Using Informatics Tools to Identify Opportunities for Precision Medicine in Diffuse Large B-Cell Lymphoma

Abstract

Introduction:

Diffuse large B-cell lymphoma (DLBCL) is genetically and clinically heterogeneous. Despite advances in genomic subtyping, standard frontline chemoimmunotherapy has remained unchanged for years. As high-throughput analysis becomes more accessible, characterizing drug-gene interactions in DLBCL can support patient-specific treatment strategies.

Methods:

Our systematic literature review compiled a comprehensive list of somatic mutations implicated in DLBCL. We extracted published and primary sequencing data for these mutations and assessed their association with signaling pathways, cell-of-origin subtypes, and clinical outcomes.

Results:

Twenty-two targetable mutations present in ≥5% DLBCL patients were associated with unfavorable outcomes, yielding a predicted population of 31.7% of DLBCL with poor-risk disease and candidacy for targeted therapy. A second literature review identified 256 articles that characterized drug-gene interactions for these mutations via in vitro studies, mouse models, and/or clinical trials.

Conclusions:

Our novel approach linking systematic review with informatics tools identified high-risk DLBCL subgroups, DLBCL-specific drug-gene interactions, and potential populations for precision medicine trials.

Microabstract:

Despite advances in genomic subtyping of DLBCL, standard frontline treatment has remained unchanged for years. A novel approach linking systematic review with informatics tools identified high-risk DLBCL subgroups and DLBCL-specific drug-gene interactions. Twenty-two targetable mutations present in ≥5% DLBCLs were associated with unfavorable outcomes, yielding a predicted population of 31.7% of DLBCL with poor-risk disease and candidacy for targeted therapy.

Introduction

Diffuse large B-cell lymphoma (DLBCL) is one of the most common non-Hodgkin lymphomas¹ and exhibits considerable genetic and clinical heterogeneity. While some DLBCL patients will be cured after receiving standard frontline therapy with rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP), ~40% will ultimately die from relapsed disease, even with the advent of molecular profiling and the availability of autologous stem cell transplantation and chimeric antigen receptor T-cell infusions.^2–4 These survival outcomes expose the dire need for management approaches extending beyond the current, uniform treatment approach into a personalized care strategy.⁵

Accordingly, much work has been devoted to investigation of the clinical and molecular factors underlying such disparate outcomes. Nearly two decades ago, conventional gene expression profiling (GEP) approaches classified DLBCL into two major subtypes based on putative cell-of-origin: activated B-cell-like (ABC) and germinal center B-cell-like (GCB), with some patients remaining unclassified.⁶ Studies distinguishing these subtypes by conventional GEP, immunohistochemistry, and novel technologies for performing GEP on paraffin-embedded tissue demonstrate superior progression free survival (PFS) and overall survival (OS) for patients with the GCB subtype.^7–9 In addition, certain genetic aberrations demonstrate preferential distribution among the two subtypes; for instance, somatic mutations in CD79A/B and MYD88 that result in constitutive B cell receptor signaling and NF-κB activation are more likely to be found in ABC-DLBCL, while mutated EZH2 is much more common in GCB-DLBCL.¹⁰

However, efforts to capitalize on these differences in molecular abnormalities and clinical outcomes with addition of targeted therapies to standard front-line chemoimmunotherapy have yet to render viable subtype-specific treatment approaches in DLBCL.¹¹ One key factor that may contribute to the lack of success in subtype-directed therapy thus far may be the substantial genetic heterogeneity observed in DLBCL; even DLBCL’s most commonly mutated genes (e.g., MYD88, BCL2, and CREBBP) are found only in a minority of cases.^12–15 Thus, although cell-of-origin subtype was a reasonable starting point for targeted therapy in DLBCL, improving outcomes for poor-risk patients may require more precise molecular targets.

