Integrated analysis of 10 lymphoma datasets identifies E2F8 as a key regulator in Burkitt’s lymphoma and mantle cell lymphoma

Abstract

Burkitt’s lymphoma (BURK), diffuse large B-cell lymphoma (DLBCL) and mantle cell lymphoma (MCL) are three main types of B-cell lymphomas. This study aimed to compare the differences of affected biological functions and pathways, as well as to explore the possible regulatory mechanisms and the potential therapeutic targets in BURK, DLBCL and MCL. We performed an integrated analysis of 10 lymphoma datasets including 352 BURK patients, 880 DLBCL patients, 216 MCL patients, and 33 controls. Our results showed that signaling pathways, amino acid metabolism and several lipid metabolism pathways varies considerably among these three types of lymphoma. Furthermore, we identified several key transcription factors (TFs) and their target genes that may promote these diseases by influencing multiple carcinogenic pathways. Among these TFs, we reported first that E2F8 displayed the most significant effects in BURK and MCL. Our results demonstrate that over-expression of E2F8 activates target genes that may promote cell cycle, mitosis, immune and other cancer related functions in BURK and MCL. Therefore, we suggest that E2F8 could be used as a biomarker and potential therapeutic target for BURK and MCL. These findings would be helpful in the study of pathogenesis, and drug discovery and also in the prognosis of B cell lymphomas.

Introduction

Non-Hodgkin’s lymphomas (NHLs) are broadly classified into two major groups: B-cell lymphomas and T-cell/NK-cell lymphomas. B-cell lymphomas develop from abnormal B lymphocytes and account for 85-90% of all NHLs [1,2]. Burkitt’s lymphoma (BURK), diffuse large B-cell lymphoma (DLBCL) and mantle cell lymphoma (MCL) are three main types of B-cell lymphomas. BURK is often associated with Epstein-Barr Virus (EBV) infection, being endemic in some parts of Africa but encountered only sporadically in the rest of the world. BURK is largely diagnosed in children. In Cameroon, the country with the highest incidence rate worldwide, the average incidence in the northwest province is 2.58 cases per 100,000 per year in children aged 15 or younger [3]. DLBCL accounts for approximately 33% of all newly diagnosed cases of NHL worldwide. It is most commonly diagnosed during the sixth decade, and is slightly more common in males [4]. Epidemiology on MCL is primarily available for the U.S. and Europe, where the average incidence is approximately 0.5 cases per 100,000 person-years. The male-to-female ratio is around 2.5:1, and the average age of onset is 65-70 years [5,6]. Immunosuppression is the main risk factor for NHLs [1,7]. Other putative environmental and lifestyle-related risk factors have also been proposed, including current cigarette smoking [8] and heavy alcohol intake [9].

In addition to the traditional risk factors, recent studies have identified several NHL risk genes. A previous study found that the co-expression of MYC/BCL6 may affect the survival in patients with DLBCL [10]. Another study suggested that subset-specific B-cell-associated gene signatures were better suited to identifications than the traditional classifications (biologic, morphologic, and clinical heterogeneity) in clinical diagnosis and prognosis of DLBCL [11]. In addition, the top discriminatory genes expressed in formalin-fixed, paraffin-embedded (FFPE) tissues of DLBCL patients also showed high statistical difference and contained novel biology with potential therapeutic insights [12]. Furthermore, gene expression signatures were performed well in distinguishing BURK and DLBCL [13,14]. Still another recent study identified that the deregulation of CUL4A, ING1 and MCPH1 affects the proliferation; and DNA damage response pathways in MCL by integrating gene expression and copy number changes [15]. Palomero et al. showed that over-expression of SOX11 activated PDGFA and promoted tumor microvascular angiogenesis in MCL cell lines, suggesting that the SOX11-PDGFA axis may serve as a potential therapeutic target of MCL [16]. Saba et al. demonstrated that B-cell receptor activation and canonical NF-kB signaling may promote tumor proliferation and affect overall survival in MCL patients [17].

