Genetic and Functional Drivers of Diffuse Large B Cell Lymphoma

Authors

Reddy A

Zhang J

Davis NS

Moffitt AB

Love CL

Waldrop A

Leppa S

Pasanen A

Meriranta L

Karjalainen-Lindsberg ML

Norgaard P

Pedersen M

Gang AO

Hogdall E

Heavican TB

Lone W

Iqbal J

Dave SS

Doi

10.1016/j.cell.2017.09.027

PMID: 28985567 · DOI: 10.1016/j.cell.2017.09.027 · Journal: Cell (2017)

TL;DR

Reddy et al. performed integrative whole-exome and transcriptome sequencing of 1001 newly-diagnosed DLBCL patients (502 with paired germline DNA) treated uniformly with rituximab-containing regimens, defining a comprehensive landscape of 150 recurrently mutated driver genes. They characterized functional dependencies via a genome-wide CRISPR screen across six lymphoma cell lines and built a multivariate prognostic model combining genetic alterations with cell-of-origin and MYC/BCL2 expression that outperformed the International Prognostic Index (IPI), cell-of-origin classification, and dual MYC/BCL2 expression alone.

Cohort & data

  • 1001 de novo DLBCL tumors with whole-exome sequencing; 400 paired tumor-normal pairs from this study plus 102 previously published pairs (Lohr 2012, Morin 2011, Pasqualucci 2011, Zhang 2013) for a total 502-pair primary discovery set.
  • All cases treated with a rituximab-containing standard regimen; complete clinical annotation including IPI, response, overall survival, gender, age, stage, performance status, and extranodal sites.
  • RNA-seq performed on 775 cases for which adequate material was available; 625 used in core integrative analysis.
  • Cell-of-origin assignment via RNA-seq classifier (Wright 2003): 313 ABC, 331 GCB, remainder unclassified. Cross-validated against Nanostring (n=200, R²=0.87, p<10⁻⁶) and the Hans IHC algorithm (n=654, p<10⁻⁶).
  • Cancer types: DLBCL NOS; one Burkitt lymphoma cell line (BL, BJAB) included in the CRISPR screen.
  • Dataset: dlbcl_duke_2017.
  • Assays: whole-exome sequencing (Agilent All Exon V5, ~75X coverage on Illumina HiSeq 2500), RNA-seq, MuTect v1.1.4 somatic variant calling, BWA mem alignment to hg19, EXCAVATOR for CNVs, ANNOVAR annotation, FISH for MYC/BCL2 translocations, IHC (IRF4, BCL6, CD10), and Sanger sequencing for variant validation (1130 events, 61 genes, 90% concordance).
  • CRISPR screen: GeCKO v2 genome-wide library (~120,000 sgRNAs, 19,050 protein-coding genes) in three replicates across six cell lines — three ABC DLBCL (LY3, TMD8, HBL1), two GCB DLBCL (SUDHL4, Pfeiffer), one Burkitt-like (BJAB). 14 population doublings between early and late timepoints; analyzed with MAGeCK-VISPR.
  • Standard treatment regimen: rituximab-containing immunochemotherapy (de facto R-CHOP).