In the last few years, several high-impact next-generation sequencing efforts have significantly refined and redefined genomic heterogeneity in DLBCL. A study by Reddy et al. published in Cell analyzed whole exome and transcriptome sequencing in a cohort of 1,001 previously untreated DLBCL patients and identified 150 genetic drivers of the disease.¹⁶ Integrative analysis of available clinical data generated a prognostic model in which these genetic alterations outperformed established prognostic markers, including cell of origin, the International Prognostic Index (IPI) score, and dual MYC and BCL2 expression. In addition, these analyses allowed authors to establish a comprehensive prognostic model incorporating genetic driver mutations, clinical factors, and histologic markers to predict outcome in DLBCL patients. In a Nature Medicine paper by Chapuy et al., authors performed whole-exome sequencing in 304 biopsies from newly diagnosed DLBCL patients, using an expanded bait set to characterize structural variants and somatic copy number alterations.¹⁷ Tumors in this series averaged 17 driver gene alterations, with a total of 158 distinct alterations identified. Five distinct categories of DLBCL were identified (C1-C5), with a sixth category, C0, exhibiting no detectable alterations. While C1 and C5 lymphomas were largely comprised of ABC-DLBCL, they differed in terms of genetic abnormalities and prognosis. Similarly, C3 and C4 lymphomas were primarily GCB subtype but harbored distinct genetic alterations and clinical outcomes. C2 contained a mix of cell-of-origin subtypes and was distinguished by frequent alterations of TP53 and/or loss of chromosome 17p, with predictably poor prognosis. In their analysis of 574 DLBCL samples published in New England Journal of Medicine, Schmitz and colleagues performed whole-exome, targeted amplicon and transcriptome sequencing and mapped specific combinations of genetic alterations onto cell-of-origin subtypes as defined by GEP.¹⁸ This work identified four novel subgroups of DLBCL accounting for 44.8% of cases in the series: MCD, based on MYD88^L265P and CD79B co-mutation; BN2, based on BCL6 fusion and NOTCH2 mutation; N1, based on NOTCH1 mutation; and EZB, based on EZH2 mutation and BCL2 translocation. MCD and N1 subgroups were highly associated with ABC subtype, while EZB was associated with GCB subtype, thus highlighting the cellular pathways most engaged by each subtype.

The recent availability of data from genomic studies involving large cohorts of DLBCL patients with survival data provides new opportunities to develop targeted treatment strategies based on variations in tumor biology. In this study, we explored an alternate method to hierarchical clustering for linking COO subtype with subtype-specific genetic alterations by defining individual DLBCL mutations that have been associated with molecularly targeted therapies in prior preclinical studies or early-phase trials. To better define opportunities for precision medicine approaches in DLBCL, we integrated systematic literature reviews with next-generation sequencing informatics and survival analyses to characterize DLBCL mutations associated with both poor clinical outcomes and currently available targeted therapies.

Material and Methods

Systematic Review

Studies providing genomic and exomic data on somatic mutations implicated in DLBCL were identified in the MEDLINE database through a series of searches using the medical subject heading (MeSH) terms “Lymphoma, Large B-Cell, Diffuse” and “genetics”, and a general text search of “exome”. The search was expanded with additional related terms and references identified from the retrieved articles. Studies analyzing genomic or exomic data for genetic aberrations implicated in DLBCL were included.

After assembling a comprehensive list for the review, we extracted published and primary sequencing and time-to-event data when available for identified genetic aberrations as well as these mutations’ association with dysregulated molecular pathways, cell-of-origin subtypes, relapsed and/or refractory status at time of tumor specimen collection, and OS. We also assessed these mutations’ frequency in the DLBCL patient cohorts analyzed by four recent high-impact exome sequencing papers.^16–19 With the exception of 3.5% (n=20) relapsed cases included in the Schmitz et al. data set, these four manuscripts characterized mutations exclusively in de novo DLBCL.

Subtype-Specific Mutational Frequencies

DLBCL subtype was classified as ABC, GCB, primary mediastinal B-cell, primary central nervous system lymphoma, leg type, or unclassified, as defined in each publication. Data for overall, GCB-specific, and ABC-specific prevalence of mutations were collected from the four recent high-impact publications of DLBCL exome sequencing when available.

Prognostic Mutation Categories

Sequencing data were used to define prognostic groups using mutation frequencies and estimated 3-year OS derived from the DLBCL survival estimation algorithm available at https://dlbcl.davelab.org.¹⁶ Mutations were categorized as either favorable, unfavorable or equivocal: favorable mutations were associated with 3-year OS ≥ 81.1% (the lower bound of the 95% confidence intervals for OS in patients with IPI score ≤1 in the Reddy dataset), whereas unfavorable mutations were associated with 3-year OS ≤ 70.7% (the upper bound of the 95% confidence intervals for OS in patients with IPI score ≥ 2 from this dataset); equivocal mutations were associated with 3-year OS between 70.7–81.1%.