Reconstruction of transcriptional regulation networks (TRNs) has been a useful approach that provides a global view of gene expression profiles and their regulations, and it also provides new insights in tumorigenesis, pathways and regulations [18,19], as well as differentiation of embryonic stem cells [20]. Our previous study has identified several tissue-specific transcription factors (TFs) that correlated with breast cancer by analyzing tissue-specific TRNs [21]. However, to our knowledge, no other studies have compared multiple types of NHLs from the perspective of TRNs. Therefore, in the present study, we integrated 10 lymphoma microarray datasets including BURK, DLBCL and MCL in order to explore common differential changes in global gene expression, pathway and TRN analyses, as well as to identify key regulators and their targets among these three types of lymphoma.

Materials and methods

Lymphoma datasets collection

Transcriptome datasets of BURK, DLBCL and MCL were downloaded from NCBI-GEO (https://www.ncbi.nlm.nih.gov/geo/) in Jun, 2017. The data selection criteria were as follows: (1) all datasets were genome-wide; (2) the samples in each dataset must be ≥ 10; (3) all samples were non-cell-line samples; and (4) complete microarray raw or normalized data were available. Based on the above criteria, we chose 10 datasets for our integrated analysis. The details of these datasets are illustrated in Table 1. In total, we collected samples from 352 BURK patients, 880 DLBCL patients, 216 MCL patients, and 33 controls.

Table 1.

Details of lymphoma datasets

GEO ID	Contributor	Title	Samples
Burkitt’s lymphoma
GSE4475	Hummel M, 2006	A Biologic Definition of Burkitt’s Lymphoma from Transcriptional and Genomic Profiling	221 cases
GSE69053	Sha C, 2015	Transferring genomics to the clinic: distinguishing Burkitt and Diffuse large B-cell lymphoma	131 cases
Diffuse large B-cell lymphoma
GSE31312	Li Y, 2011	Development and application of a new immunophenotypic algorithm for molecular subtype classification of Diffuse Large B-Cell Lymphoma (DLBCL): Report from an International DLBCL Rituximab-CHOP Consortium Program Study	498 cases
GSE56315	Bødker JS, 2014	Diffuse Large B-Cell Lymphoma Classification System That Associates Normal B-Cell Subset Phenotypes With Prognosis	89 cases 33 controls
GSE64555	Linton K, 2014	Microarray gene expression analysis of FFPE tissue permits molecular classification in diffuse large B-cell lymphoma	40 cases
GSE69053	Sha C, 2015	Transferring genomics to the clinic: distinguishing Burkitt and Diffuse large B-cell lymphoma	212 cases
GSE86613	Bødker JS, 2016	Expression data from diagnostic samples of diffuse large B-cell lymphomas (DLBCL), follicular lymphoma (FL) and primary and relapsed transformed FL	41 cases
Mantle cell lymphoma
GSE21452	Staudt LM, 2010	Integrated genomic profiling in mantle cell lymphoma	64 cases
GSE36000	Jares P, 2012	Mantle Cell Lymphoma	38 cases
GSE70910	Liu D, 2015	Direct in vivo evidence for B-cell receptor and NF-KB activation in mantle cell lymphoma: role of the lymph node microenvironment and activating mutations	55 cases
GSE93291	Staudt LM, 2017	A new molecular assay for the proliferation signature in mantle cell lymphoma applicable to formalin-fixed paraffin-embedded biopsies	59 cases

Data preprocessing and differential expression analysis

R statistical software v3.3.3 (https://www.r-project.org/) was used to perform data preprocessing and differential gene expression gene analysies. Because these datasets contain different microarray platforms and different types of lymphoma, they were first grouped into 14 batches according to study, category (different types of lymphoma and controls), and platform. Each batch contained only one study, one type of lymphoma or control, and one platform (Table 2). We used the Robust Multichip Average (RMA) algorithm in oligo package [22] to normalize the raw expression data and generate normalized gene expression intensity. Gene annotation, integration and renormalization of the 14 batches was carried out using a custom designed Python code. We removed probes with no gene annotation or those that matched multiple gene symbols. Next, we calculated the average expression value of multiple probe IDs matching an official gene symbol and took this value to represent the expression intensity of the corresponding gene symbol. The renormalization method and scripts are detailed in our previous publications [23,24]. The distributions of RMA-processed and global-renormalized gene expression values across all studies are shown in Figures S1 and S2. After global expression was renormalized, the distribution of gene expression values across all studies had a consistent range. The empirical Bayesian algorithm in the limma package [25] was used to detect differentially expressed genes between lymphoma patients and controls. Up- and down-regulated genes were defined as a log2 transformed fold-change (logFC) ≥ log2(1.5) or ≤ -log2(1.5), respectively. A P value ≤ 0.05 was considered as significant.

