CONCORDANCE BETWEEN IMMUNOHISTOCHEMISTRY AND GENE EXPRESSION PROFILING SUBTYPING FOR DIFFUSE LARGE B‐CELL LYMPHOMA IN THE PHASE 3 PHOENIX TRIAL

Abstract

Introduction: Sequencing studies identified mutational drivers in diffuse large B cell lymphoma (DLBCL), capturing outcome difference in previously unrecognized patient subsets. However, the lack of routine-applicable genomic approaches limits translation of such information to the clinic. Currently, molecular prognostication consists in cell of origin (COO) determination by the Lymph2Cx NanoString assay. We recently developed two independent prognostic signatures incorporating genes reflecting the COO, the activation of pivotal oncogenic pathways, and the composition of tumor microenvironment (TME). We aimed this study at examining the prognostic strength of a model combining the performance of each signature and developing a comprehensive NanoString assay rapidly transferable to the clinic for prognostic purposes. Methods: The expression of the genes was measured by the NanoString nCounter Analysis System using customized probes for 73 genes, including 15 COO genes, 6 additional oncogenic genes (MYC, BCL-2, NFKBIA, PIK3CA, PTEN, STAT3), 47 TME genes, and 5 housekeeping genes. The analysis was performed on 175 newly diagnosed, nodal DLBCL, homogeneously selected from the RHDS0305 and DLCL04 trials. Patients had comparable clinical features and double-hit cases were excluded. Heatmaps, Kaplan–Meier survival estimator, tree-based survival model, and P values were produced by ‘R’ statistical software. Long-rank test was used to compare overall survival (OS) and progression-free survival (PFS) among groups. Multivariate analysis was constructed through the Cox proportional hazards regression model. Results: Based on the expression of the COO and oncogenic genes, a tree-based survival model stratified patients into subgroups showing significantly different survival, with MYC, BCL-2 and NFKBIA holding additional prognostic power based on their high (H) or low (L) expression. The TME panel identified a lower gene expression cluster (C3) with significantly worse survival than those at intermediate (C2) and higher expression (C1). Integration of COO-, TME-, and MYC/BCL-2/NFKBIA-based data produced a new survival risk categorization of DLBCL. The high-risk category, showing the worst outcomes, includes ABC/H/C1-2-3, ABC/L/C3 and GCB/H/C3 cases; the intermediate-risk category comprises ABC/L/C2, GCB/H or L/C1 and UN/H or L/C1 or C3 cases; whereas the low-risk category contains GCB/H or L/C2, GCB/L/C3, UN/H/C1 or C2 and UN/L/C2 or C3 cases, with longer survival. An unsupervised clustering analysis was also performed based on the expression of the entire 74-gene panel and stratified cases into four clusters with significantly different OS (p=0.011) and PFS (p=0.009). In particular, cluster 1 and 4 showed significantly worse survival than cluster 2 and 3 (Figure 1), and a multivariate Cox analysis indicated that the prognostic performance of the panel overcomes the IPI score. Finally, such model was also validated “in silico” using a gene expression profiling dataset (GSE10846 and GSE98588) relative to a cohort of 146 DLBCL patients uniformly selected according to R-CHOP treatment. Conclusions: This study supports the idea that DLBCL heterogeneity involves both tumor and TME, resulting in diverse transcriptional subtypes with distinct outcomes and, putatively, diverse biology. Our integrative analysis prompts the development of a new survival categorization outperforming current prognostic risk-assessment. Moreover, the applicability of a unique Nanostring-based assay to routine biopsies may facilitate the stratification of patients at diagnosis and their inclusion in future trials exploring novel therapeutic approaches.