RUNX super-enhancer control through the Notch pathway by Epstein-Barr virus transcription factors regulates B cell growth

Abstract In B cells infected by the cancer-associated Epstein-Barr virus (EBV), RUNX3 and RUNX1 transcription is manipulated to control cell growth. The EBV-encoded EBNA2 transcription factor (TF) activates RUNX3 transcription leading to RUNX3-mediated repression of the RUNX1 promoter and the relief of RUNX1-directed growth repression. We show that EBNA2 activates RUNX3 through a specific element within a −97 kb super-enhancer in a manner dependent on the expression of the Notch DNA-binding partner RBP-J. We also reveal that the EBV TFs EBNA3B and EBNA3C contribute to RUNX3 activation in EBV-infected cells by targeting the same element. Uncovering a counter-regulatory feed-forward step, we demonstrate EBNA2 activation of a RUNX1 super-enhancer (−139 to −250 kb) that results in low-level RUNX1 expression in cells refractory to RUNX1-mediated growth inhibition. EBNA2 activation of the RUNX1 super-enhancer is also dependent on RBP-J. Consistent with the context-dependent roles of EBNA3B and EBNA3C as activators or repressors, we find that these proteins negatively regulate the RUNX1 super-enhancer, curbing EBNA2 activation. Taken together our results reveal cell-type-specific exploitation of RUNX gene super-enhancers by multiple EBV TFs via the Notch pathway to fine tune RUNX3 and RUNX1 expression and manipulate B-cell growth.


INTRODUCTION
The mammalian runt-related family of transcription factors (TF) (RUNX) is encoded by three separate genes (RUNX1, RUNX2 and RUNX3) located on different chromosomes that play crucial roles in the control of a range of developmental and differentiation processes (1). RUNX genes have distinct patterns of tissue-specific expression, but all bind the same DNA consensus site, through heterodimerization with the non-DNA binding CBF␤ protein, to activate or repress transcription (2,3). Disruption or misregulation of RUNX expression is associated with a wide range of human tumours (1). RUNX1 is frequently translocated in myeloid and lymphoid malignancies, with fusion of RUNX1 to the Ets family TEL TF in B-cell acute lymphoblastic leukaemia and to ETO in acute myeloid leukaemia (4). RUNX2 is essential for osteogenesis and linked to osteosarcoma (5) and RUNX3 is inactivated in a variety of solid tumours (1). RUNX1 and RUNX3 play important roles in regulating haematopoesis with loss of RUNX1 resulting in defective T and B-cell development and embryonic lethality in mice and loss of RUNX3 resulting in altered T-cell differentiation profiles (1). For all RUNX genes transcription initiates from one of two promoters located distal (P1) or proximal (P2) to the translation start site that give rise Nucleic Acids Research, 2016, Vol. 44, No. 10 4637 to protein isoforms that differ in their amino termini and alternative splicing generates further isoforms with functional differences. RUNX1 transcription is also regulated by a Gata2 and Ets protein-controlled +23 kb intronic enhancer in mouse cells and by an equivalent haemopoieticcell-specific enhancer (RE1) in human cells (6,7). The 173 kb region between P1 and P2 encompassing RE1 also functions as a CDK7-dependent RUNX1 super-enhancer in Tcell acute lymphoblastic leukaemia cell-lines (8).
Epstein-Barr virus (EBV) is a key driver in the development of a wide range of lymphomas including Burkitt's (BL), Hodgkin's and Diffuse Large B-cell (9). Its ability to immortalize resting B cells in vitro reflects its oncogenic properties and results in the generation of permanently proliferating lymphoblastoid cell lines (LCLs) in which the virus persists in its latent form (10). Latently infected LCLs express a limited set of EBV proteins comprising six nuclear antigens (EBNAs 1, 2, 3A, 3B, 3C and leader protein) and three latent membrane proteins (LMP1, 2A and 2B). In addition to regulating viral latent gene transcription, EBNA2 and the EBNA3 family of TFs (3A, 3B and 3C) drive growth transformation through epigenetic reprogramming of the host B cell (11)(12)(13)(14)(15)(16). These viral TFs do not bind DNA directly, however, but hijack B cell TFs in order to access viral and cellular gene regulatory elements. The best characterized of these interactions is between EBNA2, 3A, 3B and 3C and the Notch signalling pathway DNA-binding protein RBP-J (CBF1, CSL, Su(H)) (17)(18)(19)(20)(21). The interaction between EBNA2, 3A, 3C and RBP-J is essential for EBVdriven B cell growth demonstrating a central role for RBP-J in cellular gene reprogramming (22)(23)(24). In reporter assays, EBNA3 proteins inhibit RBP-J dependent gene activation by EBNA2 in manner involving competitive binding to RBP-J (18,21,25), although EBNA2 and EBNA3 proteins appear to bind RBP-J at different sites on the protein (26)(27)(28).
EBNA2 and EBNA3C also interact with the cellular TF PU.1 and EBNA2 activation of the EBV LMP1 promoter requires the presence of both PU.1 and RBP-J binding sites, indicating a role for PU.1 in the regulation of at least a subset of genes (29)(30)(31). Interestingly, the LMP1 promoter PU.1 site resembles a composite PU.1/IRF element and these composite sites are implicated in the EBV type-specific regulation of specific cellular genes by EBNA2 (16,32). A binding site for EBF1 is also required for activation of the LMP1 promoter by EBNA2 (33).
EBNA2 is best characterized as a transcriptional activator and harbours a classical acidic activation domain (34), although repressed gene targets have been identified (35,36). EBNA3 proteins function as activators and repressors of transcription, curbing EBNA2 activation through their associations with RBP-J, but also regulating transcription through EBNA2-independent mechanisms. Their role in epigenetic silencing through the polycomb repressor complex-mediated H3K27me3 chromatin silencing mark has been well studied (14)(15)37). We and others have shown that EBNA2 and EBNA3 proteins predominantly target cellular genes through their associations with longrange regulatory elements (15)(16)(32)(33)(38)(39). Studying the influence of EBNA binding on long-range enhancerpromoter interactions we demonstrated that EBNA3 pro-teins can repress cellular gene transcription by preventing enhancer-promoter loop formation (anti-looping) or by a repression mechanism involving the formation of loops between target gene promoters and distal EBNA3-bound sites (16).
Here, we identify the key elements within RUNX3 and RUNX1 super-enhancers through which EBNA2 and EBNA3 proteins control RUNX expression to manipulate B cell growth. Our data demonstrate that the Notch pathway component RBP-J is required for EBNA2 activation of RUNX3 and reveal additional coactivation of RUNX3 by EBNA3B and 3C. We also uncover direct feed-forward control of a novel cell-type specific RUNX1 super-enhancer region by EBNA2 through RBP-J-dependent mechanisms and show that at RUNX1, EBNA3B and 3C attenuate this activation.