Table 2.

Grouping the lymphoma datasets for data integration and global renormalization

GEO ID	Category	Platform	Batch
GSE4475	Burkitt’s lymphoma	GPL96	1
GSE21452	Mantle cell lymphoma	GPL570	2
GSE31312	Diffuse large B-cell lymphoma	GPL570	3
GSE36000	Mantle cell lymphoma	GPL570	4
GSE56315	Diffuse large B-cell lymphoma	GPL570	5
GSE64555	Diffuse large B-cell lymphoma	GPL570	6
GSE69053	Burkitt’s lymphoma	GPL8432	7
GSE69053	Diffuse large B-cell lymphoma	GPL8432	8
GSE69053	Burkitt’s lymphoma	GPL14951	9
GSE69053	Diffuse large B-cell lymphoma	GPL14951	10
GSE70910	Mantle cell lymphoma	GPL570	11
GSE86613	Diffuse large B-cell lymphoma	GPL570	12
GSE93291	Mantle cell lymphoma	GPL570	13
GSE56315	Control	GPL570	14

Gene set enrichment analysis

We used javaGSEA Desktop Application v2.2.4 to perform gene set enrichment analysis (GSEA) of the three types of lymphoma [26]. KEGG gene sets v6.0 (including 186 gene sets) was chosen to perform KEGG pathway enrichment analysis in BURK, DLBCL and MCL. The analysis parameters were set as follows: gene sets containing fewer than 15 genes or more than 500 genes were excluded, a phenotype label was set as case vs. control, and the t-statistic mean of the genes was computed in each KEGG pathway using a permutation test with 1000 replications. The up-regulated pathways were defined by a normalized enrichment score (NES) > 0 and the down-regulated pathways were defined by an NES < 0. Pathways with a P value ≤ 0.05 were chosen as significantly enriched.

Transcriptional regulation network analysis

We downloaded 1,691 human transcription factors (TFs) from the AnimalTFDB 2.0 web server (http://bioinfo.life.hust.edu.cn/AnimalTFDB/) [27] and mapped 915 TFs to our lymphoma datasets. GENIE3 (Gene Network Inference with Ensemble of trees) [28] was used to reconstruct transcriptional regulatory networks for BURK, DLBCL and MCL. We used the gene expression matrix of the three types of lymphoma and the 915 mapped transcription factors as the input data and ran GENIE3 with its default parameters. Next, we extracted the top 10,000 interactions from the original output and used Cytoscape v3.4.0 to visualize the output results. The NetworkAnalyzer tool in Cytoscape was used to analyze the three resultant networks.

Tissue-specific network analysis

GIANT web server (http://giant.princeton.edu/) [29] was implemented to construct genome-scale integrated analysis of gene networks in BURK, DLBCL and MCL. We filtered the TFs with nodes containing greater than 100 in transcriptional regulatory networks of the three types of lymphoma, and then used these TFs in each type of lymphoma as input parameters to perform gene network analysis. The tissue option chosen for the GIANT web server was “lymph node”. The server generated a network of query genes and target genes, as well as performing GO Biological Process enrichment analysis of the genes in the network. GO Biological Processes with a P value ≤ 0.05 was considered as significantly enriched.

