Extranodal natural killer/T-cell lymphoma (ENKTL) is an aggressive, rare lymphoma of natural killer (NK) cell origin with poor clinical outcomes. Here we used phenotypic and molecular profiling, including epigenetic analyses, to investigate how ENKTL ontogeny relates to normal NK-cell development. We demonstrate that neoplastic NK cells are stably, but reversibly, arrested at earlier stages of NK-cell maturation. Genes downregulated in the most epigenetic immature tumors were associated with polycomb silencing along with genomic gain and overexpression of EZH2. ENKTL cells exhibited genome-wide DNA hypermethylation. Tumor-specific DNA methylation gains were associated with polycomb-marked regions, involving extensive gene silencing and loss of transcription factor binding. To investigate therapeutic targeting, we treated novel patient-derived xenograft (PDX) models of ENKTL with the DNA hypomethylating agent, 5-azacytidine. Treatment led to reexpression of NK-cell developmental genes, phenotypic NK-cell differentiation, and prolongation of survival. These studies lay the foundation for epigenetic-directed therapy in ENKTL.
Through epigenetic and transcriptomic analyses of ENKTL, a rare, aggressive malignancy, along with normal NK-cell developmental intermediates, we identified that extreme DNA hypermethylation targets genes required for NK-cell development. Disrupting this epigenetic blockade in novel PDX models led to ENKTL differentiation and improved survival.
This article is highlighted in the In This Issue feature, p. 85
Extranodal natural killer/T-cell lymphoma (ENKTL) is an aggressive disease and one of only a limited number of malignancies known to be of natural killer (NK) cell origin. ENKTL is a universally Epstein–Barr virus (EBV)-positive non-Hodgkin lymphoma (NHL) that most frequently affects those of far Eastern Asian and Central/South American heritage. ENKTL typically presents as a nasopharyngeal mass, although it can also involve other extranodal sites such as the gut and skin and is particularly devastating when presenting at relapse or with advanced disease, which can be a common problem in developing nations (1, 2). Despite the implementation of intensive combinatorial chemotherapies, survival is less than 30% for ENKTL patients with advanced disease (1). This highlights the critical need to develop new lines of targeted therapy for the treatment of ENKTL.
Although previous reports have characterized different aspects of ENKTL etiology and presentation (3–8), several key biological aspects of the disease are incompletely understood, in particular the relationship of malignant NK tumor cells to the stage of normal NK-cell development and mechanism(s) underlying the aberrant phenotype. We have previously demonstrated that human NK-cell development occurs within mucosal-associated lymphoid tissues, such as tonsils, and proceeds through functionally distinct stages of NK-cell developmental intermediates (NKDI) that can be distinguished phenotypically according to the differential expression of CD34, CD117, CD56, CD94, NKp80, CD16, and CD57 (9–13). The process of NK-cell development is tightly controlled, involving epigenetic regulation of key genes governing specific NK-cell function, such as DNA methylation of CD16 and KIR genes (14, 15). Whereas previous studies have shown that ENKTL cells bear a CD56brightCD16−KIR− surface immunophenotype suggestive of relative immaturity (5), it is not yet known how disease-specific epigenetic events impact ENKTL gene expression and the NK-cell developmental phenotype.
Epigenetic regulation is known to play an essential role in cell lineage specification and development. Carefully controlled activities of transcription factors (TF) and other chromatin-associated proteins modify the landscape of histones and DNA methylation states to generate stable, cell type–specific global patterns of gene expression (16–18). In the setting of cancer, promoter silencing involving histone modifications, such as gain of H3K27me3 deposited by EZH2 of the polycomb repressor complex (PRC), and DNA hypermethylation can downregulate tumor suppressor genes (17, 19, 20). These two silencing mechanisms are known to cooperate in cancer, as DNA hypermethylation is commonly targeted to polycomb-marked regions (21, 22). EBV positivity is associated with DNA hypermethylation in tumor cells, with EBV+ gastric carcinoma exhibiting the most elevated degree of promoter hypermethylation of any cancer to date (23). In ENKTL, targeted and genome-wide approaches have reported several genes exhibiting promoter hypermethylation (24–26). Furthermore, recurrent genomic aberrations have been identified in ENKTL that commonly impact signaling pathways, such as somatic mutations in JAK/STAT, and epigenetic regulatory genes, such as BCOR and ARID1A (8, 27–31).
To gain new insights into the epigenetic dysregulation in ENKTL, here we evaluated genome-wide DNA methylation, TF binding, and transcriptomic profiles of primary ENKTL samples in parallel with normal NKDI stages. We further employed novel patient-derived xenograft (PDX) models of ENKTL to evaluate the impact of targeting DNA hypermethylation in vivo with the hypomethylating agent, 5-azacytidine (5-aza). Collectively, these studies lay the foundation for epigenetic-directed therapy targeting the developmental blockade in ENKTL.
Relating ENKTL and NK-cell Development Using Phenotypic and Epigenetic Features
To investigate the relationship of ENKTL to NK-cell differentiation, we first evaluated NK-cell maturation–associated surface markers on ENKTL tumor cells from freshly isolated peripheral blood (PB), bone marrow (BM), and cerebrospinal fluid (CSF) from patients with relapsed or advanced ENKTL. Normal NK cells acquire CD16 and NKp80 as they mature in tonsil-resident developmental stages (Fig. 1A) with the vast majority of mature, circulating PB NK cells expressing these markers. Of five freshly processed or thawed cryopreserved lymphoma-containing samples evaluated, all the circulating malignant NK-cell populations expressed CD56 and CD94 but lacked more mature markers such as KIRs and CD57. Four of these cases (ENKTL1, ENKTL2, ENKTL4, and ENKTL8) also lacked both NKp80 and CD16 expression, conferring a surface immunophenotype similar to tonsil-resident stage 4a NKDIs that are rarely detected in blood (ref. 13; Fig. 1A; Supplementary Fig. S1A). The other case (ENKTL12) was NKp80+CD16dim, similar to normal tonsil stage 5 NKDIs, as was an NK large granular lymphocyte leukemia (NKLGL) sample. In addition, we examined five primary ENKTL formalin-fixed, paraffin-embedded (FFPE) tumors by immunofluorescent confocal microscopy. Despite the high expression of CD56 in cells throughout the tumor samples, few NKp80+ cells were observed in 4 of 5 patients (Fig. 1B). To expand our analysis of the developmental staging of ENKTL samples, we employed an approach comparing the developmental acquisition of epigenetic marks between normal and tumor cells as used previously (16, 17). Using Illumina Methylation-EPIC arrays, we profiled genome-wide DNA methylation patterns of sorted NKDIs and found DNA methylation changes involving hypermethylation of 1.6% and hypomethylation of 2.3% of assayed CpG sites (Fig. 1C). Analysis of the 5,000 most-variable CpG sites revealed a progressive pattern of methylation changes from NKDI stage 3 to 6, demonstrating that the immunophenotypic NKDI classification is paralleled by developing epigenetic states (Fig. 1D). Indeed, hypomethylated regions were enriched for genes and TF-binding sites with known roles in NK-cell development (32) (Fig. 1E).