Cell lines
All cell lines were routinely passaged twice-weekly and cultured using the conditions previously described for each line. The DG75 cell-line originates from an EBV negative BL (40) and the RBP-J (CBF1) knockout derivative cell-line was described previously (41). The EBV-positive latency III BL cell line Mutu III (clone 48) derives from Mutu I latency I BL cells grown in culture and has been described previously (42). The EBV immortalized LCL GM12878 is an ENCODE Tier 1 cell line obtained from the Coriell Cell Repositories. The PER253 B95.8 LCL was provided by Dr H. Long and has been described previously (43).The EBV negative BL31 BL cell line series infected with wild-type recombinant EBV bacmids or EBNA 3A, 3B and 3C individual or triple knockout and revertant bacmids has been described previously and was kindly provided by Prof M. Allday (13).

Plasmid construction
The RUNX3 P2 promoter from −737 to +44 relative to the transcription start site was amplified using the following primers (5 ACGCCGCGAGGCCTGCAAGAT 3 and 5 GGCCGCAGCCCCAGAACAAA 3 ) and cloned into PCRII-TOPO. Sequencing detected a single nucleotide change relative to the published sequence (CTTCCGC-CCC has become CTTCCACCCC). A Hind III/Xho I RUNX3 P2 promoter fragment was then cloned into pGL3 basic (Promega) to generate pGL3RUNX3P2. RUNX3 enhancer regions were amplified from bacmid RP11-349B5 (Bacpac Resources) using primers designed to introduce 5 NheI and 3 XhoI sites (Supplementary Table S1). The enhancer 1 (E1) polymerase chain reaction (PCR) product (which also has an Xho1 site at its 5 end), was digested with XhoI and cloned into the Xho1 site of pGL3Runx3P2 to generate pGL3RUNX3P2 E1. All other enhancer regions (E2, E3, E4, E4 + 5, E5 and E6) were amplified by PCR and products digested with NheI and XhoI. NheI/XhoI enhancer fragments were then cloned into the Nhe1 and Xho1 sites of pGL3Runx3P2 to generate pGL3RUNX3P2E2, pGL3RUNX3P2E3, pGL3RUNX3P2E4, pGL3RUNX3P2E4+5, pGL3RUNX3P2E5 and pGL3RUNX3P2E6. pGL3RUNX3P2E1-6, containing all six enhancer regions, was cloned sequentially as follows: E4 + 5 was excised from pGL3RUNX3P1E4 + 5 as an Nhe1/EcoRV fragment, the 5 overhang from Nhe1 was filled in using Klenow (New England Biolabs) and the blunt-ended fragment cloned into pGL3Runx3P2E6 digested with EcoRV to generate pGL3Runx3P2E4-6. E3 was excised from pGL3RUNX3P2E3 as an Nhe1/EcoRV fragment and cloned into pGL3RUNX3P2E4-6 digested with EcoRV using the same strategy to give pGL3Runx3P2E3-6. E2 was then excised from pGL3Runx3P1E2 and inserted into EcoRV digested pGL3Runx3P2E3-6 as a blunted Nhe1/EcoRV fragment to give pGL3RUNX3P2E2-6. Finally, E1 was excised from pGL3RUNX3P2E1 as an XhoI fragment, and cloned into pGL3RUNX3P2E2-6 digested with XhoI to generate pGL3RUNX3P2E1-6. pGL3RUNX1P1 was described previously and contains the RUNX1 P1 promoter from −151 to +100 (44). Enhancer regions 2-5 were amplified from bacmid RP11-749I9 (Bacpac Resources) using primers designed to introduce 5 NheI and 3 XhoI sites (Supplementary Table S1). PCR products for enhancers 2-5 were digested with NheI/XhoI and cloned into pGL3RUNX1P1 cut with Nhe1/Xho1 to give pGL3RUNX1P1E2, pGL3RUNX1P1E3, pGL3RUNX1P1E4 and pGL3RUNX1P1E5. RUNX1 enhancer regions 1 and 6 were synthesized using Ge-neArt Strings R (Invitrogen). Sequences were derived from ENCODE hg19, chr21:36561619-36562555 and chr21:36669712-36670621 respectively and provided in pMA-T vectors. Enhancer region 1 (E1) was excised from the pMA-T vector using AflIII and the 5 overhangs filled with Klenow and inserted into pGL3RUNX1P1 digested with SmaI to generate pGL3RUNX1P1E1. Enhancer region 6 (E6) was excised from the pMA-T vector using MluI and SmaI and the 5 overhang filled with Klenow prior to insertion into the SmaI site of pGL3RUNX1P1 to generate pGL3RUNX1P1E6. pGL3RUNX1P1E1 + 4 + 6 containing RUNX1 enhancers 1, 4 and 6, was cloned sequentially as follows: E1 was excised from pGL3RUNX1P1E1 as an NheI/EcoRV fragment, blunt ended using Klenow and cloned into pGL3RUNX1P1E4 digested with EcoRV to generate pGL3RUNX1p1E4 + 1. E6 was excised from pGL3RUNX1P1E6 as an EcoRV fragment and cloned into pGL3RUNX1p1E4 + 1 digested with EcoRV to generate pGL3RUNX1p1E6 + 4 + 1. pGL3RUNX3P2E3 1 was created by digesting pGL3RUNX3P2E2 with BglII/Xho1 and replacing the excised fragment with a GeneArt TM Strings BglII/Xho1 fragment with the 441 bp hg19 chr1:25 348 801-25 349 241 region of enhancer 2 region deleted.

