Synthetic Lethal Screen Demonstrates That a JAK2 Inhibitor Suppresses a BCL6-dependent IL10RA/JAK2/STAT3 Pathway in High Grade B-cell Lymphoma*

We demonstrate the usefulness of synthetic lethal screening of a conditionally BCL6-deficient Burkitt lymphoma cell line, DG75-AB7, with a library of small molecules to determine survival pathways suppressed by BCL6 and suggest mechanism-based treatments for lymphoma. Lestaurtinib, a JAK2 inhibitor and one of the hits from the screen, repressed survival of BCL6-deficient cells in vitro and reduced growth and proliferation of xenografts in vivo. BCL6 deficiency in DG75-AB7 induced JAK2 mRNA and protein expression and STAT3 phosphorylation. Surface IL10RA was elevated by BCL6 deficiency, and blockade of IL10RA repressed STAT3 phosphorylation. Therefore, we define an IL10RA/JAK2/STAT3 pathway each component of which is repressed by BCL6. We also show for the first time that JAK2 is a direct BCL6 target gene; BCL6 bound to the JAK2 promoter in vitro and was enriched by ChIP-seq. The place of JAK2 inhibitors in the treatment of diffuse large B-cell lymphoma has not been defined; we suggest that JAK2 inhibitors might be most effective in poor prognosis ABC-DLBCL, which shows higher levels of IL10RA, JAK2, and STAT3 but lower levels of BCL6 than GC-DLBCL and might be usefully combined with novel approaches such as inhibition of IL10RA.

We demonstrate the usefulness of synthetic lethal screening of a conditionally BCL6-deficient Burkitt lymphoma cell line, DG75-AB7, with a library of small molecules to determine survival pathways suppressed by BCL6 and suggest mechanismbased treatments for lymphoma. Lestaurtinib, a JAK2 inhibitor and one of the hits from the screen, repressed survival of BCL6deficient cells in vitro and reduced growth and proliferation of xenografts in vivo. BCL6 deficiency in DG75-AB7 induced JAK2 mRNA and protein expression and STAT3 phosphorylation. Surface IL10RA was elevated by BCL6 deficiency, and blockade of IL10RA repressed STAT3 phosphorylation. Therefore, we define an IL10RA/JAK2/STAT3 pathway each component of which is repressed by BCL6. We also show for the first time that JAK2 is a direct BCL6 target gene; BCL6 bound to the JAK2 promoter in vitro and was enriched by ChIP-seq. The place of JAK2 inhibitors in the treatment of diffuse large B-cell lymphoma has not been defined; we suggest that JAK2 inhibitors might be most effective in poor prognosis ABC-DLBCL, which shows higher levels of IL10RA, JAK2, and STAT3 but lower levels of BCL6 than GC-DLBCL and might be usefully combined with novel approaches such as inhibition of IL10RA.
There is a need for new treatments for poor-prognosis activated B-cell-like diffuse large B-cell lymphoma (ABC-DLBCL), 6 which continues to have a cure rate Ͻ40% with conventional chemotherapy (1).
The majority of ABC-DLBCL, in contrast to germinal center B-cell-like DLBCL, have low level expression of BCL6 mRNA and protein (2). JAK/STAT3 signaling is active in ABC-DLBCL and is enhanced by constitutive activity of the NF-B pathway (3), which in turn is driven by oncogenic CARD11 mutations (4), chronic active B-cell receptor signaling (5), and MYD88 mutations (6). However, other factors are also likely to be important in determining the overall activity of JAK/STAT3 signaling. BCL6 directly represses both STAT3 (7) and NF-B p105/p50 (8) transcription, and levels of BCL6 might, therefore, be a factor independent of the known oncogenic mutations, which determines the activity of signaling pathways required by ABC-DLBCL. In this report we develop a novel B-cell line to pursue the hypothesis that genes repressed by BCL6 are components of survival signaling pathways in lymphomas with low level expression of this transcription factor.
BCL6 is a zinc finger transcription factor that is highly expressed in normal germinal center B-cells (9) and is required for high affinity antibody production (10,11). It is also constitutively expressed in ϳ40% of cases of the high grade B-cell lymphoma DLBCL due to either chromosomal translocations, mutations of a negative regulatory site in the promoter region (12)(13)(14), or abnormalities of post-translational regulation (15)(16)(17).
The N-terminal POZ domain of BCL6 associates with co-repressors NCOR1, BCOR, and SMRT (NCOR2), which in turn recruit histone deacetylases to accomplish transcriptional repression. Work largely carried out with human Burkitt lymphoma cell lines and mouse B-cell lines showed that BCL6 represses B-cell terminal differentiation through a direct effect on BLIMP1 (PRDM1) (18 -20). Effects on the cell cycle have been less clearly defined; BCL6 represses cyclin D2 transcription (18), but it also suppresses the cyclin-dependent kinase inhibitor p21 (21) and in primary cells prevents senescence and induces cyclin D1 (22). Suppression of DNA damage responses and p53 by BCL6 are believed to be required to allow somatic hypermutation in normal germinal center B-cells (23,24). As mentioned above, STAT3 and NF-B, which are both required by ABC-DLBCL, are direct targets of BCL6 transcriptional repression (7,8).
We engineered a conditional BCL6-deficient human B-cell line, DG75-AB7, from an EBV-negative Burkitt lymphoma cell line proficient for homologous recombination, DG75 (25,26). We employed DG75-AB7 in a synthetic lethal screen to determine compounds inhibiting pathways that are active in BCL6deficient cells. Synthetic lethal screening has previously been employed to find agents that are preferentially effective in the killing of cell lines bearing a transforming mutation (27) and is more widely the basis for siRNA and high throughput drug screens to discover novel therapeutics for cancer (28).
