Pan-HDAC Inhibitors Restore PRDM1 Response to IL21 in CREBBP-Mutated Follicular Lymphoma

Adaptive immune responses involve the formation of germinal centers (GC), which are specialized structures allowing the differentiation of high-affinity B cells into long-lived memory B cells and plasma cells (PC). PCs are generated through specific transcriptional programs influenced by numerous signals delivered by the microenvironment including IL21. This latter is a potent inducer of PCs and is produced by GC CD4⁺ helper T cells called follicular helper T cells (Tfh; ref. 1). Akin to most cytokines, IL21 activates mainly JAK3/STAT3 signaling pathways (2), which are known to play a highly selective role in PC differentiation (3) and apoptosis, depending on the context (4). In humans, blockade of IL21 inhibits PC generation. Tfh interacting with IL21 receptor expressing GC B cells could thus impact B-cell destiny via the delivery of IL21 signal (3, 5). Human PCs emerging through IL21/pSTAT3 signaling enhancement has concomitantly upregulated PRDM1/BLIMP1 and downregulated BCL6 gene expression (6, 7). PRDM1/BLIMP1 is considered as the master PC factor that antagonizes BCL6 which sustains the B-cell identity (8, 9). Therefore, PRDM1/BLIMP1 and BCL6 mutually inhibit each other and BCL6-related inhibition of PC generation is predominantly related to BCL6 binding at intron 3 of PRDM1 where BCL6 recruits MTA3, which acts as a corepressor (10). Moreover, this PRDM1 intronic region contains an enhancer bound by CREBBP (11). CREBBP acts as a transcriptional coactivator of many different transcription factors through its intrinsic acetylation function on histone but also on nonhistone proteins including BCL6. Indeed, CREBBP binds and acetylates BCL6, leading to the inactivation of its transcriptional repressor function (12, 13).

In follicular lymphoma, we demonstrated earlier that the tumor microenvironment is enriched for Tfhs that sustain neoplastic cells (14, 15). It is generally accepted in the field that follicular lymphomas are tumors reflective of centrocytes that fail to differentiate beyond the GC exit point. Purified follicular lymphoma B cells compared with normal GC B cells does not identify a radically opposed signature, but rather substantial modification of normal GC expression driving most likely by numerous genetic and epigenetic somatic modifications (14, 16–18). Both histone acetyl transferases, CREBBP and EP300, are commonly mutated in follicular lymphoma and CREBBP loss-of-function affects preferentially H3K27 acetylation whose depletion leads to downregulation of genes involved in GC output (19). Interestingly, the expression of these genes can be restored after inhibition of HDAC3 (20).

In this study, we assessed PRDM1 gene expression and regulation to achieve new insights on the differentiation blockade that characterizes the follicular lymphoma. We showed that despite a functional capacity to activate the IL21/pSTAT3 signaling, nonfunctional CREBBP follicular lymphoma cells were unable to increase PRDM1 expression. In these follicular lymphomas, induction of PRDM1 in response to IL21 could be restored through the use of pan-HDAC inhibitors.

Subjects were recruited under Institutional Review Board approval and informed consent process (French Minister authorization DC-2016-2565). An informed written consent was obtained from each subject or subject's guardian. The study was conducted in accordance of the Declaration of Helsinki. Normal GC-derived B and T cells were isolated from human tonsils and reactive lymph nodes. Follicular lymphoma tumors were obtained from patients that underwent a surgical biopsy during a diagnosis procedure and from patients with refractory/relapsed follicular lymphoma recruited during the phase I/II study based on an oral pan-HDACi drug (21).

For transcriptomic analysis, we used highly purified B-lymphocyte fractions that were sorted using combinations of mAbs and FACSARIA (BD Biosciences) system. Extracted RNAs were hybridized on an Affymetrix Human Genome U133 Plus 2.0 Array and strategy for raw data normalization and filtering is detailed in Supplementary Materials and Methods.

