DEK is required for homologous recombination repair of DNA breaks

DEK is a highly conserved chromatin-bound protein whose upregulation across cancer types correlates with genotoxic therapy resistance. Loss of DEK induces genome instability and sensitizes cells to DNA double strand breaks (DSBs), suggesting defects in DNA repair. While these DEK-deficiency phenotypes were thought to arise from a moderate attenuation of non-homologous end joining (NHEJ) repair, the role of DEK in DNA repair remains incompletely understood. We present new evidence demonstrating the observed decrease in NHEJ is insufficient to impact immunoglobulin class switching in DEK knockout mice. Furthermore, DEK knockout cells were sensitive to apoptosis with NHEJ inhibition. Thus, we hypothesized DEK plays additional roles in homologous recombination (HR). Using episomal and integrated reporters, we demonstrate that HR repair of conventional DSBs is severely compromised in DEK-deficient cells. To define responsible mechanisms, we tested the role of DEK in the HR repair cascade. DEK-deficient cells were impaired for γH2AX phosphorylation and attenuated for RAD51 filament formation. Additionally, DEK formed a complex with RAD51, but not BRCA1, suggesting a potential role regarding RAD51 filament formation, stability, or function. These findings define DEK as an important and multifunctional mediator of HR, and establish a synthetic lethal relationship between DEK loss and NHEJ inhibition.

a DEK-NUP214 fusion protein in AML 26 and the discovery of elevated DEK expression in breast 3,27 , colorectal 5,28 , lung 4,29 , and several other types of cancer 1 , many systems have been used to investigate the pathological consequences of DEK over-expression. A prominent phenotype in cell models is the requirement of DEK for chemotherapy and radiation resistance. For example, expression of the DEK C-terminal domain in ataxia-telangiectasia fibroblasts partially restored radiation resistance, and the loss of DEK conferred sensitivity to DNA damaging agents in multiple cell types 14,19,30 . Mechanistically, our prior report found that DEK was required for optimal kinase activity of DNA-PK. This kinase is a key mediator of canonical non-homologous end joining (NHEJ), which repairs DNA double strand breaks (DSBs) 31 , and DEK loss correspondingly suppressed NHEJ 19 .
The observed NHEJ defects in DEK-deficient cells are unlikely to fully account for the severe sensitivity to genotoxic agents, especially DNA interstrand cross linkers and topoisomerase inhibitors 14,32 . This suggests additional roles for DEK in genotoxic drug tolerance and DNA repair. A common mechanism by which genotoxic agents induce cell death is through perturbation of replication fork progression 33 . A recent study determined that DEK attenuates DNA replication stress 12 in a manner similar to RAD51 and FANCD2, factors well known for their function in both homologous recombination (HR) DSB repair and activities at arrested replication forks [34][35][36][37] .
HR requires the presence of a homologous template, often the sister-chromatid, to be used for repair, and is therefore largely confined to S/G2 phases of the cell cycle. By copying a homologous DNA sequence, HR is considered an error-free repair process that preserves genome integrity 38 . This signal cascade is initiated by the DSB sensor, ATM kinase. After localizing to a DSB, ATM autophosphorylates 39,40 , pATM catalyzes the phosphorylation of CHK2 to inhibit cell cycle progression 41 and the H2AX histone to produce gamma-H2AX (γ H2AX) epigenetic marks on both sides of the DSB. The γ H2AX mark supports DSB repair by enhancing the recruitment of BRCA1 and key nucleases including the MRN complex and CtIP 38,[42][43][44] . These factors coordinate DNA end processing into single strand DNA (ssDNA) 3′ tails 38,42,43 . The resulting ssDNA is initially coated by RPA, which is then efficiently replaced with a RAD51 filament through the combined activities of BRCA1, BRCA2, the RAD51 paralogs, and other factors 38,42 . This RAD51 filament catalyzes strand invasion, complementary strand annealing, and the formation of a stable synaptic complex with a homologous sequence on the sister chromatid 45 .
While the cellular regulation and choice between DSB-repair pathways remains incompletely understood 46-50 , NHEJ and HR factors have been shown to be mutually antagonistic as one pathway tends to compensate when the other is compromised 42,51 . For example, loss of critical HR gene functions in Fanconi Anemia mutant cells results in enhanced dependence upon NHEJ factors [52][53][54] while HR repair efficiency is increased following the disruption of essential NHEJ factors by chemical inhibition or mutation of DNA-PK 51,55,56 . Compromising both the NHEJ and HR repair pathways may be synthetic lethal to cancer cells.