Identification of Candidate Mutations for Targeted Therapy

Potential therapeutic targets were identified based on searches of the Drug-Gene Interaction database (DGIdb v3, http://www.dgidb.org),²⁰ and the Open Targets Platform (OTP, https://www.targetvalidation.org).²¹ Using these informatics tools, we screened each mutation identified from the initial literature review for association with specific therapeutic agents, thus assessing for candidate mutation/therapy pairs that could be used in precision medicine strategies. Narrowing our search to mutated genes found in ≥5% of DLBCL patients, we created a final list of candidates for pharmacogenetic targeting by selecting therapies that targeted at least one unfavorable mutation, as defined by the aforementioned criteria (Figure 1).

Figure 1.

Selection of candidate therapies for precision medicine strategies in DLBCL.

Specific drug-gene interactions from this narrowed list were characterized using a series of Pubmed searches performed for each drug-gene interaction using several combinations of the following terms: “[drug name]” “[mutated gene]” “Cancer,” “Lymphoma,” “Diffuse Large B-cell Lymphoma.” Each abstract retrieved was read to identify pertinent studies that characterized drug-gene interactions in cancer, lymphoma, or DLBCL, including in vitro assays, animal models, and clinical trials. Case reports, studies without access to full papers, and studies not involving a cancer were excluded (Supplemental Figure). Full manuscripts for the remaining studies were read to identify overall response rate (ORR) and complete response (CR) rate from relevant clinical trials. We also extracted data on associated molecular pathway dysregulation, mechanisms of action, and implications for tumor growth inhibition from relevant preclinical studies.

Finally, we predicted a population of DLBCL patients who would be potential candidates for clinical trials with molecularly targeted therapies. Specifically, we identified the proportion of patients in the Reddy dataset with ≥1 mutation associated with unfavorable OS and an identified, molecularly targetable therapy.

Results

Systematic Review

Our initial systematic literature review identified 32 studies with a total of 360 mutations implicated in DLBCL (Supplemental Table 1). Affected molecular pathways included those involved in apoptosis, immune evasion, angiogenesis, epigenetic regulation, and sucrose/tryptophan degradation (Table 1).

Table 1.

Major pathways and associated mutations identified in a systematic review of genetic alterations implicated in DLBCL.

Pathway	Mutations
Apoptosis	INPP5D, PIK3R1, CDC73, CUL4B, DNMT3A, RUNX1, BCL10, BCL2, MYBL1, FUT8, TOX, GNA12, GRHPR, HIST1H1B/C/D
Epigenetic regulation	HRAS, HDAC9, ENTPD1, KLHL6, HIST1H1E, HIST1H2BC, CREB3L2, ETV6, MECOM, MIR17HG, MLH1, MSH2, MSH6, WAC
NF-κB signaling	TBC1D4, NFKB2, CD44, XBP1, TP53, BTK, PRDM1, SYNE1, TMSB4X, YY1, NFKBIA, MYC
BCR/PI3K signaling	CD70, POU2F2, GNA12, IDO1, MPEG1, SF3B1, MYC, INO80, miR-17–92, KCNE3, SH3BP5
Immune evasion	HLA-A/B, CCDC50, IKBKB, CD58, CD22, CD79B, ATR, CHST2, TNFRSF13B
JAK/STAT signaling	JAK1, SMARCA4, STAT5B, BMF, SPIB
NOTCH signaling	NFKBIZ, NOTCH1, SOCS1, DTX1
Angiogenesis	EBF1, MLL3, PCDH17
Sucrose degradation	KCNE3, SH3BP5
Tryptophan degradation	HSK1, TCL1A

Association of specific mutations with cell-of-origin subtype varied across papers, and few mutations were restricted to one subtype. Eighty mutations were associated with relapsed/refractory status, based on the sequencing of relapsed tumors.^22–29 Of those, 40 (50%) occurred only in relapsed/refractory DLBCL samples, whereas others were associated with both previously untreated and relapsed/refractory DLBCLs.

Subtype-Specific Mutational Frequencies

Mutational frequencies of 150, 309, and 21 DLBCL driver genes were extracted from the Reddy, Schmitz, and Arthur datasets, respectively (Supplemental Table 2).^16,18,19 These datasets also included data on mutation prevalence in ABC- and GCB-DLBCLs. The Chapuy dataset¹⁷ included overall frequency data for 74 genetic aberrations but no data on mutation prevalence by cell-of-origin subtype (Table 2).

Table 2.

