Molecular subclusters of follicular lymphoma: a report from the United Kingdom's Haematological Malignancy Research Network

Abstract

Follicular lymphoma (FL) is morphologically and clinically diverse, with mutations in epigenetic regulators alongside t(14;18) identified as disease-initiating events. Identification of additional mutational entities confirms this cancer's heterogeneity, but whether mutational data can be resolved into mechanistically distinct subsets remains an open question. Targeted sequencing was applied to an unselected population-based FL cohort (n = 548) with full clinical follow-up (n = 538), which included 96 diffuse large B-cell lymphoma (DLBCL) transformations. We investigated whether molecular subclusters of FL can be identified and whether mutational data provide predictive information relating to transformation. DNA extracted from FL samples was sequenced with a 293-gene panel representing genes frequently mutated in DLBCL and FL. Three clusters were resolved using mutational data alone, independent of translocation status: FL_aSHM, with high burden of aberrant somatic hypermutation (aSHM) targets; FL_STAT6, with high STAT6 & CREBBP mutation and low aSHM; and FL_Com, with the absence of features of other subtypes and enriched KMT2D mutation. Analysis of mutation signatures demonstrated differential enrichment of predicted mutation signatures between subgroups and a dominant preference in the FL_aSHM subgroup for G(C>T)T and G(C>T)C transitions consistent with previously defined aSHM-like patterns. Of transformed cases with paired samples, 17 of 26 had evidence of branching evolution. Poorer overall survival (OS) in the aSHM group (P = .04) was associated with older age; however, overall tumor genetics provided limited information to predict individual patient risk. Our approach identifies 3 molecular subclusters of FL linked to differences in underlying mechanistic pathways. These clusters, which may be further resolved by the inclusion of translocation status and wider mutation profiles, have implications for understanding pathogenesis as well as improving treatment strategies in the future.

Graphical abstract

Figure 1.

Distribution of the number of mutations. (A) Comparison of the distributions of the number of mutations per patient in FL vs DLBCL. Each bar represents the proportion of patients with FL (red) or DLBCL patients (blue) with the corresponding number of mutated genes. (B) The proportion of patients in the FL study cohort carrying the corresponding genetic feature: the top 20 most common nonaberrant somatic hypermutation genetic features. (C) The proportion of patients in the FL study cohort carrying the corresponding aberrant somatic hypermutations.

Figure 2.

Distribution of mutation sites in the FL study cohort, by gene. The most mutated individual sites are shown for each mutated gene. The total number of mutated sites in each gene is shown by the uppermost red label; the most highly mutated sites in each gene are shown with blue labels, with their position showing the cumulative total of mutations.

Figure 3.

Heatmaps of mutated genes, by cluster. (A) For the 2 BIC-determined clusters, FL_A and FL_B. (B) for the AIC-determined clusters, FL_aSHM, FL_STAT6, and FL_Com. The panel on the righthand side shows the enrichment for mutations within each cluster with a logarithmic q-value scale. Only those mutations are shown that are identified as significantly enriched for the given group, as determined by a Benjamini-Hochberg adjusted q < 0.05 from a χ2 test of independence. “HD” indicates the homozygous deletion or a mutation in this gene.

Figure 4.

Distribution of substitutions and predicted mutation signatures. (A) Comparison of the distributions of the number of substitutions per patient in the cohort divided according to AIC cluster as indicated in the figure. (B) Left: pattern of substitutions in triplet context: observed across the cohort for genes identified as aSHM targets (upper panel), in the K1 signature described by Ye et al (middle panel), and in the RCH signature described by Alkodsi et al (lower panel). The substitution type is indicated in the color-coded squares below the graph, with individual triplet contexts plotted by relative substitution frequency with matching color-coded bars. Right: cosine similarity measures. (C) Predicted mutational Reference Signature (RefSig) contribution to the substitution pattern observed in each sample using the Signal analysis package is displayed as a hierarchically clustered heatmap, divided by BIC FL clustering (color-coded bars above the heatmap). The percentage contribution of the predicted RefSig is indicated in the white (0) to red (100) color scale, as shown in the figure. Predicted contributing signatures are shown to the left. The distribution of predicted contributing signatures is shown to the right as a scatterplot. Note mutation signatures were derived from patterns observed in the lymphoma driver panel; cases with low mutation burden fall below the threshold for signature prediction.

Figure 5.

OS of all patients with follow-up data, by cluster. (A) OS of patients by BIC cluster. (B) OS of patients by AIC cluster. (C) OS of patients initially put on “Watch-and-Wait” therapy by BIC cluster. (D) OS of patients initially put on “Watch-and-Wait” therapy by AIC cluster. (E) OS of patients initially treated with chemotherapy by BIC cluster. (F) OS of patients initially treated with chemotherapy by AIC cluster.

Figure 6.

Mutations in patients who transformed from FL to DLBCL. Distribution of mutations by individual patient stratified by mutated at FL diagnosis only, mutated at DLBCL only, and mutated at both FL and DLBCL diagnosis.

Figure 7.

Variant Allele Frequency Trajectories. (A-D) Show variant allele frequency trajectories (for samples taken at diagnosis with FL and with DLBCL) for a selection of individual patients (numbers 6, 7, 25, and 26, respectively, in [A-D]). Gene/sites are annotated. Patient 6 shows a complex set of gains with selection for HLA-B mutations; patient 7 shows a complex set of gains; patient 25 shows substantial shared mutation burdens, with a couple of gains from zero and one loss and Includes EZH2 as a shared mutation; patient 26 shows a core set of 5 mutations and major gain in CREBBP.