In this report, we demonstrate that DEK is required for the repair of DSBs by HR. Investigation of potential DEK functions in the HR pathway revealed that the protein was important for γ H2AX activation, promoted the co-recruitment of BRCA1 and RAD51 to resected ssDNA, and formed a complex with RAD51 in a BRCA1-independent manner. The dependence of HR on DEK expression was underscored by the synthetic lethal relationship between DEK loss and NHEJ inhibition. Together, these results reveal a novel, multifunctional role for DEK in HR.

DEK loss causes apoptosis in conjunction with DNA-PK inhibitors.
We have previously shown that DEK knockout (DEK− /− ) mouse embryonic fibroblasts (MEFs) harbor decreased NHEJ activity 19 . To determine the contribution of DEK to NHEJ activity in vivo, we measured the concentrations of immunoglobulins in DEK− /− mouse serum. Class switch recombination (CSR) from IgM to the IgA and IgG classes of immunoglobulins is an NHEJ-dependent process 31 , but DEK− /− mice were fully competent in producing all classes of immunoglobulins (Fig. 1a). This data suggests the residual NHEJ activity in the knockout mice is sufficient for CSR under physiologically normal in vivo conditions, despite the sensitivity of DEK-deficient cells to DNA damaging agents 14,19 . To determine if DEK-deficient cells require NHEJ for survival, we examined the need for NHEJ by inhibiting the upstream kinase, DNA-PK. We utilized two well-established DNA-PK inhibitors, NU7026 and NU7441, to prevent DNA-PKcs autophosphorylation and activation [57][58][59][60] . DEK− /− MEFs treated with the DNA-PK inhibitor NU7026 displayed dramatically more cell death than wild-type (DEK+ /+ ) or untreated samples (Fig. 1b). Analysis of cleaved caspase 3 positive cells by flow cytometry revealed that the inhibitors were sufficient to induce a significant and specific 2-3 fold increase in apoptosis in the DEK− /− cells in the absence of exogenous DNA damaging agents (Fig. 1c). Similar experiments in HeLa cells infected with a well-characterized DEK knockdown versus control adenovirus vector (AdDEKsh versus AdGFP) 18 revealed a similar 2-fold increase in apoptosis after treatment with either DNA-PK inhibitor, NU7026 or NU7441 (Fig. 1d,e). These data suggest that DEK-deficient cells, despite their attenuated NHEJ activity 19 , rely significantly on the remaining NHEJ for survival.
DEK is required for HR DSB repair. Several reports have described an antagonistic relationship between HR and NHEJ mechanisms of DSB repair 46,55,61 , and it is generally thought that loss of one pathway will result in compensation by the remaining repair modality 53,55,56 . Given the dependence of DEK-deficient cells on NHEJ and their inherent sensitivity to chemotherapeutics, we hypothesized that HR may be compromised in DEK− /− cells. To test this we first utilized two established episomal HR reporter systems 56,62 . These vectors harbor two defunct EGFP genes, the first bearing an I-SceI meganuclease cleavage site and the second bearing a truncated gene sequence (Fig. 2a). Co-transfection with an I-SceI expression vector induces a DSB that, when repaired by HR, generates a functional EGFP gene. Both the HR-EGFP/5′ EGFP and pHPRT-DR-GFP reporters demonstrated a remarkable and severe loss of HR efficiency in DEK− /− MEFs (Fig. 2b). To compare the effects of DEK loss on chromosomal HR repair following direct or replication-dependent DSB induction, we utilized the recently published 6xTer-HR reporter technology 63 . The 11CO/47 mouse embryonic stem (mES) cell line bearing a single copy of the 6xTer-I-SceI-GFP reporter cassette integrated at the Rosa26 locus quantifies HR either at an I-SceI endonuclease-induced DSB or at an adjacent Tus/Ter-induced stalled replication fork. Fork stalling is induced by Tus protein binding to an array of six Ter elements, forming a physical barrier that impedes approaching replication fork progression 63 (Fig. 2c). Using this system, we quantified the impact of DEK depletion on error-free short tract gene conversion (STGC), producing GFP + RFP − cells, and error-prone long tract gene conversion (LTGC), producing GFP + RFP + cells through duplication and thus proper alignment of two synthetic exons of RFP. After co-transfecting Dek siRNA (Fig. 2D) and I-SceI, we found that siDek treated cells were significantly compromised for both total HR and STGC, and LTGC followed the same trend (Fig. 2e). This decrease was similar in magnitude to loss of the BRCA proteins 64 , but not as severe as RAD51 deficiency 63 . However, there was no significant decrease in STGC, LTGC, or total HR with Tus-induced DNA replication fork stalling (Fig. 