Follicular lymphoma (FL) is a prototypical example of germinal center (GC) derived B-cell lymphoma. Using a mouse model recapitulating the sporadic occurrence of the FL hallmark BCL2/IGH translocation in healthy individuals, our previous work demonstrated that FL genesis is a dynamic process that requires multiple re-entries of BCL2+ memory B-cells into the GC to ultimately accumulate in lymphoid organs. In line with this, using single-cell gene expression analysis of human FL vs normal GC B-cells, we recently discovered that FL cells are not 'frozen' at a particular GC maturation stage but instead exhibit a major desynchronization of the GC-specific expression program (Milpied et al. Nat Immunol in press). Since KMT2D loss-of-function mutations and BCL2 translocations are the 2 main alterations in FL, we hypothesized these 2 genetic events might explain the GC program desynchronization we observe in humans. To explore the in vivo consequences of Kmt2d inactivation with Bcl2 overexpression in regulating GC/memory dynamics, we transduced bone marrow progenitors carrying B-cell-specific conditional Cd19-Cre Kmt2dflox/flox alleles with a retrovirus encoding human BCL2 or reporter alone, followed by iv transplantation into lethally irradiated WT recipients. Only double-mutant Kmt2d-/-Bcl2+ mice manifested with GC-derived lymphomas in chronically challenged animals, recapitulating histological and phenotypic features associated with human FL progression from early preneoplastic lesions to overt FL-like tumors. We used integrative single-cell analysis of surface phenotype (10-color panel), gene expression (88-gene panel by microfluidic RT-qPCR) and IGH clonality to deconvolute cellular heterogeneity of flow-sorted GC and memory B-cells after acute (day 10) or chronic T-cell dependent immunization in double-mutant vs. single Bcl2+, Kmt2d-/- or WT mice, retaining >4000 cells for downstream analysis. Populations of WT GC B-cells were molecularly heterogeneous and spanned a cyclic continuum of transitional B-cell states polarized between the light and dark zone where synchronized expression of gene modules characterized mouse GC functional identity. In acute responses, single and double-mutant mice formed GC similar to control chimera mice, and single GC B-cells from all genotypes clustered together with WT GCs, suggesting unperturbed GC transcriptional dynamics upon first antigen encounter. However, preneoplastic/tumoral Kmt2d-/-Bcl2+ mice after chronic challenge showed massive GC hyperplasia and single B-cells expressed a distinct transcriptional signature that clustered separately from WT GC or memory cells. Specifically, murine lymphoma B-cells sorted with a GC-like phenotype accumulated in transitional cell states where the synchrony of most GC-specific co-expression patterns was progressively lost whereas expression of cytokines (Lta) or surface markers linked to GC to memory transition (Gpr183, Cxcr3) became markedly expressed, indicating that Kmt2d-deficient lymphomas were not blocked at a particular GC stage. Given the importance of T-cell help for the fate 'decisions' of GC-to-memory B-cells, we further explored whether tumor cell-intrinsic factors may affect immune cell phenotypes thereby supporting the GC gene expression desynchronization. Using flow cytometry, we found that Kmt2d inactivation instructed a progressive remodeling of the tumor microenvironment with an increased recruitment of CD4+ T-follicular helper (TFH) (n=21, p<0.01). Using droplet-based single-cell RNA-seq to profile total spleens from 2 WT mice and 2 lymphomas (>9000 cells), we revealed a TFH cluster with an activation signature distinct from normal TFH and a concomitant expansion of exhausted CD8+ T-cells strongly co-expressing inhibitory receptors (Lag3, Tim3, Pdcd1), thereby indicating that Kmt2d inactivation in B-cells may favor lymphoma formation by shaping the FL tumor supportive niche and contributing to immune evasion. In summary, our integrative single-cell analyses in murine lymphomas revealed that Kmt2d mutations in FL not only instruct B-cell intrinsic effects involving the desynchronization of the normal GC expression program, but also trigger a concomitant re-education of a tumor-supportive immune microenvironment, establishing for the first time a key link between the most frequent epigenetic alteration and the FL microenvironment. Disclosures Salles: F. Hoffmann-La Roche Ltd: Consultancy, Honoraria, Research Funding; Celgene: Honoraria, Other: Advisory Board, Research Funding; Janssen: Honoraria, Other: Advisory Board; Novartis: Consultancy, Honoraria; Takeda: Honoraria; Servier: Honoraria, Other: Advisory Board; Epizyme: Honoraria; Merck: Honoraria; BMS: Honoraria, Other: Advisory Board; Morphosys: Honoraria; Acerta: Honoraria; Amgen: Honoraria; Gilead: Honoraria, Other: Advisory Board; Pfizer: Honoraria; Servier: Honoraria; Abbvie: Honoraria.

