PRMT5 Is Required for Lymphomagenesis Triggered by Multiple Oncogenic Drivers

Arginine methylation is becoming increasingly appreciated as an important mechanism of post-transcriptional control (1). Proteins targeted by arginine methylation contribute to a variety of cellular processes, including transcriptional regulation, chromatin regulation, RNA processing, and DNA damage repair (2–4). The type II protein arginine methyltransferase PRMT5 has been most thoroughly characterized with regard to its function as a histone 3 and 4 methyltransferase. Methylation of H3R8 and H4R3 is associated with transcription repression (5–7). However, PRMT5 also targets multiple soluble proteins, including components of the spliceosome, PIWI proteins (8, 9), RELA (10), EGFR (11), E2F1 (12), and p53 (13), thereby potentially affecting multiple cellular signaling events.

Cyclin D1, together with its binding partners cyclin-dependent kinases 4 and 6 (CDK4/6), forms an active complex that promotes cell-cycle progression by phosphorylating and inactivating the retinoblastoma protein (14). Aberrant expression and/or regulation of the D cyclins (D1, D2, and D3) has been linked to loss of cell-cycle control and is considered a driving event in many malignancies. Accumulating evidence has implicated dysregulation of cyclin D1 nuclear export and ubiquitin-dependent degradation during S-phase as key events in the genesis of neoplastic events. Cyclin D1 nuclear export and polyubiquitylation depend upon phosphorylation of a specific threonine residue (Thr-286) (15). The oncogenicity of D1T286A, a constitutively nuclear mutant, has most thoroughly been examined in the context of the Eμ-D1T286A transgenic mouse model (16, 17). In this model, D1T286A expression is targeted to the lymphoid compartment by the immunoglobulin enhancer, thereby providing an expression pattern analogous to that observed in human mantle cell lymphoma (MCL; ref. 18). Analysis of early-stage tumors reveals that nuclear accumulation of D1T286A/CDK4 triggers DNA damage and activation of the ATM–CHK2–p53 checkpoint pathway, which leads to p53-dependent apoptosis (16, 19, 20). A latency period of 4 to 21 months is required for the accumulation of cooperating mutations to counter p53 surveillance before lymphoma can develop (19). The clinical relevance of mutations that specifically disrupt phosphorylation-dependent degradation and nuclear export of cyclin D1 is highlighted by their occurrence in human cancers (21, 22).

p53 is the central regulator of cell fate following numerous stresses, including genotoxic stress and oncogene activation (23, 24). The tumor suppressor properties of p53 have been linked to its function as a transcription factor that regulates the expression of target genes linked with cell-cycle arrest, apoptosis, senescence, and DNA repair (25, 26). Tumor suppressor activities of p53 have also been attributed to its capacity to target BAX at the mitochondria, thereby directly regulating proapoptotic functions in a transcription-independent fashion (27). More than 60% of human primary tumors exhibit mutations in the TP53 gene (28). In contrast, hematologic malignancies exhibit a low frequency of p53 mutation (29, 30), implicating the existence of alternative mechanisms for bypassing p53-dependent tumor suppression.

We provide evidence for a direct link between PRMT5-dependent arginine methylation of p53, reduced expression of proapoptotic p53 transcriptional targets, and hematologic malignancy. This mechanism is engaged by multiple drivers of hematologic malignancy, where it serves as a key regulatory event that directly alters promoter engagement by p53, providing a new mechanism by which a p53 modification contributes to neoplastic transformation.