2f). This suggests that DEK, unlike most HR factors 63 , is dispensable for replication fork associated repair but required for efficient repair of DSBs. DEK is necessary for the activation of γH2AX after ionizing radiation. To understand how DEK functions in HR, we examined the consequences of DEK loss at successive steps in the repair pathway following irradiation (IR) mediated DSB induction. For these studies we used DEK− /− MEFs and AdDEKsh-infected HeLa cell systems, whose cell cycle progression was reported comparable to that of their respective DEK-proficient controls 18,19 . Beginning with the relevant upstream DNA damage sensor kinase, we found that ATM was strongly autophosphorylated in AdDEKsh-treated HeLa cells ( Supplementary Fig. S1a) and functional as indicated by phosphorylation of pCHK2 T68, an ATM substrate ( Supplementary Fig. S1b). Surprisingly, γ H2AX phosphorylation was not enhanced in the AdDEKsh HeLa cells despite robustly activated pATM ( Supplementary Fig. S1a). This held true and was even more severe in the DEK− /− MEF system where DEK was required to increase γ H2AX phosphorylation above baseline by immunofluorescence (IF) (Fig. 3a). Both the total number of cells harboring γ H2AX foci and the average number of foci per cell were severely attenuated from 3-24 hours after irradiation (Fig. 3b,c). These results were also confirmed by western blot analysis in MEFs and in DEK knockdown C33A cancer cells (Fig. 3d,e), suggesting that DEK is an important general contributor to γ H2AX phosphorylation in IR-treated cells.
Loading of RAD51 onto RPA-protected DNA is significantly reduced. DSB end-processing occurs downstream of the pATM-γ H2AX signal amplification feedback loop and generates ssDNA overhangs that are bound and protected by RPA. End processing is coordinated by several factors, including the MRN  complex and BRCA1 43,44 . Chromatin fractionation of HeLa cells revealed that chromatin recruitment of MRE11 and NBS1, components of the MRN complex, as well as BRCA1 were not affected by DEK status following IR ( Supplementary Fig. S2a,b). Following end-processing, BRCA1 both co-localizes with and is necessary for robust RAD51 IF foci development 43 . Through assessing the quality and quantity of these foci, we determined whether DEK loss disrupts RAD51 loading onto RPA-bound ssDNA 43 . Since our BRCA1 antibody does not detect murine BRCA1, we performed these experiments in HeLa cells. DEK knockdown in HeLa cells had a mild, but statistically significant delay in BRCA1-RAD51 foci formation at 3 hours following IR (Fig. 4a,b) with no difference in individual BRCA1 or RAD51 foci formation at any time point (Supplementary Fig. S3). By 6 hours the co-localization of BRCA1 and RAD51 foci was indistinguishable from control cells. To determine if the delay in RAD51-BRCA1 co-localization correlated with impaired RAD51 loading onto RPA-protected ssDNA, we also determined the kinetics of RAD51 localization to RPA foci. In line with the delay with BRCA1-RAD51 co-localization, we found a significant decrease of RAD51-RPA2 co-localized foci at 3 and 6 hours post irradiation (Fig. 4c). A 24-hour time course in DEK+ /+ and DEK− /− MEFs revealed that RAD51 loading was similarly attenuated (Fig. 4d,e). There was no difference in RAD51 foci quantity or quality ( Supplementary Fig. S4a), but RPA2 foci were generally smaller in DEK− /− cells. DEK− /− cells also retained persistent RPA2 foci at later time points, suggesting a DNA repair defect (Supplementary Fig. S4b). In summary, DEK-deficient cells have a small reduction in the recruitment of RAD51 to BRCA1 foci, which results in a mild attenuation of RAD51 loading on RPA-coated ssDNA.
DEK interacts with RAD51. While RAD51 loading was not dramatically affected by DEK loss, physical DEK interactions with the RAD51 recombinase were possible. To test this, pMIEG His-FLAG-DEK expressing HeLa cells were left untreated or received 1 mM hydroxyurea (HU) for 15hrs to induce replication fork stalling and a low level of DSB generation as published previously 65 , prior to FLAG immunoprecipitation (IP). Under these conditions, we found that RAD51 complex formation occurred with His-FLAG DEK (Fig. 5a). Validation of the complex was performed by IP of endogenous DEK and reverse IP of RAD51 in parental HeLa cells (Fig. 5b,c). No BRCA1 interaction with DEK was observed (Fig. 5b), suggesting that DEK forms a separate complex from the standard BRCA-RAD51 homologue apparatus. The DEK-RAD51 complex was also observed in untreated and irradiated cells. Thus, DEK interacts with the HR recombinase RAD51 in the presence or absence of exogenous DNA damage (Supplementary Fig. S5).