We next generated genome-wide DNA methylation profiles of 32 patients with ENKTL using available FFPE material and compared the profiles with normal tonsil- and blood-derived NKDIs. We also profiled three primary samples of NKLGL for reference. To infer the developmental status of ENKTL samples, we performed a principal component analysis using the NKDI developmental methylation signature above (Fig. 1F). The first principal component (horizontal axis) revealed the extent of NK-cell development. ENKTL samples naturally separated into two main clusters: one larger cluster most reminiscent of stage 4 NKDIs (n = 26) and the other most similar to tonsil stage 5 NKDIs (n = 8). NKLGL samples were stage 5 like. Notably, ENKTL1&8 (lacking NKp80/CD16) grouped within the stage 4–like cluster based on DNA methylation marks; in contrast, the more immunophenotypically mature ENKTL12 sample grouped within the stage 5–like cluster. To further support the relationship of DNA methylation to the NK-cell developmental immunophenotype, we determined the promoter methylation of NKp80 (KLRF1), which becomes expressed between NKDI stage 4a and 5 (13). KLRF1 promoter methylation was significantly lower in stage 5–like versus stage 4–like ENKTL samples (Fig. 1G), consistent with higher frequency of NKp80 expression in tumor tissue of stage 5–like patients (Supplementary Fig. S1B). Moreover, as T-box motifs are bound by the key developmental TFs Tbet and Eomes and were found enriched in the developmental methylation signature, we measured the cumulative methylation of T-box motifs in ENKTL samples. Stage 4–like ENKTLs ranged in methylation between NKDI stages 4a to 4b and stage 5–like ENKTL was representative of stage 5 NKDIs (Supplementary Fig. S1C).
ENKTL Exhibits Subgroup-specific Epigenetic Gene Silencing Involving Polycomb Repression
To further explore differences between ENKTL subgroups, we performed RNA sequencing (RNA-seq) on available FFPE samples (n = 15 stage 4–like, n = 5 stage 5–like). We identified 123 differentially expressed genes between subgroups (>2 fold-change, FDR q < 0.05) with the vast majority of genes showing lower expression in stage 4- versus 5-like ENKTL subgroups (Fig. 2A). We next investigated pathways and features associated with these differentially expressed genes and identified that features associated with polycomb gene repression, such as H3K27me3 and EZH2 binding, were highly enriched in chromatin immunoprecipitation sequencing (ChIP-seq) datasets from a variety of tissues (Fig. 2B). To investigate this directly in NK cells, we identified H3K27me3-marked regions in normal CD56+ NK cells from available ChIP-seq data (33) and performed gene-set enrichment analysis across all genes. H3K27me3-marked genes were highly enriched in genes repressed in stage 4–like ENKTL (Fig. 2C). In addition, core polycomb repressor complex 2 (PRC2) members EZH2 and SUZ12 were more highly expressed in stage 4–like ENKTL (Fig. 2D).
To further investigate other potential mechanisms of epigenetic silencing in ENKTL, we performed targeted sequencing with a panel of 164 genes recurrently mutated in hematologic malignances in available samples (Supplementary Table S1). We found that mutations commonly occurred in epigenetic regulators, such TET2, BCOR, EP300, and ARID1A, involving 28 of 45 (62%) patients (Supplementary Fig. S2). The next most common category was the JAK/STAT pathway and other kinase signaling pathways each involving 20 of 45 (44%) patients, supporting previous observations (8, 27, 34). However, no differences between stage 4–like and 5-like ENKTL were observed after separating patient samples when permitted by sufficient sample purity. We next explored genomic copy-number alterations (CNA) by inferring tumor-specific gains and losses from Illumina array data in all samples analyzed (n = 51). Positions and sizes of minimally gained and deleted regions of recurrent CNAs are shown in Supplementary Table S2. The most commonly observed CNA was complete or partial loss of chromosome 6q in 17 of 51 (33%) patients (Fig. 2E). Gains on chromosome 7q occurred in 16 of 51 (31%) patients, which included a minimally gained region (in 13 patients) spanning 7q31–7qter containing EZH2. Gain of 7q was only observed in the stage 4–like subgroup. Ten patients with 7q gains showed concurrent complete or partial loss of 7p, with 7p loss selectively occurring with 7q gain (P < 0.001). Other CNAs included loss of 17p13 (10/51), and gain of 8q24 (5/51), which resulted in loss of TP53 and gain of MYC, respectively, as previously reported in smaller cohorts (35, 36). Losses on 9p included 9p24.1–9p21.3 spanning the PDL1/2 and CDKN2A/B loci in 5 of 51 patients as found previously (37). Interestingly, we uncovered two previously unreported CNAs involving losses on chromosome arms 1p (6/51) and Xp (4/51), demarking minimally deleted regions of 0.76 and 4.4 Mb that include the epigenetic regulators ARID1A and BCOR, respectively (Supplementary Fig. S3A and S3B). These findings highlight a role for genomic alterations in the disruption of epigenetic regulators in ENKTL and propose a role for EZH2 gains in developmental arrest and epigenetic gene silencing.
ENKTL Samples Display Prominent CpG Island Hypermethylation
To further investigate the mechanisms underlying developmental arrest in ENKTL, we next focused on DNA tumor-specific DNA methylation differences, that is, those that occur between normal NK cells and ENKTL. We observed recurrent tumor-specific differential DNA methylation gains and losses relative to all NKDIs (Fig. 3A). Compared with the modest amount of hyper- and hypomethylation that occurs during NK-cell development, ENKTL displayed widespread disruption of normal DNA methylation patterns, with approximately 35% to 40% of all measured CpGs showing either gain or loss of methylation. Hypomethylation was more prominent in stage 5–like ENKTL, and was most frequently located in intergenic, non-CpG island (CGI) regions that were largely quiescent in normal T and NK cells (Fig. 3A; Supplementary Fig. S4A and S4B). Analysis of TF sequence motifs revealed strong enrichment for AP-1 (bZIP) motifs in hypomethylated regions, suggesting that ENKTL may involve excessive MAPK pathway activity (Supplementary Fig. S4C), supporting recent findings (31, 38). E-boxes were the second highest enriched motifs, indicative of c-MYC activation consistent with observed chromosome 8q gains. Other enriched motifs include NF-κB (p65), ARNT, and IRF TF families. These results demonstrate that hypomethylation in these ENKTL samples does not occur randomly, instead involving specific regions of the genome and potentially directed by tumor-specific processes.
