Integrated genomics with refined cell-of-origin subtyping distinguishes subtype-specific mechanisms of treatment resistance and relapse in diffuse large B-cell lymphoma

Abstract

Up to 40% of diffuse large B-cell lymphoma (DLBCL) patients do not experience a durable response to frontline immunochemotherapy, and prospective identification of high-risk cases that may benefit from personalized therapeutic management remains an unmet need. Molecular phenotyping techniques have established a landscape of genomic variants in diagnostic DLBCL; however, these have not yet been applied in large-scale studies of relapsed/refractory DLBCL, resulting in incomplete characterization of mechanisms driving tumor progression and treatment resistance. Here, we performed an integrated multiomic analysis on 228 relapsed/refractory DLBCL samples, including 24 with serial biopsies. Refined cell-of-origin subtyping identified patients harboring GCB and DZsig+ relapsed/refractory tumors in cases with primary refractory disease with remarkably poor outcomes, and comparative analysis of genomic features between relapsed and diagnostic samples identified subtype-specific mechanisms of therapeutic resistance driven by frequent alteration to MYC, BCL2, BCL6, and TP53 among additional strong lymphoma driver genes. Tumor evolution dynamics suggest innate mechanisms of chemoresistance are present in many DLBCL tumors at diagnosis, and that relapsed/refractory tumors are primarily comprised of a homogenous clonal expansion with reduced tumor microenvironment activity. Adaptation of personalized therapeutic strategies targeting DLBCL subtype-specific resistance mechanisms should be considered to benefit these high-risk populations.

Fig. 1. Genomic landscape and enrichment of genetic variants in rrDLBCL tumors.

A Oncoplot representing recurrent non-silent somatic single variants or insertions/deletions in rrDLBCL tumor samples (n = 204). Predicted functional consequence represented by color (missense, green; truncating, red; in-frame, orange; multi-hit, black). Bar plot (right) represents the sum of samples with a mutation in the respective gene. Genes depicted were filtered for genes with established potential in lymphoma (see Supplemental Table 5). B Oncoplot representing recurrent copy number alterations detected in rrDLBCL tumor samples (n = 132). Predicted event is represented by color (Amplification [Amp], dark red; Gain, light red; Deletion [Del], dark blue; Loss, light blue). Genomic regions depicted were filtered for locations with established potential in lymphoma (see Supplemental Table 6). C Forest plot and enrichment analysis comparing recurrent non-silent somatic single variants or insertions/deletions between rrDLBCL (n = 204) and ndDLBCL (n = 382) tumors. OR > 1 indicates enrichment in rrDLBCL tumors. D Forest plot and enrichment analysis comparing recurrent copy number alterations between rrDLBCL (n = 132) and ndDLBCL (n = 365) tumors. OR > 1 indicates enrichment in rrDLBCL. E Summary visualizing frequency of recurrent non-silent somatic single variants or insertions/deletions and copy number alterations in rrDLBCL and ndDLBCL tumors. Alteration type represented by color (mutation, green; gain, red; loss, blue). For clarity, only SNV and CNV data from genes/regions from C, D are shown. Bubble size represents −Log10(p value). Points above dotted line indicate enrichment in ndDLBCL tumor samples, points below dotted line indicate enrichment in rrDLBCL tumor samples. F LymphGen classification in ndDLBCL (n = 349) and rrDLBCL (n = 131) tumor samples.

Fig. 2. High-risk molecular features and poor outcomes in rrDLBCL tumors harboring dark-zone COO gene expression signature.

