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

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 also massively bind to gene bodies. Here, we demonstrate that recruitment of the NF-κB factor RELA to intragenic regions regulates alternative splicing upon NF-κB activation by the viral oncogene Tax of HTLV-1. Integrative analyses 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 an NF-κB activation-dependent manner. This leads to alternative splicing of target exons due to the RNA helicase activity of DDX17. Similar results were obtained upon Tax-independent NF-κB activation, indicating that Tax likely exacerbates a physiological process where RELA provides splice target specificity. 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.

T he human T-cell leukemia virus (HTLV-1) is the etiologic agent of numerous diseases, including adult T-cell leukemia/lymphoma (ATLL) 1 , an aggressive CD4 + T-cell malignancy, and various inflammatory diseases, such as HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) 2 . Changes at the gene expression level participate in the persistent clonal expansion of HTLV-infected CD4 + and CD8 + T-cells, leading ultimately to HTLV-1-associated diseases 3 . We recently reported that alternative splicing events help to discriminate between ATLL cells, untransformed infected cells, and their uninfected counterparts derived from patients 4 . Alternative splicing of pre-messenger RNAs is a co-transcriptional processing step that controls diversity in both the transcriptome and the proteome and that governs cell fate. Its regulation relies on a complex and still incompletely understood interplay between splicing factors, chromatin regulators, and transcription factors 5,6 . For instance, we still do not fully understand the molecular mechanisms of HTLV-1-induced splicing modifications, or whether these effects rely on interplay between transcription and splicing.
Tax is an HTLV-1-encoded protein that regulates viral and cellular gene transcription. Tax also alters host signaling pathways that sustain cell proliferation, leading ultimately to cell immortalization 7 . The nuclear factor-κB (NF-κB) signaling pathway is the most critical target of Tax for cell transformation 8 . The NF-κB transcription factors (RELA, p50, c-Rel, RelB, and p52) govern immune functions, cell differentiation, and proliferation 9 . NF-κB activation involves the degradation of IκB, which sequesters NF-κB factors in the cytoplasmic compartment; this leads to NF-κB nuclear translocation and binding of NF-κB dimers (of which RELA:p50 is the most abundant) to their target promoters 10,11 . Tax induces IKK phosphorylation and IκB degradation, leading to persistent nuclear translocation of NF-κB 12,13 . In addition, Tax interacts with nuclear NF-κB factors and enhances their effects on transcription 14,15 .
Interestingly, genome-wide analyses of NF-κB distribution have unveiled that the vast majority of RELA peaks is outside promoter regions and can be localized in introns and exons [16][17][18][19] . Some of those promoter-distant RELA-binding sites correspond to cis-regulatory transcriptional elements 20,21 but globally, there is a weak correlation between the binding of RELA to genes and regulation of their steady-state expression 17,18 . These data suggest that NF-κB could have other functions than its initially described function as a transcription factor.
Here, we show that NF-κB activation accounts for alternative splicing modifications generated upon Tax expression. These effects rely on a tight physical and functional interplay between RELA and the DDX17 splicing factor, which dynamically occurs upon NF-κB activation. Our results reveal that RELA DNA binding in the vicinity of genomic exons regulates alternative splicing through DDX17 recruitment, which then modulates exon inclusion via its RNA helicase activity.

Results
Tax regulates splicing regardless of its transcriptional effects. We performed RNA-seq analyses on 293T-LTR-GFP cells transiently transfected with a Tax expression vector. We then identified Tax-induced changes in gene expression level and in alternative splicing and annotated them as previously described 22,23 ( Supplementary Data 1 and 2). Notably, the ectopic expression of Tax affected the splicing and gene expression levels of 939 and 523 genes, respectively (Fig. 1a). A total of 1108 alternative splicing events was predicted, including 710 exon skipping events (Fig. 1b). A small percentage of genes (3.5%, 33/ 939) was altered at both their expression and splicing levels, indicating that Tax largely affects alternative splicing independently of its transcriptional activity. We validate a subset of splicing events by RT-PCR (Fig. 1c). To test whether Tax-induced splicing modifications occurred in vivo, we took advantage of a publicly available RNA-seq dataset (EGAS00001001296 24 ) from three healthy control donors, three asymptomatic carriers (AC), and 55 ATLL patients (Supplementary Data 1). Although lowly expressed, Tax mRNA was detected in the vast majority of infected samples (Supplementary Fig. 1). It is important to note that ATL can sporadically express Tax in bursts 25,26 ; thereby explaining, at least in part, their Tax-like features even though Tax is poorly detected in ATL samples 24 . Based on this, we examined the percent spliced-in (PSI) values of exons expressed in samples from donors, carriers, and patients with ATLL (Supplementary Data 1). Overall, 542 (48%) Tax-induced splicing modifications were detected at least once in 55 ATL samples (Supplementary Data 1). This low frequency was anticipated due to the high fluctuations between individuals of Tax expression levels, proviral loads, and infected cell clonality, as well as somatic alterations (e.g., gene mutations, deletions, and duplications), which accumulate in ATLL cells 24 . Some Tax-related splicing events were recurrently detected in asymptomatic and ATLL samples (Fig. 1d). In addition, using exon-specific RT-PCR assays, we confirmed that Tax promotes splicing events that have been previously described in HTLV-1-infected individuals, including in the AASS, CASK, RFX2, and CD44 genes 4,27 (Fig. 1c). Altogether, these results uncovered a large number of splicing modifications that occurred upon Tax expression, which to a degree coincide with alternative splicing events observed in HTLV-1 patients.