We also observed hypermethylation in ENKTL. Hypermethylation was greater in stage 4–like samples, involving >40% of all CpGs across the genome and targeted to CGIs and shores (Fig. 3A–C). To demonstrate the unprecedented nature of DNA hypermethylation in ENKTL, we directly compared the frequency of CGI hypermethylation to the EBV+ subtype of gastric adenocarcinoma (EBV+ GC), the tumor subtype that demonstrates the highest degree of CGI hypermethylation across all cancers profiled in The Cancer Genome Atlas (TCGA; ref. 23). This comparison revealed that ENKTLs displayed substantially higher overall CGI methylation and CGI hypermethylation frequencies compared with EBV+ GC tumors (Fig. 3C; Supplementary Fig. S5A). To gain insight into the mechanism underlying altered DNA methylation patterns in ENKTL, we analyzed the association of altered methylation with genome function by partitioning the genome using chromatin states. These states functionally define regions as active, poised, repressed, or quiescent states in combination with enhancer, promoter, transcribed, and heterochromatic function using combinations of histone modifications (39). Hypermethylated regions in ENKTLs were heavily enriched in regions containing the polycomb repressive mark H3K27me3 in normal NK cells, encompassing up to 80% of all hypermethylated CpGs (Supplementary Fig. S5B) and enriched in poised, repressed, and quiescent chromatin states (Fig. 3D). Coordinately, we found that patients with gain of EZH2 displayed additional gene hypermethylation (Supplementary Fig. S5C). Furthermore, increasing hypermethylation across samples correlated with the burden of genomic CNAs, suggesting an additional role for these global epigenetic alterations in tumor genome instability (Supplementary Fig. S5D). These results indicate that DNA hypermethylation is primarily directed to regions of polycomb repression, consistent with enhanced polycomb-associated gene silencing observed above (Fig. 2).
We next investigated how ENKTL-specific DNA hypermethylation impacts global gene expression. As polycomb-marked genes tend to be lower expressed, we performed RNA-seq on three available fresh (non–FFPE-derived) ENKTL samples isolated by FACS purification to avoid dropout of lowly expressed genes and to directly compare with NKDI profiles. We directly compared these global gene expression profiles with DNA methylation patterns from within the same samples and observed that the large majority of downregulated genes in ENKTL cells (compared with matched stage 4b NKDIs) were associated with CGI hypermethylation (Fig. 3E). Genes displaying DNA hypermethylation were largely consistent among ENKTL samples, and 42% of hypermethylated genes were expressed in either stage 4b NKDIs or lowly in ENKTL (Supplementary Fig. S5E). We next plotted differentially expressed genes between ENKTL and NKDIs within the subset of hypermethylated and expressed genes and found that the vast majority differences involved downregulation in ENKTL (Fig. 3F). We identified 88 differentially expressed genes (q < 0.1), including downregulation of genes involved in c-MYC regulation (MGA) and NK-cell development, such as TCF7, KLF4, HES4, DLL1, and IRF8 (40–43).
DNA Hypermethylation is Associated with Loss of EOMES Binding
T-box TFs Tbet and Eomes are lineage-defining TFs for NK-cell development (44–46), and we found that T-box motifs were the most highly enriched motifs in the NK-cell developmental DNA methylation signature (Fig. 1D). To further investigate whether reduced epigenetic programming of T-box–binding sites (Supplementary Fig. S1C) was due to loss of T-box TFs in ENKTL, we measured Tbet and Eomes gene expression and protein abundance. We found that gene expression for both TFs in ENKTL was comparable with normal stage 5/6 NK cells, along with robust protein abundance detected by intracellular flow cytometry (Fig. 4A and B), indicating that lack of programming may be due to disruption of TF binding to chromatin. To investigate whether binding to T-box sites is disrupted and if loss of binding is associated with site-specific DNA hypermethylation, we performed ChIP-seq for EOMES in normal NKDIs and primary ENKTL cells. Comparing stage 5 NKDIs (representing mature NK cells) with ENKTL, we observed that the majority of EOMES peaks in ENKTL were overlapped with stage 5 NKDIs (Fig. 4C). The majority (56%) of peaks present in NKDIs were lost in ENKTL and approximately one-third of these peaks overlapped CGIs. DNA hypermethylation was significantly associated with EOMES loss at these CGIs, as peaks lost in ENKTL were 2.6-fold more likely to be hypermethylated compared with peaks that were unchanged between NKDIs and ENKTL (37% vs. 14% of CGIs, P < 1.0E−12). To illustrate an example of the association between changes of DNA methylation during NK-cell development and hypermethylation associated with loss of EOMES binding and gene expression in ENKTL, we captured these features at the IRF8 locus, a TF required for NK-cell development and function (47, 48). NK-cell development was associated with reduced DNA methylation across the IRF8 locus involving enhancer regions immediately downstream of the promoter CGI and throughout the gene body, as indicated by the colocalization of H3K4me1 and H3K27ac in mature NK cells (Fig. 4D). In ENKTL, developmentally acquired DNA demethylation was reversed, resulting in hypermethylation of these enhancer elements as well as the promoter CGI. EOMES binding in mature NK cells was lost, corresponding with DNA hypermethylation at these key regulatory sites. The increasing expression of IRF8 that occurred during normal NK-cell development was strongly suppressed in ENKTL (Fig. 4E).
