Arginine methylation of DDX5 RGG/RG motif by PRMT5 regulates RNA:DNA resolution

Aberrant transcription-associated RNA:DNA hybrid (R-loop) formation often lead to catastrophic conflicts during replication resulting in DNA double strand breaks and genome instability. To prevent such conflicts, these hybrids require dissolution by helicases and/or RNaseH. Little information is known about how these helicases are regulated. Herein, we identify DDX5, an RGG/RG motif containing DEAD-box family of RNA helicase, as a crucial player in R-loop resolution. We define at the mechanistic level the function of DDX5 in R-loop resolution. In vitro, recombinant DDX5 resolves R-loops in an ATP-dependent manner leading to R-loop degradation by the XRN2 exoribonuclease. DDX5 deficient cells accumulated R-loops at loci known to form R-loops using RNA:DNA immunoprecipitation (DRIP)-qPCR and increased RNaseH sensitive RAD51 foci. PRMT5, an arginine methyltransferase, associated with DDX5 and methylated its RGG/RG motif. This motif was required to associate with XRN2 and resolve cellular R-loops. Furthermore, PRMT5 deficient cells accumulated R-loops, as detected by DRIP-qPCR resulting in increased gH2AX foci. Our findings define a new mechanism by which an RNA helicase, DDX5, is modulated by arginine methylation to resolve R-loops.


Introduction
In mammalian cells, there are nine protein arginine methyltransferases (PRMTs) responsible for transferring methyl groups from S-adenosyl-methionine to the nitrogen atoms of arginine (Bedford & Clarke, 2009). Arginine methylation modulates multiple biological processes from gene expression, pre-mRNA splicing, the DNA damage response and signal transduction (Blanc & Richard, 2017). PRMT5 is the major type enzyme catalyzing symmetrical dimethylarginines and as such multiple responses have been observed with inhibition of its function including the induction of the p53 response, DNA damage and cell death. PRMT5 regulates gene expression by methylating histones and transcription factors (Cho, Zheng et al., 2012, Friesen, Massenet et al., 2001, Pal, Vishwanath et al., 2004. Methylation of the Sm proteins regulates the p53 response by modulating the alternative splicing of MDM4 and MDM2 (Bezzi, Teo et al., 2013).
Little is known of the post-translational modifications regulating R-loop formation and resolution. It is known that the pausing of RNA polymerase II (RNA PolII) increases DDX23 phosphorylation by SRPK2 enhancing R-loop suppression (Sridhara et al., 2017). Acetylation of DDX21 by CBP regulates its helicase activity (Song et al., 2017). Methylation of RNA PolII subunit POLR2A by PRMT5 regulates Senataxin recruitment at transcription termination regions (Zhao, Gish et al., 2016), while methylation of TDRD3 by CARM1 regulates the recruitment of topoisomerase IIIB at c-Myc locus preventing negative supercoiling and R-loops (Yang, McBride et al., 2014). In this study, we define a new role for arginine methylation in the regulation of R-loops. We show that the methylation of the RNA helicase DDX5 by PRMT5 regulates its association with XRN2 to suppress R-loops and maintain genomic stability.