Mutation frequency across recent major sequencing studies in DLBCL. Abbreviations: ABC, activated B cell-like subtype; GCB, germinal center B cell-like subtype.

Dataset	Reddy (2017)	Schmitz (2018)	Arthur (2018)	Chapuy (2018)
Number of Mutations	150	300	277	74
Mean Mutation Frequency (Overall)	4.7%	3.9%	4.4%	8.8%
Mean Mutation Frequency among ABC Cases	6.9%	3.9%	-	-
Mean Mutation Frequency among GCB Cases	6.2%	3.9%	-	-

Prognostic Mutation Categories and Possible Precision Medicine Candidates

Informatics tools were used to identify possible precision medicine candidates. Table 3 lists therapeutic agents based on their pharmacology mechanism and mutational frequency of identified drug targets. Mutations that could be targeted by BTK inhibition (e.g., by ibrutinib) had the highest cumulative frequency, with 20% of patients in the Reddy dataset harboring mutations in BTK and/or MYD88. Our Medline searches to characterize drug-gene interactions for mutations of interest yielded 256 pertinent articles, including 97 clinical trials and 163 preclinical studies (in vitro studies and/or mouse models; Supplemental Table 3). Fifteen studies investigated potential therapeutic targets of these mutations in DLBCL (13 preclinical studies and 2 clinical trials, Table 4). There were 83 mutations associated with at least one type of targeted therapy. Figure 2 depicts the overlap of mutations that are associated with targeted therapies and unfavorable survival outcomes. In the Reddy dataset, 23 mutations were associated with favorable outcomes with an expected 3-year OS of ≥81.1%, and 52 mutations were associated with unfavorable outcomes with an expected 3-year OS of ≤70.7%. Twenty-two of the mutations associated with unfavorable outcomes were also targetable. Based on these results, we predicted that 31.7% of the DLBCL population harbor mutations associated with both unfavorable prognosis and the potential for targeted therapy.

Table 3.

Candidate targeted therapies and associated mutations in DLBCL. Candidate targeted therapies were identified in the Drug-Gene Interaction database and the Open Targets Platform by searching for association with unfavorable mutations occurring in ≥ 5% DLBCL patients. Cumulative mutation frequency denotes frequency of cases with any mutation or combination of mutations associated with the given therapy as observed in the Reddy dataset (n = 1,001).

Mechanism of Action	Therapeutic Agent	Mutation(s)	Cumulative Mutation Frequency
RAR agonism	Alitretinoin Etretinate Tazarotene	PTPRK, RARA	11%
	Tretinoin	PTPRK, SMARC4A	5%
	Adapalene Acitretin Isotretinoin NRX195183	PTPRK	3%
VEGF inhibition	Bevacizumab	TP53	10%
BTK inhibition	Ibrutinib	BTK, MYD88	20%
	ONO-4059 HM-71224 Acalabrutinib Spebrutinib MSC-2364447	BTK	3%
ATR inhibition	AZD-6738 Berzosertib	ATR	4%
Alpha-glucosidase inhibition	Miglitol Acarbose	MGA	8%
MET inhibition	Crizotinib Capmatinib	MECOM, MET	7%
CDK4/6 inhibition	Palbociclib	CCND1, CCND3, CDKN2A, RB1	17%
PI3K / Casein kinase II inhibition	LY-294002	PIM1	17%
PKC inhibition	GF-109203
PARP inhibition	Olaparib	ATM, ATR, BRCA1, P2RY8, PARP2, PTEN	13%
DNA methyltransferase inhibition	Decitabine	DNMT3A, TET2	11%
Wee1 kinase inhibition	AZD-1775	TP53	10%
PIM kinase inhibition	SGI-1776	PIK3CD, PIM2	9%
NOTCH inhibition	Tarextumab	NFKBIZ, NOTCH2	8%
NTRK inhibition	Cabozantinib	MECOM, MET	7%
Nitric oxide donation	Isosorbide	MCL1	6%
ROCK inhibition	Fasudil	CBLB	2%
TRK inhibition	Larotrectinib	ETV6	4%
ALK/TRK inhibition	Entrectinib
JAK inhibition	Tofacitinib	JAK3	4%
BRAF inhibition	Vemurafenib	BRAF	2%

Table 4.

