Oncogenic MYC activation promotes proliferation in Burkitt lymphoma, but also induces cell-cycle arrest and apoptosis mediated by p53, a tumor suppressor that is mutated in 40% of Burkitt lymphoma cases. To identify molecular dependencies in Burkitt lymphoma, we performed RNAi-based, loss-of-function screening in eight Burkitt lymphoma cell lines and integrated non-Burkitt lymphoma RNAi screens and genetic data. We identified 76 genes essential to Burkitt lymphoma, including genes associated with hematopoietic cell differentiation (FLI1, BCL11A) or B-cell development and activation (PAX5, CDKN1B, JAK2, CARD11) and found a number of context-specific dependencies including oncogene addiction in cell lines with TCF3/ID3 or MYD88 mutation. The strongest genotype–phenotype association was seen for TP53. MDM4, a negative regulator of TP53, was essential in TP53 wild-type (TP53wt) Burkitt lymphoma cell lines. MDM4 knockdown activated p53, induced cell-cycle arrest, and decreased tumor growth in a xenograft model in a p53-dependent manner. Small molecule inhibition of the MDM4–p53 interaction was effective only in TP53wt Burkitt lymphoma cell lines. Moreover, primary TP53wt Burkitt lymphoma samples frequently acquired gains of chromosome 1q, which includes the MDM4 locus, and showed elevated MDM4 mRNA levels. 1q gain was associated with TP53wt across 789 cancer cell lines and MDM4 was essential in the TP53wt-context in 216 cell lines representing 19 cancer entities from the Achilles Project. Our findings highlight the critical role of p53 as a tumor suppressor in Burkitt lymphoma and identify MDM4 as a functional target of 1q gain in a wide range of cancers that is therapeutically targetable.
Targeting MDM4 to alleviate degradation of p53 can be exploited therapeutically across Burkitt lymphoma and other cancers with wild-type p53 harboring 1q gain, the most frequent copy number alteration in cancer.
Burkitt lymphoma is an aggressive B-cell lymphoma that is characterized by translocation of the MYC gene to immunoglobulin loci (1). Although oncogenic MYC promotes cell growth and proliferation, it also evokes failsafe mechanisms such as p53 activation that have to be overcome for transformation (2). About 40% of Burkitt lymphoma acquire TP53 mutations evading MYC-induced stress signals (3, 4).
Recent mutational cartography efforts in Burkitt lymphoma identified additional recurrent mutations in TCF3, ID3, GNA13, RET, PIK3R1, DDX3X, FBXO11, and the SWI/SNF genes ARID1A and SMARCA4 (5–8). Burkitt lymphoma also display copy number alterations (CNA) in addition to the MYC translocation, targeting chromosomes 1q, 13q31, 17p13 (including TP53), and 9p21.2 (including CDKN2A; refs. 9, 10). A gain of 1q is found in 30% of Burkitt lymphoma and often affects large regions (11), which has contributed to the limited understanding of oncogenic mechanisms involved. The implications of these mutations and CNAs are currently unclear.
RNAi-based genomics screens allow querying of functional dependencies in an unbiased fashion and in high throughput. Using panels of representative cell lines, context-specific vulnerabilities have been linked to genetic and pathologic subgroups (12). The Achilles Project reported comprehensive screening data in 501 cell lines using RNAi (13, 14). While activating mutations caused direct oncogene addiction, as seen in cell lines with BRAF, KRAS, or PI3K mutation, secondary gene dependencies were observed for loss-of-function mutations in tumor suppressor genes, such as ARID1A (15). Integration of gene expression and drug sensitivity profiles may provide further insight into the molecular basis of diseases and might be used to tailor targeted therapies (16).
For a comprehensive dissection of molecular dependencies in Burkitt lymphoma, we performed a loss-of-function RNAi screen across a panel of genetically characterized Burkitt lymphoma cell lines and intersected our findings on genotype-specific essential genes with the genetic profile of a well-annotated patient cohort.
Materials and Methods
Raw shRNA read counts from the RNAi screen and scripts used for processing are available upon request.
Microarray data are available at ArrayExpress under the accession number E-MTAB-7134.
Supplementary Methods and Tables are available with the online version of this article.
BJAB, BL-2, CA46, Namalwa, Ramos, Raji, BL-41, DogKit, DG-75, and Gumbus were obtained from DSMZ; BL7, BL60, LY47 were provided by G.M. Lenoir (IARC); Salina, Seraphine, and Cheptanges were provided by A. Rickinson (Birmingham, UK); and 293T/17 by Stefan Fröhling (DKFZ). All cell lines were maintained under standard conditions. Cell line authentification was performed using Multiplex Cell Authentification and cell cultures were tested for contamination and Mycoplasma using the Cell Contamination Test (Multiplexion).