To directly assess the potential of PRMT5 to drive neoplastic growth, we chose to first assess whether PRMT5 would cooperate with a cancer-derived allele of cyclin D1 to drive lymphomagenesis; this strategy was fueled by previous reports of PRMT5 overexpression in cyclin D1–driven malignancy (5). Initially, 5-fluorouracil (5-FU)–treated bone marrow hematopoietic stem/progenitor cells (HSPC) transduced with retroviral supernatants encoding PRMT5 and cyclin D1T286A were injected into lethally irradiated, syngeneic C57BL/6 mice. Surprisingly, recipient mice reconstituted with HSPCs overexpressing only D1T286A developed fatal pancytopenia with a remarkable reduction in the white blood cells, red blood cells, and platelet counts by 2 weeks after reconstitution (Fig. 1A; Supplementary Fig. S1A). The spleen and thymus of D1T286A reconstituted mice exhibited significant atrophy (Supplementary Fig. S1B). These results indicated failure of bone marrow reconstitution by D1T286A. However, all animals transplanted with cells coexpressing D1T286A and PRMT5 survived hematopoietic failure and succumbed to leukemia/lymphoma by 170 days, with a median survival age of 147 days (Fig. 1A). Macroscopic examination of tumor-burdened mice revealed thymic, splenic, and liver involvement; the involvement of peripheral blood leukocytosis and increased blast circulation in bone marrow was also readily apparent (Fig. 1B–D). Histologic analyses revealed extensive infiltration of lymphoblastoid cells within liver, spleen, thymus, lung, and kidney, and almost-complete effacement of the normal tissue architecture (Fig. 1E). D1T286A/PRMT5 chimeric mice (n = 7) exhibited accumulation of CD4⁺ lymphocytes in the bone marrow and spleen (Fig. 1F and G). Tumor cells were GFP⁺/nerve growth factor receptor (NGFR)⁺, demonstrating maintenance of transgenes (Fig. 1F). The tumors analyzed were CD3⁺TCR Vβ⁺ CD4⁺ CD8⁻ (Supplementary Fig. S2A and primarily CD25⁻ CD69⁻; Supplementary Fig. S2B), consistent with their identity as mature T cells. T-cell clonality was further assessed through both immunophenotypic analysis and PCR-based analysis of the T-cell receptor Vβ repertoire (TCR-Vβ-R) (Table S1; Supplementary Fig. S1D). Whereas CD4⁺ T cells from a wild-type mouse used a variety of Vβ chains as expected, those from the tumor-bearing mice did not exhibit outgrowth of a monoclonal TCR Vβ clone, suggesting that the tumors are oligoclonal. However, because these results could reflect technical issues pertaining to antibody selectivity, we further addressed the suggested oligoclonal nature of tumors. The clonality of the TCR repertoires of 22 individual Vβ gene families (from Vβ 1-20, with the subfamilies Vβ 8.1, 8.2, and 8.3) was assessed by a PCR amplification assay. An oligoclonal pattern was observed in all tumors derived from D1T286A+PRMT5 mice (Supplementary Fig. S1D). In addition, the CD4⁺ tumor cells have phenotypes of memory T cells (CD44^highCD62L^low; Supplementary Fig. S2C). Interestingly, PRMT5 alone was not sufficient for transformation (Fig. 1A; Supplementary Fig. S1C). The generation of mitotic spreads from dispersed tumors and normal lymphocytes revealed chromosomal gains (>40N) and increased chromatid breaks associated specifically with the tumor (Supplementary Fig. S2D–S2E), demonstrating that coexpression of PRMT5 had not reduced DNA damage associated with D1T286A expression (5).

To ensure that the phenotype reflected neoplastic growth, cells from the bone marrow of primary leukemia/lymphoma burdened mice were transplanted into sublethally irradiated secondary and tertiary recipients. All secondary recipients receiving more than 1 × 10⁵ cells succumbed to CD4⁺ leukemia/lymphoma, with an average latency of 62 days (1 × 10⁶, brown) and 79 days (1 × 10⁵, black; Supplementary Fig. S2F). Notably, 1 × 10⁴ cells were sufficient following secondary transplantation for disease manifestation, albeit with reduced penetrance (60%) and longer latency (15–20 weeks; blue, Supplementary Fig. S2F). Tertiary recipients died rapidly between 18 and 27 days (red, Supplementary Fig. S2F). The immunophenotype of the leukemia/lymphoma cells in secondary recipients is analogous with that of primary disease, with most cells retaining high expression of CD4 (Supplementary Fig. S2G). In addition, the tumor cells were LIN⁻, c-KIT⁺, and SCA1⁺ (Supplementary Fig. S2H). Collectively, this work demonstrates that PRMT5 can function as a driver oncogene in the context of nuclear cyclin D1.

Cyclin D1T286A-dependent regulation of PRMT5 reflects phosphorylation of MEP50 on Thr-5 (5). If MEP50 phosphorylation serves as the point of integration for D1T286A, then inhibition of MEP50 phosphorylation should inhibit tumorigenesis. Because endogenous MEP50 levels remain stable after D1T286A and PRMT5 transduction (Fig. 1H), we coexpressed either wild-type MEP50 or MEP50T5A (an alanine mutation previously shown to make PRMT5/MEP50 complexes refractory to regulation by cyclin D1/CDK4 (5), with D1T286A/PRMT5. D1T286A/PRMT5/MEP50^WT mice developed CD4⁺ leukemia/lymphoma with complete penetrance and reduced latency relative to D1T286A/PRMT5 (Fig. 2A–C; log-rank test P = 0.01). In striking contrast, mice reconstituted with D1T286A/PRMT5/MEP50T5A died within 2 weeks of transplant (Fig. 2A), similar to what occurred with D1T286A alone. This again likely reflects hematopoietic failure, as survival can be supported with normal bone marrow. Mice reconstituted without sorting (under these conditions 70% of cells were normal bone marrow) did not die and survived within the observation period. FACS revealed that D1T286A+MEP50T5A+PRMT5 cells are eliminated by 4 weeks after transplantation (Fig. 2D). These results support a model in which MEP50 phosphorylation serves as the point of integration of cyclin D1 with PRMT5 to drive neoplastic growth.

