Selective Inhibition of HDAC3 Targets Synthetic Vulnerabilities and Activates Immune Surveillance in Lymphoma

Diffuse large B-cell lymphoma (DLBCL) and follicular lymphoma are the two most frequent subtypes of non-Hodgkin lymphoma. These diseases originate from germinal center B (GCB) cells, which are at a stage of development that naturally allows for the proliferation and affinity maturation of antigen-experienced B cells to produce terminally differentiated memory B cells or plasma cells. The germinal center (GC) reaction is regulated by B cell–intrinsic activation and suppression of genes by master regulators such as the BCL6 transcription factor (1), and extrinsically via the interaction of GCB cells with follicular helper T (T_FH) cells and other immune cells within the GC (2). The BCL6 transcription factor is critical for GCB-cell development and coordinately suppresses the expression of large sets of genes by recruiting SMRT/NCOR corepressor complexes containing HDAC3 (3), the LSD1 histone demethylase (4), and tethering a noncanonical polycomb repressor 1–like complex in cooperation with EZH2 (5). These genes are normally reactivated to drive GC exit and terminal differentiation in the GC light zone, but the epigenetic control of these dynamically regulated GC transcriptional programs is perturbed in DLBCL and follicular lymphoma via the downstream effects of somatic mutation of chromatin-modifying genes (CMG; ref. 6).

The second most frequently mutated CMG in both DLBCL and follicular lymphoma is the CREBBP gene, which encodes a histone acetyltransferase that activates transcription via acetylation of histone H3 at lysine 27 (H3K27Ac) and other residues. We have previously found that these mutations arise as early events during the genomic evolution of follicular lymphoma and reside in a population of tumor-propagating cells, often referred to as common progenitor cells (CPC; ref. 7). We have also noted an association between CREBBP inactivation and reduced expression of MHC class II in human and murine lymphomas (7, 8). The expression of MHC class II is critical for the terminal differentiation of B cells through the GC reaction (9). The interaction with T helper cells via MHC class II results in B-cell costimulation through CD40 that drives NFκB activation and subsequent IRF4-driven suppression of BCL6. However, in B-cell lymphoma, tumor antigens may also be presented in MHC class II and recognized by CD4⁺ T cells that drive an antitumor immune response (10, 11). The active suppression of MHC class II expression in B-cell lymphoma may therefore be driven by evolutionary pressure against MHC class II–binding tumor antigens, as recognized in other cancers (12). In support of this notion, the reduced expression of MHC class II has been found to be associated with poor outcome in DLBCL (13, 14).

Recently, MHC class II expression has been defined as an important component of IFNγ-related signatures that are predictive of the activity of PD-1–neutralizing antibodies (14–17). This is consistent with a prominent role for CD4⁺ T cells in directing antitumor immunity and responses to immunotherapy (18). Despite this, current immunotherapeutic strategies largely rely on the preexistence of an inflammatory microenvironment for therapeutic efficacy. Here, we have characterized the molecular consequences of CREBBP mutations and identified BCL6-regulated cell cycle, differentiation, and IFN signaling pathways as core features that are aberrantly silenced at the epigenetic and transcriptional level. We show that HDAC3 inhibition specifically restores these pathways, thus suppressing growth and most critically enabling T cells to recognize and kill lymphoma cells. Together, these findings highlight multiple mechanisms by which selective inhibition of HDAC3 can drive tumor-intrinsic killing as well as activate IFNγ signaling and antitumor immunity that extends to both CREBBP wild-type and CREBBP-mutant tumors.

In B-cell lymphomas, the CREBBP gene is predominantly targeted by point mutations that result in single amino acid substitutions within the lysine acetyltransferase (KAT) domain (7, 19), with a hotspot at arginine 1446 (R1446) that leads to a catalytically inactive protein (20, 21). However, all of the prior studies characterizing the effects of CREBBP mutation have been performed using knockout (KO) or knockdown of Crebbp, resulting in loss of protein (LOP; refs. 8, 19, 22–25). Furthermore, mutations of R1446 have not been documented in any lymphoma cell line. We therefore opted to investigate whether there may be unique functional consequences of KAT domain hotspot mutations in CREBBP. To achieve this, we utilized CRISPR/Cas9-mediated gene editing with two unique guide RNAs (gRNA) to introduce the most common CREBBP mutation, R1446C, into a CREBBP wild-type cell line bearing the t(14;18)(q21;q32) translocation RL (Fig. 1A). This allowed us to generate clones from each gRNA that had received the constructs but remained wild-type (CREBBP^WT), those that edited their genomes to introduce the point mutations into both alleles (CREBBP^R1446C), and those that acquired homozygous frameshift mutations resulting in LOP (CREBBP^KO; Supplementary Fig. S1). These isogenic sets of clones differ only in their CREBBP mutation status, and therefore allow for detailed functional characterization in a highly controlled setting.

Western blot analysis confirmed that the CREBBP^R1446C protein was still expressed and that CREBBP^KO mutations resulted in a complete loss of protein expression (Fig. 1B). Western blotting with densitometry of the H3K27Ac mark revealed significant reduction in H3K27Ac in CREBBP^R1446C cells versus isogenic CREBBP^WT controls (P < 0.001; Fig. 1C). Although CREBBP^KO cells also showed lower H3K27Ac abundance than isogenic CREBBP^WT controls, this reduction was not statistically significant (P = 0.106). We performed chromatin immunoprecipitation sequencing (ChIP-seq) for H3K27Ac to define the physical location of these changes and identify potentially deregulated genes. This revealed 2,022 regions with significantly reduced acetylation, and 2,304 regions with significantly increased acetylation in CREBBP^R1446C cells compared with isogenic CREBBP^WT controls (fold change >1.5, q-value < 0.01; Fig. 1D; Supplementary Fig. S2; Supplementary Table S1). This pattern was mirrored by another CREBBP-catalyzed mark, H3K18Ac, and the loss or gain of histone acetylation was also accompanied by gain or loss of H3K27me3, respectively (Supplementary Fig. S3A–S3D). Regions with loss of H3K27Ac were observed to normally bear this mark in human GCB cells (ref. 24; Fig. 1D), suggesting that CREBBP^R1446C mutations lead to loss of a normal GCB epigenetic program. Notably, CREBBP^KO resulted in a reduction of H3K27Ac in these regions also, but at a lower magnitude than that observed with CREBBP^R1446C mutations (Fig. 1D and E). This was not observed for regions with increased H3K27Ac, which showed little change in CREBBP^KO cells.

