Clinicopathologic features are poor at predicting the risk of CNS relapse after initial treatment in DLBCL, or at guiding CNS prophylactic therapy. However, mutational profiles may correlate with primary extranodal disease, including primary CNS lymphoma (Ollila T, Curr Treat Options Oncol 2018). We hypothesized that specific genomic profiles in DLBCL would also correlate with CNS relapse or with systemic-only (non-CNS) relapse. We examined 26 DLBCLs with CNS-only (n=13) or systemic-only relapse (n=13) using a 592 gene next-generation sequencing (NGS) assay (Illumina NextSeq, average coverage depth >750x; Caris Life Sciences) performed on formalin-fixed tissue. We also determined cell of origin (COO) by IHC; MYC/BCL2/BCL6 rearrangements by FISH, and karyotype by cytogenetics. We used the novel LymphGen genomic DLBCL classifier (Wright GW, Cancer Cell 2020) to group tumors into the MCD, ST2, EZB, BN2, and A53 subtypes. We then constructed a simplified hierarchical classifier (hc) usable with common clinical multigene NGS panels to identify 3 relevant subtypes: (1) hc-MCD subtype (defined as MYD88L265P or >2 mutations in CD79B, PIM1, ETV6, BTG1, TBL1XR1, or PRDM1), (2) hc-TP53 subtype characterized by TP53 mutations +/- complex karyotype, and (3) hc-GCB subtype with ≥2 mutations in BCL2, CREBBP, EZH2, KMT2D, TNFRSF14, GNA13, MEF2B, or PTEN. We compared prevalence of these subtypes between our groups and a reference dataset of unselected DLBCL tumors combining data from Chapuy (Nat Medicine 2018; n=135; dbGaP phs000450) and Reddy (Cell 2018; n=1001; EGA: EGAS00001002606). There was no difference between DLBCL with CNS-only or systemic recurrence in any standard clinicopathologic characteristics, including COO, FISH, or karyotype. The LymphGen MCD subtype was significantly more frequent in DLBCL with CNS-only relapse compared with reference datasets (39% vs. 7%, P=.001), but not in DLBCL with systemic relapse (P=.05). Conversely, the ST2 subtype was more frequent in DLBCL with systemic relapse (23% vs. 3%, P=.010) and completely absent in CNS relapse (0%). The A53 or EZB subtypes did not significantly differ between any groups. The simplified NGS classifier identified 96% of LymphGen MCD cases and 100% of A53 cases. The hc-MCD subtype was also significantly associated with CNS relapse (46% vs. 18%, P=.018) but not with systemic relapse (P=.26), whereas hc-TP53 and hc-GCB subtypes did not differ from the reference dataset (P=.15 and P=.70, respectively). Our data demonstrate that the MCD DLBCL subtype is specifically associated with the risk of CNS relapse. Full molecular classification of DLBCL is not yet incorporated into current clinical practice, as it requires integration of single nucleotide, structural chromosomal variants, and copy alterations. We show that a meaningful molecular predictor signature for the MCD subtype can be obtained from clinically validated multigene NGS assays. If validated, this signature may prove useful for selecting patients for CNS-directed prophylactic therapy.

Citation Format: Thomas A. Ollila, Habibe Kurt, Jozal Waroich, John Vatkevich, Ashlee Sturtevant, Nimesh Patel, Diana O. Treaba, Patrycja M. Dubielecka, Adam J. Olszewski. Genomic subtypes correlate with the risk of central nervous system (CNS) recurrence in diffuse large B-cell lymphoma (DLBCL) [abstract]. In: Proceedings of the AACR Virtual Meeting: Advances in Malignant Lymphoma; 2020 Aug 17-19. Philadelphia (PA): AACR; Blood Cancer Discov 2020;1(3_Suppl):Abstract nr PO-38.