Given the critical function of PRMT5 in D1-dependent lymphomagenesis, we ascertained whether PRMT5 contributes to neoplastic outgrowth triggered by other oncogenic drivers such as NOTCH 1 [intracellular domain (ICN)], c-MYC, and MLL–AF9. Ectopic expression of PRMT5 failed to accelerate disease, increase the penetrance of leukemia/lymphoma, or alter the phenotype of tumors driven by the ICN domain of NOTCH1, c-MYC, or MLL–AF9 (Fig. 3A; Supplementary Fig. S3A–S3D). However, upon examination of PRMT5 levels in tumors versus normal splenic lymphocytes, we noted a significant increase in endogenous PRMT5 levels that greatly exceeded that achieved by retroviral transduction (Fig. 3B; Supplementary Fig. S3E).

The significant increase in endogenous PRMT5 highlighted a potential requirement for increased PRMT5 activity downstream of NOTCH1, c-MYC, and MLL–AF9. To specifically assess a potential requisite role for PRMT5 in the context of these oncogenic drivers, we utilized either a dominant-negative Prmt5 allele (PRMT5Δ) or an shRNA validated as Prmt5-specific (5). Prmt5 knockdown and PRMT5Δ overexpression were confirmed in transduced HSPCs before transplantation into irradiated mice (Fig. 3C and D). Surprisingly, Prmt5 knockdown and PRMT5Δ overexpression were without effect on the latency of disease (Fig. 3A). However, upon Western blot analysis, we noted that c-MYC– and ICN-driven tumors had restored PRMT5 expression in cells transduced with shPrmt5 (Fig. 3C; note PRMT5 expression after transplantation). Likewise, in tumors driven by cells transduced withMYC-tagged PRMT5Δ, PRMT5Δ expression was undetectable (Fig. 3D). Similar results were observed with MLL–AF9 (Supplementary Fig. S3F). These results demonstrate strong selection to maintain PRMT5 expression. To independently assess the requisite role for PRMT5 in neoplastic growth driven by c-MYC and ICN, we evaluated the impact of PRMT5Δ expression on colony expansion and hematopoietic stem cell (HSC) renewal in methylcellulose. Consistent with a requisite functional role in c-MYC– and ICN-driven tumor cell proliferation and survival, PRMT5Δ significantly inhibited colony formation during serial passage (Fig. 3E). Similar results were observed when PRMT5Δ was coexpressed with MLL–AF9 (Supplementary Fig. S3G). Collectively, these results suggest that PRMT5 plays a critical role in supporting neoplastic transformation in this setting.

The strong cooperative generation of a leukemic phenotype in mice transplanted with HSPCs coexpressing D1T286A and PRMT5 was in stark contrast to mice transplanted with HSPCs expressing only D1T286A, wherein all mice died within the first 14 days after transplantation (Fig. 1A). This result prompted us to carry out a more extensive analysis of the mechanistic contribution of PRMT5 to D1T286A-driven disease. Tumorigenesis driven by constitutively nuclear cyclin D1 mutants is opposed by p53-dependent apoptosis (16). Although apoptosis can be overcome through genetic ablation of Trp53 in mice (19), wild-type p53 is retained in 70% of human MCL cases (31). Sequencing of Trp53 in lymphomas that arise in Eμ-D1T286A single transgenic mice reveals retention of wild-type Trp53 in 40% of resultant tumors (16), suggesting the existence of uncharacterized mechanisms that permit the bypass of p53-dependent tumor suppression. To gain insights into the regulation of p53 in cyclin D1–driven neoplasia, we generated bone marrow chimeras using HSPCs derived from Trp53^+/+ or Trp53^−/− mice. HSPCs were transduced with retroviruses expressing wild-type cyclin D1 or D1T286A. In contrast to mice reconstituted with Trp53^+/+ HSPCs expressing D1T286A, which die by 14 days after reconstitution (Fig. 1A; Fig. 4A), the use of Trp53^−/− donors expressing D1T286A permitted hematopoietic reconstitution (Fig. 4A and B). To determine whether this reflected graft failure, we examined the contribution of D1T286A-expressing Trp53^−/− HSPCs to recipient bone marrow in a competitive setting. We transplanted a mixture of transduced (GFP⁺) and control HSPCs at around a 1:1 ratio into recipient mice and analyzed donor contribution to bone marrow cells 1 month later. Donor Trp53^+/+ cells transduced with cyclin D1 contributed to approximately 55% of bone marrow cells 1 month after transplant, whereas Trp53^+/+ cells transduced with D1T286A contributed to only 5% of recipient bone marrow cells (Fig. 4B). Therefore, D1T286A impairs the engraftment of HSPCs, and this is dependent on the presence of p53. Previous studies showed that expression of D1T286A causes p53-dependent apoptosis (19). Consistent with this, transduction of wild-type HSPCs with a retrovirus encoding D1T286A triggered extensive p53-dependent apoptosis, whereas infection with an equivalent multiplicity of infection of virus encoding wild-type cyclin D1 was without effect (Fig. 4C).

