An integrated approach to dissecting oncogene addiction implicates a Myb-coordinated self-renewal program as essential for leukemia maintenance

An inducible mouse model of MLL fusion AML

To explore oncogene addiction programs in a genetically defined leukemia context, we developed an AML mouse model driven by a conditional MLL-AF9 allele together with oncogenic Ras, which enabled us to examine the consequences of MLL-AF9 inhibition in an established disease. In order to produce a tightly regulated system that was easy to monitor, we constructed two retroviral vectors containing dsRed-linked MLL-AF9 under control of a tetracycline (tet) response element (TRE) promoter, and Kras^G12D linked to the “Tet-off” tet transactivator (tTA), which activates TRE expression in a doxycycline (dox)-repressible manner (Fig. 1A). Leukemias were generated by retroviral cotransduction of both vectors into hematopoietic stem and progenitor cells (HSPCs), which were transplanted into syngeneic mice. Cells harboring both constructs (dsRed⁺) induced an aggressive myelomonocytic leukemia similar to existing mouse models based on constitutive MLL oncogene expression, which recapitulated key features of human MLL fusion-driven AML (Zuber et al. 2009).

The impact of withdrawing expression of MLL-AF9 on leukemia cell proliferation was dramatic. Dox treatment rapidly turned off dsRed, MLL-AF9, and HoxA9 expression (Supplemental Fig. 1) and triggered cell cycle arrest and terminal differentiation of leukemia cells (Fig. 1B; Supplemental Fig. 2). Withdrawal of MLL-AF9 in vivo led to disease regression and clearance of leukemia cells from all infiltrated organs (Fig. 1C; Supplemental Fig. 3) and prolonged the survival of leukemia-bearing mice (median, 16 vs. 127 d; P < 0.0001) (Fig. 1D). Most animals remained in remission upon discontinuing dox treatment after 45 d, and those with apparent “relapses” actually succumbed to a dsRed-negative host-derived lymphoid malignancy (Supplemental Fig. 4). These results, together with a previous report (Horton et al. 2008), demonstrate that leukemia cells require the continued expression of MLL-AF9.

Identifying critical mediators of leukemia maintenance

To identify molecular mechanisms underlying addiction to MLL-AF9, we performed genome-wide analysis of gene expression changes on leukemia cells following oncogene withdrawal. Six days following dox treatment, five independent Tet-off-regulatable MLL-AF9;Nras^G12D leukemias showed a complex pattern of gene expression changes (Fig. 1E; Supplemental Table 1). Gene ontology analysis revealed a down-regulation of genes associated with cell cycle and mitosis, whereas those involved in mature myeloid cell functions were induced, suggesting that leukemia cells underwent a terminal differentiation program (Supplemental Table 2). In addition, gene set enrichment analysis (GSEA) (Subramanian et al. 2005) revealed a strong inverse correlation between global gene expression changes following MLL-AF9 withdrawal and a LSC-associated signature identified in MLL oncoprotein-expressing leukemias (normalized enrichment score [NES], 1.7 and −1.9; false discovery rate [FDR], 0.05 and 0) (Supplemental Fig. 5; Somervaille et al. 2009). Hence, the aberrant self-renewal and “stemness” programs induced by MLL-AF9 require continued oncogene expression.

To identify genes that might mediate these effects, we focused our subsequent analysis on differentially expressed transcription factors, as these molecules govern cell fate in normal hematopoiesis (Rosenbauer and Tenen 2007). Out of 17 strongly down-regulated transcription factors (P < 0.005, FC_log2 < −1.5), 16 were validated by quantitative RT–PCR (qRT–PCR) analysis, including established MLL-AF9 targets such as HoxA9, HoxA5, and Meis1 (Supplemental Fig. 6). To distinguish between direct and indirect effects, we produced AMLs expressing a Flag-tagged MLL-AF9 transgene (Supplemental Fig. 7A,B) and analyzed MLL-AF9 promoter occupancy by chromatin immunoprecipitation (ChIP). In addition to binding the promoter regions of its canonical targets (HoxA9 and Meis1), MLL-AF9 was also enriched at the promoters of five additional transcription factors (Tshz1, Myc, FoxP1, Myb, and Irx5) (Fig. 1F; Supplemental Fig. 7C), suggesting that these also may be direct MLL-AF9 targets. The most differentially expressed direct MLL-AF9 target was the transcription factor c-Myb (Myb), which has been shown to be induced by MLL oncoproteins through an indirect mechanism (Hess et al. 2006). However, consistent with our results in murine leukemias, ChIP analysis of human THP1 AML cells expressing a Flag-tagged MLL-AF9 also showed strong occupancy of the fusion protein on the Myb promoter (Supplemental Fig. 7D). Besides Myb itself, several of its known transcriptional targets were among the most down-regulated genes following MLL-AF9 withdrawal (Fig. 1E; Supplemental Fig. 8; Ramsay and Gonda 2008), suggesting that Myb-dependent transcriptional programs may play a pivotal role in addiction to MLL-AF9.

