INTRODUCTION: Diffuse large B cell lymphoma (DLBCL) is a malignant non-Hodgkin's lymphoma with approximately one third of patients respond to front-line treatment such as R-CHOP. However, most of the patients often relapse within 2-3 years, and their 3-year progression free survival rate is dropped from 70%-10%. Therefore, understanding of the molecular mechanisms that associated with DLBCL relapse to develop the novel therapeutic approach is in an urgent need.

METHODS: Diagnosis (n=34) and paired relapse samples (n=37) from 34 patients with DLBCL treated with R-CHOP were retrieved from the pathology archives under a protocol approved by the Institutional Review Board. Of these, matched normal germline controls are also available in 10 cases. We performed whole exome sequencing (WES) using the illumine Hiseq2000 platform. Short sequencing reads were aligned to human genome assembly GRCh37/hg19 using the BWA aligner. Variant calls were made and determined to be either 'diagnosis-specific", "relapse-specific" or shared based on variant allele frequency (VAF) in the diagnosis, relapsed and normal controls. To identify relapse specific variants that may play a driver role in relapse, the potential pathogenicity of all relapse specific variants with VAF > 20 is assessed by the UMD predictor and all predicted pathogenic variants are further annotated by Ingenuity variant analysis. Genes with mutations in 2 relapse samples and none of the diagnostic samples, or mutations in >3 relapse sample and <1 pre-treatment sample are selected regardless of the outcome of the UMD predictor. Pathway enriched analysis was performed by Ingenuity analysis.

RESULTS: Analysis of the paired DLBCL samples with matched normal controls demonstrates that relapsed DLBCL develop via divergent evolution from a common precursor as the pre-treatment DLBCL. A total of 661 relapse-specific variants from 595 genes are identified from the above pairwise analysis. 548 of them are missense mutations; 69 are nonsense mutations, and 29 are frameshift mutations. Recurrently mutated genes include TP53 (19% of cases), TNFRSF14 (13.5%), ACTG1 (11.4%), CELSR3 (11.4%), LPR1B (11.4%), GUCY2F (11.4%); ARHGEF10L, DNAH9, DSP, EHD2, HIST1H1B, HIVEP1, KIAA2022, MYC, SKI, SLC29A4, SLIT3, EPHB6, PCDHGA6, BRWD1, C17ORF70 (each 8.6%), and 21 other genes (each 5.7%). Notably, a recurrent missense mutation (p.E50V) in EIF4A2 was identified in 2 cases. About 50% of these recurrently mutated genes are more frequently mutated (p<0.05) in our relapsed DLBCL cohort compared to previously published large DLBCL cohorts comprised predominantly of treatment-naïve DLBCL. Among the top enriched pathways in the set of recurrently mutated genes are aryl hydrocarbon receptor signaling, protein kinase A signaling, ILK signaling, EIF2 signaling, axonal guidance signaling and MYC-mediated apoptosis signaling, while the top enriched pathways in all the mutated genes are axonal guidance signaling, ephrin receptor signaling, actin cytoskeleton signaling and ILK signaling. In addition, ultra-deep sequencing of several relapse-specific variants in pre-treatment samples has identified pre-existing relapse-precursor clones in the pre-treatment tumors, suggesting that these variants may play in role in chemoresistance.

CONCLUSION: WES study on a large cohort of relapsed DLBCL with matched pre-treatment tumors provides a comprehensive view of the mutation landscape of relapsed DLBCL. Novel drivers for relapsed tumors are identified; and alterations in pathways previously not implicated in DLBCL pathogenesis are discovered. shRNA and CRISPR screening using a subset of our candidate genes is underway to functionally validate genes that may mediate chemo-resistance. Our study provides a framework for identifying the genetic mechanisms in DLBCL therapeutic resistance and relapse formation.


Tam:Takeda: Consultancy; Paragon Genomics: Consultancy.

Author notes


Asterisk with author names denotes non-ASH members.