Published studies of drug-gene interactions in DLBCL. (A) Published preclinical studies of drug-gene interactions in DLBCL. Abbreviations: CLL, chronic lymphocytic leukemia; FL, follicular lymphoma; MCL, mantle cell lymphoma. (B) Published clinical trials involving DLBCL participants selected by targeted mutation status. Abbreviations: R/R, relapsed/refractory; ORR, overall response rate; CR, complete response rate.

A.
Author	Year	Drugs Tested	Implicated Genes	Cell Lines?	Mouse Model?	Comments
Liu Z	2018	Crizotinib, Ricolinostat	HDAC6	-	Yes
Ren L	2016	Ibrutinib	BTK	CLL	Yes
Mondello P	2017	Ibrutinib, Panobinostat	MYD88	-	Yes
Esteve-Arenys A	2018	Venetoclax, CPI203	MYC, BCL2	-	Yes	Limited to double-hit lymphomas
Ravà M	2018	Venetoclax, Tigecycline	MYC, BCL2	-	Yes	Limited to double-hit lymphomas
Klanova M	2016	Venetoclax, Homoharringtonine, ABT-737	BCL2, MCL1	-	Yes
Li L	2015	Venetoclax, Dinaciclib	BCL2, MCL1	-	Yes
Barr PM	2012	Fostamatinib	SYK	FL, MCL	-
Yang G	2016	Ibrutinib, A419259	MYD88	-	-
Chen JG	2018	Ibrutinib, IL-6/IL-10 Antibodies	MYD88	-	-
Sun Y	2018	Ibrutinib, P13I	BTK	-	-
Yu H	2018	Tirabrutinib	BTK	CLL	-	Development of laboratory assay for use in clinical trials

B.
Author	Year	Genes of Interest	Drugs Tested	R/R	ORR	CR	Sample Size (n)	Comments
Friedberg JW	2010	SYK	Fostamatinib	Yes	22%	4%	23
Jaglowski SM	2015	BTK	Ibrutinib, Ofatumumab	No	33%	0%	71	Included multiple lymphoma subtypes

Figure 2.

Potentially targetable mutations associated with unfavorable survival in DLBCL. (A) Intersection of mutations in the targeted therapy cluster and in the unfavorable outcome cluster (expected 3-year overall survival [OS] < 70.7%), occurring in ≥ 5% DLBCL. Mutations implicated in DLBCL were compiled from a systematic literature review and were considered targetable with currently available therapies based on queries of the Drug-Gene Interaction database and the Open Targets Platform. (B) Center panel and colored bars on the right represent prevalence of genetic alterations associated with unfavorable 3-year OS and targeted therapies in the Reddy et al. study population. Grey bars on the left indicate weighted prevalence of these genetic alterations across the Reddy, Schmitz, and Chapuy datasets. Abbreviations: SNV, single nucleotide variant.

Discussion

Our study helps to establish a foundation and framework for developing precision medicine strategies for DLBCL treatment using a novel, combinatorial informatics approach. Our systematic review compiled a comprehensive list of mutations implicated in DLBCL, and we utilized extracted OS data from a large sequencing study to establish prognostic subgroups defined by mutation status. Next, we examined unfavorable mutations occurring in ≥5% DLBCL patients, using two bioinformatics platforms to identify mutations associated with candidate therapeutic agents. Finally, to help inform the development path for precision medicine in DLBCL, we characterized relevant drug-gene interactions for candidate therapeutic targets to determine the current understanding of the activity of these specific agents in DLBCL and other cancers.

Our approach seeks to address a key issue in DLBCL treatment: namely, that current subtype-specific therapies have not yet proved sufficient as a personalized medicine strategy. For instance, although phase II clinical trial results suggested that adding the proteasome inhibitor bortezomib to R-CHOP (VR-CHOP) could improve poor outcomes in non-GCB-DLBCL patients, data from a subsequent randomized trial demonstrated no difference in OS between R-CHOP and VR-CHOP in this population.³⁰ A randomized trial examining the addition of the Bruton tyrosine kinase (BTK) inhibitor ibrutinib to R-CHOP has also been reported but not published, and shows no survival benefit. Subtype-specific studies adding lenalidomide or the second-generation proteasome inhibitor carfilzomib to R-CHOP are ongoing, but rational design of trials that take into account the genomic heterogeneity of DLBCL and its implications for clinical outcomes is needed.^5,31 DLBCL subgroupings using genomic clustering have been proposed by two groups.^17,18 Here we present an alternative approach. By linking systematic review with informatics tools, we have identified 22 potentially targetable mutations that are associated with unfavorable survival outcomes. Using these mutations as biomarkers could inform selection of targeted agents for study in DLBCL and identify poor-risk DLBCL patients most likely to benefit from trials investigating combinations of such agents with frontline chemoimmunotherapy.