DDX5 deficiency leads to an increase in RNaseH sensitive RAD51 foci.
A genome-wide screen identified RNA binding proteins including the DEAD box helicase DDX17 to localize to DNA damage sites (Adamson, Smogorzewska et al., 2012), however its function in the DNA double strand break (DSB) repair was not determined. DDX17 is often found in a complex with its ortholog DDX5 (Fuller-Pace, 2013). To investigate whether these RNA helicases play a role in the DNA damage response (DDR) pathway, we initially examined whether U2OS cells deficient in either helicase had altered ionizing radiation (IR)-induced DNA damage foci. Depletion of DDX5, but not DDX17, by siRNA transfection led to a significant increase of RAD51, RPA2 and BrdU foci at different time points after IR treatment ( Figure 1A-C and S1A-D), while, DDX5 knockdown had little effect on γH2AX or 53BP1 foci formation ( Figure S1E). In more details, we found that each RAD51 focus was much brighter (increasing from 1 to 4h) than in siCTL cells ( Figure 1B and data not shown), suggesting that longer RAD51 filaments were forming next to each DSB. Moreover, there were more foci per cell at each time point, suggesting that without DDX5, RAD51 accumulated at lesions without completing repair.
Next, we examined whether the helicase activity of DDX5 was required for RAD51 foci accumulation. DDX5 deficient cells were transfected with siRNA resistant expression vectors encoding FLAG epitope tagged wild type and helicase dead DDX5. Wild type DDX5, but not FLAG-DDX5 DEAD mutant, significantly restored RAD51 foci to lower levels, as observed in siCTL cells ( Figure S2A, S2B), demonstrating that the DDX5 RNA helicase activity is required to maintain proper RAD51 foci dynamics.
Many RNA helicases have been identified to regulate R-loops and it is known that a proportion of RAD51 foci are sensitive to RNaseH treatment (Wahba, Gore et al., 2013), as they are preferentially localized in transcriptionally active regions (Aymard et al., 2017). To test whether the defect of DDR in DDX5-deficient cells is associated with DDX5 function in the resolution of RNA:DNA hybrids, we investigated the effect of RNase H1 overexpression on IR-induced RAD51 foci formation. RNase H1 digests the RNA strand in the RNA:DNA hybrid, revealing Rloop dependent RAD51 foci (Wahba et al., 2013). Following siRNA control or DDX5 knockdown, U2OS cells were transfected with HA-tagged RNase H1 and visualized by indirect immunofluorescence with anti-HA (red) and anti-RAD51 or anti-53BP1 (green) antibodies. The data was analyzed in both HA-RNase H1-positive and negative populations, respectively. In the HA-RNase H1-negative population (RNase H1-), DDX5 depletion led to significant increase of the percentage of cells with more than 10 RAD51 foci, but not 53BP1 foci ( Figure 1D and 1E).
However, in the HA-RNase H1-transfected cells (RNase H1+), DDX5 depletion had no effect on RAD51 foci compared to control siRNA-transfected cells ( Figure 1D and 1E, siCTL, RNase H1+). These results suggest that the defect of DSB response and repair in DDX5-deficient cells is associated with increased RNA:DNA hybrid formation.