For this study we defined synthetic lethality as reduction of survival of BCL6-deficient DG75-AB7 cells while relatively sparing BCL6-replete cells. To carry out the synthetic lethal screen we employed a library of small molecule inhibitors that are either in clinical use or are good candidates for clinical use. Hits from the screen could, therefore, potentially be rapidly introduced into clinical trials. We demonstrate that IL10RA and JAK2 are transcriptionally regulated by BCL6 and suggest (7,29) that low BCL6 is a determinant of an IL10RA/JAK2/ STAT3 pathway that might be important in survival of some ABC-DLBCL.

Results
Characterization of a Conditional BCL6-deficient B-cell Line-BCL6 deficiency reduces survival and causes accumulation in G 1 .
We produced conditional BCL6-deficient Burkitt lymphoma cell lines. On the addition of doxycycline BCL6 was effectively repressed, and proliferation was reduced in three separate clones (Fig. 1, C and D). Further work was carried out with clone DG75-AB7. When doxycycline was washed out of the culture medium, the effects on growth were reversed, suggesting a continuing requirement for BCL6 (Fig. 1E). After 7 days of culture doxycycline-treated DG75-AB7 cells had undergone fewer cell divisions than untreated cells ( Fig. 2A) with an accompanying reduction in survival and an increase in cells in G 1 at the expense of G 2 M and S phases (Fig. 2B). Overall BCL6-deficient cells showed reduced proliferation due to G 1 growth arrest.
Gene expression profiling was carried out to obtain a comprehensive view of BCL6 gene target alterations in DG75-AB7. 205 genes with a Ն2-fold up-or down-regulation in at least one of the samples cultured with doxycycline (16, 48, or 96 h) compared with the basal (Ϫdoxycycline) sample were identified. 14/205 genes were not annotated and were hence excluded from subsequent analysis, leaving 191 genes. Of these, 162 genes (85%) were up-regulated in response to BCL6 depletion (Table 1). Validation by RT-polymerase chain reaction (PCR) was carried out for a subset of these genes ( Fig. 3 and Table 2).
Work by others has demonstrated specific functionally important BCL6 targets (18). Analysis of changes in DG75-AB7 FIGURE 1. A conditional BCL6-deficient human B-cell line. A, scheme of disruption of the endogenous BCL6 loci of DG75. A zeocin resistance cassette flanked by loxP sites was inserted into exon 3 of the BCL6 gene, containing the initiation codon by homologous recombination. After Cre recombinase mediated removal of the antibiotic resistance gene the process was repeated on the second BCL6 allele. B, Southern demonstrating disruption of the endogenous BCL6 loci and the presence of the inserted BCL6 transgene. Genomic DNA was digested with SphI and probed with labeled probes A and B. Wild-type cells (BCL6 ϩ/ϩ , lane 1) gave an 11.8-kb band. After the first round of targeting and before removal of the zeocin resistance cassette (BCL6ϩ/Ϫ Zeo r , lane 2) the 11.8 kb band was present with 7.9-kb and 5.1-kb bands from the targeted locus. After removal of the zeocin resistance cassette but before targeting of the second locus (BCL6 ϩ/Ϫ , lane 3) 11.8-kb and 5.1-kb bands were present. After insertion of the BCL6 transgene and targeting of the second locus (BCL6 Ϫ/Ϫ Zeo r ϩ pTRE-BCL6HA, lane 4) the 11.8-kb band was no longer present, bands of 7.9 kb and 5.1 kb from the disrupted BCL6 loci were present together with a band of 4.1 kb from the BCL6 transgene. C, growth curves (cumulative doublings) of wild-type DG75 (WT) and three targeted clones (AB7, CD9, and FA8) in the absence (solid line) and presence (dotted line) of doxycycline. D, Western showing BCL6 expression in DG75-AB7 in the presence and absence of doxycycline. BCL6 expression in the DLBCL cell line, SUDHL4 (BCL6 expressing), and HEK293 cells (BCL6 non-expressing) are also shown. E, the effects of doxycycline on proliferation are reversible. DG75-AB7 cells were cultured in the presence or absence of doxycycline or with doxycycline for various times (2, 4, or 6 days) before exchanging with culture medium lacking the antibiotic. Proliferation rate (cumulative doublings) were normalized to proliferation of cells cultured without doxycycline.
to the mRNA expression of these genes showed Ն30% induction of expression in 14/19 (73%) on the addition of doxycycline (Table 3A). In an alternative method to validate the gene expression changes observed in DG75-AB7, we utilized published data, which in another Burkitt lymphoma cell line has demonstrated a set of genes that bind BCL6 at their genomic loci (34). 39 of the 44 genes that bound BCL6 by ChIP-quantitative PCR are represented on the gene expression microarrays we employed, and 18 (46%) of these showed altered mRNA expression (Ն30% induction or repression) at one or more time points after the addition of doxycycline to DG75-AB7 (Table  3B). Although not all genes discovered through ChIP-chip will be functionally important, our data support a functional role for several e.g. TNFAIP8, TAP1, SUB1, and CD53, that have not yet been investigated in detail.
