Longitudinal Analyses of Diagnostic-Relapse Biopsies of Diffuse Large B Cell Lymphoma Reveal a Poor Risk Subset of ABC Patients Based on the Expression of a 30 Gene Panel

Background: Although diffuse large B cell lymphoma (DLBCL) can be cured using immuno-chemotherapy, 40% of patients experience relapse or refractory disease. Large-scale profiling studies have mainly focused on DLBCL at diagnosis, resolving different outcome groups based on gene expression (e.g. cell-of-origin (COO) or molecular high grade), MYC/BCL2 translocations (double-hit lymphoma) or gene mutations and copy number aberrations (Schmitz et al, NEJM 2018; Chapuy et al, NatureMedicine 2018). In comparison, longitudinal studies have been hindered by the limited availability of sequential biopsy samples. To date, the relapse-specific gene mutations identified are limited and inconsistent across studies. In our study, we have focussed attention on the changes in gene expression profile (GEP) accompanying DLBCL relapse.

Methods: We retrospectively collected archival paired diagnostic/relapse formalin fixed paraffin embedded tumor biopsies from 38 de novo DLBCL patients collected from multiple UK sites treated with rituximab-based immuno-chemotherapy, where partial or complete remission was reported following treatment. COO classification was performed by the Lymph2Cx assay on NanoString to distinguish activated B-cell-like (ABC) and germinal center B-cell-like (GCB) subtypes. The Ion AmpliSeq™ Transcriptome Human Gene Expression Kit was used to measure the expression levels of > 20,000 genes on the paired samples.

Results: COO remained stable from diagnosis to relapse in 17 ABC-ABC pairs, 11 GCB-GCB pairs and 4 unclassified (UNC)-UNC pairs. Frank COO switching was observed in 6 cases (1 ABC-GCB, 2 ABC-UNC, 2 GCB-UNC, 1 UNC-ABC). Pairs with stable COO were taken forward for further analysis. Gene expression analysis using the limma R package identified 163 and 136 genes as differentially expressed (DE) (p <= 0.01 and absolute log₂FC > 1) between the diagnostic and relapse biopsies in ABC and GCB tumors respectively, with only a one gene overlap. Gene Set Enrichment Analysis further suggested that ABC and GCB relapses are mediated via different mechanisms, with tumor growth and proliferation signatures enriched in ABC relapses, whilst adaptive immunity-related signatures accompanied GCB relapses.

Next, we aimed to utilise our relapse-specific genes to identify outcome predictors at diagnosis using publicly available GEP datasets. In order to increase our discovery power and accuracy, a larger set of DE genes from the paired differential analysis (796 genes in ABC pairs and 387 from GCB pairs) were selected (p <= 0.05) and subsequently used in a training cohort (GEP from Reddy et al, Cell 2017). The Prediction Analysis for Microarrays R (PAMR) algorithm identified a 30-gene signature within DE genes from ABC pairs (Fig1.A), capable of separating the 249 ABC cases into 136 low and 113 high-risk cases with significantly inferior overall survival (Hazard Ratio (HR)=1.89, log-rank p=0.0017, measure of goodness-of-fit C-index=0.71; Fig1.B). No equivalent signature was found in the GCB cases using this approach.

The prognostic significance of this 30-gene discriminator was successfully validated using a linear predictor in two independent GEP datasets: 1) a population-based cohort (Lenz et al, NEJM 2008) with 93 R-CHOP-treated ABC cases identifying 47 low and 46 high-risk cases (HR=1.92, p=0.046, C-index=0.77; Fig1.C) and 2) a clinical trial dataset (REMoDL-B, Davies et al, Lancet Oncol 2019) with 255 ABC cases identifying 110 low and 145 high-risk ABC cases (HR=1.95, p=0.0051, C-index=0.70; Fig1.D).

Conclusions: Here we describe a 30-gene discriminator in ABC-DLBCL, derived from genes differentially expressed between diagnosis and relapse, that allowed the definition of clinically distinct high and low risk subgroups in ABC-DLBCLs at diagnosis. The clinical translation of such a tool may be useful to guide therapy for this unfavourable subgroup of ABC-DLBCLs. Validation of this signature is currently underway in additional datasets and further study is required to understand the contribution of these genes in DLBCL pathology.

View largeDownload slide

View largeDownload slide

Close modal