DDX39B inhibits NF-κB activity
Given the importance of the κB-site in regulating the response to alkylating chemotherapy [28], we performed streptavidin-agarose pull-down using biotin-tagged oligonucleotides containing κB binding sequences. Gel electrophoresis and silver staining of the pull-down product revealed a band that differentially bound κB DNA probes that vary only at the − 1 nucleotide (Additional file 1: Fig. S1a). MS/MS analysis of this band identified DDX39B as one of the only non-keratin peptides (Additional file 1: Fig. S1b). To validate DNA binding of DDX39B, we expressed and purified DDX39B protein (Additional file 1: Fig. S1c) and found that this protein bound to the -1C probe more than the -1A probe (Additional file 1: Fig. S1d). Given the propensity of DDX39B to bind κB DNA, we examined whether loss of DDX39B altered NF-κB activity. Several regions of DDX39B were targeted with short-hairpin (sh) vectors and a series of cell lines expressing either control or sh-DDX39B constructed (Additional file 1: Fig. S1e). Using a luciferase reporter under the control of the -1C κB-site, we found that loss of DDX39B increased NF-κB activity compared to control (Fig. 1a). Conversely, overexpression of DDX39B reduced NF-κB activity from this reporter (Fig. 1b). Of note, DDX39B also inhibited expression from a reporter under the control of a -1A κB-site (Additional file 1: Fig. S1f).
To examine the mechanism for inhibition of NF-κB by DDX39B, we analyzed IκBα, a primary regulator of the NF-κB response. Knockdown of DDX39B in GBM cells resulted in a decrease in IκBα protein (Fig. 1c). This decrease was not due to reduced mRNA expression as loss of DDX39B actually increased NFKBIA mRNA (Additional file 1: Fig. S2a). In addition, loss of DDX39B did not alter the fraction of NFKBIA in the cytoplasm (Additional file 1: Fig. S2b), an important finding given that DDX39B regulates mRNA nuclear export [13]. Given these findings, we examined IκBα phosphorylation and noted that loss of DDX39B resulted in increased phosphorylation (Fig. 1c). To confirm this unexpected finding in distinct cells, we used the patient-derived glioma stem-like cells (GSCs), GBM34 and GBM44 [27, 29]. We first determined the basal DDX39B protein abundance in these cells and noted that GBM44 GSCs have substantially more DDX39B than GBM34 (Fig. 1d). We then knocked down DDX39B in GBM44 GSCs and again observed both increased IκBα phosphorylation and decreased steady state IκBα protein (Fig. 1e). These results suggested that the decrease in IκBα protein was related to its phosphorylation. These findings raised the question of whether the increase in NF-κB activity with loss of DDX39B was due to increased nuclear p65 and p50 protein. Both immunoblot and immunofluorescence (IF) analysis did not show an observable change in nuclear p65 or p50 with loss of DDX39B (Fig. 1f and Additional file 1: Fig. S2c), suggesting that additional alterations mediated the change in NF-κB activity. Given the increase in IκBα phosphorylation, we examined p65 phosphorylation. Knockdown of DDX39B resulted in a substantial increase in p65 Ser536 phosphorylation in both GBM44 GSCs (Fig. 1g) and A172 GBM cells (Additional file 1: Fig. S2d). Conversely, overexpression of DDX39B in GBM34 GSCs that have low baseline DDX39B attenuated p65 phosphorylation (Fig. 1h). These findings indicate that loss of DDX39B increases NF-κB activity via a general effect on p65 phosphorylation.