Identify key regulators and target genes

We filtered key transcription factors based on the relationship confidence in the GIANT network in BURK, DLBCL and MCL, respectively. A Venn diagram was used to screen the differentially expressed target genes in the GIANT networks in the three types of lymphoma. We used univariate linear model (the “lm” function in R) to calculate the correlation between the filtered key transcription factors and differentially expressed target genes in each type of lymphoma. The absolute value of correlation coefficient ≥ 0.5 and a P value ≤ 0.05 was considered as significant. We used the “plot” function in R to show the log2 transformed fold changes and the correlation coefficients of each paired transcription factors and target genes in the three types of lymphoma.

Results

Differentially expressed genes in three types of lymphoma

In total, we found 11,511 genes in the integrated lymphoma datasets. Differentially expressed genes in the three types of lymphoma were illustrated in Table 3. We obtained 216 up-regulated and 205 down-regulated genes, 198 up-regulated and 89 down-regulated genes, as well as 230 up-regulated and 67 down-regulated genes in BURK, DLBCL, and MCL, respectively. The top 20 differentially-expressed genes in three types of lymphoma were shown in Table S1. In addition, we noted that 178 common genes were deregulated in the three types of lymphoma (Figure S3A).

Table 3.

Differentially expressed genes in three types of lymphoma

Lymphoma	Cases/Comtrols	Up-regulated	Down-regulated
Burkitt’s lymphoma	352/33	216	205
Diffuse large B-cell lymphoma	880/33	198	89
Mantle cell lymphoma	216/33	230	67

Commonly and differentially deregulated KEGG pathways

We screened the pathways that were significantly enriched in at least one type of lymphoma. The results were shown in Figure 1. There were 13 commonly enriched pathways (10 up-regulated and 3 down-regulated) in BURK, DLBCL, and MCL. These pathways include cell cycle, amino sugar and nucleotide sugar metabolism, PPAR signaling pathways, pyrimidine metabolism, Fc gamma R-mediated phagocytosis, N-glycan biosynthesis and other cancer-related pathways. Furthermore, we observed tyrosine metabolism, GnRH signaling pathway, vascular smooth muscle contraction, neuroactive ligand-receptor interaction, fatty acid and other lipid metabolism were common down-regulated in BURK and DLBCL, whereas these pathways showed no change in MCL (Figures 1 and S4). In addition, the JAK-STAT signaling pathway, adherents junction and chemokine signaling pathways were commonly deregulated in DLBCL and MCL, whereas these pathways did not reach significant difference in BURK (Figures 1 and S4). Furthermore, there were 16, 10 and 5 pathways that were only significantly enriched in BURK, DLBCL, and MCL, respectively. These results suggest a relatively large difference in these three types of lymphoma.

Figure 1.

GSEA results of enriched KEGG pathways in Burkitt’s lymphoma, diffuse large B-cell lymphoma and mantle cell lymphoma.

Expression profiles of transcription factors and transcriptional regulatory networks

There were 29, 14 and 16 differentially expressed transcription factors in BURK, DLBCL, and MCL, respectively (Figure 2A). Among them, 11 deregulated transcription factors were shared by all the three types of lymphoma (Figure S3B). Transcriptional regulatory networks (TRNs) of three types of lymphoma were shown in Figures S5, S6, S7. Network properties of the three TRNs were listed in Table S2. We found that MCL TRN had the highest clustering coefficient of 0.120; the clustering coefficient in BURK TRN was 0.049 and in DLBCL it was 0.079. Furthermore, MCL TRN showed the highest connected component of 101, followed by DLBCL TRN (76) and BURK TRN (56). In addition, MCL TRN had the most multi-edge node pairs at 130, followed by DLBCL TRN (89) and BURK TRN (68). We sorted transcription factors for each of the TRNs according to the number of nodes, and then extracted the transcription factors with more than 100 nodes. In total, we got 8, 11 and 24 top transcription factors for BURK, DLBCL, and MCL, respectively (Figures 2B-D, S5, S6, S7). We revealed that no top TF was shared by all three types of lymphoma, only E2F8 was shared by BURK and MCL, and CDX2 was shared by BURK and DLBCL (Figure 2E). We combined these differentially expressed TFs and top-node TFs in each type of lymphoma and found the intersected TFs in BURK was E2F8, and in MCL were PRDM1, E2F8, VDR and IRF4 (Figure S8). No deregulated top-node TF was found in DLBCL.