The relative quantification of gene expression was determined using the 2^−ΔΔC_t method, then normalized to at least an internal control gene (ABL1, GAPDH, and/or HPRT1) and relative to a calibrator control sample corresponding to a mix of cDNA of peripheral blood mononuclear cells (PBMC) from several healthy donors. Statistical analysis using GraphPad Prism Software used Mann–Whitney test.

All antibodies and technical details concerning protein expression and FISH analyses are in Supplementary Materials and Methods, Supplementary Table S4, and Supplementary Figs. S3, S4, and S8.

Cells were cultured for 2 hours before receiving indicated treatments for 24 hours, and then CD19/CD20 viable B cells were collected and used for subsequent DNA and RNA extraction. Detailed procedures are given in Supplementary Materials and Methods.

We performed the SureSelect targeted-capture strategy from Agilent Technologies using a panel of 34 genes described previously (22) with subsequent paired-end sequencing on an Illumina HiSeq 1500 Platform (Illumina). For variants detection, SureCall software from Agilent technologies was used (see Supplementary Materials and Methods, Supplementary Tables S2B and S3B).

Statistical analyses were performed with the GraphPad Prism software V5 (GraphPad Software) as indicated in Supplementary Materials and Methods section.

B-cell differentiation process is supported by the microenvironment through the delivery of soluble and membrane signals (23, 24). Among them, CD40L and IL21 are major contributors to the transcriptional emergence of PC identity genes in committed B cells (3, 7, 25). To decipher specific CD40L- and IL21-targeted genes, freshly purified centroblasts (CB) were cultured for 3 hours in presence or not of CD40L and IL21 before transcriptome analysis were performed. We found a total of 4,932 genes differentially expressed with 2,154 up- and 2,778 downregulated genes between unstimulated and stimulated conditions (FDR < 0.05; Fig. 1A). In particular, PRDM1 expression is increased 3.5 times upon CD40L and IL21 stimulation (P = 0.001; Fig. 1B). In parallel, using the same approach, we analyzed highly purified human centroblastic and centrocytic GC-derived B cells from tonsils and highly purified follicular lymphoma B cells. At first, a total of 4,654 genes (FDR < 0.05) with 2,347 up- and 2,307 downregulated genes were found differentially expressed between centroblasts (CB) and centrocytes (CC; Fig. 1C). We display list comparisons using Venn diagrams to find which subsets of genes are involved in both CB/CC transition and CB response to IL21/CD40L signaling. A total of 804 up- and 965 downregulated genes (total of 1,769 genes) were common to both lists (Fig. 1C) and major hallmark pathways associated with these genes were related to immune responses and cell proliferation, respectively (Supplementary Table S1). We next applied this 1,769-gene signature to compare CB, CC, and follicular lymphoma B-cell transcriptomes. Unsupervised hierarchical analysis clustered follicular lymphoma B cells separately from other cells with a particular and an opposite pattern of expression compared with CBs (Fig. 1D). We delimited two boxes of probe sets based on the clustering and found probe sets that positively expressed in Box1 were vastly (>95%) connected to CBs, whereas the Box2-positive probe sets were associated to follicular lymphoma B cells (>82%). In addition, Box2 contained 95.8% of the 804 genes upregulated in the 1,769-gene signature. Interestingly, in this analysis, the CCs presented somewhere an intermediate position, suggesting a polarized axis where follicular lymphoma cells occupy the most advanced position in the B-cell differentiation (Fig. 1E). To complete our characterization, we use Ingenuity Pathways Analysis (IPA, Ingenuity Systems, www.ingenuity.com) and focused on previously described molecular pathways enriched in B cells and PCs (26). Globally, we confirmed that Box1 genes are related to CB signatures (cell cycle and FOXM1 transcription factor network), whereas Box2 presented plasmablast and early PC features (Fig. 1F). In this context, we noticed with interest the strong repression of PRDM1 in follicular lymphomas (highlighted in Fig. 1D). This finding was confirmed using an independent cohort of 23 follicular lymphomas previously explored and where total CD19⁺ B cells were compared with B cells issued from reactive lymph nodes (Fig. 1G; ref. 14). IHC for BLIMP1 protein showed a totally negative staining in follicular lymphoma, whereas normal GCs showed very weak expression and tumor plasmocytoma, a strong positivity (Supplementary Fig. S1C). Our transcriptome findings in follicular lymphoma are in contradiction with the transcriptional network of normal B cells (27) and also with our previous data (28) on the hierarchical clustering, showing a transcriptional switch between CB and CC subtypes. Altogether, our data suggest the existence of a deregulation of the transcriptional BCL6/PRDM1 balance in follicular lymphomas. Indeed, follicular lymphomas maintain the expression of B-cell identity genes including BACH2 and BCL6 (PAX5 is missing in our 1,769-gene signature) besides low PRDM1 expression, whereas, on the other hand, these cells express IRF4 and XBP1, both tightly connected to the PC identity. By IPA Ingenuity Systems, we found that the Box2 was significantly enriched for the CD40L/CD40 pathway (Supplementary Fig. S1A). Gene set enrichment assays (GSEA) applied on the 1,769-gene signature identified a significant enrichment in follicular lymphomas compared with CB for CD40L, IL21, and plasma cell upregulated genes signatures (Supplementary Fig. S1B). Collectively, our results indicate that follicular lymphoma B cells are most likely blocked at a terminal B-cell differentiation step characterized by a low expression of PRDM1, despite the presence of substantial CD40L and IL21 microenvironment signaling.