Using RNA sequencing (RNA-seq), we observed broad changes in transcription, with 766 genes showing significantly increased expression and 733 genes showing significantly decreased expression in CREBBP^R1446C cells compared with CREBBP^WT isogenic controls (fold change >1.5, q-value < 0.01; Fig. 1F; Supplementary Table S2). The genes that were proximal to regions of H3K27Ac loss showed a coordinate reduction in transcript abundance and vice versa (FDR < 0.001; Fig. 1G), suggesting that these broad changes in transcription are directly linked to altered promoter/enhancer activity as a result of differential H3K27Ac. Importantly, we observed significant enrichment of the transcriptional signature induced by shRNA knockdown of CREBBP in murine B cells or human DLBCL cell lines (ref. 8; Fig. 1H). However, an even more significant enrichment was observed for the signature of genes we defined as being lost in association with CREBBP mutation in primary human CREBBP mutant follicular lymphoma B cells (ref. 7; FDR < 0.001; Fig. 1H). Consistent with our CRISPR cell line results, primary follicular lymphoma B cells with CREBBP LOP mutations were observed to have significantly less repression of this signature by single-sample gene set enrichment analysis (ssGSEA) compared with those with CREBBP KAT domain mutation (P = 0.039; Supplementary Fig. S4).

We used epigenetic and transcriptional profiles and hypergeometric analysis to gain insight into biological functions disrupted by CREBBP^R1446C mutations (Fig. 1I; Supplementary Tables S3 and S4). This confirmed a significant enrichment of genes downregulated in patients with CREBBP-mutant lymphoma, and further revealed an enrichment of BCL6–SMRT and BCL6 targets, consistent with the proposed role of CREBBP in opposing BCL6-mediated transcriptional repression. The biological functions of these genes included B-cell receptor (BCR) and NFκB signaling in addition to IFN signaling and antigen presentation (Fig. 1I). In line with the enrichment of antigen-presenting pathways and the genome-wide differences in H3K27Ac patterning, we observed more severe reduction of H3K27Ac over the MHC class II locus in CREBBP^R1446C versus CREBBP^KO cells (Fig. 1J). This was associated with an approximately 2-fold reduction of cell surface MHC class II in CREBBP^KO cells compared with isogenic CREBBP^WT controls, but an approximately 37-fold reduction in CREBBP^R1446C cells (Fig. 1K). Notably, EP300 is expressed at a high level in all of the CRISPR-modified cell lines, and ChIP-qPCR showed that the MHC class II locus is bound by both CREBBP and EP300 in these cells (Supplementary Fig. S5A–S5C). Moreover, inhibition of EP300 activity in CREBBP^KO cells reduced the expression of MHC class II to levels similar to that in CREBBP^R1446C cells (Supplementary Fig. S5D), suggesting that the redundant activity of EP300 may partially sustain MHC class II expression in CREBBP^KO cells. These data further support a stronger epigenetic and transcriptional suppression of BCL6 target genes, including those involved in antigen presentation, with CREBBP^R1446C mutations compared with CREBBP^KO.

To confirm that the CREBBP KAT mutation is biologically active in primary GC B cells in a similar manner to that observed in CRISPR-engineered cell lines, we generated a novel genetically engineered mouse model with Cre-inducible expression of Crebbp^R1447C (equivalent to human R1446C). The endogenous Crebbp locus was engineered to switch from WT to mutant gene expression using a floxed inversion cassette (Supplementary Fig. S6A). These animals were crossed to mice engineered with the Cγ1-Cre allele to specifically induce recombination in GC B cells and bone marrow transplanted into irradiated CD45.1 recipients (Supplementary Fig. S6B and S6C). Transplanted mice were immunized with sheep red blood cells and sacrificed 10 days later when the GC reaction was at its peak (Supplementary Fig. S6C), whereupon flow cytometry was performed to assess B-cell populations. As expected, there was no perturbation of earlier stages of B-cell development (Supplementary Fig. S6D and S6E). Cre recombination was validated in sorted GC B cells (Supplementary Fig. S6F). Cγ1Cre;Crebbp^R1447C/+ GC B cells showed significantly reduced MHC class II expression compared with Cγ1Cre;Crebbp^+/+ controls (P = 0.02, Supplementary Fig. S6G), whereas (as expected) there was no difference in MHC II among naïve B cells from either genotype (Supplementary Fig. S6H). Finally, Cγ1Cre;Crebbp^R1447C/+ mice manifested a significantly skewed light-zone:dark-zone ratio in favor of increased dark-zone cells (P = 0.01; Supplementary Fig. S6I–S6L), which are the cells that most reflect the actions of BCL6, although the overall abundance of GC B cells was not increased (Supplementary Fig. S6M and S6N). These results confirm in primary cells our MHC class II findings from CRISPR/Cas9-edited clones and hence their value as a platform for mechanistic studies.

Finally, mutations in CREBBP have been previously associated with adverse outcome in follicular lymphoma and are incorporated into the M7-FLIPI prognostic index (26). However, in these analyses, all CREBBP mutations were considered collectively without discriminating between KAT domain point mutations and nonsense/frameshift LOP mutations. We therefore reevaluated these data in light of our observed functional differences between these mutations. This showed that there was a significant difference in failure-free survival (Fig. 1L) between patients with these two classes of CREBBP mutations. This was not significant for overall survival (Fig. 1M). Specifically, patients bearing KAT domain point mutations in CREBBP (22% of which were R1446 mutations) had a significantly reduced failure-free survival compared with patients with LOP mutations in CREBBP (log-rank P = 0.026), with no other clinical factors being significantly different between these groups (Supplementary Table S5). Collectively, these results provide the first direct experimental description of the role of CREBBP^R1446C mutations in lymphoma B cells. Our data suggest that CREBBP KAT domain mutations may have a potential dominant repressive function on loci that can be targeted by redundant acetyltransferases, thereby driving a more profound molecular phenotype that is associated with a worse patient outcome.