Figure 2.

Differentially expressed and multiple connected transcription factors in three types of lymphoma. A. log2 transformed fold change of all transcription factors in three different types of lymphoma. B-D. Transcription factors with nodes ≥ 100 in Burkitt’s lymphoma, diffuse large B-cell lymphoma and mantle cell lymphoma, respectively. E. Venn diagram of transcription factors with nodes ≥ 100 in three types of lymphoma.

GIANT networks and enriched biological functions

Figure 3 shows the GIANT networks of top-node TFs and their biological functions in each of the three types of lymphoma. We listed the top 20 enriched GO Biological Processes in each type of lymphoma (only 5 significantly enriched items in DLBCL). In BURK, only E2F8 showed the highest relationship confidence with target genes. In DLBCL, there were 6 genes (YY1, STAT1, HMGB1, PMS1, RNF138 and YBX1) which showed a relatively high relationship confidence with target genes. Furthermore, WHSC1, E2F8, HMGB3, E2F1 and FOXM1 had the highest relationship confidence with target genes in MCL. The enriched biological functions in BURK were mainly cell division, mitosis, microtubule cytoskeleton organization, nuclear division, chromosome segregation and other cell cycle related functions (Figure 3A). In DLBCL, there were 5 enriched DNA replication and transcription related biological processes. In addition, several immune-related biological functions (interleukin-13 production, interleukin-4 production, vitamin D receptor signaling pathway and other functions) and cell cycle related functions were significantly enriched in MCL. Based on the GIANT networks, we therefore speculate that these top-node TFs may play key roles in BURK and MCL.

Figure 3.

Genome-scale integrated analysis of gene networks in Burkitt’s lymphoma (A), diffuse large B-cell lymphoma (B) and mantle cell lymphoma (C). The top 20 enriched biological functions of the genes in the networks are shown in the right panel.

Key transcription factors and their target genes in three types of lymphoma

We combined target genes in the GIANT network and deregulated genes in BURK, DLBCL and MCL (Figure 4A-C). There were 15, 3 and 8 target genes that were deregulated in BURK, DLBCL and MCL, respectively. Based on the above results, we calculated the correlations between the key TFs and deregulated target genes in each type of lymphoma. Interestingly, these target genes were all up-regulated. In BURK, there were significant correlations between E2F8 and 15 downstream target genes (Figure 4D). However, the key TFs in DLBCL showed no expression changes and relatively lower correlations with target genes (Figure 4E). Furthermore, the key TFs and target genes in MCL presented the highest correlations of all (Figure 4F). Among these key TFs, E2F8 was significantly up-regulated, while E2F1 and HMGB3 showed boundary significance.

Figure 4.

Correlation of key transcription factors and deregulated target genes in three types of lymphoma. A-C. Venn diagram of target genes in GIANT network and deregulated genes in Burkitt’s lymphoma, diffuse large B-cell lymphoma and mantle cell lymphoma, respectively. D-F. Correlation of key transcription factors and deregulated target genes in Burkitt’s lymphoma, diffuse large B-cell lymphoma and mantle cell lymphoma, respectively. The black/yellow color bar indicates the correlation coefficients of key transcription factors and deregulated target genes. The blue/red color bar indicates the log2 transformed fold change of these genes. Significantly differentially expressed genes are indicated by an asterisk.

Possible regulatory mechanisms in three types of lymphoma

Through integration the results of differentially expressed genes, affected pathways and biological functions, transcriptional regulation networks and TF-target interactions, we propose the possible regulatory mechanisms in each type of lymphoma (Figure 5). In BURK, the up-regulated E2F8 activated 15 downstream target genes, and these genes further affected multiple cell cycle related pathways, eventually resulting in cancer. However, in DLBCL, the effect of these transcription factors may not be carcinogenic. Furthermore, in MCL, the highly-expressed E2F8, E2F1 and HMGB3 activated 8 downstream target genes, and influenced cell cycle and immune functions, also resulting in cancer.

Figure 5.