Site-directed mutagenesis and deletion
The Q5 R Site Directed Mutagenesis Kit (New England Biolabs) was used to generate enhancer mutations. All primers were designed using NEBaseChanger TM software (Supplementary Table S1). To create the RUNX3 enhancer 2 deletion mutant ( 2) 391 nt from 25349261-25349614 inclusive were deleted by designing primers that PCR out from the edges of this region in pGL3RUNX3E2. The NF-B motif in RUNX3 enhancer 2 was mutated by designing primers to make the required substitution (GCAGGGAAGGCCCCA to GCAGGGAAGGGAATA).

Transient transfections
For RUNX1 and RUNX3 promoter reporter assays, DG75 cells were electroporated with plasmid DNA at 230 V and 950 F (BioRad Gene Pulser II) and luciferase assays carried out as described previously (45) using sequential injection on a Glowmax multi detection system (Promega). Cells were transfected with 2 g of the pGL3 luciferase reporter plasmids and 0.5 g pRL-CMV (Promega) as a transfection control, in the absence or presence of 10 or 20 g of the EBNA2-expressing plasmid pSG5 EBNA2A. One tenth of each transfection was processed for western blotting to analyse protein expression levels.

ChIP-QPCR
ChIP-QPCR was carried out as described previously for EBNA2, EBNA3A, EBNA3B and EBNA3C using antibodies verified as specific for each EBNA (16,43,46) and primer pairs across RUNX enhancer regions (Supplementary Table S2) and previously described positive (CTBP2 enhancer) and negative controls (PPIA) (16). RBP-J ChIP was carried out using 4 l STL84 JK, a rabbit polyclonal antibody to RBP-J (provided by Prof E Kieff), following the protocol previously described for polyclonal antibodies (43) with the exception that protein A Sepharose beads were blocked with 0.5% bovine serum albumin (w/v) in phosphate buffered saline.

ChIP-sequencing
EBNA2 was immunoprecipitated from 30 × 10 6 crosslinked GM12878 cells as described previously using the PE2 mouse monoclonal antibody and a rabbit anti-mouse secondary antibody (15,16). A control immunoprecipitation was carried out in parallel using a 1:1 mix of sheep and mouse IgG (Dako). Libraries were prepared using the NEB-Next ChIP-seq library prep reagent set for Illumina and NEBNext Index primers (New England Biolabs) and samples subjected to 50 bp single-end read sequencing with an Illumina Genome Analyzer IIx with a total of seven samples per lane. Data analysis was performed as described previously (15,16). Data are available via GEO accession number GSE76869. Capture Hi-C Previously described capture Hi-C data from GM12878 and CD34+ cells was examined for interactions at RUNX3 that were captured using a 20.1 kb HindIII P1 promoter fragment as bait (chr1:25273787-25293947) (47).

EBNA2 binds a functional long-range RUNX3 superenhancer
To elucidate the mechanism of EBNA2 activation of RUNX3 in EBV infected cells we examined EBNA2 binding data obtained by ChIP-sequencing from two EBV-infected cell lines, one a Burkitt's lymphoma cell line expressing the full panel of EBV latent genes (Mutu III (16)), and the other the Tier 1 ENCODE EBV-immortalized lymphoblastoid cell line GM12878. We identified a cluster of five to six EBNA2 binding sites in an 18-kb region centred at −97-kb upstream from the RUNX3 P2 promoter ( Figure  1A), the promoter active in EBV-infected B lymphoblastoid cells (44). Five main EBNA2 binding sites were detected in Mutu III cells, but an additional sixth site was present in GM12878 cells. EBNA2 binding at these sites was confirmed by ChIP-QPCR ( Figure 1B and C). Examination of ENCODE ChIP-sequencing data for GM12878 revealed high-level H3K27 acetylation (H3K27Ac) across this region ( Figure 1A) and chromatin segmentation analysis is consistent with an active regulatory function. Chromatin landscape analysis (dbSUPER, http://bioinfo.au.tsinghua. edu.cn/dbsuper/ (48)) and recent reports classify this region as a highly-active and characteristically large TF binding site cluster indicative of a lineage-specific super-enhancer (49). The super-enhancer classification of this element is thus far restricted to EBV-infected cells (GM12878), early haematopoietic and T cell lineages (CD3, CD56, CD34+ primary, CD8 primary and CD4 T-cell subsets) and not CD19 primary and CD20 B cells, pointing to EBV infection in driving its activation in B cells.
To further examine the functionality of this superenhancer as a control element for RUNX3, we examined long-range promoter interaction data obtained for CD34+ haematopoietic progenitor and GM12878 cells using capture Hi-C (CHi-C). This technique is a newly-developed modification of the Hi-C genome-wide chromosome conformation method that selectively enriches for interactions involving promoters (47). In GM12878 cells, long-range interactions of the RUNX3 P1 promoter bait captured both the P2 promoter and the super-enhancer region indicating the presence of enhancer-promoter looping consistent with the function of this region as a RUNX3 enhancer in EBVinfected B cells (Figure 2A). In contrast in CD34+ cells, RUNX3 promoter-super-enhancer interactions were absent ( Figure 2B), indicating that the super-enhancer does not make significant contacts with RUNX3 promoters in these cells, despite its prediction as a CD34+ cell super-enhancer region.