Myb is required for the maintenance of leukemia

c-MYB was first identified as a proto-oncogene via its association with oncogenic retroviruses and as a target for insertional mutagenesis in mice (Klempnauer and Bishop 1984; Shen-Ong et al. 1986). More recent studies indicate that Myb is the target of somatic mutations and translocations in human cancers, and thus is an important human oncogene (Clappier et al. 2007; Persson et al. 2009). Functionally, Myb is a member of the basic helix–loop–helix family of transcription factors (Nomura et al. 1988), an important transcriptional regulator in normal stem and progenitor cell compartments (Emambokus et al. 2003; Malaterre et al. 2007), and is also a component of the LSC signature described above (Somervaille et al. 2009). Moreover, Myb inhibition is known to prevent the initiation of leukemogenesis by MLL-ENL (Hess et al. 2006) and has anti-proliferative effects in some leukemia cells (Luger et al. 2002; Somervaille et al. 2009). However, the role of Myb in leukemia maintenance has not been well characterized, in part because systems to efficiently inhibit gene expression in established disease in vivo were not available.

To study the role of Myb and other genes in leukemia maintenance, we took advantage of new methods to regulate gene expression using tet-responsive mir30-based shRNAs. We generated a “Tet-on RNAi-competent” model of MLL-AF9;Nras^G12D AML that expresses an improved reverse tTA (rtTA3) linked to MLL-AF9, as well as Nras^G12D-linked Luciferase to facilitate leukemia monitoring in vivo (Fig. 2A; Zuber et al. 2011a). These leukemia cells were then transduced with a new Tet-regulatable dual-color shRNA expression vector (TRMPV-Neo) (Zuber et al. 2011a). In this system, expression of the Venus fluorescent protein reports integration of the retrovirus, while dsRed fluorescence tracks expression of the shRNA, enabling precise fluorescent tracking and isolation of shRNA-expressing cells as the Venus⁺dsRed⁺ population.

In competitive proliferation assays, TRMPV-Neo-transduced leukemia cells were mixed with uninfected leukemia cells and treated with dox to induce shRNA expression, and the percentage of shRNA-expressing (Venus⁺dsRed⁺) cells was monitored over time using flow cytometry. As expected, TRMPV-Neo-transduced leukemia cells expressing a neutral shRNA (targeting Renilla luciferase, Ren) were maintained in the cell population; in contrast, potent Myb shRNAs were rapidly depleted relative to non-shRNA-expressing cells (dsRed⁻) (Fig. 2B,C). These results mirrored the effects of suppressing replication protein A3 (Rpa3), an essential DNA replication factor required for proliferation (McJunkin et al. 2011; Zuber et al. 2011a). In contrast to the Rpa3 shRNAs, Myb shRNAs had no effect on immortalized rtTA-expressing mouse embryonic fibroblasts (RRT-MEFs) (Fig. 2B; Fellmann et al. 2011), demonstrating that Myb shRNAs are not generally toxic.

To extend these observations to human cancer cells, we evaluated the effects of Myb inhibition in 10 human leukemia cell lines. Suppression of Myb by two independent shRNAs impaired proliferation of all four MLL fusion-expressing cell lines tested, two of which coexpress MLL-AF9 with oncogenic Ras, as well as two other AML lines without MLL fusion proteins (Fig. 2D,E; Supplemental Fig. 9A). In four of six leukemia lines without MLL aberrations, Myb suppression had only minimal or no effect on cell proliferation, and these differences in sensitivity did not correlate with Myb expression levels (Supplemental Fig. 9B). Together, these results indicate that the sensitivity to Myb inhibition extends to human leukemia cells, particularly, but not exclusively, to MLL-rearranged contexts.