One of the important effects of BCL6 is suppression of DNA damage responses partly through transcriptional repression of ATR (23). To show that DG75-AB7 reproduces this aspect of BCL6 deficiency, DNA damage responses in response to x-irradiation were determined. Culture in doxycycline caused induction of ATR protein in AB7 and significant (Mann-Whitney U test) reductions in DNA damage in response to x-irradiation (as determined by H2AX phosphorylation) at 1 Gy (p ϭ 0.003), 2 Gy (p ϭ 0.007), and 4 Gy (p ϭ 0.01) (Fig. 2C). Therefore, DG75-AB7 is a model system demonstrating gene expression and functional changes on the addition of doxycycline, in line with known BCL6 effects.
The JAK2 Inhibitor, Lestaurtinib, Reveals a Survival Pathway in BCL6-deficient Cells-We utilized a drug sensitivity screen to determine survival pathways in BCL6-deficient DG75-AB7. The effect of each compound in the library on cell viability was after culture with doxycycline measured by a luminescence cell viability assay and cell cycle changes after culture of DG75-AB7 in the absence (Ϫ) or presence of doxycycline (Dox). Cells were stained with propidium iodide, and cell cycle analysis was carried out with FlowJo Software. C, Western blot demonstrating changes in expression of the BCL6 target gene, ATR on culture of DG75-AB7 with doxycycline (Do). DNA repair capacity in DG75-AB7 was repressed by BCL6. Levels of phosphorylated H2AX after x-irradiation in either the absence (red bars) or presence (blue bars) of doxycycline were determined by flow cytometry (n ϭ 3). H2AX levels specifically in G 0 /G 1 of the cell cycle are presented. There are significant differences in levels of phosphorylated H2AX (Mann-Whitney U test) at 1 Gy (p ϭ 0.003), 2 Gy (p ϭ 0.007), and 4 Gy (p ϭ 0.01). *, p Ͻ 0.05, **, p Ͻ 0.01.

TABLE 1
Genes, whose expression altered <0.5-fold or >2-fold at one or more time-points after the addition of doxycycline F-Fold change in expression compared to baseline conditions is presented at 16, 48, and 96 h. For several genes, e.g. CCL3L1 and TAP1, multiple probes are present on the Affymetrix chip, and all data are presented in the table. Conditional formatting is employed such that induced gene expression is colored red, and repressed expression is colored blue. estimated both in the presence and absence of doxycycline. In each case the effect was quantified as a z-score, with negative z-scores representing inhibition of cell survival (Fig. 4A). As expected, the majority of compounds do not show an effect on cell survival, but comparison of z-score data from DG75-AB7 cell screens in the presence of doxycycline (BCL6 deficient) or absence of doxycycline (BCL6 replete) allowed identification of compounds that preferentially reduced survival of BCL6-deficient cells while having relatively little effect on BCL6 replete cells (Table 4). Seven compounds demonstrated z-scores ՅϪ2 at two or more concentrations. Of these compounds, paclitaxel and vinorelbine reduced survival of both BCL6 replete and deficient cells, whereas others, 2-methoxyestradiol, dasatinib, canertinib, lestaurtinib, and sunitinib appeared to preferentially suppress growth of BCL6 deficient cells (Fig. 4B). By comparison doxorubicin and flavopiridol ( Fig. 4C) reduced survival without a differential effect on BCL6deficient cells, whereas cyclophosphamide and olaparib showed no effect on either BCL6 replete or deficient cells.
BCL6 Deficiency Induces a Transcriptional Increase in JAK2 Levels-We focused further work on the JAK2 inhibitor, lestaurtinib, because BCL6 directly represses STAT3 (7), which is a principle target of phosphorylation by JAK2, and we wondered whether BCL6 also repressed JAK2 to cause increased overall inhibition of JAK2 and STAT3. The addition of doxycycline caused a 4-fold increase in JAK2 mRNA and induction of JAK2 protein (Fig. 5, A and B). Induction of phosphotyrosine 705-STAT3 accompanied these changes, and lestaurtinib and two other JAK2 inhibitors, fedratinib and ruxolitinib, prevented phosphorylation ( Fig. 5C). At the doses employed, ruxolitinib and to a lesser extent fedratinib, stabilized JAK2 phosphorylation (in line with their mechanism of action as type I inhibitors (35)), whereas all three agents repressed STAT3 phosphorylation. STAT3 transcription factor activity is largely associated with phosphorylation on tyrosine 705. Westerns blots, utilizing tyrosine 705-and serine 727-specific anti-phospho-STAT3 showed STAT3 tyrosine 705 after induction of BCL6 deficiency, whereas levels of phosphorylated STAT3 serine 727 remain unchanged (Fig. 5D). Overall, the data demonstrated preferential reduction of survival of BCL6-deficient DG75-AB7 by lestaurtinib, and consistent with this effect being mediated by STAT3, JAK2 inhibitors repressed the STAT3 phosphorylation induced by BCL6 deficiency.
In order to confirm that the reduction in viability observed with lestaurtinib was mediated through JAK2 we transfected DG75-AB7 either with siRNA directed against JAK2 or a negative control siRNA (Fig. 5, E and F). Repression of JAK2 caused a significant (t test; p ϭ 0.008) reduction in cell viability in the presence of doxycycline.