RNAi screen and shRNA-mediated knockdown
The RNAi screen was performed as described previously (17) with modifications using the DECIPHER Human Module I pooled lentiviral shRNA library (#DHPAC-M1-P) targeting 5,045 genes in key signaling pathways with four to five shRNAs per gene (Cellecta). shRNA representation was determined two and 14 days posttransduction using high-throughput sequencing. P values for shRNA depletion were calculated using the edgeR package (18) and collapsed into gene scores using weighted Z-transformation (19). P values for differential shRNA viability effects were calculated as described previously using public software and collapsed into gene scores using Kolmogorov–Smirnov statistics (https://software.broadinstitute.org/GENE-E/index.html). RNAi results in non-Burkitt lymphoma cell lines screened with the same library were provided by Cellecta as raw read counts and genome-wide RNAi results in 216 cell lines were publically available as log2-transformed shRNA fold changes (13). Single shRNAs were coexpressed with RFP constitutively from the pRSI12-U6-(sh)-UbiC-TagRFP-2A-Puro vector backbone. shRNA cytotoxicity was determined by transduction of 50% of cells and relative RFP-loss compared with a scrambled shRNA (shNT).
Genetic annotation of cell lines
Mutations in Burkitt lymphoma cell lines were identified from genomic DNA using a self-designed amplicon panel (20) or from RNA sequencing on the Illumina HiSeq2000. Sequences were mapped against the human reference genome hg19 using the STAR alignment tool. Mutations were called as described previously (21). Genetic information for non-Burkitt lymphoma cell lines was extracted from Cancer Cell Line Encyclopedia (CCLE; https://portals.broadinstitute.org/ccle/home) and COSMIC (GDSC, http://www.cancerrxgene.org/).
Total RNA was isolated with RNeasy Mini Kit (Qiagen) and on-column DNase I (Qiagen) digestion. RNA was reverse-transcribed by SuperScript III First-Strand Synthesis SuperMix (Invitrogen) and quantified using QuantiFast SYBR Green RT-PCR (Qiagen) or Power SYBR Green Master Mix (Applied Biosystems) on a LightCycler 480 Real-Time PCR System, software v1.5 (Roche Applied Sciences).
Antibodies were from Merck Millipore (anti-MDM4 04-1555; anti-MDM2 OP46), abcam (anti-GAPDH, ab9485), BD Pharmingen (anti-p53 554294), Cell Signaling Technology [anti-cleaved PARP 9546; anti-mouse IgG DyLight800 5257; anti-rabbit IgG (H+L) DyLight680 5366], or Santa Cruz (anti-p21 556431; anti-PUMA sc-28226). The LI-COR Odyssey Infrared Imaging System (Cell Signaling Technology) was used for detection and ImageJ (NIH) for band quantification.
CRISPR/Cas9 gene knockout
sgRNAs were coexpressed with Cas9 from lentiCRISPRv2 (Addgene, plasmid #52961). Seraphine cells with effective p53 knockout were selected using puromycin and Nutlin-3.
Cells were incubated for 2 hours with BrdUrd and analyzed in flow cytometry using anti-BrdUrd-APC and 7-AAD from the BrdU Flow Kit (552598; BD Pharmingen).
Gene expression profiling
Total RNA of cell cultures with >80% shRNA+/RFP+ cells was hybridized on a Illumina BeadChip HumanHT-12-v4 containing >47,000 probes for 31,000 annotated human genes. Gene set enrichment analysis (GSEA) was performed for C2 and H gene sets from the MSigDB database using software provided by the BroadInstitut (http://software.broadinstitute.org/gsea/msigdb; ref. 22).
Animal studies were performed in agreement with the Guide for Care and Use of Laboratory Animals published by the US NIH (NIH Publication no. 85–23, revised 1996), in compliance with the German law on the protection of animals, and with the approval of the regional authorities responsible (Regierung von Oberbayern). The in vivo experiments were performed as published previously (23). Briefly, Seraphine-TP53wt, Seraphine-TP53ko, and Raji cell lines were infected in vitro with shNT or shMDM4 aiming at >80% transduction efficiency. A total of 1 × 107 cells were subcutaneously injected into flanks of immunodeficient mice. Tumor growth was monitored by FDG-PET after 11 or 16 days depending on the graft efficiency and mice were sacrificed.
ATP-based growth assay
Cell content of DMSO and drug-treated cells was determined by ATP level after 48 hours incubation using CellTiter-Glo luminescent assay (Promega) as described (24). After normalization to DMSO, IC50 values were calculated with GraphPad Prism using nonlinear regression to fit the data to the log(inhibitor) versus response (variable slope) curve as described in the manual of the software.