DDX17 interacts with RELA upon Tax-induced NF-κB activation. As Tax is a well-known trans-acting transcription regulator, we first analyzed whether Tax could affect gene expression levels of splicing factors. However, no significant change was measured for 227 genes encoding splicing regulators (Supplementary Data 1, Fig. 2a), thereby suggesting a direct role of Tax in alternative splicing regulatory mechanisms. To tackle this question, we focused on the auxiliary component of the spliceosome DDX17, which has been previously identified, but not validated, in a recent mass spectrometry screen for putative protein partners of Tax 28 .
We therefore aimed to validate the interaction between Tax and DDX17. As shown in Fig. 2b, Tax co-immunoprecipitated with the two endogenous isoforms of DDX17, namely p72 and p82. Reciprocal IP confirmed this interaction (Fig. 2c). Due to the involvement of NF-κB signaling in Tax-positive cells (Fig. 1d 11 ), we examined whether DDX17 interacts with a Tax mutated form, namely M22 (G137A, L138S), which is defective for IKK and NF-κB activation [29][30][31][32] . Despite similar expression levels and immunoprecipitation efficiencies of Tax and M22 (Fig. 2d), we failed to detect any interaction between M22 and DDX17 (Fig. 2b,c), suggesting that NF-κB is required for recruiting DDX17. In this setting, RELA co-immunoprecipitated with DDX17 and Tax, but not with M22 (Fig. 2b, c). Moreover, DDX17 was coimmunoprecipitated with RELA in a Tax-dependent manner (Fig. 2e). This interaction did not require RNA as the DDX17: RELA complex remained detected when cell extracts were pretreated with RNAse A (Fig. 2f).
As DDX17:RELA complexes were observed neither in control cells (that do not expressed Tax) nor in M22-expressing cells, this suggested that NF-κB activation is necessary for the binding of DDX17 to RELA. This hypothesis was confirmed by exposing M22-expressing cells to TNFα, a potent NF-κB activator that allowed retrieving DDX17:RELA complexes (Fig. 2g). Altogether, these results revealed that Tax dynamically orchestrates interactions between the transcription factor RELA and the splicing regulator DDX17 by activating the NF-κB signaling pathway (Fig. 2h).
Tax-mediated effects on splicing depend on DDX5/17. To estimate the role of DDX17 in Tax-regulated splicing events, RNA-sequencing was performed using 293T-LTR-GFP cells expressing or not Tax and depleted or not for DDX17 and its paralog DDX5, which cross-regulate and complement each other 22,33,34 . Tax had no effect on the expression of DDX5 and DDX17 (Figs. 2a and 3a) and RELA protein level was not significantly changed upon both Tax expression and DDX5/17 silencing (Fig. 3b).
Overall, 58.5% (648/1108) of Tax-regulated exons were affected by DDX5/17 knockdown, a significantly higher proportion than expected by chance (Fig. 3c, Supplementary Fig. 2a (Fig. 3d). Tax, as well as siRNA-mediated DDX5/17 depletion, had no marked effects on gene expression levels of those genes ( Supplementary Fig. 2c, d). Of note, splicing specific RT-PCR assays permitted to validate the effect of DDX5/17 on Tax-dependent splicing changes for CD44, ADD3, and EIF4A2 transcripts, even though their predicted differential inclusion fell below the arbitrary computational threshold ( Fig. 3d and Supplementary Fig. 2b). This suggested that the contribution of DDX5/17 to Tax-mediated alternative splicing regulation might be under-estimated.