Key findings

  • Identified 150 recurrent driver genes in DLBCL (mean 7.75 mutations per case), of which 27 (including SPEN, KLHL14, and MGA) had not been previously implicated in DLBCL.
  • Genes CDKN2A and RB1, historically described as primarily copy-number altered, were also recurrent targets of function-altering point mutations.
  • 20 driver genes were differentially mutated between ABC and GCB subtypes: EZH2, SGK1, GNA13, SOCS1, STAT6, and TNFRSF14 enriched in GCB; ETV6, MYD88, PIM1, and TBL1XR1 enriched in ABC.
  • Pairwise overlap analysis using Fisher’s test and mutual exclusion (Leiserson 2016) found 61 of 150 driver genes had statistically significant relationships with other drivers (p<0.01). Notable mutually exclusive pairs include KMT2D (MLL2) vs MYC, and TP53 vs KLHL6.
  • CRISPR screen identified 1956 essential genes across at least one cell line. Driver genes from WES were disproportionately enriched at the extremes of the CRISPR-score distribution (p=3×10⁻⁵).
  • 35 driver genes were functional oncogenes (knockout reduced viability); top oncogenic dependencies across most DLBCLs included MYC, RHOA, SF3B1, MTOR, and BCL2. Knockout-enriched (tumor-suppressor) hits included TP53, MGA, PTEN, and NCOR1.
  • Subtype-specific essentiality: knockout of EBF1, IRF4, CARD11, MYD88, and IKBKB was selectively lethal in ABC DLBCL; knockout of ZBTB7A, XPO1, TGFBR2, and PTPN6 was selectively lethal in GCB DLBCL.
  • 9 of 35 CRISPR-validated driver genes are direct targets of drugs in clinical trials or already approved for other indications; 36% of DLBCL patients harbor genetic alterations in these 9 drug targets.
  • Integrative gene-set analysis (KEGG, Reactome, MSigDB plus lymphoma-specific signatures from Lenz 2008, Monti 2005, Shaffer) reduced ~9500 gene sets to 31 non-redundant clusters via affinity propagation. CRISPR hits were enriched in cancer-related processes (oxidative phosphorylation, DNA replication, cell cycle, RNA processing) and depleted in immune/stromal sets.
  • IPI was highly prognostic (p<10⁻⁶); ABC vs GCB cell-of-origin was prognostic (P=0.002 log-rank); MYC/BCL2 dual-high expressors had worse survival; ABC/GCB status did not further stratify dual expressors.
  • FISH analysis: MYC translocations associated with MYC mutations and high MYC expression; BCL2 translocations associated with BCL2 mutations and amplifications.
  • Mutations associated with poorer survival across all DLBCL: MYC, CD79B, ZFAT. Favorable: NF1, SGK1.
  • Within ABC DLBCL: KLHL14, BTG1, PAX5, and CDKN2A alterations associated with poorer survival; CREBBP alterations with favorable survival.
  • Within GCB DLBCL: NFKBIA, NCOR1 alterations associated with poorer survival; EZH2, MYD88, and ARID5B alterations associated with significantly better survival.
  • A multivariate Cox proportional-hazards model combining 150 driver genes with cell-of-origin, MYC, and BCL2 expression — using all combinations of up to 3 markers affecting ≥20 patients — yielded a 3-tier genomic risk classifier validated on an independent 20% test set (p=8×10⁻⁵) and via 5-fold cross-validation × 100 (median log-rank p=8×10⁻⁶).
  • The integrative DNA+RNA combinatorial model outperformed DNA-only, RNA-only, and DNA+RNA-without-combinations models, IPI alone, cell-of-origin alone, and MYC/BCL2 dual-expressor models. The genomic risk model retained prognostic value within IPI strata and within complete-response patients.
  • The IPI was strongest at predicting early mortality; the genomic risk model retained prognostic value for both early and late (≥5-year) mortality.
  • MYC genetic alterations + high MYC expression marked the worst-prognosis subset; GCB DLBCL with CD70 alterations marked the best-prognosis subset.