DDX5 resolves R-loops in vitro and in vivo.
DDX5 is an RGG/RG motif containing helicase previously shown to unwind RNA/RNA and RNA:DNA duplexes (Hirling, Scheffner et al., 1989, Rossler, Straka et al., 2001, Xing, Wang et al., 2017. Dbp2, the S. cerevisiae homolog of DDX5, was shown to resolve RNA:DNA hybrids in the context of R-loops (Cloutier, Wang et al., 2016), however, whether DDX5 shares this Rloop resolving activity is unknown. To examine whether DDX5 resolves R-loops, we purified recombinant DDX5 to homogeneity ( Figure 2A) and performed in vitro R-loop (RNA:DNA hybrid) and D-loop (DNA:DNA hybrid) unwinding activity assays using radiolabeled nucleotide substrates. The addition of increasing DDX5 doses led to the appearance of a faster migrating species on native gels representing the DNA strands without the bound RNA fragment and this occurred in an ATP-dependent manner ( Figure 2B). Remarkably, DDX5 did not resolve the Dloop substrate ( Figure 2B). These observations show that DDX5, like its yeast homolog Dbp2, resolves R-loops in vitro.
To examine whether DDX5 resolves R-loops in vivo, we generated DDX5-deficient U2OS cells using three siRNAs and we measured the accumulation of RNA:DNA hybrids by immunofluorescence using the monoclonal antibody (S9.6) known to recognize RNA:DNA hybrids within the mitochondria, nucleoli, and the nucleoplasm (Bhatia et al., 2014, Ginno, Lott et al., 2012, Hamperl, Bocek et al., 2017, Hodroj et al., 2017, Sollier et al., 2014. We measured the S9.6 signal in the nucleoplasm as the total nuclear signal subtracting the nucleolar contribution. The nucleus and nucleolus were detected with DAPI and anti-nucleolin antibodies, respectively ( Figure S3A). Depletion of DDX5 led to a significant increase of the S9.6 signal in the nucleoplasm of U2OS cells with all siRNAs tested ( Figure 2C). We also visualized the accumulation of RNA:DNA hybrids from isolated genomic DNA using slot-blot analysis with the S9.6 antibody in the presence or absence of RNase H. Again, we observed a significant increase in the S9.6 signal within the genomic DNA isolated from DDX5-deficient cells compared to control siRNA treated cells. siSenataxin (siSTEX) was used as a positive control ( Figure S3B, S3C).
To examine the role of the DDX5 in R-loop resolution at precise genomic loci, we used the RNA:DNA immunoprecipitation (DRIP) method. By qPCR analysis we quantified R-loop accumulation at selected six specific loci: HIST1H1E, NEAT1, RPPH1, EGR1, MALAT1 and HIST1H2BG, previously known to have a propensity to form R-loops (García-Rubio et al., 2015, Hodroj et al., 2017, Yang et al., 2014. The knockdown of DDX5 resulted in a significant increase of R-loops when compared to control siRNA conditions and these were sensitive to RNase H treatment ( Figure 2D). These findings suggest that DDX5 deficient cells accumulate Rloops.
DDX5 deficiency leads to spontaneous DNA damage and hypersensitivity of U2OS cells to replication stress.
As unresolved R-loops result in DNA damage (Skourti-Stathaki et al., 2011), we next investigated whether DDX5-deficient cells exhibit spontaneous DNA damage due to unresolved R-loops and indeed this was the case, as assessed by an increase in phosphorylation and foci formation of histone variant H2AX (termed gH2AX), a marker of DNA damage ( Figure S4A and S4B). Cells deficient in DDX5 are known to be sensitive to ionizing radiation (Nicol, Bray et al., 2013), however it is unknown whether these cells are sensitive to DNA replication stress such as hydroxyurea, as would be predictive of R-loop accumulation during DNA replication. Indeed, depletion of DDX5 using two different siRNAs led to a significant reduction of cell survival to that of hydroxyurea treatment compared with the control siRNA-transfected cells ( Figure S4C).
To further confirm this effect, we also performed a FACS (fluorescence-activated cell sorting)based survival analysis of co-cultured cells ( Figure S4D). This analysis enables direct comparison in the same culture of the proliferative fitness of DDX5-expressing and depleted cells. The U2OS cells without or with stable GFP expression were transfected with control and DDX5 siRNAs, respectively. The cells were mixed and co-plated two days after transfection and then treated with hydroxyurea. Compared to non-treated cells, a decrease in the percentage of GFP-positive cells after treatment indicates an increase in sensitivity to hydroxyurea in the target siDDX5-transfected cells ( Figure S4E). Depletion of DDX5 caused significant increase in cell sensitivity to hydroxyurea, either at low dose (0.4 mM) with long-time incubation (24h) or at high dose (10 mM) and short exposure (2h; Figure S4F, S4G). Taken together, these results suggest that DDX5 deficiency causes cell hypersensitivity to the DNA replication stress agent hydroxyurea.