Discussion
Herein we demonstrate that DEK is required for the repair of DSBs by HR, and regulates multiple steps in the HR cascade by promoting γ H2AX activation, enabling robust RAD51 loading, and forming a complex with RAD51 (Fig. 5d). This importance of DEK in HR was further underscored by the exquisite sensitivity of DEK-deficient cells to NHEJ loss through DNA-PK inhibition. In summary, this is the first report to describe DEK as a HR factor and to identify a synthetic lethal relationship that exists between DNA-PK inhibition and DEK loss.
According to the 6xTer reporter assays (Fig. 2c-f), DEK appears to be required for HR only in the context of conventional DSBs, wherein DEK-deficient cells had a similar HR deficiency to BRCA1 loss 63 , but not in the context of stalled replication forks. However, this selectivity differs when DEK activities are compared to those of BRCA1, BRCA2, and RAD51, factors which are required for the HR repair triggered by both I-SceI induced DSBs and Tus-stalled replication forks 63 . Based on these observations, it is likely that DEK is expendable for replication-associated HR repair, but this by no means precludes non-HR functions at stalled replication forks. Indeed, published DNA fiber assays have elegantly shown that DEK can promote stalled replication fork restart and reduce γ H2AX response to DNA replication stress 12 .
Differential γ H2AX responses to irradiation (Fig. 3) versus chemotherapy-induced damage 12,19 lends further support to a model where DEK is specifically required for HR repair of DSBs. Previous studies have found that loss of DEK enhances γ H2AX activation following treatment with replication fork stalling agents 12,19 . While similar pATM activation was observed in response to stalling or irradiation, γ H2AX phosphorylation was specifically not engaged after IR exposure. This suggests that DEK is specifically required for ATM-H2AX signal transduction under conditions of DSBs. While it is known that γ H2AX is not strictly required for HR, its loss can slow the kinetics of the pathway and lead to a moderate decrease in repair efficiency 44,66 . Thus, the attenuation of BRCA1-RAD51 and RAD51-RPA2 foci co-localization (Fig. 4) is likely a consequence of failed γ H2AX activation. These pulldown results, in which DEK interacted with RAD51 in a BRCA1-free complex (Fig. 5), further support this hypothesis as they suggest DEK is not part of the BRCA1-Palb2-BRCA2-RAD51 complex that facilitates RAD51 loading. However, we cannot presently rule out weak interactions or a role for BRCA in promoting the DEK-RAD51 interaction. Also, since these interactions were identified and confirmed in cervical cancer cells, it is uncertain if the DEK-RAD51 complex is a component of normal cell biology or specific to cancer cells.
Taken together these data support a model in which DEK has two distinct roles in the repair of DSBs by HR. First, DEK has an important function in γ H2AX activation by ATM, but this is unlikely to be sufficient for the degree of HR deficiency observed in DEK-deficient cells 44,66 . Therefore, DEK must have a second function involving the DEK-RAD51 complex. With regard to a possible mechanism of action, DEK might play an early role in homologous recombination by supporting the initial formation of the RAD51-DNA filament. However, since BRCA1 and BRCA2 play key roles in this step and, since BRCA1 was not detectable in the DEK-RAD51 complex (Fig. 5b), we do not favor this scenario. Based on previous structural and cell-free studies, DEK binding to DNA stimulates self-multimerization and potential interactions with other proteins 22,67,68 . Furthermore, DEK preferentially binds cruciform over conventional linear DNA structures 24 , and harbors three putative DNA binding motifs 67,68 . Thus, one possible scenario is that DEK interacts with both the invading RAD51 filament and  the sister chromatid to stabilize resulting D-loop structures. Such activities can now be readily explored through established biochemical assays 69 .
Since this is the first report to establish a role for DEK in HR, our work opens up multiple avenues for future study. At the molecular level, mechanisms whereby DEK promotes γ H2AX activation specifically at DSBs and the nature and function of the complex with RAD51 remain unclear. Secondly, in the absence of known DEK homologues, detailed studies of the evolutionary history and origin of DEK are lacking despite its strong conservation across mammals and presence in plants. Interestingly, DEK is absent in bacteria and yeast, wherein mechanisms of HR are well described. It is possible that the early multicellular eukaryotes required a novel chromatin regulator to stabilize their increasingly complex genomes during DNA repair. Considering its role in chromatin modification and ATM-mediated γ H2AX activation, its affinity for RAD51 and DNA structures similar to Holliday junctions, and its ability to stabilize DNA breaks for in vitro DNA ligation, DEK is a likely candidate to fill this niche 1,24 .
