Intragenic recruitment of NF-κB drives alternative splicing modifications upon activation by the viral oncogene TAX of HTLV-1

The chronic NF-κB activation in inflammation and cancer has long been linked to persistent activation of NF-κB responsive gene promoters. However, NF-κB factors such as RELA also massively bind to gene bodies. Here, we demonstrate that the recruitment of RELA to intragenic regions regulates alternative splicing upon activation of NF-κB by the viral oncogene TAX of HTLV-1. Integrative analysis of RNA splicing and chromatin occupancy, combined with chromatin tethering assays, demonstrate that DNA-bound RELA interacts with and recruits the splicing regulator DDX17 in a NF-kB activation-dependent manner, leading to alternative splicing of target exons thanks to DDX17 RNA helicase activity. This NF-kB/DDX17 axis accounts for a major part of the TAX-induced alternative splicing landscape that mainly affects genes involved in oncogenic pathways. Collectively, our results demonstrate a physical and direct involvement of NF-κB in alternative splicing regulation, which significantly revisits our knowledge of HTLV-1 pathogenesis and other NF-κB-related diseases.


Introduction 1
The Human T-cell leukemia virus (HTLV-1) is the etiologic agent of Adult T-cell 2 Leukemia/Lymphoma (ATLL) 1 , an aggressive CD4+ T-cell malignancy, and of various 3 inflammatory diseases including the HTLV-1-associated myelopathy/tropical spastic 4 paraparesis (HAM/TSP) 2 . It has long been established that changes in gene expression 5 level participate to the persistent clonal expansion of HTLV-infected CD4+ and CD8+ T-6 cells, leading ultimately to HTLV-1 associated diseases 3 . We recently reported that 7 alternative splicing events help to discriminate between ATLL cells, untransformed 8 infected cells and their uninfected counterparts derived from patients 4 . Alternative 9 splicing of pre-messenger RNAs is a cotranscriptional processing step that controls both 10 the transcriptome and proteome diversity and governs in turn cell fate. Its regulation 11 relies on a complex and still incompletely understood interplay between splicing factors, 12 chromatin regulators and transcription factors 5,6 . In this setting, the molecular 13 mechanisms of HTLV-1-induced splicing modifications and whether these effects rely on 14 an interplay between transcription and splicing is not known . 15 16 TAX is an HTLV-1-encoded protein that regulates viral and cellular gene transcription. 17 TAX also alters host signaling pathways that sustain cell proliferation and lead ultimately 18 to cell immortalization 7 . The Nuclear factors kB (NF-kB) signaling pathway is the most 19 critical target of TAX for cell transformation 8 . The NF-kB transcription factors (RELA,p50,20 c-Rel, RelB, and p52) govern immune functions, cell differentiation and proliferation 9 . NF-21 kB activation involves the degradation of IκB that sequesters NF-kB factors in the 22 cytoplasmic compartment, leading to NF-kB nuclear translocation and binding of NF-kB 23 dimers (e.g., RELA:p50 for the most abundant) to their target promoters 10,11 . TAX induces 24 IKK phosphorylation and IκB degradation, leading to persistent nuclear translocation of 25 NF-kB 12,13 . In addition, TAX interacts with nuclear NF-kB factors and enhances their 26 effects on transcription 14,15 . 27 28 Interestingly, genome-wide analyses of NF-kB distribution have unveiled that the vast 29 majority of RELA peaks is outside promoter regions and can be localized in introns and 30 exons [16][17][18][19] . Some of those promoter-distant RELA binding sites correspond to cis-31 regulatory transcriptional elements 20,21 but globally, there is a weak correlation between 32 the binding of RELA to genes and regulation of their steady-state expression 17,18 . These 33 Results 1 TAX induces alternative splicing modifications irrespectively of its effects on 2 transcription. 3

