Genome-Defined African Ancestry is Associated with Distinct Mutations and Worse Survival in Diffuse Large B-Cell Lymphoma

Abstract

Background.

Significant racial differences are observed in incidence and clinical outcomes of diffuse large B-cell lymphoma (DLBCL) in the United States, but it remains unclear whether genomic differences contribute to these disparities.

Methods.

To understand the influences of genetic ancestry on tumor genomic alterations, we estimated the genetic ancestry of 1,001 previously described DLBCL patients using unsupervised model-based ADMIXTURE applied to exome sequencing data and examined the mutational profile of 150 DLBCL driver genes in tumors from this cohort.

Results.

Global ancestry prediction identified 619 patients with >90% European ancestry, 81 with >90% African ancestry, and 50 with >90% Asian ancestry. Compared with DLBCL patients with European ancestry, patients with African ancestry were >10 years younger at diagnosis and were more likely to present with B symptoms, elevated serum lactate dehydrogenase, extranodal disease, and advanced stage disease. Patients with African ancestry exhibited worse overall survival compared to patients with European ancestry (median 4.9 years vs 8.8 years, p=0.04). We found recurrent mutations of MLL2, HIST1H1E, MYD88, BCL2, and PIM1 across all ancestry groups, suggesting shared mechanisms underlying tumor biology. We also identified six DLBCL driver genes that were more commonly mutated in patients with African ancestry compared to patients with European ancestry: ATM (21.0% vs 7.75%, p<0.001), MGA (19.7 vs 5.33%, p<0.001), SETD2 (17.3% vs 5.17%, p<0.001), TET2 (12.3% vs 5.82%, p=0.029), MLL3 (11.1% vs 4.36%, p=0.013), and DNMT3A (11.1% vs 4.52%, p=0.016).

Conclusions.

Distinct prevalence and patterns of mutation highlight an important difference in the mutational landscapes of DLBCL arising in different ancestry groups. Our work provides the first-ever characterization of genetic alterations among DLBCL patients of African descent.

Precis:

In a cohort of 1,001 DLBCL cases, genome-defined African ancestry was associated with distinct driver mutations in ATM, MGA, SETD2, TET2, MLL3, and DNMT3A in DLBCL tumors. Patients with African ancestry were more likely to present with B symptoms, elevated serum LDH, extranodal disease, and advanced stage disease, and had inferior overall survival compared to patients with European ancestry.

Introduction

Diffuse large B-cell lymphoma (DLBCL) is the most common subtype of non-Hodgkin lymphoma (NHL) in the United States, comprising approximately one-third of lymphomas in adults.¹ The clinical course of DLBCL can be aggressive, with a median survival of less than one year in untreated patients.² With standard immunochemotherapy regimens such as rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP), 75% of patients achieve complete remission, and 64% of those individuals are alive and disease-free at 10 years.³ Despite these high cure rates, DLBCL outcomes are heterogeneous, and 30–40% of patients are either refractory to or relapse after first-line treatment and have a lower likelihood of long-term survival. In particular, individuals who relapse within 24 months of diagnosis or who have a short diagnosis-to-treatment interval have worse outcomes.^{4, 5} Accordingly, much work has been devoted to characterizing the biologic, clinical, and environmental factors that underlie this heterogeneity of DLBCL.

The genomic landscape of DLBCL has been robustly studied through integrative analyses of genomics and transcriptomics. Gene expression profiling of DLBCL identified molecularly distinct DLBCL subgroups defined by cell-of-origin: activated B-cell-like (ABC), germinal center B-cell-like (GCB), and primary mediastinal DLBCL^{6, 7} that have been recapitulated by immunohistochemistry^8–10 and have divergent clinical outcomes. Three-year overall survival (OS) is approximately 80% for GCB-DLBCL and 30–40% for ABC-DLBCL.^{8, 11, 12} Several recent landmark studies have defined genetic drivers, clarified their functional roles, identified patterns of abnormalities associated with GCB- and ABC-DLBCL, and linked these with clinical outcomes.^12–15 However, no analysis has provided detailed information on distribution of genomic alterations or DLBCL subtypes by race or ancestry.