The Notch pathway DNA-binding protein RBP-J is necessary but not sufficient for EBNA2 activation of the RUNX3 super-enhancer
To determine which regions of the RUNX3 super-enhancer mediate EBNA2-responsiveness, we cloned each of the six EBNA2 binding regions into luciferase reporter constructs containing the RUNX3 P2 promoter and created a single construct containing all six enhancer regions together. Transient transfection of these constructs into B cells in the absence and presence of EBNA2 demonstrated that the 1.5 kb enhancer region 2 was the key mediator of EBNA2 activation ( Figure 3A). EBNA2 was able to activate transcription up to 3.8-fold via enhancer 2 and up to 8.4-fold when all enhancer regions were combined ( Figure 3A). Enhancers 4 and 6 were activated by EBNA2 up to 2-fold and thus contributed to the increased activation observed in the presence of all six enhancers. The EBV C promoter was used as a positive control and was activated up to 5.5-fold by EBNA2 ( Figure 3A). Interestingly, none of the enhancer regions were able to increase basal transcription from P2 in the absence of EBNA2 and some enhancer regions (E1, E3, E4 and E5) decreased basal transcription ( Figure 3B). Western blotting confirmed expression of EBNA2 across the different transfections at similar levels (Supplementary Figure  S1).
Since enhancer 2 was the major mediator of EBNA2 responsiveness, we next investigated which cellular TFs were responsible for mediating EBNA2 binding and activation via this enhancer. Examination of ENCODE GM12878 Factorbook ChIP-sequencing data for TFs with a motif within binding sites in enhancer 2 revealed EBF1, PAX5, USF1 and RUNX3 had maximum cluster scores for binding, with BATF also bound at high levels (Supplementary Figure S2). A number of other TFs bound at sites with motifs but with lower cluster scores for binding, including the NFB TF RelA (Supplementary Figure S2) previously implicated as a RUNX3 transcription regulator (49). Since a previous ChIP-sequencing study detected binding of the cellular Notch pathway DNA-binding protein and EBNA2 binding partner, RBP-J, at the RUNX3 super-enhancer in the EBV-immortalized LCL IB4 (33), we investigated whether RBP-J bound the RUNX3 superenhancer in Mutu III and GM12878 cells. ChIP-QPCR analysis demonstrated that RBP-J bound to the RUNX3 super-enhancer with maximum binding at enhancer 2 in both cell types ( Figure 3C and D) and we identified three candidate RBP-J binding sites within this region ( Figure  4A).
To determine which of the candidate cellular TFs may play a role in EBNA2 activation via enhancer 2, we performed an initial deletion analysis. Since the TF binding sites at enhancer 2 were arranged in two clusters, we created enhancer 2 reporter constructs deleted for each cluster ( Figure 4A). Transient reporter assays using these deletion mutants demonstrated that loss of either cluster ablated the ability of EBNA2 to activate transcription via enhancer 2 ( Figure 4B and Supplementary Figure S1). The RUNX3 super-enhancer is bound by multiple NF-B subunits and RUNX3 gene expression is reduced in EBV-infected cells upon inactivation of NF-B (49), implicating NF-B in  RUNX3 super-enhancer control. To determine whether the NF-B site in cluster 2 of RUNX3 enhancer 2 contributed to EBNA2 activation of the RUNX3 super-enhancer, we performed luciferase assays using an enhancer 2 reporter construct with the NF-B site mutated ( Figure 4A). We found that the ability of EBNA2 to activate RUNX3 P2 transcription via enhancer 2 was unaffected by mutation of this site, indicating that NF-B subunits do not play a role in mediating the effects of EBNA2 ( Figure 4C). Interestingly however, basal transcription levels of the RUNX3 P2 enhancer 2 construct in the absence of EBNA2 were reduced by ∼50% leading to an equivalent reduction in level of RUNX3 transcription even in the presence of EBNA2 ( Figure 4D). These data therefore support the previous observations that loss of NF-B leads to a 50% reduction in overall RUNX3 transcription in EBV-infected cells but rule out a role for NF-B in RUNX3 activation by EBNA2 via enhancer 2.
To further investigate a role for RBP-J in the activation of RUNX3 by EBNA2, we examined the ability of EBNA2 to activate transcription via RUNX3 super-enhancer elements in an RBP-J knock-out B-cell-line (41). Our results demonstrated that in the absence of RBP-J EBNA2 was no longer able to activate RUNX3 either via enhancer 2 or all six combined enhancer regions ( Figure 5A and B). Control experiments using the EBV C promoter which is activated by EBNA2 in an RBP-J-dependent manner showed the same ablation of EBNA2 activation in RBP-J knockout cells (50) (Figure 5C and Supplementary Figure S1). Con- sistent with the previously observed partial dependency of the viral LMP1 promoter on RBP-J (30), this promoter was still partially responsive to EBNA2 in RBP-J KO cells, displaying a 2.8-fold response to EBNA2 compared to a 6.7fold activation in wild-type cells ( Figure 5D and Supple- mentary Figure S1). Taken together, our data indicate that RBP-J is necessary but not sufficient for EBNA2 activation of RUNX3 and implicate further TFs other than NF-B in cluster 2 of enhancer 2 in mediating EBNA2 responsiveness in cooperation with RBP-J.