BCL6 Deficiency Induces IL10RA Expression-Serum IL-10 levels are prognostic in DLBCL (36), and IL10 receptor expression was increased in ABC-DLBCL as compared with germinal  Table 2. RT-PCR was carried out from cDNA produced from DG75-AB7 cultured in the presence (ϩDox) or absence (ϪDox) of doxycycline (1 g/ml) for 96 h. cDNA dilutions of 1, 1:5, 1:25, and 1:125 (left to right as indicated by the black triangles) were prepared and used as the template for the PCR. Columns to the right show the -fold induction obtained for each gene from the microarray results and the corresponding inductions calculated from the RT-PCR. Conditional formatting indicates induced genes (red) and repressed genes (blue) (Microsoft Excel v14.4.7).   center (GC)-DLBCL (29). IL10 receptor activation promoted STAT3 phosphorylation and DLBCL survival (29). We wondered whether JAK2/STAT3 phosphorylation required IL10 receptor expression and engagement in DG75-AB7. Induction of BCL6 deficiency induced IL10RA mRNA expression (Fig.  5G), and flow cytometry showed elevated surface expression of IL10RA in a population of DG75-AB7 cells (Fig. 5H). It is not clear why IL10RA expression was not induced in all cells. Exogenous IL10 induced STAT3 tyrosine phosphorylation, which is repressed by a blocking anti-IL10 antibody (Fig. 5I), demonstrating that IL10RA engagement is required for STAT3 phosphorylation. Collectively, our data implicates BCL6 in suppressing an IL10RA/JAK2/STAT3 pathway. Lestaurtinib Combined with Induced BCL6 Deficiency Is Effective in a Mouse Xenograft Model-To evaluate in vivo efficacy of lestaurtinib in combination with BCL6 deficiency, we utilized SCID-beige mouse xenografts. Animals were flank-injected with DG75-AB7 and divided into four groups (n ϭ 8).
Although there were no obvious histological differences between tumors in untreated animals and those that had received either doxycycline in the drinking water or lestaurtinib by intraperitoneal injection (Fig. 6), those tumors from animals that received both agents unexpectedly showed massive central necrosis. As compared with untreated animals, mice that received either doxycycline or lestaurtinib alone showed reduced growth, and the combination of agents produced further repression (Fig. 6C). Immunohistochemical assessment of proliferation demonstrated that the combination of doxycycline and lestaurtinib increased the fraction of non-proliferating Ki-67 negative cells (Fig. 6D). Therefore, BCL6 deficiency was sufficient to markedly repress tumor growth without altering the fraction of Ki-67-expressing cells, but the combination of BCL6 deficiency and lestaurtinib significantly (Mann-Whitney U test; p ϭ 0.007) increased numbers of Ki-67-negative cells and was associated with tumor necrosis.
JAK2 Is a Direct Target of BCL6 Transcriptional Repression-Inspection of the JAK2 promoter region identified a possible BCL6 binding site at Ϫ1185 bp from the transcription start site (Fig. 7A). To demonstrate BCL6 binding to this sequence in vitro, we utilized gel shift assays employing lysates from the BCL6 expressing Ramos Burkitt lymphoma cell line and labeled JAK2 promoter BCL6 binding sequence (J2B6BS) oligonucleotide. A protein-DNA complex was effectively competed by a consensus BCL6 binding sequence (FB20) (7), and an anti-BCL6 antibody caused disappearance of the shifted band (Fig.  7B), therefore, identifying the complex as containing BCL6. FB20 oligonucleotide was able to compete more effectively than J2B6BS for binding to BCL6, suggesting a relatively weak interaction at the JAK2 promoter. To show in vivo BCL6 binding at the JAK2 promoter, we analyzed publicly available ChIP-seq databases (37). Statistically significant peaks (Table 5) of BCL6 binding corresponded with BCOR co-repressor binding without evidence of SMRT or NCOR binding (Fig. 7, C and D), supporting both a direct effect of BCL6 and association with BCOR at the JAK2 locus. To demonstrate BCL6 binding at the JAK2 locus in DG75-AB7, a single site ChIP assay was employed and showed ϳ30-fold more binding in the absence of doxycycline (Fig. 7E). Next, reporter assays were carried out in

Small molecule inhibitor screen
Compounds at concentrations from 1, 10, 100, and 1000 nM were administered to DG75-AB7 in the presence or absence of doxycycline, and the surviving fraction was measured. Drug-effect z-scores were calculated. Standardized value (drug effect z-score) is calculated from x (log 2 median raw value without doxycycline Ϫlog 2 median raw value with doxycycline), (log 2 median plate raw values), and (S.D. of differences between log 2 raw values in the presence and absence of doxycycline) according to the formula z-score ϭ (x Ϫ )/. z-scores of Ϫ2 or less at two or more concentrations were obtained for 2-methoxyestraduiol, dasatinib, canertinib, lestaurtinib, paclitaxel, and sunitinib (indicated by the brown bar to right of table). z-scores of Ϫ2 or less at one concentrationonly were obtained for a further group of compounds (blue bar to right of table), and compounds for which no significant z-scores were obtained are indicated by the orange bar to right of table. Individual z-scores were color-coded utilizing a conditional formatting tool (Microsoft Excel v14.4.7) with red indicating low, white indicating mid-range, and blue indicating high scores.  HEK293 cells to demonstrate the functional importance of the binding site (Fig. 7F). Transcription was repressed from the wild-type promoter, in response to a transfected BCL6 expres-sion vector, and there was relief of repression on mutation of the BCL6 binding site. This result was confirmed in DG75-AB7 (Fig. 7G) Therefore, there is a functional BCL6 binding site in the JAK2 proximal promoter region, and BCL6 binds to this region in vivo, supporting the direct transcriptional repression of JAK2 by BCL6.
Expression of JAK2 and BCL6 mRNA Are Inversely Correlated in Human DLBCL-To obtain data on relative JAK2 and BCL6 mRNA expression in primary human lymphoma, we analyzed a publicly available gene expression database (38). JAK2 mRNA is expressed significantly more highly in ABC-DLBCL compared with GC-DLBCL (Fig. 8). This adds to previously published data that both STAT3 and IL10RA are found in significantly greater amounts in ABC-DLBCL (7,29).