While racial differences in the incidence of lymphoma are well known,¹⁶ and many groups have contributed to a developed understanding of disparities in cancer outcomes for solid tumors,^{17, 18} our group has systematically examined disparities in the presentation and outcomes of lymphoma.^19–21 In two national cancer datasets, the Surveillance, Epidemiology and End Results registry (SEER) and the National Cancer Database (NCDB), we discovered that African American patients in the United States with DLBCL were diagnosed 10 years younger than whites, were significantly more likely to have advanced stage disease, and were less likely to be insured, less likely have received standard of care therapy, and less likely to be alive at 5 years (38% vs 46%).^{19, 21} Studies of DLBCL patients from two academic centers that collected detailed information on administration of chemotherapy and immunotherapy found no racial differences in treatment, but African American patients still experienced inferior survival.²² Given the established biological subtypes of DLBCL described above and their association with disparate survival outcomes, we hypothesized that differences in the prevalence of ABC-DLBCL between African American and whites could underlie racial disparities in outcomes, and that recently described gene mutations associated with ABC-DLBCL would more common in African American patients. Pilot data from our two-institution study (n=89) suggested racial differences in the prevalence of non-GCB DLBCL by immunohistochemistry (64% in African American vs 37% in white, p=0.01).^{23, 24} However, no studies have taken into account genome-based ancestry, mutation patterns, and outcome in DLBCL patients.

Admixture mapping is a powerful genome association strategy for investigating potential genetic factors that contribute to racial difference in complex diseases. It evaluates the relationships between different ancestral populations and allele frequencies of multiple single nucleotide polymorphisms. Further analyses can also distinguish chromosomal segments that denote distinct ancestry via statistical enumeration of ancestry informative markers, with the rationale that the genetic variation associated with pathogenesis of a disease could present more frequently on chromosome segments derived from a parental population with a higher incidence of that disease.²⁵ Admixture mapping analyses may be particularly useful in DLBCL given that the incidence of B-cell lymphoma is markedly lower in African Americans compared with Americans of European ancestry,²¹ and since African Americans in the United States represent a highly admixed population involving African, European, Native American, and Asian ancestry.²⁶

The purpose of this study is to identify ancestry-specific cancer driver gene mutations and to describe the genomic landscape of DLBCL across ancestry groups.

Methods

DLBCL data

DLBCL whole exome sequencing (n=1,001) and RNA sequencing (n=775) datasets were downloaded from European Genome-Phenome Archive (EGA: EGAS00001002606) from the genomic analysis performed by Reddy and colleagues.¹² This dataset involved patients who were uniformly treated with standard-of-care therapy with R-CHOP and includes clinical data. Clinical information was obtained from Reddy et al supplementary materials, including age, gender, serum lactate dehydrogenase (LDH), disease stage, ECOG performance status, number of extranodal sites of disease, B symptoms, International Prognostic Index (IPI) score, RNA-sequencing-based molecular subtype (GCB, ABC, unclassified), genomic risk model, initial response to therapy, date of last follow-up, and date of death.²⁷ Data that could be used to calculate disease-free survival, progression-free survival, or event-free survival was not available from this dataset. Of the 1,001 patients included in the dataset, 159 patients were from Emory University. Self-reported race of selected Emory University patients was available from institutional medical records and collected under an IRB-approved protocol (n=63; 42 African-American, 18 non-Hispanic white, and 3 Hispanic).

Global ancestry estimation

We applied unsupervised model-based ADMIXTURE²⁸ global ancestry analysis with K=3 postulated ancestral populations to identify global ancestry groups using exome sequencing data. We used K=2–12 and cross-validation error method implemented in ADMIXTURE to estimate the most preferable K. We categorized patients into one of seven genetically-defined ancestry groups: >90% European ancestry, >90% African ancestry, >90% Asian ancestry, 50–90% European ancestry, 50–90% African ancestry, 50–90% Asian ancestry, and unclassified. We utilized the 63 Emory patients with known self-reported race to describe the overlap between self-reported African ancestry and African ancestry determined by genomic assessment of ADMIXTURE.