Our genomic analysis showed that BCL6 target genes were significantly enriched among those with reduced H3K27Ac and gene expression in CREBBP^R1446C cells. We therefore evaluated CREBBP and BCL6 binding over these regions using data from normal GC B cells (24) and found that the regions with CREBBP^R1446C mutation–associated H3K27Ac loss are bound by both proteins at significantly higher levels than H3K27Ac peaks that remain unchanged with CREBBP mutation (CREBBP Wilcoxon P = 1.2⁻⁴¹; BCL6 Wilcoxon P = 2.83⁻⁵²) or peaks with increased H3K27Ac in CREBBP^R1446C mutants (CREBBP Wilcoxon P = 3.01⁻¹³⁷; BCL6 Wilcoxon P = 3.17⁻¹²⁸; Fig. 2A). This suggests that these genes may be antagonistically regulated by CREBBP and BCL6, the latter of which mediates gene repression via recruitment of HDAC3-containing NCOR/SMRT complexes (3, 27). The epigenetic suppression of gene expression in CREBBP-mutant cells may therefore be dependent upon HDAC3-mediated suppression of BCL6 target genes. Using a selective HDAC3 inhibitor, BRD3308 (28), we found that CREBBP^R1446C and CREBBP^KO clones showed greater sensitivity to HDAC3 inhibition compared with isogenic WT controls in cell-proliferation assays (Fig. 2B). We confirmed this as being an on-target effect of BRD3308 by performing shRNA knockdown of HDAC3 and observing a greater effect on cell proliferation in CREBBP^R1446C cells compared with isogenic controls (Fig. 2C; Supplementary Fig. S7). More over, HDAC3 inhibition was able to efficiently promote the accumulation of H3K27Ac in a dose-dependent manner in both CREBBP^WT and CREBBP^R1446C cells, as compared with the inactive chemical control compound BRD4097 (Fig. 2D). This suggests that the increased sensitivity to HDAC3 inhibition in CREBBP^R1446C cells may be linked with an acquired addiction to an epigenetic change driven by CREBBP mutation. We posited that one of these effects may be the suppression of CDKN1A (encoding p21) expression, which is a key BCL6 target gene (29) that has reduced levels of H3K27Ac in both CREBBP^R1446C and CREBBP^KO cells (Fig. 2E). In support of this, we observed a marked induction of p21 expression by BRD3308 (Fig. 2F) and observed that shRNA-mediated silencing of CDKN1A partially rescued the effect of BRD3308 on cell proliferation (Fig. 2G). Therefore, CREBBP mutations, regardless of type, sensitize cells to the effects of HDAC3 inhibition in part via the induction of p21.

We aimed to confirm this trend using a larger panel of DLBCL cell lines with either WT (n = 6) or mutant CREBBP (n = 6). This revealed a significantly lower effective dose 50 (ED₅₀) of BRD3308 in CREBBP mutant compared with WT cell lines (P = 0.002, Fig. 2H–I). We did not observe this trend using the nonspecific HDAC inhibitors romidepsin and SAHA (Supplementary Fig. S8). Notably, none of these cell lines harbor KAT domain missense mutations in CREBBP, providing further evidence that sensitivity to HDAC3 selective inhibition is independent of mutation type (i.e., KAT domain missense vs. nonsense/frameshift). Furthermore, we also observed the induction of p21 expression and apoptosis in CREBBP WT cell lines, although to a lesser degree (Supplementary Fig. S9). To gain greater insight into this, we performed RNA-seq of CREBBP WT (OCI-Ly1) and CREBBP-mutant (OCI-Ly19, OZ) cell lines treated with BRD3308. Notably, the ability of HDAC3 inhibition to induce the expression of genes involved in the terminal differentiation of B cells was conserved in both WT and mutant cell lines (Fig. 2J). These results are consistent with the role of BCL6 in controlling checkpoints, terminal differentiation, and other functions (1) and point toward induction of these transcriptional programs as contributing to the antilymphoma response induced by HDAC3 inhibition in both the CREBBP WT and CREBBP-mutant settings. Although targetable by HDAC3 inhibition in WT cells, the BCL6–HDAC3 target gene set is more significantly perturbed in the context of CREBBP mutation leading to an enhanced cell-intrinsic effect of HDAC3 inhibition.

Our observation that HDAC3 inhibition can affect both CREBBP mutant and WT B cells led us to test its efficacy in primary patient-derived xenograft (PDX) models of DLBCL. To achieve this, we first expanded CREBBP WT tumors in vivo and transitioned them to our novel in vitro organoid system for exposure to BRD3308 (30). All tumors that were tested showed a dose-dependent reduction in cell viability when cultured with BRD3308, compared with the vehicle control (Fig. 3A). We therefore treated 3 of these CREBBP WT tumors in vivo with either 25 mg/kg or 50 mg/kg of BRD3308, which led to a significant reduction in growth of PDX tumors treated with either dose compared with vehicle control (Fig. 3B). We were able to obtain only a single primary human CREBBP-mutant (R1446C) DLBCL PDX model, which could be grown only by renal capsule implantation. Treatment of these tumors with 25 mg/kg BRD3308 and monitoring by weekly MRI (Fig. 3C) showed a significant reduction in growth compared with vehicle control–treated tumors (Fig. 3D). In line with our cell line models, quantitative PCR analysis of treated PDX tumors showed an induction of BCL6 target genes with a role in B-cell terminal differentiation, including IRF4, PRDM1, CD138, and CD40 compared with vehicle-treated tumors (Supplementary Fig. S10). Therefore, selective inhibition of HDAC3 may be a rational approach for targeting the aberrant epigenetic silencing of BCL6 target genes in primary human B-cell lymphoma.

We aimed to take a deeper look at the molecular consequences of HDAC3 inhibition by performing H3K27Ac ChIP-seq and RNA-seq of CREBBP^R1446C-mutant cells after exposure to BRD3308, as compared with the inactive negative control compound BRD4097. This showed that selective HDAC3 inhibition promoted the gain of H3K27Ac over a broad number of regions (n = 6,756; Fig. 4A). Strikingly, HDAC3 inhibition significantly either restored or further increased the abundance of H3K27Ac at 28% (558/1,975) of sites that became deacetylated in CREBBP^R1446C (hypergeometric enrichment P = 3.02⁻¹⁷⁸), consistent with the role of HDAC3 in opposing CREBBP functions (Fig. 4B). Indeed, a more quantitative analysis indicated HDAC3 inhibition coordinately increased the level of H3K27Ac over the same loci that showed reduced levels in CREBBP^R1446C compared with CREBBP^WT cells, although this restoration of H3K27Ac was not sufficient to completely revert the epigenomes of CREBBP^R1446C cells to the level that was observed in CREBBP^WT cells (Fig. 4C). In line with the role of HDAC3 and BCL6 in transcriptional repression, BRD3308 induced an expression profile that was markedly skewed toward gene upregulation (n = 1,467 vs. 208 genes downregulated; Fig. 4D; Supplementary Table S6 and S7), including upregulation of IFN-responsive genes such as antigen-presenting machinery and PD-L1 (CD274). Notably, the genes with increased expression following HDAC3 inhibition were significantly enriched for those that lose H3K27Ac in CREBBP^R1446C-mutant compared with WT (FDR < 0.001; Fig. 4E), further supporting the conclusion that HDAC3 inhibition directly counteracts changes associated with CREBBP mutation. A quantitative assessment of ChIP-seq signal indicated that enhancers manifested greater gain of H3K27Ac as compared with promoters in cells exposed to HDAC3i (Wilcoxon P < 0.001; Fig. 4F), although MHC class II genes also showed coordinate increases in promoter H3K27Ac (Fig. 4G). Analysis of critical gene loci that are deregulated by CREBBP mutation, such as the MHC class II gene cluster and CIITA, exemplify this induction of H3K27ac (Fig. 4H–I). We validated the increased H3K27Ac and expression of these genes in independent experiments wherein CREBBP^R1446C or isogenic control cells were treated with BRD3308 or vehicle, H3K27Ac was measured by ChIP-qPCR (Supplementary Fig. S11), and transcript abundance was measured by qPCR (P < 0.001; Fig. 4J). We further validated that this was an on-target effect of BRD3308 by performing shRNA-mediated knockdown of HDAC3, which resulted in the increased expression of these genes relative to control shRNA (Fig. 4K). Together, these data indicate that the aberrant mutant CREBBP epigenetic and transcriptional program can be restored by selective pharmacologic inhibition of HDAC3.