Background: Follicular lymphoma (FL) is an incurable B-cell malignancy that constitutes a quarter of all lymphomas. Although RCHOP immuno-chemotherapy induces high rates of complete remission, almost all FL patients experience multiple relapses post-treatment. The limited understanding of treatment response heterogeneity is due to the absence of in vitro or in vivo experimental models, primarily because tumor cells heavily rely on their microenvironment to survive. In this study, we present an innovative xenograft model of primary FL cells in avian embryos, circumventing these limitations. Methods: We developed the FL-AVI-PDX model by transplanting 20 biopsy FL samples, including good (n=11) and poor clinical responders (POD24, n=9), into chicken embryos. Each set of embryos was treated with RCHOP or vehicle intravenously. We evaluated the effect of immuno-chemotherapy on tumor volume by light sheet microscopy and on tumor biology by transcriptomic analysis at the single-cell level. Results: We successfully engrafted all samples in avian embryos. We found that RCHOP treatment in ovo led to tumor volume reduction, which predicted progression-free survival in multivariate analysis, demonstrating the model's capacity to capture clinical heterogeneity at the patient level. The FL-AVI-PDX model also provided a unique opportunity to analyze the transcriptomic impact of RCHOP on FL cells using single-cell RNA sequencing. We identified a signature of 21 genes upregulated after RCHOP exposure, displaying significant intra-tumoral heterogeneity. As a proof of concept, we validated the functional involvement of BAX, a gene from the RCHOP-induced signature, as a critical effector of immuno-chemotherapy in vitro and in avian embryos. Conclusions: The FL-AVI-PDX model is a platform for functional precision oncology in primary FL cells that captures both inter- and intra-patient heterogeneity of clinical response to a complex therapeutic regimen. It offers a unique opportunity to better understand FL biology, opening perspectives for the development of new drugs.

No abstract

Introduction: The risk of developing a follicular lymphoma (FL) begins to be better understood with the identification in large epidemiological studies of familial predisposition, some occupational exposures and genetic factors. Genome wide-association studies (GWAS) identified constitutional single nucleotide polymorphisms (SNPs) at risk of FL in HLA region (rs12195582), in 11q23.3 (near CXCR5), in 11q24.3 (near ETS1), in 3q28 (near LPP), in 18q21.33 (near BCL2), and 8q24 (near PVT1); three suggestive loci are localized at 17q25.3 (near CYBC1), 3q13.33 (CD86), 18q12.3 (SLC14A2) (Skibola, Am J Hum Genet. 2014). High t(14;18) frequency in blood years before diagnosis from healthy individuals was also defined as a predictive biomarker for FL (Roulland, J Clin Oncol 2014). It is currently unknown whether any relationship exists between inherited genetic variants associated with FL susceptibility and t(14;18) frequency and if the combination of the two biomarkers could be useful for a better stratification of risk of FL development in healthy individuals. Methods: We used quantitative PCR assays to estimate t(14;18) frequency in prediagnostic blood samples from 105 individuals that were obtained on average 6.4 years before FL diagnosis (pre-FL group) together with 236 age and gender-matched individuals (control group) that were issued from the participants in the EPIC cohort ( European Prospective Investigation Into Cancer and Nutrition). Constitutional DNA was analyzed for the genotyping of the nine SNPs associated with FL risk (HLA, rs12195582; CXCR5, rs4938573; ETS1, rs4937362; LPP, rs6444305; BCL2, rs17749561; PVT1, rs13254990; CYBC1, rs3751913; CD86, rs2681416; SLC14A2, rs11082438). Genotyping were performed in duplicate using TaqMan® assays on Fluidigm platform. The nine SNPs were analyzed individually and combined in a polygenic risk score (PRS). PRS is a weighted average of the number of risk alleles with the weights being the log of the odds-ratio (OR) reported in the FL GWAS (Skibola, Am J Hum Genet. 