Mice transplanted with Trp53^−/− HSPCs that express D1T286A developed aggressive and widespread lymphomas involving the lung, liver, kidney, spleen, thymus, and bone marrow, with a significantly reduced latency compared with recipients reconstituted with Trp53^−/− HSPCs (P = 0.003 by log-rank test; Fig. 4D; Supplementary Fig. S4A–ES4). Mice reconstituted with Trp53^−/− HSPCs expressing cyclin D1 developed malignancies not significantly different from those transplanted with Trp53^−/− HSPCs in latency, frequency, and molecular phenotype (P = 0.36; Fig. 4D; Supplementary Fig. S4A), demonstrating that wild-type D1 provided no overt oncogenic activity even on the Trp53^−/− background. Malignancies were not observed in mice transplanted with Trp53^+/+ or Trp53^+/+ cells expressing D1 (Fig. 4D). Critically, mice transplanted with syngeneic HSPCs co-overexpressing a dominant-negative allele of Trp53 (p53DN) and nuclear D1T286A developed CD4⁺ lymphoma by 180 days (Fig. 4E; Supplementary Fig. S4F). Taken together, these data suggest that p53 inactivation contributes to sustained growth in the presence of D1T286A, thereby facilitating tumorigenesis.

The synergism of PRMT5 and D1T286A in cells harboring wild-type p53 prompted us to ascertain whether PRMT5 attenuated p53-dependent apoptosis. Coinfection of HSPCs with viruses encoding PRMT5 and D1T286A reduced apoptosis of D1T286A-expressing cells to a similar degree as a dominant-negative p53DN (Fig. 5A). Expression of D1T286A and PRMT5 was confirmed by Western blot analysis and expression of the IRES-linked GFP and mCherry markers (Supplementary Fig. S5A–S5B). Expression of a catalytically inactive PRMT5, PRMT5Δ (32), failed to protect D1T286A HSPCs from death (Fig. 5A), thereby demonstrating the requirement for PRMT5 methyltransferase function.

These findings support a model in which PRMT5 inactivates p53, thereby promoting survival of D1T286A-expressing cells. To further interrogate this model, the ability of bone marrow–derived HSPCs to form colonies in methylcellulose was determined (Fig. 5B). Both PRMT5 and p53DN increased the number of colonies generated by D1T286A-expressing cells through 5 rounds of serial replating. These results demonstrate that increased PRMT5 methyltransferase activity is as effective as dominant-negative p53 with regard to inhibiting D1T286A-triggered apoptosis and increasing self-renewal and cell transformation by D1T286A in vitro (Fig. 5B) or in vivo (Fig. 1A).

MEP50, being the direct substrate of D1T286A/CDK4, should be required for inhibition of apoptosis. Expression of wild-type MEP50 decreased D1T286A-dependent apoptosis, as did coexpression of MEP50 and PRMT5. In contrast, MEP50T5A failed to inhibit D1T286A-induced apoptosis (Fig. 5C). Consistent with this, expression of MEP50T5A inhibited the PRMT5-dependent increase in serial replating ability (Fig. 5D). These data are consistent with the impact of MEP50T5A on D1T286A/PRMT5–driven lymphoma (see Fig. 2A) and suggest a model in which D1T286A-dependent activation of PRMT5, via MEP50 phosphorylation, is necessary for inhibition of p53-dependent apoptosis.

PRMT5/MEP50–dependent inhibition of apoptosis in D1T286A-expressing cells implies that tumors driven by D1T286A/PRMT5 should maintain wild-type p53. Indeed, DNA sequencing of D1T286A/PRMT5–expressing tumors revealed intact, wild-type p53 in 12 of 12 tumors analyzed. To mechanistically interrogate D1T286A/PRMT5 regulation of p53, we assessed the ability of D1T286A-phosphorylated MEP50/PRMT5 to methylate p53. Wild-type p53 or a mutant p53 harboring arginine to lysine mutation at positions corresponding to reported sites of PRMT5 methylation (333, 335, 337, denoted p53RK; ref. 13) were utilized as substrates. PRMT5/MEP50 catalyzed methylation of wild-type p53, whereas methylation was abrogated in the p53RK mutant (Supplementary Fig. S6A). Catalytically dead PRMT5Δ also failed to methylate p53 (Supplementary Fig. S6B). To assess whether D1T286A-dependent phosphorylation of PRMT5/MEP50 induced p53 methylation, immunopurified PRMT5/MEP50 complexes were mixed with purified, active cyclin D1T286A/CDK4 kinase and ATP in CDK4 kinase buffer. PRMT5/MEP50 complexes were then washed extensively and mixed with recombinant p53 and ³H-SAM. Similar to published results using a histone H4 substrate (5), D1T286A/CDK4 kinase triggered a significant increase in PRMT5-dependent methylation of p53 (Supplementary Fig. S6C). To confirm that this reflects phosphorylation of MEP50 on Thr-5, we performed an analogous experiment, using PRMT5/MEP50, PRMT5/MEP50T5A, or PRMT5/MEP50T5D complexes [MEP50T5D is a phosphomimetic mutant in which threonine 5 was mutated to aspartic acid (D) to mimic constitutive phosphorylation; Supplementary Fig. S6D)]. Although D1T286A/CDK4 increased catalysis by PRMT5/MEP50, the MEP50T5A complexes retained only basal activity. By contrast, PRMT5/MEP50T5D complexes exhibited activity equivalent to phosphorylated wild-type PRMT5/MEP50 complexes and were refractory to further activation.