RNA-seq analyses were performed on 293T-LTR-GFP cells transiently transfected with a 4
TAX expression vector. TAX-induced changes in gene expression level and in alternative 5 splicing were identified and annotated as previously described 22,23 (Table S1). As shown 6 in Figure 1A, the ectopic expression of TAX affected the splicing and gene expression 7 levels of 939 and 523 genes, respectively. A total of 1108 alternative splicing events were 8 predicted including 710 exon skipping events ( Figure 1B). A minority of genes (3.5%, 9 33/939) was altered at both the expression and splicing levels, indicating that TAX largely 10 affects alternative splicing independently of its transcriptional activity. A subset of 11 splicing events was validated by RT-PCR ( Figure 1C). We took advantage of RNA-seq 12 datasets (EGAS00001001296 24 ) for assessing whether TAX-related alternative splicing 13 could pertain to asymptomatic carriers (AC) and ATLL patients. Overall, 542 (48%) TAX-14 induced splicing modifications were detected at least once across 55 clinical samples 15 (Table S1). Hierarchical clustering of these exons based on their inclusion rate (PSI) 16 identified TAX-regulated exons that discriminate AC and ATLL samples from uninfected 17 CD4+ T-cells ( Figure 1D). We furthermore confirmed that TAX promotes splicing events 18 previously detected in HTLV-1 infected individuals, including AASS, CASK,RFX2 and CD44 19 4,25 . We firmly established that the expression of the splicing variant CD44v10 previously 20 identified in HAM/TSP patients 25 fully relies on TAX expression ( Figure 1C and Figures 21 S1A-C). Altogether, these results uncovered a large number of splicing modifications upon 22 TAX expression that for a part coincide with alternative splicing events observed in HTLV-23 1 patients. 24 25 Gene ontology analysis of quantitatively altered genes revealed several signaling 26 pathways that are well described in TAX expressing cells, including NF-kB, TNF, and 27 MAPK signaling ( Figure 1E) 26,27 . In contrast, genes modified at the splicing level belong 28 to membrane-related regulatory processes including focal adhesion and ABC transporters 29 ( Figure 1E). In this setting, we observed that TAX-expressing cells displayed switched cell 30 adhesion properties from hyaluronate-to type IV collagen-coated surfaces, which is in 31 accordance with the substrate affinity of the CD44v10 isoform 28 ( Figure S1D). 32

33
The splicing factor DDX17 interacts with RELA and TAX in a NF-kB dependent 1

manner. 2
Since Tax is a well-known trans-acting transcription regulator, we first analysed whether 3 TAX could affect gene expression levels of splicing factors. However, no significant change 4 was measured for 227 genes encoding splicing regulators (Table S1, Figure 2A), thereby 5 suggesting a direct role of TAX in alternative splicing regulatory mechanisms. To tackle 6 this question, we focused on the auxiliary component of the spliceosome DDX17, which 7 has been previously identified, but not validated, in a recent mass spectrometry screen 8 for putative protein partners of TAX 29 . 9