EBNA3B and EBNA3C coactivate RUNX3 by targeting super-enhancer element 2
In addition to the identification of EBNA2 human genome binding sites, our previous ChIP-sequencing experiments also examined binding of EBNA3A, 3B and 3C across the genome (15,16). Examination of EBNA3 binding at the RUNX3 locus revealed that EBNA3 proteins also target the RUNX3 super-enhancer. Consistent with the key role of enhancer 2, EBNA3 binding was predominantly local-ized to enhancer 2 ( Figure 6A). Since we have previously demonstrated that distinct subsets of EBNA3 proteins target specific gene regulatory elements, we performed ChIP-QPCR using individual antibodies specific to EBNA3A, 3B and 3C to examine the binding of these three viral TFs at RUNX3 enhancer 2. We found that only EBNA3B and 3C bound to this region ( Figure 6B-G). Examination of previously published microarray analysis (11) confirmed a role of EBNA3 proteins in the regulation of RUNX3 transcription. Previous analysis of EBV negative BL cells infected with wild type EBV bacmids or EBV bacmids deleted for the EBNA3A, 3B or 3C genes individually or in combination showed that infection with wild-type EBV led to a 3.3fold upregulation of RUNX3 mRNA levels, but infection with viruses lacking all EBNA3 proteins resulted in 2.4-fold lower expression than in wild-type EBV infected cells (11). We confirmed that these effects were also evident at the protein level ( Figure 6H). RUNX3 protein levels were upregulated 2.5-fold on infection of EBV negative BL31 cells with wild-type EBV, with cells infected with revertant viruses expressing similar levels to wild type infected cells, as expected ( Figure 6H). In contrast, cells infected with a virus lacking EBNA3A, 3B and 3C displayed a 1.8-fold reduction in RUNX3 protein expression compared to the corresponding revertant cell-line ( Figure 6H). Loss of EBNA3B or EBNA3C individually had no effect on the upregulation of RUNX3 indicating that EBNA3B and EBNA3C can independently activate RUNX3 transcription at the mRNA or protein level (11) ( Figure 6H). RUNX3 transcription in EBV-infected cells is therefore upregulated through targeting of the upstream super-enhancer by EBNA2, EBNA3B and EBNA3C.

Feed-forward activation of RUNX1 by EBNA2 via an upstream super-enhancer
Previous studies have shown that upregulation of RUNX3 by EBNA2 results in reduced RUNX1 transcription through the repressive effects of RUNX3 binding to the RUNX1 P1 promoter, the RUNX1 promoter active in B cells (44). In addition to this indirect control of RUNX1 transcription, our ChIP-sequencing analysis has now revealed that EBNA2 also binds to six sites in a region 139-250kb upstream from RUNX1 P1 pointing to a role in the direct control of RUNX1 (Figure 7). Interestingly, EBNA2 appears to target this region in a cell-type specific manner, with binding detected in Mutu III BL cells and not in GM12878 or other LCLs (Figure 7 and data not shown) as confirmed by ChIP-QPCR. This region has not been previously described as a control region for RUNX1 but analysis revealed that areas within this region encompassing EBNA2 binding sites are classified as super-enhancers in specific cell backgrounds, including the Diffuse large B cell lymphoma cell-lines Toledo and HBL1, the Breast cancer line HCC1954, mammary epithelial cells and skeletal muscle (dbSUPER, http://bioinfo.au.tsinghua.edu.cn/dbsuper/ (48)). Consistent with the lack of H3K27Ac and EBNA2 binding in the GM12878 LCL (Figure 7), this region is not predicted to be a super-enhancer in these cells.
Cloning the six EBNA2 binding regions into luciferase reporter constructs containing the RUNX1 P1 promoter re-   vealed that the 2.6-kb Enhancer 4 region was the major determinant of EBNA2 responsiveness in the RUNX1 superenhancer, with EBNA2 able to activate transcription via enhancer 4 up to 4.3-fold ( Figure 8A). Since enhancers 1 and 6 directed small responses to EBNA2 (although not reaching significance for enhancer 1), we also created a reporter construct containing enhancers 1, 4 and 6 together. EBNA2 activation of this construct was slightly increased to 5-fold ( Figure 8A). None of the enhancer regions were able to increase basal transcription from RUNX1 P1 in the absence of EBNA2 and in fact E1, E4 and E1,4,6 regions decreased basal transcription from this promoter ( Figure 8B). Western blotting confirmed expression of EBNA2 across the different transfections at similar levels (Supplementary Figure  S1). Our data therefore implicate RUNX1 super-enhancer targeting by EBNA2 in the activation of RUNX1 expression in certain cell backgrounds.