Treatment of ENKTL PDX Mice with 5-aza Enhances Differentiation and Survival
To examine the role of DNA hypermethylation in ENKTL, we developed novel PDX models using primary ENKTL cells obtained from two patients experiencing aggressive, refractory, or relapsed disease. DNA methylation analysis of both patients revealed high rates of CGI hypermethylation and stage 4–like disease, however, with differing degrees of maturation (Fig. 5A; Supplementary Fig. S6A). DFTL-85005 lacked NKp80, yet showed KIR, CD57, and CD16dim positivity along with a more advanced epigenetic differentiation status, whereas ENKTL1 showed an immature phenotype lacking NKp80, KIRs, CD57 and CD16, and less epigenetic differentiation (Figs. 1A and 5A and B; Supplementary Fig. S6B). ENKTL cells were engrafted into busulfan-treated NOD-scid IL2Rgammanull (NSG) immunodeficient mice, and mice developed lymphocytosis and hepatosplenomegaly, with organ infiltration by similar populations of atypical, minimally differentiated malignant human NK cells that maintained their pretransplant surface immunophenotypes as well as EBER expression (ref. 49; Fig. 5B; Supplementary Fig. S6C–S6F). DNA methylation analysis of ENKTL cells before and after expansion in NSG mice revealed that tumor-specific hyper- and hypomethylation was stable (Supplementary Fig. S6G). In contrast, healthy donor blood stage 4b NKDIs (CD94+CD16−CD57−KIR−) differentiated normally into late-stage NK cells when engrafted in mice (Fig. 5B), demonstrating the intrinsic nature of the developmental blockade in ENKTL.
We next targeted DNA hypermethylation in ENKTL PDX mice with 5-aza, a hypomethylating agent (HMA) widely used in hematologic malignancies. 5-aza treatment resulted in a significant decrease in tumor burden and increase in overall survival as compared with control mice (Fig. 5C). Engraftment of DFTL-85005 cells resulted in mice expiring in approximately 45 days; however, following HMA therapy, all mice were alive at 4.5 months with no signs of circulating disease. At this point (129 days postengraftment) 5-aza was withdrawn resulting in mice expiring at an average of 37 days postwithdrawal, suggesting that 5-aza provided highly effective disease control in this model. Engraftment of ENKTL1 cells resulted in a similar time to reach moribund in control-treated mice (mean 33 days); yet, although HMA therapy significantly extended overall survival, all mice expired. ENKTL cells obtained from ENKTL1-PDX mice undergoing 5-aza treatment showed upregulation of mature NK-cell markers, including KIRs, CD16, and CD57, suggesting 5-aza–induced differentiation (Fig. 5B). Analysis of ex vivo ENKTL cells showed an on-target global reduction in DNA methylation after treatment (Fig. 5D), impacting tumor-specific hypermethylated CpGs as well as those normally methylated in NKDIs (Supplementary Fig. S6H). 5-aza largely resulted in gene upregulation in bulk harvested ENKTL cells, including cancer-testis (CT) antigens and methylated tumor suppressor genes, such as CDKN1A (Fig. 5E). Despite the global impact of 5-aza on the methylome, drug treatment primarily restored a gene expression signature highly enriched for NK-cell maturation and identity (Fig. 5F), suggesting that differentiation of ENKTL cells is a central facet of the drug's mechanism of action in vivo. As ENKTL1 cells display +7q/EZH2 gain, EZH2 overexpression, and DNA hypermethylation associated with H3K27me3 (Fig. 2E; Supplementary Fig. S7A and S7B), we next treated ENKTL cells with the EZH2 inhibitor tazemetostat (EPZ-6438). Treatment of ENKTL1 cells in vitro did not result in upregulation of developmental markers (Supplementary Fig. S7C). Inhibition of EZH2 alone and in combination with 5-aza did not improve survival in ENKTL1-PDX mice (Supplementary Fig. S7D).
Developmental Phenotypic Diversity Following 5-aza Treatment
As genes belonging to the KIR family are upregulated during the differentiation of NKDIs from stage 4a/b to 5 (Supplementary Fig. S8), and increased KIR expression is paralleled by increased mature NK cell–associated antigens, such as CD16 and CD57 (refs. 5, 7, 45; Fig. 5B and E), we used KIR positivity as a marker of ENKTL differentiation and sorted KIR+/− fractions that arose following 5-aza treatment. To test whether phenotypic differences between KIR+/− subfractions were associated with DNA methylation, subfractions were separately analyzed using Illumina arrays. The KIR+ subset exhibited consistently reduced global DNA methylation compared to KIR− cells during 5-aza treatment (Fig. 6A), suggesting a relationship of 5-aza–induced hypomethylation with differentiation. Gene expression analysis of KIR+ versus KIR− cells revealed selective upregulation of other (non-KIR) genes involved in NK-cell maturation and function in KIR+ cells, including CD16 (FCGR3A) and NKp44 (NCR2) (Fig. 6B). In vivo–expanded ENKTL cells were serially transplantable, and thus we tested whether 5-aza–induced phenotypic changes were transient or stable by separately retransplanting KIR+ and KIR− ENKTL cells obtained from ENKTL1-PDX mice that had been treated with 5-aza for 14 days. KIR+/− subfractions largely retained their differentiation status during expansion in mice despite the absence of treatment, suggesting largely stable phenotypic reprogramming by 5-aza (Fig. 6C). However, we observed that KIR+ cells displayed less Ki67 positivity than KIR− cells (Fig. 6D), consistent with decreasing proportions of KIR+ versus KIR− cells throughout treatment (Fig. 6E), suggesting competitive expansion of 5-aza–resistant KIR− cells in this model. Collectively, these findings demonstrate that targeting DNA hypermethylation in ENTKL induces differentiation and reprogramming of tumor cells along with increased survival in PDX models, raising the possibility of therapeutic benefit to patients.
Malignant counterparts to normal human B-cell and T-cell developmental intermediates have been known for decades (50, 51); however, the normal developmental pathway for human NK cells in secondary lymphoid tissue has only recently been elucidated (9–12, 32, 45). This permitted our investigation showing that ENKTL represents the malignant counterpart to a specific range of human NKDIs normally present in mucosal tissues. Our work provides new insights into the developmental, phenotypic, and molecular features of ENKTL, and highlights the importance of profound DNA hypermethylation that contributes to the blockade of NK -cell differentiation. This global epigenetic disruption resulted in gene silencing relative to expression patterns found in normal NKDIs and in mature NK cells. Employing novel PDX models uncovered that targeting the epigenetic aberration partially reversed gene silencing, promoted NK-cell differentiation, and extended survival, a potential new therapeutic strategy for this fatal disease.