Genes & alterations

  • MYC — combined genetic alteration plus high RNA expression defines the worst-prognosis DLBCL subset; CRISPR-essential across most cell lines; mutually exclusive with KMT2D (MLL2) mutations.
  • BCL2 — recurrent oncogenic missense and copy gains; FISH translocations correlate with mutations and amplifications; CRISPR-essential dependency; component of dual-expressor poor-prognosis signature.
  • BCL6 — measured by IHC for cell-of-origin (Hans algorithm); recurrent driver.
  • CARD11 — oncogenic missense pattern; CRISPR-essential selectively in ABC DLBCL.
  • IRF4 — oncogenic missense; CRISPR-essential selectively in ABC DLBCL; used in Hans IHC algorithm.
  • SPEN — newly implicated tumor suppressor (truncating mutations / CN loss).
  • CDKN2A — recurrent CN loss plus function-altering point mutations; alterations associated with poorer survival in ABC DLBCL.
  • TNFAIP3 — tumor-suppressor pattern.
  • EZH2 — preferentially mutated in GCB; mutations associated with better survival in GCB.
  • SGK1 — preferentially mutated in GCB; mutations associated with favorable survival across DLBCL.
  • GNA13, SOCS1, STAT6, TNFRSF14 — preferentially mutated in GCB DLBCL.
  • ETV6, MYD88, PIM1, TBL1XR1 — preferentially mutated in ABC DLBCL. MYD88 mutations associated with better survival in GCB; CRISPR-essential in ABC.
  • TP53 — recurrent driver; mutually exclusive with KLHL6; CRISPR-enriched (loss of TP53 increases fitness).
  • KMT2D (formerly MLL2) — recurrent driver, mutually exclusive with MYC mutations.
  • RHOA — CRISPR-essential dependency in DLBCL; associated with proliferation gene-expression signatures.
  • SF3B1, MTOR — top CRISPR-essential dependencies; MTOR mutations are prevalent but those patients tend to have generally good outcomes (especially in GCB).
  • MGA, PTEN, NCOR1 — CRISPR-enriched (tumor-suppressor pattern). NCOR1 alterations associated with poorer survival in GCB.
  • EBF1, IKBKB — ABC-selective CRISPR essentiality (B-cell-development/NF-κB signaling).
  • ZBTB7A, XPO1, TGFBR2, PTPN6 — GCB-selective CRISPR essentiality.
  • CD79B, ZFAT — mutations associated with poorer overall survival.
  • NF1 — mutations associated with favorable survival.
  • KLHL14, BTG1, PAX5 — alterations associated with poorer survival in ABC DLBCL.
  • CREBBP — alterations associated with favorable survival in ABC DLBCL.
  • NFKBIA, ARID5B — survival modifiers in GCB; ARID5B associated with better survival.
  • PIK3R1, PIM2, BTK, CHD8, POU2F2, YY1, H1-4 (HIST1H1E) — driver-gene cluster members across signaling, cell growth, B-cell development, and transcription/translation functional groups.
  • NOTCH2, PIK3CD, JAK2 — recurrent drivers but CRISPR knockout did NOT impair growth, suggesting they may act in early pathogenesis or via non-targetable functions in established DLBCL.
  • RB1 — recurrent point mutations in addition to historically described copy-number losses.
  • CD70 — alterations within GCB DLBCL define the most favorable-prognosis subset.

Clinical implications

  • Risk stratification. A multivariate combinatorial model integrating 150 driver genes with cell-of-origin and MYC/BCL2 expression outperforms IPI, cell-of-origin, and MYC/BCL2 dual-expression scores. The genomic risk model retains prognostic value within complete responders and across IPI strata, identifying high-risk patients who otherwise appear low-risk by clinical criteria.
  • Biomarker deployment. The authors note the model can be applied clinically using existing assays — RNA-based cell-of-origin, MYC and BCL2 expression measurement, plus targeted sequencing of a panel of DLBCL driver genes. An interactive web tool was released at dlbcl.davelab.org.
  • Therapeutic targeting. ~36% of DLBCL patients harbor alterations in one of 9 CRISPR-validated essential drivers that are direct targets of approved or trial-stage drugs, suggesting actionable subsets for targeted therapy.
  • Subtype-directed therapy hypotheses. ABC-selective CRISPR dependencies on CARD11, MYD88, IKBKB, IRF4, and EBF1 reinforce NF-κB pathway targeting (e.g., BTK inhibitors, IRAK inhibitors) for ABC patients. GCB-selective dependence on XPO1, PTPN6, TGFBR2, and ZBTB7A suggests distinct opportunities (e.g., selinexor for XPO1).
  • Caution on naive ontology-based targeting. CRISPR knockout of NOTCH2, PIK3CD, and JAK2 — all recurrently mutated in DLBCL and pharmacologically targetable — did not impair growth, suggesting therapeutic targeting based on mutation alone may fail in DLBCL.
  • Trial-design implication for MTOR. Although MTOR mutations are prevalent (and CRISPR-essential), MTOR-mutant DLBCLs tend to have favorable prognosis (especially in GCB). Trials enrolling relapsed patients may systematically miss the responder population.
  • Treatment baseline. All 1001 patients received rituximab-containing immunochemotherapy as standard of care, providing a uniformly-treated reference cohort for future biomarker discovery.