Discussion
Burkitt lymphoma cell lines have been useful in determining mechanisms of action and gene targets of BCL6 (18, 39 -41). to the JAK2 BCL6 binding sequence. Gel shift assay were carried out with lysates from DG75-AB7 cells in the absence of doxycycline and labeled JAK2 BCL6 binding sequence oligonucleotide (J2B6BS) (50 fmol). The addition of a super-shifting anti-BCL6 antibody (aBCL6) caused disappearance of the specific band and formation of the specific BCL6-DNA complex was effectively competed by unlabeled FB20 (a consensus BCL6 binding site 7) and also by J2B6BS. Nonspecific (ns) bands that were neither super-shifted nor competed by unlabeled oligonucleotide were observed. C, ChIP-seq demonstrating in vivo BCL6 binding at the JAK2 locus of the OCI-Ly1 diffuse large B-cell lymphoma cell line employing data from 37. The gray bar indicates the peak of BCL6 and BCOR binding at the proximal promoter region. D, zoom-in view of the ChIP-seq data to demonstrate the BCL6 binding site (B6BS). E, single site ChIP assay. ChIP was carried out with anti-BCL6 and anti-actin antibodies followed by real-time PCR for the BCL6 binding site at the JAK2 locus and a site at the CD20 locus, which does not bind BCL6. -Fold enrichment is compared between anti-BCL6 and anti-actin antibodies. DOX, doxycycline. F, luciferase reporter assays. HEK293 cells were transfected with 0, 50, or 200 ng of a mammalian expression vector bearing full-length BCL6 cDNA (as indicated by the black triangles) and luciferase vectors bearing either 2 kb of the proximal JAK2 promoter (WT) or this region with mutations of the BCL6 binding site 1 (Mut) or a vector bearing a luciferase gene without any inserted promoter sequence (Ctrl). Each experiment (n ϭ 3) was carried out in triplicate. For WT, Mut, and Ctrl results are normalized to the condition without transfected BCL6 expression vector. There are significant differences (paired t test) between mutated and wild-type JAK2 promoter sequences for both 50 ng (p ϭ 0.015) and 200 ng (p ϭ 0.03) of transfected BCL6 expression plasmid. *, p Ͻ 0.05. G, luciferase reporter assay. DG75-AB7 cells were transfected with luciferase reporter constructs as before. Cells cultured in the presence of doxycycline showed significantly enhanced (paired t test; p ϭ 0.02) reporter activity. *, p Ͻ 0.05. (n ϭ 3). Irr. Ab., irrelevant antibody.

TABLE 5 Statistical summary of ChIP enrichment data at the JAK2 locus from GEO data, GSE29282 (37)
The statistical test used compares each sample to the reference using a one-tailed binomial test. A false discovery rate of 0.001 was also applied. Partek Genomics Suite software version 6.6 (Copyright 2014; Partek Inc., St. Louis, MO) was employed to do the analysis. Absolute signal (read number) and computed significance of enrichment (p value) are presented. Both BCL6 (42,43) and the genes whose transcription it directly inhibits (3,7,8) are potential targets for therapy in DLBCL. To produce a model system for the systematic evaluation of BCL6and BCL6-regulated pathways for therapy, we produced a conditional BCL6-deficient Burkitt lymphoma cell line that reproduces the functional effect of BCL6 on DNA damage responses (23) and many of the gene expression changes known to be due to BCL6. This novel cell line has significant advantages over previous methodologies (18) because minimum perturbation is required to suppress BCL6. BCL6 deficiency suppressed proliferation of DG75-AB7 and reduced the surviving fraction of cells, but similar to work using a peptide inhibitor to abrogate BCL6 function in a Burkitt lymphoma cell line (44), ϳ50% of cells remain viable. We wondered whether specific pathways, normally repressed by BCL6, became active under these conditions to maintain survival. Others have shown that genes normally repressed by BCL6 can contribute to survival specifically in ABC-DLBCL (7). To reveal pathways required for survival of BCL6-deficient DG75-AB7, which might prove to be important in ABC-DLBCL, we carried out a synthetic lethal screen of DG75-AB7 against a library of small molecule inhibitors, many of which are in clinical use or are candidates for clinical use, and sought a reduction in the surviving fraction of cells, which was greater than that produced by BCL6 deficiency alone.

BCL6 BCOR
Several agents from the library screen were found to enhance the effects of BCL6 deficiency. We chose to focus on lestaurtinib, a JAK2 inhibitor. JAK2 is activated in DLBCL (36) but has not previously been reported to be a direct target of BCL6 transcriptional repression. We produced several lines of evidence to support a direct effect of BCL6 on JAK2; 1) JAK2 mRNA levels increased in response to BCL6 deficiency, 2) BCL6 bound to the JAK2 promoter in vivo and in vitro, and 3) a BCL6 binding site in the JAK2 promoter region was required for BCL6-mediated repression in luciferase reporter assays.