Genetic profile of primary Burkitt lymphoma patients
CNAs were analyzed by CGH using a BAC/PAC array consisting of 2799 DNA fragments as described elsewhere (25, 26) and by SNP array (GSE21597). Interphase FISH analysis was performed on paraffin-embedded or frozen tissue sections to determine MYC, BCL2, and BCL6 translocations to IG regions. TP53 mutations were determined by DHPLC and sequencing of exons 4 to 10 of the coding region (27). The expression data of primary samples was downloaded from Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo, GSE43677). Patients were classified into Burkitt lymphoma, DLBCL, and an intermediate group based on a previously described molecular signature (28). For all samples, tumor cell content exceeded 70%. The study was performed as part of the “Molecular Mechanisms in Malignant Lymphomas” Network Project of the Deutsche Krebshilfe and was approved by a central ethics commission (University Hospital, Göttingen, Germany). Written informed consent was obtained in accordance with the Declaration of Helsinki.
Landscape of essential genes in Burkitt lymphoma
To identify therapeutic targets in Burkitt lymphoma, we investigated molecular dependencies in Burkitt lymphoma cell lines using RNAi-based loss-of-function screening. We used a pooled shRNA library to silence 5,045 genes including members of signal transduction pathways, drug targets, and disease-associated genes with four to five shRNAs per gene and assessed changes in shRNA abundance after culturing the cells for 2 weeks (Fig. 1A). On average 24% of shRNAs were depleted at least two-fold and shRNAs targeting core essential complexes, including the ribosome and the proteasome, were specifically lost (68% and 47%, respectively; Fig. 1B). To evaluate the viability effect of individual gene knockdowns, we calculated weighted z-scores that combine the effect of shRNAs targeting the same gene and emphasize strong fold changes (18, 19). Common essential genes, as defined on the basis of previous RNAi screens (29), showed significantly lower scores compared with nonessential genes (P < 0.001, Fig. 1C). Notably, although a subset of genes was essential in all cell lines, we also observed cell line–specific viability effects (Supplementary Fig. S1A).
To investigate essential genes in the context of Burkitt lymphoma, we probed our data against RNAi screening results using the same set of shRNAs in six carcinoma cell lines (C4-2, DU145, PC3, R22v1, MDA-MB-231, A2780) and three cell lines of myeloid and lymphoid origin (AML193, THP1, U937; Supplementary Fig. S1B). We ranked shRNAs based on their differential effects between two cell line groups and calculated a gene classification score as a measurement of their strength to distinguish between the groups (Supplementary Table S1; ref. 12). We then selected genes that were predictors of an entity group and showed strong differential viability effects based on the weighted z-scores. To exclude core essential genes, gene scores in eight Burkitt lymphoma cell lines were first compared with the six carcinomas. We identified 76 genes essential in Burkitt lymphoma, including genes associated with hematopoietic cell differentiation (FLI1, BCL11A) or B-cell development and activation (PAX5, CDKN1B, JAK2, CARD11; Fig. 1D, left). We therefore investigated, if these viability genes were classifiers of Burkitt lymphoma or of the blood lineage (Supplementary Fig. S1C). Knockdown of FLI1, a transcriptional regulator of the hematopoietic system and B-cell development (30), was also toxic to blood-lineage derived non-Burkitt lymphoma cell lines, whereas PAX5, a marker of early B-cell development, was an essential gene exclusively in Burkitt lymphoma (Fig. 1D, middle/right).
Genotype-specific dependencies in Burkitt lymphoma
We next investigated essential genes in the context of a specific gene mutation. We performed RNA sequencing of the Burkitt lymphoma cell lines included in the RNAi screen, and compared essential genes in the respective genotype groups focusing on genes that are recurrently mutated in Burkitt lymphoma, such as TP53, ID3, TCF3, DDX3X, FOXO1, and GNA13 (Supplementary Table S2; refs. 5–8). Mutations in the transcription factor TCF3 lead to oncogene activation and loss-of-function mutations of its inhibitor ID3 are often observed as a complementary mechanism of TCF3 activation (7). Therefore, cell lines carrying either TCF3 or ID3 mutation were treated as one group. The four cell lines with TCF3/ID3 mutation were strongly dependent on TCF3 expression, indicating oncogene addiction (P < 0.01; Fig. 1E). In line with the loss-of-function effect of mutations in ID3, ID3 silencing was not toxic (Fig. 1E, left). The cell line BL2 harbors the activating p.S219C mutation in MYD88, an adaptor protein involved in Toll-like receptor signaling and NF-κB activation. shRNAs targeting MYD88 or its direct downstream mediator IRAK1 were specifically toxic in the MYD88mut context (Fig. 1F). Encouraged by the ability to uncover oncogene addiction, we expanded our analysis of genotype-specific vulnerabilities to DDX3X, FOXO1, GNA13, and TP53 (Supplementary Table S1; Supplementary Fig. S1D). TP53 mutation was associated with the strongest differential viability effects (gene classification scores >2; Supplementary Table S1) and we therefore focused on TP53-specific vulnerabilities.
