Diffuse large B cell lymphoma (DLBCL) is the most common form of non-Hodgkin’s lymphoma. Many efforts have characterised DLBCL subtypes that account for some of the heterogeneity in clinical outcome following standard therapy. However, not much is known regarding intratumoural heterogeneity. Case studies show that subclones can emerge after targeted therapy and have different drug responses. Thus, our question is: what are the genotypes and phenotypes of DLBCL subclones? We hypothesise that DLBCL tumours contain genetic subclones, with phenotypic variation that reflects B cell development. Here, I present the largest single-cell DLBCL dataset to date, consisting of 91 cases with paired single-cell RNA (scRNA) and single-cell ATAC (scATAC) sequencing. This work identified genetic subclones based on scRNA copy number. Interestingly, 85% of tumours have genetic subclones (mean 2, range 1–6). We identified seven clusters of gene expression signatures, or themes, that characterise intratumoural subclone diversity. Four themes (memory B cell, plasma cell, germinal centre identity, and B cell identity) recapitulate B cells blocked in differentiation and account for most of the phenotypic heterogeneity between subclones. Three additional themes (MYC, cell cycle, metabolic) characterise transitory cell states. I used an independent discovery dataset (n = 562 samples) to associate genetic alterations with theme expression. A key observation was that amplification of REL (which encodes c-REL, a NF-κB transcription factor subunit), is negatively associated with expression of the memory B cell theme in the discovery dataset (t-test, p = 1×10⁻¹¹) and this was validated in scRNA of genetic subclones (t-test, p = 0.03). Amplification of REL is a recurrent alteration in germinal centre B cell subtype (GCB) DLBCL, and its functional consequences have not yet been characterised. NF-κB transcription factors (c-REL, RELA) have unique functions in regulating normal B cell development. I hypothesise that REL amplification is an oncogenic event that restricts B cells to the germinal centre and prevents memory cell differentiation. I have designed cell-line and mouse models that will be used to test this hypothesis. In summary, my work has identified DLBCL genetic subclones with highly variable phenotypes that may have distinct therapeutic vulnerabilities. It offers a multi-platform single cell resource for the DLBCL community.