IFN signaling and antigen-presenting genes have not been well investigated as downstream targets of BCL6–HDAC3 complexes, but were enriched in genes suppressed by CREBBP mutation and restored by HDAC3 inhibition. Given their critical role in antitumor immunity, we evaluated whether HDAC3 inhibition may be sufficient to restore or promote the expression of these BCL6-repressed immune signatures. Using MHC class II protein expression on CREBBP^R1446C-mutant cells as a proxy for the CREBBP/BCL6 counter-regulated IFN signaling pathway, we evaluated the activity of HDAC inhibitors for promoting the expression of immune response genes. Although HDAC inhibitors with broader specificities were able to induce MHC class II expression, selective inhibition of HDAC3 was sufficient for the robust and maximal (>10-fold) restoration of MHC class II expression in CREBBP^R1446C cells (Fig. 5A; Supplementary Fig. S12) in line with our observation that these genes are silenced by BCL6–HDAC3 complexes. These selective inhibitors each show some inhibition of other HDACs (28), and their activity in opposing BCL6 function may in part be linked to inhibition of HDAC1/2, which are recruited by BCL6 via CoRest and NuRD (4, 31, 32). However, HDAC1/2 have important functions in normal hematopoiesis (33), and hence compounds that target these enzymes induce toxic effects against these cells that are not elicited by BRD330829 (29), suggesting that selective inhibition of HDAC3 may limit hematologic toxicities that are observed with pan-HDAC inhibitors. Furthermore, although some of the less specific HDAC inhibitors were toxic to normal human CD4⁺ and CD8⁺ T cells, the selective inhibition of HDAC3 was not (Fig. 5B; Supplementary Fig. S13). This suggests that selective HDAC3 inhibition may be capable of eliciting immune responses by on-target on-tumor induction of antigen presentation without on-target off-tumor killing of T cells.

Based upon our observations with MHC class II, the magnitude of this induction appeared to be dependent upon the baseline of expression. We therefore hypothesized that CREBBP mutation status may determine the baseline expression for IFN and antigen-presenting pathways, but may not be a prerequisite for this response due to the conserved activity of BCL6–HDAC3 in regulating this axis in WT cells. We comprehensively evaluated this theory using our RNA-seq data from CREBBP^WT and CREBBP^R1446C cells treated with either the HDAC3 inhibitor BRD3308 or the inactive control compound BRD4097. Consistent with our observations of MHC class II protein expression, BRD3308 treatment coordinately induced the transcript expression of MHC class II and IFN pathway genes in both CREBBP^WT and CREBBP^R1446C CRISPR-edited cells (Fig. 5C). This trend included a significant enrichment for the genes that were epigenetically suppressed in association with CREBBP mutations, resulting in their significant increase in expression in CREBBP^WT cells (Fig. 5D), similar to our observations from CREBBP^R1446C cells (Fig. 4E). Moreover, we observed similar enrichment of transcriptionally induced pathways in both CREBBP^WT and CREBBP^R1446C cells that included the same pathways that were suppressed by CREBBP mutation (Fig. 5E; Supplementary Tables S7–S10). Consistent with the almost exclusive function of HDAC3 as a BCL6 corepressor in GC B cells (3), and the importance of BCL6 activity in both CREBBP WT and CREBBP-mutant cells, we observed highly significant enrichment for genes regulated by BCL6–SMRT complexes among genes with induced H3K27Ac and expression after BRD3308 treatment (Fig. 5E). Also significantly enriched were canonical BCL6 target gene sets such as p53-regulated genes, and signaling through BCR, CD40, and cytokines such as IL4 and IL10. Finally, we observed significant enrichment for BCL6 target gene sets linked to IFN signaling, antigen presentation via MHC class II, and PD-1 signaling.

Although there were conserved patterns of gene activation in both CREBBP^WT and CREBBP^R1446C cells, we observed that the magnitude of this induction was greatest in CREBBP^R1446C cells, which started from a lower baseline of expression linked to mutation-associated epigenetic suppression (Fig. 5F). We identified IRF1 as a BCL6-regulated transcription factor that is critical for IFN responses, and is induced by HDAC3 inhibition in CREBBP^WT and CREBBP^R1446C cells (Fig. 5C). We therefore hypothesized that IRF1 may contribute to the different magnitude of induction in MHC class II genes following HDAC3 inhibition in these two genetic contexts. We investigated this by measuring IRF1 activity in a luciferase reporter assay in CREBBP^WT and CREBBP^R1446C-mutant cells. Treatment of CREBBP^R1446C cells with BRD3308 led to a significant increase of IRF1 activity that was similar in magnitude to that induced by IFNγ treatment (Fig. 5G). In contrast, this effect was not observed in isogenic CREBBP^WT control cells following HDAC3 inhibition, despite these cells showing similarly increased IRF1 activity in response to IFNγ treatment. Exposing CREBBP^R1446C cells to IFNγ-neutralizing antibodies only partially ameliorated MHC class II induction after BRD3308 in CREBBP-mutant cells (Fig. 5H). This suggests that preferential induction of antigen presentation by HDAC3 inhibition in CREBBP-mutant cells likely depends on a combination of mechanisms, including directly opposing BCL6 repression of these genes as well as secondary induction through IFNγ signaling (which is also directly regulated by BCL6).