The arrest of ENKTL cells at precursor stages of NK-cell development is supported by the striking similarity of tumor cells with (i) the immunophenotypic markers present in normal tissue-associated NKDIs, and (ii) the global and site-specific DNA methylation patterns present in normal tissue-associated NKDIs. As normal NK-cell development was primarily associated with reductions in DNA methylation associated with increased accessibility for key developmental TFs, ectopic epigenetic silencing and repression resulting from primary EBV infection likely prevents (or reverses) the acquisition of key developmental events in tumor cells. We found that patients with a more immature, stage 4–like phenotype exhibited a greater degree of gene silencing, which was associated with regions displaying polycomb repression in normal cells. EZH2 overexpression is associated with enhanced proliferation and poor prognosis in ENKTL (52, 53), and targeting of DNA hypermethylation to polycomb-marked regions has been described broadly in cancer (21, 22). As we observed higher expression of PRC2 subunits including EZH2 in stage 4–like ENKTL, one possibility explaining the greater degree of DNA hypermethylation in stage 4–like ENKTL is that higher PRC2 activity provides more H3K27me3 template for DNA hypermethylation to occur. Polycomb silencing is used by normal cells as a reversible switch to repress/poise genes where expression may be required at a later stage of development. The switch to DNA methylation–based repression may result in more robust silencing and prevent key developmental genes from becoming expressed at the appropriate later stage of development. The lack of differentiation and cytotoxicity following EZH2 inhibition may reflect predominant DNA methylation–based silencing once DNA hypermethylation has occurred during disease progression. DNA hypermethylation thus may prevent key TFs that coordinate NK-cell development, such as TBET and EOMES, from being able to activate target genes necessary to facilitate differentiation. Our findings suggest that functional gene repression/silencing of developmental genes occurs at the level of DNA and chromatin. In support, when DNA hypermethylation was reduced in ENKTL cells, the most prevalent gene expression signature was that of mature NK cells, despite the global nature of 5-aza–induced demethylation.
Our finding of the near-uniform, severe DNA hypermethylation observed in the PDX models and our ENKTL samples provides rationale for HMA use in this disease. The mechanism of acute cytoreduction following 5-aza treatment appears to involve differentiation and reduction of proliferation affecting the vast majority of cells and provided long-term disease control in one PDX model. The differentiation of ENKTL cells following 5-aza treatment is reminiscent of the effects of single agent all-trans retinoic acid on acute promyelocytic leukemia whose further development has led to an astounding cure rate. Once ENKTL cells underwent demethylation, the altered phenotype and DNA methylation were stable following therapy cessation and subsequent transplantation, suggesting an opportunity for combining HMAs with other approaches for ENKTL treatment. Here we show that 5-aza upregulates many genes with immunoregulatory function, along with CT antigens, which may improve responses to immunotherapy. In summary, our work provides insight into the pathogenesis of a rare but fatal lymphoid malignancy and as such, lays the foundation for epigenetic-directed therapy targeting the developmental blockade in ENKTL.
We used material from n = 50 patients with ENKTL after excluding samples with insufficient nucleic acid quality and tumor purity (see below sections for QC and purity determination and thresholds for specific assays). ENKTL-involved FFPE tissue samples were obtained from The Ohio State University (OSU, Columbus, OH; n = 9), Stanford University (Stanford, CA; n = 7), the Laboratorio de Patologia in Guatemala (n = 8), Liga Nacional Contra el Cáncer (INCAN) (n = 11), the Instituto Nacional de Enfermedades Neoplasticas in Peru (INEN) (n = 5), and the National Taiwan University Hospital (n = 6). We also obtained viable tumor cell samples from OSU (ENKTL n = 3, NKLGL n = 2) and from the University of Virginia (Charlottesville, VA; ENKTL n = 1, NKLGL n = 1). Cells were cryopreserved and stored in liquid nitrogen until use unless otherwise noted. Viable tumor cells were purified from PB, BM, or CSF using FACS (for sorting details, see below). Pathologic diagnosis was confirmed by hematopathology review according to WHO classification specifications (54), including confirmation of EBER positivity. Additional immunofluorescence staining for developmental staging markers included CD56, NKp80, and IRF8 (see below). Written informed consent was obtained for research material evaluated in this study (reviewed by OSU Institutional Review Board, Protocol #2017C0070) with the assistance of the OSU Leukemia Tissue Bank Shared Resource, and in accordance with the Declaration of Helsinki.
Flow Cytometric Analysis and FACS of NKDIs and Patient Samples
NKDIs were isolated from fresh tonsil tissues or peripheral blood as previously described (13, 55). ENKTL cells were isolated from fresh or freshly frozen ENKTL patient samples by standard Ficoll paque isolation technique (13, 56). Single-cell suspensions were labeled with NK-associated antibodies (see Supplementary Table S3 for a complete list of antibodies) and sorted to purity using a BD FACS Aria II cell sorter. Purities were validated postsort and all samples had purity greater than 99%. For in vivo cell isolation, lymphoid tissue or peripheral blood was isolated from NSG mice engrafted with primary ENKTL cells (see description below). Cells were mechanically dissociated into a single-cell suspension as described previously (56), and labeled with murine CD45.1 or human CD45 (hCD45) to distinguish ENKTL populations. Among hCD45+ cells, all ENKTL samples expressed CD56 and were lineage negative (CD3/14/20), and they were further divided into KIR+ and KIR− subsets based on the expression of pan-KIR2D and KIR3DL1/2 for cell isolation. For apoptosis evaluation by annexin V staining, single-cell suspensions were labeled for surface expression of hCD45, CD56, and KIRs and then subsequently labeled with anti-annexin V and SYTOX live/dead viability dye (Invitrogen). Samples were incubated for 15 minutes and run immediately on a BD LSR II flow cytometer. For Ki67 expression, cells were labeled for surface marker expression and were then fixed, permeabilized, and stained for Ki67 expression according to the manufacturer's recommendations (BD Biosciences). Samples were then analyzed on a BD LSRII flow cytometer. All flow cytometric analysis was performed using FlowJo (TreeStar) software.