10
We therefore aimed to validate the interaction between TAX and DDX17. As shown in 11 Figure 2B, TAX co-immunoprecipitated with the two endogenous isoforms of DDX17, 12 namely p72 and p82. Reciprocal IP confirmed this interaction ( Figure 2C). Due to the 13 involvement of NF-kB signaling in TAX positive cells ( Figure 1D, 27 ), we examined whether 14 DDX17 interacts with a TAX mutated form, namely M22 (G137A, L138S), which is 15 defective for IKK and NF-kB activation 30-33 . Despite similar expression levels and 16 immunoprecipitation efficiencies of TAX and M22 ( Figure 2D), we failed to detect any 17 interaction between M22 and DDX17 ( Figures 2B and 2C), suggesting that NF-kB is 18 required for recruiting DDX17. In this setting, RELA co-immunoprecipitated with DDX17 19 and TAX, but not with M22 ( Figures 2B and 2C). Moreover, DDX17 was co-20 immunoprecipitated with RELA in a TAX-dependent manner ( Figure 2E). This interaction 21 did not require RNA since the DDX17:RELA complex remained detected when cell extracts 22 were pre-treated with RNAse A ( Figure 2F). 23 24 As DDX17:RELA complexes were observed neither in control cells (that do not expressed 25 TAX) nor in M22 expressing cells, this suggested that NF-kB activation is necessary for the 26 binding of DDX17 to RELA. This hypothesis was confirmed by exposing TAXM22-27 expressing cells to TNFa, a potent NF-kB activator that allowed to retrieve DDX17:RELA 28 complexes ( Figure 2G). Altogether, these results revealed that TAX-induced NF-kB 29 activation dynamically orchestrates the interations between TAX, the transcription factor 30 RELA and the splicing regulator DDX17 ( Figure 2H). 31 32 TAX-mediated effects on splicing depend on DDX5/17. 33 To estimate the role of DDX17 in TAX-regulated splicing events, RNA-sequencing was 1 performed using 293T-LTR-GFP cells expressing or not TAX and depleted or not for 2 DDX17 and its paralog DDX5, which cross-regulate and complement each other 22,34,35 . 3 TAX had no effect on the expression of DDX5 and DDX17 (Figures 2A and 3A) and RELA 4 protein level was not significantly changed upon both TAX expression and DDX5/17 5 silencing ( Figure 3B). 6 7 Overall, 58.5% (648/1108) of TAX-regulated exons were affected by DDX5/17 8 knockdown, a significantly higher proportion than expected by chance ( Figure 3C, Figure  9 S2A). Of particular significance, 423 TAX-induced splicing events were completely 10 dependent on the presence of DDX5/17 (Table S3). For example, DDX5/17 silencing 11 completely abolished the TAX-mediated effect on splicing of SEC31B, CASK, MYCBP2, 12 CCNL1, ROBO1, ADD3 and CD44 transcripts ( Figure 3D). Of note, splicing specific RT-PCR 13 assays permitted to validate the effect of DDX5/17 on TAX-dependent splicing changes 14 for CD44, ADD3 and EIF4A2 transcripts, even though their predicted differential inclusion 15 fell below the arbitrary computational threshold (Table S3, Figure 3D and Figure S2D). 16 This suggested that the contribution of DDX5/17 to TAX-mediated alternative splicing 17 regulation might be under-estimated. 18