As NF-κB activation modified the interactions between DDX17, RELA, and Tax (Fig. 2), we next examined the interplay between NF-κB activation and DDX17-mediated splicing regulation. As shown in Fig. 3d, M22 did not have any effect on DDX5/ 17-sensitive splicing events, arguing that Tax splicing targets are regulated by RNA helicases DDX5/17 in an NF-κB-dependent manner. This was further confirmed using siRNA-mediated depletion of RELA that abolished DDX17-dependent splicing regulations by Tax (Fig. 3d).    Functional insight of Tax and DDX17-dependent spliced genes. Gene ontology (GO) analysis of quantitatively altered genes revealed several signaling pathways that are well described in Tax-expressing cells, including NF-κB, TNF, and MAPK signaling ( Supplementary Fig. 3C) 11,35 . In contrast, genes modified at the splicing level belong to membrane-related regulatory processes including focal adhesion and ABC transporters (Supplementary Fig. 3C). In particular, the term Focal Adhesion was shared between Tax splicing targets identified in infected samples (RNA-seq datasets, Supplementary Data 1) and those regulated by DDX17 (Supplementary Data 3) ( Supplementary Fig. 3C). In addition, using an exon ontology (EO) approach that we recently developed to estimate enrichment in protein features encoded by exons 36 , we found that Tax-and DDX17-regulated exons encoded for regions involved in functionally validated posttranslational modifications (PTM), protein structure and    Supplementary  Fig. 3C). These data are reminiscent of PTM related to protein regions encoded by variable CD44 exons that also control the cell-adhesion properties of CD44 ( Supplementary Fig. 3D). Accordingly, we observed that Tax-expressing HEK cells displayed switched cell-adhesion properties from hyaluronate-to type IV collagen-coated surfaces, which is in accordance with the substrate affinity of the CD44 v10 isoform 37 ( Supplementary  Fig. 3E).
RELA recruits DDX17 at the vicinity of genomic exons. The results described above prompted us to focus on CD44 as gene model to further examine the mechanisms underlying RELA and DDX17 splicing regulations upon Tax. CD44 is composed of 10 constitutive exons and 10 variable exons. The constitutive exons 1-5 and 15-20 encode the standard CD44 transcripts, while CD44 variants (CD44v) are produced by extensive splicing leading to alternative inclusion of variable exons 5a-14 also named v1-v10 ( Fig. 4b) 38 . As shown in Fig. 4a, the exon v10 inclusion rate is markedly influenced by Tax in a DDX5/17-and NF-κB activation-dependent manner. The importance of NF-κB in this process was further confirmed as the inactivation of NF-κB via the ectopic expression of the IκBα super repressor (IκBSR) abolished the effects of Tax on CD44 v10 inclusion ( Supplementary Fig. 4A). Remarkably, in line with previous work suggesting that NF-κB directly regulates the CD44 promoter, we also noticed a slight reduction in CD44 expression in Tax-expressing cells knocked down for RELA ( Supplementary Fig. 4B), thereby defining CD44 regulations as an appropriate situation to address whether or not RELA-dependent transcription and splicing are two interrelated processes.
Using quantitative ChIP (qChIP) analyses, we first observed that Tax expression led to recruit RELA not only to the CD44 promoter, but also to a genomic region spanning the alternative exon v10, but not a downstream constitutive exon (E16) (Fig. 4b, c, left panel). To assess whether RELA occupancies at the v10 exon and CD44 promoter are interrelated, a stable cell line was generated in which the κB site localized at -218 bp from the transcription start site (TSS) was deleted using a CRISPR-Cas9 approach. Positive clones (CD44ΔkB) were screened and sequenced to confirm the 40 bp deletion in the promoter region (Fig. 4b). As expected, Tax expression failed to promote RELA binding at the promoter in CD44ΔkB cells (Fig. 4c, right panel), coinciding with a decreased CD44 expression comparable to that observed upon RELA depletion ( Supplementary Fig. 4B, C). Nevertheless, Tax still promoted RELA binding at the v10 region. Importantly, Tax expression induced v10 inclusion at a similar level in both CD44ΔkB and parental cells (Fig. 4d). These results suggested that Tax-mediated effect on exon v10 splicing could depend on RELA binding in the vicinity of the alternative v10 exon. Supporting this hypothesis, the analysis of publicly available RELA ChIP-seq datasets revealed that intragenic RELA peaks are significantly closer to alternative exons than to constitutive exons ( Supplementary Fig. 4D). More particularly, alternative exons regulated by Tax, including those sensitive to DDX5/17 depletion, were identified in genomic regions enriched in NF-κB-binding sites compared to alternative exons non-regulated by Tax (Fig. 4e).   