EBNA2 activation of the RUNX1 super-enhancer requires RBP-J
To investigate the role of RBP-J in activation of the RUNX1 super-enhancer region, we examined RBP-J binding at the six RUNX1 super-enhancer peaks in Mutu III cells using ChIP-QPCR and found that RBP-J bound at the highest levels at enhancer 1, 4 and 6 ( Figure 8C). To determine whether EBNA2 was able to activate the RUNX1 enhancer in the RBP-J knockout B-cell-line, we carried out luciferase assays using a RUNX1 P1 reporter construct and constructs also containing enhancer 4 or enhancers 1, 4 and 6 (Figure 9A). Similar to our observations for EBNA2 activation of RUNX3 enhancer elements, we found that EBNA2 ac-tivation of RUNX1 super-enhancer elements was also dependent on RBP-J ( Figure 9A). EBNA2 activation via enhancer 4 and enhancers 1, 4 and 6 in RBP-J knockout cells was reduced to levels that were not significantly different from the low-level EBNA2 activation observed for constructs containing the RUNX1 P1 promoter alone (Figure 9A). To determine whether the lack of EBNA2 binding to RUNX1 super-enhancer elements in GM12878 cells was due to low RBP-J levels, we examined RBP-J protein expression in Mutu III and GM12878 cells ( Figure 9B and C). In fact, RBP-J was expressed at 1.9-fold higher levels in GM12878 so the availability of RBP-J was not a likely explanation. However, RUNX3 protein levels were also 3fold higher in GM12878 cells ( Figure 9B and C). It is therefore possible that higher levels of RUNX3 lead to increased repression of the RUNX1 locus in GM12878 cells through the previously described inhibitory effects of RUNX3 on the RUNX1 P1 promoter (44). Taken together our results indicate that the major mediator of EBNA2 activation of RUNX1 enhancer regions is RBP-J, pointing to RBP-J as the key cellular TF hijacked by EBV to control RUNX1 and RUNX3 transcription in EBV-infected cells.

RUNX1 super-enhancer binding by EBNA3B and EBNA3C attenuates EBNA2 activation in BL cells
EBNA3 ChIP-sequencing also identified binding of EBNA3 proteins at enhancers 1, 3 and 4 in the RUNX1 super-enhancer in Mutu III BL cells ( Figure 10A). ChIP-QPCR with individual EBNA3 antibodies demonstrated that EBNA3B and EBNA3C but not EBNA3A bound these super-enhancer sites ( Figure 10B-D). In fact, different patterns of binding were observed for EBNA3B and 3C, with peak binding of EBNA3B at enhancer 4 and peak binding of EBNA3C at enhancer 3. Our results therefore demonstrate that EBNA2, EBNA3B and EBNA3C target both RUNX3 and RUNX1 super-enhancers, but at RUNX1 binding is cell-type specific. To investigate what effects EBNA3B and 3C binding may have on RUNX1 expression in BL cells, we again examined microarray data from the EBV negative BL cell series infected with wild-type EBV or EBNA3 knockout viruses (11). These data revealed a clear role for EBNA3B and 3C in repressing RUNX1, since in the absence of EBNA3B there was a 2.2-fold increase in RUNX1 mRNA levels compared to wild-type EBV-infected cells and in the absence of EBNA 3C a 1.8-fold increase in mRNA levels was detected (11). We confirmed that the repressive effects of EBNA3B and EBNA3C on RUNX1 in this cell series were also evident at the protein level ( Figure 10E). Infection of EBV negative BL31 cells with wild-type EBV or revertant viruses reduced RUNX1 protein expression as expected, but infection with EBNA3B knockout viruses resulted in 6.7-fold higher protein levels than wild-type EBV infected cells (Figure 10E). RUNX1 protein expression in cells infected with EBNA3C knockout viruses was also 2.2-3.7-fold higher than in cells infected with wild-type EBV ( Figure 10E). Since RUNX3 represses RUNX1 expression, the role of EBNA3 proteins as activators of RUNX3 expression would be predicted to lead to a reduction in RUNX1 mRNA levels. However, in cell lines infected with individual EBNA3B or 3C knockout viruses, no reduction in RUNX3 mRNA levels were observed ( Figure 6H). Thus indirect elevation of RUNX1 expression due to reduced RUNX3 expression in these cell lines can be ruled out. The observed effects of EBNA3B and 3C on RUNX1 expression are therefore most likely to reflect direct effects via their binding to RUNX1 super-enhancer elements. Our data therefore highlight EBNA3B and 3C as additional regulators of RUNX3 and RUNX1 expression in EBV-infected cells. Consistent with their previously reported dual roles as activators and repressors of cellular gene expression (11), in the context of RUNX gene regulation they function as activators of RUNX3 but repressors of RUNX1.