To independently address D1T286A-enhanced PRMT5 activity toward p53, we generated a p53-me2–specific antibody that specifically recognizes p53 symmetrically methylated on arginines 333, 335, and 337 (Supplementary Fig. S5C–S5D). Using this antibody, elevated methyl p53 levels were noted following expression of D1T286A (indicated with double arrows) in NIH3T3 cells that harbor wild-type p53 compared with that in untransfected cells (single arrows; Supplementary Fig. S6E). We also noted that cotransfection of PRMT5 with D1T286A (triple arrows compared with double arrows) resulted in a strong increase in p53-me2 staining (Supplementary Fig. S6F). Importantly, p53-me2 and PRMT5 were nuclear, demonstrating that methylation of p53 did not trigger nuclear exclusion (Supplementary Fig. S6E–S6F).

Tumor suppression by p53 can reflect either transcription-dependent (nuclear) activities and/or transcription-independent (cytoplasmic) activities (27, 33). No differential targeting of p53 to the cytoplasm was observed, suggesting a transcription-dependent impact (Supplementary Fig. S6E–S6F). PRMT5-dependent methylation of p53 occurs within the p53 oligomerization domain, suggesting a direct influence on p53 DNA binding and transcriptional output. To test this, we assessed the gene expression profiles of PRMT5, D1T286A, or D1T286A+PRMT5 in HSPCs using a quantitative PCR array containing 84 p53 target genes. D1T286A expression triggered varying induction of p53 target genes, including factors involved in apoptosis, cell cycle, and DNA repair (Fig. 6A). PRMT5 expression alone had little or no impact on expression of these genes. In contrast, coexpression of PRMT5 with D1T286A antagonized induction of a majority of the genes induced by D1T286A alone (Fig. 6A). We independently confirmed that D1T286A expression triggered strong increases in mRNAs encoding Apaf1, Bax, Pmaip1 (Noxa), Casp9, and Gadd45a (Fig. 6B). We also noted increased expression of Bbc3 (Puma), which was not represented on the array (Fig. 6B). Importantly, coexpression of PRMT5 with D1T286A reduced expression of these proapoptotic genes. Decreased expression did not reflect the absence of D1T286A (Fig. 1H; Supplementary Fig. S5A). We also noted a strong D1T286A-dependent increase in Cdkn1a expression, which was also antagonized by PRMT5 (Fig. 6B).

Chromatin immunoprecipitation (ChIP) was performed using bone marrow cells isolated from 5-FU–treated C57BL/6 mice transduced with D1T286A, PRMT5, or D1T286A+PRMT5 to address p53 recruitment. Expression of D1T286A resulted in a significant increase in p53 occupancy on all promoters tested (Fig. 6C). Although expression of PRMT5 alone failed to influence p53 occupancy, coexpression with D1T286A significantly reduced p53 occupancy on Cdkn1a, Apaf1, and Bax promoters, but had no statistically significant effect on p53 occupancy on the Pmaip1 promoter (Fig. 6C).

Recent work has also demonstrated that E2F1, analogous to p53, can be methylated by PRMT5 (34); here methylation appears to destabilize E2F1, thereby reducing its transcriptional activity and ultimately expression of proapoptotic genes. Most of the genes induced by D1T286A are targets of both E2F1 and p53 (12, 34). To determine whether E2F1 might differentially engage promoters cooperatively with p53 following D1T286A expression, we performed ChIP using an E2F1-specific antibody. D1T286A transduction resulted in a small but significant recruitment of E2F1 to the Apaf1 and Cdkn1a promoter regions; recruitment to the Cdkn1a promoter was sensitive to PRMT5 (Supplementary Fig. S7A). Strikingly, D1T286A triggered a much stronger induction of E2F1 recruitment to the Pmaip1 promoter, and this was entirely reversed by PRMT5 (Supplementary Fig. S7A), suggesting that E2F1 likely contributes more to Pmaip1 expression. No enrichment of E2F1 was observed at the Bax promoter (Supplementary Fig. S7A).

PRMT5-dependent dimethylation of histone H4 arginine 3 (H4R3) is associated with transcriptional repression (5–7). We determined whether direct histone methylation might also contribute to PRMT5-mediated gene silencing of Apaf1, Cdkn1a, Bax, and Pmaip1 by performing ChIP with an antibody specific for dimethylated histone H4R3 and using primers specific to the proximal promoter regions (∼500 bp upstream of the first coding exon). Cyclin D1T286A and PRMT5 coexpression resulted in increased methylation of H4R3 at both the Apaf1 and Pmaip1 promoters (Supplementary Fig. S7B). These data suggest that Apaf1 and Pmaip1 suppression reflects coordinated repressive histone modification and reduced occupancy of E2F1 and p53, all of which are PRMT5/D1T286A dependent.