Limitations & open questions

  • CRISPR functional screen was performed in only six cell lines (3 ABC, 2 GCB, 1 Burkitt-like), which may not capture the full genetic and microenvironmental diversity of the 1001-patient cohort.
  • Expression of many signaling pathways (e.g., PI3-kinase) did not reliably correlate with mutations in pathway components (PIK3R1, PIK3CD, PTEN), suggesting bulk RNA expression from non-malignant infiltrating cells confounds these associations.
  • Lethal effects of driver gene knockout were not restricted to cell lines harboring mutations in those genes, indicating these drivers represent broadly favored proliferative pathways rather than synthetic-lethal vulnerabilities for mutated subsets.
  • The authors note ongoing mortality even among complete responders, implying a substantial fraction of clinically-defined “cured” patients still harbor genomic risk.
  • The genomic model was validated only on a 20% held-out subset of the same cohort plus cross-validation; external multi-institutional validation remains a future need.
  • No direct comparison with other contemporary classifications (e.g., LymphGen genetic subtypes, Chapuy clusters) is provided in this paper.
  • Some recurrently mutated, pharmacologically-targetable genes (NOTCH2, PIK3CD, JAK2) lack functional dependence in DLBCL cells, leaving open whether they act earlier in lymphomagenesis or whether the CRISPR screen missed context-dependent essentiality.

Citations from this paper used in the wiki

  • “we performed an integrative analysis of whole exome sequencing and transcriptome sequencing in a cohort of 1001 DLBCL patients to comprehensively define the landscape of 150 genetic drivers of the disease” (Summary).
  • “We identified 1956 ‘essential genes’ whose silencing resulted in significantly decreased cell fitness in at least one cell line.” (Results, CRISPR screen).
  • “In all, there were 35 driver genes whose knockout resulted in decreased viability of DLBCL cells, identifying them as functional oncogenes.” (Results).
  • “Knockout of EBF1, IRF4, CARD11, MYD88 and IKBKB was selectively lethal in ABC DLBCL whereas knockout of ZBTB7A, XPO1, TGFBR2 and PTPN6 was selectively lethal in GCB DLBCL.” (Results).
  • “Of the 35 CRISPR driver gene hits, 9 genes are direct targets of the therapeutic drug targets either in human clinical trials or already in use for another indication. Importantly, 36% of the DLBCL patients have genetic alterations in these 9 drug targets” (Results).
  • “MYC genetic alterations combined with MYC expression defined the subset with the least favorable prognosis in DLBCLs, while GCB DLBCLs with CD70 alterations had the most favorable prognosis.” (Results, Figure 5D).
  • “The integrative model strongly outperformed the other models based on genetic alterations (‘DNA only’) or expression (‘RNA only’) alone.” (Results, Figure 5F).
  • “while MTOR mutations are prevalent in DLBCL, those patients tend to have generally good outcomes, especially in the GCB subtype” (Discussion).
  • “Of the 150 driver genes that we identified, 27 genes including SPEN, KLHL14 and MGA that have not been previously implicated in DLBCL to our knowledge.” (Discussion).
  • “CRISPR-based knockout of several therapeutically targetable NOTCH2, PIK3CD and JAK2 did not have a significant impact in the growth of DLBCL cells.” (Discussion).

This page was processed by crosslinker on 2026-05-15.