p53 pathway susceptibilities in Burkitt lymphoma
We identified seven genes (MDM4, CDKN3, BRCA2, BHMT2, SRC, PPP2R1A, PPM1D) that were essential in TP53wt Burkitt lymphoma cell lines (Fig. 2A). Notably, as Epstein–Barr virus (EBV) associated proteins deregulate cell-cycle checkpoints and quench the p53 pathway by deubiquitination of the p53 inhibitor MDM2 (31), we confirmed a balanced distribution of EBV infection status among TP53wt and TP53mut Burkitt lymphoma cell lines (Supplementary Table S2). To test the p53-specificity in a larger set of cell lines, we analyzed gene effect scores in 19 TP53wt and 42 TP53mut cell lines of hematopoietic/lymphoid origin from a combined RNAi screen of the DepMap project (Fig. 2B; ref. 14). All candidate genes showed a trend towards lower gene effect scores in TP53wt cell lines. We did not identify robust vulnerabilities for the mutant p53 context (Fig. 2A; Supplementary Fig. S2). Genes with a significantly lower effect score in TP53mut cell lines of the DepMap project were associated with the TP53 pathway and portrayed a growth advantage to TP53wt cell lines (Supplementary Fig. S2A–S2D).
We chose the two most robust hits, MDM4 and CDKN3, for validation experiments. CDKN3 is a spindle checkpoint phosphatase essential for G1–S transition during the cell cycle (32). shRNAs targeting CDKN3 efficiently reduced CDKN3 mRNA level (Fig. 2C). Using two nonoverlapping shRNAs, we tested the screen findings in a growth competition assay in five TP53wt and seven TP53mut Burkitt lymphoma cell lines. shRNAs were coexpressed with red fluorescent protein (RFP) in nearly 50% of cells and the fraction of RFP+/shRNA+ cells was monitored over time. The knockdown of CDKN3 was toxic to 4/5 TP53wt cell lines (Fig. 2D). To further test whether the observed effects were dependent on p53, we generated a p53 knockout cell line based on the TP53wt cell line Seraphine (Supplementary Fig. S3A). The toxicity of CDKN3 knockdown was partially rescued with one shRNA in Seraphine-TP53ko (Fig. 2D).
MDM4 inactivates p53-mediated transcription by blocking of its transactivation domain (33). shRNAs targeting MDM4 efficiently reduced MDM4 mRNA and protein levels (Fig. 2E). The knockdown was toxic in 3/4 TP53wt cell lines, but not in seven TP53mut Burkitt lymphoma cell lines, and the effect was completely rescued in isogenic Seraphine-TP53ko with one shRNA and partially rescued with a second shRNA (Fig. 2F). The BL2 cell lines that was less responsive to CDKN3 and MDM4 knockdown carries a deletion of the CDKN2A locus encoding for p53 activator p14 and p16 and shows a lower basal p53 pathway activity, which might explain the milder effect (Supplementary Fig. S3B).
MDM4 promotes cell-cycle progression by p53 inactivation
To understand the downstream effects of MDM4 depletion in Burkitt lymphoma, we assessed protein levels of p53 and known p53 targets. MDM4 knockdown in TP53wt cells increased p53 protein level and induced the pro-apoptotic Bcl-2 family member PUMA and the cell-cycle inhibitor p21 (Fig. 3A). Because MDM4 downregulation did not cause apoptosis as determined by absence of PARP cleavage (Fig. 3A), we analyzed the cell-cycle profile in the presence or absence of functional p53 after MDM4 silencing. In the TP53wt context, shRNAs targeting MDM4 decreased cycling cells compared with a nontargeting shRNA (shNT, P < 0.001), which was not observed in the TP53mut cell line Raji and rescued in the Seraphine-TP53ko cell line (Fig. 3B). Further cell-cycle profiling in additional cell lines confirmed p53-specific induction of cell-cycle arrest following MDM4 knockdown (Supplementary Fig. S3C).
We next determined global gene expression changes after MDM4 and MDM2 silencing in the TP53wt and TP53ko Seraphine cell lines (Fig. 3C; Supplementary Table S3). Silencing of MDM4 or MDM2 induced strong changes only in the presence of p53 and affected similar pathways. Using gene set enrichment analysis for cancer hallmark genes (MSigDB), we identified p53 targets as the strongest upregulated pathway, whereas prominent survival and proliferation pathways, including MYC and E2F targets, were downregulated. These suggest that most effects were mediated by p53 activation, in accordance with a previous report on genes commonly regulated after MDM4 or MDM2 knockdown (34). We also compared genes differentially regulated by MDM2 or MDM4 silencing (Supplementary Fig. S4). Downregulation of MYC and upregulation of CCND1 were exclusively seen after MDM4 knockdown, indicating potential differences in pathway contribution exerted by MDM4 over MDM2.