2014). A model for FL risk was developed using multivariable logistic regression. Predictive ability was assessed by area under Receiver Operating Characteristic (ROC) curve, with 10-fold cross-validation. This work is supported by the French NCI (INCA, PRT-K16-167). Results: t(14;18) frequency as a log-transformed continuous variable is predictive of FL risk (OR: 1.50; 95%CI: 1.29-1.78, P<0.001); PRS is also strongly associated with FL risk (OR: 3.31; 95%CI: 2.01-5.62, P<0.001). Age at screening (OR: 0.98; 95%CI: 0.95-1.01, P=0.24) did not influence FL risk. A weak but statistically significant correlation between t(14; 18) frequency and PRS was observed (Pearson correlation=0.18, P=0.002). In multivariable analysis, both PRS (OR: 2.84, 95%CI: 1.66-4.99, P<0.001) and t(14;18) frequency (OR: 1.45, 95%CI: 1.18-1.84, P<0.001) remained statistically significant. No departure from log-linear effect was observed in the modeling nor statistical interaction between PRS and log-transformed t(14;18), confirming that t(14; 18) frequency and genetic markers (PRS) were two independent factors of FL risk. Sensitivity analyses showed that these results were not influenced by the delay between the date of the screening and the FL diagnosis. Combining t(14;18) frequency and the PRS in a prediction model allowed the identification of some individuals at very high risk of developing FL and a shiny web application was developed to provide an easy-to-use tool to obtain individualized risk predictions for new patients (Figure). Area under ROC curve (AUC) showed that the model that integrated t(14;18) frequency and PRS (AUC: 0.69, 95%CI: 0.62-0.76) had a better prediction of FL risk than t(14;18) frequency (AUC: 0.62, 95%CI: 0.55-0.70) and PRS alone (AUC: 0.68, 95%CI: 0.61-0.75). Conclusions: Genetic variants combined in PRS and t(14;18) frequency allowed the identification of individuals at high-risk of FL development and provided a better way, when these two biomarkers were combined, to discriminate healthy individuals from pre-FL cases many years before malignant transformation. These findings could be used for screening test of populations with some environmental exposures positively associated with FL development in epidemiological studies and may contribute in the future to monitoring or early intervention. Figure Disclosures Salles: Roche, Janssen, Gilead, Celgene: Consultancy, Honoraria, Membership on an entity's Board of Directors or advisory committees, Other: Educational events; Autolus: Consultancy, Membership on an entity's Board of Directors or advisory committees; BMS: Honoraria; Takeda: Consultancy, Honoraria, Membership on an entity's Board of Directors or advisory committees, Other: Educational events; Amgen: Honoraria, Other: Educational events; Novartis, Servier, AbbVie, Karyopharm, Kite, MorphoSys: Consultancy, Honoraria, Membership on an entity's Board of Directors or advisory committees, Other: Educational events; Merck: Consultancy, Honoraria, Membership on an entity's Board of Directors or advisory committees; Epizyme: Consultancy, Honoraria.