During the differentiation of CCs into PCs, the upregulation of PRDM1 through the IL21 signaling was previously demonstrated in normal and lymphoma B cells (4, 29). On the basis of these data and for reasons of sample saving to complete our study on primary follicular lymphoma cells, we decided to focus our investigations solely on IL21 signaling. Highly purified follicular lymphoma B cells of 18 patients (clinical details are in Supplementary Table S2A) were cultured for 24 hours in the presence or not of IL21 and cell viability was systematically monitored (Supplementary Fig. S3). As control, we used L3055 centroblastic cell line (named hereafter, control L3055), an EBV-negative Burkitt lymphoma cell line, phenotypically and functionally similar to the normal CBs (30) carrying besides the rearranged MYC locus no additional alterations affecting BCL6 and PRDM1 loci, as shown by CGH array (Supplementary Fig. S2). Control L3055 significantly increased the PRDM1 expression upon IL21 without effect on BCL6 (Fig. 2A and B). Protein BLIMP1/PRDM1 was detectable by flow cytometry for control tonsil–derived GC B cells and by Western blot analysis for L3055 cells after 24 hours of IL21 stimulation (Supplementary Fig. S2). In contrast, the 18 follicular lymphomas did not significantly modify their expression of PRDM1 and BCL6 (Fig. 2A and B; Supplementary Fig. S3C and S3D). Spearman analysis to find a correlation between the expression of PRDM1 and BCL6, showed that a large majority of follicular lymphomas did not respond for both genes after IL21 stimulation (Fig. 2C). These results are in line with our transcriptome analysis findings and confirm that most follicular lymphoma B cells exhibit a defect in PRDM1 response.