TaxM22 and siRNA-mediated RELA depletion were used in order to assess the dependency of splicing events on NF-κB activation. 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). All genes but MYCBP2 were unmodified at the whole transcript level upon Tax expression ( Supplementary Fig. 2c). Data are presented as the mean ± SEM values from biological replicates. Each black square represents a biological replicate. Statistical significance was determined with two-way ANOVA followed by Fisher's LSD test (**p < 0.01, ****p < 0.0001 Accordingly, we observed that RELA-binding sites are often found in the vicinity of Tax-regulated exons (Fig. 4f). Using the MEME-ChiP suite as motif discovery algorithm 39 , we uncovered that RELA-binding sites located within the closest range (<1 kb) of Tax-regulated exons coincided with the typical NF-κB consensus motif (Fig. 4g). Furthermore, this subset of Taxregulated exons displayed weak 3′ and 5′ splice sites together with significant low minimum free energy (MFE) value (Supplementary Fig. 4E) and high GC-content ( Supplementary Fig. 4F) when compared to all human exons. This emphasizes the high potential of these splice sites to form stable secondary RNA structures, a typical feature of exons regulated by RNA helicases DDX5/17 33 . Taken together, these data define a signature of splicing target specificity for RELA, and they suggest that RELA and DDX17 might control together the inclusion of a subset of Tax-regulated exons. We therefore investigated the genomic occupancy of some target exons by RELA and DDX17 by qChIP analysis of cells expressing or not Tax. For all tested genes (CD44, SEC31B, CASK, and MYCBP2), both RELA and DDX17 bound specifically the regulated alternative exon in a Tax-dependent manner, compared to a downstream constitutive exon ( Fig. 5a and Supplementary 4g). Remarkably, the knockdown of RELA expression by siRNA affected DDX17 chromatin occupancy on Tax-regulated exons (Fig. 5a), despite its slight positive impact on DDX17 expression ( Supplementary Fig. 2d). We further validated the RELA/DDX17dependent alternative splicing events in C91PL and ATL2 cells compared to the non-infected MOLT4 cells (Fig. 5c). The magnitude of splicing regulation was higher in ATL2 than in C91PL cells, and appeared to positively correlate to a higher expression level of both DDX17 and RELA, and a high level of constitutive NF-κB activation, as reflected by IL8 expression (Fig. 5b). As formerly observed in HEK cells expressing Tax, these  splicing events in HTLV-1-infected cells coincided with a high and significant increase in chromatin occupancy of both RELA and DDX17 at the vicinity of regulated exons (Fig. 5d). These data reveal that the RELA:DDX17 axis in the splicing regulatory network pertains to both HEK293T and CD4 + -infected cells expressing Tax.
Causal relationship linking RELA and DDX17 to splicing regulation. To more confidently assess the causative relationship linking RELA and DDX17 to alternative splicing, we experimentally tethered DDX17 or RELA to the CD44 v10 exon locus using modified TALE (Transcription-Activator-Like-Effector) 40 . For this, we designed a TALE domain that recognizes specifically an exonic 20 bp DNA sequence located 12 bp upstream from the 5′ splice site (SS) of exon v10. This TALE domain was fused to either RELA or DDX17 proteins. We also used an additional construct consisting in the same TALE fused to GFP to rule out non-specific effects resulting from the DNA binding of the TALE. After transient transfection of each TALE construct into 293T-LTR-GFP cells (Fig. 6a), we monitored the relative effects on the recruitment of endogenous RELA and DDX17 as well as on exon v10 splicing; all results were normalized and expressed as relative effects compared to the TALE-GFP (Fig. 6a-c). As expected, and validating our approach, TALE-RELA tethering to the exon v10 led to a significant chromatin recruitment of RELA to its target site, and not to the downstream exon E16 used as control (Fig. 6a, left panel). Further, a significant and specific DDX17 enrichment was observed at exon v10 after expression of TALE-RELA as compared to TALE-GFP (Fig. 6a, left panel), indicating that tethering RELA to exon v10 induced local recruitment of the endogenous DDX17 protein. At the RNA level, this TALE-RELA-mediated recruitment of DDX17 coincided with a significant increase in the exon v10 inclusion rate (Fig. 6a, right  panel).   