We next examined the basal protein and mRNA expression levels of p53, MDM4, and MDM2 in a panel of Burkitt lymphoma models (Fig. 3D). p53 protein was detected at higher level in all TP53mut cell lines (P < 0.01) as described previously (35), whereas p53 mRNA levels were lower (P = 0.045). Wild-type p53 is rapidly turned-over in a negative feedback loop mediated by MDM2 and mutant p53 protein accumulates as a result of disrupted proteasomal decay (36). MDM4 mRNA was significantly higher in TP53wt Burkitt lymphoma cell lines (P = 0.027) and was correlated with protein expression (P < 0.01; Fig. 3D).
MDM4 is a therapeutic target in TP53wt Burkitt lymphoma
To evaluate the potential of MDM4 as a therapeutic target in TP53wt Burkitt lymphoma in vivo, we determined the effect of MDM4 silencing on tumor growth in a mouse xenograft model. After transduction, cell lines representing TP53wt (Seraphine), TP53ko (Seraphine-TP53ko) and TP53mut (Raji) were injected subcutaneously into the flanks of immunodeficient mice (23). To quantify tumor formation and dynamic growth, we measured fludeoxyglucose (FDG) uptake in positron emission tomography (PET). In vivo tumor formation was significantly reduced after MDM4 knockdown in the presence of wild-type p53 (P < 0.05; Fig. 4A and B).
Restoration of p53 activity is an attractive therapeutic approach for treatment of cancer (37). The small molecule inhibitor Nutlin-3 is targeting the p53 inhibitor MDM2 and therefore restores signaling through the p53 pathway (38). TP53wt Burkitt lymphoma cell lines were sensitive towards Nutlin-3 with an average IC50 value of 4 μmol/L, while the average IC50 for TP53mut cell lines was 27 μmol/L. The reduction in cell numbers was significantly stronger in TP53wt cell lines starting from a concentration of 1.11 μmol/L (1.11 μmol/L: P = 0.016 *, 3.33 μmol/L: P = 1.60e−04 ***, 10 μmol/L: P = 2.98e−06 ***, 30 μmol/L: P = 1.86e−03 **; Fig. 4C). We tested the specificity of Nutlin-3 in the isogenic cell lines Seraphine-TP53wt and Seraphine-TP53ko and observed an increase of p53 levels in the TP53wt cell line (Supplementary Fig. S3A) and p53-dependent induction of apoptosis using 10 μmol/L Nutlin-3 (Supplementary Fig. S3D).
Despite the high sequence homology of MDM2 and MDM4, Nutlin-3 targets MDM2 with a much higher binding affinity (39). Moreover, overexpression of MDM4 can lead to resistance against MDM2-targeting drugs (39). We therefore tested the dual-specificity inhibitor RO-5963, which targets MDM2 and MDM4 (40), and observed a higher sensitivity in TP53wt Burkitt lymphoma cell lines starting at a concentration of 1.11 μmol/L (1.11 μmol/L: P = 0.017 *, 3.33 μmol/L: P = 0.0014 **, 10 μmol/L: P = 0.002 **; Fig. 4D). The average IC50 in TP53wt cell lines was 4.6 μmol/L. The highest concentration tested was 10 μmol/L and IC50 was not reached for most TP53mut cell lines. These data provide a rational for targeting MDM4/2 in TP53wt Burkitt lymphoma.
Gain of MDM4 on chr1q provides an alternative to TP53 mutations in Burkitt lymphoma
To understand the role of the p53 pathway in Burkitt lymphoma, we analyzed the genetic profile of aggressive B-cell lymphoma patients classified into Burkitt lymphoma, diffuse large B-cell lymphoma (DLBCL), or cases with intermediate phenotype (Supplementary Table S4; ref. 28). TP53 mutations were found in 28/61 (45.9%) of Burkitt lymphoma samples and were significantly more frequent in Burkitt lymphoma than in DLBCL (P < 0.001; Fig. 5A). MYC box I mutations were previously reported to be mutually exclusive with TP53 mutations and to serve as an alternative mechanism to escape apoptotic pathways in the presence of wild-type TP53 (4). MYC mutations were present in 37/56 Burkitt lymphoma samples (66.1%) and the MYC box I residues 56 to 58 were affected in 20 (35.7%) cases (Fig. 5B). Notably, MYC box I mutations frequently co-occurred with TP53 mutations (Fig. 5B).
