|Home | About | Journals | Submit | Contact Us | Français|
The transcription factor IRF4 is required during an immune response for lymphocyte activation and the generation of immunoglobulin-secreting plasma cells1-3. Multiple myeloma, a malignancy of plasma cells, has a complex molecular etiology with several subgroups defined by gene expression profiling and recurrent chromosomal translocations4,5. Moreover, the malignant clone can sustain multiple oncogenic lesions, accumulating genetic damage as the disease progresses6,7. Current therapies for myeloma can extend survival but are not curative8,9. Hence, new therapeutic strategies are needed that target molecular pathways shared by all subtypes of myeloma. Using a loss-of-function, RNA-interference-based genetic screen we show here that IRF4 inhibition was toxic to myeloma cell lines, regardless of transforming oncogenic mechanism. Gene expression profiling and genome-wide chromatin immunoprecipitation analysis uncovered an extensive network of IRF4 target genes and identified MYC as a direct target of IRF4 in activated B cells and myeloma. Unexpectedly, IRF4 was itself a direct target of MYC transactivation, generating an autoregulatory circuit in myeloma cells. Though IRF4 is not genetically altered in most myelomas, they are nonetheless addicted to an aberrant IRF4 regulatory network that fuses the gene expression programs of normal plasma cells and activated B cells.
Recently, we developed a genetic method to identify therapeutic targets in cancer in which small hairpin RNAs (shRNAs) that mediate RNA interference are screened for their ability to block cancer cell proliferation and/or survival10. Here we report the results of such an “Achilles heel” screen in multiple myeloma (Supplementary Table 3). We used myeloma cell lines from three molecular subtypes: KMS12 (CCND1 translocation), H929 (FGFR3/MMSET translocation), and SKMM1 (MAFB, IRF4 translocations). Myeloma cells that received an shRNA targeting the coding region of IRF4 were depleted from cultures by 2-8 fold (Fig.1a). Lymphoma cell lines were largely unaffected by IRF4 knockdown, with the exception of OCI-Ly3, an activated B cell-like diffuse large B cell lymphoma line that expresses IRF4 highly11.
We next identified two additional shRNAs against IRF4 that were toxic to myeloma cell lines, one directed against the IRF4 3' untranslated region (UTR, Supplementary Fig.1). The toxicity of this shRNA was associated with a 50-75% decrease in IRF4 mRNA and protein (Supplementary Fig. 2a, b, c). Cell death occurred within 3 days, as measured by an increase in sub-G1 DNA content; there was, however, no effect on the cell cycle (Supplementary Fig. 2d, e, f, g). Expression of a cDNA containing only the coding region of IRF4 was able to rescue myeloma cells from the toxicity of the 3'UTR-directed IRF4 shRNA, confirming that the toxicity of this shRNA was specific (Fig.1b).
Strikingly, knockdown of IRF4 killed 10 myeloma cell lines, but had minimal effect on 5 lymphoma cell lines (Fig.1c). These myeloma lines bear many of the recurrent genetic aberrations typical of this cancer, including translocations of CCND1, MYC, MAF, MAFB, FGFR3: MMSET, NIK and IRF4, as well as RAS mutations, inactivation of TP53 and CDKN2C, and genetic abnormalities that activate the NF-κB pathway (Supplementary Table 1). Resequencing of the IRF4 coding exons in these lines revealed that 9 had a wild type sequence and one had a heterozygous mutation in exon 8 resulting in a missense substitution whose functional significance is unknown. Moreover, no amplification of the IRF4 locus was detected by array-based comparative genomic hybridization and no translocations involving IRF4 were detected by cytogenetics, with the exception of the previously documented IRF4 translocation in SKMM1 cells (data not shown). Thus, IRF4 dependency spans many myeloma subtypes and does not require genetic abnormalities in the IRF4 locus.
To understand the molecular basis for this dependency, we identified downstream targets of IRF4 by profiling gene expression changes in myeloma lines following induction of IRF4 shRNAs (Supplementary Fig. 3). A total of 308 genes were consistently downregulated following IRF4 knockdown (Supplementary Table 2). This list was significantly enriched for genes that are more highly expressed in primary myeloma samples than in normal mature B cells, based on gene set enrichment analysis12 of published gene expression profiling data (p=0.002 ; Fig. 2a)13 . Thus, IRF4 directs a broad gene expression program that is characteristic of primary myeloma cells.