Immunofluorescence Staining of Paraffin-Embedded Tissue and Confocal Microscopy
Paraffin-embedded tissue sections were stored at -20°C and brought to room temperature before dewaxing via three 10-minute xylene (catalog no. X5–1; Fisher Scientific) washes. Dewaxed tissue sections were rehydrated in serial washes of 100%, 95%, and 70% ethanol (catalog no. BP2818100; Fisher Scientific) for 10 minutes, followed by 35% ethanol wash for 5 minutes and excess ethanol was removed by washing tissue with deionized water. Heat-activated antigen retrieval was performed on rehydrated tissue sections in Agilent Dako antigen retrieval solution (catalog no. S2367; Agilent) in a steamer (Black & Decker). Samples were permeabilized with 0.2% Triton X-100 (catalog no. BP151–100; Fisher Scientific) in PBS (catalog no. 14–190–235; Thermo Fisher) for 20 minutes and washed with PBS. Prior to immunostaining, tissue sections were incubated with blocking buffer [PBS/10% donkey serum; catalog no. D9663; Millipore Sigma-Aldrich)/5% BSA (catalog no. BP1605–100; Fisher Scientific)/0.1% Triton X-100] for 1 hour at room temperature. Anti-NKp80 (catalog no. MBS9412114; MyBioSource) was diluted in antibody diluent buffer (PBS/1% donkey serum/0.5% BSA/0.01% Triton X-100) to a concentration of 3.5 μg/mL. Tissue sections were incubated with NKp80 primary antibody overnight at 4°C followed by four washes with antibody diluent buffer and a second round of blocking for 1 hour. Goat anti-rabbit IgG Alexa Fluor 647 (catalog no. A27040; Invitrogen) secondary antibody was used at a dilution of 1:100. Blocked tissue was incubated with anti-CD56 FITC-conjugated antibody (clone no. 123A8; Novus Biologicals). To amplify fluorescence signal detection of CD56, samples were incubated with rabbit anti-FITC/GFP IgG AlexaFluor 488 secondary antibody (catalog no. A-11090; Thermo) diluted 1:200. Three-dimensional image acquisition was performed at 1.87 μm step using a 20 × 0.5 NA objective or 0.17-μm steps with a 100 × 1.46 NA objective on a Zeiss AxioObserver Z1 microscope stand equipped with a Yokogawa W1 spinning disk and a Prime 95B sCMOS camera for detection and 3i Slidebook software for data acquisition and export. Confocal imaging data were processed using Imaris File Converter (Bitplane), Fiji (57), and QuPath 0.2.3 (58).
DNA Methylation Analysis
DNA was extracted from freshly sorted cells (ENKTL, NKLGL, NKDIs) using the Gentra Puregene Cell Kit (Qiagen) or using the QIAamp DNA FFPE Tissue Kit (Qiagen) for FFPE tissues. Genome-wide DNA methylation was generated using Infinium MethylationEPIC Kit (Illumina) according to manufacturer's recommended procedures. For FFPE samples, DNA was preprocessed using the Infinium FFPE QC and DNA Restoration Kit (Illumina). Data were processed using the RnBeads software (59) and normalized using BMIQ (60). Illumina probes located on X/Y chromosomes, non-CpG probes, and probes with a SNP (>0.01 MAF) located ≤5 bp to the analyzed CpGs were censored for a total of 804,572 analyzed CpGs. Sample purity was assessed using reference-based cell type deconvolution (61). After exclusion of samples with <70% tumor purity and samples that failed to pass data quality control thresholds, n = 32 ENKTL samples were used for DNA methylation analyses. All NKDI (n = 21), NKLGL (n = 3), and NK-cell line (n = 3) samples were used. NK92 cells were obtained from ATCC, and NKL cells were a generous gift from Dr. Michael Robertson (Indiana University, Indianapolis, IN), and YT cells were obtained through the DSMZ (German Cancer Cell Repository). YT and NK92 cells were verified by PCR genotyping of short tandem repeat markers. All cell lines were utilized within six passages from original stock and cultured less than 14 days prior to studies. Cell lines were maintained in RPMI media supplemented with 10% FBS (Invitrogen). Mycoplasma testing was routinely performed as quality control on an annual basis, and no additional modifications were performed to the cells prior to analysis. Targeted analysis of the KLRF1 (NKp80) promoter DNA methylation was performed using the MassARRAY EpiTYPER assay (Agena Biosciences). Briefly, DNA was converted by sodium bisulfite using the EZ DNA Methylation Kit (Zymo Research). Bisulfite DNA–specific PCR primers were designed to amplify CpGs in vicinity of the KLRF1 promoter, and following PCR amplification, PCR products were subject to the standard EpiTYPER workflow. Primer sequences (for bisulfite-converted DNA) used were (forward) 5′-TGGTGTGAATTAATATGGTAGTTTTT, (reverse) 5′-ACACCCAAATACTAAAACAATCCAA; and (forward) 5′-TTTTTTAGTTTGTTATTAAGTTTAGGGAGA, (reverse) 5′-TCAATACCCTATATAATCCAATAACCAC. Genomic copy number for ENKTL (n = 50), NKLGL (n = 3), and NK cell line (n = 3) samples was derived from Illumina array data using Conumee (https://github.com/hovestadt/conumee).
DNA Methylation Data Analysis
The total number of differentially methylated CpGs during NK-cell development was determined by comparing stage 3 and stage 6 NKDIs and identifying CpGs that differed by 20% and with an FDR of q < 0.1. Tumor cell (ENKTL)-specific differential methylation was identified by comparing ENKTL subgroups versus matched NKDI stage likewise showing >20% difference and q < 0.1. To determine the proportion of hyper- or hypomethylated CpGs in ENKTL, only probes that showed less or greater than 50% methylation in NKDIs, respectively, were considered. Enrichment and annotation of DNA sequence motifs proximal (±100 bp) to CpGs of interest were determined using HOMER (62). CpGs of interest were compared with a background probe set composed of probes matched for CpG, GC, and DNA methylation content; universe appropriateness was evaluated using the LOLA R package (63). Principal component analysis comparing ENKTL and NKDIs was generated using the 5,000 most-variable CpGs among NKDIs. Positions of CpGs relative to transcriptional start sites (TSS), gene body, introns, etc. and CpG island (CGI) features were obtained from Illumina probe annotation. An occupancy plot of all CpG methylation ±2 kb of CGIs was produced by considering all CGIs >300 bp, for a total of n = 21,306 islands. For comparisons including gastric adenocarcinoma subtype samples, EPIC/850K data was downsampled to overlap the 450K probe set. Raw Illumina methylation data for TCGA gastric adenocarcinoma samples was obtained from the Genomic Data Commons (GDC) data portal and were processed identically. Likewise, samples with <70% tumor cell nuclei in TCGA/GDC annotation were excluded. CGIs were considered hypermethylated if it contained a CpG that gained ≥50% methylation versus the matched normal subtype. Chromatin states for normal blood CD56+ NK cells, CD4+ and CD8+ T cells, and CD34+ HSCs were obtained from the Epigenetics Roadmap Project (64). Overlap enrichment and P values was determined using Epiannotator software (65). We used the 15-state model and did not display 3 states due to low representation. Histone ChIP-seq data for CD56+ NK cells from cord and peripheral blood was obtained from Blueprint Epigenome DCC portal. Global methylation was determined by taking the mean beta value across all nonfiltered CpGs passing QC thresholds above. Heatmaps and PCA plots were generated using the Qlucore Omics Explorer software (Qlucore). Histograms, scatterplots, and ad hoc statistical analyses were done using Prism (GraphPad Software Inc).
