Hypoxia-inducible factors 1 and 2 (HIF1 and HIF2) are heterodimeric transcription factors consisting of α regulatory subunits and a constitutively expressed β subunit. The expression of α regulatory subunits is promoted by hypoxia, cancer-associated mutations, and inflammatory cytokines. Thus, HIF1 and HIF2 provide a molecular link between cancer and inflammation. We have recently identified novel small molecules that selectively inhibit translation of the HIF2a message and thereby powerfully inhibit the expression of HIF2a target genes. We report here that Connectivity Map analysis links three of these compounds to the anti-inflammatory cytokine 15-deoxy-Δ12,14-prostaglandin J2 (PGJ2). As with our identified compounds, PGJ2 inhibits translation of the HIF2a message in a mammalian target of rapamycin–independent manner by promoting the binding of iron regulatory protein-1 (IRP1) to a noncanonical iron responsive element (IRE) embedded within the 5′-untranslated region of the HIF2a message. The IRE is necessary and sufficient for mediating the effect. Mutation of the IRE sequence, or downregulation of IRP1 expression, blocks the effect of PGJ2 on HIF2a translation. This is the first report of an endogenous natural molecule regulating HIF2a translation, and it suggests that part of the anti-inflammatory and putative antineoplastic effects of PGJ2 may be mediated through inhibition of HIF2a within tumor epithelial cells themselves and/or mesenchymal cells of the tumor microenvironment. Cancer Res; 70(8); 3071–9. ©2010 AACR.
Hypoxia-inducible factor (HIF) is directly linked to cancer progression, angiogenesis, and inflammation. HIF is critical for the transcriptional regulation of the cellular response to hypoxia. In addition to hypoxia, tumor-associated mutations that activate the phosphoinositide 3-kinase pathway, inactivate the von Hippel-Lindau (VHL) gene, or disrupt the mitochondrial Krebs cycle result in increased HIF activity (1). HIF is upregulated in the majority of human tumors and it has been proposed as a prognostic factor for aggressive disease. Many preclinical models indicate that HIF inactivation may lead to suppression of tumor growth (2–4) as well as regression of pathologic angiogenesis in nonmalignant diseases such as macular regeneration (5).
HIF has also been implicated in promoting inflammatory responses. Targeted deletion of HIF1a from macrophages of transgenic mice led to significant attenuation of experimentally induced serum sickness or chemically induced dermatitis (6). Recently, HIF1a has been identified as a direct transcriptional target of NF-κB (7). It is therefore not surprising that HIF has been proposed, and to a great extent validated, as a therapeutic molecular target for antineoplastic and anti-inflammatory interventions (8).
In an effort to discover HIF-based antineoplastic and anti-inflammatory compounds, we conducted a cell-based screen and discovered four small molecules that decrease HIF2a expression under conditions of normoxia and hypoxia (9). Microarray analyses indicated that these small molecules powerfully inhibited the expression of signature HIF2a target genes. As part of our studies to understand the mechanism of action of these inhibitors, we used the Connectivity Map to identify other small molecules that might mimic the gene expression changes induced by the HIF2a inhibitors. The Connectivity Map is a reference collection of gene expression profiles from cultured human cell lines (breast cancer epithelial cell line MCF7, prostate cancer cells PC3, nonepithelial leukemia line HL60, and melanoma cell line SKMEL5) treated with a large number of diverse bioactive small molecules. The current collection (build 02) contains data for 6,100 treatment instances representing 1,309 discrete small molecules. Comparative analysis between a “query” gene expression signature (generated by profiling cells treated with small molecules or corresponding to a physiologic or a disease specific process) and the Connectivity Map collection may highlight similar patterns of gene expression changes and therefore lead to the discovery of functional connections between drugs, genes, and diseases. In the case of our HIF2a inhibitor “query” signatures, the connectivity analysis strongly linked the activity of three of our four inhibitors to the endogenous anti-inflammatory 15-deoxy-Δ12,14-prostaglandin J2 (PGJ2).