DDX39B blocks NF-κB as part of the response to RNA
The increase in p65 phosphorylation in the absence of external stimulation suggested that DDX39B modulated NF-κB via an internal signaling response. Consistent with such a hypothesis, DDX39B was recently linked to the PRR pathway [8]. Given that DDX39B is known to bind mRNA and we find that it also binds DNA (Additional file 1: Fig. S1d), we examined whether the response to either of these nucleic acids mediated the effect of DDX39B on NF-κB. To this end, primary mouse embryonic fibroblasts (MEFs) deleted of either stimulator of IFN genes (STING), encoded by Tmem173, or mitochondrial antiviral signaling protein (MAVS) were obtained. These adaptor proteins mediate the response to cytoplasmic DNA and dsRNA, respectively. In wild-type (wt) MEFs, like human cells, knockdown of Ddx39b increased NF-κB activity (Fig. 2a). However, whereas loss of Ddx39b induced NF-κB in Tmem173−/− MEFs, in MAVS−/− MEFs, depletion of Ddx39b did not (Fig. 2a). These findings indicated that DDX39B modulated the NF-κB response to cytoplasmic dsRNA, not DNA. In addition to MAVS, two other adapter proteins mediate signaling associated with cytosolic RNA, myeloid differentiation primary response 88 (MYD88) and TIR-domain-containing adapter-inducing interferon-β (TRIF) [30]. While knockdown of Ddx39b increased NF-κB activity in Myd88−/− MEFs, in Trif−/− MEFs, no increase in NF-κB activity was seen (Fig. 2b). To further study these factors, we examined changes in p65 phosphorylation. While p65 phosphorylation was increased in wt MEFs following knockdown of Ddx39b, in both MAVS−/− and Trif−/− MEFs, no increase in phospho-p65 was seen (Fig. 2c). These findings support the role of MAVS and TRIF in mediating the regulation of NF-κB by DDX39B.
As a further specificity control to ensure that the changes in NF-κB were not a consequence of lentiviral infection, especially given that PRR signaling is modulated by viral infection, we used CRISPR/Cas9 technology to delete DDX39B in U87 GBM cells. Several CRISPR clones were isolated, and loss of DDX39B verified by immunoblot (Additional file 1: Fig. S2e). Similar to shRNA knockdown, CRISPR-mediated depletion of DDX39B resulted in both increased NF-κB activity and increased p65 phosphorylation (Fig. 2d and Additional file 1: Fig. S2f) without any appreciable change in p65 nuclear translocation (Additional file 1: Fig. S2g). Using these clones, we examined the role of MAVS in human cells and found that while depletion of DDX39B by CRISPR increased p65 phosphorylation compared to control, knockdown of MAVS, in two independent clones, reduced this phosphorylation (Fig. 2e). Consistent with this finding, knockdown of MAVS also blocked the increase in NF-κB activity induced by loss of DDX39B (Additional file 1: Fig. S2h). In addition, loss of DDX39B by CRISPR rendered U87 cells highly resistant to TMZ, an effect that was reversed by re-expression of DDX39B (Additional file 1: Fig. S2i). These results indicate that DDX39B blocks NF-κB activity by a mechanism involving the response to dsRNA.
DDX39B inhibits NF-κB in association with LGP2
MAVS primarily mediates signaling downstream of the dsRNA sensors, retinoic acid-inducible gene I (RIG-I), melanoma differentiation-associated gene 5 (MDA5), and LGP2 [30]. We examined whether these RIG-I-like receptors (RLRs) were required for the effect of DDX39B on NF-κB. Although knockout of either Rig-i or Mda5 in MEFs did not attenuate the increase in NF-κB induced by sh-Ddx39b, deletion of Lgp2 did (Fig. 3a), suggesting that DDX39B acted via LGP2 to regulate NF-κB activity. Consistent with this, in Lgp2−/− MEFs, knockdown of Ddx39b failed to increase p65 phosphorylation (Fig. 3b). To examine whether LGP2 was required for the effect of DDX39B in human cells, we used siRNA targeting human LGP2. Whereas sh-DDX39B increased NF-κB activity in cells expressing a control siRNA, in the presence of si-LGP2, knockdown of DDX39B did not increase NF-κB activity (Fig. 3c). These results indicate that LGP2, but not other RLRs, is required for the increase in NF-κB activity induced by DDX39B loss.
We next examined whether loss of DDX39B altered the abundance of the specific proteins involved in this pathway. Knockdown of DDX39B did not significantly change the protein or mRNA level of LGP2, MAVS, RIG-I, or MDA5 (Fig. 3d and Additional file 1: Fig. S3a). Protein kinase R (PKR) is another PRR previously shown to be activated by loss of DDX39B [31]. We examined whether PKR was required for the effect of DDX39B and found that knockdown of PKR did not significantly alter activation of NF-κB by DDX39B loss (Additional file 1: Fig. S3b). Similarly, toll-like receptor 3 (TLR3) was not required for this response (Additional file 1: Fig. S3b), an important observation given that TRIF mediates signaling downstream of TLR3 in response to endosomal dsRNA [30]. In addition, given the role of LGP2 and MAVS in mediating the response to dsRNA, we examined whether the increase in NF-κB was due to a change in the amount of cytoplasmic dsRNA, a finding previously reported with knockdown of DDX39B in the setting of influenza A infection [31]. Using a monoclonal antibody against dsRNA, we found that loss of DDX39B did not significantly increase the amount of cytoplasmic dsRNA (Fig. 3e and Additional file 1: Fig. S3c). Together, these findings indicate that although DDX39B regulated NF-κB via the PRR response to dsRNA, this was not a non-specific effect related to changes in dsRNA signaling.