We used total cell suspensions from follicular lymphoma tumors to evaluate the functional capacities of B cells and Tfhs. Control cell counterparts came from nonmalignant tonsils or reactive lymph nodes (rLN). The gating strategy of the flow cytometry analysis defined a specific CD3/CD4/CxCR5/PD-1–positive cell population corresponding to Tfhs (Supplementary Fig. S4). As described previously, Tfhs increased in number in follicular lymphomas compared with rLNs (Fig. 3A; refs. 14, 31). Functional experiments on Tfhs showed an enhanced production of IL21 after stimulation by PMA–ionomycin (Fig. 3B; ref. 32) as well as a significant increase of the pSTAT3 response after 10 minutes of IL21 (Fig. 3C). Follicular lymphoma B cells from 20 patients showed after IL21 stimulation a significant increase in pSTAT3 compared with controls. Interestingly, this tonic response was detected in malignant and nonmalignant B cells, both discriminated with BCL2 intracellular staining (Fig. 3D; Supplementary Fig. S4). In addition, both B-cell populations for the same follicular lymphoma proportionally increased their pSTAT3 expression (Fig. 3E). Moreover, 24 hours of IL21 stimulation led to the expression of BATF, a specific pSTAT3-target (ref. 33; Fig. 3F). We completed our analysis by evaluating the expression of pSTAT3 by IHC in 7 follicular lymphomas and compared the results with 2 rLNs. In normal GC, B cells presented pSTAT3 staining mainly in medium and large B cells, localized within the light zone of the GC in the vicinity of PD1-positive T cells (Fig. 3G, left). In follicular lymphoma, this pattern of expression was lost and pSTAT3-positive B cells were more numerous and formed clusters (Fig. 3G, right). Overall, we found a mean number of 40 pSTAT3-positive cells per GC in follicular lymphoma (ranged from 29 to 51) compared with a mean of 19 in rLNs. Unlike rLNs, some follicular lymphoma Tfhs were positive for pSTAT3 in agreement with our above flow cytometry data (Fig. 3G, right). Altogether, our data showed that follicular lymphomas present an enhanced and functional IL21/pSTAT3 signaling.

Genetic alterations of CREBBP may alter its acetyltransferase activity, which may abolish BCL6 acetylation and maintain its repressor activity on PRDM1 (10, 12, 13). We next sought for somatic mutations in CREBBP-coding region included in our 34-gene lymphopanel beside other chromatin-modifier genes (EP300, EZH2, KMT2D, and MEF2B; ref. 34) used for capture-targeted deep-sequencing strategy. Fourteen follicular lymphomas of 20 were mutated for CREBBP at least once (70%), including the 10 nonsynonymous variants (SNV) located in the HAT catalytic domain (histone acetyltransferase domain), 2 SNVs outside the HAT, and 2 frameshifts (fs) in the N-terminal region. For 12 of these variants (HAT-located and fs), a nonfunctional CREBBP protein is encoded (Fig. 4A; Supplementary Table S2C; ref. 12). EP300 (another gene with a HAT domain) was mutated in two cases but outside HAT domain (Supplementary Table S2C). Altogether, 12 of 20 follicular lymphomas presented a functional loss of CREBBP owing to genetic alterations.

To determine the binding capacity of BCL6 protein to intron 3 (INT3) of PRDM1, we used a chromatin immunoprecipitation (ChIP) approach followed by qPCR (see procedure details in Supplementary Fig. S5). Highly purified follicular lymphoma B cells and control L3055 were cultured for 24 hours in the presence or absence of IL21. BCL6 binding to PRDM1 was decreased in L3055 upon IL21, and correlates with its increased expression due to a probable releasing BCL6-mediated transcriptional repression (Fig. 4B). We then explored 10 follicular lymphoma of 18 previously explored for PRDM1 expression (Fig. 2A), for which we had sufficient viable B cells for ChIP assays. For two (FL_6108 and FL_5511) out of four wild-type follicular lymphomas for CREBBP, the IL21-induced PRDM1 expression was associated with a decrease of BCL6 enrichment at INT3 of PRDM1, thus behaving like control L3055. Six other follicular lymphomas, characterized by a positive BCL6 protein IHC staining and a loss-of-function variant of CREBBP, we observed a strong enrichment of the BCL6 binding, except for FL_5008. This finding correlated with the absence of IL21 induction of PRDM1 gene expression and data indicate that in response to IL21 stimulation, the tumor cells operate an active silencing of the PRDM1 gene, possibly facilitating local recruitment of the BCL6 repressor (Fig. 4B). Globally, our data suggest that a link exists between BCL6 binding, PRDM1 response to IL21, and the functional activity of CREBBP. Therefore, we decided to explore further the binding of BCL6 protein to INT3 of PRDM1 in response to IL21 with the use of a histone acetyltransferase inhibitor (HDACi).

