of p-value 0.05. If the authors account for multiple testing in the large dataset and use a p-adjusted cut-off, how many transcripts exhibit splicing changes that are affected by ADAR1 and ADAR2?
-Thank the reviewer for raising this question. We reanalysed our data by setting a FDR-adjusted pvalue cut-off of 0.05, 0.1 or 0.2 and found that the number of ADAR1/2-regulated splicing events dropped (please see the tables below). For the following two reasons, we decided to set an adjust pvalue cut-off of 0.2 (FDR<0.2) and amended main text and Fig.1b-c accordingly. 1) If we set FDR<0.1 or FDR<0.05, most of our validated cassette exon events was filtered out, such as CCDC15 exon 9, ZNF517 exon 3, RELL2 exon 3 and AKAP9 exon 19. 2) As discussed in Handbook of Biological Statistics 1 , low FDR cut-off may lead to high rate of false negative. We hope to convince the reviewer that in our study, we have set stringent requirements, including 1) total junction reads ≥ 15, 2) splicing index change (|ΔSI|) ≥ 10% with a FDR value <0.2; and 3) particularly SI change from both knockdown and overexpression experiments must be in opposite direction. All six randomly selected targets could be experimentally validated in both EC109 cells and HEK293T cells (Fig. 1d,e and Supplementary Fig. 1b). Tang  -The reviewer has commented "in many cases, the representative image is difficult to ascertain changes". In this study, we did observe small batch-to-batch variations in the baseline PSI values of EV or Scr control cells, which was most likely due to the difference in cell conditions arising from different passage number, cell confluency or transfection efficiency and so on. That was also why for each batch of RT-PCR experiments, we included at least 3 biological replicates, in order to make solid conclusions.
-The reasons why we used the paired t-test include: 1) we intended to compare each treatment to its own empty vector (EV) control (e.g WT vs EV, DeAD vs EV, and EAA vs EV) for each biological replicate; 2) as abovementioned, the baseline PSI values of EV control cells may be varied; and 3) ANOVA and other multiple variable tests are used to compare more than two groups and show significance as long as one of the comparing groups shows the difference. Therefore, as we intended to show whether there was a repressive effect of WT, DeAD or EAA on exon inclusion, we think the paired t-test is more suitable than ANOVA.
3. The results regarding ADAR2 binding to the GA motif in RELL2 and inhibits U2AF65 binding are not convincing. The RNA binding experiments with ADAR2 and U2AF65 alone as well as together (Figure 5g) seem to suggest that ADAR2 binding to the RELL2 Py3 RNA is enhanced in Tang and  the presence of U2AF65. Can the authors provide relative affinities (using a titration curve) for ADAR2 binding to this sequence in the presence and absence of U2AF65? -We thank the reviewer for making valuable comments. We first repeated the original UV-crosslink experiment twice and confirmed there was no obvious enhancement of ADAR2 binding to RELL2 in the presence of U2AF65 (revised Fig. 5g). Next, we conducted the UV-crosslink assay with increasing amount of U2AF65 protein and we also did not observe a dose-dependent increase in ADAR2 binding to RELL2 (Supplementary Fig. 4h).
The authors also state that ADAR2 binding to the RELL2 Py3 RNA was diminished by the exonic GA double mutations, but the gel shift provided (Supplemental Figure 4f) shows a large amount of binding. Can the authors provide relative affinities (using a titration curve) for ADAR2 binding to the mutant and wild-type RELL2 Py3 RNA? -Thank the reviewer for pointing this out. We performed REMSA with increasing amount of ADAR2 protein for both wild-type and mutant RELL2 probes (revised Fig. 5e). Our new REMSA data showed that ADAR2 protein binds more strongly to the wild-type probe than the mutant probe by measuring the intensity of RNA-protein shift band over the total intensity (sum of the intensities of free probe and shift band) (revised Fig. 5e). The REMSA data suggests that binding of ADAR2 to RELL2 was reduced by double mutations. We have replaced the original Fig. 5e with this new data including both gel image and band intensity measurements.