A COO composition in ndDLBCL (ABC, n = 102; GCB, n = 122; DZsig+, n = 44; Unclassified, n = 37) and rrDLBCL (ABC, n = 68; GCB, n = 39; DZsig+, n = 24; unclassified, n = 15) tumors. B Kaplan–Meier estimation of overall survival (OS, from the time of original diagnosis) for patients from ndDLBCL cohort (left, n = 305) and rrDLBCL cohorts (right, n = 102) stratified by COO subtype. Patients analyzed at the indicated time points are shown in the table. ABC Blue, GCB yellow, Unclassified green, DZsig+ red. P value determined via Log-rank test. C Ternary plot visualizing the relative frequency of recurrent non-silent somatic single variants or insertions/deletions in rrDLBCL tumors stratified by COO subtypes ABC (n = 58, blue), GCB (n = 37, yellow), and DZsig+ (n = 22, red). Mutations affecting >15% of tumors within each respective subtype were labeled. D Ternary plot visualizing the relative frequency of recurrent copy number alterations in rrDLBCL tumors stratified by COO subtypes ABC (n = 37, blue), GCB (n = 25, yellow), and DZsig+ (n = 14, red). CNVs affecting >25% of tumors within each respective subtype were labeled. E Forest plot and enrichment analysis comparing recurrent non-silent somatic single variants or insertions/deletions between rrDLBCL and ndDLBCL tumors stratified by COO subtypes ABC (left; n = 58 rrDLBCL, n = 86 ndDLBCL), GCB (middle; n = 37 rrDLBCL, n = 110 ndDLBCL), and DZsig+ (right; n = 22 rrDLBCL, n = 38 ndDLBCL). OR > 1 indicates enrichment in rrDLBCL. F Forest plot and enrichment analysis comparing recurrent copy number alterations between rrDLBCL and ndDLBCL tumors stratified by COO subtypes ABC (left; n = 38 rrDLBCL, n = 87 ndDLBCL), GCB (middle; n = 25 rrDLBCL, n = 110 ndDLBCL), and DZsig+ (right; n = 14 rrDLBCL, n = 38 ndDLBCL). OR > 1 indicates enrichment in rrDLBCL.

Fig. 3. Transcriptional programming in rrDLBCL.

A Differential gene expression analysis from bulk RNA-sequencing in rrDLBCL (n = 144) and ndDLBCL (n = 303) tumor samples. Genes with significant enrichment (FDR < 0.05, |Log2FC|>0.5) in rrDLBCL tumors (maroon) or ndDLBCL tumors (gray) are indicated by color. Dotted line indicates −Log10(FDR) = 0.05 and Log2|FoldChange| = 0.5. B Gene set enrichment analysis (GSEA) describing disrupted pathways as indicated from differential gene expression in A. X axis represents the absolute value of normalized enrichment score (NES), y axis represents −Log10(adj p value). Gene set size indicated by bubble size. NES indicated by color (Blue–negative enrichment, red–positive enrichment). Dotted line indicates −log₁₀(0.05). C Leading edge plots visualizing enrichment score for DLBCL_RNASig-DOWN and DLBCL_RNASig-UP gene sets from GSEA in (B). D Classification of lymphoma microenvironment (LME) signatures inferred from gene expression profiling in ndDLBCL (n = 303) and rrDLBCL (n = 143) tumor samples. LME-DE red, LME-GC blue, LME-IN teal, LME-ME purple. E Cell type abundance of the associated lymphoma microenvironment inferred from gene expression analysis using the CIBERSORTx tool in ndDLBCL (n = 303) and rrDLBCL (n = 144) tumor samples. Cell types are grouped by rows. COO annotation is indicated below. F Enrichment of inferred T-cell subtypes from E. p-value determined using Wilcoxon rank sum test. G Enrichment of inferred T-cell subtypes among GCB (n = 39 rrDLBCL, n = 121 ndDLBCL) and DZsig+ (n = 22 rrDLBCL, n = 44 ndDLBCL) tumors from E. p-value determined using Wilcoxon rank sum test.

Fig. 4. Relapse timing and tumor biology.