We postulated that CREBBP-mutated follicular lymphomas might present a diminished acetylated form of BCL6 protein conferring a potential oncogenic activity (12, 13). Thus, we hypothesized that by restoring acetylation state of histones and proteins in follicular lymphoma B cells using an HDAC inhibitor, we would be able to restore PRDM1 response to IL21. We used vorinostat, also known as suberanilohydroxamic acid (SAHA), a potent pan-HDACi, on 6 CREBBP-mutated follicular lymphomas and on control L3055. Purified cells were analyzed after 24 hours of culture using three different conditions: IL21 alone, IL21 plus vorinostat, and vorinostat alone. None of these conditions triggered cell death (Supplementary Fig. S7). In control L3055, addition of SAHA to IL21 induced a significant increase of BLIMP1/PRDM1 mRNA and protein expression without modification of BCL6 binding to PRDM1 gene (Fig. 5; Supplementary Fig. S6). For 4 of 6 follicular lymphomas, PRDM1 expression increased under IL21 plus vorinostat compared with IL21 alone and was associated with a BCL6 expression decrease (Fig. 5A and B) leading overall to a significant positive effect on the PRDM1/BCL6 expression balance as shown for the control L3055 (P = 0.041; Fig. 5D). In addition, we detected for all 6 follicular lymphomas a clear BCL6 occupancy decrease to the INT3 of PRDM1 under IL21 plus vorinostat (Fig. 5C). The vorinostat-only condition compared with IL21 alone did not significantly increase the PRDM1/BCL6 ratio (P = 0.93; Fig. 5D), whereas a decrease of BCL6 binding occurred in five of six cases (Fig. 5C and D). In conclusion, our observations could be summarized schematically using a scale where SAHA addition to IL21 will reverse the abnormal BCL6/PRDM1 transcriptional gene expression equilibrium in follicular lymphomas by increasing PRDM1 while decreasing BCL6 expression (Fig. 5E).

To determine whether HDACi therapy in follicular lymphoma may affect the expression of PC master genes, we analyzed 4 of 7 patients with follicular lymphoma included in a multicenter phase I/II study testing a new pan-HDACi in refractory/relapsed B-cell lymphoproliferative disease (21). Somatic mutation screening was performed on tumor DNA at diagnosis and when patients reprogressed (Supplementary Table S3). FL_CAN patient still in complete remission 5 years after the inclusion (the drug discontinued after 4 years) was CREBBP wild-type at diagnosis, but mutated in the HAT domain of EP300, which encoded a protein that, like CREBBP, sustains acetylation-mediated inactivation of BCL6 repressive activity (ref. 12; Fig. 6A). For patients FL_CON and FL_DAD, we detected nonsynonymous variants within, respectively, the HAT domain of EP300 or CREBBP before and after treatment (Fig. 6A). These two patients, one with 7 months of stable disease (FL_DAD) and the other without response to the drug (FL_CON), presented a clonal evolution in their rebiopsy (Fig. 6A), probably leading to the therapeutic resistance. FL_CAS patient who had an objective 28-month response to the drug before relapse was not mutated for CREBBP or EP300. However, he showed also a clonal evolution upon treatment and notably the loss of the CD79B mutation, which plays a role in antigenic stimulation in chronic active BCR signaling (Fig. 6A).

