BCL6 Antagonizes NOTCH2 to Maintain Survival of Human Follicular Lymphoma Cells

Follicular lymphoma (FL) is the second most common form of non-Hodgkin lymphoma (1). It is generally an indolent and slow-growing disease that is nonetheless mostly incurable with currently available chemo-immunotherapy regimens (1). FLs arise from germinal center (GC) B cells. GCs are transient structures that develop after exposure to T cell–dependent antigen. To form GCs, B cells aggregate within lymphoid follicles and initiate a program of rapid proliferation and somatic hypermutation of their immunoglobulin loci for the purpose of generating high-affinity antibodies. These rapidly proliferating B cells are called centroblasts and are dependent on the presence of the transcriptional repressor BCL6, which is a master regulator of the GC reaction. As clones of B cells emerge within the GC, they encounter T cells and follicular dendritic cells. Signaling events that ensue select B cells with high-affinity antibody for terminal differentiation into memory and B cells. During this signaling process, the B cells slow down and adopt an intermediate state between pre- and post-GC phenotypes, where they are called centrocytes. FL phenotype and gene expression profiles reflect aspects of centroblasts and centrocytes of the GC reaction.

From the molecular standpoint, FLs almost universally harbor t(14;18) translocations involving fusion of BCL2 to regulatory elements associated with immunoglobulin heavy chain locus (2). Constitutive expression of BCL2 suppresses apoptosis, which would otherwise occur physiologically in GC B cells. Mice engineered to express Bcl2 under the control of the Vav2 promoter develop an FL-like disease, albeit with a long latency period (3). BCL2 is a direct transcriptional target of BCL6, which causes its expression to be completely silenced during the GC reaction. Translocation of BCL2 enables its escape from BCL6 repression. This leads to a situation where both proteins BCL2 and BCL6 are expressed together. Along these lines, it has been reported that >90% of FL cases express BCL6 (4, 5). The implication of BCL6 expression in FL has not been explored.

In normal GC B cells, the most established function of BCL6 is to repress critical checkpoint and DNA-damage repair pathway genes, such as ATR, CHEK1, TP53, CDKN1A, etc. Through this mechanism, GC B cells can proliferate and tolerate the DNA damage associated with somatic hypermutation and class switch recombination (6). The survival and growth checkpoint functions of BCL6 are also maintained in diffuse large B-cell lymphomas (DLBCL), which like FL derive from GC B cells. BCL6 expression in DLBCL is maintained in part through chromosomal translocations, although most DLBCLs express BCL6 regardless of genetic lesions. Functional assays demonstrate that DLBCL cells are dependent on BCL6 regardless of translocations (6). Hence, BCL6 more than an oncogene is actually a lineage factor for DLBCL. BCL6 is a member of the BTB–POZ family of transcription factors and mediates transcriptional repression in large part by recruiting the SMRT, NCOR, and BCOR corepressors via the BTB domain (6). Specific peptidomimetic inhibitors of the BCL6 BTB domain kill DLBCL cells in vitro and in vivo (7–9).

Traditionally, BCL6 has not been considered a phenotypic driver in FL, because these tumors, particularly the low-grade ones, only rarely display BCL6 translocations in their early stages and have an indolent phenotype. However, the potent oncogenic functions of BCL6 make it unlikely that its constitutive expression in FL is merely a passenger marker. BCL6 biological functions are dependent on the target genes that it regulates. The biological functions of BCL6 are not likely limited to repressing cell growth and DNA-damage checkpoints. It is entirely possible that other sets of target genes might be crucial for putative roles of BCL6 in FL. Indeed, previous work showed that BCL6 may function through partially different target genes in DLBCL as compared with normal GC B cells (10). Based on these considerations, we hypothesized that BCL6 might also function as an oncoprotein in FL and that any such role would be linked to repression of specific sets of target genes. Discovery of BCL6 target genes in FL seemed like an appropriate starting point to address these questions. Through this approach, we report a novel function for BCL6 in binding and repressing expression and activity of NOTCH2 in FL cells. Repression of NOTCH2 by BCL6 is required to maintain the survival of FL cells. We show that this function is inherited from GC B cells and is required for development of GCs during the humoral immune response. Finally, we find that BCL6-targeted therapy potently kills FL-derived cell lines both in vitro and in vivo and, most importantly, also kills primary human FL patient specimens ex vivo.

As a first approach to exploring BCL6 functions in FL, we performed chromatin immunoprecipitation (ChIP)-on-chip to identify direct target genes relevant to this disease. Because no cell lines are available that reflect FL biology in its indolent phase, we performed these studies using CD20-purified B cells from four independent primary FL patient samples with >80% tumor cell content. ChIP products were cohybridized with their respective inputs to microarrays representing 25,000 promoters. BCL6 binding sites were identified by random permutation analysis and a peak overlap algorithm (10). BCL6 binding sites (48.4%) overlapped between the four FL specimens, amounting to a total of 1,529 probesets and corresponding to 1,712 genes (Supplementary Table S1). DNA motif analysis confirmed that BCL6 canonical DNA binding sequence was highly enriched at these BCL6 binding sites (P < 1.7⁻⁷, FIRE algorithm with hypergeometric test, ref. 11; data not shown). To determine whether BCL6 targets in FL lymphoma cells were associated with particular biological functions, we queried curated gene signatures relevant to lymphomagenesis (12). The top 5 gene sets captured by this method using Fisher exact tests with Benjamini–Hochberg (BH) correction included known direct BCL6 targets from experiments in the Ramos cell line (9), a DLBCL proliferation signature (13), a cell-cycle gene set (14), a NOTCH-induced gene signature (15), and genes repressed by BLIMP1 (ref. 16; Fig. 1A and Supplementary Table S2). We noted that repression of NOTCH was not a previously recognized function of BCL6 in the context of B-cell lymphomas.