19
Finally, since NF-kB activation modified the interactions between DDX17, RELA, and TAX 20 (Figure 2), we examined the interplay between NF-kB activation and DDX17-mediated 21 splicing regulation. As shown in Figure 3D, M22 did not have any effect on DDX5/17-22 sensitive splicing events, arguing that TAX splicing targets are regulated by RNA helicases 23 DDX5/17 in an NF-kB dependent manner. 24 25 RELA binds to genomic exons and recruits DDX17 to regulate splicing in an RNA 26 helicase-dependent manner. 27 The results described above prompted us to hypothesize that the nuclear translocation of 28 RELA upon TAX expression might promote the chromatin recruitment of DDX17 to RELA 29 target genes. To test this hypothesis, the CD44 gene was used as a gene model. CD44 is 30 composed of 10 constitutive exons and 10 variable exons. The constitutive exons 1-5 and 31 15-20 encode the standard CD44 transcripts, while CD44 variants (CD44v) are produced 32 by extensive splicing leading to alternative inclusion of variable exons 5a-14 also named 33 v1-v10 ( Figure 4A) 36 . As shown above ( Figure 3D), the exon v10 inclusion rate is 1 markedly influenced by TAX in a DDX5/17-and NF-kB activation-dependent manner. The 2 importance of NF-kB in this process was further confirmed as the inactivation of NF-kB 3 via the ectopic expression of the IkBa super repressor (IkBSR) abolished the effects of 4 TAX on CD44 v10 inclusion ( Figure S3A). 5 6 Quantitative ChIP (qChIP) analyses revealed that RELA was recruited upon TAX 7 expression not only to the CD44 promoter, but also to a genomic region spanning the 8 alternative exon v10, but not a downstream constitutive exon (E16) (Figures 4A and 4B,9 left panel). To assess whether RELA occupancies at the v10 exon and CD44 promoter are 10 interrelated, a stable cell line was generated in which the kB site localized at -218 bp from 11 the transcription start site (TSS) was deleted using a CRISPR-Cas9 approach. Positive 12 clones (CD44DkB) were screened and sequenced to confirm the 40 bp deletion in the 13 promoter region ( Figure 4A). As expected, TAX expression failed to promote RELA 14 binding at the promoter in CD44DkB cells ( Figure 4B, right panel). Nevertheless, TAX still 15 promoted RELA binding at the v10 region. Importantly, TAX expression induced v10 16 inclusion at a similar level in both CD44DkB and parental cells ( Figure 4C). These results 17 suggested that TAX-mediated effect on exon v10 splicing could depend on RELA binding 18 in the vicinity of the alternative v10 exon. Supporting this hypothesis, the analysis of 19 publicly available RELA ChIP-seq datasets revealed that intragenic RELA peaks are 20 significantly closer to alternative exons than to constitutive exons ( Figure S3B). In this 21 setting, we observed that RELA binding sites are often found in the vicinity of TAX 22 regulated exons ( Figure 4D). Using the MEME-ChiP suite as motif discovery algorithm 37 , 23 we uncovered that RELA-binding sites located within the closest range (<1kbp) of TAX-24 regulated exons coincided with the typical NF-kB consensus motif ( Figure 4E). 25 Furthermore, this subset of TAX-regulated exons displayed weak 3' and 5' splice sites 26 together with significant low MFE value ( Figure 4F) and high GC-content ( Figure 4G) 27 when compared to all human exons. This emphasizes the high potential of these splice 28 sites to form stable secondary RNA structures, a typical feature of exons regulated by RNA 29 helicases DDX5/17 34 . Taken together, these data define a signature of splicing target 30 specificity for RELA, and they suggest that RELA and DDX17 might control together the 31 inclusion of a subset TAX-regulated exons. We therefore investigated the genomic 32 occupancy of some target exons by RELA and DDX17 by qChIP analysis of cells expressing 33 or not TAX. For all tested genes (CD44, SEC31B, CASK, and MYCBP2), both RELA and 1 DDX17 bound specifically the regulated alternative exon in a TAX-dependent manner, 2 compared to a downstream constitutive exon ( Figure 4H). Furthermore, RELA binding 3 was lost in cells depleted for DDX5/17, indicating that RNA helicases contribute to 4 stabilize DNA-bound RELA ( Figure 4I). 5 6 A causal relationship between exonic DNA-binding of RELA, chromatin recruitment 7 of DDX17, and splicing regulation. 8 To assess the causative relationship that links RELA and DDX17 to alternative splicing, we 9 intended to experimentally tether DDX17 or RELA at the CD44 v10 exon locus using 10 modified TALE (Transcription-Activator-Like-Effector) 38 . We designed a TALE domain 11 that recognizes specifically an exonic 20 bp DNA sequence located 12 bp upstream from 12 the 5′ splice site (SS) of exon v10. This TALE domain was either fused to RELA or DDX17 13 proteins ( Figure 5). We also used an additional construct consisting in the same TALE 14 fused to GFP to rule out non-specific effects resulting from the DNA binding of the TALE. 15 Each TALE construct was transiently transfected into 293T-LTR-GFP cells, and we 16 monitored their relative effects both on the recruitment of endogenous RELA and DDX17, 17 and on the splicing of exon v10. All results shown in Figure 5 were normalized and 18 expressed as relative effects compared to the TALE-GFP. As expected, and validating our 19 approach, TALE-RELA tethering to the exon v10 led to a significant chromatin recruitment 20 of RELA to its target site, and not to the downstream exon E16 used as control ( Figure 5A, 21 left panel). A significant and specific enrichment of DDX17 was also observed on exon v10 22 upon expression of the TALE-RELA compared to TALE-GFP ( Figure 5A, left panel), 23 indicating that tethering RELA to exon v10 induced a local recruitment of endogenous 24 DDX17 proteins. At the RNA level, this TALE-RELA-mediated recruitment of DDX17 25 coincided with a significant increase in exon v10 inclusion rate ( Figure 5A). 26