From the group of 22 potentially targetable mutations that our analysis identified as associated with both unfavorable survival, MYC, BTK, MCL1, TP53, and DNMT3A are of particular interest. MYC, BTK, and MCL1 are oncogenes that accelerate tumor growth,^32–34 and TP53 and DNMT3A are tumor suppressor genes generally inactivated during tumorigenesis.³⁵

In normal cells, MYC regulates a large variety of cellular functions (including proliferation, protein synthesis, and metabolism) and microRNAs with roles in tumorigenesis. While MYC translocation is associated with aggressive lymphomas and unfavorable prognosis, dysregulated MYC alone does not cause lymphomagenesis.³⁶ In DLBCL, MYC is most frequently translocated with co-occurring rearrangements in BCL2 and/or BCL6, commonly called double-hit lymphoma. Preclinical models have investigated the role of combination therapies targeting double-hit lymphomas in murine models. Rava et al. showed synergistic antitumoral effects of antibiotics (tigecycline, tetracycline, and doxycycline) in combination with venetoclax, a potent BCL2 inhibitor, in treatment of double-hit lymphoma cells.³⁷ In addition, the BET bromodomain inhibitor CPI203 has been shown to overcome double-hit lymphoma cells’ resistance to venetoclax when the two drugs are administered in combination.³⁸ Other strategies in development target MYC more indirectly: e.g., by decreasing MYC transcription through BRD4, by inhibiting the kinases and deubiquitinases that lead to MYC’s evasion of the ubiquitin-proteasome system, or by selective interruption of MYC’s activation partner, Max.³⁹ Overall, more work is needed to identify therapies that overcome the poor prognosis associated with MYC translocation and overexpression in DLBCL.

BTK facilitates normal peripheral B cell survival via B cell receptor (BCR) signaling, and dysregulated BTK signaling is an especially important mechanism of pathologic survival in many ABC-DLCBLs.³³ Preclinical studies and clinical trials have been used to justify BTK inhibition as a potential therapeutic strategy in DLBCL. Importantly, Wilson et al identified that response to the BTK inhibitor ibrutinib in ABC-DLBCL was dependent on the mutation statuses of several members of the BCR and MYD88 pathways, including CD79A/B, CARD11, and MYD88.⁴⁰ Ren et al. demonstrated that ibrutinib could synergistically kill cancer cells with anti-CD20 monoclonal antibody therapy by suppressing FcγR-mediated cytokine production.⁴¹ Jaglowski et al. presented evidence from a phase 1b/2 study that ibrutinib and ofatumumab worked synergistically to achieve an overall response rate of 1 out of 3 (33%) chronic lymphocytic leukemia (CLL) patients who had a Richter’s transformation to aggressive lymphoma.⁴² In addition, clinical trials investigating the addition of ibrutinib to frontline therapy for DLBCL in selected populations are ongoing. While our study focused on identifying candidate therapies based on the presence of individual mutations, future work to develop therapeutic strategies in DLBCL must also take into account how interactions between concomitant genetic abnormalities impact sensitivity to targeted agents, as exemplified by mutation-dependent BTK inhibitor sensitivity in ABC-DLCBL.⁴³

MCL1 regulates mitochondrial homeostasis as a member of the antiapoptotic BCL2 family. It is a frequently amplified gene in several cancers, and has been shown to reduce cell sensitivity to a variety of chemotherapeutic agents.⁴⁴ Preclinical models have indicated some efficacy for targeting co-expressed MCL1 and BCL2 with various therapies.^45,46 Klanova et al. effectively targeted MCL1 and BCL2 using venetoclax and homoharringtonine in vitro and in vivo. Moreover, MCL1-specific BH3-mimetics (e.g. AZD5991) have shown recent preclinical activity in multiple myeloma and acute myeloid leukemia, indicating that MCL1 may be a rational target in DLBCL.⁴⁷ In the clinical setting, venetoclax has quickly moved into clinical trials for relapsed/refractory DLBCL following success in treatment of CLL and mantle cell lymphoma.