DDX5 is arginine-methylated in U2OS cells
Interestingly, in a separate study, we identified DDX5 by mass spectrometry analysis as a PRMT5-interacting protein in two prostate cancer cell lines (Table S1). PRMT5 is known to catalyze the mono-and symmetrical dimethylation of arginine residues in proteins (Branscombe, Frankel et al., 2001). We first confirmed the physical association between PRMT5 and DDX5 in U2OS cells. Cells were lysed and either PRMT5 or DDX5 was immunoprecipitated followed by immunoblotting. Significant amounts of DDX5 was co-immunoprecipitated with anti-PRMT5 antibodies, but not control IgG ( Figure 3A). The reverse was also observed, PRMT5 was coimmunoprecipitated with anti-DDX5 antibodies ( Figure 3B). These findings confirm the association of PRMT5 with DDX5.
Next, to determine whether DDX5 is arginine methylated, we first immunoprecipitated DDX5 from U2OS cells and analyzed it for the presence of methylarginines by mass spectrometry. We identified R502 within the C-terminal RGG/RG motif of DDX5 to harbor a monomethyl ( Figure   3C). U2OS cells transfected with Flag-epitope tagged DDX5 were immunoprecipitated with anti-Flag antibodies and the bound proteins immunoblotted with anti-methylarginine antibodies. We detected that Flag-DDX5 was arginine methylated ( Figure 3D). RGG/RG motifs are methylated mainly by PRMT1 and PRMT5, the two major enzymes responsible for cellular protein arginine methylation (Blanc & Richard, 2017). To identify which PRMT leads to R-loop accumulations, we turned to R-loop visualization using immunofluorescence assays with S9.6 antibody in siPRMT1 or siPRMT5 U2OS cells. We observed that the knockdown of PRMT5, but not PRMT1, led to significant increase of nuclear S9.6 staining in U2OS cells ( Figure 3E). DDX5 siRNA-1 was used as a positive control.
Similarly, DRIP-qPCR analysis demonstrated that depletion of PRMT5 caused significant RNA:DNA hybrid accumulation, as observed with siDDX5 ( Figure 3F) and this lead to a significant increase in gH2AX foci ( Figure 3G). We next performed in vitro arginine methylation analysis using glutathione-S-transferase (GST)-DDX5 fusion proteins to identify the methylation region. DDX5 has two RGG/RG motifs: One located at its N-terminus and the other at its C-terminus. Interestingly, both DDX5 RGG/RG motifs are conserved in the yeast homolog Dbp2, although the rest of the C-terminal region is truncated in Dbp2 ( Figure 4A). We generated three truncation mutants of DDX5, including the N-terminal region (residues 1-100; F1), the central catalytic enzyme domain (92-471; F2) and the C-terminal region (residues 466-614; F3; Figure 4B). Only the C-terminal region (F3), encompassing the RGG/RG motif, was methylated by PRMT5 ( Figure 4C). We then substituted DDX5 R478, R482, R484, R486, R502 within the RGG/RG motif with lysines in a smaller region 466-555; F4). The 5R to 5K mutation within DDX5 (RK) completely abolished DDX5 arginine methylation by PRMT5 ( Figure 4D).

DDX5 helicase activity and RGG/RG motif is required for R-loop resolution.
Next, we examined whether R to K substitutions of DDX5 affected its R-loop resolution function. Interestingly, ectopic expression of DDX5-RK as well as the catalytically inactive mutant (DDX5-XD) caused significant increase of R-loop accumulation in U2OS cells, as determined by S9.6 immunofluorescence ( Figure 4E). These findings suggest that the ectopic expression of DDX5-RK and DDX5-XD likely behave as dominant negatives. We further investigated whether WT DDX5, DDX5-RK, and DDX5-XD can rescue the R-loop accumulation observed in DDX5 deficiency. The ectopic expression of WT DDX5 reversed the R-loop accumulation observed in siDDX5 cells, however, this was not observed with the expression of DDX5-RK ( Figure 4F). Taken together, these results suggest that the RGG/RG motif is required for R-loop resolution.

DDX5 associates with known R-loop regulatory proteins.
To identify interacting partners of DDX5 which may function in R-loop metabolism, we performed stable isotope labeling with amino acids in cell culture (SILAC)-based mass spectrometry analysis. U2OS cells expressing Flag-epitope tagged DDX5 were grown in the 'heavy' mediurm and the control U2OS cells in light medium ( Figure S5A). Many RNA binding proteins belonging to the heterogeneous nuclear ribonucleoproteins (hnRNPs) and DEAD/DEAH-box families were identified (Table S2 and Figure S5B). These included known DDX5 interactors including DHX9, DDX3X, and DDX17 (Choi & Lee, 2012, Ogilvie, Wilson et al., 2003, Wilson & Giguere, 2007. Interestingly, we also identified proteins known to influence R-loops resolution: DDX1, DHX9, XRN2, SRSF1 and Aly/REF an exon junction complex protein (Table S2). PRMT5 was also identified, but since PRMT5 is a contaminant of anti-FLAG immunoprecipitates, we could not claim association with DDX5 based on this approach.