We next investigated whether the IRF4 target genes in myeloma are also upregulated in other normal hematopoietic cells that require high IRF4 expression, including plasma cells3, mitogenically activated mature B cells1, and dendritic cells14. Human bone marrow-derived plasma cells expressed 22% of the IRF4 target genes at higher levels than mature blood B cells (Fig. 2a)13. Likewise, 25% of the IRF4 targets were more highly expressed in plasmacytoid dendritic cells than in monocytes (Supplementary Fig. 4)15. Blood B cells activated by anti-IgM crosslinking expressed one third of the IRF4 target genes more highly than resting B cells (Fig. 2a).
However, IRF4 regulates a broader set of genes in myeloma than in individual hematopoietic subsets. Roughly one quarter of the IRF4 target genes in myeloma were upregulated in activated B cells but not plasma cells, including genes known to be important in cellular growth and proliferation, such as MYC (Fig. 2a). Conversely, one sixth of the myeloma IRF4 target genes were highly expressed in plasma cells but not activated B cells.
To identify direct IRF4 targets, we performed genome-wide chromatin immunoprecipitation (ChIP-CHIP), using DNA microarrays with probes spanning ~10kb at the 5' end of 17,574 human genes. Specific IRF4 binding to 558 genes was detected in a myeloma cell line (KMS12) but not a lymphoma cell line (OCI-Ly19). Of these, 35 were also IRF4 targets by gene expression profiling, a highly significant overlap (p=1.0 × 10−16, Chi-square; Fig. 2b, Supplementary Fig. 5), and were considered presumptive direct IRF4 targets (Supplementary Table 2). Direct binding of IRF4 was confirmed by conventional chromatin immunoprecipitation (ChIP) for 22 genes, leading us to conclude that all 35 genes are likely IRF4 direct targets (Fig. 2b, and data not shown). This list of IRF4 direct targets is a conservative estimate since the ChIP-CHIP arrays interrogate limited regions around each gene. Indeed, direct ChIP experiments demonstrated that two other IRF4 target genes, PRDM1 and SQLE, were directly bound by IRF4 in regions not covered by our ChIP-CHIP analyses (Supplementary Fig. 5). IRF4 bound to the promoter and fourth intron of PRDM1, which encodes Blimp-1, another key regulator of plasmacytic differentiation (Supplementary Fig. 5). These observations support the proposal that IRF4 lies genetically upstream of PRDM1 in the regulatory hierarchy of terminal B cell differentiation3. Notably, IRF4 bound to its own promoter, supporting a positive feedback mechanism by which plasma cells can maintain high IRF4 expression3 (Supplementary Fig. 5).
A direct IRF4 target of particular interest is MYC, given its prominent role in the pathogenesis of myeloma16. Knockdown of IRF4 reduced MYC mRNA levels by more than 2-fold in myeloma cell lines and caused MYC DNA binding activity to decrease in nuclear extracts of myeloma cells. Conversely, ectopic expression of IRF4 in a lymphoma cell line increased MYC mRNA levels (Fig. 3a, Supplementary Fig. 6). By ChIP, we surveyed regions of the MYC locus for binding by IRF4 in myeloma cells and detected a peak of binding around −1.6 kb upstream of the MYC start site, coinciding with a region detected by ChIP-CHIP (Fig. 3b, Supplementary Fig.7). Knockdown of IRF4 expression diminished the amount of IRF4 bound to this region of the MYC promoter (Fig. 3c). In human B cells, phorbol myristate acetate (PMA) and ionomycin (P/I) treatment induces transcription of IRF4 and MYC (data not shown). Correspondingly, a sharp increase in IRF4 binding to the MYC promoter was detected after 3 and 20 hours of P/I activation (Fig. 3d). Genetic evidence that Myc is an IRF4 target was provided by analysis of mitogenically-stimulated wild-type and IRF4-deficient mouse B cells (Fig. 3e). In IRF4-deficient cells, both Myc and Prdm1 failed to be fully induced by P/I treatment whereas the immediate early gene fos was normally induced, and a housekeeping gene, Usf2, did not change in expression. Finally, ectopic expression of IRF4 in a lymphoma cell line was able to transactivate a reporter construct in which GFP is under the control of the MYC promoter (Fig. 3f).