Targeted mutation analysis was performed for ENKTL samples with available material (n = 45) at the Stanford Molecular Genetic Pathology Clinical Laboratory using a clinically validated, targeted next-generation sequencing assay. Sequencing libraries were prepared using the KK8232 KAPA LTP Library Preparation Kit Illumina Platforms (KAPA Biosystems), and target enrichment with custom-designed oligonucleotides (Roche NimbleGen). Oligonucleotide design was performed as described previously using the Cancer Personalized Profiling by deep Sequencing (CAPP-Seq) method (66). The panel covers, partially or fully, 164 genes that are recurrently mutated and clinically relevant in hematolymphoid malignancies (Supplementary Table S1). Pooled libraries were sequenced on an Illumina sequencing instrument (MiSeq or NextSeq 550, Illumina). Mapping was performed against the human reference genome hg19 with BWA in paired-end mode using the bwa-mem algorithm and standard parameters. Samtools BAQ adjustment was enabled for single-nucleotide variants (SNV) calling but disabled for deletion/insertion calling, and VarScan v2.3.6 was used for variant calling, specifically for SNVs and deletion/insertion <20 bp. Variants were annotated using Annovar and Ensembl reference transcripts with a sensitivity cutoff of 5% variant allele fraction. Quality control was performed in compliance with the laboratory accreditation requirements outline by the College of American Pathologists (Molecular Pathology-Checklist, CAP accreditation program). Some specific quality control metrics monitored for every run and over time include: on-target rate, median depth, percent positions over 100×/200×/500×, fragment size distribution, depth of positions where clinically actionable variants lie, minimum/maximum/mean/median read depth of exons and other regions captured. The quality control data was reviewed and approved by a molecular pathologist before data interpretation was initiated.
RNA-seq and Analysis
Freshly sorted blood and tonsil NKDI samples (three samples of each NKDI population; n = 21 samples total), available fresh sort-purified ENKTL samples (n = 3), and ex vivo ENKTL cell samples harvested from PDX mice (n = 18) were pelleted and total RNA was isolated using the Qiagen RNeasy Mini Kit (Qiagen). RNA was extracted from FFPE samples (n = 20) using the RNeasy FFPE kit (Qiagen). Directional poly-A RNA-seq libraries were prepared using the NEBNext Ultra II Directional RNA Library Prep Kit Cat#E7760 (New England Biolabs) and sequenced as 42-bp paired-end reads on an Illumina NextSeq 500 instrument (Illumina) to a depth of 33.2–48.0 million read pairs. Alignment to human genome version hg38 was done using TopHat (67). Transcriptome assembly and analysis was performed using Cufflinks and expression was reported as FPKM. For FFPE samples, low-quality reads (q < 10), and adaptor sequences were eliminated from raw reads using bbduk version 37.64. Samples were aligned to Ensemble Human reference GRCh37.97 using the RNA-seq aligner STAR version 2.6.0c. Features were identified from Ensemble (grch37/release-97). Feature coverage counts were calculated using featureCounts, FPKMs values were generated on the basis of overall alignment. Differential gene expression was determined using DEseq2 (68). To determine pathway and feature enrichment between ENKTL subgroups, a gene set of 410 significantly lower expressed genes in stage 4–like versus 5–like ENKTL (FDR q < 0.1) was interrogated using Enrichr (69). Gene-set enrichment analysis (70) was performed using a custom gene set generated from the intersection of H3K27me3 peaks in ChIP-seq profiles from normal CD56+ cells (obtained from Blueprint) and assigned to the most proximal TSS (<3 kb). Differentially expressed genes following 5-aza treatment in ENKTL cells and differential expression between KIR+ and KIR− subsets were determined by separating the groups and applying a threshold of >2.0 log2 fold-change FDR q < 0.1. Heat maps display median-normalized, log2-transformed expression values visualized using Qlucore. Enrichment of upregulated genes following 5-aza treatment was evaluated against the GeneAtlas gcrma dataset (http://biogps.org/dataset/GSE1133), which includes a panel of 176 individual tissues and cell types, using Enrichr (71).
ChIP-seq was performed by Active Motif Inc. In brief, 5–10 million ENKTL or stage 5 NKDI cells were fixed with 1% formaldehyde for 15 minutes and quenched with 0.125 mol/L glycine followed by lysis and disruption with a Dounce homogenizer. Lysates were sonicated and the DNA sheared to an average length of 300–500 bp using the EpiShear probe sonicator (catalog no. 53051). Genomic DNA (Input) was prepared by treating aliquots of chromatin with RNase, proteinase K, and heat for de-crosslinking, followed by clean up using SPRI beads (Beckman Coulter), and quantitation by Clariostar (BMG Labtech). Chromatin (10 μg) was immunoprecipitated with 4 μg of TBR2/EOMES antibody from Abcam (catalog no. ab23345, lot no. GR3304549–1). Complexes were washed, eluted from the beads with SDS buffer, and subjected to RNase and proteinase K treatment. Crosslinks were reversed by incubation overnight at 65°C, and ChIP DNA was purified by phenol–chloroform extraction and ethanol precipitation. Paired-end Illumina sequencing libraries were derived from ChIP and Iinput DNAs as described previously (72) using an automated system (Apollo 342, Wafergen Biosystems/Takara). Illumina libraries were sequenced on an Illumina NextSeq 500 instrument using 75-nt single-end sequencing generating >30 million reads per sample. Reads were aligned to hg38 using BWA, and reads that did not pass Illumina's purity filter, duplicate reads, aligned with >2 mismatches, and were nonuniquely mapped to the genome were eliminated for subsequent analysis. EOMES peaks were called using the MACS2 algorithm (73), resulting in 13,023 and 19,942 called peaks in ENKTL and stage 5 NKDI, respectively. Peaks were visualized using the Integrative Genomics Viewer (IGV).
