Molecular Correlates of Central Nervous System Relapse in Diffuse Large B-Cell Lymphoma

Introduction: Central nervous system (CNS) relapse is a rare phenomenon in diffuse large B-cell lymphoma (DLBCL), occurring in less than 5% of all patients, but is associated with disproportionate morbidity and mortality. Indeed, the median survival of patients diagnosed with CNS relapse is as low as 2-4 months. Individual risk factors for CNS relapse are well established, and include clinical parameters such as stage, number/type of extranodal sites and elevated lactate dehydrogenase. These and other clinical risk factors have been integrated into a risk score that is reproducible and easy to calculate (CNS International Prognostic Index). Moreover, molecular attributes such as double-hit translocation status, MYC/BCL2 dual protein expression and the activated B-cell-like subtype have been associated with a higher risk of CNS relapse. However, while experts recommend prophylactic interventions for high-risk patients, the major shortcoming of available risk tools is their limited sensitivity. Herein, we evaluated whether gene expression and/or mutational profiles can identify those patients that will ultimately experience CNS relapse, and whether intratumoral heterogeneity impedes accurate prognostication.

Methods: We accrued diagnostic FFPET samples from 230 newly diagnosed DLBCL patients, selected to fall into 3 clinical groups: 1) cases with CNS relapse/CNS involvement at diagnosis (n=58); 2) cases with systemic relapse but without CNS relapse (n=64) and 3) cases without any relapse (n=108). These 230 samples were subjected to microarray-based gene expression profiling and differential gene expression analysis. Pathway analysis was performed using Gene Set Enrichment Analysis on ranked gene lists. We assembled a partially overlapping dataset with mutation data of 45 genes in 213 diagnostic samples (n=65 with CNS relapse, 62 with systemic relapse and 86 without relapse). Lastly, we performed exome sequencing in 5 pairs (peripheral and CNS parenchymal tumors) of patients with CNS relapse or CNS involvement at diagnosis, and reconstructed clonal phylogenies using PyClone.

Results: Focusing on gene expression data at first, we did not observe significant differential expression between CNS relapse and non-relapse cases at the individual gene level. This was in contrast to the comparison between systemic relapse vs. non-relapse cases where 368 genes were differentially expressed (adjusted P<0.05). In terms of pathway analysis, minimal gene set enrichment was seen in CNS relapse cases, whereas functional annotations such as translation, ribosome biogenesis and MYC targets were significantly enriched in cases with systemic relapse. In keeping with these observations, the percentage of cases that were positive for the recently published double hit signature was highest in cases with systemic relapse (64% vs. 39% in CNS relapse cases and 27% in cases without relapse, P=0.012). However, CNS relapsing cases were defined by down-regulation of numerous immune signatures (e.g. interferon and multiple T cell signatures), suggesting that an intact immune response may have a protective effect on CNS relapse. Considering mutation data, we found that TP53 and SGK were most commonly mutated in systemic relapse cases, while TNFRSF14 and KTM2D were most commonly mutated in non-relapse cases (all adjusted P<0.05). The only gene mutation with a borderline significant trend for enrichment in CNS relapse cases was MYD88 (adjusted P=0.05). We then performed exome sequencing of 5 tumor pairs. A subset of high-confidence somatic variants and tumor purity were used as input for PyClone to infer clonal population structures. In all pairs, we documented the existence of common ancestral mutations, as well as significant clonal divergence, with CNS-exclusive mutations not identified in diagnostic specimens.

Conclusion: In summary, we have documented that CNS and systemic relapse result from distinct biological processes that, in part, may be associated with the underlying taxonomy of DLBCL. Our findings further show that CNS relapse results from the dissemination of sub-clones that may not be readily sampled at the time of diagnosis, and that intratumoral heterogeneity may limit our ability to predict CNS relapse. Large-scale, integrative analyses and in-depth characterization of clonal trajectories hold the promise to increase our ability to predict dissemination of DLBCL into the CNS.