FRANKFURT. With more than 150,000 new cases worldwide each year, diffuse large B-cell lymphoma (DLBCL) is the most common aggressive form of lymphoma. Following diagnosis, patients typically receive a standard treatment regimen consisting of a therapeutic antibody and chemotherapy (R-CHOP or Pola-R-CHP), and nearly two-thirds of patients have a good chance of being cured. However, more than one-third of patients experience a relapse after treatment, or their tumors fail to respond to therapy, necessitating alternative treatments such as CAR T-cell therapy.

The varying effectiveness of standard therapy is due to the considerable molecular heterogeneity of the disease. Researchers have therefore long been searching for molecular tumor characteristics that would allow them to distinguish between different DLBCL subtypes and treat them more specifically.

A heterogeneous disease

To date, diffuse large B-cell lymphoma has been extensively investigated at the genetic level. This has led to classification systems that distinguish subtypes according to genetic alterations and patterns of gene expression.

An international research team led by Goethe University Frankfurt, Universitätsmedizin Frankfurt, the German Cancer Consortium (DKTK), and the Frankfurt Cancer Institute has now identified novel tumor characteristics beyond genetics that characterize DLBCL tumors. In the future, these features may enable the identification of high-risk patients for whom standard therapy is unlikely to be successful.

To achieve this, researchers analyzed tumor samples from 478 patients by examining the mutations in the tumors and the expression of each gene. In addition, they determined which proteins were produced in the tumor cells and in what quantities (proteomic analysis).

They then evaluated these data using AI models to identify patterns within the datasets. Professor Florian Büttner from the Faculty of Medicine and the Institute of Computer Science, whose team developed the machine learning models, explains: “Our model demonstrates how interpretable machine learning can reveal relationships across different molecular layers: we succeeded in correlating mutation and protein patterns with treatment outcomes.” This enabled the research team to classify patients into groups that both describe the biology of the disease and provide insights into potential therapeutic options. The findings were validated using high-resolution single-cell tumor analyses.

Characteristics of high-risk patients

Dr. Julius Enssle, a physician-scientist at Universitätsmedizin Frankfurt and the National Institutes of Health in the United States, one of the study’s three first authors alongside biochemist Dr. Björn Häupl and computer scientist Arber Qoku, explains: “We can now much better understand the biological characteristics of DLBCL tumors that determine patients’ clinical prognosis and are independent of previously established risk factors. Our data show that different genetic mutations can lead to similar tumor cell characteristics in DLBCL, and we now have a much clearer understanding of these mechanisms. This is particularly important for high-risk patients.” According to Enssle, the tumors in this group – referred to in the study as PG4 (proteogenotype 4) – are centered around the gene MYC, which drives tumor cell growth and division. Furthermore, very few immune cells are present in the microenvironment of these tumors: “The tumors of high-risk patients are immunologically ‘cold’ – in particular, the function of cytotoxic T cells is suppressed, which normally recognize and eliminate tumor cells.”

Building on these findings, the research team succeeded in pharmacologically inhibiting the molecular programs involving MYC in cultured PG4 lymphoma cells, thereby selectively eliminating the lymphoma cells. Enssle states: “This has enabled us to identify potential targets for the development of precision diagnostics and therapies.” Professor Thomas Oellerich, Director of the Department of Medicine 2 at Universitätsmedizin Frankfurt and lead investigator of the study, is convinced: “Although there is still a long way to go, we have taken an important step toward personalized medicine for aggressive lymphoma. In the long term, our findings may help identify high-risk patients earlier and tailor their treatment more precisely to the underlying tumor biology.”