TP53 regulates cell-cycle arrest, DNA repair, apoptosis, and autophagy, and its mutated form contributes to lymphomagenesis by dysregulating these pathways. Patients with TP53 mutations demonstrate worse survival than those without, in DLBCL and in many other malignancies.⁴⁸ DNMT3A is another prominent tumor suppressor gene that interferes with DNA methylation epigenetic regulation when mutated.⁴⁹ Some work has been done to elucidate DNMT3A’s role as a negative prognostic factor in DLBCL, but further study is needed.⁵⁰

The four large sequencing studies included in our analyses defined mutations almost exclusively in de novo DLBCL. While targeted therapy has the potential to improve outcomes in frontline treatment of DLBCL, precision-medicine approaches are also needed for the significant proportion of DLBCL patients who relapse after initial therapy. Several groups have identified prevalent mutations in relapsed and/or refractory (R/R) DLBCL. Notably, Melchardt et al. performed whole-exome sequencing in 28 R/R DLBCLs and described three clonal evolutional patterns over time: large global change, subclonal selection, and minimal-to-no change that could suggest primary resistance. In light of these distinct patterns and the finding that having fewer non-synonymous mutations was significantly correlated to a better median OS, the authors concluded that resequencing for new insights after disease relapse is a rational clinical decision.⁵¹ In another study, Nijland et al. applied whole-exome sequencing to paired DLBCL tumors prior to treatment and at relapse. They identified 264 mutations (including PIM1, SOCS1, and MYC) that could potentially be related to therapy resistance. A minority of the mutations detected in the pretreatment biopsy (median 7.6%) were not detected in the matching relapse sample. Conversely, relapsed biopsies exhibited a mild increase in mutations (median 12.5%) when compared to primary tumor, with novel mutations representing a significant driver of the observed difference.⁵²

Our study identifies opportunities for precision medicine in lymphoma management, but several other obstacles currently hinder implementation of precision oncology into clinical trials and practice. Lack of standardized sequencing methods and platforms may limit comparison of data collected across institutions and studies. Multi-center collaborations that utilize common pipelines for tumor sequencing and integration of clinical data allow for more comprehensive understanding of how genomic alterations influence outcomes in this heterogeneous disease. An emphasis on bioinformatics and computational methods to analyze cross-talk between pathways within gene groups will be crucial for increasing drug efficacy and combating drug resistance mechanisms. In addition, how clonal evolution and tumor microenvironment may impact DLBCL outcomes remains unclear. Finally, ensuring that treatment can be initiated quickly in this aggressive disease is essential, thus mandating rapid sample analysis in precision medicine clinical trials in DLBCL.^53–56

Potential limitations of this study include many of the problems commonly associated with systematic reviews and meta-analyses, such as publication bias and study selection bias. Analysis of the included papers did not reveal any overt publication biases. Heterogeneous methods for subtyping and analysis across genomic analyses introduced another relevant challenge for our analysis by making standardized conclusions harder to draw. Fortunately, our clustering system minimized the lack of standardization problem by using informatics tools and mutation frequencies only from major publications with large datasets. Moreover, the results generally aligned with other major papers. In terms of discrepancies in mutation frequency between the 4 major sequencing studies, we suspect that these largely stem from differences in sequencing platforms and gene selection bias.

Conclusions

Methods that incorporate clinical and genomic features are needed to determine how individualized treatment should be employed in DLBCL, and this paper defines a genomic basis for this endeavor. Moreover, our study predicts a generally large population that could benefit from precision clinical trials using this strategy. Ultimately, our work identifies important mutations associated with DLBCL progression and specific targetable therapeutic candidates that may inform precision clinical trial opportunities in DLBCL.

Clinical Practice Points.

While some DLBCL patients will be cured after receiving standard frontline therapy, ~40% will ultimately die from relapsed disease. Several recent, high-impact next-generation sequencing efforts have significantly refined and redefined genomic heterogeneity in DLBCL, but their findings have yet to be translated into changes to treatment strategy in this disease. We linked systematic review with informatics tools that queried these large sequencing studies to identify high-risk DLBCL subgroups, characterize DLBCL-specific drug-gene interactions, and identify potential populations for precision medicine trials. From the group of 22 potentially targetable mutations that our analysis identified as associated with unfavorable survival, BTK, MCL1, and DNMT3A are of particular interest. Future applications of this mixed-methods approach could include identification of target populations for clinical trials aimed at improving poor-risk DLBCL patients’ response to frontline therapy with novel therapeutic regimens.