If arginine methylation antagonizes p53 activity in response to expression of oncogenic cyclin D1 alleles, tumors should exhibit increased p53-me2. We utilized the antibody reactive against p53 dimethylated on arginines 333, 335, and 337 (Supplementary Fig. S5C–S5D) to assess p53-me2 status in tumors. Immunoblot of tumor lysates revealed a marked induction of p53-me2 in the D1T286A/PRMT5 but not in D1T286A/p53DN lymphoid tumors (Fig. 7A). These data support a mechanism in which survival of D1T286A-expressing cells is driven by and may require PRMT5-dependent methylation of p53. We also examined p53-me2 in murine lymphomas driven by either ICN or c-MYC. Increased arginine p53-me2 was noted in all ICN-driven tumors examined (Fig. 7B and C).

We next examined p53 arginine methylation status in primary human cancers wherein cyclin D1 is a driver. Immunohistochemical staining of lymph-node sections from human MCL patients revealed a marked increase of p53-me2 relative to normal lymph node (Fig. 7D). Increased staining was readily apparent in 6 of 8 primary MCL specimens. PRMT5 overexpression (Supplementary Fig. S8A) and increased symmetrical dimethylation histone H4R3 (Supplementary Fig. S8B), a marker for PRMT5 methyltransferase activity, were also noted in these samples. Forty percent of esophageal squamous cell carcinoma (ESCC) is associated with nuclear cyclin D1 as a driver (35). Immunohistochemistry in tissue microarrays (TMA) containing paired tumor and adjacent nonneoplastic clinical specimens revealed that 44% of the ESCCs exhibited concurrent high p53-me2 and high nuclear D1 (Fig. 7E); 46% simultaneously exhibited high expression of PRMT5 and high nuclear cyclin D1 (Supplementary Fig. S8C–S8D). Finally, we also assessed p53-me2 status in a panel of T-cell leukemia/lymphoma–derived cell lines relative to normal peripheral lymphocytes. Arginine methylation of p53 was readily detected in cancer cell lines with wild-type p53, whereas cell lines harboring mutant p53 exhibited low to undetectable arginine methylation (Fig. 7F). These data support the utilization of PRMT5-dependent methylation of p53 as an alternative to the acquisition of inactivating p53 mutations.

PRMT5 levels are frequently high in human lymphoid cancers, and are thought to contribute directly to manifestation of the malignancy. Although high levels of PRMT5 are associated with increased proliferation (36), mechanistic insights into its contribution will be critical to determine its potential as a therapeutic target. We have assessed the capacity of PRMT5 to regulate lymphomagenesis triggered by four distinct oncogenic drivers: cyclin D1, c-MYC, NOTCH1 (ICN), and MLL–AF9. Notably, overexpression of PRMT5 was found to cooperate with an oncogenic allele of cyclin D1 (D1T286A). Intriguingly, coexpression of PRMT5 failed to notably increase the penetrance or decrease the onset of disease triggered by any of the other three driver oncogenes. However, expression of a PRMT5 dominant-negative allele or knockdown of endogenous PRMT5 significantly inhibited the ability of all three to induce neoplastic growth in the colony-forming assay. Although seemingly a paradox, this likely reflects the capacity of c-MYC, NOTCH, and MLL–AF9 to potently induce high levels of endogenous PRMT5 expression, thereby abrogating the need for coexpression. Collectively, these data identify PRMT5 as a point of convergence during lymphomagenesis.

Although cyclin D1 has been considered a driver oncogene since its identification as the PRAD1 oncogene in parathyroid adenoma and the BCL1 oncogene in MCL (37, 38), we have only recently gained insights into the molecular underpinnings whereby it triggers a neoplastic switch. Current evidence suggests that failure to inactivate the nuclear cyclin D1/CDK4 kinase during S-phase dysregulates the “once and only once” regulation of DNA replication initiation, a direct consequence of CDT1 stabilization (19). Activation of PRMT5 is a key to this phenotype in that it generates repressive histone marks within the Cul4A/B promoter. The loss of Cul4A/B ultimately leads to overexpression of CDT1, a key component of the replication licensing machinery (39). This is in turn sensed by the DSB checkpoint effector ATM (20) that phosphorylates and activates p53.