The frequency of MHC class II loss in DLBCL exceeds the frequency of CREBBP mutations in this disease (13, 21), through unknown mechanisms. The ability of HDAC3 inhibition to induce MHC class II expression in CREBBP WT DLBCL cells may therefore have potentially broad implications for immunotherapy. Using RNA-seq data from WT (OCI-Ly1) and mutant (OCI-Ly19 and OZ) DLBCL cell lines (Fig. 6A; Supplementary Tables S11 and S12) and qPCR analysis of PDX tumors treated in vivo (Supplementary Fig. S14A–S14D), we confirmed that gene expression induced by HDAC3 inhibition was largely conserved in both contexts. The expression of CD274 (encoding PD-L1) was significantly higher in 4 of 4 PDX tumors treated with 25 mg/kg BRD3308 in vivo compared with vehicle control–treated tumors, and MHC class II expression was significantly higher in 3 of 4 PDX tumors (Supplementary Fig. S14A–S14D). We extended upon this using IHC staining for MHC class II in tumors from a CREBBP R1446C–mutant PDX model (DFBL; Fig. 6B) and a CREBBP WT PDX model that was MHC class II negative at baseline (DR-NY2, Fig. 6C). In tumors from both models, in vivo treatment with BRD3308 led to a marked induction of MHC class II expression. In a broader panel of CREBBP WT and CREBBP-mutant cell lines, we observed that a core set of genes including HLA-DR, CIITA, and CD274 had consistently higher expression in BRD3308-treated cells compared with the matched control, but with a higher magnitude of induction in CREBBP-mutant cell lines (Fig. 6D). This trend was also observed by flow cytometry for MHC class II, which extends upon our observations in CRISPR/Cas9-modified cells by showing a reproducible increase in expression in a larger panel of DLBCL cell lines, and a higher magnitude of induction in CREBBP-mutant cell lines (Fig. 6E). Together, these results show that selective inhibition of HDAC3 using BRD3308 can promote the expression of IFN and antigen-presenting pathway genes in both the CREBBP WT and CREBBP-mutant settings. However, the magnitude of induction is greatest in CREBBP-mutant cells, owing in part to the preferential induction of IRF1 activity in these cells.

IFN signaling and antigen presentation are central to antitumor immune responses. We therefore investigated whether HDAC3 inhibition could promote antigen-dependent antitumor immunity. To test this, we implanted OCI-Ly18 DLBCL cells into NSG mice, and once tumors formed we injected human peripheral blood mononuclear cells (PBMC), including T cells, to expose them to antigens presented by the tumor cells. These are likely to be histocompatibility antigens rather than tumor neoantigens, but are nonetheless presented and recognized through MHC–T-cell receptor (TCR) interactions. After in vivo priming, tumor-infiltrating lymphocytes (TIL) were cocultured in vitro with OCI-Ly18 cells that were pretreated for 72 hours with increasing concentrations of BRD3308 to assess the effects on T-cell activation and tumor-cell killing (Fig. 7A). The DLBCL cells that were epigenetically primed for antigen presentation by BRD3308 significantly increased the activation of CD4⁺ T cells, as determined by CD69 expression (P < 0.05; Fig. 7B). As in prior experiments, we observed cell-intrinsic effects of BRD3308 on OCI-Ly18 cells in the absence of TILs, resulting in declining cell viability with higher concentrations of the inhibitor. However, the effects of BRD3308 were markedly and significantly increased in the presence of TILs, consistent with T cell–directed killing of the tumor cells (P < 0.001; Fig. 7C). To confirm that this killing was dependent on MHC–TCR interactions, we also performed this experiment in the presence of blocking antibodies for MHC class I, MHC class II, or both. Blocking one or the other of MHC class I or II rescued some of the cytotoxicity observed in this assay, but blocking both MHC class I and class II together completely abrogated the TIL-associated effect (Fig. 7C). These data show that HDAC3 inhibition can potentiate antitumor immune responses that are likely to be antigen-dependent because they are driven by MHC–TCR interactions. Our identification of the IFN signaling pathway, and IFNγ itself, as a central component of the response to HDAC3 inhibition led us to test whether IFNγ levels may rise as a result of treatment. An ELISpot analysis of the IFNγ levels from the TIL and OCI-Ly18 coculture experiment revealed a striking and significant increase in IFNγ levels, even at the lowest concentrations of HDAC3 inhibitor (Fig. 7D; Supplementary Fig. S15).

Considering the manner in which to best harness the antilymphoma immunity effect of HDAC3 inhibitors, we noted that induction of MHC class II is mechanistically linked to IFNγ-associated PD-L1 upregulation, which could potentially limit maximal antitumor response. We therefore posited that PD-1/PD-L1 blockade may be an attractive combination regimen. To test this hypothesis, we used a murine lymphoma transplantation model in which splenocytes from IμBcl6;Ezh2^Y641F mice (5) were transplanted into irradiated syngeneic WT recipients (Fig. 7E). Aggressive tumors formed within these mice, which we treated with either vehicle control, BRD3308 alone, αPD-L1 alone, or a BRD3308 + αPD-L1 combination. Treatment with BRD3308 led to a significant increase in serum IFNγ within these mice, which was also observed with αPD-L1 treatment and was even further enhanced in an additive manner by the combination treatment (Fig. 7F). Immunofluorescence staining for CD4 and CD8 showed significantly increased infiltration within the spleens from BRD3308-treated mice, which was further increased by the combination therapy (Fig. 7G–J). The ability of BRD3308 and αPD-L1 combination to increase TILs was likely linked with the IFNγ-mediated PD-L1 induction that was observed in BRD3308-treated tumors (Fig. 7K and L). In line with this interaction, BRD3308 and αPD-L1 each showed a small degree of single-agent efficacy, but the combination led to the most significant reduction in B220 cells within the spleens of tumor-bearing mice (Fig. 7M and N). Together, these data show that HDAC3 inhibitor–mediated reversal of the BCL6-repressed IFN pathway leads to joint induction of MHC class II and PD-L1, which, although eliciting significant antitumor immune response, can be further enhanced by combination with PD-1/PD-L1 blockade to yield a more potent immunotherapy strategy that is superior to immune checkpoint inhibition alone.

Precision medicine and immunotherapy have led to significant breakthroughs in a variety of cancers, but have lacked success in B-cell lymphoma. For precision medicine, this is largely due to a paucity of “actionable” genetic alterations or, rather, the lack of current therapeutic avenues to target the mutations that have been defined as being important for disease biology. For immunotherapy, the mechanisms driving lack of response or early progression are not well understood, but are likely to be underpinned by the complex immune microenvironment and genetic alterations that drive immune escape. The exception to both of these statements is the use of PD-1–neutralizing antibodies in classic Hodgkin lymphoma, which opposes genetically driven immune suppression by the malignant Reed–Sternberg cells through DNA copy-number gain of the PD-L1 locus (34) and elicits responses in the majority of patients (35). This stands as an example of the potential success that could be achieved by the characterization and rational therapeutic targeting of genetic alterations and/or the neutralization of immune escape mechanisms. However, we are not yet able to successfully target some of the most frequently mutated genes or overcome the barriers of inadequate response to immunotherapy in the most common subtypes of B-cell lymphoma. These are important areas of need if we hope to further improve the outcomes of these patients.