We next explored the profile of CNAs in Burkitt lymphoma stratified by TP53 mutation status (Fig. 5C). The most frequent gains were on 1q21-q23 (TP53wt: 39%/TP53mut: 20%), 1q24-q25 (32%/8%), 1q32.1 (29%/12%), 2p16.1 (23%/20%), 11q12.3-q13.1 (13%/20%), 6p22 (14.3%), and 3q27.3 (29%/36%%), and the most frequent loss was on 17p13 (4%/20%). Deletion of 17p13 included the TP53 gene and co-occurred with TP53 mutation in five of six cases, resulting in biallelic p53 inactivation. Notably, loss of the MDM2 inhibitor ARF (CDKN2A locus on 9p21.3), that has been described as an alternative mechanism of p53 inactivation in Burkitt lymphoma cell lines (41), was rare in primary Burkitt lymphoma biopsies (n = 1). Chr1q gain was the most frequent CNA in TP53wt Burkitt lymphoma, which was not seen in DLBCL (Supplementary Fig. S5A) or intermediated cases (Supplementary Fig. S5B), and besides of 1q21, chromosomal gains frequently affected 1q32, including the MDM4 locus (Fig. 5D).
As 1q gain affected a large region with further oncogenes, we tested if Burkitt lymphoma cell lines from the RNAi screen were more dependent on genes on 1q (Fig. 5E and F). The RNAi library covered 235 genes located on 1q including known oncogenes. All four TP53wt Burkitt lymphoma cell lines were previously reported to carry a 1q gain (42). In Seraphine, the whole chromosomal arm was affected (+1q21.1qter), whereas partial gains were seen in BL-2 (+1q21.1q31.3), LY47 (+1q43q44), and Seraphine (+1q21.1qter). The TP53mut cell lines were diploid for 1q (Supplementary Table S2). Genes on 1q were not enriched for viability genes in the group of TP53wt or TP53mut Burkitt lymphoma cell lines, respectively (Fig. 5E). Notably, MDM4 was the only gene showing TP53-specific viability effects after silencing (Fig. 5F).
Altogether, our data support a critical role for quenching of the p53 pathway in Burkitt lymphoma preferably by mutations of TP53 or amplification of MDM4, thereby identifying p53 signaling as the critical failsafe checkpoint in Burkitt lymphoma.
TP53 mutations and MDM4 gain inactivate the p53 pathway in primary Burkitt lymphoma
To study the functional consequences of p53 pathway aberrations, we generated a molecular signature that distinguished TP53wt and TP53mut B-cell non-Hodgkin Lymphoma (B-NHL, n = 430) using supervised hierarchical clustering (Fig. 6A). The gene CDKN2A was significantly repressed in TP53wt Burkitt lymphoma (P < 0.01), intermediate lymphoma (P < 0.01), and DLBCL (P < 0.01) samples (Fig. 6B). Within the 50 most differentially expressed gene probes with lower expression in TP53mut patients, 28 were located on chr17p13 and four gene probes were located on chr1q (Fig. 6A). These findings reflect the gene dosage effect as a result of chr17p13 deletion in TP53mut and chr1q gain in TP53wt patients. Nine probes corresponding to six p53 target genes were expressed in TP53wt samples, demonstrating that a portion of aggressive B-NHL retain active p53 signaling. Therefore, elevated MDM2 levels in TP53wt DLBCL (P < 0.01) and Burkitt lymphoma (P < 0.01) might be a consequence of a p53 activity (Fig. 6C). Notably, high MDM4 mRNA expression was specific to Burkitt lymphoma with TP53wt (P < 0.01, Fig. 6D). MDM4 expression was high in all Burkitt lymphoma with chr1q gain, but also in some TP53wt Burkitt lymphoma without 1q gain, indicating that additional mechanisms regulate MDM4 expression (Supplementary Fig. S6). Combined, these data provide evidence for upregulation of MDM4 in TP53wt Burkitt lymphoma as a disease driver.
MDM4 and TP53 mutation across cancer models
To investigate the role of chr1q gain in context of TP53 mutations across a range of cancer types, we analyzed the associations between genetic aberrations in 789 cell lines with available SNP6.0 data and TP53 mutation data within the Cancer Cell Line Encyclopedia (43). Chr1q32 gain was identified in 122 cell lines (15.5%) and was associated with wild-type p53 (P < 0.001, 23% in TP53wt and 12% in TP53mut) (Fig. 7A). We further combined genetic information with functional genomics data and investigated p53-dependent vulnerabilities in a set of 216 cell lines representing 19 cancer entities from the Achilles Project (13). TP53 and chr1q32 status were available for 182 cell lines. TP53 mutations were present in 70% of all cancer cell lines and chr1q32 was also significantly associated with TP53wt (P < 0.001; Fig. 7B; Supplementary Table S5). Notably, MDM4 was the top ranked gene leading to impaired viability of TP53wt cell lines out of more than 10,000 genes investigated (P < 0.001; Fig. 7C; Supplementary Table S6). All shRNAs targeting MDM4 were strongly depleted in TP53wt cell lines (Fig. 7D). MDM2 also showed significant shRNA depletion in TP53wt cell lines (P = 0.004, rank 51; Fig. 7C).