Relative exon inclusion was measured as described in Fig. 3. d Relative RELA and DDX17 occupancies of regulated exons in ATL2 cells as compared to MOLT4 cells. Each occupancy of regulated exon by RELA and DDX17 is represented as a fold of that measured at its neighboring constitutive exon. Source data are provided as a Source Data file. Data are presented as the mean ± SEM values from biological replicates. Each black square represents a biological replicate. Statistical significance was determined with two-way ANOVA followed by Fisher's LSD test (a, d) and two-tailed unpaired t-test (b, c) (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). We next investigated whether DDX17 tethering could result in similar effects. Quantitative ChIP analysis demonstrated that DDX17 was properly tethered to exon v10 when fused to the designed TALE, but TALE-DDX17 had no effect on RELA recruitment (Fig. 6b, left panel). This was expected since the formation of RELA:DDX17 complexes only occurs upon NF-κB activation (Fig. 2). Nevertheless, TALE-DDX17-expressing cells exhibited a reproducible and significant increase in v10 inclusion   Fig. 6 Chromatin relationship between RELA and DDX17. a-c Chromatin and splicing regulation upon TALE-mediated tethering of RELA and DDX17. The TALE domain was designed to bind the v10 exon of CD44 and fused to either GFP (a-c), RELA (a), DDX17, (b), or a helicase-deficient mutant DDX17_K142R (c). The effect of TALEs on RELA and DDX17 chromatin enrichment (left panels) and on the relative v10 exon inclusion (right panels) was monitored in HEK cells by qChIP and qRT-PCR, respectively. Results were normalized to measures obtained in TALE-GFP assays. (d) Relative exon inclusion rate of exon v10 of CD44 in HEK cells expressing or not the Tax mutant M22 and the TALE-DDX17 construct. e Relative exon inclusion rate of exon v10 of CD44 in HEK cells exposed to TNFα and PMA. f Relative exon inclusion rate of exon v10 of CD44 in HEK cells transiently transfected with increasing amounts of RELA expression vector (200 and 500 ng). g Model of NF-κB-dependent regulation of alternative splicing. Upon NF-κB activation, DNA-bound RELA proteins act as chromatin anchors for DDX17, which then provides splicing target specificity due to its RNA helicase activity. Data are presented as the mean ± SEM values from biological replicates. Each black square represents a biological replicate. Statistical significance was determined with two-tailed unpaired t-test (qChIP in a-c) and one-way ANOVA followed by Fisher's LSD test (relative exon inclusion (REI) in a-f) (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). In a-c, ( Fig. 6b, right panel), indicating that chromatin-bound DDX17 alone (e.g., without RELA) can modulate splicing efficiency. Of note, the levels of v10 exon inclusion induced by TALE-RELA and TALE-DDX17 were comparable to that measured in cells transiently transfected with a Tax expression vector ( Fig. 4d and Supplementary Fig. 4A). Although less quantitative, a nested RT-PCR assay clearly confirmed these results (Supplementary Fig. 5b). Strikingly, however, the mutant TALE-DDX17_K142R (a DDX17 helicase mutant 33,[41][42][43] failed to increase the levels of exon v10 inclusion, despite a clear chromatin enrichment of the DDX17 mutant ( Fig. 6c and Supplementary Fig. 5b). Collectively, these results demonstrate that RELA binding in the vicinity of genomic exons recruits the RNA helicase DDX17, which positively regulates the inclusion rate of the target exon due to its RNA helicase activity. Of note, this causal relationship between RELA, DDX17, and splicing regulation was observed irrespectively of the expression of Tax. Indeed, TALE-DDX17 rescued the inclusion of exon v10 in cells expressing the NF-κB defective mutant M22 (Fig. 6d). In addition, Tax-independent activation of NF-κB in HEK cells (using either TNFα or Phorbol 12-myristate 13-acetate (PMA)) mimicked the effects of Tax on splicing regulation (Fig. 6e). More importantly, ectopic RELA expression had similar effects, in a dose-dependent manner (Fig. 6f). Further, we observed a dynamic induction of RELA:DDX17 interactions after RELA expression or PMA activation ( Supplementary Fig. 5c), but not after TNFα activation, likely due to a lower level of RELA nuclear translocation in this condition ( Supplementary Fig. 5d). Collectively, these results suggest that the mechanistic role of Tax is to promote DDX17-dependent splicing regulation by promoting constitutive activation of NF-κB pathway.

Discussion
Since the finding of splicing dysregulations in HTLV-1-infected individuals 4,24,27,44 , deciphering how HTLV-1 interferes with the splicing regulatory network has become a new challenging issue for improving our knowledge of HTLV-1 infection and its associated diseases. Here, we provide the molecular evidence that upon Tax-induced NF-κB activation, RELA directly regulates splicing by binding to gene bodies at the vicinity of GC-rich exons and by locally recruiting the splicing factor DDX17, which regulates splicing via its RNA helicase activity.
