Characterization of DLBCL cell of origin-phenotypes based on tumor microenvironment features

Diffuse large B-cell lymphoma (DLBCL) is an aggressive form of non-Hodgkin lymphoma with a high recurrence rate.

The molecular profiling of DLBCL tumors culminated in several immunohistochemistry algorithms for prognostic stratification.

Among those, the Hans classifier is widely used for classifying DLBCL into germinal center B-cell-like (GCB) and non-germinal center/activated B-cell-like (non-GCB/ABC) subtypes.

The Hans classifier primarily evaluates protein expression of tumor-associated markers, however the tumor microenvironment (TME) of DLBCL includes a myriad of immune and stromal cells, cytokines, and extracellular matrix components that contribute to tumor growth, immune evasion, and recurrence rate.

Although the Hans classifier provides a practical method for subtype identification, incorporation of TME information may improve risk stratification and further refine patient groups.

Here, we present an unbiased deep learning-based approach to extract meaningful features from TME of DLBCL tumors for the automated processing and analysis of multiplexed images of a DLBCL patient cohort.

Our pipeline quantifies a range of features that describe tumor sample cell composition, morphology, and its spatial organization.

We point to alterations in the proportions of several cell populations between GCB and ABC tumors including increased immune cell proportions of the ABC and its preferential interaction with the M2-macrophages.

Our analysis offers an in-depth characterization of the DLBCL subtypes and is exemplary of how our pipeline can be used for detailed quantitative analysis of a tumor and its subtypes.