Still, further investigations into the activities of DEK in HR and NHEJ repair will remain important for deepening our knowledge of the evident evolutionary pressure on DEK and HR, understanding the multiple activities of DEK in HR, and for the development of small molecule inhibitors to target relevant functions of this oncogene. In support of these future therapeutic efforts, we identified a synthetic lethal relationship between DEK loss and canonical NHEJ inhibition, as well as a cellular phenotype that can be assessed for DEK inhibition in vivo. This is important as DEK is largely dispensable for cell division, and DEK− /− mice are healthy and viable 12,18,20 , suggesting a high therapeutic index for a potential anti-DEK drug. Thus, it is conceivable that DEK overexpressing tumors may regress without a need for adjuvant radiation or traditional chemotherapy if substituted for tumor-specific co-targeting of DEK and DNA-PK.

Materials and Methods
Cell culture, adenoviral infections, and viral transductions. HeLa and C33A cells were grown in Dulbecco's Modified Essential Medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and antibiotics. MEFs, generated previously 19 , were cultured in DMEM with 10% heat inactivated FBS, 100 μ M MEM non-essential amino acids, 0.055 mM β -mercaptoethanol (BME), 2 mM L-glutamine, and 10 μ g/ml gentamycin. Mouse embryonic stem (ES) cells were grown in ES media on fibroblast or gelatin substrate as previously described 63 . To induce DNA damage, cells were treated with γ -IR from a 137 Cs source or 1 mM hydroxyurea (HU). Both 40 μ M NU7026 and 2 μ M NU7441 (Tocris, Bristol, UK) were used to specifically inhibit DNA-PK as previously published 59,70 . DEK knockdown was accomplished using the adenoviral AdDEKsh as compared to control AdGFP vectors at 10 infectious units (IU) per cell for 48 hr prior to IR treatment, as described previously 18 . The pMIEG His-FLAG-DEK retroviral vector was described recently 71 , and cells were transduced with virus for 24 hrs prior to sorting for GFP positive cells on a BD-FACSAria II flow cytometer.
Serum Immunoglobulin ELISA Assays. Dek+ /+ and Dek− /− littermate mice were born from heterozygous crosses on a mixed C57Bl6/S129 background as previously described 20 . Usage and handling of mice was performed with the approval of the Cincinnati Children's Institutional Animal Care and Use Committee and complied with institutional, state, and federal guidelines and regulations as well as AAALAC accreditation standards. All mice were housed in specific pathogen free housing with ad libitum access to food and water. Blood was obtained by cardiac puncture of three DEK+ /+ and three DEK− /− mice. To measure immunoglobulin (Ig) in the blood serum by ELISA, plates were coated with the following Southern Biotechnology Associates antibodies (Birmingham, AL, USA): anti-mouse IgM (no. 406501), IgA (no.556969), or IgG (no. 1030-01), and Ig was detected with horseradish peroxidase (HRP)-conjugated goat anti-mouse IgG1 (no. 1070-05), IgG3 (no. 1100-05), IgG2a (no.1080-05), IgG2b (no. 1090-05), IgA (no. 1040-05), or IgM (no. 1020-05). In all cases, wells were developed with the Ultra TMB peroxidase substrate system (Thermo Scientific) and OD was measured at 450 nm using a Fluostar Omega microplate reader (BMG-Labtech, Ortenberg, Germany).
Cleaved caspase 3 flow cytometry. Cells were trypsinized, fixed in 4% PFA for 10 min at 37 °C, permeabilized in 90% ice-cold methanol for 30 min, and incubated with the cleaved caspase 3 cell signaling antibody for 1 hr. Analysis was performed on a BD FACSCanto II instrument.

Western blot analysis.
Chromatin fractionation. Chromatin fractionation was performed as described previously 73 according to the schematic in Fig. 3d. Briefly, HeLa cells were infected with AdGFP or AdDEKsh 48 hr prior to receiving 10 Gy of IR. Six hours following IR, cells were counted to ensure equal numbers of cells per sample. Cell lysis and nuclear isolation was performed in a wash buffer (10 mM PIPES pH = 7.0, 1 mM EGTA, 0.1 M NaCl, 0.3 M sucrose, 0.5 M NaF, 0.5 mM Na 3 VO 4 , and 1x protease inhibitor cocktail) plus 1% Triton X-100. Serial separation of the precipitated nuclear fraction was performed first by treatment with wash buffer + 20 U DNAseI (AM2235, ThermoFischer Scientific), followed by a 5 min incubation with wash buffer + 0.5 M (NH 4 )SO 4 on the resulting pellet. The resulting fractions were treated with 4x loading buffer.