Our investigations on tumor RNA extracts showed globally an increase in the expression of the 3 tested PC transcription factors, PRDM1, IRF4, and XBP1 in rebiopsies compared with the initial tissues, whereas B-cell identity factors remain stable except for BCL6 and BACH2 in patient FL_CON only (Fig. 6B). The expression of PRDM1 in FL_CON and FL_CAS increased during treatment, whereas FL_DAD showed stable expression but had a higher level of PRDM1 expression at the time of inclusion compared with other patients. In addition, FL_DAD was the only one which showed a marked increase in the expression of IRF4. The 3 patients for whom the treatment failed showed a clear increase of XBP1 expression. Interestingly, for patients with FL_DAD and FL_CON, we observed an elevation of the expression of the active form of XBP1 and XBP1s, associated with a clear upregulation of the unfolding protein response (UPR) sensor ERN1, a gene encoding the endoplasmic reticulum kinase involved in XBP1 maturation (Fig. 6C; ref. 35). These mechanisms are essential for PC differentiation (36). Overall, increased expressions of key PC-identity genes under pan-HDACi treatment suggest that follicular lymphomas underwent some differentiation, but that tumor cells still escape the drug (Fig. 5D).

In 2006, two different groups (37, 38) described the presence of PRDM1/BLIMP1 inactivation in 50% of non-GC DLBCL owing to alterations on both alleles by either deletion or mutations. This loss of function is critical for the lymphomagenesis process giving to BLIMP1 the role of a bona fide tumor suppressor. This inactivation is mutually exclusive with BCL6 alterations (39). However, they also identified few non-GC DLBCLs without PRDM1 genetic alterations but with a lack of BLIMP1 protein, suggesting that other mechanisms could be responsible for PRDM1/BLIMP1 inactivation. In our study, we identified in follicular lymphoma B cells compared with normal counterparts a downregulation of PRDM1 expression, whereas our screening for somatic mutations rules out the existence of PRDM1 genetic alterations. Our transcriptome analysis of normal GC B cells subjected to IL21/CD40L stimuli identified a specific signature representative of the Tfh-delivery signal promoting terminal B-cell differentiation, that is, upregulation of PC identity genes including PRDM1 associated to downregulation of B-cell identity genes (6, 7). In comparison, the IL21/CD40L signature in follicular lymphomas transcriptome is reversed with upregulation of some PC identity genes (e.g., IRF4 and XBP1) and downregulation of PRDM1, whose gene expression is mandatory for PC maturation and maintenance (40, 41). Therefore, our observations that follicular lymphoma cells are low PRDM1 expressers suggest a blockade of follicular lymphoma cells maturation that might be due to the impossibility of tumor cells to express PRDM1/BLIMP1.

Studies of human and mouse models revealed that IL21 plays a crucial role in the development of B-cell immunoglobulin responses through the induction of PRDM1 expression (42, 43). Our functional experiments in control L3055 found a positive IL21-mediated PRDM1 response while follicular lymphoma cells showed low baseline expression of PRDM1 and a lack of induction with IL21. This finding suggested that follicular lymphoma cells take advantage of the GC's “fertile soil” to develop and initiate the terminal differentiation; however, the process is stopped in a step of differentiation characterized by their inability to positively regulate the expression of PRDM1 in absence of any genetic alteration of this gene (44). In chronic lymphocytic leukemia, Duckworth and colleagues described a transcriptional repression of PRDM1 upon IL21 owing to the loss of chromatin active marks (45). In some aggressive lymphomas, silencing of PRDM1 was related to DNA hypermethylation in regulatory regions of the gene (46). In our study, we investigated the proximal promoter and intron 3 of the gene and found very low level of methylation, ruling out this mechanism (data not shown).

The IL21 signaling goes mostly through a potent induction of the STAT3 pathway involving the binding to a pSTAT3-IRF4 consensus site of PRDM1 leading to the gene upregulation (2, 29). Herein, we confirmed previous descriptions in follicular lymphoma with the presence of a Tfh-enriched microenvironment (14), an enhanced capacity of Tfhs to secrete IL21 (15), and the increased proportion of both follicular lymphoma B cells and Tfhs to produce pSTAT3 in response to IL21 compared with their normal counterparts. These findings may reflect the existence of a chronically active stimulation in follicular lymphoma microenvironment leading to a prompt pSTAT3 functional response. Indeed, we found an enhanced expression of BATF, a target gene of pSTAT3, after IL21 stimulation and the presence of pSTAT3-positive cell clusters in tumor tissues.