The CREBBP gene is mutated in approximately 15% of DLBCL (21) and approximately 66% of follicular lymphoma (7), and is therefore a potentially high-yield target for precision therapeutic approaches. Our use of CRISPR/Cas9 gene editing to generate isogenic lymphoma cell lines that differ only in their CREBBP mutation status allowed us to perform detailed characterization of the epigenetic and transcriptional consequences of these mutations. Using this approach, we identified for the first time functional differences between the most frequent KAT domain point mutation, R1446C mutations, and frameshift mutations that result in KO. Although similar regions of the genome showed reduced H3K27Ac in R1446C and KO mutants compared with isogenic WT controls, the magnitude of these changes was markedly reduced in CREBBP KO. This suggests that acetyltransferases such as EP300 may compensate for the loss of CREBBP protein in the setting of CREBBP nonsense/frameshift mutations, consistent with observations of functional redundancy between Crebbp and Ep300 in B cell–specific conditional KO mice (25, 36). But the presence of a catalytically inactive CREBBP^R1446C protein may dominantly suppress histone acetylation by preventing the participation of redundant acetyltransferases such as EP300 in transcriptional activating complexes. This yields a more severe epigenetic and transcriptional phenotype in CREBBP^R1446C mutants compared with CREBBP^KO mutants, and an inferior clinical outcome.

Despite differences in the magnitude of molecular changes between R1446C and KO cells, these mutations yielded a similar degree of synthetic vulnerability to HDAC3 inhibition. This was likely driven by the increased suppression of BCL6 target genes that we observed in both contexts, including CDKN1A (encoding p21). This gene has also been highlighted as a critical nexus in the oncogenic potential of EZH2 (37), which cooperates with BCL6 to silence gene expression (5). One of the important mechanisms for BCL6-mediated gene silencing is through the recruitment of HDAC3 as part of the NCOR/SMRT complex (3), thereby highlighting HDAC3 inhibition as a rational therapeutic avenue for counteracting BCL6 activity. Virtually all of the HDAC3 corepressor complexes present in DLBCL cells are bound with BCL6, suggesting that HDAC3 inhibitors' effect is largely explained by their depression of the subset of BCL6 target genes regulated through this mechanism (3). Importantly, as a variety of GCB-derived malignancies rely on BCL6 function independently of CREBBP or EZH2 mutation (38, 39), opposing this function through HDAC3 inhibition may also be effective in tumors without these genetic alterations. We have shown preliminary evidence in the primary setting using DLBCL PDX models.

One of the important transcriptional programs that is regulated by BCL6 is the IFN signaling pathway (3), which we observed to be significantly repressed in CREBBP-mutant cells. It has been also long been recognized that IFNγ cooperates with lymphocytes to prevent cancer development (40). IFN signaling supports T cell–driven antitumor immunity via a variety of mechanisms, including the induction of antigen presentation on MHC class II (41), and the expression of MHC class II–restricted tumor antigens is required for the spontaneous or immunotherapy-induced rejection of tumors (14). We have shown that the selective inhibition of HDAC3 is sufficient for broadly restoring the reduced H3K27Ac and gene expression that is associated with CREBBP mutation, including the IFN signaling and antigen-presenting programs. This was in part driven by the increased production of IFNγ following HDAC3 inhibition, but also via the induced expression and activity of the IRF1 transcription factor. Together, these factors led to a robust restoration of MHC class II expression on CREBBP-mutant cells and drove dose-dependent potentiation of antitumor T-cell responses. However, we also noted that the same molecular signature is promoted by HDAC3 inhibition in CREBBP WT cells, also with an associated increase of MHC class II expression. As with the cell-intrinsic effects of HDAC3 inhibition, the effects on immune interactions may therefore be active in both CREBBP WT and CREBBP-mutant cells as a result of the conserved molecular circuitry controlling these pathways in each genetic context.

A variety of HDAC inhibitors were capable of restoring antigen presentation in our models, and clinical responses are observed with these agents in patients with relapsed/refractory follicular lymphoma and DLBCL. However, grade 3 to 4 hematologic toxicities such as thrombocytopenia, anemia, and neutropenia were frequent, and the responses tended not to be durable (42, 43). We posit that specific inhibition of HDAC3 may be accompanied by reduced toxicity, as HDAC1 and HDAC2 have important roles in hematopoiesis (33), and avoiding the inhibition of these HDACs may therefore avoid the undesired hematologic effects associated with pan-HDAC inhibition. In line with this, we observed that BRD3308 was less toxic to CD4⁺ and CD8⁺ T cells than pan-HDAC inhibitors, and was able to elicit MHC-dependent T-cell responses against a CREBBP WT DLBCL cell line in vitro. Although these responses are likely driven by histocompatibility antigens rather than tumor neoantigens, they support the premise that selective HDAC3 inhibition is capable of promoting antigen presentation and immune responses. We also speculate that the clinically observed lack of durability for HDAC inhibitors may be the result of adaptive immune suppression through mechanisms such as PD-L1 induction, which dampens T-cell responses through the PD-1 receptor (44), as well as direct toxicity of pan-HDAC inhibitors to T cells. We found evidence for adaptive immune suppression within our model systems, showing that HDAC3 inhibition leads to increased IFNγ production and the upregulation of PD-L1 expression. This is in line with recent observations that CD274 (encoding PD-L1) is a BCL6-supressed gene (45) and with a well-characterized role for PD-L1 as an IFNγ-responsive gene (44). In other cancers in which a florid antigen-driven immune response and concomitant adaptive immune suppression via PD-L1 exist within the tumor microenvironment, blockade of the PD-1 receptor is an effective therapeutic strategy (15, 17). Recent studies have also shown that the efficacy of PD-1 blockade is inextricably linked with the existence of an IFN-driven immune response and expression of MHC class II (15, 17). We have shown some evidence for this in a syngeneic, BCL6-driven murine model of B-cell lymphoma, wherein the combination of HDAC3 inhibition and αPD-L1 led to enhanced levels of CD4⁺ and CD8⁺ T-cell infiltration and clearance of tumor cells. Together, these observations suggest that the greatest potentiation of antitumor immunity in GCB-derived malignancies may be achieved through stimulation of IFN signaling and MHC class II expression by HDAC3 inhibition, in combination with the blockade of adaptive immune suppression using PD-L1/PD-1 neutralizing antibodies. However, this concept requires further exploration in future studies.