RGG/RG motif of DDX5 mediates association with XRN2
XRN2 has been shown to function with helicases such as Senataxin (Cristini et al., 2018, Skourti-Stathaki et al., 2011 and DHX9 (Cristini et al., 2018) and is required to resolve R-loops (Morales et al., 2016). We hypothesized that XRN2 also assists DDX5 with R-loop resolution of a subset of genes. We first confirmed the interaction between the two proteins. U2OS cells expressing Flag-DDX5, were immunoprecipitated with anti-Flag antibodies and the presence of XRN2 was detected by immunoblotting ( Figure 5A), confirming our mass spectrometry data.
We next mapped the region of DDX5 required to associate with XRN2. DDX5 is composed of a central helicase domain with N-and C-terminal RGG/RG motifs ( Figure 4A). The XRN2 interaction region was mapped between amino acids 435 and 554 of DDX5, where the Cterminal RGG/RG motif resides ( Figure 5C). We next determined whether mutation of these arginine residues within the RGG/RG motif affected recognition with anti-methylarginine antibody and interaction with XRN2. Indeed, Flag-DDX5:RK was not recognized with the antimethylarginine antibody and had reduced association with XRN2, unlike wild type DDX5 ( Figure 5D). These findings suggest that arginine methylation of the RGG/RG motif of DDX5 regulates the interaction with XRN2.

DDX5 and XRN2 function together in R-loop resolution
To test whether DDX5 and XRN2 function together in R-loop resolution, we performed DDX5 R-loop resolving in vitro assays in the presence or absence of purified XRN2 ( Figure 5E). The presence of recombinant XRN2 (20 nM) increased the RNase product (i.e. degraded RNA) component of the RNA:DNA hybrid unwound by DDX5 ( Figure 5F). However, XRN2 did not influence the overall ability of DDX5 to resolve R-loops, as observed by the generation of the unwound product which was the same in the absence or presence of XRN2 ( Figure 5F). These results show that XRN2 degrades the RNA released by DDX5 in vitro.
The activity of the XRN2 exonuclease is required to promote transcriptional termination of a subset of genes (Kim, Krogan et al., 2004, West, Gromak et al., 2004. Given the requirement of XRN2 for efficient termination, we examined whether DDX5 might play a role in coupling Rloops resolution for transcription termination of certain genes. To test this possibility, we performed ChIP assays to measure XRN2 occupation along the β-actin gene in the DDX5 depleted cells. HEK293 cells transfected with siCTL or siDDX5 showed XRN2 distribution throughout β-actin, as previously shown (Kaneko, Rozenblatt-Rosen et al., 2007). However, a significant decrease of XRN2 occupancy was observed in siDDX5 depleted cells ( Figure 5G