27
We next investigated whether DDX17 tethering could result in similar effects. 28 Quantitative ChIP analysis demonstrated that DDX17 was properly tethered to exon v10 29 when fused to the designed TALE ( Figure 5B) but TALE-DDX17 had no effect on RELA 30 recruitment ( Figure 5B). This was expected since the formation of RELA:DDX17 31 complexes only occurs upon NF-kB activation ( Figure 2). Nevertheless, TALE-DDX17-32 expressing cells exhibited a reproducible and significant increase in v10 inclusion ( Figure  33 5B), indicating that chromatin-bound DDX17 alone can modulate splicing efficiency. It is 1 worth to underline that the level of v10 exon inclusion induced by the TALE-RELA and -2 DDX17 was comparable to that measured in cells transiently transfected with a TAX 3 expression vector ( Figure 4C and Figure S1C). Although it is less quantitative approach, a 4 nested RT-PCR assay clearly confirmed these results ( Figure S4). Strikingly however, the 5 TALE-DDX17_K142R (a DDX17 helicase mutant 34,39-41 ) failed to influence exon v10 6 inclusion despite a clear chromatin enrichment of DDX17 ( Figure 5C, Figure S4). 7 Collectively, these results demonstrate that the binding of RELA at the vicinity of genomic 8 exons recruits the RNA helicase DDX17 that positively regulates the inclusion rate of the 9 target exon thanks to its RNA helicase activity. 10

DISCUSSION 1
Since the finding of splicing dysregulations in HTLV-1 infected individuals 4,24,25,42 , 2 deciphering how HTLV-1 interferes with the splicing regulatory network has become a 3 new challenging issue for improving our knowledge of HTLV-1 infection and its associated 4 diseases. Here, we provide the first molecular evidence that upon TAX-induced NF-kB 5 activation, RELA directly regulates splicing by binding to gene bodies at the vicinity of GC-6 rich exons and by locally recruiting the splicing factor DDX17, which regulates splicing via 7 its RNA helicase activity. 8 9 Our results demonstrate for the first time that TAX deeply impacts alternative splicing 10 independently from its effects on transcription. In addition, TAX-regulated exons were 11 found in transcripts enriched in functional pathways that are distinct from those enriched 12 by TAX transcriptional targets, suggesting that splicing reprogramming may constitute an 13 additional layer of regulations by which HTLV-1 modifies the host cell phenotype. Arguing 14 for this, we showed that the TAX-induced splicing variant CD44v10, which was previously 15 identified in circulating blood of HAM/TSP patients 25 and confirmed here ex vivo in 16 infected CD4+ T-cell clones, contributes to modulate cell adhesion affinity in vitro. GO 17 analyses of TAX splicing targets also pointed to the phosphatidylinositol signaling system 18 and to the inositol phosphate metabolism, two processes that are particularly connected 19 to NF-kB signaling and that play critical roles in oncogenesis and disease progression of 20 malignant diseases, including ATLL 43,44 . This suggests that, besides its transcriptional 21 effects, splicing regulatory functions of TAX might account for its oncogenic properties. 22 Accordingly, a large number of TAX-regulated exons could be observed in ATLL samples, 23 which rarely express TAX but typically exhibit NF-kB addiction for survival and 24 proliferation 24,45,46 . 25