We next examined whether DDX39B interacted with the PRRs involved. Overexpressed DDX39B did not associate with either MAVS or TRIF but did interact with LGP2 (Fig. 3f). Moreover, this interaction was evident even in the presence of nuclease, ruling out a nucleic acid bridge as the mechanism (Additional file 1: Fig. S3d). Subsequently, we confirmed the endogenous association of DDX39B with LGP2 in both human and mouse cells (Fig. 3g). Importantly, although LGP2 is cytoplasmic and DDX39B nuclear, we found that endogenous DDX39B was also present in the cytoplasm (Fig. 3h), a finding previously noted [32,33,34]. Finally, to validate the role of LGP2 in promoting NF-κB activation in the setting of DDX39B loss, we re-expressed LGP2 in Lgp2−/− MEFs. Whereas knockdown of DDX39B did not increase NF-κB in Lgp2−/− MEFs, when LGP2 was re-expressed, loss of DDX39B induced an increase in NF-κB (Fig. 3i). In addition, while overexpression of DDX39B in Lgp2−/− MEFs did not inhibit NF-κB activity, re-expression of LGP2 in these cells enabled inhibition of NF-κB by DDX39B expression (Fig. 3j). In sum, these results demonstrate that DDX39B inhibits NF-κB by a mechanism involving its interaction with LGP2.
DDX39B promotes cytotoxicity by temozolomide in GBM cells
As NF-κB plays a prominent role in the cytotoxic response to DNA damage, we examined whether modulating DDX39B level altered the response to the alkylating agent, temozolomide (TMZ). DDX39B was knocked down in U87 GBM cells, and clonal survival examined in multiple independent sh-DDX39B clones. Loss of DDX39B consistently resulted in significantly greater survival following treatment with TMZ compared to control (Fig. 4a). We next examined induction of cell death in patient-derived GSCs. In GBM44 GSCs, TMZ treatment induced significantly less cytotoxicity following knockdown of DDX39B than that seen with control shRNA (Fig. 4b). Consistent with the ability of DDX39B loss to attenuate cytotoxicity, overexpression of DDX39B in U87 GBM cells resulted in decreased clonal survival following TMZ treatment compared to control (Fig. 4c). Moreover, in GBM34 GSCs that have low basal DDX39B and were very resistant to TMZ, overexpression of DDX39B enabled induction of cytotoxicity by TMZ (Fig. 4d). These results indicate that DDX39B promotes cytotoxicity in response to alkylating chemotherapy.
DDX39B is regulated by sumoylation and PIASx-β
The propensity of DDX39B to block NF-κB activity raised the question of how this action is regulated. Sumoylation is a post-translational modification (PTM) that plays an important role in modulating the activity and metabolism of DExD-box helicases [35, 36]. To examine DDX39B sumoylation, we obtained HeLa cells stably expressing His-tagged SUMO 1 and 2 [37]. Using these cells, we saw a unique band in SUMO 2 expressing cells after nickel affinity purification and anti-DDX39B immunoblot (Fig. 5a). This finding was also noted when Flag-DDX39B was overexpressed in these cells (Additional file 1: Fig. S4a). To further examine this, we expressed SUMO 1, 2, and 3 in HEK293T cells. Despite the close homology between SUMO 2 and 3, DDX39B was only modified by SUMO 2 (Fig. 5b). In addition, given the role of DDX39B in the cytotoxic effect of alkylating DNA damage, we examined whether TMZ treatment affected its sumoylation. DDX39B was sumoylated in response to TMZ, and this effect was maximal 12 h after treatment (Fig. 5c).