A Kaplan–Meier estimation of overall survival (OS) for rrDLBCL patients stratified by the time at which their first relapse sample (R1) was collected. Patients available for analysis at the indicated time points are shown in the table. Primary refractory (n = 23)–Purple, early relapse (n = 34)–blue, late relapse (n = 20)–teal. P value determined via Log-rank test. B Proportion of COO among bins grouping rrDLBCL tumors by time at which the relapse sample was collected. Primary refractory, n = 31; early relapse, n = 51; late relapse, n = 26. ABC–blue, DZsig+–red, GCB–orange, Unclassified (UNC)–green. C Oncoplot representing recurrent non-silent somatic single variants or insertions/deletions detected in rrDLBCL tumor samples (n = 127) grouped by time at which the relapse sample was collected. Primary refractory, n = 38; early relapse, n = 60; late relapse, n = 29. Predicted functional consequence represented by color (missense, green; truncating, red; in-frame, orange; multi-hit, black). The bar plot (right) represents the sum of samples with a mutation in the respective gene. Genes depicted were filtered for genes with established potential in lymphoma (see Supplemental Table 5). D Ternary plot visualizing the relative frequency of recurrent non-silent somatic single variants or insertions/deletions in rrDLBCL tumors stratified by time at which the relapse sample was collected. Primary refractory (n = 38, purple), early relapse (n = 60, blue), late relapse (n = 29, teal). Mutations affecting >10% of tumors within each respective group were labeled. E Ternary plot visualizing the relative frequency of recurrent copy number alterations in rrDLBCL tumors stratified by time at which the relapse sample was collected. Primary refractory (n = 22, purple), early relapse (n = 26, blue), late relapse (n = 13, teal). Mutations affecting >25% of tumors within each respective group were labeled. F Differential gene expression analysis from bulk RNA-sequencing in rrDLBCL comparing primary refractory tumors (n = 30) against early + late relapse tumors (n = 77). Genes with significant enrichment (FDR < 0.05, |Log2FC|>0.5) are indicated by color. Genes with −Log(FDR) > 1.5 & |Log2FC|>2.5 are labeled in gray, while genes with established potential in lymphoma are labeled in black. G Gene set enrichment analysis (GSEA) describing disrupted pathways as indicated from differential gene expression between cases from F. X-axis represents absolute value of normalized enrichment score (NES), y axis represents −Log10(FDR). Dotted line indicates −Log₁₀(0.05). Gene set size indicated by bubble size. NES indicated by color (Blue negative enrichment, red positive enrichment).

Fig. 5. Patterns of clonal evolution in rrDLBCL tumors.

A Time from diagnosis to relapse sample acquisition (months) from samples from A. Sample type indicated by color (Dx gray, R1 red, R2 dark red, R3 brown). Line breaks indicate a previous relapse event was clinically annotated, but no sample was available for analysis. The dashed line visualizes 24 months from diagnosis. B Sankey plots visualizing LymphGen (n = 18) molecular classifications in paired diagnostic and relapsed/refractory tumor samples. C Sankey plots visualizing COO (n = 18) and Refined COO (n = 14) molecular classifications in paired diagnostic and relapsed/refractory tumor samples. D Raw vaf assessed in BCL2, CREBBP, KMT2D, and TP53 genes between paired diagnostic and relapsed/refractory tumor samples. Connecting lines indicate the same variant present in both paired samples from respective patients. E Clonal population structure in representative paired ndDLBCL and rrDLBCL samples inferred from PyClone-VI and ClonEvol tools using bulk SNV and CNV data as input, visualized by bell plot and sphere of cells. Colors represent distinctly inferred clones. Colored dots comprising spheres of cells are representative of the clonal abundance depicted in respective bell plots. Samples relapsing before 24 months are shown on the left, and samples relapsing after 24 months are shown on the right. Mutations in select lymphoma-driving genes are shown in affected clones. F IGH clonality score determined from the TRUST4 tool using bulk RNA-sequencing as input in paired ndDLBCL and rrDLBCL (R1 and R2) tumor samples. Dots represent individual samples. Lines connect paired samples. G IGH clonality score determined from the TRUST4 tool in ndDLBCL (n = 303) and rrDLBCL (n = 114) tumor samples. Dots represent individual samples. P value determined using the Wilcoxon rank sum test.