To distinguish BCL6 target genes likely to contribute to the FL phenotype, we sought to identify those targets most strongly repressed in FL. Analysis of gene expression profiles from 191 patients with FL (17) demonstrated that 184 FL BCL6 target genes displayed significant inverse correlation with BCL6 expression, including NOTCH2 (Spearman correlation, P < 0.05, Fig. 1B and Supplementary Table S3). To determine whether these 184 genes were enriched for any particular pathway category, we explored their functional annotation using the Database for Annotation, Visualization and Integrated Discovery (DAVID; Supplementary Fig. S1A). This analysis again highlighted NOTCH2 as well as NOTCH pathway genes involved in cell cycle, apoptosis, cellular morphogenesis, lymphoid organ development, or transcription (Supplementary Fig. S1B). These data suggested that BCL6 might be a repressor of the NOTCH2 and NOTCH signaling pathways. In further support of this notion, we observed an inverse correlation between expression of BCL6 and expression of a curated list (15, 18, 19) of NOTCH cofactors and target genes among which NOTCH2 was the most inversely correlated (Spearman correlation, P < 0.05, Fig. 1C and Supplementary Table S4). Examination of BCL6 read densities at the NOTCH2 promoter in the four FL specimens showed enrichment as compared with negative control genes (HPRT and COX6B; Fig. 1D and Supplementary Fig. S1C), similar in magnitude to its enrichment at canonical BCL6 targets like TP53 and BCL6 itself (Supplementary Fig. S1C). Moreover, we identified canonical BCL6 DNA binding sites in the regulatory regions of the NOTCH cofactor genes MAML1, MAML2, and RBPJ, all members of the NOTCH coactivator complex, as well as HES1, a transcriptional repressor and the prototypic NOTCH pathway transcriptional target (Fig. 1E). To validate whether these are true BCL6 targets, we performed QChIP and confirmed that BCL6 is indeed bound to these loci in two independent FL-derived cell lines (Fig. 1F). Primers used to this analysis are found in Supplementary Table S5.

Because GC B cells are the cell of origin of FL, and FL gene expression profiles reflect GC B-cell transcriptional programming we wondered if BCL6 could bind to the NOTCH2 locus in this setting as well. Binding of BCL6 to the promoters of NOTCH2, MAML2, and RBPJ was confirmed by performing QChIP in independent primary human GC B cells (Supplementary Fig. S1D). In addition to FL, GC B cells give rise to DLBCLs. To determine whether BCL6 could bind and regulate NOTCH2 and related genes in DLBCLs, we used BCL6 ChIP-on-chip data performed and analyzed on the same platform as the FLs (10). Among BCL6 target genes in DLBCL cells, the NOTCH-induced gene signature was not significantly enriched (Supplementary Tables S6–S8), although other gene sets overrepresented in FL BCL6 target genes were also enriched in DLBCL (Supplementary Fig. S1E and Supplementary Table S6). Analysis of gene expression profiles from 71 DLBCL patient samples (20) showed that 245 DLBCL BCL6 target genes displayed significant inverse correlation with BCL6 expression (Spearman correlation, P < 0.05; Supplementary Table S7). Nonetheless, analysis of the curated list of NOTCH cofactors and target genes from Fig. 1C indicated that NOTCH2 and other NOTCH pathway partners were inversely correlated (Spearman correlation, P < 0.05; Supplementary Fig. S1G and Supplementary Table S8). Furthermore, less than 15% of the genes found in the DLBCL subset were shared with the ones found on the FL subset (Spearman correlation, P < 0.05; Supplementary Table S9). Altogether, these results point to NOTCH2 and its cofactors as bona fide BCL6 target genes with potential relevance to the phenotype of FL tumors as well as formation of GCs during the humoral immune response. Although repression of the NOTCH pathway is not as strongly linked with DLBCL, it is evident that BCL6 represses NOTCH2 in this disease subtype as well.

Upregulation of BCL6 is required for mature follicular B cells to differentiate into GC B cells during the humoral immune response. In contrast, NOTCH2 plays a critical role in marginal zone differentiation, which is an alternative cell fate for follicular B cells (21). Hence, we wondered whether upregulation of BCL6 in human GC B cells would be associated with silencing of NOTCH2. We purified primary human naïve B cells (NB) and GC B cells from human tonsils and confirmed their purity by IgD⁺ and CD38⁺ staining, respectively (Supplementary Fig. S2A). We then measured the relative transcript abundance of NOTCH2, MAML1, and MAML2 as well as BCL6 by qPCR (Fig. 2A). Whereas BCL6 was upregulated in GC B cells, NOTCH2, MAML1, and MAML2 (but not RBPJ) were concordantly downregulated (Fig. 2A). Examination of available gene expression profiles obtained from five independent sets of NB and GC B cells (22) confirmed downregulation of NOTCH2, MAML1, and NOTCH target genes in GC B cells and the inverse correlation with BCL6 (Fig. 2B and Supplementary Table S10; probesets for MAML2 were not present on this array). NOTCH2 protein downregulation in GC B cells was further confirmed by immunoblotting (Supplementary Fig. S2B). During the GC reaction, B cells first become proliferative centroblasts (CB) and then become centrocytes (CC) as they interact with T cells in the GC light zone. We analyzed gene expression in these cell types using RNA sequencing (RNA-seq), and again observed inverse correlation between BCL6 and NOTCH2 in both CB and CC (Fig. 2C and D).

In order to determine whether these changes in gene expression are linked to GC activation signals, we purified human and murine mature B cells, independently cocultured them with stromal cells (OP9) and exposed them to IL4 and IL21 (23, 24). Murine mature B-cell purity was confirmed by CD45/B220⁺ staining by flow cytometry (Supplementary Fig. S2C). In both human and murine cells, we observed significant BCL6 upregulation (P < 0.0001; P = 0.0004 human and murine cells, respectively) associated with downregulation of NOTCH2 (P = 0.0136 and 0.044) and MAML2 (P = 0.0029 and 0.0255), although expression of MAML1 (P = 0.1069 and 0.0609) and RBPJ (P = 0.0784 and 0.1841) was more variable (Fig. 2E and F). These data suggest that BCL6 repression of NOTCH2 is an integral feature of normal GC B-cell activation and may be critical to specifying the GC phenotype in opposition to marginal zone differentiation.