Myb suppression is tolerated in normal hematopoiesis

If Myb mediates cellular addiction to MLL-AF9, then myeloid leukemia cells should be more sensitive to Myb inhibition than their normal counterparts. To investigate the effects of RNAi-mediated Myb suppression in normal hematopoietic cells, we established a two-color competitive reconstitution assay in which HSPCs transduced with either a GFP-tagged experimental shRNA or an mCherry-tagged neutral internal control shRNA (Ren) are monitored for their ability to reconstitute the hematopoietic system of lethally irradiated recipient mice (Fig. 3A). As expected, 6 wk after transplantation, donor-derived bone marrow contained GFP- and mCherry-positive shRen-expressing cells, while cells expressing a GFP-tagged shRNA targeting Rpa3 were completely depleted (Fig. 3B,C).

In contrast, cells expressing Myb shRNAs were able to efficiently reconstitute recipient mice (Fig. 3B,C) even though Myb was suppressed to similar levels as in MLL-AF9;Nras^G12D leukemia cells (Fig. 3D). These results differ from the consequences of complete Myb ablation observed in knockout mice (Mucenski et al. 1991; Lieu and Reddy 2009) and suggest that hypomorphic states produced by shRNA-mediated Myb suppression are better tolerated. Lineage analysis revealed that Myb inhibition does not impede normal erythropoiesis and myelopoiesis but, consistent with previous reports (Bender et al. 2004; Fahl et al. 2009), impairs normal lymphopoiesis (Supplemental Fig. 10). Interestingly, one of the shRNAs (shMyb.2652) showed enrichment in myeloid and progenitor compartments (Supplemental Fig. 10), which, together with a previous study (Emambokus et al. 2003), suggests that hypomorphic states of Myb induce expansion rather than depletion in the myeloid compartment. Of note, cells with Myb suppression are capable of supporting long-term myeloid reconstitution, and animals remained healthy and disease-free for >12 wk (data not shown). These results indicate that the consequences of suppressing Myb are dosage- and lineage-dependent and can result in highly cancer-specific sensitivities in the myeloid lineage. They also establish a rapid and general assay for exploring the impact of gene function during normal hematopoiesis.

Suppression of Myb eradicates AML in vivo

To explore the sensitivity to Myb inhibition in established AML in vivo, we transplanted a series of Tet-on-competent MLL-AF9;Nras^G12D, TRMPV-Neo-transduced leukemias harboring shRNAs targeting Myb, the Ras effector Braf (another putative target in our model), or Ren, and induced shRNA expression with dox following leukemia detection in bioluminescence imaging (Supplemental Fig. 11A). Even using mixed populations of transduced cells, Myb suppression resulted in a delay in disease progression (Supplemental Fig. 11A) and a significant survival benefit (P < 0.005, Supplemental Fig. 11B). At terminal disease stage, bone marrow of mice transplanted with shRen and shBraf-transduced leukemia was completely infiltrated with shRNA-expressing cells (Venus⁺dsRed⁺), while leukemia cells expressing Myb-shRNAs were entirely gone (Supplemental Fig. 11C,D). Thus, in mice transplanted with shMyb-transduced MLL-AF9;Nras^G12D leukemia, progressive disease resulted from the outgrowth of cells that failed to induce Myb shRNA expression (dsRed⁻).

Even more dramatic results were observed when the experiment was repeated using single cell-derived clones that uniformly induce shRNA expression. While induction of Ren or Braf shRNAs had no effect on disease course, suppression of Myb induced the clearance of leukemia cells from all infiltrated organs, leading to complete remissions (Fig. 4A–C; Supplemental Fig. 11E). These remissions were durable, and most animals remained disease-free even after discontinuing dox treatment after 40 d (Fig. 4C). Again, those mice that subsequently developed disease harbored leukemia cells that were dsRed-negative, and thus did not express the Myb shRNA and no longer inhibited Myb expression (Supplemental Fig. 11F). This dramatic anti-leukemic response was in contrast to the minimal effects of a combined chemotherapy regimen mimicking clinical AML therapy (Ara-C + Doxorubicin) (Zuber et al. 2009) in mice bearing the same leukemias (Fig. 4C). Thus, transient inhibition of Myb invariably induces complete remission of MLL-AF9-driven AML in vivo, leading to eradication of an otherwise aggressive and treatment refractory disease.