We demonstrated that STAT3, a major target of phosphorylation by JAK2, was induced and phosphorylated in BCL6deficient DG75-AB7. STAT3 mRNA expression is directly repressed by BCL67, and BCL6 also represses IL-6 production in macrophages to inhibit STAT3 signaling (45), but a role for BCL6 in regulating JAK2 has not been previously demonstrated. JAK-STAT signaling has been investigated in DLBCL; this pathway can be activated by IL-10 (36) and by diverse genetic mechanisms such as aberrant MYD88 signaling (6) or inactivating mutation of SOCS1 (46). Overall there may be constitutive STAT3 signaling in ϳ50% of DLBCL (3,7), but activating mutations of JAK2 appear to be very rare, and mechanisms for JAK2 activation in DLBCL other than activating mutations have been suggested (47). Here we show that JAK2, like STAT3, is a direct BCL6 target gene in a Burkitt lymphoma cell line, and this raises the possibility that BCL6 contributes to the overall regulation of JAK2-STAT3 signaling.
It was surprising that JAK2 was not only increased in amounts on induction of BCL6 deficiency in DG75-AB7 but was also phosphorylated. We noted that IL10RA mRNA increased on induction of BCL6 deficiency in DG75-AB7, showing that it is directly or indirectly regulated by BCL6 (Table 1), and we wondered whether engagement of this receptor could mediate JAK2 phosphorylation. Experiments with IL10RA blocking antibodies supported the notion that BCL6 deficiency activated an IL10RA/JAK2/STAT3 pathway.
Our data suggested that an IL10RA/JAK2/STAT3 pathway is repressed by BCL6. IL10RA, JAK2, and STAT3 have been shown separately to be components of pathways that are specifically active in ABC-DLBCL. Both STAT3 (7) and IL10RA (29) have been suggested as targets for therapy in ABC-DLBCL, and ABC-DLBCL cell lines are more sensitive to JAK2 inhibition than GC-DLBCL cell lines (3). In addition, high serum IL10 levels carry a poor prognosis in DLBCL (36). Through our work we were able to link this disparate data to show that there is a role for BCL6 in the regulation of all these genes. Therefore, BCL6 appears to repress a survival pathway in ABC-DLBCL, and systematic future studies of BCL6 target genes might reveal other important survival mechanisms in ABC-DLBCL.
There has been interest in employing JAK2 inhibitors in DLBCL (48,49), but the place of these agents has not been defined. We suggest that they might be useful adjuncts to conventional treatment in cases of ABC-DLBCL expressing relatively low levels of BCL6 and that BCL6 mRNA expression might be a component of a biomarker panel to predict patient populations that will benefit from these drugs.

Experimental Procedures
Construction of DG75-AB7-Both BCL6 alleles of the parental EBV negative Burkitt lymphoma cell line, DG75-354 (gift of Dr. Berthold Henglein), which contains a tetracycline transactivator, were disrupted by homologous recombination, and a tetracycline repressible BCL6 cDNA was inserted to produce DG75-AB7. Exon 3, which contains the translation start site of BCL6, was disrupted (Fig. 1, A and B). The targeting construct contains a zeocin (zeo) resistance cassette that is flanked by loxP sites followed by a downstream SphI site with flanking arms of homology targeting it to the BCL6 gene. The construct contains a unique KpnI restriction site facilitating linearization before gene targeting. After transformation of the parent DG75-354 cells i.e. DG75 containing a tetracycline transactivator construct, with this construct and subsequent selection in zeocin, targeted integration events were detected by PCR screening and confirmed by Southern blotting. Gene disruption was targeted to the 11763-bp genomic SphI fragment, which flanks exon 1b and finishes in exon 6. The arms of homology were subcloned as a smaller fragment derived from this initial big construct. This 10-kb region was amplified from genomic DNA. This 9992-bp KpnI/SacII fragment was ligated into pBluescript KSϩ (Stratagene), resulting in vector p10kb-homology. The 8918-bp ApaI/EagI fragment of p10kb-homology was then subcloned from p10kb-homology into pBluescript KSϩ generating p8.9kb-homology. For disruption of the BCL6 allele, a zeocin cassette was inserted into the SbfI site of p8.9kb-homology (end of exon 3). This 1245-bp zeocin cassette (including its promoter) was amplified from pCMV/Zeo utilizing primers adding loxP sites to both ends of the zeocin cassette and an SphI restriction site to its 3Ј end. This SbfI-digested product was cloned into the SbfI site of p8.9kb-homology to generate the final targeting construct pTarg-BCL6. The orientation of the zeocin cassette was verified by restriction enzyme digestion, and the region surrounding the 5Ј loxP site was sequenced. After targeting of the first BCL6 allele, cells were transiently transfected with Cre-recombinase (pMC-Cre kindly supplied by H. Gu, University of Köln), which excised the region between the loxP sites. The premature stop codon generated by targeted BCL6 gene disruption is likely to result in the mRNA being degraded by nonsense-mediated decay. The potential translation of the message would result in a 55-amino acid truncated protein, which was expected to be nonfunctional and quickly be degraded. Once the zeocin cassette and resistance was lost, the same targeting construct was used to knock-out the second allele.
Gene Expression Analysis-Gene expression changes due to the absence of BCL6 were measured by Affymetrix (Santa Clara, CA) microarrays. Total RNA was extracted using RNeasy minipreps (Qiagen, Hilden, Germany), and 100 ng of total RNA was processed with the GeneChip Eukaryotic Whole Transcript Sense Target Labeling Assay kit (Affymetrix). Briefly double-stranded cDNA was synthesized from total RNA using random hexamers tagged with a T7 promoter sequence and transcribed in vitro by T7 RNA polymerase to produce antisense cRNA. This was subsequently purified and used as template for a further round of cDNA synthesis. After RNA hydrolysis, treatment of sense cDNA with uracil DNA glycosylase and apurinic/apyrimidinic endonuclease 1 (APE1) fragments the DNA strand. cDNA fragments are labeled by terminal deoxynucleotidyltransferase (TdT) with biotin and hybridized to the array (Human GeneChip 1.0 ST Array, Affymetrix), which was subsequently scanned.