Our results demonstrate that Tax deeply impacts alternative splicing independently from its effects on transcription. In addition, Tax-regulated exons were found in transcripts enriched in functional pathways that are distinct from those enriched by Tax transcriptional targets, suggesting that splicing reprogramming may constitute an additional layer of regulation by which HTLV-1 modifies the host cell phenotype. Supporting this, we also showed that the Tax-induced splicing variant CD44 v10, which was previously identified in circulating blood of HAM/TSP patients 27 and confirmed here ex vivo in infected CD4 + T-cell clones, contributes to modulating cell-adhesion affinity in vitro. GO analyses of Tax splicing targets also identified the GO terms phosphatidylinositol signaling system and inositol phosphate metabolism, two processes that are particularly connected to NF-κB signaling and that have critical roles in oncogenesis and disease progression of malignant diseases, including ATLL 45,46 . Furthermore, exon ontology analysis predicted that critical changes at the protein level would affect the experimentally validated protein structure and post-translational modifications, which together are likely to affect the connectivity network between proteins, and subsequently contribute to modifying cell phenotypes. This suggests that, beside its transcriptional effects, splicing regulatory functions of Tax might contribute to its oncogenic properties. Indeed, a large number of Tax-regulated exons could be observed in ATLL samples, which rarely express Tax but typically exhibit NF-κB addiction for survival and proliferation 24,26,47 .
At the molecular level, we showed that increased chromatin occupancy of RELA upon Tax expression is not restricted to promoter regions but also occurs in the vicinity of exons that are regulated at the splicing level (Figs. 4 and 5). Exons regulated by Tax, especially those localized within 1 kb of intragenic RELAbinding sites, are characterized by a high GC-content, a typical feature of exons regulated by the DDX5 and DDX17 RNA helicases 33 (Supplementary Fig. 4E, F). Accordingly, we found that a majority of Tax-regulated exons depend on the expression of these proteins (Fig. 3c). A local chromatin recruitment of DDX17 and RELA was validated on several Tax-regulated exons (Figs. 4  and 5). More importantly, we identified a confident causal relationship between exon tethering of RELA, local recruitment to chromatin of DDX17, and subsequent splicing regulation via DDX17 RNA helicase activity (Fig. 6). This catalytic activity of DDX17 was strictly required for its splicing regulatory functions (Fig. 6), as previously reported 33 . Indeed, the RNA helicase activities of DDX5 and DDX17 have been implicated in resolving RNA structures, facilitating the recognition of the 5′ splice site (which can be embedded in secondary structures), and exposing RNA-binding motifs to additional splicing regulators 33,42,[48][49][50] . However, even though some RNA-binding specificity has been reported for DDX17 51,52 , these RNA helicases are devoid of a proper RNA-binding domain, and their activity in splicing likely depends on additional factors that are able to provide target specificity. Here, we suggest that RELA is a DDX17 recruiter, acting as a chromatin anchor for DDX17 in the vicinity of exons dynamically selected upon NF-κB activation. While performing complementary experiments, we notified that siRNA-DDX5/17 affected the RELA chromatin occupancy of both exons and promoters regulated by Tax (Supplementary Fig. 6). This is reminiscent of recent report indicating that DDX5/17 depletion affects interactions between the transcription factor REST and DNA, defining DDX5/17 as REST transcriptional coregulators 22 . Altogether, these data indicate that the RELA-DDX17 chromatin interplay relies on complex mechanisms that deserve future investigations.
The target specificity of NF-κB factors remains a complex question. It has been estimated that~30-50% of genomic RELAbinding sites do not harbor a typical NF-κB site, and only a minority of RELA-binding events have been associated with transcriptional change [16][17][18][19] , thereby indicating that neither having a consensus site nor significant NF-κB occupancy are sufficient criteria for defining RELA's target specificity. Here, we identified a typical NF-κB consensus motif at RELA-binding loci that are close to alternatively spliced exons but we also uncovered that weak splice sites, low MFE, and significant GC-content bias of exons likely contribute to RELA's target specificity. Because low MFE and high GC-content confer a high propensity to form stable RNA secondary structures, the recognition and the selection of such GC-rich exons with weak splice sites by the splicing machinery typically depend on the RNA helicases DDX5/17 33 . Based on these observations, we propose a model of RELAinduced splicing target specificity (Fig. 6g), whereby, upon NF-κB activation, RELA binds to intragenic-binding consensus motifs and locally recruits DDX17. When the RELA:DDX17 complex is located in close proximity of GC-rich exons flanked by weak splice sites, DDX17 can increase their inclusion rate by unwinding GC-rich secondary structures of the nascent RNA transcript, and potentially also by unmasking binding motifs for additional splicing regulators. Of note, although Tax was identified in both RELA-and DDX17-containing complexes, we demonstrated that Tax-independent NF-κB activation is sufficient for promoting interactions between RELA and DDX17 and the corresponding splicing regulation. Without ruling out possible additional effects of Tax on the RELA:DDX17 complex, such as favoring dimer formation (as already proposed for NF-κB dimers 15 ), these data indicate that Tax exacerbates a dynamic and physiologic process involving RELA in splicing target specificity.