Statistical analysis

Differences in baseline characteristics at diagnosis across different ancestry groups were analyzed using pairwise analyses (African vs European; Asian vs European). Univariate and multivariable Cox proportional hazards models were used to determine predictors for survival and included the following variables: ancestry, age, gender, LDH, stage, performance status, number of extranodal sites of disease, B symptoms, IPI score, and RNA-sequencing-based molecular subtype. OS curves were constructed using the Kaplan-Meier method and were compared by a two-sided log-rank test. A level of significance of 0.05 was defined as statistically significant. All statistics were computed using R statistics software (R 3.5.1).

Genetic drivers in DLBCL patients with African ancestry

After stratifying patients by ancestry groups as determined by ADMIXTURE, we examined mutational frequency of the 150 DLBCL driver genes previously identified in the Reddy et al dataset. Each mutation was further categorized by type of alteration: truncation, missense, frameshift, copy number gain, or copy number loss. Differences in mutational frequency between ancestry groups were analyzed using Fisher’s exact test. We identified putative driver mutations on OncoPrints (CBioPortal)²⁹, based on mutation hotspots³⁰ and OncoKB.³¹

Results

Global Ancestry Estimation

As self-reported race was not part of the clinical information collected in Reddy et al dataset, we performed ADMIXTURE global ancestry on 1,001 DLBCL cases. We identified patients with >90% European ancestry (n=619), >90% African ancestry (n=81), >90% Asian ancestry (n=50), 50–90% European ancestry (n=164), 50–90% African ancestry (n=34), 50–90% Asian ancestry (n=47), and unclassified (n=6; Figure 1). Global ancestry prediction correlated with self-reported race for 96.7% of patients with available data (n=63).

Figure 1. Global ancestry estimation of 1,001 DLBCL patients.

ADMIXTURE plot showing genome-wide ancestry fractions for individuals from admixed populations. Each column corresponds to an individual patient. Abbreviations: EU, European ancestry; AF, African ancestry; AS, Asian ancestry.

Baseline characteristics

The clinical characteristics of DLBCL patients are described in Table 1. The mean age at diagnosis was 62 years (range, 3–93) for patients with >90% European ancestry, compared to 52 (9–83) for patients with >90% African ancestry (p<0.001). A greater proportion of patients with >90% African ancestry presented with B symptoms (p=0.001), elevated LDH (p<0.001), advanced stage disease (p=0.007), extranodal sites of involvement (p<0.001) and B symptoms (p=0.001), as compared to patients with >90% European ancestry. There were no statistically significant differences between these ancestry groups in gender, ECOG performance status, IPI score, or cell-of-origin subtype. Clinical characteristics of patients with >90% European ancestry and >90% Asian ancestry were remarkably similar except for elevated LDH observed in Asian ancestry group (p<0.05). No significant differences were observed with respect to molecular subtypes across all ancestry groups (approximately 40% of patients had ABC-DLBCL).

Table 1. Clinical characteristics by genetic ancestry in 1001 DLBCL cases.

For each variable, cases with missing or unknown data were excluded in calculating the P value.

		Number of patients (%)			P value
Characteristics	All patients N=750	>90% EU N=619	>90% AF N=81	>90% AS N=50	>90% AF vs >90% EU	>90% AS vs >90% EU
Age
Mean	62	63	52	60	<0.001	0.91
≤ 60	318 (44.5)	244 (41.2)	52 (72.2)	22 (44.0)
> 60	396 (55.5)	348 (58.8)	20 (27.8)	28 (56.0)
Gender
Male	420 (56.1)	349 (56.5)	41 (50.6)	30 (60.0)	0.38	0.81
Female	329 (43.9)	269 (43.5)	40 (49.4)	20 (40.0)
LDH
Elevated	363 (54.8)	277 (50.7)	51 (76.1)	35 (71.4)	<0.001	0.008
Normal	299 (45.2)	269 (49.3)	16 (23.9)	14 (28.6)
Stage
I/II	277 (38.1)	242 (40.3)	18 (23.7)	17 (34.0)	0.007	0.47
III/IV	450 (61.9)	359 (59.7)	58 (76.3)	33 (66.0)
ECOG PS
0–2	488 (71.9)	396 (71.4)	51 (68.9)	41 (82.0)	0.77	0.15
3–4	191 (28.1)	159 (28.6)	23 (31.1)	9 (18.0)
>1 Extranodal sites
Yes	166 (23.8)	121 (20.6)	29 (46.8)	16 (34.0)	<0.001	0.05
No	531 (76.2)	467 (79.4)	33 (53.2)	31 (66.0)
B symptoms
Yes	252 (35.9)	205 (35.0)	28 (41.2)	19 (38.8)	0.001	0.71
No	450 (64.1)	380 (65.0)	40 (58.8)	30 (61.2)
Subtype
ABC	239 (40.0)	202 (39.8)	21 (42.9)	16 (40.0)	0.31	0.57
GCB	261 (43.7)	232 (45.7)	17 (34.7)	12 (30.0)
Unclassified	97 (16.2)	74 (14.6)	11 (22.4)	12 (30.0)