Given that BCL6 is required for the development of GC formation and directly represses NOTCH2, we wondered whether expression of an active form of NOTCH2 (intracellular domain, ICN2) in GC B cells might disrupt GC formation. To address this question, we studied GC formation in a mouse strain engineered to contain an ICN2-IRES-YFP (ICN2) cassette with a LoxP flanked start site knocked-in to the ROSA26 locus (25). These mice were crossed with a tamoxifen-inducible ROSA26-Cre-ER^T2 strain or in ROSA26–wild-type (WT) control mice. GC formation was induced by immunization with the T cell–dependent antigen NP₆₅-CGG. ICN2 expression was induced the following day by tamoxifen injection. Animals were sacrificed 14 days later, and spleens were resected for analysis (Supplementary Fig. S3A). Paraffin-embedded spleen sections from WT and ICN2 mice were stained for the GC-specific marker peanut agglutinin (PNA) or for BCL6 and B220. GCs were defined as clusters of PNA⁺ or BCL6⁺/B220⁺ cells (Fig. 3A and B). As compared with WT, ICN2 conditional mice exhibited significant reduction in PNA⁺ or BCL6⁺/B220⁺ GC per spleen section (mean of 29 vs. 4, P = 0.0005, unpaired two-tailed t test; and mean 22 vs. 4 P = 0.0003, respectively; Fig. 3C and D). The average size of the GCs was also reduced by 2- to 3-fold versus WT controls (mean of 262 ± 36 vs. 91 ± 10 μm² for PNA⁺, P = 0.0015, and 177 ± 30 vs. 84 ± 13 μm² for BCL6⁺/B220⁺, P = 0.0336 respectively; Fig. 3E–G). A more quantitative analysis of GC B cells was generated by flow cytometry. In the presence of ICN2, the abundance of B220⁺GL7⁺CD95⁺ GC B cells was reduced 3-fold as compared with WT animals (1.7% vs. 0.6% mean GC B cells vs. total splenocytes, P = 0.0003 unpaired two-tailed t test; Fig. 3H and I).

Formation of GCs requires cooperation between different cell types. Hence, it is conceivable that ER-induced ICN2 impairment in GC formation could be attributed to non–B-cell effects. Therefore, we established a second mouse model that specifically limited conditional expression of ICN2 to GC B cells. In this case, the C1γCre mouse strain, which activates Cre expression in GC B cells (26), was crossed to ROSA26-WT or ROSA26-ICN2-IRES-YFP (hereafter referred to as WT and ICN2, respectively). These animals were immunized with the T cell–dependent antigen sheep red blood cells (SRBC). Paraffin-embedded spleen sections from WT and ICN2 mice were stained for PNA, or BCL6/B220, to identify GCs (Supplementary Fig. S3B and S3C). These experiments again yielded significant reductions in the numbers of GCs in ICN2 conditional mice, with a mean of 13 versus 4 PNA⁺ GC per spleen section (P = 0.0001, unpaired two-tailed t test) and 14 versus 5 BCL6⁺/B220⁺ GC per spleen (P = 0.0001; Supplementary Fig. S3D and S3E). The average size of the GCs was also significantly diminished in ICN2-expressing mice (mean of 301 ± 35 vs. 68 ± 17 μm² for PNA⁺, P < 0.0001, and 226 ± 24 vs. 66 ± 16 μm² for BCL6⁺/B220⁺, P = 0.0002 respectively; Supplementary Fig. S3F–S3H). Quantitative assessment of GC B cells by flow cytometry yielded a 3-fold reduction in the abundance of B220⁺GL7⁺CD95⁺ GC B cells as compared with WT animals (3.28% vs. 1.08% mean GC B cells vs. total splenocytes, P < 0.0001, unpaired two-tailed t test; Supplementary Fig. S4A and S4B).

Consistent with the reduction in GCs (which contain cells that proliferate and undergo apoptosis), there was also reduction in the abundance of proliferating cell clusters as shown by PCNA and Ki67 immunohistochemistry (Supplementary Fig. S4C), as well as clusters of cells with apoptotic markers caspase-3 and TUNEL in the ICN2 mice (Supplementary Fig. S4D). Given the crucial role of ICN2 in driving marginal zone B-cell differentiation, we next stained for MZB markers CD21 and CD23. We observed an increase in the MZB cell population in ICN2 conditional mice in detriment to the follicular B (FoB) cell population (mean of 65% ± 1.5% FoB vs. 15% ± 2.6% MZB) compared with WT (mean of 70% ± 3% FoB vs. 10% ± 1.3% MZB; Supplementary Fig. S4E). In contrast, as expected, there was no effect on T lymphoid (single- or double-positive populations stained for CD8/CD4; top) or myeloid lineages (CD11b/Gr-1; bottom, Supplementary Fig. S4F). Altogether, these data indicate that ICN2 expression is incompatible with B cells forming GCs and that repression of Notch2 is a critical function of BCL6 in enabling the GC phenotype.

