Neoantigens in Patients with De Novo Follicular Lymphoma: Results from the PRIMA Study

Introduction: Follicular lymphoma (FL) is a slow growing lymphatic cancer characterized by translocations in and overexpression of BCL-2, which inhibits apoptosis. This pathogenesis may generate neoantigens (NAs) that are recognizable by T-cells as part of a patients’ immune surveillance and are unique to specific FL mutations. Tumor mutation burden (TMB), and NA prevalence and prognostic nature, have previously been characterized for diffuse large B-cell lymphoma (DLBCL), identifying that TMB correlated with NA burden (NAB). While TMB did not correlate with outcomes, the presence of a NA, especially BCL2 NA, correlated with outcomes in de novo DLBCL. In addition, the majority of patients were predicted to have ≥1 NA (Paulson, EHA 2019). However, to date, the prevalence of NAs in FL, and their association with clinical outcomes, have not been characterized. The aim of our study was to characterize NA prevalence and evaluate the prognostic value of NA biomarkers, assessed by a targeted, comprehensive genomic profiling (CGP) platform, on progression-free survival (PFS) for patients with de novo FL.

Methods: CGP data on 465 genes were available from patients with FL who provided biopsy samples at screening for the Phase III PRIMA trial (NCT00140582; intent-to-treat [ITT] population =1018; patients received rituximab [R] maintenance vs observation after response to initial first-line treatment with R-CVP [cyclophosphamide, vincristine, prednisone], R-CHOP [cyclophosphamide, doxorubicin, vincristine, prednisone] or R-FCM [fludarabine, cyclophosphamide, mitoxantrone]). CGP was used to calculate TMB, HLA type (OptiType) and NA prediction (NetMHCpan). The prevalence of NAs at time of screening was analyzed by the number and proportion of patients. The prognostic value of NAs were evaluated against the Cox Proportional Hazards null model for PFS including terms for treatment, country and response to induction treatment. The Akaike Information Criterion (AIC), a likelihood ratio test (LRT) p-value vs the null when including an additional NA term and its associated hazard ratio (95% CI), were calculated. A visualization of the genetic mutational landscape was generated using dimension reduction methods.

Results: In total, 247 of the 1202 enrolled patients were assayed with CGP. Baseline characteristics and survival in the biomarker evaluable population (BEP) were consistent with the ITT population. We calculated a median TMB of 6.7 mutations per megabase for the PRIMA cohorts with a median of two predicted NAs per patient. The majority of patients (83%) were predicted to have ≥1 NA. TMB moderately correlated with NAB (0.42 pearson coefficient), Figure A. TMB was not associated with outcomes (hazard ratio [HR] 95% CI: 0.98 [0.94–1.02]) and similarly the presence of a NA was not associated with PFS (HR [95% CI]: 0.90 [0.56–1.44]). The most prevalent predicted NAs (n [patients]; %) were in BCL2 (54; 21.9%), CREBBP (49; 19.8%), EZH2 (14; 5.7%), KMT2D (13; 5.3%), and CARD11 (10; 4.0%). In a pooled analysis of patients from both arms, only presence of EZH2 NA improved the AIC of the null model (AIC=1135.6) to 1134.7 (LRT p=0.09 better than 0.1), Table. The HR (95% CI) of EZH2 NA in this model was 0.46 (0.17–1.25), indicating better survival for the EZH2 NA-negative group. Within each treatment group, HR (95% CI) for EZH2 was 0.27 (0.07–1.13) in the observation group and 1.05 (0.25–4.44) in the R group, Figure B. Inspecting the tSNE mutations by gene, a neighborhood pattern could be seen between EZH2 and BCL2 connected by KMT2D and flanked by CARD11 and CREBBP, Figure C.

Conclusions: We observed that similarly to de novo DLBCL, TMB and NAB were also correlated in patients with de novo FL. While NAB was associated with clinical outcomes in de novo DLBCL, we did not observe these associations in FL. In patients with de novo FL,who discontinued R after initial treatment response, the absence of EZH2 predicted neoantigens were associated with PFS. This is consistent with results on EZH2 mutation status reported by Huet, et al.(Blood Cancer J 2017). A caveat to this hypothesis generating analysis is the small number of PFS events in the EZH2 neoantigen group. These insights may inform future personalized strategies in FL.