A Novel Genetics-Based Classification of Advanced Follicular Lymphoma Identifies Prognostic Subgroups Following Bendamustine-Rituximab Immunochemotherapy

Introduction

Follicular lymphoma (FL) is the most common indolent non-Hodgkin lymphoma and is considered incurable when presenting in advanced stage. When treatment is required, immunochemotherapy is typically associated with excellent disease control. However, most patients eventually experience disease relapse and ~10-20% experience histological transformation (HT), typically to diffuse large B-cell lymphoma. Progression of disease within 24 months (POD24) and HT are the 2 major events that dictate dismal patient outcomes. Available risk models are not able to reproducibly identify, at the time of diagnosis, these high-risk patients following different treatment regimens. We performed an analysis using genomic data to identify biologically and clinically unified subgroups within a real-world cohort of FL patients with advanced stage disease uniformly treated with bendamustine-rituximab (BR)– a modern standard-of-care regimen.

Methods

The patient cohort was identified in the British Columbia Cancer Lymphoid Cancer database with the following criteria: age 18 years or older, FL grades 1-3A, symptomatic advanced stage disease (Ann Arbor stages III-IV or not amenable to radiation) and received BR as first systemic therapy. DNA (n = 193 [whole genome (n = 51) and exome (n = 142)]) sequencing was performed on diagnostic biopsies alongside fluorescent in situ hybridization (FISH) for BCL2 rearrangement. Manta and GRIDSS2 were used to identify structural variants. Somatic point mutations were obtained using an ensemble of 4 variant callers. Selected coding variants, hotspots, mutations in aberrant somatic hypermutation (aSHM) targets, and BCL2 rearrangement status were converted to a binary matrix for non-negative matrix factorization (NMF) clustering. Battenberg was used to identify copy number alterations from WGS data. Sigprofiler was run on samples with WGS data to identify mutation signatures. Using the clustering results, a random forest (RF) classifier was trained to assign samples to genetic classes.

Results

Histone modifying genes were the most frequently mutated genes with KMT2D, CREBBP, and EZH2 mutated in 68%, 61%, and 22% of tumors, respectively. Rearrangements in BCL2 were detected in 90% of tumors using FISH and/or sequencing methods. Application of NMF clustering resolved an optimal solution with 4 genetic clusters. The first group was characterized by an increased burden of mutations at regions commonly affected by aSHM (aSHM cluster; 33% of the cohort). The second group was characterized by samples with missense mutations in the CREBBP lysine acetyltransferase (KAT) domain (C-KAT cluster; 23% of the cohort). The third group was enriched for EZH2 Y646 hotspot mutations and TNFRSF14 mutations (EZT cluster; 31% of the cohort). The final group had a paucity of recurrent driver mutations, including KMT2D (KWT cluster; 14% of the cohort). Although the NMF clustering did not include copy number, we identified copy number alterations associated with certain subgroups. Amplifications of the 1q region containing FCGR2B were enriched in the aSHM cluster. Amplifications in chromosome 7 containing genes associated with lymphoma, such as EZH2, were present in all clusters except the C-KAT cluster. Deletions of the 1p region containing TNFRSF14 were enriched among EZT tumors. Mutation signature analysis revealed higher levels of SBS84, a signature associated with AID activity, in the aSHM cluster.

The m7-FLIPI risk score failed to predict POD24 in our cohort (16.5% vs. 20% POD24 rate in patients with low vs. high risk; P = 0.60). There was a significant difference in the POD24 proportions between the genetic subgroups (P < 0.001) with only one event in the EZT cluster (2%) compared with 29%, 24% and 8% in the aSHM, C-KAT and KWT clusters, respectively. Moreover, the EZT cluster had a lower cumulative rate of HT (5-year rate 2% compared with 18% aSHM, 16% C-KAT, and 16% KWT; log-rank P = 0.02 across the groups).

Conclusions

We identified 4 novel genetic subgroups of advanced stage FL with prognostic implications. In the context of front-line BR, patients in the EZT subgroup had excellent outcomes, including a very low rate of HT. Development of an RF classifier allows application to other patient cohorts to validate these clusters beyond symptomatic advanced stage disease and determine whether the clusters are predictive, identifying selective advantage to specific treatment regimens.