To confirm that BCL6 directly represses NOTCH2 and related genes, we depleted BCL6 from FL-derived cell lines using an siRNA that we validated to be specific for BCL6 (27) compared with scrambled control, followed by qPCR assessment of NOTCH2, MAML1, MAML2, and RBPJ transcripts (Fig. 4A). We observed approximately 2-fold derepression of NOTCH2 and variable derepression of the other genes. This magnitude of derepression is similar to that reported for other BCL6 targets (7, 9, 10, 28, 29). siBCL6 was confirmed to deplete BCL6 protein (Supplementary Fig. S5A). To further confirm this result using an independent approach, we treated the FL cell lines with the specific BCL6 inhibitor RI-BPI, which binds to the BCL6 BTB repression domain to block the transcriptional effects of BCL6 (7). We first validated that we could reproduce the known effect of Retro-inverted BCL6-peptide inhibitor (RI-BPI) in blocking the repressor activity of the BCL6 BTB domain in the context of lymphoma cell lines using a BCL6 BTB domain reporter assay (Supplementary Fig. S5B). We then measured the effect of RI-BPI on derepressing NOTCH2, MAML1, MAML2, and RBPJ as compared with control peptide and observed a similar effect as that seen with siRNA (Fig. 4B). There was no induction of NOTCH1, which was expressed at very low levels in these cells (data not shown). Given that NOTCH2 was also inversely correlated with BCL6 in GCB-DLBCL cells, we examined the effect of RI-BPI on two such cell lines and observed a generally similar degree of derepression of NOTCH2, MAML1, and MAML2 (Supplementary Fig. S5C). To determine whether upregulation of NOTCH2 was functionally significant, we performed NOTCH reporter assays in the FL-derived cell lines after BCL6 siRNA. BCL6 knockdown resulted in significant induction of NOTCH reporter activity (P < 0.0001 in both cell lines, unpaired two-tailed t test), but did not affect a control reporter (Fig. 4C).

In addition to induction of expression of the NOTCH2 transcriptional complex, NOTCH activation involves signaling through NOTCH ligands (DLL1, DLL3, DLL4, JAG1, and JAG2) and cleavage by metalloproteases (ADAM10 and ADAM17). To determine possible sources of NOTCH signaling in FLs and GC B cells, we measured expression of these genes in the principal FL and DLBCL cell lines used for this study (DoHH2, Sc-1, OCI-Ly1, and SUD-HL-4), purified primary mature B-cell populations: NB, GC, CB, CC, memory B (MB), tonsillar plasma cell (TPC), bone marrow plasma cell (BMPC), and in a cohort of patients with primary FL by RNA-seq (30, 31). Expression of NOTCH ligands DLL1, DLL3, DLL4, and JAG1 was low in all of four cell lines, although JAG2 was expressed, and possibly relevant to NOTCH activation in vitro (Fig. 4D and E). In contrast, NOTCH ligand expression was essentially absent in the relevant mature B-cell subsets, although DLL4 and JAG1 were upregulated later in BMPCs (Fig. 4F and G). Only a small subset of FLs manifested higher levels of NOTCH ligands. In contrast, ADAM10 and ADAM17 are expressed in the cell lines, primary mature B cells and FLs, especially ADAM17 (Fig. 4D and H). Hence, in the in vivo setting, NOTCH ligand delivery to lymphoma cells likely comes mostly from the lymph node microenvironment, where NOTCH ligands are known to be expressed (32), whereupon FLs or mature B cells are then competent to cleave NOTCH2. Indeed, we observed that coculture of FL cells with a stromal cell line engineered to express DLL1 but not the same cell line without DLL1 reproducibly induced apoptosis, consistent with NOTCH2 signaling being deleterious to these cells (Fig. 4I).

To determine whether BCL6 repression of NOTCH2 was important to its actions in FL cells, we first wished to establish whether FL cells are biologically dependent on BCL6. We therefore exposed DoHH2, Sc-1, and WSU-DLCL2 FL-derived cell lines to increasing concentrations of RI-BPI and measured cell viability using a fluorometric resazurin reduction method. DoHH2 and Sc-1 cells displayed a GI₅₀ of 11.7 μmol/L and 15.2 μmol/L, respectively, which is comparable to the GI₅₀ of BCL6-dependent DLBCL cells (7), whereas WSU-DLCL2 cells were more resistant (Fig. 5A). Similar to the case of DLBCL (7, 9), not all FL cells were responsive to RI-BPI, but the ones that were sensitive underwent apoptosis, as shown in caspase-3/7 cleavage assays and annexin V/7AAD flow cytometry (Fig. 5B and Supplementary Fig. S6A). To determine whether NOTCH2 repression contributes to the effect of BCL6 in maintaining survival of FL cells, we attempted to rescue our panel of BCL6-dependent FL and GCB-DLBCL lymphoma cell lines (DoHH2, Sc-1, SU-DHL-4, and OCI-Ly1) from the effects of BCL6 depletion by preventing NOTCH2 upregulation using an siRNA approach. We verified knockdown of both transcripts (BCL6 and NOTCH2) in each cell line, using two independent siRNA for both BCL6 and NOTCH2 (Supplementary Fig. S6B). BCL6 knockdown resulted in a ∼30% to 60% loss of viability in all four cell lines at 48 hours, whereas NOTCH2 siRNA alone did not affect cell viability (Fig. 5C and Supplementary Fig. S6C). Notably, concordant knockdown of NOTCH2 prevented its upregulation in response to BCL6 siRNA and significantly rescued all of four cell lines from BCL6 siRNA–induced loss of viability. We used two independent siRNA sequences for BCL6 and NOTCH2. The rescue of siBCL6-1 sequence by both siNOTCH2 RNA sequences is shown in Fig. 5C and the rescue of siBCL6-2 in Supplementary Fig. S6C. In contrast, the viability of the BCL6-independent t(14;18) lymphoma cell line OCI-Ly8 (8) was not affected by BCL6 or NOTCH2 siRNA, even though they manifested a similar degree of knockdown (Supplementary Fig. S6D and S6E). NOTCH2 siRNA also, at least partially, rescued FL cells from loss of viability induced by the BCL6 inhibitor RI-BPI (Fig. 5D and data not shown). In a reciprocal experiment, to determine whether induction of NOTCH2 was sufficient to suppress the growth of FL cells downstream of BCL6 inhibition, we transduced DoHH2 and Sc-1 cells with a retrovirus expressing ICN2 and GFP, or GFP alone. We measured the relative depletion of GFP-positive cells from the total population of cells by flow cytometry over the course of 10 days. NOTCH2 expression was clearly growth suppressive and sufficient to inhibit FL cells, because 95% of ICN2-GFP DoHH2 cells were depleted by 10 days, as were 75% of ICN2-GFP Sc-1 cells (Supplementary Fig. S6F). Finally, to confirm the importance of NOTCH2 pathway suppression by BCL6 in FL cells, we exposed lymphoma cells to the NOTCH2 antagonist antibody NRR2, after confirming its specificity of action in vivo against NOTCH2 but not NOTCH1, consistent with previous reports (Supplementary Fig. S7; refs. 33, 34). NRR2, but not control antibody, could also rescue DoHH2 and SU-DHL-4 cells from cell death induced by RI-BPI to variable degrees (Fig. 5E). Repression of NOTCH2 is thus a critical downstream mechanism of action of BCL6, required for its ability to maintain the survival of FL cells.