Discussion
In this study, we define a new function for DDX5 in R-loop resolution and reveal a regulatory role for the methylation of its RGG/RG motif in modulating its function. Depletion of either DDX5 or PRMT5 in cellulo lead to increased accumulation of RNA:DNA hybrids, as detected with the anti-S9.6 antibody by immunofluorescence, slot blot analysis and DRIP-qPCR. The Cterminal RGG/RG motif of DDX5 was methylated by PRMT5 and required for R-loop resolution, as amino acid substitution of its arginines in the RGG/RG motif with lysines (RK) failed to reverse the accumulation of R-loops in siDDX5 cells. Purified human DDX5 resolved R-loop structure in vitro and its catalytically inactive mutant failed to rescue the effect of DDX5 deficiency on R-loop accumulation, suggesting that DDX5 represses R-loop accumulation by directly resolving the RNA:DNA hybrid structures. Moreover, proteomic analysis revealed that DDX5 associates with many RNA binding proteins including helicases, as well as with XRN2, an exonuclease known to function in transcriptional termination (Kim et al., 2004, West et al., 2004. This is in line with a few recent studies for high-through-put identification of R-loop associated proteins, in which DDX5 was identified as a top hit of R-loop associated proteins (Cristini et al., 2018, Wang, Grunseich et al., 2018, although DDX5 function was not addressed in these studies. We show that one mechanism of action of DDX5 is to mediate regulated interaction with XRN2. DDX5-deficient cells harbored less XRN2 at the b-actin transcriptional termination site, accompanied with an increase of RNA pol II accumulation. These data are consistent with pausing and R-loop accumulation and DDX5 functioning to recruit XRN2 at certain genomic regions in a PRMT5 dependent manner. It is known that p54nrb and PSF function to recruit XRN2 at transcription termination sites (Kaneko et al., 2007), therefore, DDX5 may function with these factors for optimal XRN2 recruitment.
Consistent with DDX5 function in R-loop repression, DDX5 knockdown lead to an increase number and a sustained presence of RNase H-sensitive RAD51 foci in response to IR-induced DSBs. Recently, a few studies suggest that HR proteins, such as RAD51, RAD52 and RPA1 are selectively recruited to transcriptionally active DNA damage sites (Aymard et al., 2017).
Genome-wide mapping reveals that RNA:DNA hybrids accumulate at DSB sites and Senataxin can remove DSB-induced RNA:DNA hybrids (Cohen, Puget et al., 2018). The increase of IR-induced RAD51 foci in DDX5-deficient cells is likely associated with an increase of RNA:DNA hybrid accumulation surrounding DSBs as DDX5 has a similar function as Senataxin for R-loop removal. Transcription-associated R-loop formation at DSBs frees non-template single-stranded DNA (ssDNA) which may facilitate recruitment of ssDNAbinding proteins such as RPA and RAD51 to DSBs and HR repair. Alternatively, R-loop removal and formation at DSBs may be associated with DSB-induced transcription silencing (Shanbhag, Rafalska-Metcalf et al., 2010) or activation (Francia, Michelini et al., 2012), which in turn determine DSB repair pathway choice. For example, DSB-induced transcription silencing may facilitate NHEJ protein recruitment and DNA end ligation, and the transcription activation may facilitate RNA:DNA accumulation and HR protein recruitment as well as RNA-mediated HR repair.
The methylation of the RGG/RG motif of DDX5 likely serves additional functions besides recruiting XRN2 and this is especially pertinent in R-loops that are not associated with transcriptional termination regions. Furthermore, the function we ascribe to the methylation of DDX5 probably functions in a coordinate fashion with the methylation of other substrates such as RNA polII subunit POLR2A to regulate R-loops suppression (Zhao et al., 2016). Besides DDX5, another DEAD box helicase, DDX21 recently shown to regulate R-loops (Song et al., 2017) harbors an RGG/RG motif and may be regulated in a similar manner as DDX5. As Tudor domains are known to interact with methylated arginines (Bedford & Clarke, 2009), we do not know the protein interface that XRN2 uses to interact with methylated DDX5. Of note, XRN2 itself also contains an RGG/RG motif, however, we do not know its function.
Why so many helicases are required to resolve R-loops? Certain RNA helicases may function on different types of R-loops or regulate R-loops for a particular process or at different genomic locations. For example, DDX1 converts G4 RNA structures to R-loops for IgH class switch recombination (Ribeiro de Almeida et al., 2018). DDX1 also functions in tRNA, mRNA and microRNA processing, transport of RNAs from nucleus to cytoplasm, and AU-rich elementmediated mRNA decay in cytoplasm. Its function in R-loop resolution facilitates homologous recombination repairs (Li et al., 2016). DDX19 is well-known for its functions in mRNA nuclear export at the nuclear pore. A role for DDX19 in R-loop suppression has been reported, suggesting a functional link between mRNA nuclear export and R-loop clearance (Hodroj et al., 2017). DDX23 is a component of the spliceosomal U5 small nuclear ribonucleoprotein (U5 snRNP) and required for integration of U4/U6-U5 tri-snRNP into the splicesome. However, the role of DDX23 in suppressing R-loops does not require a functional U5 snRNP (Sridhara et al., 2017). DDX21 is known to be involved in multiple steps of ribosome biogenesis as well as mRNA transcription (Song et al., 2017).
PRMT5 deficient cells exhibit induction of the p53 response, DNA damage and cell death and as such PRMT5 is an interesting therapeutic target for many cancers. The upregulation of the p53 response was shown to be the result of an imbalance in the methylation of Sm proteins and other RNA binding proteins (e.g. SRSF) that regulate MDM4 and MDM2 alternative splicing (Bezzi et al., 2013). PRMT5 deficient cells also accumulate DNA damage leading to cellular senescence or cell death (Bezzi et al., 2013, Clarke et al., 2017, Hamard et al., 2018. What leads to the increased DNA damage in these cells is still not clear. Our observations herein suggest that PRMT5 defective cells fail to suppress R-loops leading to increase DNA damage and it may represent one the earliest response to PRMT5 deficiency. The DNA damage observed from the R-loop accumulation may also contribute to increasing the p53 response, as DNA damage is known to upregulate p53.