To identify the sumoylation site, SUMOsp 2.0 software was used [38]. One lysine, K53, was identified within a consensus sumoylation motif (ΨKXE; where Ψ represents a bulky aliphatic residue), and three other lysines, K32, K155, and K156, were found within non-consensus sites (Fig. 5d). We constructed a K53R mutant obliterating the consensus sumoylation site and also an E55A mutant that removed the K53 sumoylation motif without disturbing the lysine itself. We also mutated all four potential sumoylation sites together, 4-mutant (4M). Expression of K53R with HA-SUMO 2 resulted in substantially reduced basal sumoylation of DDX39B (Fig. 5e). Similarly, sumoylation of E55A and 4M was reduced compared to wt. To validate the role of these residues, we used HeLa SUMO 2 cells. Again, mutation of K53 or its motif (E55A) blocked DDX39B sumoylation compared to wt (Additional file 1: Fig. S4b). Although a low level of sumoylation was seen with the mutants in some experiments, there was no difference in sumoylation of K53R and 4M suggesting that K53 was the important site. In addition, mutation of K53 blocked the increase in sumoylation induced by TMZ treatment (Fig. 5f).
To identify the potential E3 ligase involved in DDX39B sumoylation, we noted that in a previous yeast two-hybrid screen, protein inhibitor of activated statx-Beta (PIASx-β) was identified as a factor that interacts with DDX39B/Bat1 [39]. To screen for the potential role of PIAS proteins in DDX39B sumoylation, we expressed five PIAS constructs in HeLa SUMO 2 cells. Only PIASx-β increased the sumoylation of DDX39B (Fig. 5g). Moreover, when overexpressed, only PIASx-β not PIASx-α or PIASγ interacted with DDX39B (Fig. 5h). We also examined endogenous PIASx using a general PIASx antibody and found that this ligase interacted with DDX39B (Fig. 5i), and consistent with its ability to induce DDX39B sumoylation, treatment with TMZ increased the interaction of PIASx with DDX39B (Fig. 5i). In addition, knockdown of PIASX expression blocked both basal and TMZ-induced sumoylation of DDX39B (Fig. 5j). Together, these results indicate that DDX39B is sumoylated by a mechanism involving PIASx-β and that sumoylation occurs at K53.
DDX39B sumoylation promotes its degradation
Sumoylation of DExD-box helicases has been linked to transcriptional repression. We examined whether DDX39B sumoylation was required for its ability to inhibit NF-κB. While wt-DDX39B inhibited NF-κB activity, mutation of either K53 or E55 did not significantly alter this effect (Additional file 1: Fig. S4c) suggesting that sumoylation of DDX39B does not directly mediate inhibition of NF-κB. Sumoylation also regulates protein stability. To study the role of sumoylation in modulating DDX39B stability, we examined the kinetics of its degradation in the presence of the protein synthesis inhibitor, cycloheximide (CHX). Whereas the amount of wt-DDX39B protein was decreased at 4 h and disappeared within 8 h, mutant DDX39B was not decreased until 8 h (Fig. 6a and Additional file 1: Fig. S4d). Notably, treatment with TMZ did not alter DDX39B mRNA expression (Additional file 1: Fig. S4e). These findings suggested that blocking K53 sumoylation increased the stability of DDX39B protein. As protein stability is associated with poly-ubiquitination, we examined ubiquitination of DDX39B. Consistent with the increased stability of DDX39B sumo-mutants, these mutants had decreased basal poly-ubiquitination compared to wt (Fig. 6b).
We next examined the role of PIASx in this response and found that knockdown of PIASX reduced DDX39B poly-ubiquitination compared to control (Fig. 6c). Subsequently, using overexpression studies, we found that while PIAS1 did not increase DDX39B ubiquitination, PIASx-β substantially increased ubiquitin addition (Fig. 6d). Moreover, in the presence of the proteasome inhibitor, MG132, the stability of wt-DDX39B was substantially extended (Fig. 6e). These results indicated that sumoylation of DDX39B at K53 in the presence of PIASx-β resulted in its poly-ubiquitination and proteasomal degradation. Finally, as TMZ also promoted sumoylation of DDX39B, we examined its effect on ubiquitination and found that treatment with TMZ increased DDX39B poly-ubiquitination (Fig. 6f). Consistent with this, TMZ treatment also led to decreased DDX39B protein stability (Fig. 6g). In sum, these findings indicate that DDX39B protein abundance is regulated by sumoylation-dependent ubiquitination and that alkylation damage leads to a decrease in DDX39B protein.