Both, BCL6 and BLIMP1 are known to negatively cross-regulate each other. Therefore, during B-cell differentiation, crossing the restriction point, thanks to a reverse transcriptional balance (i.e., PRDM1 goes up while BCL6 decreases), will be crucial to complete the final cell commitment (6, 47). The role played by IL21 in this control has been little explored mainly because of the absence of reliable B-cell lines. Our cell culture experiments on control L3055 showed that the IL21-mediated PRDM1 upregulation was associated with a decrease of BCL6 occupancy on PRDM1. For primary cultured follicular lymphoma B cells, only two cases presented similar results whereas the other eight showed a decrease in their PRDM1 expression by IL21 and, in parallel, they increased the binding of BCL6 protein to PRDM1. Among them, 6 follicular lymphomas had a functional loss of CREBBP, which has an impact on the BCL6 acetylation status. Indeed, in normal cells, CREBBP binds and acetylates BCL6, which disrupts its ability to recruit histone deacetylases, thereby enhancing its capacity to repress transcription of target genes, such as PRDM1 (12, 13). We, therefore, speculate that HDACi treatments could lead to accumulation of inactive acetylated BCL6, cell-cycle arrest, and apoptosis in B-cell lymphoma cells (13). Our experiments with vorinostat on nonfunctional CREBBP follicular lymphoma cells showed globally an increase in PRDM1 expression after IL21 exposure. This gain of PRDM1 response was associated to a decrease of BCL6 protein occupancy on INT3 of PRDM1 and a decrease in BCL6 gene expression leading indirectly to an enhanced PRDM1 functional response through the PRDM1/BCL6 balance, which drives the B-cell terminal differentiation (26). It is of interest to notice that wild-type follicular lymphoma cases for CREBBP maintained a positive PRDM1 response to IL21, whereas 3 of 4 presented a deregulated BCL6 gene (3q27-positive) with limited modifications of BCL6 binding to PRDM1 (Fig. 4). The fourth case wild-type for CREBBP and without 3q27 abnormality responded to IL21 like the control L3055, that is, a positive PRDM1 response to IL21 accompanied with a decrease of BCL6 enrichment at INT3 of PRDM1. Taken altogether these findings, despite the lack of protein verification on follicular lymphoma cells due to material scarcity, we speculate that the decrease of BCL6 binding to PRDM1 in response to IL21, in nonfunctional CREBBP follicular lymphomas treated by vorinostat, is likely not due to the decrease of BCL6 gene expression. We suspect that additional modifications on the BCL6 protein complex could have been induced by the vorinostat allowing the decrease of BCL6 enrichment at INT3 and thereby suppressing the repression on PRDM1.