FL cell lines may not necessarily accurately represent the biology of primary indolent FL in human patients at the time of diagnosis. We obtained a set of 17 diagnostic primary human FL specimens from patients with nontransformed disease, made single-cell suspensions, exposed them to RI-BPI or vehicle ex vivo for 48 hours, and then assessed for viability. Immunohistochemistry analysis indicated that 10 patients were clearly BCL6 positive and 7 were borderline positive to negative for BCL6 (data not shown). All 17 samples were exposed to 10 μmol/L RI-BPI or vehicle. Although the BCL6-negative/low FLs were resistant to BCL6 inhibitors, 9 of 10 of the BCL6-positive FLs responded with a 20% to 70% loss of viability (Fig. 6A). Consistent with the actions of BCL6 in normal GC B cells and FL cell lines, we observed that in primary human FLs, RI-BPI induced derepression of NOTCH2, as well as induction of the NOTCH2 targets HES1 and HES6 (Fig. 6B). Moreover, we also observed reexpression of ATF5, APOL6, CCR6, and HOXA13, all of which are direct BCL6 targets inversely correlated with BCL6 expression in patients with FL (Fig. 1B and Supplementary Table S3), and of STAT3, a positive control BCL6 target, but not of HPRT, which is a negative control (Supplementary Fig. S8A).

To determine whether RI-BPI could also suppress FL tumors in situ in animals, we xenotransplanted the DoHH2 and Sc-1 cell lines into SCID mice. Once palpable tumors formed, pairs of DoHH2 or Sc-1 tumor–bearing mice were randomized to receive either RI-BPI 25 mg/kg/day intraperitoneally or vehicle control (5 mice per treatment condition). Both animals of each pair were sacrificed when one of them reached maximal permitted tumor burden. In all cases, the RI-BPI–treated tumors were considerably smaller than controls for both the more sensitive DoHH2 (Fisher exact test P = 0.03) and the less sensitive Sc-1 cell line (Fisher exact test P = 0.04, Fig. 6C). Immunohistochemical analysis of these tumors showed an increase in apoptosis in the RI-BPI–treated tumors by TUNEL assays from 7% to 12% in DoHH2 (P < 0.001, Fisher exact test), and from 10% to 18% (P < 0.001, Fisher exact test) in Sc-1 tumors, respectively. Analysis of these same tumors by caspase-3 yielded similar results, from 6% to 22% in DoHH2 (P < 0.0001, Fisher exact test), and from 11% to 21% (P < 0.0001, Fisher exact test) in Sc-1 (Fig. 6D and Supplementary Fig. S8B–S8D). Examination of mRNA extracted from tumor xenografts revealed upregulation of NOTCH2, MAML1, MAML2, and HES1 (P = 0.0086, 0.0011, 0.0929, and 0.0079, respectively, Mann–Whitney test) in RI-BPI–treated mice versus vehicle (Fig. 6E). BCL6 is thus a bona fide therapeutic target in FL at least in part through its repression of NOTCH2, which we show is a growth suppressor in FL.

The role of BCL6 in FL has not been previously explored, in part because of considerations such as (i) the indolent phenotype of FL distinct from the more aggressive DLBCLs typically associated with BCL6, (ii) the frequent t(14:18) translocation focused attention on BCL2, and (iii) BCL6 is not often translocated in FL (5). However, our previous work indicated that DLBCL cells are dependent on BCL6 regardless of whether its locus is affected by mutations (7, 8). Therefore, analysis of BCL6 in this disease seemed warranted (35, 36). Analysis of the BCL6 cistrome in primary FL specimens suggested new mechanisms of action for BCL6 not previously gleaned from studies in DLBCL.

In particular, we focused on BCL6 repression of NOTCH2, a growth suppressor of pre–B leukemia cells and Hodgkin lymphoma cells (37). BCL6 was bound to NOTCH target genes in FL, and NOTCH2 levels were inversely correlated with BCL6 in FL. We find that BCL6 is a direct repressor of NOTCH2, MAML1, and MAML2 expression, and that BCL6 inhibition induces NOTCH2 transcriptional activity in reporter assays and endogenous NOTCH2 target genes in FL cells. BCL6 was also shown to antagonize NOTCH signaling in the context of Xenopus embryonic development (38). In that setting, BCL6 binding and repression of certain NOTCH target genes as well as its direct interference in the NOTCH1–MAML1 interaction was essential in the determination of axis symmetry during development (38). In mammals, BCL6 does not play a role in axis symmetry because Bcl6-null mice do not show this developmental defect, and antagonism with NOTCH signaling is more linked to NOTCH2, at least in B cells. More recently, it was also reported that BCL6 may control neurogenesis through SIRT1-dependent repression of selective NOTCH targets (39). BCL6 prevented transcriptional activation of the Hes5 promoter, by excluding Maml1 and instead recruiting Sirt1 to NOTCH transcriptional complex to downregulate its expression but without impairing NOTCH signaling during the transition from neuronal progenitor stem cells to differentiated neurons (39). This additional protein interference mechanism may be relevant to B cells as well and could be the subject of further investigation into BCL6 and NOTCH cross-talk.