In conclusion, this work defines a molecular circuit that controls the survival and differentiation of lymphoma B cells and their interaction with T cells. This circuit is antagonistically regulated by CREBBP and BCL6, and can be pushed toward promoting tumor cell death and antitumor immunity via selective inhibition of HDAC3. This highlights HDAC3 inhibition as an attractive therapeutic avenue, which may be broadly active in follicular lymphoma and DLBCL due to the near-ubiquitous role of BCL6, but which may have enhanced potency in CREBBP-mutant tumors.

For detailed methods, please refer to the Supplementary Methods.

The RL cell line (ATCC CRL-2261) was modified by electroporation with one of two unique gRNA sequences in the pSpCas9(BB)-2A-GFP vector (Addgene plasmid #48138, gift from Feng Zhang, Massachusetts Institute of Technology; ref. 46) with a single-stranded oligonucleotide donor template. Single GFP-positive cells were sorted 3 to 4 days after transfection, and colonies expanded and evaluated for changes in the targeted region using Sanger sequencing. The process was repeated until point mutants were retrieved from cells treated with each of the two gRNAs, totaling 742 single clones. Potential off-target sites of each gRNA were determined by BLAST, and all sites with ≥16/20 nucleotide match to either of the gRNA sequences were interrogated by Sanger sequencing (Supplementary Table S13). All cells were maintained as subconfluent culture in RPMI medium with 10% FBS and penicillin/streptomycin and revalidated by Sanger sequencing prior to each set of experiments. Basic phenotyping of the cell lines is presented in Supplementary Fig. S16A–S16E.

Cells were washed and fixed in formaldehyde and chromatin sheared by sonication. An antibody specific to H3K27Ac (Active Motif) was coupled to magnetic protein G beads and incubated with chromatin overnight, and immunoprecipitation was performed. Input controls were reserved for comparison. Nucleosomal DNA was isolated and used either as a template for qPCR (ChIP-qPCR) or to generate NGS libraries using KAPA Hyper Prep Kits (Roche) and TruSeq adaptors (Bioo Scientific) using 6 cycles of PCR enrichment. Libraries were 6-plexed and sequenced with 2 × 100bp reads on a HiSeq-4000 (Illumina). The data were mapped using BWA, peaks called using MACS2, and differential analyses performed using DiffBind. For GSEA, the gene with the closest transcription start site to the peak was used. The statistical thresholds for significance were q < 0.05 and fold change > 1.5.

Total RNA was isolated using AllPrep DNA/RNA Kits (Qiagen) and evaluated for quality on a Tapestation 4200 instrument (Agilent). Total RNA (1 μg) was used for library preparation with KAPA HyperPrep RNA Kits with RiboErase (Roche) and TruSeq adapters (Bioo Scientific). Libraries were validated on a Tapestation 4200 instrument (Agilent), quantified by Qubit dsDNA Kit (Life Technologies), 6-plexed, and sequenced on a HiSeq4000 instrument at the MD Anderson Sequencing and Microarray Facility using 2 × 100 bp reads. Reads were aligned with STAR, and differential gene expression analysis performed with DEseq2. The statistical thresholds for significance were q < 0.05 and fold change > 2.

Cells were seeded in 96-well plates at 50,000 cells/100 μL/well with either vehicle (DMSO 0.1%) or increasing concentrations of drugs. Cell viability was assessed with the fluorescent redox dye CellTiter-Blue (Promega). The reagent was added to the culture medium at 1:5 dilution, according to the manufacturer's instructions. Procedures to determine the effects of certain conditions on cell proliferation and apoptosis were performed in three independent experiments. The two-tailed Student t test and Wilcoxon rank-sum test were used to estimate the statistical significance of differences among results from the three experiments. Significance was set at P < 0.05. The PRISM software was used for the statistical analyses.

A CREBBP R1446C–mutant tumor (DFBL13727) was obtained from the public repository of xenografts (47), expanded for 12 weeks in 1 mouse by surgical implantation into the renal capsule, and then implanted into the renal capsule of 12 additional mice for efficacy studies. Tumors were allowed to grow for 6 weeks, the size measured by MRI, and mice randomized to treatment and control groups to similar average tumor sizes. Mice were treated with BRD3308 (25 mg/kg) or vehicle control twice per day, 5 days on and 2 days off for a total of 3 weeks, and tumor size was assessed by MRI every 7 days. Two mice per group were euthanized on day 10 and the tumors harvested for biomarker analysis. Mice were cared for in accordance with guidelines approved by the MD Anderson Institutional Animal Care and Use Committee.

For CREBBP WT tumors (NY-DR2, DANA, and TONY), 6-week-old female NSG mice were implanted subcutaneously and treatments started when tumors reached 100 mm³. Mice (12/group) were randomized and dosed via oral gavage with BRD3308 (25 mg/kg) or control vehicle (0.5% methylcellulose, 0.2% Tween 80) twice daily for 21 consecutive days. Mice were cared for in accordance with guidelines approved by the Memorial Sloan Kettering Cancer Center Institutional Animal Care and Use Committee and Research Animal Resource Center.

For in vitro organoid culture, the tumors were dissociated in single cells, stained with carboxyfluorescein diacetate succinimidyl ester (CFSE), washed, and mixed with irradiated 40LB cells at a 10:1 ratio of Primary:40LB. The cell mixture was then used to fabricate organoids in a 96-well plate as described previously (48), with 20 μL organoids containing 3% silicate nanoparticles and a 5% gelatin in Iscove's Modified Dulbecco's Medium (IMDM) solution. The organoids were cultured in IMDM containing 20% FBS supplemented with antibiotics and normocin (Invivogen) for 6 days, doubling the volume of medium after 3 days. The cell mixture was exposed to four 1:3 serial dilutions of BRD3308 starting at 5 μmol/L or vehicle control (DMSO) in triplicate for 6 days, treating a second time at 3 days. After 6 days of exposure, cell viability and proliferation were assessed by flow cytometry using DAPI staining gating on CFSE-positive cells.

The OCI-LY18 cells were subcutaneously implanted in NSG mice and allowed to become established before the mice were injected with PBMC. After 2 weeks, the tumors were dissociated to single cells and CD3⁺ TILs were positively selected. T cells were expanded with a single administration of immunomagnetic microbeads coated with mouse anti-human CD3 and CD28 mAbs and rhIL2 and rhIL15 for 5 days. OCI-LY18 cells were treated with either DMSO or 10 μmol/L BRD3308 for 3 days, washed, and resuspended in fresh media with or without T cells at a ratio of 1:10. For MHC blocking, OCI-LY18 cells were treated with 10 μg/mL of either isotype Ig or blocking antibody against HLA-ABC W6/32, HLA-DR/DP/DQ, or the combination of the two. After 24-hour coculture, cell viability was analyzed using CellTiter-Blue.