26
At the molecular level, we showed that the increased chromatin occupancy of RELA upon 27 TAX expression is not restricted to promoter regions but also occurs in the vicinity of 28 exons that are regulated at the splicing level ( Figure 4). Exons regulated by TAX, especially 29 those localized within 1 kb of intragenic RELA binding sites, are characterized by a high 30 GC-content, a typical feature of exons regulated by the DDX5 and DDX17 RNA helicases 34 31 ( Figure S3D). Accordingly, we found that a majority of TAX-regulated exons depend on 32 the expression of these proteins ( Figure 3). A local chromatin recruitment of DDX17 and 33 RELA was validated on several TAX-regulated exons ( Figure 4). More importantly we 1 identified a confident causal relationship between the exonic tethering of RELA, the local 2 chromatin recruitment of DDX17, and the subsequent splicing regulation via DDX17 RNA 3 helicase activity. This catalytic activity of DDX17 is strictly required for its splicing 4 regulatory functions ( Figure 5), as previously reported 34 . Indeed, the RNA helicase 5 activity of DDX5 and DDX17 has been involved in resolving RNA structures, facilitating 6 the recognition of the 5' splice site, that can be embedded in secondary structures, and 7 exposing RNA binding motifs to additional splicing regulators 34,40,47-49 . However, even 8 though some RNA binding specificity has been reported for DDX17 50,51 , these RNA 9 helicases are devoid of a proper RNA binding domain and their activity in splicing may 10 depend on additional factors that are able to provide target specificity. Here, we suggest 11 that RELA may be also regarded as a DDX17 recruiter by acting as a chromatin anchor for 12 DDX17 in the vicinity of exons dynamically selected upon NF-kB activation. 13

14
The target specificity of NF-kB factors remains a complex question. It has been estimated 15 that approximately 30 to 50% of genomic RELA binding sites do not harbor a typical NF-16 kB site, and that only a minority of RELA-binding events associate with transcriptional 17 change 16-19 , thereby indicating that neither a consensus site nor significant NF-kB 18 occupancy are sufficient criteria for defining RELA's target specificity. Here, we identified 19 a typical kB consensus motif at RELA-binding loci that are close to alternatively spliced 20 exons but we also uncovered that weak splice sites, low MFE, and significant GC-content 21 bias of exons likely contribute to RELA's target specificity. Because low MFE and high GC-22 content confer a high propensity to form stable RNA secondary structures, the recognition 23 and the selection of such GC-rich exons with weak splice sites by the splicing machinery 24 typically depend on RNA helicases DDX5/17 34 . Based on these observations we propose 25 that, upon TAX-induced NF-kB activation, RELA binds to intragenic binding consensus 26 motifs and locally recruits DDX17. When the RELA:DDX17 complex is located at a close 27 proximity of GC-rich exons flanked by weak splice sites, DDX17 can impact on their 28 inclusion rate by unwinding GC-rich secondary structures of the nascent RNA transcript, 29 and by potentially unmasking binding motifs for additional splicing regulators. 30

31
In conclusion, our results provide conceptual advance for understanding how cell 32 signaling pathways may drive target specificity in splicing by dynamically recruiting 33 cognate transcription factors at the vicinity of target exons that act as chromatin anchor 1 for splicing regulators. In the context of NF-kB signaling, such mechanism likely has a 2 significant impact on cell fate determination and disease development associated with 3 HTLV-1 infection, but also on other situations linked to chronic NF-kB activation, as 4 numerous human inflammatory diseases and cancer.

Cell Culture and Transfections 2
Peripheral blood mononuclear cells (PBMCs) were obtained by Ficoll separation of whole 3 blood of HTLV-1 infected individuals. T-cell limiting dilution cloning was performed as 4 previously described 4 . The human embryonic kidney 293T-LTR-GFP cells 52 , which 5 contain an integrated GFP reporter gene under the control of the TAX-responsive HTLV-6 1 LTR, were cultured in DMEM+Glutamax medium supplemented with decomplemented 7 10% FBS and 1% penicilline/streptomycine. This cell line was used to measure 8 transfection efficiency in TAX and TAX M22 conditions. In standard transfection 9 experiments, siRNAs (Table S4)  Tween, membranes were incubated 1h at room temperature with the secondary 33 antibodies conjugated with the HRP enzyme and washed 3 times as above. Finally, the 1 HRP substrate (GE Helathcare) was applied to the membrane for 5 minutes and the 2 chemiluminescence was read on Chemidoc (Biorad). 3 4