DDX39B inhibits expression of secreted factors associated with the extracellular matrix, migration, and angiogenesis
The interaction of DDX39B with the PRR response raised the question of whether DDX39B modulated innate immune signaling in general. To begin to study this, we examined expression of interferon beta (IFNB1), a primary interferon-stimulated gene (ISG). Loss of DDX39B in GBM cells did not significantly alter IFNB1 expression (Additional file 1: Fig. S5a), suggesting that despite inhibition of NF-κB, DDX39B did not modulate general interferon signaling. Given this finding, to more comprehensively study the DDX39B-dependent response, we examined genome-wide expression in GBM. Using GBM44 GSCs that express high levels of DDX39B, we studied differential gene expression following knockdown of DDX39B compared to control (Fig. 7a and Additional file 2: Table S1). Gene ontology (GO) term analysis of the most significantly altered genes (FDR < 0.01) revealed that genes associated with the ECM and migration were among the most significantly upregulated (Fig.7b), while transcripts associated with cytokine-mediated signaling were downregulated (Additional file 1: Fig. S5b). In addition, we interrogated the list of differentially expressed genes (DEGs) with known NF-κB target genes. Of the 430 NF-κB target genes identified by the Gilmore lab [40], 117 were present in the DEGs from our analysis (Additional file 2: Table S1) underlining the relevance of DDX39B to regulation of NF-κB. Consistent with the lack of change of IFNB1, no general change in innate immune pathways or IGSs was seen. To validate the RNA-seq data, we performed qPCR analysis following DDX39B knockdown and confirmed the changes in expression of many of the most significantly altered genes (Fig. 7c). In addition, we found similar changes in the expression of several of the upregulated genes in a distinct GBM cell line (Additional file 1: Fig. S5c).
To further study the role of DDX39B, we noted that in a previous study, genome-wide analysis had been performed in HeLa cells following knockdown of DDX39B [24]. In that study, differential gene expression was not reported. We analyzed the raw data from that study (GSE94730). Notably, GO term analysis revealed that similar to our findings, the subgroups of factors most significantly upregulated with loss of DDX39B included genes that were either secreted or associated with the ECM (Additional file 1: Fig. S5d and Additional file 3: Table S2). The primary terms linked to downregulated genes included GTP associated factors (Additional file 3: Table S2).
Given that the RNA-seq studies were performed on whole cell lysates and that loss of DDX39B leads to nuclear retention of a subset of mRNAs, we examined whether the upregulated transcripts demonstrated any change in cellular distribution. To this end, mRNA levels were quantified in nuclear and cytosolic fractions. Although loss of DDX39B attenuated the cytoplasmic fraction of HSP70, an mRNA previously shown to be exported by DDX39B [41], it did not significantly change the amount of cytoplasmic PDGFRA, PLAU, or SMAD3 (Fig. 7d). These results indicated that transcripts upregulated with loss of DDX39B were also exported out of the nucleus where they could be translated. To further study this response, we expressed exogenous DDX39B in GBM34 GSCs that have low basal DDX39B. Overexpression of DDX39B decreased the level of several of the genes that were identified in the RNA-seq analysis (Fig. 7e) validating the ability of DDX39B to attenuate expression of these factors. In addition, to determine whether the link between DDX39B and LGP2 was important in modulating endogenous gene expression, we examined expression of several genes following knockdown of LGP2 in the presence of sh-DDX39B. Notably, depletion of LGP2 reduced the increase in ADGRG1, PLAU, and PTX3, but not PDGFA, induced by DDX39B loss (Fig. 7f). These results indicate that DDX39B attenuates the expression of certain genes involved in the ECM and angiogenesis in association with LGP2.
Finally, as DDX39B regulates multiple aspects of mRNA metabolism including splicing, we were interested in whether the changes in mRNA expression were reflected by similar alterations at the protein level. To screen for changes in multiple proteins, we used a membrane-based array that contained antibodies against many of the factors that were significantly altered in the RNA-seq analysis. Knockdown of DDX39B resulted in an increase in the protein abundance of many of the factors that were increased at the mRNA level (Fig. 7g and Additional file 1: Fig. S5e). Moreover, the transcript that was most significantly downregulated on RNA-seq following loss of DDX39B, IL1B, was also strongly decreased at the protein level (Fig. 7g). These findings confirmed that the mRNA changes seen with loss of DDX39B were recapitulated at the protein level. Notably, the proteins induced with loss of DDX39B act to promote angiogenesis, cell migration, and interaction with the ECM (e.g., VEGF, uPA, IL6, PTX3, and PDGFB), supporting the mRNA data and indicating that in GBM cells, DDX39B attenuates expression of secreted factors that are associated with these processes.