Eight cancer entities were represented with at least two TP53mut and two TP53wt cell lines, which allowed us to explore MDM4 dependency in different cancer subtypes (Fig. 7E; Supplementary Table S6). We observed entity-specific preference for MDM4 over MDM2: MDM4 was identified as an essential gene in TP53wt cell lines derived from the hematopoietic/lymphoid system (rank 1), large intestine (rank 3), breast carcinoma (rank 25), and ovarian carcinoma (rank 62; Fig. 7F). p53-specific dependency on MDM2 were strongest in ovarian carcinoma (rank 20) and CNS (rank 8; Fig. 7F). Combined these data suggest a functional role for MDM4 as a critical cancer driver targeted by 1q gain across cancers.
The combination of sequencing efforts and functional genomics serves as a powerful tool to understand the pathogenesis of diseases and to discover molecular targets. This study dissected specific vulnerabilities in Burkitt lymphoma using RNAi screening. We observed a strong dependency of Burkitt lymphoma on PAX5, a key B-cell transcription factor previously linked to B-cell lymphomagenesis (44), in accordance to a genome-wide CRISPR/Cas9 screen in two Burkitt lymphoma cell lines (45). These findings identify PAX5 as a “lineage-survival oncogene” (46) and demonstrate the power of genetic perturbation screens in dissection of “non-oncogene addictions” (47) that may not be predicted from the genetic profile. The increased capacity to drug transcription factors (48) and the recent demonstration of the role of PAX5 as a metabolic gatekeeper (49) suggests that PAX5 targeting may provide a novel therapeutic strategy.
Previously, a RNAi interference screen using a targeted shRNA library was used to characterize the B-cell receptor pathway in Burkitt lymphoma cell lines (7). This study also revealed gene mutation-specific dependencies and found Burkitt lymphoma lines rely on cyclin D3/CDK6 for cell-cycle progression and cyclin D3 mutants augment this effect. We add to these data by systematically querying genotype-specific vulnerabilities of Burkitt lymphoma. We identified oncogene dependency on TCF3 in TCF3/ID3 mutant Burkitt lymphoma, and dependency on MYD88 and IRAK1 in a cell line with MYD88 mutation, consistent with previous results in Burkitt lymphoma and DLBCL (7, 50). The strongest dependency was observed for MDM4 in TP53wt cell lines and further underscores the importance of suppressing p53-mediated stress signals in the pathogenesis of Burkitt lymphoma with activation of the MYC oncogene. Reactivation of p53 by inhibition of MDM4 is a promising therapeutic approach in melanoma (51) and breast carcinomas (52). We validated MDM4 as a potential target in TP53wt Burkitt lymphoma using a mouse xenograft model and showed effective p53-specific cytotoxicity for MDM2/MDM4 dual inhibition.
Chromosome 1q gain is the most frequent copy number across cancer (53), but functional evidence for the disease drivers affected by 1q gain has been lacking. Cytogenetic studies in Burkitt lymphoma identified gains for 1q25.1 and 1q31.3 and suggested PTPRC, a regulator of B-cell receptor and cytokine signaling, and two annotated miRNA genes (hsa-mir-181b-1 and -213) as strong candidates (9). A study of primary tumors and cell lines identified BCA2 and PIAS3 on 1q21.-1q21.3, MDM4 on 1q32.1 and AKT3 on 1q44 as possible drivers (42). In an unbiased approach, we now identified an association of 1q gain with wild-type p53 in primary Burkitt lymphoma, a finding not observed for DLBCL. Although DLBCL develops diverse mechanisms of p53 and cell-cycle deregulation (54), our genetic perturbation screen provides functional evidence that 1q gain and TP53 mutation are specifically selected for in Burkitt lymphoma to inactivate p53 activity. A pan-cancer analysis also revealed entity-specific dependency on MDM4 in TP53wt cancer cells with important clinical implications for p53 reactivating compounds.
MDM2 and MDM4 have been reported to be frequently deregulated in cancer [reviewed in Eischen and Lozano (55)]. We identified entity-specific preferences for MDM4 or MDM2 dependency. Our data suggest that among lymphomas, Burkitt lymphoma exhibits disease-specific mechanisms of p53 pathway suppression via TP53 mutation and MDM4 overexpression. A major open question pertains to the selective advantage of MDM4 or MDM2 overexpression in TP53wt cancers. MDM4 and MDM2 are highly homologous and closely interact to regulate the p53 pathway (55). In addition, p53-independent oncogenic activities were described for both proteins. MDM4, for example, was shown to promote pRb degradation by MDM2 and therefore enhances cell-cycle progression by E2F1 activation (56). In our study, we identified downregulation of MYC and upregulation of CCND1 after MDM4, but not MDM2 knockdown, indicating differences in pathway contribution exerted by MDM4 over MDM2 that need further exploration.