The activation of p53 in this context is highly apoptotic (Figs. 1 and 2; ref. 40), and it effectively inhibits tumorigenesis triggered by D1T286A (19). Inactivating mutations in p53 occur at a surprisingly low frequency in animal models of cancer driven by cyclin D1 and even lower frequency in human MCL (31). The retention of wild-type p53 is generally associated with the loss of key effectors such as p19^ARF (p14^ARF in human cells) or with the overexpression of MDM2 (18, 40–42). Because neither of these mechanisms was observed in D1T286A-driven tumors (ref. 16; data herein), we considered alternative mechanisms for bypassing p53 function. Among the potential nonhistone targets of PRMT5/MEP50 is p53. Given that arginine methylation was suggested to modify the proapoptotic response of p53, we considered whether D1T286A-dependent activation of PRMT5/MEP50 might directly inhibit p53 and p53-dependent cell death, thereby precluding the selection for inactivating p53 mutations. Consistent with this hypothesis, coexpression of PRMT5 with D1T286A bypassed p53-dependent apoptosis and resulted in the rapid acquisition of an aggressive CD4⁺ lymphoma that retained wild-type p53. The need for coexpression of PRMT5 in this system likely reflects its very low expression in HSCs relative to progenitor lineages. This is in contrast to the Eμ-D1T286A model in which D1T286A expression is targeted to lineage-committed IgM⁺/IgD^low B-lymphocytes that express PRMT5 (3). In the latter model, PRMT5 levels are at a threshold wherein phosphorylation-dependent increases in its function are sufficient for disease manifestation.

Importantly, we also noted increased PRMT5 expression and p53 methylation in primary human MCL and in ESCC. The physiologic importance of arginines 333, 335, and 337 in the regulation of p53 function is further emphasized by their mutation in Li Fraumeni families that present with a wide spectrum of tumors (43, 44).

Similar to D1T286A, we noted increased p53 methylation on arginines targeted by PRMT5 in NOTCH- and MLL–AF9-triggered tumors, but not those generated by c-MYC expression, suggesting that PRMT5 may contribute to p53 inactivation in tumors where NOTCH or MLL–AF9 function as drivers, thereby alleviating selection for p53 mutation. In the context of c-MYC, it has already been established that p53 mutation and biallelic deletion of ARF are the primary genetic events for p53 bypass (42). Importantly, however, our data suggest that PRMT5 may represent a unique therapeutic target in multiple neoplastic settings. In point of fact, PRMT5 inhibition can induce lymphoma cell death (45), and it has also been suggested as a therapeutic target for glioblastoma (46). Collectively, the data support a model wherein PRMT5 exhibits broad proproliferative and prosurvival activities and wherein the precise mode of action by PRMT5 likely reflects genetic and perhaps tissue specificity.

An open question that remains is why do certain oncogenic drivers utilize mechanisms to bypass p53-dependent tumor suppression such as methylation (e.g., cyclin D1), whereas others select for mutation of p53 (e.g., c-MYC)? One possibility is that each mechanism, while reducing p53 function, permits the maintenance of key p53 functions that are important for tumor progression. Data demonstrating that mutant TP53 alleles frequently have neopmorphic activities and thus do not equate with TP53 deletion support this conclusion (28). Likewise, methylated p53, although it has significantly reduced transcriptional activity at many gene targets (Cdkn1a and Apaf1), exhibits less sensitivity at other targets (e.g., Pmaip1). A second more applicable consequence of maintenance of wild-type p53 might reflect in how tumors respond to therapeutic intervention. With cyclin D1-driven tumors such as MCL, inhibition of PRMT5 should not only directly affect many transcriptional programs, but should also permit functional reactivation of wild-type p53 and p53-dependent tumor-suppressive activities. The development of PRMT5-selective inhibitors will allow further investigation of these concepts in many distinct tumor contexts.

HEK 293T (obtained from and authenticated by the ATCC) and NIH3T3 cells (a gift from Charles J. Sherr, St. Jude, authenticated by Southern blot) were cultured in DMEM supplemented with 10% FBS and 1% penicillin/streptomycin. HSB2, LOUCY, 8402, PF382, and TALL-1 (obtained from and authenticated by the ATCC) cells were maintained in RPMI-1640 containing 10% FBS, 1% glutamine, and 1% penicillin/streptomycin supplemented with 0.05 mmol/L of 2-mercaptoethanol. MOLT-3, MOLT-4, and CEM cells (obtained from and authenticated by the ATCC) were maintained in RPMI-1640 containing 10% FBS and 1% penicillin/streptomycin.

Cyclin D1 and D1T286A were flag-tagged and subcloned into MSCV-IRES-GFP (MigR1). The 6X MYC-tagged human PRMT5 from pCS2-PRMT5 and MYC-tagged human MEP50 from pcDNA3-myc-MEP50 (5) were subcloned into pcDNA3, MSCV-IRES-tNGFR (tNGFR), or MSCV-IRES-mCherry (mCherry subcloned with MigR1 and pCS2-mCherry) vectors. p53R175H (p53DN) was subcloned into tNGFR and mCherry vectors. PRMT5Δ and p53RK mutants were generated with the QuikChange Site-Directed Mutagenesis Kit (Stratagene) according to the manufacturer's instructions. All clones were sequenced in their entirety. Retroviral supernatants were generated by transient transfection of 293T cells with Lipofectamine.