In conclusion, our results provide conceptual advances for understanding how cell signaling pathways may drive target specificity in splicing by dynamically recruiting cognate transcription factors at the vicinity of target exons that act as chromatin anchor for splicing regulators. In the context of NF-κB signaling, such a mechanism likely has a significant impact on cell fate determination and disease development associated with HTLV-1 infection and other situations linked to chronic NF-κB activation, such as human inflammatory diseases and cancer.

Methods
Cell culture and transfections. Peripheral blood mononuclear cells (PBMCs) were obtained by Ficoll separation of whole blood from HTLV-1-infected individuals in the context of a Biomedical Research Program approved by the Committee for the Protection of Persons, Ile-de-France II, Paris (2012-10-04 SC). All individuals gave informed consent. PBMCs were cloned by limiting dilution in RPMI 1640 supplemented with penicillin and streptomycin, sodium pyruvate, non-essential amino acids, 2-mercaptoethanol, 10% filtered human AB serum, 100 U/ml recombinant IL-2 (Chiron Corporation), and 75 μM HTLV-1 integrase inhibitor L-731,988. Clones were phenotyped by flow cytometry using antibodies against CD4 (Dako-Cytomation) and isotype-matched controls on a FACScan system using CellQuest software (Becton Dickinson). HTLV-1-positive clones were assessed by PCR. The human embryonic kidney 293T-LTR-GFP cells 53 , which contain an integrated GFP reporter gene under the control of the Tax-responsive HTLV-1 LTR, were cultured in DMEM + glutamax medium supplemented with 10% heat-inactivated FBS and 1% penicillin/streptomycin. This cell line was used to measure transfection efficiency in Tax and TaxM22 conditions. In standard transfection experiments, siRNAs (Supplementary Data 5) and/or expression vectors (pSG5M empty, pSG5M-Tax-WT, pSG5M-M22) were mixed with JetPrime (Polyplus Transfection) following the manufacturer's instructions, and cells were harvested 48 h after transfection. TNFα exposure consisted in treating cells with 10 ng/ml of TNFα for 24 h. HTLV-1-chronically infected lymphocytes ATL2 (kind gift from Masao Matsuoka (Kyoto University, Kyoto, Japan) and Roberto Accolla (Università degli Studi dell'Insubria, Varese, Italia)), C91PL (kind gift from Cynthia Pise-Masison (National Cancer Institute, NIH, Bethesda, MD)) and Renaud Mahieux (Center for Research in Infectious Diseases, Lyon, France) and non-infected MOLT4 cells (CRL-1582, ATCC) were grown in RPMI 1640 medium (Gibco, Life Technologies) supplemented with 10% heat-inactivated fetal calf serum, 20 IU/ml penicillin, 20 μg/ml streptomycin, and 25 mM HEPES.
Cell-adhesion assays. Culture plates were prepared by coating with 40 µg/ml hyaluronic acid from human umbilical cord (Sigma) and 25 µg/ml type IV collagen from human placenta (Sigma) overnight at 4°C. Non-specific binding sites were blocked for 1 h with PBS containing 1 mg/ml heat-denatured BSA. After three washes with 1× PBS, 5 × 10 4 cells transiently transfected with pSG5M-Tax vector or its empty control were added at 48-h post-transfection. Cell adhesion was allowed to proceed for 20 min at room temperature. Non-adherent cells were removed with three washes with 1× PBS, and adherent cells were quantified. All experiments were done in triplicate.
Ontology analysis were performed using DAVID software for gene ontology and Exon Ontology v1.5.0 36 .
In silico screening of NF-κB-responsive elements in the CD44 promoter sequence was carried out using the PROMO database (based on TRANSFAC v8.3) 58 . The MEME-ChIP suite was used to discover the regulatory motifs in the NF-κB ChIP-seq data 39 .