The impacts of lack of the CCDC15 exon 9 included transcript on soft agar, clonogenic growth and tumor volume are impactful. However, it is unclear whether ADAR1 has a similar function? Or ADAR1 status contributes to the effects of CCDC15 exon 9 included? Similar questions for ADAR2 and RELL2 exon 3 included are also important to address.
-We thank the reviewer for raising these questions. To further confirm the functional impacts of CCDC15 exon 9 inclusion, we have included new data showing the functional role of CCDC15 exon 9-included (full-length) isoform in Fig. 8f-h. Consistently, overexpression of CCDC15 exon 9-included isoform enhanced the tumorigenicity of EC109 cells.
-Functions of ADAR1 and ADAR2 in ESCC (esophageal squamous cell carcinoma) have been extensively studied previously 2, 3 . ADAR1 generally acts as an oncogene in ESCC progression; while ADAR2 acts as a tumor suppressor. In this study, our main focus was to investigate whether and how ADAR1 and ADAR2 regulate cancer-associated splicing events, rather than study if ADARs are the main contributors to tumorigenic or tumor suppressive effect of CCDC15-ex9included and RELL2-ex3-included isoforms. Moreover, ADAR1 and ADAR2 are involved in many cellular processes such as protein recoding and miRNA biogenesis. Modulation of ADAR1 or ADAR2 expression level would affect multiple downstream targets. In addition, as we discussed in the manuscript, splicing of CCDC15 and RELL2 can be regulated by other splicing factors and their splicing patterns in tumors may be a combinational outcome of different splicing regulators.
-In this study, we showed that both ADAR1 and ADAR2 repress CCDC15 exon 9 inclusion. As shown in Supplementary Fig. 5b, oncogenic CCDC15-ex9-included isoform was upregulated in 60% (20/33) of primary ESCC tumor samples, suggesting that loss of ADAR2 may contribute more to the increased CCDC15 exon 9 inclusion in tumors than overexpression of ADAR1. Therefore, we did not perform experiments to study whether that ADAR1 status contributes to the oncogenic effect of CCDC15 exon 9 included isoform.
-For RELL2, we found that ADAR2 represses RELL2 exon 3 inclusion in EC109 cells. Although ADAR2 is thought to be a tumor suppressor and downregulated in ESCC, we did observe that 30% (10/33) of ESCC patients showed higher expression of ADAR2 in tumors (Supplementary Fig.  5b). For this group of patients, ADAR2 may contribute to oncogenic effect of RELL2 ex3 exclusion. Tang and Shen et al. NCOMMS-19-25121-T, Page 4 For the remaining ESCC patients, there may be other splicing factors contributing to the tumor suppressive effect of RELL2 ex3 exclusion, which remains for our further investigation.

Minor points with experiments/results:
1. The western blot analysis of ADAR1 in Figure 1a seems to be out of alignment with the known sizes of ADAR1 p110 and ADAR1 p150.
-Thank you for pointing out the misalignment. We have corrected the marker labelling. Figure 2 has parallel experiments performed in HEK293T and EC109 cells, except for the dosedependent response. Is there a reason the authors did not include that cell line? Also, the statistical analysis of the dose-dependent response is lacking.

2.
-We have applied the linear regression analysis to show that ADAR proteins could cause significant CCDC15 splicing change in HEK293T dose-dependently (Fig. 2b).
-We were facing difficulties in obtaining a good dose-dependent overexpression of ADAR1 and ADAR2 proteins in EC109 cells, we thus observed a saturated repressive effect on CCDC15 exon inclusion upon ADAR1/2 overexpression (please see the figure below). Therefore, we did not include these data in the manuscript.

For the MEME analysis (Figure 6), can the authors speculate on what it means that 3 of the 4 most enriched motifs in the ADAR1/2 cassette exons do not contain adenosine?
-We thank the reviewer for raising this question. The frequency of adenosine in our input sequences is about 25% and therefore lack of adenosine enrichment within the four discovered motifs is not due to the depletion of adenosine in our inputs. Adenosine is not completely absent from the other three motif but rather it is less frequent and conserved in these three motifs. One possibility for this observation could be that these cassette exons are regulated by splicing factors that recognize pyrimidine-rich sequences and therefore we could observe relatively pyrimidine-rich motifs in our analysis. Another possibility is that adenosine is relatively less conserved in the length of motif we set because we could see frequent conserved adenosine nucleotide in longer motif sequences. a, RT-PCR analysis of CCDC15 exon 9 splicing in EC109 cells overexpressed with dose-increase ADAR1 or ADAR2. b, Western blot analysis of ADAR1 and ADAR2 expression in EC109 cells.

A Western blot showing downregulation of RELL2 protein would unequivocally demonstrate the physiological impact of exon 3 skipping.
-As suggested by the reviewer, we have tried different antibodies against RELL2 including the one used in a previous publication (Wang et al. Eur J Pharmacol 2019) for western blot analysis, but we could not obtain a good WB result. Due to the poor antibody quality and low endogenous expression of RELL2 in EC109 cells, we could only detect a weak RELL2 band with high background signals. As shown in the figure below, there was a subtle decrease in RELL2 protein upon ADAR2 overexpression. In addition, ADAR2 overexpression could only repress RELL2 exon 3 inclusion level by 20% at maximum (Fig. 2c). Therefore, it is challenging to observe a pronounced reduction in RELL2 protein expression by WB analysis. Due to the poor quality of data, we decided not to include this data in the manuscript. Nonetheless, we could include the images in the paper if requested.