The loss of PRDM1 contributes to the lymphomagenesis by blocking PC differentiation (39), knowing, however, that in follicular lymphomas, the loss of CREBBP is not sufficient to drive lymphomagenesis, but need the cooccurrence of BCL2 translocation (11). Our analysis of 4 massively pretreated patients with follicular lymphoma with a new pan-HDACi showed that the drug might increase PC identity genes expression within the tumor, including PRDM1, in agreement with our in vitro data. The restoration of PC identity genes with a pan-HDACi is in line with the recent report of Jiang and colleagues who described broad effects of CREBBP on the transcriptional regulation of B cells (19) and here, in our study, specifically on a master gene of the B-cell differentiation, that is, PRDM1. Pan-HDAC inhibitors may allow a new step in follicular lymphoma cell differentiation with varying efficacy depending on the presence of somatic abnormalities and clonal tumor diversity. Rebiopsied tissues presented an increase of the BCL6 expression, which could be due to the reprogression status of the disease connected with the proliferation properties of BCL6 rather than its induction by the drug. Despite this effect, the drug allowed a positive increase in the PRDM1/BCL6 balance. Among the identified upregulated PC identity genes, we found XBP1 gene, which occupies a downstream position in the transcriptional cascade that governs B-cell differentiation (26). Indeed, PRDM1/BLIMP-1 is known to regulate UPR components like ATF6 and ERN1 that are required for full-length XBP1 expression and the production of subsequent active spliced-form XBP1s (48, 49). Two of 3 patients had increased XBP1s and ERN1 expressions with the pan-HDACi drug supporting the idea of a successful BLIMP1/PRDM1 protein restoration in these cells. Recently, Bujisic and colleagues described that impaired ERN1 expression and XBP1 activation contribute to tumor growth in GC-type diffuse large B-cell lymphoma (50).

HDACi drugs seem to be promising medications in largely pretreated follicular lymphomas as confirmed recently by Even and colleagues, which showed a significant clinical activity in this disease, including highly durable responses (51). In this context, our study shows that monitoring effects of such drug by analyzing specific gene expressions, for example, PRDM1, IRF4, XBP1, and ERN1, thanks to iterative fine-needle punctures to easily accessible tumors, might be of interest. Furthermore, additional studies will also be needed to confirm that pan-HDAC inhibitors restore both gene and protein expression for PRDM1/Blimp1 and other key players in PC differentiation, thus providing the ultimate proof of the effect of these drugs in follicular lymphoma. Collectively, our data uncover a new mechanism by which pan-HDAC inhibitors may act positively to treat patients with follicular lymphoma and in particular, those with nonfunctional CREBBP.

V. Ribrag reports receiving commercial research grants from ArgenX and Epizyme and speakers bureau honoraria from Servier, and is a consultant/advisory board member for Bristol-Myers Squibb, Epizyme, Gilead, Infinity, MSD, Nanostring, Pharmamar, and Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: F. Desmots, T. Fest

Development of methodology: F. Desmots, E. Guiheneuf, T. Fest

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Roussel, C. Pangault, C. Pastoret, E. Guiheneuf, J. Le Priol, G. Caron, C. Henry, M.-A. Belaud-Rotureau, P. Godmer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Desmots, M. Roussel, C. Pangault, C. Pastoret, E. Guiheneuf, M.-A. Belaud-Rotureau, V. Ribrag, T. Fest

Writing, review, and/or revision of the manuscript: F. Desmots, C. Pastoret, E. Guiheneuf, T. Lamy, F. Jardin, K. Tarte, V. Ribrag, T. Fest

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Desmots, J. Le Priol, V. Camara-Clayette, T. Lamy, K. Tarte

Study supervision: F. Desmots, T. Fest

Others (histologic interpretation): F. Llamas-Gutierrez

Others (assisted with experiments): V. Camara-Clayette

This work was supported by an internal grant from the Hematology Laboratory, CHU de Rennes, France, and the Ligue Régionale contre le Cancer de l'Ouest. The NGS experiments are part of the RELYSE project supported by an NCI translational grant. Sequencing was performed in the Biogenouest Genomics/Human and Environmental Genomics core facility of Rennes (Biosit/OSUR). Human samples were obtained from the processing of biological samples through the Centre de Ressources Biologiques (CRB)-Santé of Rennes (BB-0033-00056, http://www.crbsante-rennes.com). Cell sorting was performed by flow cytometry facility, Biosit, University of Rennes 1 (Rennes, France). Part of this work was supported by the Carte d'identité des Tumeurs (CIT) program (http://cit.ligue-cancer.net/index.php/en) from the Ligue Nationale Contre le Cancer. The research protocol was conducted under French legal guidelines and fulfilled the requirements of the local institutional ethics committee.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.