Reagents and Antibodies
Mouse anti-DDX5 monoclonal antibodies (A-5, sc-166167 and clone204, 05-850) were purchased from Santa Cruz Biotechnology and Millipore. Rabbit anti-RAD51 (H-92, sc-8349) antibody was purchased from Santa Cruz Biotechnology. Rabbit anti-53BP1 antibody (NB100-304) was from Novus Biologicals. Rabbit anti-nucleolin antibody (ab50279) and mouse anti-RPA2 monoclonal antibody (ab16850) was from Abcam. Anti-BrdU (B44) monoclonal antibody was from BD Biosciences. Anti-XRN2 antibodies (A301-102A and Ab72181) were purchased from Bethyl Laboratories Inc and Abcam (ab72181) for western blot, and from Proteintech The gBlock DNAs were synthesized by Integrated DNA Technologies (IDT). The catalytic inactive DDX5 mutant with replacement of both Arg403 at the motif Va and Arg428 at the motif VI with Leucine and the RK mutant with replacement of the five arginine residues with lysine at the RGG/RG motif were constructed by two-steps PCR using the siRNA-resistant DDX5 construct as template. The plasmids for expressing GST fusion proteins of DDX5 fragments were constructed by inserting PCR-amplified DDX5 cDNA fragments in pGEX-6P1 vector at Bam HI and Sal I sites. HA tagged RNase H1 was a kind gift from Dr. Yangzhong Yang (Beckman Research Institute, City of Hope, CA).

Cell Lysis and Immunoprecipitation
For co-immunoprecipitation experiments, cells were lysed with a lysis buffer containing 50 mM HEPES, pH 7.4, 150 mM NaCl, 1% Triton X-100 and a cocktail of protease inhibitors and phosphatase inhibitors. After high speed centrifugation, the supernatant was precleared and then incubated with antibodies for one hour and then protein-A or protein G agarose beads for another hour at 4ºC. In some experiments, the supernatant was incubated directly with agarose beads on which antibodies were pre-attached covalently. The beads were washed five times with lysis buffer for western blot analysis or four time with lysis buffer and then twice with PBS for mass spectrometry analysis.

Analysis
U2OS stable cells with Flag-DDX5 expression were grown in standard heavy (Lys4/Arg6) SILAC medium and the control U2OS cells in light medium (Lys0/Arg0) for ten days. Similarly, U2OS stable cells with Flag-DDX5 expression were grown in the heavy SILAC medium and the control U2OS cells in light medium for ten days. The cells were lysed and subjected to immunoprecipitation with anti-Flag antibody as described above. The beads with bound proteins were sent as described previously (Thandapani, O'Connor et al., 2013).