Co-immunoprecipitation 5
Cells were harvested in IP lysis buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 2 mM EDTA, 6 1% NP40, 10% Glycerol). Extracts were incubated overnight with 5 µg of antibodies 7 recognizing either RELA (C20 sc-372, Santa Cruz), Tax (1A3, Covalab), DDX17 8 (ProteinTech) in the presence of 30 μL Dynabeads® Protein A/G (Thermo Fisher). 9 Isotype IgG rabbit (Invitrogen) or mouse (Santa Cruz) was also used as negative control. 10 The immunoprecipitated complexes were washed three times with IP lysis buffer. NaCl , 20mM Tris HCL pH8,1mM EDTA). The immunoprecipitated chromatin was purified 29 by phenol-chloroform extraction and quantitative PCR was performed using Rotor-Gene 30 3000 cycler (Corbett) or LightCycler 480 II (Roche, Mannheim, Germany). Values were 31 expressed relative to the signal obtained for the immunoprecipitation with control IgG. 32 Primers used for ChIP experiments were designed in exon/intron junction (Table S4). For 33 TALE ChIP experiment, DDX17 and RelA enrichment were normalized to the signal 1 observed with V5 antibody corresponding to TALE recruitment. Then, the TALE GFP 2 condition was used as control and set to 1. 3 4 RNA extraction, classical PCR and Real-time quantitative PCR. 5 Total RNAs were extracted using TRIzol (Invitrogen). RNAs (2.5 μg) were retro-6 transcribed with Maxima First Strand cDNA Synthesis Kit after treatment with dsDNase 7 (Thermo Scientific) following the manufacturer's instructions. PCRs were performed 8 using 7.5 ng of cDNAs with GoTaq polymerase (Promega, Madison, WI, USA). PCR 9 products were separated by ethidium bromide-labeled agarose gel electrophoresis. Band 10 intensity was quantified using the ImageLab software (Bio-Rad). Quantitative PCR was 11 then performed using 5 ng of cDNAs with SYBR® Premix Ex Taq TM II (Tli RNaseH Plus) 12 on LightCycler 480 II. Relative level of the target sequence was normalized using the 18S 13 or GAPDH gene expression (∆Ct) and controls were set to 1(∆∆Ct). We calculated the 14 inclusion rate of alternative exons using the following method: 2 −∆∆Ct (included 15 exon)/2 −∆∆Ct (constitutive exon). The oligonucleotide sequences used are listed in Table  16 S4. 17