A recent study proposed that RELL2 acts as a tumor suppressor (Wang et al Eur J Pharmacol 2019)
. This work should be cited and discussed, as it re-enforces the findings reported here.
-Thank you for your suggestion. We have added a short discussion into the revised manuscript.
Western blot analysis of RELL2 expression. 1) There are discrepancies in the number of bands and sizes of ADAR1 in WB data. Figure 1A shows two bands for ADAR1, with one band ~130kD and another below 100kD. However, there is a single band for ADAR1 at ~130kD in Figure S2A and a single band at 100kD in Figure S2B. Could this be due to cell type differences and why? Similarly, ADAR2 overexpression sometimes produces one band (Fig 1A and Fig S2A for EC109) or two bands (Fig S2A 293T, Fig S2B). The authors should also provide uncropped WB's in the supplementary figures.
-We thank the reviewer for raising this question and apologize that the discrepancies of p110 band among Fig. 1a, Supplementary Fig. 2a and 2b were due to misalignment when we labeled the images and we have corrected the marker labeling. -There are two ADAR1 isoforms, p110 and p150, and the major isoform is the p110 isoform at ~110kD. The p150 band is not always visible as its expression is induced by interferon 4, 5 and therefore can be affected by cell conditions. Similarly, ADAR2 has different isoforms, mainly due to alternative splicing 6,7 , and different ADAR2 isoforms can be differentially expressed under different cell conditions. In addition, when we prolonged the exposure time for western blot, we were able to observe two bands of ADAR2. -As requested by the reviewer, we have provided uncropped WB images in the Source Data file. Figure 3D, the authors showed that mutation of site 2 from A to G facilitates exon 9 exclusion, and even more so when SRSF7 is overexpressed. This implied that SRSF7 binding is affected by ADAR1 editing. Can the authors validate whether RNA binding of ADAR1/2 affects SRSF7 binding? Do the mutations introduced in the intron 9 region (Figure 4), particularly del5, affect SRSF7 binding to RNA? -Thank the reviewer for making this valuable suggestion. We have performed REMSA to show that binding of ADAR1 to the CCDC15 In8-9 dsRNA probe did not affect binding of SRSF7 and included this new data as Fig. 4g. In addition, we have conducted RIP assay to check if SRSF7 binds to the mutant del5 minigene, in order to address the reviewer's second question. We could observe a similar binding affinity of SRSF7 to both wild-type and In9 del5 mutant ADAR1 (please see the figure below). We feel this new REMSA data is sufficient to support that ADAR1 binding does not affect the binding of SRSF7 to CCDC15 transcript, so we only included the REMSA data into the revised manuscript. Nonetheless, we could include the RIP data into the manuscript if requested. a, RT-PCR analysis of CCDC15 exon 9 splicing in HEK293T cells transfected with ADAR1 or/and ADAR2 shRNAs. b, QPCR analysis of ADAR1 and ADAR2 expressions in indicated cells.

2) In
Tang and Shen et al. NCOMMS-19-25121-T, Page 10 patient samples would be helpful (e.g. a bar graph of PSI of CCDC15 and RELL2 comparing ADAR1/2 high and normal/low groups). This way, statistical analyses comparing the PSI of different patient groups can be done, although the small sample size might limit this type of analysis.
-As suggested by the reviewer, we have provided new figure showing the ΔPSI of CCDC15 and RELL2 cassette exons for all patient samples (Supplementary Fig. 5b, c). However, due to the small sample size, there was no statistical significance between different groups by using Student's t-test.

5)
WB showing CCDC15 exon9 knockdown should be done to show that the effect of knockdown is specific for that particular isoform at the protein level, and that other isoforms are unaffected. In addition, these experiments (foci, soft agar, mouse experiments) should also include overexpression of CCDC15 containing exon9.
-As suggested by the reviewer, we have provided a WB image showing the knockdown of CCDC15 at protein level (Supplementary Fig. 5f). Although we could only observe one single band at the predicted molecular weight, the CCDC15 antibody seemed not to work very well to detect endogenous CCDC15 protein (Supplementary Fig. 5f).
-There is only 4.85kD difference in the protein size between two isoforms (exon 9-included andskipped), it is not likely to differentiate them by WB. Therefore, we have used qRT-PCR to show the expressions of two isoforms in CCDC15 exon 9 knockdown cells in Supplementary Fig. 5d.
-We also included the functional role of CCDC15 exon9-included isoform. As shown in Fig. 8f-h, both in vitro and in vivo tumorigenicity assays supported that overexpression of CCDC15 exon9included isoform in EC109 cells promotes tumorigenesis. 6) Although unrelated to the main focus of the manuscript, it would be interesting to discuss what makes CCDC15 exon 9 isoform oncogenic in esophageal cancer. A brief discussion in the Discussion section would be helpful.
-Thank the reviewer for the suggestion. We have added a short discussion regarding the possible functional disruption by exon 9 skipping in the "Discussion" (in red font).

7)
Addition of a schematic summarizing the findings in the manuscript can be added to help the reader.
-Thank the reviewer for the suggestion. We have provided a schematic diagram summarizing our findings in Fig. 9.