These data provide strong evidence implicating MYC as a direct target gene of IRF4. Accordingly, the list of IRF4 targets was highly enriched for genes that are directly transactivated by MYC17-19 (n=23; p=1 × 10−8, Chi-square; Supplementary Table 2 and Supplementary Fig. 9). These genes encode key components of glycolysis (LDHA, HK2, PDK1) and mitochondrial respiration (ATP5D, CYCS), as well as important regulators of cellular senescence (BMI1, TERT). Since MYC is a key coordinator of cellular growth, metabolism and proliferation20, we examined whether knockdown of MYC expression was toxic to myeloma cells. An shRNA targeting the MYC 3'UTR knocked down MYC expression and DNA binding by ~2-fold (Supplementary Fig. 6). This shRNA was toxic to both myeloma and lymphoma cell lines but had little effect on the myeloma cell line U266, consistent with its high expression of MYCL1 instead of MYC (Fig. 4a)21. Expression of the MYC coding region was able to rescue cells from the toxicity of the MYC shRNA, confirming its specificity (Fig. 4b). Thus, loss of MYC expression may contribute to the toxicity of IRF4 shRNAs for myeloma cells.
Using two independent MYC shRNAs, we identified the targets of MYC in myeloma cells. Following MYC shRNA induction, the expression levels of many direct MYC targets decreased (Fig. 4c). Unexpectedly, the expression of IRF4 also decreased, as did the expression of many IRF4 target genes (Fig. 4c, d). ChIP demonstrated binding of MYC to a region of the IRF4 first intron in two myeloma cell lines expressing MYC (KMS12, H929) but not in a cell line with very low MYC expression (U266, Fig. 4e). Further, we detected MYC binding to IRF4 in mitogenically activated B cells, which express MYC, but not resting B cells, which do not (Fig. 4f).
These data reveal a positive regulatory loop in myeloma cells in which IRF4 and MYC mutually reinforce each other's expression (Fig. 5a). In keeping with this model, myeloma patient samples express both MYC and IRF4 mRNA more highly than normal plasma cells (p=5.1 × 10−7 for IRF4; Fig. 4g). Moreover, MYC and IRF4 mRNA levels showed significant positive correlation across 451 primary myeloma samples4 (r= 0.24, p=2.5×10−7, Supplementary Fig. 7). This moderate correlation was remarkable since IRF4 is likely to be only one of many factors regulating MYC transcription in myelomas22. Although the MYC locus in myeloma is often amplified and inserted at ectopic genomic locations, especially within and near the immunoglobulin loci16, the MYC breakpoints in these chromosomal rearrangements are many kilobases from the MYC transcriptional start site and thus preserve the IRF4 binding region. Our data suggest that the oncogenic activation of MYC in myeloma upregulates IRF4, which in turn drives expression of MYC and other IRF4 target genes (Fig. 5a).
In some respects, the dependency of myeloma cells on IRF4 is reminiscent of the function of “lineage-survival” oncogenes23. These genes are primarily transcription factors that provide essential functions in a particular cellular lineage but are also dysregulated in cancers derived from that lineage. IRF4 differs from lineage survival oncogenes in two respects. First, many lineage survival oncogenes are altered by mutations or chromosomal structural alterations whereas the IRF4 locus appears to be unaltered in most myelomas. Second, the regulatory network that IRF4 controls in myeloma is decidedly abnormal, not merely reflecting the genetic program of normal plasma cells but also borrowing from the genetic program of antigen-stimulated mature B cells (Figs. (Figs.2a,2a, ,5b).5b). This transcriptional promiscuity is exemplified by the direct IRF4 targets MYC, SCD, SQLE, CCNC, and CDK6, which are not highly expressed in normal plasma cells but are upregulated in mature B cells upon antigen receptor signaling (Figs. (Figs.2a,2a, ,5b).5b). Thus, myelomas have broadened the genetic repertoire of IRF4, perhaps due to epigenetic alterations that allow IRF4 access to loci that are normally silenced in plasma cells. Hence, the dependency of myeloma on IRF4 may be best described as “nononcogene addiction” i.e. the aberrant function of a normal cellular protein that is required for cancer cell proliferation or survival24.