Abbreviations: EU, European ancestry; AF, African ancestry; AS, Asian ancestry;ECOG PS, Eastern Cooperative Oncology Group performance status; ABC, activated B-cell-like; GCB, germinal center B-cell like.

Survival Analysis

Figure 2 presents the Kaplan-Meier survival curves stratified by estimated genomic ancestry. DLBCL patients with >90% African ancestry exhibited worse survival compared to patients with >90% European ancestry (median OS 4.9 years vs 8.8 years, p=0.04). The median OS for patients with >90% Asian ancestry was 8.6 years. Table 2 shows univariate and multivariable results of Cox regression analyses of factors predicting mortality. In our multivariable model, age at diagnosis (hazard ratio [HR] 1.91, p<0.001), elevated LDH (HR 1.46, p=0.02), ECOG performance status >2 (HR 1.94, p<0.001), and ABC subtype (HR 1.47, p=0.04) predicted worse survival. OS was worse among patients with >90% African ancestry compared to >90% European ancestry in univariate analyses (HR 1.56, p=0.02), but the strength of this association decreased in the multivariable model (p=0.054).

Figure 2.

Overall survival of DLBCL patients treated with R-CHOP, by ancestry.

Table 2. Cox regression models of mortality predictors in patients with DLBCL.

Covariates with adjusted p values < 0.05 in the univariate model were assessed in the multivariable model, with adjustment for age, ancestry, LDH, disease stage, ECOG PS, B symptoms, and subtype.

	Univariate Model		Multiple Variable Model
Factors	HR (95% Cl)	p-value	HR (95% Cl)	p-value
Age
≤ 60	Reference		Reference
> 60	1.86 (1.42–2.45)	<0.001	1.91 (1.38–2.64)	<0.001
Gender
Female	Reference		Reference
Male	0.98 (0.69–1.18)	0.48
Ancestry
> 90% European	Reference		Reference
> 90% African	1.56 (1.09–2.23)	0.02	1.59 (0.99–2.56)	0.054
> 90% Asian	1.22 (0.80–2.23)	0.34	1.18 (0.72–1.91)	0.506
LDH
Normal	Reference		Reference
Elevated	1.96 (1.47–2.61)	<0.001	1.46 (1.05–2.04)	0.02
Stage
I/II	Reference		Reference
III/IV	1.87 (1.40–2.48)	<0.001	1.36 (0.97–1.95)	0.3
ECOG PS
0–2	Reference		Reference
3–4	2.45 (1.92–3.25)	<0.001	1.94 (1.43–2.66)	<0.001
>1 Extranodal sites
No	Reference		Reference
Yes	1.28 (0.95–1.72)	0.10
B symptoms
No	Reference		Reference
Yes	1.40 (1.08–1.83)	0.01	1.10 (0.81–1.50)	0.55
Subtype
GCB	Reference		Reference
ABC	1.56 (1.13–2.14)	0.005	1.47 (1.04–2.16)	0.04

Abbreviations: HR, hazard ratio; Cl, confidence interval; ECOG PS, Eastern Cooperative Oncology Group performance status; LDH, lactate dehydrogenase.