DISCUSSION
Studying how RUNX transcription is regulated by the EBNA2 and EBNA3 family of EBV TFs in EBV-infected B cells, we have discovered that the Notch pathway DNAbinding protein RBP-J is the key cellular factor hijacked by EBNA2 to direct activation of both RUNX3 and RUNX1 transcription via upstream super-enhancers. We also show that EBNA3B and 3C regulate both RUNX3 and RUNX1 transcription in opposing directions to fine-tune RUNX gene expression in EBV-infected cells, maintaining high RUNX3 expression and low RUNX1 expression to prevent inhibition of EBV-immortalized cell growth by RUNX1 ( Figure 11).
Analysis of genome-wide sequencing data for long-range chromatin interactions captured by promoter baits (CHi-C) in the GM12878 EBV-immortalized LCL provided clear evidence that the RUNX3 super-enhancer region interacts with RUNX3 promoters. This supports the functional relevance of the RUNX3 super-enhancer as a long-range regulatory element hijacked by EBV to control RUNX3 transcription. It was not possible to obtain information on RUNX1 P1 or P2 promoter interactions from the CHi-C GM12878 data set since a far upstream alternative and inactive RUNX1 promoter was used as bait. Given that the Figure 11. RUNX gene regulation by the EBNAs in EBV-infected cells. EBNA2, EBNA3B and EBNA3C activate RUNX3 expression by binding to a distal upstream super-enhancer. RUNX3 then represses RUNX1 transcription by binding to the RUNX1 P1 promoter. EBNA2 can activate RUNX1 transcription via an upstream super-enhancer, but this is celltype specific. EBNA3B and EBNA3C repress RUNX1 transcription in a cell-type specific manner. Total RUNX1 expression levels depend on the balance between the level of RUNX3 expression (driving RUNX1 repression) and EBNA2 activation and EBNA3B and 3C repression mediated via super-enhancer binding. Nucleic Acids Research, 2016, Vol. 44, No. 10 4647 EBNAs do not bind the upstream RUNX1 super-enhancer region in GM12878 cells, these data would have been likely to provide evidence for a lack of enhancer-promoter interactions in this cell line in any case.
We have delineated the regions of RUNX superenhancers that are the major determinants of EBNA2 responsiveness. For both genes one key region (enhancer region 2 for RUNX3 and enhancer region 4 for RUNX1), corresponding to an EBNA2 binding peak detected in ChIP-sequencing, directs the majority of the EBNA2 activation in enhancer-promoter reporter assays using endogenous RUNX promoters. Consistent with the RBP-Jdependent EBNA2 activation of these enhancer elements, they were both bound by RBP-J in EBV-infected cells, along with a number of other B cell TFs, supporting their roles as active regulatory elements. Interestingly, our ChIPsequencing analysis detected five binding sites for EBNA2 at the RUNX3 super-enhancer in the latency III BL cell line Mutu III, but six sites in GM12878, cells consistent with the histone acetylation pattern at this region in GM12878. Sequencing of these regions from both cell-lines detected a substitution in the Mutu III sequence within a putative low affinity RBP-J motif that reduced RBP-J binding to this site in vitro (data not shown). RBP-J bound this site in GM12878 and not Mutu III cells, but at lower levels than enhancer 2, consistent with the weak responsiveness of enhancer region 6 to EBNA2 in reporter assays. It is therefore likely that the lack of binding of EBNA2 to this region in Mutu III cells results from this sequence variation, although the overall activity of the RUNX3 super-enhancer is unlikely to be particularly affected as a result.
For RUNX3, additional cellular TFs in a region adjacent to a cluster of three potential RBP-J sites in enhancer region 2 also appear to play a cooperative role in gene activation by EBNA2 as deletion of this region also attenuates EBNA2 effects. We focused on TFs with ChIP-sequencing data available through ENCODE for the GM12878 LCL that also have a motif at their binding site to reduce the number of candidates. This highlighted EBF1, RUNX3, BATF and RelA as potential cooperative factors. Since NF-B subunits including RelA have been previously implicated in RUNX3 transcription control in EBV-infected cells (49), our initial analysis examined the effect of mutating the NF-B site in enhancer 2. Consistent with this previous study, we found that loss of this site reduced overall enhancermediated transcription from the RUNX3 P2 promoter, but did not impair the ability of EBNA2 to activate transcription. NF-B binding to enhancer 2 within the RUNX3 super-enhancer therefore appears to play a key role in maintaining enhancer-driven but EBNA2-independent RUNX3 transcription in B cells. A full understanding of which regions or TF binding sites within the region adjacent to the cluster of RBP-J sites contribute to EBNA2 activation will require further extensive and systematic mutation of the TF binding sites we have identified based on initial screens for high-level binders with motifs from ENCODE, but likely also deletions and combinations of site mutations across this ∼400-bp region, since many other TFs also bind at lower levels (Supplementary Figure S2), and an even larger number bind in the absence of their canonical motifs.
We have uncovered a new role for EBNA3B and 3C acting alongside EBNA2 in EBV-infected cells to maintain high RUNX3 expression. The effects of EBNA3 proteins on RUNX3 transcription would not have been apparent in previous studies that detected RUNX3 as a direct target for activation by EBNA2 using EBV infected cell-lines expressing conditionally active EBNA2 in the context of the full complement of latent proteins including the EBNA3s, since the loss of EBNA2 alone is clearly sufficient to reduce RUNX3 transcription (51). The independent ability of EBNA2 to activate RUNX3 transcription is also evident from the RUNX3 enhancer reporter assays described here. In contrast to our detection of the binding of endogenous EBNA3B and 3C and not EBNA3A at the RUNX3 super-enhancer, a previous study documented EBNA3A (and EBNA3C) binding to this region in LCLs infected with recombinant viruses expressing HA/Flag-tagged EBNA3A or 3C proteins (49). There is however, no evidence for a role for EBNA3A in the regulation of RUNX3 gene expression in BL cells or LCLs to date (11,52).