The Quality of Microarray Data Were Evaluated by Box Plot and Scatter Plot Analysis-Real time RT-PCR-JAK2 and IL10RA mRNA levels were determined by real time PCR performed using an Applied Biosystems 7500 Real-Time PCR machine (Applied Biosystems, Foster City, CA) and Taqman Universal PCR Master Mix. JAK2 primers were: forward (5Ј-CAGGCAACAGGAACAAGATG) and reverse (5Ј-CCATTC-CCATGCAGAGTCTT); IL10RA primers were forward (5Ј-CAGGAACTGACGGATTGGGAA) and reverse (5Ј-GCTTC-AAACCACACAGACGG).
Small Molecule Inhibitor Screen-DG75-AB7 was seeded at a density of 5 ϫ 10 4 cells/ml in a 384-well plate in a volume of 50 l/well on day 0 and continuously cultured in the presence or absence of doxycycline (1 g/ml) as well as a library of small molecule inhibitors (Table 4) for 4 days at which point cell survival was estimated. The small molecule library encompassed conventional chemotherapy and targeted therapy agents at four different concentrations. In total, six replica experiments were performed with the data from the replicas combined in the final analysis. Staurosporine (10 M) was used as a positive control for the induction of apoptosis. Cell viability was estimated at the end of the experiment using the CellTiter-Glo (Promega, Southampton, UK) luminescence assay. 20 l of CellTiter-Glo reagent was added to each well, the plate contents were mixed for 10 min at room temperature, and the luminescence was measured. Each culture condition was replicated six times. Raw luminescence values were log 2 -transformed and then z-score-standardized according to the log 2 median effect and the variance of effects, which was estimated by calculation of the median absolute deviation.
Cells excluding trypan blue were counted, and cultures were split every 2 days so that concentrations did not exceed or fall below the preferred cell density of between 1 and 9 ϫ 10 5 cells/ ml. The doubling time (T d ) between time point 1 (t 1 ) and time point 2 (t 2 ) (in h) was calculated by inserting cell concentrations at time point t 1 (c 1 ) and time point t 2 (c 2 ) into the formula T d ϭ (t 2 Ϫ t 1 ) ϫ (ln 2 /ln(c 2 /c 1 )).
For cell cycle analysis, 10 6 cells were collected by centrifugation, washed once in PBS, and resuspended in PBS (300 l). Ice-cold ethanol (700 l) was added, and the fixed cells were collected by centrifugation, washed once in PBS, and resuspended in 500 l of PBS containing RNase A (0.2 mg/ml) (Qiagen) and digested at 37°C for 1 h. Propidium iodide was added to a final concentration of 40 g/ml, and cells were analyzed by flow cytometry.
Expression of IL10RA was determined by flow cytometry (FACSCanto, BD Biosciences) utilizing anti-IL10RA antibody (R&D Systems, Minneapolis, MN; MAB2742) in the presence of human Fc block (BD Biosciences; 564219). Detection was with secondary anti-mouse IgG-FITC (Jackson Immuno-Research, West Grove, PA; 315-096-003). A blocking anti-IL10RA antibody (Novus Biologicals, Oxford, UK; NBP1-42534) was employed at 5 g/ml to demonstrate functional effects. IL10 (R&D Systems) was employed at 10 ng/ml. DG75-AB7 cells were cultured for 4 days in the presence or absence of doxycycline after which DNA double-strand breaks were induced by irradiation at a dose rate of 1 Gy/min at 250 kV constant potential and half-value layer of 1.5 mm copper (Pantak Industrial x-ray machine (Pantak Inc., East Haven, CT)). 2 h after irradiation DNA double-strand breaks were identified using fluorescent detection of the phosphorylated form of H2AX (␥H2AX). 5 ϫ 10 5 cells were fixed using ice-cold ethanol for at least 24 h at 4°C and then permeabilized with 0.1% Triton X-100 on ice for 10 min with 0.2 mg/ml RNase (Sigma; R6513); cells were then incubated at 37°C for 1 h, blocked with 4% FCS, and stained with a 1:500 dilution of anti-␥H2AX (Ser-139) (clone JBW301; Millipore 05-636) for 2 h on ice and then with a 1:200 dilution of secondary antibody (AlexaFluor488 goat anti-mouse IgG; Invitrogen A21121) with rotation at room temperature (in the dark) for 1 h. After washing, cells were stained with propidium iodide (0.5 g/ml) and analyzed on a FACSCanto II (BD Biosciences) and characterized for cell cycle stage using FlowJo Software v7.6.4 (Treestar Inc., Ashland, OR). The mean fluorescent intensity of ␥H2AX was determined for cells within G 1 of the cell cycle.