We also examined the association between ancestry and initial response to therapy. Of patients with >90% African ancestry, 81.3% of patients achieved complete response, 6.7% attained partial response, and 12.0% showed no response. The proportion of patients with >90% European ancestry achieving complete response was 91.8%, and no response was seen in 8.2% of patient (both non-significant).

Genetic Drivers in DLBCL Patients by Global Ancestry

We compared the mutation frequencies of 150 DLBCL driver genes identified in the prior analysis. The 40 most frequently mutated driver genes are shown in Figure 3A. The mean frequencies of mutations in patients with >90% African ancestry and patients with >90% European ancestry were similar (7.5% vs 7.4%, respectively). Patients with >90% Asian ancestry had the highest mutation frequency (9.9%). We found alterations in MLL2, HIST1H1E, MYD88, BCL2, and PIM1 in at least 10% of cases in each ancestry group. We also found genes that were differentially mutated between ancestry groups. ATM, MGA, SETD2, TET2, MLL3, and DNMT3A were mutated more frequently in patients with African ancestry compared to patients with European ancestry (Figure 3B, p<0.05). These six genes accounted for 86 mutations across 57% of patients with African ancestry (p<0.001), 29 mutations across 40% of patients with Asian ancestry (p=0.02), as compared to 172 mutations across 24% of patients with European ancestry. Conversely, patients with African ancestry were less likely to have TP53 (2.5% vs 10.8%, p=0.021) mutations compared to patients with European ancestry.

Figure 3. Frequencies of mutations and copy number variations for top 40 DLBCL driver genes by patient ancestry.

A. Heatmap of genetic alterations across ancestral groups. Columns correspond to tumors and rows correspond to DLBCL driver genes. Individual boxes are colored according to the type of alteration for that gene in that tumor. Genes are ordered by frequency, and mutations are stratified by mutation types. Color bars show gender, International Prognostic Index (IPI) score, cell-of-origin subtype, and genomic risk model. B. Frequencies of gene alterations were compared between DLBCL tumors from European, African, and Asian ancestry using the Fisher exact test. (*p <0.05; **p <0.001). C. Top 20 most frequently mutated DLBCL driver genes by ancestry and type of alteration. Abbreviations: EU, European ancestry; AF, African ancestry; AS, Asian ancestry; ABC, Activated B-cell-like (ABC); GCB, germinal center B-cell-like (GCB).

Figure 4A depicts the heatmap of the six highly mutated genes in patients with African ancestry (n=81) along with corresponding clinical features. The most frequently mutated gene in the African ancestry group was ATM (21% of cases), with a large proportion of missense mutations, including five likely oncogenic mutations. MGA had a Q240 hotspot mutation in the T-box domain. TET2 had four mutational hotspots, three localized in the catalytic domain of its J-binding protein. Mutations observed in SETD2, DNMT3A and MLL3 were missense variants of unknown significance. The distributions of mutations in ATM, MGA, and TET2 are shown in Figure 4B.

Figure 4. The genomic landscape of DLBCL drivers in 81 patients with >90% African ancestry.

A. Mutations in ATM, MGA, SETD2, TET2, DNMT3A and MLL3 observed in 81 DLBCL patients with African ancestry. Frequencies of alteration are expressed as a percentage of all cases. Clinical features of the corresponding patients are indicated below the mutational heatmap, including the International Prognostic Index (IPI) score, response to therapy (complete[CR], partial[PR], or no response[NR]), activated B-cell-like (ABC) versus germinal center B-cell-like (GCB) DLBCL subtype, and gender. B. Stem plot showing distribution of mutations in ATM, MGA, and TET2. Known driver hotspot mutations and oncogenic mutations determined by OncoKB are annotated. X axis indicates amino acid position of each mutation. Y axis indicates number of patients with the same mutation.