Myb regulates a LSC-associated program

We next performed another iteration of expression profiling and functional studies to explore the transcriptional networks and effector programs underlying the leukemia-specific dependency on Myb. Immunophenotyping and gene expression analysis revealed that suppression of Myb induces terminal differentiation into mature neutrophils as well as monocytes/macrophages (Fig. 4D; Supplemental Fig. 12), precisely phenocopying effects observed after inactivating the driving MLL-AF9 oncogene (Fig. 1B; Supplemental Fig. 12). The global gene expression changes following Myb suppression were remarkably similar to those observed after MLL-AF9 withdrawal (Spearman correlation, 0.50) (Fig. 5A,B; Supplemental Fig. 13A,B), as determined by a recently developed method to compare global expression signatures (Rank–Rank Hypergeometric Overlap [RRHO]) (Plaisier et al. 2010). Indeed, >40% of significantly altered genes in both signatures overlap (Fig. 5A; Supplemental Fig. 13C), and these influence the same biological processes (Supplemental Tables 3, 4). Such strong effects indicate that Myb is a dominant factor involved in aberrant transcription and impaired differentiation associated with MLL-AF9.

The transcriptional module associated with Myb suppression also showed a strong inverse correlation with the LSC signature (NES, 1.5 and −1.7; FDR, 0.05 and 0.004) (Fig. 5C; Supplemental Fig. 13D), indicating that Myb is a master regulator of LSC-associated programs. Notably, the Myb-mediated expression profile was also significantly enriched for two gene sets associated with poor prognosis in pediatric and adult AML (NES, 1.8 and 1.5; FDR, <0.01 and 0.05) (Fig. 5D; Yagi et al. 2003; Metzeler et al. 2008), implying that this program also contributes to aggressive disease in humans. Interestingly, some canonical MLL target genes such as HoxA9 and Meis1 are excluded from the Myb-specific signatures (Fig. 5A; Supplemental Fig. 13C), indicating that Myb controls only a subset of MLL-AF9 targets and does not predominantly act as a cofactor in MLL complexes as proposed (Jin et al. 2010). These findings also suggest that HoxA9 and Meis1, despite their known role in leukemia initiation by MLL fusion proteins (Zeisig et al. 2004), may be less important for maintaining aberrant self-renewal. Supporting this notion, multiple HoxA9 and Meis1 shRNAs did not affect MLL-AF9;Nras^G12D AML cells in competitive proliferation assays (Fig. 5E).

Components of the Myb-mediated self-renewal program are required for leukemia maintenance

Several known Myb target genes have been associated with self-renewal in different contexts, and therefore might mediate Myb-dependent aberrant self-renewal in AML. For example, Kit, a target of somatic mutations in human AML (Estey and Dohner 2006), serves as a LSC marker in murine MLL-AF9-induced leukemia (Krivtsov et al. 2006). Myc contributes to self-renewal in normal stem cells (Smith et al. 2010) and may play a pivotal role in establishing the LSC signature described above (Kim et al. 2010). Bcl2 promotes the self-renewal of normal hematopoietic stem cells (Milyavsky et al. 2010), and its pharmacologic inhibition is purported to target LSCs (Konopleva et al. 2006). Finally, in different analyses integrating LSC-, Myb-, and MLL-AF9-associated signatures, we consistently identified the methyltransferase Smyd2 as a prominent LSC-associated gene suppressed upon Myb and MLL-AF9 withdrawal (Supplemental Tables 5, 6) and subsequently validated it as a Myb-regulated gene in mouse and human AML cells (Fig. 5F,G). Interestingly, Smyd2 is amplified in esophageal squamous cell carcinoma (Komatsu et al. 2009) and methylates and inhibits the p53 and Rb tumor suppressors (Huang et al. 2006; Saddic et al. 2010), two molecules that negatively regulate self-renewal (Galderisi et al. 2006; Zhao et al. 2010). Indeed, serial replating assays indicate that enforced Smyd2 expression enhances self-renewal of hematopoietic progenitors, similar to the known effects of Myc (Fig. 6A).

Our system enables the rapid evaluation of candidate tumor maintenance genes without the production of new germline transgenic mice, simply by making potent shRNAs targeting a gene of interest and assessing their impact in our transplantable model. To evaluate a possible role of the Myb-regulated genes described above in maintaining aberrant self-renewal, we analyzed the impact of their TRMPV-shRNA-mediated suppression on AML proliferation and differentiation. Suppression of Kit by multiple potent shRNAs (Supplemental Fig. 14A) had no effect on MLL-AF9;Nras^G12D AML (Fig. 6B). In contrast, leukemia cells expressing potent shRNAs targeting either Myc, Smyd2, or Bcl2 (Supplemental Fig. 14B–D) underwent terminal myeloid differentiation and were depleted from the population over time (Fig. 6B,D). Inhibition of Myc, but not Smyd2 or Bcl2, also impeded proliferation of RRT-MEF (Fig. 6C), indicating that strong RNAi-mediated suppression of Myc confers general anti-proliferative effects.