Possible regulatory mechanisms of key transcription factors, target genes, and pathways in three types of lymphoma. A. Up-regulated E2F8 activated several target genes and then affected microtubule cytoskeleton organization, mitotic cell cycle, cell division, and other cancer-related pathways, finally caused Burkitt’s lymphoma. B. The expression of these transcription factors may have less effect on diffuse large B-cell lymphoma. C. Up-regulated HMGB3, E2F8 and E2F1 activated several target genes and then affected cell cycle process, mitosis, interleukin-13 production, and other cancer-related pathways, finally resulting in mantle cell lymphoma.

Discussion

In the present study, we observed there were 13 commonly enriched pathways and 31 specific enriched pathways in BURK, DLBCL and MCL. Furthermore, through reconstruction of TRNs and GIANT networks analysis in the three types of lymphoma, we identified several key TFs and their target genes in BURK, DLBCL and MCL, respectively. In particular, E2F8 was up-regulated in all the three types of lymphoma and was also the top TF (nodes ≥ 100) in BURK and MCL. We further found that several E2F8 target genes were over-expressed in BURK and MCL, and that these genes play crucial roles in cell cycle and immune-related functions. However, the top TFs may not cause a carcinogenic effect in DLBCL.

Previous studies have demonstrated some important pathways result in a carcinogenic effect. Multiple tumor-related metabolic pathways (energy metabolism, amino acid metabolism, fatty acid metabolism and choline phospholipid metabolism) were identified in BURK mouse models [30]. Fatty acid oxidation and up-regulated carnitine palmitoyltransferase I (CPTI) have been found to promote cancer growth, regulate gene expression and apoptosis in most of cancers, including BURK and DLBCL [31]. In addition, up-regulated fatty acid synthase (FASN) affected lipogenesis and multiple signaling pathways in several cancer cell lines and was suggested as a potential therapeutic target [32]. Our study showed several lipid metabolism pathways (fatty acid metabolism, glycerolipid metabolism, steroid hormone biosynthesis, etc.) were significantly down-regulated in BURK and DLBCL, whereas these pathways showed no change or were slightly elevated in MCL (Figure 1). The JAK-STAT signaling pathway is another tumor-related signaling pathway. Deregulation of JAK-STAT has been found in several lymphoma cell lines [33,34], primary mediastinal B-cell lymphoma [35], DLBCL [36], MCL and follicular lymphoma [37]. Furthermore, the JAK-STAT signaling pathway was also found to be influenced in other hematological malignancies (leukemia, multiple myeloma, myeloproliferative neoplasms, etc.) and suggested that several JAK inhibitors have broad clinical application prospects [38]. In our study, the JAK-STAT signaling pathway was significantly up-regulated in DLBCL and MCL, whereas it exhibited no change in BURK. Furthermore, our study also showed several significantly down-regulated amino acid metabolism pathways (tyrosine metabolism etc.), sensory systems (olfactory transduction and taste transduction) and xenobiotic biodegradation and metabolism (drug metabolism - cytochrome P450 and metabolism of xenobiotics by cytochrome P450) in BURK and DLBCL. However, all these pathways showed no change or even modest increases in MCL. Based on these findings, we propose that these impairment pathways have a certain carcinogenic effect and vary widely in different types of lymphoma.