We further investigated the distribution of DLBCL driver genes mutations across all ancestry groups (Figure 3C). Genes exhibited a pattern of either predominant missense and/or copy number gains of oncogenes (e.g., BCL2, CARD11) or truncating mutations and/or copy number losses of tumor suppressor genes (e.g., SPEN, TP53). Patients with African ancestry harbored a higher proportion of missense mutations compared to patients with European ancestry (77.7% of total mutations vs 64.9%; p<0.001), and frameshift substitution (8.1% vs 5.4%, p=0.03). In contrast, patients with European ancestry had higher proportions of truncation (11.3% vs 4.2%, p<0.001), copy number gain (14.2% vs 8.4%, p=0.002), and copy number loss (4.1% vs 1.7%, p=0.02) compared to patients with African ancestry.

In our examination of the association of driver gene alterations and OS in patients with African ancestry, none of the individual mutated genes had a prognostic significance. Kaplan Meier analysis showed that alterations in MGA, DNMT3A, and MLL3 showed a trend toward worse OS, but this cohort size was insufficiently powered to establish statistical significance.

Discussion

This study represents a comprehensive effort to examine DLBCL cancer genomics among different ancestral populations, with integrated clinical characteristics, tumor cell-of-origin subtype, outcome, mutations, and ancestry information from whole exome sequencing. We used genetically determined African ancestry rather than self-reported race to examine differences in DLBCL clinical features and outcomes. Our results are consistent with previous epidemiologic studies of DLBCL in that African American patients were diagnosed at a younger age compared to non-Hispanic white patients, and were more likely to present with B symptoms and advanced stage disease.^{21, 22} Our study shows significant differences in survival based on ancestral origin, supporting with prior findings that African American patients with DLBCL have worse OS. However, contrary to our hypothesis, we did not observe any differences in the frequency of ABC-DLBCL by ancestral origin. Although African ancestral origin was a predictor of OS in univariate models, the strength of this association decreased in multivariable models that included factors from the IPI score³² and cell-of-origin subtype, suggesting an interplay between these factors or lack of power to identify significance with this sample size.

Genome-wide association studies (GWAS) have identified common variants that are associated with the risk of various types of cancer.^{33, 34} It is becoming increasingly apparent that host genomic factors such as germline variants can act as expression quantitative trait loci (eQTLs), thereby affecting the expression of cancer driver genes.³⁵ These germline variants may be preferentially found in certain populations, which could potentially explain the differences in disease severity and therapeutic response observed between some population groups. Importantly, however, most of the DLBCL cases examining risk variants have been conducted primarily in European³⁶ and Asian.^{37, 38} Investigation of lymphoma genetics in African Americans has been lacking, and the relationships between variants and risk of DLBCL in African Americans has been unknown. We identified overlapping alterations of MLL2, HIST1H1E, MYD88, BCL2, and PIM1 across ancestry groups, indicating shared oncogenic mechanisms. We also found differentially mutated genes across ancestry groups. The surprisingly lower incidence of TP53 mutations in African ancestry patients suggests the possibility of alternative cellular pathway being of greater importance. These observations underscore the considerable genetic heterogeneity across ancestry groups that may contribute to the observed disparities in the clinical presentation and outcomes of DLBCL.

ATM was the most frequently mutated gene in African ancestry group. ATM codes for a protein critically important in the cellular response to DNA damage, which acts upstream of p53. Mutations in ATM have been reported in B cell chronic lymphocytic leukemia (CLL)³⁹ and mantle cell lymphoma,⁴⁰ and ATM mutations were associated with worse overall and treatment-free survival in CLL patients.³⁹ Although the strongest evidence for association has been found with truncation and splice-junction variants, some studies have identified missense variants leading to depletion of functional ATM protein,⁴¹ and induction of downstream components in the p53 pathway following irradiation.⁴² In the current study, ATM mutations were mainly missense mutations and seemed to be independent of p53 alterations.