During B-cell development, induction of NOTCH2 activity by its ligand DLL1 in the murine splenic vasculature directs B cells toward marginal zone differentiation and away from the pool of naïve follicular B cells from which GCs form (40). We confirm that NOTCH2, MAML1, and several key NOTCH targets are downregulated in GC B cells in comparison to their precursor naïve B cells. Indeed, we show that NOTCH2 must be silenced for the development of fully established GCs, and that it is the transcription factor BCL6, which is a master regulator of GC formation, that mediates this downregulation. Induction of BCL6 through IL21 and IL4 in human and murine naïve B cells (23, 24) thus resulted in downregulation of NOTCH2, MAML2, and RBPJ, suggesting additional mechanisms through which NOTCH2 activity is controlled in divergent cell fates in secondary lymphoid tissues. BCL6 repression of NOTCH2 in FL is therefore derived from a normal function of BCL6 in GC B cells. It is notable that studies in human GC B cells cultured in the presence of the follicular dendritic HK cell line showed that Jg1-mediated NOTCH signaling contributes to the survival mediated by follicular dendritic cells (32). In murine B cells, NOTCH signaling through DLL1 was shown to enhance formation of IgG1 class switched plasma cells (41). However, in experiments with B cells in their physiologic context, we show that induction of ICN2 profoundly suppresses GC formation. Along with NOTCH2, BCL6 represses other growth and survival genes in GC B cells, including BCL2 and MYC (10). Collectively, the data suggest a scenario whereby BCL6 may attenuate NOTCH2 signaling in GC centroblasts in GC dark zone, but later, as BCL6 levels are downregulated and B cells undergo class switch recombination and interact with follicular dendritic cells of the GC as they transition to memory or plasma cells, the NOTCH pathway may cooperate with other survival signals to maintain the survival and proliferation of post-GC B cells. Along these lines, Lee and colleagues (42) reported five cases of activating NOTCH2 mutations in non-GCB DLBCLs, which originate from post-GC B cells. These findings underline the importance of cell context in determining whether certain genes function as oncogenes or tumor suppressors.

It should be noted that a recent study reported five NOTCH1 and two NOTCH2 gain-of-function mutations in FL, accounting for a total of 6.3% among 112 patients with FL. However, NOTCH-mutated FL cases more frequently included a DLBCL histologic component than the WT cases (43). Hence, these mutations could occur in very specific subsets of patients who may be borderline DLBCL. Along these lines, Dr. Paulli's work points to a bias toward patients with DLBCL positive for hepatitis C virus who carry NOTCH2 mutations (20%) versus patients negative for the virus (44). They could also be linked to FLs with features of late GC lymphomas. Cell-type context is clearly important because even though NOTCH activating mutations are known to drive T-cell leukemias, NOTCH is by contrast a tumor suppressor in myeloid leukemia (45). Suppression of NOTCH is evidently a critical function of BCL6 in FLs, because NOTCH2 siRNA or antagonist could rescue cell death induced by BCL6 blockade, and NOTCH2 expression killed FL cells. NOTCH2 is thus a growth suppressor of FLs, and BCL6 mediates FL pathogenesis in part through suppression of this pathway.

Finally, we show that BCL6 is a bona fide therapeutic target in FL, and not just a passenger marker. This was confirmed using two different loss-of-function strategies (siRNA and RI-BPI) and was relevant not only to cell lines but also to primary human BCL6-positive FLs. BCL6 inhibitors also induced NOTCH2 and suppressed the growth of FL xenografts, suggesting that cell-autonomous inhibition of an FL survival pathway is able to suppress FL tumor growth in vivo. This is an example of “non-oncogene” dependence, in that even though the BCL6 locus is not usually mutated in FL, cells that express it are biologically dependent on its continued presence to maintain their survival. Because RI-BPI blocks only the BTB domain lateral groove and does not affect other BCL6 functions, FL survival is clearly dependent on BCL6 recruitment of corepressors to the BCL6 BTB domain (7). The presented data suggest that BCL6 targeted therapy could be a useful approach for treatment of FLs. Importantly, because RI-BPI affects only certain BCL6 functions (46), it does not induce toxicity or inflammation in animals even when administered long term and so is suitable as a therapeutic agent for diseases that might require chronic dosing (7). Likewise, animals engineered to express a BCL6 mutant that mimics the loss of function induced by RI-BPI live normal healthy lives (47). The data expand the spectrum of patients who are candidates for RI-BPI or other BCL6 inhibitor clinical trials and offer a potential approach for more effectively eradicating these incurable tumors.

Publicly available gene expression microarray data were obtained from 191 primary FLs (9). Data processing and normalization were performed as previously described (9). For the examination of available gene expression profiles obtained from five independent sets of NB and GC B cells, data publicly available were obtained from GC B-cell array accession number GSE2350 (22). BCL6 ChIP-on-chip data have been submitted to GEO GSE29165.

DoHH2, Sc-1, WSU-DLCL2, and SU-DHL-4 cell lines were obtained from the DSMZ German collection of microorganisms and cell cultures. They were grown in RPMI-1640 medium (CellGro) supplemented with 10% FBS, 1% penicillin/streptomycin, 2 mmol/L l-glutamine and 10 mmol/L HEPES (all from Gibco). OCI-Ly1 and OCI-Ly8 cell lines obtained from OCI were grown in Iscove's Medium (CellGro) supplemented with 10% FBS, 1% penicillin/streptomycin as above. Stable cultures for DoHH2 and Sc-1 cell lines were established by retroviral infection of pMIGR1-GFP control and pMIGR1-ICN2-GFP (the intracellular domain of NOTCH2 protein that is fused to GFP). The OP9 cell line was grown in DMEM (CellGro) supplemented with 20% FBS, 1% penicillin/streptomycin, 2 mmol/L l-glutamine, 10 mmol/L HEPES, 55 μmol/L β-mercaptoethanol, and 50 μg/mL gentamicin (Gibco).

We performed DNA genotyping to identify and authenticate all the cell lines before use, December 2016 being last time they were authenticated. DNA extraction, short repeat profiling, and comparison with known cell line profiles from ATCC were performed by BioSynthesis Inc.

