Individual projects

Research project

DC5 Deciphering the spatial architecture of lymphoma


In this project we will understand how tumor heterogeneity in B-Cell lymphoma contributes to poor clinical outcomes. To that end, we will explore a unique dataset that has whole-exome sequencing and spatial transcriptomics data for 100 lymphoma samples covering the major subtypes.

Keywords: Lymphoma, genomics, spatial transcriptomics, artificial intelligence 

See more

B-cell lymphomas are a group of very heterogeneous malignancies that, together, are the most common form of hematological cancer (5% of all cancer diagnoses worldwide). Broadly speaking, they can be categorized into two large groups: Hodgkin lymphoma (HL, 10% of all cases) and non-Hodgkin lymphoma (nHL, most of the remaining 90%). Furthermore, within nHL, there are over 70 sub-entities, the two most common ones being diffuse-large B-cell lymphoma (or DLBCL) and Follicular Lymphoma (FL).

This heterogeneity is also seen from a molecular perspective, as the genetic alterations associated with each of these subtypes are also different. In HL, the most common somatic mutations are in the JAK-STAT and NFKB pathways. In FL they tend to occur in the B-cell receptor pathway and epigenetic modifiers. Finally, in DLBCL they are a mixture of both, with some in chromatin remodelers (MLL2, CREBBP, ARID1A), and others to the JAK-STAT pathway (SOCS1) and classic driver genes such as TP53, CDKN2A or BCL2.

The tumor microenvironment (TME) is also strikingly different across B-cell lymphoma subtypes. The overall architecture of non-tumoral cells in HL consists of a supportive milieu of non-malignant cells whose composition is completely different from that of normal lymphoid tissue. In contrast, in FL, the cell composition and spatial arrangement within the TME resemble that of normal lymphoid tissue, with, neoplastic nodules similar to germinal centers.

The consequence of this heterogeneity is that we currently lack a clear clinical strategy to treat this group of diseases. Within hematological tumors, the recent clinical progress mostly comes from therapies based on monoclonal antibodies and CAR-T cells, which enabled the successful treatment of several types of the disease with poor prognosis. However, many of the resulting benefits are not general to all hematological tumors, with the situation being particularly dire in the specific case of B-cell lymphomas. For this reason, it is critical to embrace the complexity of these tumor types with new tools to understand their oncogenic processes.

The main objective of this project is to understand the relationship between the different factors driving oncogenesis and heterogeneity in the major subtypes of B-cell lymphomas (HL, DLBCL and FL). To that end, we will characterize a large cohort of B-cell lymphoma samples (n = 100) that represent the major subtypes of the disease (HL, DLBCL, and FL) different of -omics layers (whole-exome sequencing and spatial transcriptomics). Then, we will use the spatial transcriptomics datasets from the different samples to find the location of the cell types of each sample, allowing us to also find the specific state of the different cells. Finally, we will build a multi-scale, multi-omics model of each of the three subtypes of B-cell lymphomas. To that end, we will use a revolutionary new approach to integrate biological data, biology-driven deep neural networks, to merge the WES and spatial transcriptomics profiles, allowing us to identify how the elements of the tumor interact with each other across biological scales and B-cell lymphoma subtypes.

See less


Dana-Farber Cancer Institute (Boston, USA)