In this article, we confirm this linkage experimentally and we show that PGJ2 inhibits HIF2a translation in a mammalian target of rapamycin (mTOR)–independent manner. PGJ2 acts by enhancing the binding of the iron regulatory protein-1 (IRP1) to the recently identified iron responsive element (IRE) located within the 5′-untranslated region (UTR) of HIF2a message (9, 10). Mutational analyses and reporter hybrid experiments indicate that the IRE element is both necessary and sufficient for the ability of PGJ2 to inhibit HIF2a translation. There is a growing body of evidence suggesting that PGJ2 is a key endogenous negative regulator of inflammation and angiogenesis (11) and that it constitutes one of the mechanisms used by cells to resolve the inflammatory response (12–14). We therefore propose that HIF2 is likely an important target through which PGJ2 exerts its anti-inflammatory and antiproliferative effects, and that this interaction is one of the endogenous mechanisms by which cells downregulate HIF2a levels upon restoration of normal ambient oxygen tension.
Materials and Methods
Plasmids, cell culture conditions, and luciferase assays
Complete description of plasmids and cell lines used in this study is provided elsewhere (9). Buffers were made as described in ref. 15, and chemicals ordered from Sigma. Cells were grown on DMEM-10% Fetal Clone and transfected with Lipofectamine 2000. Unless stated differently, 25% confluent cells were treated with either fresh DMEM or DMEM supplemented with DMSO-only or PGJ2 for 24 h and then changed into fresh DMEM or DMEM supplemented with DMSO-only or PGJ2 for an additional 24 h, before harvesting at 90% to 100% confluence. For hypoxic experiments, cells were incubated at the indicated oxygen concentration for the latter 24 h. All cell lines used in this work were purchased from American Type Culture Collection, aliquoted, immediately frozen, and replated for the purposes of these experiments within the last 9 mo. Luciferase assays were done using the Luciferase Assay Reporter System (Promega). Stable clones were normalized to medium-only–treated wells.
Gene expression and Connectivity Map analysis
Gene expression profiles of 786-0 cells, treated with small-molecule inhibitors or vehicle control, were generated using Affymetrix U133 Plus 2.0 microarrays and compound signatures were derived as described elsewhere (9). Raw gene expression data have GEO series accession no. GSE13818. Compound signatures from which non–HG-U133A probe sets were removed (Supplementary Table S1) were used to query the Connectivity Map (build 02). Details of the Connectivity Map data set and analytics are provided elsewhere (16).
Proteins were immunoblotted as described before (4). Primary antibodies (1:1,000) include anti-HIF1a and anti-HIF2a (Novus); anti–Glut-1 (Alpha Diagnostics); anti-p70S6K, anti–phospho-T389 p70S6K, anti–phospho-S235/6 S6, and anti-RhoB (Cell Signaling Technologies); or anti-actin (Novus). Gel densitometry was done using SynGene Gene Tools software package (Synoptics Ltd.).
Cycloheximide (10 μg/mL) was added to DMSO- or PGJ2-treated cells at t = −4, −3, −2, and −1 h and cells were lysed at t = 0. HIF2a expression was quantified by Western blot.
Quantitative real-time PCR
The following intron-spanning primers were used (forward and reverse, respectively): for β2-microglobulin, 5′-TTTCATCCATCCGACATTGA-3′ and 5′-ATCTTCAAACCTCCATGATG-3′; for HIF2a, 5′-GGATCAGCGCACAGAGTTC-3′ and 5′-GTACTGGGTGGCGTAGCACT-3′; and for EGLN3, 5′-ATCAGCTTCCTCCTGTCCCT-3′ and 5′-GGGCTGCACTTCGTGTGGGT-3′. Nascent HIF2a message levels were determined using intronic primers 5′-AGACGGTGGACTCCGCCA-3′ and 5′-TTAAAGGGAGGGGTACAC-3′.
In vivo [35S]methionine pulse-label experiments
Cells were treated with medium only, DMSO, or PGJ2; pulsed with [35S]methionine (NEN) for 30 min; and chased with 3 mg/mL methionine for 1.5 and 3 h. Cell lysate (500 μg) was immunoprecipitated with 2 μg of anti-HIF2a or control anti-HA Y-11 (Santa Cruz) antibodies.
Polysomal profile analysis (17), electrophoretic mobility shift assays (EMSA), and in vitro iron competition assays (9) were done as described. Prostaglandin D2-MOX EIA Kit (Cayman Chemical Company) was used for prostaglandin D2 (PGD2) ELISA (18).