The RI-BPI corresponds to sequence S6.2 as previously published (7). Control and RI-BPI peptides were synthesized by Biosynthesis Inc.

Tonsillar tissue was obtained as discarded material from routine tonsillectomies at the Montefiore Children's Hospital and Weill Cornell Medical College (with the approval of the Institutional Review Boards of Albert Einstein College of Medicine, Montefiore Hospital, and Weill Cornell Medical College and in accordance with the Helsinki protocols). Briefly, tonsillar mononuclear cells were separated by density centrifugation with Fico/Lite LymphoH (Atlanta Biologicals). Samples were divided in two and the mononuclear cells rophatPro Separator system as follows: Naïve B cells were stained sequentially with anti–IgD-FITC followed by FITC microbeads followed by a positive selection. GC B-cell isolation was done sequentially with anti-CD77, followed by anti-MARM, and rat anti-mouse IgG1 microbeads followed by a positive selection. The purity of the isolated B-cell populations was determined by flow-cytometry LSRII system. Naïve B cells were IgD⁺CD38^lo and GC B cells were IgD⁻CD77⁺CD38^hi (see Supplementary Fig. S3 for purity of samples). The FlowJo software from Treestar, Inc. was used for the flow-cytometry analysis. Following purification, the samples were processed for mRNA and protein extraction. Naïve B cells were cocultured with a stromal layer of OP9. Cocultures were grown in RPMI with 20% FBS, 1% penicillin/streptomycin, 2 mmol/L l-glutamine and 10 mmol/L HEPES. Cytokine treatment was done using 30 ng/mL of IL4 and 30 ng/mL of IL21 or vehicle for up to 4 days. Cell viability was assessed every day by Trypan blue dye exclusion, and cytokines were added to media every other day. Alternatively, resting B lymphocytes from BL57/6 mouse spleens were isolated. B220⁺ splenocytes were obtained by negative selection with anti-CD43 and anti–Mac-1/CD11b monoclonal antibodies coupled to magnetic microbeads. Additional information is given in the Methods section of Supplementary Figures.

Experiments were performed in accordance with the guidelines of the New York University Institutional Animal Care and Use Committee. ROSA26-ICN2-IRES-YFP knock-in mice were generated by insertion of a LoxP flanked splice acceptor NEO-ATG cassette with two polyA sites followed by the ICN2-IRES-YFP cassette into the ROSA26 locus, allowing the ROSA26 promoter to drive the expression of the NEO-ATG cassette. In order to express the transgene (ICN2-IRES-YFP), these mice were crossed with tamoxifen-inducible ROSA26-CreER^T2 mice expressing CreER^T2 from the ubiquitously expressed ROSA26 promoter or with the CγCre. For GC studies, serum of preimmunized mice (ROSA 26 WT and ROSA26-ICN2-IRES-YFP knock-in) was collected 1 day prior to the start of the experiment (day −1). The next day, mice were injected intraperitoneally with 100 μL of ⁴Hydroxy-³nitrophenylacetyl hapten-chicken gamma globulin (NP₆₅-GCC) in alum (day 0). For CreER^T2 induction, tamoxifen was solubilized in corn oil at a concentration of 20 mg/mL, and 1 day after NP immunization, a single intraperitoneal injection of 0.2 mg/g body weight was administered. Alternatively, mice were immunized intraperitoneally with 500 μL 2% SRBC in PBS. Serum was collected at days 7 and 14. Mice were sacrificed, and spleens were harvested for IHC and flow-cytometry analysis. A schematic representation of the experiment is given in Supplementary Fig. S4A. Genotyping primers are listed in Supplementary Table S5. Additional information is available in the Methods section of Supplementary Figures.

B220⁺ splenocytes from WT and ICN2 mice were obtained as described above. For the GC B-cell population, we gated on B220⁺ and GL7 eFluor674 and CD95(Fas) PECy7. For plasma cell population (CD38⁺ CD138⁺), we used CD38 APC and CD138 PE. ICN2 expression was assessed by B220⁺ YFP⁺. Immunohistochemistry of spleens was performed on formalin-fixed paraffin-embedded sections with the following primary antibodies: PNA, BCL6 (N3), and CD45R/B220. BCL6 shows a purple pattern, whereas B220 staining is pale brown. To count the GC, we used CellSens Software (Olympus America Inc.). Additional information is available in the Methods section of Supplementary Figures.

Deidentified primary FL specimen tissues were obtained in accordance with the guidelines and approval of the Institutional Review Board of Weill Cornell Medical College and in accordance with the Helsinki protocols. We selected the specimens based on estimated tumor content >80% by our collaborating pathologist Dr. Wayne Tam. Single-cell suspensions from lymph node biopsies were obtained by physical disruption of tissues (using scalpels and cell strainers), followed by cell density gradient separation (Fico/Lite LymphoH; Atlanta Biologicals). Cell number and viability were determined by Trypan blue dye exclusion, and cells were cultivated in medium containing 80% RPMI and 20% human serum supplemented with antibiotics, l-glutamine 4 mmol/L and HEPES 10 mmol/L. Cells were exposed in duplicate to control and RI-BPI at indicated concentrations for 48 hours. Viability was determined as detailed above. The BCL6 protein status was determined in paraffin-embedded samples by immunohistochemistry using anti-BCL6 (Dako North America).

All animal procedures followed NIH protocols and were approved by the Animal Institute Committee of the Weill Cornell Medical College. Six- to 8-week-old male SCID mice were purchased from the National Cancer Institute (NCI) and housed in a clean environment. Mice were subcutaneously injected in the left flank with low-passage 10⁷ human lymphoma cells (DoHH2 and Sc-1). Tumor volume was monitored every other day using electronic digital calipers in two dimensions. Tumor volume was calculated using the formula: Tumor volume (mm³) = (smallest diameter² × largest diameter)/2. When tumors reached a palpable size (approximately 75–100 mm³ after 21 days post-injection), the mice were randomized to two different treatment arms. RI-BPI was stored lyophilized at −20°C until reconstituted with sterile pure water immediately before use. RI-BPI was administered by intraperitoneal injection. Mice were weighed twice a week. All mice were euthanized by cervical dislocation under anesthesia when at least 2 of 10 tumors reached 20 mm in any dimension (equivalent to 1 g), which was generally on day 9 or 10 of the treatment schedule. At the moment of euthanasia, the tumors and other tissues were harvested and weighed.