We next transplanted drug-selected TRMPV-Neo-transduced MLL-AF9;Nras^G12D leukemias, induced shRNA expression, and evaluated the bone marrow for the presence of shRNA-expressing cells at terminal disease stage (Fig. 6E,F). Suppression of Myc, Bcl2, or Smyd2 led to the depletion of leukemia cells in vivo, although in no case were the effects as dramatic or complete as observed when targeting Myb. We also performed cDNA rescue experiments using a modified competition assay in which the relative frequency of shMyb-expressing cells is monitored over time in cells that ectopically express different cDNAs (Supplemental Fig. 15A). Consistent with RNAi-based phenotypes, overexpression of Myc, Bcl2, or Smyd2 delayed Myb-dependent depletion to varying degrees, but was unable to fully rescue the inhibitory and prodifferentiation effects of suppressing Myb (Fig. 6G; Supplemental Fig. 15B). While we cannot rule out that the less-dramatic effects observed after suppressing Myc, Smyd2, or Bcl2 are due to differences in shRNA knockdown efficiency, the inability of cDNAs encoding these genes to completely rescue the effects of Myb suppression suggests that none of them can completely replicate the effects of Myb. Instead, these results indicate that Myb exerts its central role in maintaining aberrant self-renewal through a coordinated program involving Myc, Bcl2, Smyd2, and, undoubtedly, additional target genes.

Retroviral constructs

All constructs were derived from either the Murine Stem Cell Virus backbone (MSCV, Clontech) or the pQCXIX self-inactivating retroviral vector backbone (pSIN, Clontech) using standard cloning techniques. TRMPV vectors are available through Addgene (#27990, #27991, #27993, and #27996). shRNAs were designed using the Biopredsi algorithm and subcloned as described (for sequences, see Supplemental Table 7; Zuber et al. 2011a). See the Supplemental Material for more detailed vector information.

Generation and monitoring of AML

Primary leukemias were generated by cotransduction of retroviral constructs into fetal liver cells and transplantation into syngeneic recipient mice as described previously (Zuber et al. 2009). Leukemia cells were harvested from bone marrow and spleen and cultured in RPMI-1640 + 10%FBS. Peripheral blood smears and bone marrow cytospins were stained with Wright-Giemsa (Sigma); tissue specimens were fixed in 10% formalin and stained with Hematoxylin and eosin (Sigma). For immunostaining, cells were stained with specified fluorophore-conjugated antibodies (BioLegend) and analyzed by flow cytometry (LSRII, BD BioSciences). Mice were monitored and treated as described (Zuber et al. 2011a). Statistical evaluation of overall survival was based on the log-rank (Mantel-Cox) test for comparison of the Kaplan-Meier event time format. More detailed methods are provided in the Supplemental Material.

Gene expression analysis

Cultured and treated primary leukemia cells were sorted for shRNA-expressing cells when necessary using an Aria-II (BD Biosciences). Total RNA was extracted using Trizol reagent (Invitrogen) according to the manufacturer's instructions. cDNA was synthesized using TaqMan reverse transcriptase reaction (Applied Biosystems). qRT–PCR analysis was performed on an iCycler mounted with an iQ5 multicolor real-time PCR detection system (Bio-Rad). All signals were quantified using the ΔΔCt method and normalized to β-actin levels. Microarray experiments were performed on Mouse Genome 430A 2.0 arrays (Affymetrix). All raw microarray data files are available from the Gene Expression Omnibus (GSE30747). For Western blot analysis, the following antibodies were used: c-Myb (clone1-1, 1:500; Millipore), Flag M2 (#1804, 1:1000; Sigma), Braf (F-7, 1:200; Santa Cruz Biotechnolgy), Bcl2 (BCL/10C4, 1:1000; BioLegend), Smyd2 (#4251, Cell Signalling), and actin (HRP-conjugated, 1:10000; Sigma). ChIP was performed as described (Steger et al. 2008). Detailed methods are provided in the Supplemental Material.