Immunofluorescence
For immunostaining with S9.6 and anti-nucleolin antibody, the cells were fixed with cold methanol for 10 minutes at room temperature. Slides were then washed three times with PBS containing 0.1% Tween-20 (PBST) and blocked with blocking buffer (3% bovine serum albumin in PBST) for 1 h at room temperature or overnight at 4°C. Slides were then incubated with antibodies S9.6 (1/200) and anti-nucleolin (1/1000) diluted in the blocking buffer for 2 hours.
Images were taken using a Zeiss M1 fluorescence microscope with 63X amplification and Zeiss LSM800 confocal system with 40X magnification where indicated..

Slot-blotting
Nucleic acids were extracted from U2OS cells by SDS/proteinase K treatment at 37°C overnight followed by phenol-chloroform extraction and ethanol precipitation. The nucleic acids were blotted onto Hybond-N Nylon membrane in duplicate using a Slot-Blot apparatus (Schleicher & Schuell). One half of the membrane was treated with 0.5 N NaOH, 1.5 M NaCl for 10 min to denature the DNA and neutralized for another 10 min in 0.5 M Tris-HCl buffer (pH7.0) containing 1 M NaCl. After UV-crosslinking (0.12 J/m2), the non-treated membrane was subjected to western blot analysis with S9.6, and the treated membrane, served as loading control, with the single-stranded DNA antibody. The S9.6 signal was normalized by the loading control.

DRIP (RNA:DNA Immunoprecipitation)-qPCR
DRIP assays were performed as described (Ginno et al., 2012). Briefly, nucleic acids were extracted from U2OS cells by SDS/proteinase K treatment at 37°C overnight followed by phenol-chloroform extraction and ethanol precipitation. The harvested nucleic acids were digested for 24 h at 37°C using a restriction enzyme cocktail (50 units/100 µg nucleic acids, each of BsrGI, EcoRI, HindIII, SspI, and XbaI)

ChIP (Chromatin Immunoprecipitation)
HEK293 cells were crosslinked with 1% formaldehyde at room temperature for 10 min. ChIP Pol II results, the relative signal was normalized to the IgG, and for ChIP XRN2 results, the signal was normalized to the Pol II.

FACS-Based Cell Survival Assay
For FACS-based cell survival assay, a U2OS cell line with stably transfected GFP (U2OS-GFP) was generated. The U2OS and U2OS-GFP cells were transfected with control and target siRNAs respectively, or reversely the U2OS cells transfected with target siRNAs and U2OS-GFP cells with control siRNA. The cells were trypsinized and mixed with an approximately 1:1 ratio two days after transfection. The cells were then co-plated and treated with DNA damage agents or left untreated. After seven to ten days of recovery, the cells were subjected to FACS analysis to determine the ratio of GFP+/GFP-(green/non-green) cells which reflects the relative survival of the two cell populations.

Clonogenic cell survival assay
For clonogenic assay, 200-1000 cells/10-cm dish were seeded and the cells were treated with hydroxyurea (HU) for 20h. Ten to Fourteen days after the treatment, Cells were fixed with 4% paraformaldehyde and stained with 0.05% crystal violet (Sigma-Aldrich) and colonies were counted.          Wild type U2OS and U2OS stably expressing transfected GFP were transfected with siCTL and siDDX5. The cells were then harvested two days after transfection and mixed in an approximately 1:1 ratio. These cells were plated and exposed to hydroxyurea. Then FACS analysis was performed following seven to ten days of recovery.
(E) A FACS profile shows DDX5 deficiency leads to a decrease of GFP+ cell population and an increase of GFP-cell population in the hydroxyurea-treated cells, compared to untreated cells.
(F and G) Cell survival following treatment with 10 mM hydroxyurea (HU) for 2 h or 0.4 mM 24 h. Data are plotted as a ratio in GFP+ to GFP-cells normalized to untreated control. The graphs show the average and standard error of the mean (SEM) from four independent experiments.

Figure S5. Identification of DDX5-interacting proteins by SILAC MS/MS spectrometry.
(A) U2OS cells with stably or transiently transfected Flag-DDX5 or empty vector were subjected to SILAC analysis as described in the "Materials and Methods" section.
(B) The identified proteins, including XRN2 with a heavy/light signal ratio were presented.