Our EBNA3B and 3C binding data indicate that the most responsive EBNA2 enhancer, enhancer 2, is the major EBNA3 binding site. Since this enhancer is bound by RBP-J, its activation by EBNA2 is RBP-J dependent, and EBNA3B and 3C bind RBP-J, it is likely that the activation of RUNX3 transcription by these EBNA3 proteins is also mediated by RBP-J. EBNA2 and EBNA3 proteins associate with RBP-J in a mutually exclusive manner (25) and we have previously shown using re-ChIP that EBNA2 and 3 proteins do not co-occupy the same sites on DNA at the same time (16). In addition, further studies have now convincingly demonstrated that EBNA2 and EBNA3 proteins bind competitively to at least a specific subset of RBP-J-bound cell chromatin sites (53,54). For example, at an intergenic RBP-J-bound enhancer site located between CXCL9 and CXCL10 also bound by both EBNA2 and EBNA3A, conditional inactivation of EBNA3A led to a 6-fold increase in EBNA2 binding (53). Recent similar studies of RBP-J-bound genomic sites also bound by either EBNA3A (near HDAC7 or CDH1) or EBNA3C (near BACH2, JAK1, CXCR5) found that conditional inactivation of these EBNA3 proteins allowed EBNA2 binding via RBP-J (54). Thus at the RUNX3 superenhancer, EBNA2 and EBNA3 proteins likely bind this site in different cells in the population used for ChIP-sequencing. For RUNX3 it appears that EBNA3 binding, like EBNA2 binding, leads to RUNX3 activation. Consistent with our previous unpublished observations and those of others, we have been unable to reproduce EBNA3 protein-mediated cellular gene activation in reporter assays to date so the mechanisms and factors involved in EBNA3-mediated RUNX3 activation will require further analysis in a chromatin context that recapitulates the observed in vivo effects.
Our ChIP-sequencing analysis uncovered further levels of regulation of the RUNX gene network that involve the direct effects of both EBNA2 and EBNA3B and 3C on RUNX1 transcription, with the effects of EBNA2 again mediated through RBP-J. This direct regulation was not evident from previous studies using transcriptomics approaches alone due to the upregulation of RUNX3 in EBVinfected cells and the resulting repression of RUNX1 by RUNX3 that would have masked any effects of the EBNAs on RUNX1. Our data reveal that this direct regulation of the RUNX1 super-enhancer also occurs in a cell-type dependent manner, consistent with the previously reported celltype specific nature of super-enhancers (55).
This cell-type specificity could be controlled by the overall level of RUNX3 in the cells which would determine the level of repression of the RUNX1 locus. Although RUNX3 represses RUNX1 transcription through a site in the P1 promoter (44), it is possible that this repression would lead to an inhibition of enhancer-promoter interactions and the subsequent 'closure' of enhancer chromatin. This would result in reduced binding of cellular TFs like RBP-J and therefore prevent EBNA binding. This theory is supported by the higher level of RUNX3 expression in GM12878 compared to Mutu III. Further understanding of how RUNX3 may repress the entire RUNX1 locus and limit super-enhancer activity will require studies of the effects of RUNX3 on RUNX1 promoter-super-enhancer interactions and cellular TF binding in B cell systems lacking EBNA2 to allow RUNX3 repression to be examined independently of EBNA2 activation. It is also possible that MYC may play a role in the differential activity of the RUNX1 superenhancer in BL cells compared to LCLs, since BL cells express high levels of MYC as a result of the MYC-IG translocation. ENCODE ChIP-seq data indicates that both MYC and its partner MAX bind at many sites across the large RUNX1 super-enhancer region in a number of cell-types. Interestingly, MYC/MAX binding is not detected in this region in the GM12878 LCL where MYC expression is low. Given the co-operative pioneer activity of MYC (56), binding of MYC/MAX to the RUNX1 super-enhancer region in BLs could increase accessibility for other TFs.
The role of EBNA3B and 3C as repressors of RUNX1 transcription probably serves as a mechanism to curb highlevel activation of RUNX1 by EBNA2, although why the remaining expression of RUNX1 induced by EBNA2 does not result in the growth arrest of BL cells is unclear. It is possible that BL cells tolerate higher level RUNX1 expression than LCLs because BL cell growth is driven by deregulated MYC. Since RUNX1 has conflicting roles in the regulation of cell growth in different backgrounds, it is also possible that there may be some advantage to higher level of expression of RUNX1 in BL. A full understanding of the role of RUNX1 in controlling B cell growth will await the outcome of genome-wide binding studies that are currently underway.
RBP-J plays an essential role in the immortalization of B cells by EBV and EBNA2, 3A or 3C proteins unable to interact with RBP-J fail to immortalize B cells and/or support the proliferation of infected cells (22)(23)(24)57). Our data indicate that regulation of the RUNX gene network through super-enhancer control via RBP-J may contribute to the dependency on RBP-J for growth deregulation by EBNA2 and EBNA3C, given the repressive effects of RUNX1 on B cell growth. EBNA3B plays a non-critical role in EBVmediated transformation since EBNA3B deleted viruses are able to transform B cells (58), although it can also associate with RBP-J (21). Nonetheless EBNA3B is a major controller of cellular gene expression, often acting together with other EBNA3 proteins and may have a tumour suppres-sive role (11,59). Interestingly, extensive interplay between Notch and RUNX pathways has been reported across a number of developmental systems. Notch signalling during Drosophila haematopoiesis induces Lozenge (RUNX1) expression in specific precursor cells (60), Notch1 upregulates RUNX1 transcription in NIH3T3 cells and Notch1-RBP-J null mice display reduced RUNX1 expression in para-aortic splanchnopleural cells and impaired haemopoietic potential (61). Notch signalling also controls RUNX1 expression and haematopoietic stem cell development in Zebrafish (62) and RUNX1 was identified as a direct target induced by activated Notch1 signalling in murine mesodermal cells, but not embryonic stem cells (63). RUNX3 has also been shown to be a direct target of Notch in murine endothelial cells (64). Our data therefore identify an important role for RBP-J and therefore the Notch pathway in the control of RUNX transcription via long-range regulatory elements in human B-cells that is likely to extend to other cell-types.
In summary, our data reveal a complex cross-regulatory RUNX gene network controlled by the Notch signalling component RBP-J that is hijacked through interactions with EBV-encoded TFs to fine-tune RUNX3 and RUNX1 gene expression and control B-cell growth. We identify RUNX1 super-enhancer control regions involved in RUNX1 regulation in B cells that may also function as key control regions in other cell-types. A greater understanding of how RUNX genes are controlled by long-range enhancers will help to delineate how these genes are regulated during normal developmental processes and become deregulated in disease.