Luciferase Reporter Assays-A 2.0-kb region upstream of the JAK2 transcription start site was cloned from DG75-AB7 genomic DNA. The gene was amplified by PCR using MyFi polymerase (Bioline, London, UK) and the primers 5Ј-taaggtaccccttgtccaccttg-3Ј (forward, with a KpnI site) and 5Ј-cctgctagcttcgaactcagc-3Ј (reverse, with a NheI site). The site was also cloned in the reverse orientation using the primers 5Јtaagctagcccttgtccaccttg-3Ј (forward, with a NheI site) and 5Ј-cctggtaccttcgaactcagc-3Ј (reverse, with a KpnI site). The amplified product was digested with KpnI and NheI and cloned into the empty pGL4 Luciferase Reporter Vector, pGL4.10 (Promega, Fitchburg, WI). Site-directed mutagenesis was performed using the QuikChange II XL kit (Agilent, Santa Clara, CA) to create a mutated binding site, GGCCGAGAA. All mutations were confirmed by sequence verification. HEK-293 cells were transiently transfected with Turbofect transfection reagent (Thermo Scientific, Waltham, MA) following the manufacturer's instructions. Cells were seeded in 96-well whitebottomed plates 24 h before transfection. The cells were then co-transfected with BCL6-FLAG expression construct (50 or 100 ng) and/or empty pGL4.10 reporter vector to 200 ng total DNA. After 8 h the cells were then co-transfected with 100 ng of the various reporter constructs and 100 ng of pGL4.73 Renilla Luciferase construct. All transfections were performed in triplicate and luciferase activities were measured using the Dual-Glo Luciferase Assay System (Promega) 24 h after reporter construct transfection. All luminescence readings were normalized to Renilla luciferase control readings. Reporter assays (without co-transfection of BCL6 expression construct) were similarly carried out in DG75-AB7 in the presence and absence of doxycycline.
Electromobility Shift Assays (EMSA)-EMSAs were carried out utilizing a LightShift Chemiluminescent EMSA kit (Thermo Scientific, Waltham, MA; 20148). EMSAs were performed using the LightShift Chemiluminescent EMSA kit (Thermo Scientific). Binding reactions containing 20 g of nuclear protein lysate were preincubated for 15 min at room temperature in EMSA binding buffer (10 mM Tris, 50 mM KCl, 1 mM DTT, 10 mM MgCl 2 , 10 M ZnCl 2 , 100 g/ml BSA, 4% glycerol, 2 g of Poly (dI⅐dC)) before the addition of a biotinlabeled probe and a subsequent 20-min incubation at room temperature. Complexes were separated on 5% Tris borate-EDTA polyacrylamide gels before electrophoretic transfer to nylon membranes and detection as per the manufacturer's instructions. For specific-competition analysis a 200-fold excess of unlabeled probe was added before the preincubation. For supershift analysis, 2 g of antibody was incubated with the binding reaction for 20 min after the preincubation and before the addition of the labeled probe. The test probe was designed to the putative BCL6 binding site in the JAK2 promoter region (5Ј-ACGATTTCCTAGAATAAGTG-3Ј). The canonical BCL6 probe FB20 was used as a positive control (5Ј-GAAA-ATTCCTAGAAAGCATA-3Ј).
siRNA Knockdown-DG75-AB7 were transfected with siRNA directed against JAK2 (Life Technologies, 4392420) or a negative control siRNA (#4390843) using siPORT NeoFX (Life Technologies) after which they were cultured for 24 h. Doxycycline was then added to half the cells, which were then cultured for a further 24 h and harvested, and CellTiter-Glo luminescence (Promega) was determined.
In Vivo Adoptive Transfer-18 -22-Week-old SCID-Beige mice (Charles River, Burlington, MA) were inoculated subcutaneously in the flank with DG75-AB7 (1 ϫ 10 7 cells in 100 l) premixed with equivolume amounts of Matrigel HC (Sigma, E1270) under general anesthesia. Tumors were left to establish for 7 days, and then the mice were assigned to one of four groups (n ϭ 8). Group 1 received no treatment and regular drinking water and provided a baseline for tumor development and morphology. Group 2 received doxycycline (Sigma D9891) in the drinking water at 1 mg/ml. Group 3 received regular drinking water and once daily subcutaneous injections of lestaurtinib (10 mg/kg), and Group 4 received doxycycline in the drinking water (as before) in combination with once daily subcutaneous injections of lestaurtinib (10 mg/kg).
All groups received water ad libitum, and the mean water intake per mouse was calculated (6.5 ml/mouse/day) corresponding to a mean dose of doxycycline ϳ300 mg/kg/mouse/ day. No differences in fluid intake were noted between the groups. Animals were typically caged in single sex pairs, in conditions appropriate for immune-deficient mice (irradiated diet and bedding). After tumor initiation, body condition and tumor volumes were recorded daily, and water intake was monitored weekly. Tumor growth was monitored by measurement of the length (L) and the width (W) with Vernier calipers. Tumor volume was calculated by the formula volume ϭ W 2 ϫ 0.5 liter.
When tumor volumes in two or more of the control mice exceeded 10 mm in size (in more than 2 dimensions), the experiment was terminated (under veterinary advice). Body mass and tumor mass measurements were made, and tissues were collected from all mice from all groups. All animal procedures were carried out in the Central Research Facility, University of Leicester, under the UK Home Office License Number PPL60/ 4399 and following National Cancer Research Institute guidelines with regard to maximum permissible tumor size.
Spleen samples were formalin-fixed (10% neutral-buffered formalin), dehydrated (70% ethanol), and impregnated with wax, and 4-m sections were cut (following standard protocols). Staining with hematoxylin and the eosin was performed using Shandon Varistain staining machine according to the manufacturer's instructions.
The sections were dewaxed in xylene and rehydrated in alcohol. Antigen retrieval was carried out using a high temperature unmasking technique with diaminobenzidine chromagen solution (DAKO) high pH target retrieval solution in the microwave. Sections were quenched with peroxidase block, and sections were incubated with the monoclonal anti-BCL6 or anti-Ki-67 antibodies (DAKO) (1/10 dilution in TBS) for 45 min and rinsed with TBS/Tween buffer followed by incubation with the HRP-labeled polymer for 45 min and counterstaining with hematoxylin.