Six-week-old female C56BL/6J mice were irradiated at day −1 and 0 with 250 rad, and after 2 hours transplanted with 1 × 10⁶ splenocytes from IμBcl6;Ezh2^Y641F mice (5) and 0.2 × 10⁶ bone marrow cells from a healthy age-matched donor. Three weeks after transplantation, mice became sick and treatment was initiated with either vehicle or BRD3308 (25 mg/kg twice daily, every day for 21 days) via oral gavage, with or without αPD-L1 delivered by intraperitoneal injection (250 μg twice weekly for 4 administration). Mice were monitored daily and euthanized when they became moribund. Spleen and liver were measured, weighed, and analyzed by flow cytometry for evaluating the disease. Immunofluorescence staining and imaging were performed at the Molecular Cytology Core Facility of Memorial Sloan Kettering Cancer Center (New York, NY). Sections were stained with either anti-CD4 (R&D Systems, catalog no. AF554, 2 μg/mL) or anti–PD-L1 (R&D Systems catalog no. AF1019, 2 μg/mL) or B220 (BD Bioscience, catalog no. 550286, 0.3 μg/mL) or anti-CD8 (eBiosciences catalog no. 14–0808, 2.5 μg/mL) for 5 hours, followed by 60-minute incubation with biotinylated horse anti-goat IgG (for CD4 and PD-L1; Vector Laboratories, catalog no. BA-9500) or biotinylated goat anti-rat IgG (for CD8 and B220; Vector Laboratories, catalog no. BA9400) at 1:200 dilution. The detection was performed with Streptavidin-HRP D (part of DAB Map kit, Ventana Medical Systems), followed by incubation with Tyramide Alexa 488 (Invitrogen, catalog no. B40953), and counterstaining with DAPI (Sigma Aldrich, catalog no. D9542, 5 μg/mL). The slides were scanned with a Pannoramic Flash P250 Scanner (3DHistech) using a 20×/0.8NA objective lens. The fluorescence channels were imaged using DAPI, FITC, and TRITC filters sequentially with manually adjusted exposure times. Images were then exported into .tifs using Caseviewer (3DHistech) to be analyzed.

The RNA-seq and ChIP-seq data are available at the Gene Expression Omnibus database under accession number GSE142357.

A.Y. Chang is a postdoctoral fellow at Pfizer Inc. (but was not during the time of this manuscript study). S.S. Neelapu is a consultant at Kite/Gilead, Novartis, Incyte, Legend Biotech, Merck, Celgene, Unum Therapeutics, Pfizer, Precision Biosciences, Cell Medica, Allogene, and Calibr and reports receiving commercial research grants from Kite/Gilead, Novartis, Merck, BMS, Cellectis, Poseida, Karus, Acerta, Allogene, and Unum Therapeutics. J. Baselga is an EVP, Oncology R&D, at AstraZeneca, is a member of the Board of Directors at Foghorn, Varian Medical Systems, and Bristol-Myers Squibb, is a consultant for Grail, Aura, Infinity, PMV Pharma, Juno, and Seragon, reports receiving a commercial research grant from Roche, and has ownership interest (including patents) in Apogen, Tango, and Venthera. A. Younes is a consultant at Xynomics, BMS, Merck, and Janssen. A. Dogan is a consultant at Roche, Seattle Genetics, EUSA Pharma, Takeda, and Corvus Pharmaceuticals, reports receiving a commercial research grant from Roche, and is a consultant/advisory board member for Amgen. D.A. Scheinberg is on the SAB at Eureka, Oncopep, KLUS, Sellas, Actinium, and Pfizer, reports receiving a commercial research grant from BMS, and has ownership interest (including patents) in IOVA. A.M. Melnick is a consultant for Epizyme and Constellation, reports receiving commercial research grants from Janssen and Daiichi Sankyo, and is a consultant/advisory board member for KDAC. M.R. Green is a consultant at Verastem Oncology, has ownership interest (including patents) in KDAC Therapeutics, and is a consultant/advisory board member for KDAC Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P. Mondello, S. Tadros, E. de Stanchina, J. Baselga, A.M. Melnick, M.R. Green

Development of methodology: P. Mondello, S. Tadros, H.-Y. Ying, E. Toska, A.M. Melnick, M.R. Green

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Mondello, S. Tadros, L. Fontan, A.Y. Chang, N. Jain, H. Yang, S. Singh, H.-Y. Ying, C.-S. Chu, S. Alig, M. Durant, E. de Stanchina, S. Ghosh, A. Mottok, S.S. Neelapu, O. Weigert, G. Inghirami, A. Younes, A. Dogan, R.G. Roeder, A.M. Melnick

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Mondello, S. Tadros, M. Teater, L. Fontan, A.Y. Chang, H. Yang, H.-Y. Ying, C.-S. Chu, M.C.J. Ma, S. Alig, S. Ghosh, A. Mottok, L. Nastoupil, S.S. Neelapu, O. Weigert, J. Baselga, A. Dogan, D.A. Scheinberg, R.G. Roeder, A.M. Melnick, M.R. Green

Writing, review, and/or revision of the manuscript: P. Mondello, S. Tadros, M. Teater, H.-Y. Ying, A. Mottok, L. Nastoupil, S.S. Neelapu, G. Inghirami, J. Baselga, A. Younes, C. Yee, D.A. Scheinberg, A.M. Melnick, M.R. Green

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Mondello, M.C.J. Ma, L. Nastoupil, S.S. Neelapu, A.M. Melnick, M.R. Green

Study supervision: A. Younes, D.A. Scheinberg, A.M. Melnick, M.R. Green

This work was supported by R01 CA201380 (to M.R. Green), R01 CA055349 (to D.A. Scheinberg), R01 CA178765 (to R.G. Roeder), U54 OD020355 01 (to E. de Stanchina and G. Inghirami), NCI SPORE P50 CA192937 (to A. Dogan and A. Younes), the MD Anderson Cancer Center (P30 CA016672) and Memorial Sloan Kettering Cancer Center (P30 CA008748) NCI CORE Grants, the Chemotherapy Foundation (to A.M. Melnick), the Star Cancer Consortium (to A.M. Melnick), the Jaime Erin Follicular Lymphoma Research Consortium (to A.M. Melnick, M.R. Green, and S.S. Neelapu), the Schweitzer Family Fund (to M.R. Green), and the Futcher Foundation (to L. Nastoupil and M.R. Green). H. Yang is a Fellow of the Leukemia and Lymphoma Society.