All animal experiments were conducted in compliance with the Animal Care and Use Committee of the University of Pennsylvania and Medical University of South Carolina. Bone marrow transplantation (BMT) experiments were performed as previously described (47). Briefly, bone marrow cells were collected from 6- to 8-week-old C57BL/6 or B6.129S2-Trp53tm1Tyj/J (Trp53^−/−; The Jackson Laboratory) mice 4 days after i.v. administration of 5-FU (150 mg/kg) and retrovirally transduced ex vivo in the presence of IL3, IL6, and stem cell factor (SCF). Retroviral supernatants with equal titers were used to produce similar transduction efficiencies. GFP⁺ NGFR⁺ cells (0.5–1 × 10⁶) were then injected i.v. into lethally irradiated (900 rad) B6 recipients. Chimeric mice were maintained on antibiotics for 2 weeks.

Single-cell suspensions prepared from bone marrow and spleen were stained on ice in PBS plus 2% FBS and analyzed on FACSCalibur, FACSVerse, LSRII, or FACSAria (BD Biosciences). Files were analyzed in FlowJo (TreeStar).

For assessing the total hematopoietic progenitor cell activity, bone marrow was harvested from 5-FU–treated C57BL/6 mice. After red blood cell lysis using ACK lysis buffer (Lonza), 2 × 10⁴ nucleated cells were plated in triplicates into methylcellulose medium (MethoCult 3234; Stem Cell Technologies) supplemented with 50 ng/mL FLT3L, 50 ng/mL SCF, 10 ng/mL IL3, 10 ng/mL IL6, and 10 ng/mL IL7 (Stem Cell Technologies). The colony number was counted 7 days after replating.

Tissue was fixed in 4% buffered formalin and subsequently was dehydrated, paraffin embedded, and sectioned. Tissue sections were immunostained as described previously (19), with meR-p53 at 1:200 dilution and PRMT5 at 1:150 dilution as the primary antibody. For immunofluorescence, cells were fixed, blocked, and immunostained as described (48).

Total RNA was extracted using the RNeasy Micro Kit (Qiagen), and used as template to synthesize cDNA (RT² First Strand Kit) for quantitative RT-PCR (qRT-PCR) analysis with the Mouse p53 Signaling Pathway PCR Array (PAMM-027; SuperArray Bioscience Corporation). Primers for Apaf1, Cdkn1a, Bax, Pmaip1, Casp9, Gadd45a, and Bbc3 were generated according to the RT² qPCR Primer Assay (Qiagen).

Western blot analysis was carried out as previously described (5). Chromatin was prepared using the truChIP Low Cell Chromatin Shearing Kit (Covaris) and sheared into 200- to 700-bp fragments using a Covaris S2 instrument (duty cycle, 2%; intensity, 3; 200 cycles per burst; 4 min). Immunoprecipitation was performed using the IgG, p53 (FL-393), E2F-1 (C-20), and H4R3 (Abcam) antibodies with a Quick Chip Kit (Imgenex). Quantification of the precipitated DNA was determined with qPCR (Qiagen, QuantiTect SYBR Green Mastermix) and normalized with the input genomic DNA. Primers used were: Apaf1 (E2F-1) forward, 5′-TAGTTTTGTAGGCACACAGCTCTAAATAGGAG-3′; Apaf1 (E2F-1) reverse, 5′-CGGATGAGTTTGCTCACACCCTCCACC-3′; Pmaip1 (E2F-1) forward, 5′-GCCCCAGCAATGGATACGA-3′; Pmaip1 (E2F-1) reverse, 5′-TGCTCAACCCCCAAATTGCT-3′. The other primers are from Qiagen EpiTect ChIP qPCR primers.

Statistical significance was determined by the Student t test using the Excel software; P < 0.05 was considered statistically significant.

No potential conflicts of interest were disclosed.

Conception and design: Y. Li, A.K. Rustgi, J.A. Diehl

Development of methodology: Y. Li, Y. Yang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Li, N. Chitnis, H. Nakagawa, Y. Kita, Z. Li, M. Wasik, A.J.P. Klein-Szanto

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Li, N. Chitnis, H. Nakagawa, Y. Kita, S. Natsugoe

Writing, review, and/or revision of the manuscript: Y. Li, A.K. Rustgi, J.A. Diehl

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Li, A.J.P. Klein-Szanto

Study supervision: J.A. Diehl

The authors thank Nancy Speck for advice and technical expertise for bone marrow reconstitution experiments, and Zhaorui Lian for technical assistance.

This work was supported by grants CA11360 (to J.A. Diehl) and P01-CA098101 (to J.A. Diehl, A.K. Rustgi, H. Nakagawa, and A.J.P. Klein-Szanto). This study was also supported by NIH-P30-DK050306 and the Molecular Pathology and Imaging, Molecular Biology and Cell Culture Core Facilities.

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.