To predict splice site strengthes, scores were computed using MaxEntScan 59 for the sequence covering both sides of the splicing site (using 3 bases into the exon and 6 bases into the intron for 5′ splice sites, and 20 bases into the intron and 3 bases into the exon for 3′ splice sites). MaxEntScan uses Maximum Entropy Models (MEMs) to compute log-odds ratios. The minimum free energy was computed from exon-intron junction sequences using RNAFold from the ViennaRNA package (v 2.4.1; http://rna.tbi.univie.ac.at/cgi-bin/RNAWebSuite/RNAfold.cgi). Analyzed sequences include 25 nucleotides within the intron and 25 nucleotides within the exon. The GC-content was calculated for exons defined in FasterDB 54 .
The distribution of RELA peaks across alternative and constitutive exons, and the average distance between RELA peaks and Tax exon targets was measured using ChiPseq datasets from GEO 60 , ENCODE 61 , and CISTROME 62 databases: from GEO GSE63736, GSM1239484, GSM486271, GSM486293, GSM486298, GSM486318, GSM847876, GSM847877, GSM2394419, GSM2394421, GSM2394423, from ENCODE ENCFF002CPA, ENCFF002CQB, ENCFF002CQJ, ENCFF002CQN, ENCFF580QGA, and from CISTROME 53597, 5388, 5389, 4940, 36310, 36316, 4971. For another GEO dataset, GSM2628088, reads were mapped to the hg19 build of the human genome with Bowtie2 63 and RELA peaks were identified with Macs2 64 . Alternative and Constitutive spliced exons were obtained from FasterDB 54 . To focus on intragenic RELA peaks, we used the bedtools 65 intersect command to remove all intergenic RELA peaks and all RELA peaks localized on first exon (or at least at less than 500 nt distance) for each gene. A Perl script was specifically created to measure the distance between RELA peaks and Tax-regulated exons. Briefly, RELA peaks and exons are provided as BED files, and the script reports for each exon the distance in nucleotides of the nearest RELA peak. Closest peak distances from the 710 Tax-regulated exon-cassettes were compared to closest peak distances from 710 exons chosen by chance (10 5 runs). Using bootstrap analysis, the random samples were compared with the set of Tax-regulated exons, and the normal distribution of these counts was used in a sample t-test to assess the significance of the RELA-binding sites enrichment at the vicinity of Tax-regulated exons. p = 2e − 4 was considered significant.
TALE design and construct. The TALE constructs were obtained from Ther-moFisher Scientific. TALEs were constructed using the Golden Gate Assembly method 40 . The RVDs HD, NI, NG, and NN were chosen to specifically recognize the nucleotides C, A, T, and G, respectively. The TALE-targeting CD44 v10 sequence was 5′-TCCAACTCTAATGTCAATC-3′. This TALE construct was fused to a V5 sequence and a SV40 NLS at its 5′ end and cloned in the NotI-HindIII fragment of the pXJ41 backbone plasmid. DDX17-WT and DDX17-K142R cDNA were obtained by PCR from pcDNA3-HA-DDX17 and pcDNA3-HA-DDX17-K142R and were cloned in the HindIII-BglII fragment in the MCS downstream to the TALE sequence.
CRISPR design and construct. The sequence-specific sgRNA for site-specific interference of genomic targets were designed using CRISPRseek R package, and sequences were selected to minimize off-target effect 66 . Two complementary oligonucleotides were annealed and cloned into BbsI site of pSpCas9(BB)-2A-Puro (PX459) V2.0 (Addgene plasmid #62988) 67 for co-expression with Cas9 using 5U of T4 DNA ligase, T4 DNA ligase buffer (1×) (Roche). 293T-LTR-GFP cells were transfected with the mix of equimolar ratio of PX459-sgRNA1 and PX459-sgRNA2 (Supplementary Data 5). At 24-h post-transfection, the medium was changed, 1 μg/ml puromycin was added for selection, and cells were cloned by serial dilution method.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
RNA-seq data produced in this study have been deposited on NCBI GEO under the accession number GSE123752. The source data underlying Figs. 1c, 2b-f, 3a, b, d, 4a, c-f, 5a-d, 6a-f, and Supplementary Figs. 1, 2b-d, 3a, b, e, 4a-g, 5a-d, and 6a-c are provided as a Source Data file. All data supporting the findings of this study are available within the article and its supplementary information files and from the corresponding author upon reasonable request. Source data are provided with this paper.

Code availability
FaRline is publicly available (http://fasterdb.ens-lyon.fr/FARLINE.tgz). Other custom codes used in this study are available from the corresponding author upon reasonable request. Source data are provided with this paper.