RNA-seq and bio-informatic analysis 19
RNA-seq analyses were performed as previously described 22 . Briefly, poly-A transcripts 20 were extracted from 293T-LTR-GFP cells transfected with pSG5M-Tax or pSG5M empty 21 vectors and knockdown or not for DDX5-17. RNA-seq libraries were generated at the Aros 22 Applied Biotechnology (Aarhus, Denmark) using Stranded mRNA Sample Prep kit 23 (Illumina) and sequenced using the illumina HiSeq 2500 technology. Each sample have in 24 average 6.10 7 of paired-end pairs of reads. These RNA-seq data were analyzed using 25 FaRLine, a computational program dedicated to analyzing alternative splicing with 26 FasterDB database 23,53 . The gene expression level in each sample was calculated with 27 HTSeq-count (v0.7.2) 54 and differential expression between conditions was computed 28 with DESeq2 (v1.10.1) ( abs(log2FoldChange) ≥ 0.4, pvalues ≤ 0.05) 55 . In silico screening 29 of NF-kB responsive elements in the CD44 promoter sequence was carried out via PROMO 30 database (based on TRANSFAC v8.3) 56 . The MEME-ChIP suite was used to discover the 31 regulatory motifs in the NF-kB ChIP-seq data 37 . 32 For the prediction of splice site strength, scores were computed using MaxEntScan 57 for 1 sequence (3 bases in the exon and 6 bases in the intron for 5' splice sites; 20 bases in the 2 intron and 3 bases in the exon for 3'splice sites) covering both sides of the splicing site. The distribution of RELA peaks across alternative and constitutive exons, and the average 11 distance between RELA peaks and TAX exon targets was measured using ChiP-seq 12 datasets from GEO 58 , ENCODE 59 and CISTROME 60 databases: from GEO GSE63736, 13 GSM1239484, GSM486271, GSM486293, GSM486298, GSM486318, GSM847876, 14 GSM847877, GSM2394419, GSM2394421, GSM2394423, from ENCODE ENCFF002CPA, 15 ENCFF002CQB, ENCFF002CQJ, ENCFF002CQN, ENCFF580QGA and from CISTROME 16 53597, 5388, 5389, 4940, 36310, 36316, 4971. For another GEO dataset, GSM2628088, 17 reads were mapped to the hg19 build of the human genome with Bowtie2 61 and RelA 18 peaks were identified with Macs2 62 . Alternative and Constitutive spliced exons were 19 obtained from FasterDB 53 . In order to focus on intragenic RELA peaks, we used the 20 bedtools 63 intersect command to remove all intergenic RELA peaks and all RELA peaks 21 localized on first exon (or at least at less than 500nt) for each gene. A Perl script was 22 specifically created to measure the distance between RELA peaks and TAX-regulated 23 exons. Briefly, RELA peaks and exons are provided as BED files and the script reports for 24 each exon the distance in nucleotides of the nearest RELA peak. Closest peak distances 25 from the 710 TAX-regulated exon-cassettes were compared to closest peak distances 26 from 710 exons chosen by chance (10 5 runs). 27

TALE design and construct 29
The TALE constructs were obtained from ThermoFisher Scientific. TALEs were 30 constructed using the Golden Gate Assembly method as previously described 38 . The RVDs 31 HD, NI, NG and NN were chosen to specifically recognize the nucleotides C, A, T and G, 32 respectively. The TALE targeting CD44 v10 sequence was 5' TCCAACTCTAATGTCAATC 3'. 33 This TALE construct was fused to a V5 sequence and a SV40 NLS at its 5' end and cloned 1 in the NotI-HindIII fragment of the pXJ41 backbone plasmid. DDX17-WT and DDX17-2 K142R cDNA were obtained by PCR from pcDNA3-HA-DDX17 and pcDNA3-HA-DDX17-3 K142R and were cloned in the HindIII-BglII fragment in the MCS downstream to the TALE 4 sequence. The RELA cDNA was amplified from a library of cDNA of 293T-LTR-GFP cells 5 and was cloned in the HindIII-BamHI fragment. 6 7 CRISPR design and construct 8 The sequence-specific sgRNA for site-specific interference of genomic targets were 9 designed using CRISPRseek R package 1 , and sequences were selected to minimize off-10 target effect 64 . Two complementary oligonucleotides were annealed and cloned into BbsI  (Table S4). At 24h post-transfection, the medium was changed and 1μg/ml 15 puromycin was added for selection and cells were cloned by serial dilution method.   Histograms represent the results of exon specific quantitative RT-PCR measurements computed as a relative exon inclusion (alternatively spliced exon vs constitutive exon reflecting the total gene expression level) from three biological replicates ± s.d.. All of these genes but MYCBP2 were unmodified at the whole transcript level upon TAX expression ( Figure S2C).  ChIP-seq datasets were analysed as detailed in method section. We excluded RELA peaks localized in intergenic regions and exons linked to specific events like pomoters, alternative first/last and mutually exclusive exons. The groups "Constitutive exons" and "Alternative exons" contained 41873 and 103000 exons, respectively. The window was fixed to 3 kb upstream and downstream of each exon coordinates. P-value was calculated using the Mann-Whitney test. Oligonucleotides were previously described (36). The first round of amplification consisted in 15 cycles of PCR with the primers C13 and C12A, the second round consisted in 35 cycles with primers pv10 and C2A. Final PCR products were resolved on Agarose gel (1%).