Introduction Follicular Lymphoma (FL), the second most frequent lymphoma in adults, often presents as a disseminated disease at diagnosis. Despite a generally slow progression and a median overall survival of more than 15 years with current chemo-immunotherapies, FL patients often suffer from multiple relapses. Yet, the biological mechanisms promoting FL dissemination, progression and relapse are still poorly understood. FL, like most B-cell lymphomas, originates from germinal centers (GC) where B-cells physiologically undergo clonal expansion, antibody affinity maturation, and differentiation into antibody-producing plasma cells (PC) or recirculating memory (Mem) B-cells. Recently, we provided evidence that FL B-cells are not blocked in a GC B-cell state but might adopt new dynamic modes of functional diversity (Milpied et al., Nature Immunology 2018), yet the main sources of intratumoral heterogeneity within FL remained to be identified. Methods Frozen live cell suspensions were obtained from the CeVi collection of the Institute Carnot/Calym (ANR, France). We initially applied a plate-based 5'-end single-cell RNAseq (scRNAseq) method for deep integrative single-cell analyses of transcriptome, B-cell receptor (BCR) sequence, and surface phenotype on FACS-sorted FL B-cells (4 patients, lymph node biopsies) and their non-malignant counterparts (6 adult healthy donors, spleen and tonsil samples). We confirmed our findings on additional FL samples with high-throughput droplet-based 3'-end scRNAseq (9 patients, lymph node biopsies), and 5'-end scRNAseq paired with BCRseq (5 patients, lymph node biopsies). Custom and existing bioinformatics analysis pipelines were combined for quality control and cell filtering, dimensionality reduction (PCA, t-SNE, UMAP), clustering, pseudo-time analysis, BCR sequence analysis and integrative data analysis. We further validated our transcriptomic data with FACS-based surface and intracellular protein analysis (8 patients, lymph node biopsies). Results Consistent with our previous findings, FL B-cells were transcriptionally diverse, with most cells exhibiting a patient-specific gene expression profile distinct from PC, GC and Mem cells. Challenging the mainstream view of a differentiation blockade in FL, we identified rare FL B-cells carrying a PC-like profile (including low expression of MS4A1/CD20, high expression of XBP1, MZB1, PRDM1). PC-like FL B-cells expressed high levels of the tumor clonal BCR heavy and light chain mRNA, and BCR sequence phylogenetic analysis revealed that those cells did not branch out from a specific tumor subclone. Most importantly, we found that the molecular profiles of the vast majority of FL B-cells spanned a continuum of transitional states between proliferating GC-like and quiescent Mem-like gene expression states. Principal component analysis and pseudo-time reconstruction revealed that pseudo-immune differentiation axis was consistently the main source of intra-sample transcriptional heterogeneity. On top of cell cycle related genes, GC-like FL B-cells notably expressed AICDA, BCL6, RGS13, NANS, CD81, and CD38 genes. By contrast, Mem-like FL B-cells expressed CD44, GPR183, CD69, CXCR4, CCR7, SELL, KLF2, suggesting that those cells may not be confined to the FL follicles. Flow cytometry analysis of dissociated FL tumors confirmed that only the CD38hiCD81hi subset of FL B-cells (GC-like cells), expressed Ki67 and high levels of Bcl6, whereas only CD38negCD81neg FL B-cells (Mem-like cells) consistently contained CD44+ and GPR183+ cells. Conclusions Our study suggests that FL B-cells hijack the physiological GC differentiation process to dynamically alternate between GC-like and Mem-like states that might be responsible for FL progression and dissemination, respectively. We anticipate that such FL-specific clonal dynamics may be orchestrated by extrinsic signals delivered by tumor-infiltrating T cells. Disclosures Milpied: Innate Pharma: Research Funding; Institut Roche: Research Funding.