The direct targets of IRF4 reveal it to be a master regulator influencing metabolic control, membrane biogenesis, cell cycle progression, cell death, transcriptional regulation and plasmacytic differentiation (Fig. 5b). Given this pleiotropic program, we believe that loss of IRF4 in a myeloma cell results in “death by a thousand cuts”. For example, several key cell cycle regulators are IRF4 targets, in keeping with its role in lymphocyte activation1, including STAG2, CDK6, and MYC. STAG2 encodes a component of the cohesin complex crucially involved in the segregation of chromosomes during mitosis25. Two different shRNAs targeting STAG2 were toxic for both a myeloma and a lymphoma cell line (Supplementary Fig. 8), as were shRNAs targeting MYC (Fig. 4a). Likewise, myeloma cells were specifically killed by 2 different shRNAs targeting SUB1, an IRF4 direct target that encodes a transcriptional coactivator26. It seems likely, therefore, that decreased expression of each of these IRF4 direct targets contributes to IRF4 shRNA toxicity. A prominent role for IRF4 in regulating membrane biogenesis was indicated by the many enzymes and regulators of sterol and lipid synthesis under its control (Supplementary Fig. 9), including SQLE and SCD, which encode rate-limiting enzymes in these pathways. Further, the IRF4 target gene set was strikingly enriched for genes encoding components of glucose metabolism and ATP production, many of which are targets of MYC (Supplementary Fig. 9). It is therefore plausible that metabolic collapse also contributes to cell death caused by IRF4 deprivation.
Our data demonstrate that myelomas are addicted to an abnormal regulatory network controlled by IRF4, irrespective of their molecular subtype and underlying oncogenic abnormalities. Hence, IRF4 emerges as a master regulator of an aberrant and malignancy-specific gene expression program relevant to all molecular subtypes of this cancer. Since mice lacking one allele of irf4 are phenotypically normal1 and since a ~50% knockdown of IRF4 mRNA and protein was sufficient to kill myeloma cell lines, a therapeutic window may exist in which IRF4-directed therapy might kill myeloma cells while sparing normal cells. Though transcription factors have been considered intractable therapeutic targets, recent successful targeting of p5327 and BCL-628 provides hope that IRF4 can be exploited as an Achilles heel of multiple myeloma.
Lines were engineered to express the ecotropic retroviral receptor and the bacterial tetracycline repressor as described10. The retroviral constructs for shRNA expression and the design of shRNA library sequences have been described10; in some vectors, the puromycin selectable marker (puro) was replaced by a fusion between puro and green fluorescent protein (GFP) for tracking by flow cytometry. Doxycyline (20 ng/ml) was used for shRNA induction. IRF4 and MYC were expressed using retroviral vectors as described3. Primary human resting blood B cells were purified by magnetic separation (CD19+ beads Miltenyi) and grown at 2 million cells/ml in IMDM+10%FBS; primary mouse splenic, resting B cells were purified by magnetic separation (B cell kit, Miltenyi) and grown at 2 million cells/ml in RPMI+10%FBS. Lymphocytes were activated with PMA (40 ng/ml) and ionomycin (2μM). Gene expression profiling was performed using Agilent 4×44k or Lymphochip29 microarrays. ChiP-CHIP experiments were performed using Agilent Human Promoter Set microarrays.
This research was supported by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research. We wish to thank Kathleen Meyer for her assistance with GEO submissions, David Levens, Juhong Liu, Hye-Jung Chung for assistance with MYC ChIP assay design and the MYC promoter-GFP reporter construct, Keiko Ozato and Lakshmi Ramakrishna for IRF4-deficient mice, and Mike Kuehl and the members of the Staudt lab for their assistance and helpful discussions.