For siRNA knockdown experiments, BCL6-specific siRNAs (Cat# HSS100966) and control nontargeting siRNA (Cat# 1299003) were purchased from Invitrogen. siRNA sequences for NOTCH2 (Cat# J-012235) were purchased from Dharmacon-ThermoScientific. siRNA (20 pmol/L) was suspended in 20 μL of Solution SF and introduced into 3 × 10⁶ cells using the Amaxa 96-well nucleofector (Lonza). For Western blot experiments, rabbit antibody raised against BCL6-N3 (sc-858) and anti–Actin-HRP conjugated (sc-1615) were purchased from Santa Cruz Biotechnology; rabbit antibody raised against ICN2 (ab72803) that recognizes the cleaved intracellular fragment of NOTCH2 was purchased from Abcam. For experiments using cell lines treated with RI-BPI, 10 to 20 μmol/L final concentration of the drug was used for 24 and 48 hours. Doses were selected based on the relative GI₅₀s for each cell line. For DoHH2 and Sc-1 anti-NRR2 experiments, 2 × 10⁶ cells were treated with 2 μg/mL of anti-NRR2 from Genentech, Inc. or a negative control, anti-IgG1 from Southern Biotech. Twenty-four hours after treatment, cells were replated to 96-well plates and RI-BPI was added. Cell viability was measured 24 or 48 hours after RI-BPI treatment as detailed below. Experiments were performed in triplicate, and the figures represent the average of three experiments ± SEM.

Cell viability on lymphoma cell lines was determined using a fluorometric resazurin reduction method (CellTiter-Blue, Promega) and relative fluorescence (560_exitation/590_emission) detected with a Synergy4 Microplate Reader (BioTek Instruments). The number of viable cells was calculated by extrapolating from the standard curve. Fluorescence was measured for three replicates per treatment condition and cell viability in drug-treated cells normalized to their respective controls. Experiments were performed in triplicate. The figures represent the average of three experiments and the SEM. To determine growth inhibition of lymphoma cell lines exposed to different doses of RI-BPI, cells were plated at concentrations sufficient to keep untreated cells in exponential growth over the 48-hour drug exposure time. Cell viability was measured as described above. Trypan blue dye exclusion was used as a secondary method to confirm the results. Fluorescence was determined for 6 replicates per treatment condition, and cell viability in drug-treated cells was normalized to their respective controls. Unless stated otherwise, the experiments were carried out in biological triplicate. The CompuSyn software package (Biosoft) was used to plot dose–effect curves and determine the drug concentration that inhibits the growth of cell lines by 50% compared with control (GI₅₀). The linear correlation coefficient was higher than 0.90 for each curve in the median-effect plot.

Additional experimental procedures including quantitative RT-PCR, plasmids and reporter assays, ChIP and ChIP-on-chip assay, bioinformatics analysis of Gene Expression data, primary cultures, flow cytometry, and mouse studies can be found on Supplemental Experimental Procedures.

A. Melnick reports receiving commercial research grants from Janssen, GSK, Lilly, and Roche, and is a consultant/advisory board member for Epizyme and Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E. Valls, H. Geng, L. Cerchietti, I. Aifantis, A. Melnick

Development of methodology: E. Valls, H. Geng, M. Rivas, S.N. Yang, X. Agirre, W. Ci, A. Melnick

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Valls, C. Lobry, L. Cerchietti, P. Oh, S.N. Yang, E. Oswald, C.W. Graham, Y. Jiang, K. Hatzi, E. Perkey, Z. Li, W. Tam, K. Bhatt, P.A. Zweidler-McKay, I. Maillard, W. Ci, A. Melnick

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Valls, C. Lobry, H. Geng, M. Cardenas, L. Cerchietti, E. Oswald, Y. Jiang, K. Hatzi, X. Agirre, Z. Li, I. Maillard, O. Elemento, W. Ci, I. Aifantis, A. Melnick

Writing, review, and/or revision of the manuscript: E. Valls, C. Lobry, P. Oh, W. Tam, J.P. Leonard, I. Aifantis, A. Melnick

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. Valls, L. Wang, M. Cardenas, W. Tam, K. Bhatt, J.P. Leonard, I. Aifantis

Study supervision: L. Cerchietti, A. Melnick

We thank the Weill Cornell Epigenomics Core Facility for support with ChIP-on-chip assays. We thank Dr. Artavanis-Tsakonas for sharing the ROSA26-NOTCH2IC animals.

This research was funded by a follicular lymphoma research grant from the Lymphoma Research Foundation to A. Melnick. A. Melnick is also supported by NCIR01CA104348, NCIR01CA143032, and the Chemotherapy Foundation. E. Valls has been the recipient of an IDIBAPS Postdoctoral Fellowship-BIOTRACK, supported by the European Community's Seventh Framework Programme (EC FP7/2007-2013) under the grant agreement number 229673. I. Aifantis is supported by the NIH (RO1CA216421, RO1CA169784, RO1CA202025, R01CA133379, and R01CA149655) and the NYSTEM program of the New York State Health Department. C. Lobry is supported by a Leukemia and Lymphoma Society Career Development Award and a French government ATIP-Avenir grant. W. Ci is supported by the National Basic Research Programme (Grant No. 2016YFC0900303); the National Natural Science Foundation of China (Grant No. 81422035, 91519307, and 81672541); and the Chinese Academy of Science (CAS; Grant No. QYZDB-SSW-SMC039 and KJZD-EW-L14).