MDM2 overexpression by enhanced translation was described in TP53wt Burkitt lymphoma cell lines (41). In pediatric Burkitt lymphoma (pBL), which shows p53 mutations at a lower frequency than adult Burkitt lymphoma, MDM2 overexpression and p53 mutation accounted for 55% of cases (57). MDM4 mRNA was shown to be overexpressed in TP53wt pBL, some of which harbored a 1q gain (58). Our results extend these findings in adult Burkitt lymphoma.
Oncogenic MYC activation provokes p53-mediated apoptosis (2) and MYC-induced lymphomagenesis in transgenic mice is dependent on secondary lesions that promote survival (59). Mutations in the conserved Myc box I were shown to prevent the induction of apoptosis via Bim in a mouse xenograft model and to occur mutually exclusively to TP53 mutations in primary Burkitt lymphoma samples (4). In our study, however, TP53 mutations occurred independent of MYC box I mutations.
Based on the incidence of TP53 mutation and 1q gain in the disease, our findings suggest a widespread mechanism to suppress p53 activity in Burkitt lymphoma to overcome p53-mediated cell-cycle arrest and apoptosis caused by MYC overexpression. This provides critical biological and therapeutic rationale for targeting MDM4 in TP53 wild-type diseases.
Disclosure of Potential Conflicts of Interest
L.H. Trümper is a consultant/advisory board member of Takeda Pharma. R. Siebert has received speakers bureau honoraria from Roche and AstraZeneca. No conflicts of interest was disclosed by the other authors.
Conception and design: J. Hüllein, M. Słabicki, M. Rosolowski, M. Zapatka, R. Siebert, T. Zenz
Development of methodology: J. Hüllein, M. Słabicki, R. Scholtysik, D. Tedesco, W. Huber, T. Zenz
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Hüllein, M. Słabicki, A. Jethwa, S. Habringer, K. Tomska, S. Scheinost, R. Wagener, M. Lukas, R. Küppers, W. Klapper, C. Pott, S. Stilgenbauer, B. Burkhardt, L.H. Trümper, M. Hummel, M. Zapatka, R. Siebert
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Hüllein, M. Słabicki, M. Rosolowski, A. Jethwa, S. Habringer, R. Kurilov, J. Lu, Z. Huang, O. Yavorska, H. Helfrich, R. Scholtysik, D. Tedesco, S. Stilgenbauer, M. Hummel, B. Brors, M. Zapatka, R. Siebert, M. Kreuz, U. Keller, W. Huber, T. Zenz
Writing, review, and/or revision of the manuscript: J. Hüllein, M. Słabicki, M. Rosolowski, R. Wagener, S. Stilgenbauer, B. Burkhardt, M. Löffler, L.H. Trümper, M. Hummel, M. Zapatka, R. Siebert, M. Kreuz, U. Keller, T. Zenz
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Hüllein, R. Scholtysik, K. Bonneau, R. Küppers, M. Löffler, M. Hummel
Study supervision: M. Słabicki, T. Zenz
The work was supported by the Helmholtz Virtual Institute, “Understanding and overcoming resistance to apoptosis and therapy in leukemia,” the Helmholtz initiative iMed on Personalized Medicine, the European Union (FP7 projects Radiant, Systems Microscopy, Horizon 2020 project SOUND), and the “Monique Dornonville de la Cour – Stiftung.” The “Deutsche Krebshilfe” supported T. Zenz (“Mildred-Scheel” Professorship), M. Löffler (“Mildred-Scheel” Fellowship), the Monique-Dornonville de la Cour Stiftung and the “Molecular Mechanisms of Malignant Lymphoma – MMML” consortium. R. Scholtysik/R. Wagener received infrastructural support by the “KinderKrebsInitiative Buchholz Holm-Seppensen.”
We thank the microarray unit of the DKFZ Genomics and Proteomics Core Facility for providing the Illumina Whole-Genome Expression Beadchips and related services, and the high-throughput sequencing unit for providing RNA sequencing services. We thank Hanno Glimm, Stefan Fröhling, Daniela Richter, Roland Eils, Peter Lichter, Stephan Wolf, Katja Beck, and Janna Kirchhof for infrastructure and program development within DKFZ-HIPO and NCT POP, and Tina Uhrig for technical assistance and Agnes Hotz-Wagenblatt for shRNA alignment. We thank Anna Jauch for FISH analysis in Burkitt lymphoma cell lines. We thank Henry-Jacques Delecluse and Astrid Hofmann for staining of EBV proteins in Burkitt lymphoma cell lines to determine the EBV status and latency phase.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.