This study identified 11 up-regulated TFs in all the three types of lymphoma. Among them, PRDM1 and IRF4 were most over-expressed and also found in the top 20 list of differentially expressed genes (Figure 2 and Table S1). Most of these TFs play dominant roles in TRNs and their target genes are related to cell cycle, mitosis, immune and other biological functions (Figures 3 and S5, S6, S7). PRDM1 belongs to the PRDM gene family of transcription repressors containing Kruppel-type zinc fingers. The SET-related PR (PRDI-BF1-RIZ) domain is a master transcriptional regulator for terminal differentiation of B cells into plasma cells [39,40]. Previous studies have demonstrated that PRDM1 is over-expressed in Hodgkin lymphomas [39], T-cell lymphoma [41] and DLBCL [40,42]. Furthermore, other studies have demonstrated that over-expression of miR-9 or let-7a reduced PRDM1 levels in multiple cell lines and thus these factors are suggested as potential therapeutic targets for lymphomas [39,40]. IRF4 (also called MUM1) is a lymphocyte-specific member of the interferon regulatory factor (IRF) family of transcription factors and plays crucial roles in the activation of innate and adaptive immune systems [43]. IRF4 is involved in the t(6;14)(p25;q32) translocation, which causes the juxtaposition of the IRF4 gene (mapping at 6p25) to the IgH locus on 14q32 [43]. Over-expression of the IRF4 is detected in multiple myelomas, HIV-related lymphomas, lymphoplasmacytic lymphomas, peripheral T-cell lymphomas, DLBCL and MCL [43-46] and may represent a possible therapeutic target [46]. Furthermore, we also identified several top-node TFs in the TRNs of the three types of lymphoma, including E2F1, E2F8, HMGB3 etc. A previous study found E2F1 is highly expressed in BURK cell lines and sporadic BURK specimens, suggesting E2F1 as a promising therapeutic target for sporadic BURK [47]. However, recent research showed over-expression of E2F1 has no relationship with prognosis in T lymphoblastic lymphoma or leukemia [48]. The expression of E2F1 in this study was also up-regulated in BURK, DLBCL and MCL, despite all did not reach a significant difference.

E2F8 is a recently identified member of the E2F family. It controls cell cycle genes through homodimers or heterodimers from DNA-binding domains [49]. Furthermore, E2F8 plays a variety of regulatory roles including regulation of placenta architecture and embryonic development, mediation of the DNA damage-correlated cell-cycle, promotion of angiogenesis and lymphangiogenesis, and also influences polyploidization [49]. The traditional view is that E2F8 functions as a transcriptional repressor. Its mediated transcriptional repression is a critical mechanism during postnatal liver development [50] and it also represses a large set of E2F target genes in skin cancer [51]. However, new perspectives noted that E2F8 acts as a transcription activator which confers cisplatin resistance by promoting MASTL mediated mitotic progression in ER+ breast cancer cells [52]. Another breast cancer study demonstrated that E2F8 transcriptionally upregulates CCNE1 and CCNE2 via directly interacting with their respective gene promoters, which accelerates the transition from G1 to S phase of breast cancer cells [53]. Furthermore, a recent study suggested that E2F8 promotes papillary thyroid cancer progression by activating CCND1 and other cell cycle related genes [54]. In our study, E2F8 was up-regulated in all the three types of lymphoma, and was the most important TFs in BURK and MCL TRNs (Figures 2, S5 and S7). In addition, over-expressed E2F8 target genes influenced cell cycle and disrupted immune functions in BURK and MCL (Figure 4). However, to our knowledge, no other study reports the role of up-regulated E2F8 in lymphoma. Our study also discloses that HMGB3 is a key regulator in MCL, although the expression did not achieve significance (Figure 4). High mobility group-box 3 (HMGB3) is a member of the HMG superfamily and is classified with HMGB1 and HMGB2 into the HMGB subfamily. It plays important roles in DNA recombination, repair, replication, and transcription [55]. Recent studies showed that high-expression of HMGB3 promotes growth and migration in colorectal cancer [56] and reduces patient survival in esophageal squamous cell carcinoma [57]. Knockdown of HMGB3 also showed a potential therapeutic effect in gastric cancer cells [55]. However, no studies have reported the association between HMGB3 and lymphoma so far.

In summary, this study identifies a series of commonly and specifically deregulated pathways in BURK, DLBCL and MCL, respectively. Furthermore, several key TFs may activate their target genes, influencing cell cycle, mitosis, immune functions and other oncogenic pathways in BURK and MCL. However, these top TFs may have less impact on DLBCL. It is noteworthy that we are the first to report that E2F8 plays a dominant role in BURK and MCL. Therefore, we suggest that E2F8 can be used as a biomarker and a potential therapeutic target for BURK and MCL. Future studies are required to uncover its potential mechanisms.

Disclosure of conflict of interest

None.