Generation and Treatment of ENKTL PDX Mice
NSG mice were obtained through the Ohio State Vector Core facility. All murine studies were reviewed and performed in compliance with OSU IACUC regulations (approved OSU IACUC protocol no. 2009A0033). Tumor cells used for DFTL-85005 PDX mice were obtained from the Public Repository for Xenografts (PRoXe), PDX name DFTL-85005-V4 (74). Cells were obtained from a patient with stage 4 ENKTL with skin, testicular, bone marrow, and likely nasopharyngeal involvement. The sample was collected during progressive disease phase following cycle 1 of SMILE chemotherapy in accordance with OSU IRB Protocol #2017C0070 as described above and in accordance with the Declaration of Helsinki. Tumor cells used for ENKTL1 PDX mice were obtained from a relapsed ENKTL patient at OSU following treatment with VIPD+XRT (etoposide, ifosfamide, cisplatin and dexamethasone plus radiotherapy), AspaMetDex (l-asparaginase, methotrexate, and dexamethasone), and GEMOX (gemcitabine and oxaliplatin) lines of chemotherapy. Neither patient received HMA therapy. Six- to 10-week-old female and male mice were pretreated with 25 mg/kg busulfan (Selleckchem, catalog no. S1692) 1 day prior to ENKTL engraftment. The next day, primary ENKTL cells (2 × 106 cells/mouse) were injected intravenously into the tail veins of the NSG mice. To promote initial engraftment, the first passage of mice were given intraperitoneal injections of recombinant human IL15 (0.5 μg; NCI Frederick) at day 0 following tumor inoculation and twice weekly thereafter. Subsequent experiments were performed using cells from passage two without cytokine support. For time course studies, blood was obtained via peripheral cheek bleed. ENKTL cells were harvested from spleens and purified by FACS at the indicated timepoints for transcriptomic and genomic analysis as described above. For studies with 5-aza (Sigma, catalog no. A2385), ENKTL cells were allowed to engraft and starting on day 7, mice were treated with 5 mg/kg 5-aza delivered intraperitoneally three times per week for 14 days, rested for 7 days, and then repeating the 14-day cycle. For EZH2 inhibition studies, ENKTL1 cells were injected as indicated above, and starting on day 7, mice were treated with vehicle control, 5 mg/kg 5-aza delivered intraperitoneally three times per week for 14 days, 200 mg/kg tazemetostat (EPZ-6438) delivered by oral gavage daily, or both drugs in combination. For retransplant studies, ENKTL1 PDX mice were treated for 14 days with 5-aza followed by transplantation into new NSG mice for 21 days before reextraction.
Illumina array and sequencing data are available at the Gene Expression Omnibus (GEO) website under accession number GSE169646.
B.L. Mundy-Bosse reports grants from the Pelotonia Foundation during the conduct of the study, as well as grants from the NCI and American Cancer Society outside the submitted work. A.P. Nalin reports grants from the NIH/NCI during the conduct of the study. E.L. Briercheck reports a Conquer Cancer Foundation Global Oncology Young Investigator Award and a Fulbright Scholar Award during the conduct of the study. T.P. Loughran reports personal fees and other support from Dren Bio; personal fees from Kymera Therapeutics; and other support from Keystone Nano and Bioniz outside the submitted work. D.M. Weinstock reports grants from Daiichi Sankyo and Abcuro; other support from Ajax and Travera; personal fees from Bantam and AstraZeneca; and grants and personal fees from Secura and Verastem outside the submitted work. M.A. Caligiuri reports grants from the NIH during the conduct of the study, as well as nonfinancial support from CytoImmune Therapeutics Inc. and personal fees from Cytovia, OncoC4, and CBMG outside the submitted work. J.E. Brammer reports grants from the NIH/National Center for Advancement of Translational Science (NCATS), a Pelotonia Intramural Research Grant, and grants from Division of Hematology Sponsored Research Program during the conduct of the study, as well as grants from Bristol Myers Squibb/Celgene and other support from Viracta outside the submitted work. A.G. Freud reports grants from the NIH/NCI and Pelotonia (OSU) during the conduct of the study, as well as personal fees from ImmuneBridge, LLC outside the submitted work. No disclosures were reported by the other authors.
B.L. Mundy-Bosse: Conceptualization, data curation, formal analysis, supervision, funding acquisition, validation, methodology, writing–original draft, writing–review and editing. C. Weigel: Data curation, methodology. Y. Wu: Data curation, formal analysis, methodology. S. Abdelbaky: Data curation. Y. Youssef: Data curation, writing–original draft. S. Beceiro Casas: Data curation. N. Polley: Data curation. G. Ernst: Data curation. K.A. Young: Data curation, investigation. K.K. McConnell: Data curation, investigation. A.P. Nalin: Data curation, investigation. K.G. Wu: Data curation, investigation. M. Broughton: Data curation, investigation. M.R. Lordo: Data curation, investigation. E. Altynova: Data curation, investigation. E. Hegewisch-Solloa: Methodology, writing–original draft. D.Y. Enriquez-Vera: Resources. D. Duenas: Resources. C. Barrionuevo: Resources. S. Yu: Resources. A. Saleem: Resources. C.J. Suarez: Resources, data curation. E.L. Briercheck: Resources, data curation. H. Molina-Kirsch: Resources. T.P. Loughran: Resources. D. Weichenhan: Data curation, formal analysis. C. Plass: Data curation, formal analysis, methodology. J.C. Reneau: Resources, data curation, writing–original draft. E.M. Mace: Formal analysis, visualization, methodology, writing–original draft. F. Valvert Gamboa: Resources. D.M. Weinstock: Resources, writing–review and editing. Y. Natkunam: Resources, data curation, formal analysis, writing–review and editing. M.A. Caligiuri: Conceptualization, writing–review and editing. A. Mishra: Conceptualization. P. Porcu: Conceptualization, resources, writing–review and editing. R.A. Baiocchi: Conceptualization, writing–review and editing. J.E. Brammer: Resources, data curation, writing–review and editing. A.G. Freud: Conceptualization, resources, data curation, methodology, writing–original draft. C.C. Oakes: Conceptualization, resources, data curation, supervision, methodology, writing-original draft, writing–review and editing.
We would like to acknowledge the expertise and thank Matthias Schick and Roger Fischer at the German Cancer Research Center (DKFZ) Genomics and Proteomics Facility (GPCF). This work was supported by grants from the NIH/NCI (R01CA255860, to B.L. Mundy-Bosse and C.C. Oakes; L30CA199447 and R01CA208353, to A.G. Freud; K22CA218466, to B.L. Mundy-Bosse; KL2TR002734, to J.E. Brammer; F30CA236063, to A.P. Nalin; and R01AI137073, to E.M. Mace), the American Cancer Society (RSG-21-150-01-CDP), the OSUCCC/Pelotonia Organization, and the OSU Division of Hematology. C.C. Oakes is supported by the Gabrielle's Angel Foundation for Cancer Research. We thank the Cooperative Human Tissue Network of Nationwide Children's Hospital for providing us with human pediatric tonsil samples. We also thank The Ohio State University Global One Health Initiative and the OSUCCC Leukemia Tissue Bank and Analytic Cytometry Shared Resources, supported by NCI P30 CA16058.
Note: Supplementary data for this article are available at Blood Cancer Discovery Online (https://bloodcancerdiscov.aacrjournals.org/).