Our study identified epigenetic regulators that are associated with DLBCL in the African ancestry group, including histone-modifying enzymes and regulators of DNA methylation. Among these genes, SETD2 and TET2 are of special interest because of their known involvement in myeloid/lymphoid oncogenic pathways. SETD2 codes for a histone H3 lysine 36 trimethyltransferase (H3K36me3), and loss of function of SETD2 has been associated with accelerated tumor cell growth in leukemia.⁴³ Mar et al described the effect of inactivating mutations of SETD2 in conferring drug resistance against DNA-damaging agents in cell lines and in a murine model of AML.⁴⁴ Consequently, treatment with an inhibitor of the H3K26me3 demethylase restored sensitivity to cytarabine, suggesting potential therapeutic avenue to investigate. TET2 is a wells-established tumor suppressor gene in myeloid malignancies and TET2 mutations have been associated with advanced stage disease and worse progression-free survival in AML and T-cell lymphoma.⁴⁵ Several groups have identified rare variants in TET2 that were associated with aggressive prostate cancer in African Americans but were protective in European Americans.⁴⁶ Studies investigating the prevalence of TET2 mutations across cancer in African American populations and the biological significance of these mutations may provide new insights for pathogenesis and therapy.

Most of the six DLBCL driver gene mutations in individuals of >90% African ancestry (ATM, MGA, SETD2, TET2, MLL3, and DNMT3A) were missense, and their functional impact in cancer biology has yet to be established. Future studies are required to establish the role of these missense mutations in tumor development, progression, and differential responses to chemotherapy across ancestry groups. None of these genes had a prognostic impact on OS, suggesting that the integrated analysis of pathways may be more relevant than individual genes. This approach may overcome the challenges of low frequency of these genes.

There are several limitations to this study. First, we predicted each patient’s ancestral information based on whole exome sequencing data. Although ancestral origin was highly correlated with small subset of self-identified race among patients where these data were available, we have limited clinical information on the remaining patients to confirm ancestry groups. Small mismatch between genetic ancestry and self-reported race could be due to limited ability of ancestry informative markers to predict ancestry. However, it is important to recognize that genetic information provides more precise estimates of ancestral origin, while self-reported race encompasses social and environmental dimensions that may be relevant to disease risk. Notably, socioeconomic status and other environmental factors were not available in our dataset. Thus, it is unclear to what extent these factors might have acted as confounders or interacted with genetic abnormalities to impact the observed racial disparities in outcome. Secondly, our evaluation was limited to somatic mutations and copy number variations of 150 previously identified DLBCL driver genes. We cannot rule out the possibility that other genes outside of our selection display substantially different mutational frequencies between patients with African and European ancestry. However, such associations would likely occur only in mutations that are uncommon in DLBCL. In addition, many of the gene variants identified are rare and of unknown function, and the number of patients of African ancestry was relatively small. Sample size may not be adequate to comprehensively characterize racial differences in genetic alterations despite the large size of the 1,001 DLBCL cases as a whole. Larger cohorts with patients with African ancestry will be necessary to identify driver genes and pathways implicated in pathogenesis of DLBCL in this particular population; at the time of our analysis, no cohort with significant numbers of samples or sequencing data from DLBCL patients of African descent was available. Large cohort studies such as the Lymphoma Epidemiology of Outcomes Cohort Study (LEO; NCT02736357) are now underway that may make additional analyses possible. Finally, eQTL analysis could potentially determine the functional significance of the identified genes variants and identify downstream targets of validated susceptibility genes in DLBCL patients.

This study highlights the striking genetic heterogeneity underlying DLBCL across ancestry groups. We have identified certain alterations that are common across ancestry groups, but also found driver genes with significantly different mutational profiles. These findings suggest involvement of different oncogenic genes and pathways in African American populations and constitute possible mechanisms that may contribute to racial differences in incidence, patterns of presentation, and survival. Our work adds to existing literature identifying differences in DLBCL outcomes by race, insurance status, socioeconomic status, clinical, and environmental factors.^{1, 19, 21, 22, 47–49} Larger epidemiologic studies that integrate these factors along with gene expression, copy number, and genome sequencing data are needed to better understand the complex interactions between these variables and their contributions to disparities in DLBCL outcomes. Ideally, such data should be collected from population-based samples to provide a representation of general clinical practice and to avoid skewing toward academic settings that may over-represent patients with poor-risk genomics. Such studies also may require oversampling of ethnic minority groups to adequately evaluate ethnic diversity. The current study provides a model for disentangling factors that contribute to racial differences in outcomes for hematological malignancies and supports the development of interventions that overcome these barriers and improve survival.