Connectivity Map analysis links HIF2a inhibitors to PGJ2
In previous work, we described how we identified several small-molecule HIF2a inhibitors (9). To gain insight into the signaling pathways perturbed by these inhibitors, we generated gene expression signatures for these compounds in the human renal cell carcinoma (RCC) cell line 786-O and subjected these signatures to Connectivity Map analysis. 786-O cells are VHL-deficient cells and therefore constitutively expressed HIF2a. Comparison of gene expression changes induced by the small-molecule HIF2a inhibitors to the profiles included in the Connectivity Map database strongly suggested that three of four compounds (40, 41, and 76) shared a functional similarity with the anti-inflammatory PGJ2 (Fig. 1A).
PGJ2 decreases HIF2a activity in a dose-dependent manner
To validate the Connectivity Map–derived prediction that PGJ2 is involved in HIF signaling, we treated 786-O–derived, hypoxia response element (HRE)-luciferase–expressing (7H4) and SV40-luciferase–expressing (7SV) control cells with increasing doses of PGJ2. Fresh medium containing the indicated concentrations of PGJ2 or vehicle-only DMSO control was applied to cells at 24-hour intervals. Luciferase activity was measured at 48 hours. PGJ2 decreased HRE-driven luciferase activity (Fig. 1B) in a dose-dependent manner, with an IC50app of 5 μmol/L, whereas SV40-driven luciferase activity was not affected. This action of PGJ2 seems to be independent of NF-κB activation. Application of the IκB kinase inhibitor BMS-345541 did not alter the HRE-luciferase activity in 786-O cells (Supplementary Fig. S1A and B). Moreover, this seems also to be a peroxisome proliferator–activated receptor-γ (PPAR-γ)–independent function of PGJ2; the PGJ2/HIF inhibitor linkage holds strong even in Connectivity Map comparisons using expression profiles obtained from PPAR-γ–negative cells (MCF7), and application of a PPAR-γ inhibitor, GW9662, did not block the effect of PGJ2 on HIF2a activity (data not shown). This decrease in HRE-mediated luciferase activity closely matched the decrease in HIF2a protein expression as well as that of the HIF2a target genes Glut-1 (as measured by Western blot) and EGLN3 [as measured by quantitative reverse transcription-PCR (RT-PCR); Fig. 1C]. To investigate whether the effect of PGJ2 on HIF2a protein expression is due to decreased transcription of the HIF2a gene, we performed quantitative RT-PCR to examine HIF2a mRNA expression in PGJ2- or vehicle-only DMSO–treated 786-O cells. We observed no concomitant decrease in HIF2a mRNA expression, indicating that the transcription of HIF2a mRNA was not affected (Fig. 1D). To investigate whether PGJ2 alters the stability of HIF2a protein, we examined the half-life of HIF2a in PGJ2- or vehicle-only DMSO–treated 786-O cells following the addition of cycloheximide. These results show no indication that PGJ2 affects HIF2a protein half-life (Supplementary Fig. S2A and B).
The effect of PGJ2 on HIF2a expression and activity is not a cell line–specific phenomenon; treatment of several VHL-deficient human RCC lines with PGJ2 similarly resulted in decreased HIF2a protein expression (Supplementary Fig. S3A). Moreover, consistent with the fact that the HIF1a candidate IRE sequence seems to be nonfunctional (9), at least in the cell lines tested, we see no decrease in HIF1a expression in UMRC2 cells treated with PGJ2 (Supplementary Fig. S3B).
PGJ2 inhibits HIF translation in a mTOR-independent manner
To determine if PGJ2 decreases translation of the HIF2a message, we evaluated the synthesis rate and stability of the newly synthesized HIF2a by an [35S]methionine pulse-chase experiment on PGJ2- or vehicle-only DMSO–treated 786-O cells. PGJ2 was found to significantly decrease the amount of protein translated from the mRNA without decreasing the half-life of the protein (Fig. 2A, top). Loading control is 1/1,000th of the cell lysate used in the respective immunoprecipitation directly loaded into an SDS-PAGE gel. This loading control shows not only that equal protein was used in each immunoprecipitation but also that PGJ2 does not globally decrease cellular translation (Fig. 2A, bottom). Consistent with the interpretation that PGJ2 does not affect global translation, polysome analysis was done on vehicle-only DMSO–, rapamycin-, or PGJ2-treated 786-O cells. These experiments show that PGJ2 had little effect on the number of ribosomes actively engaged in translation, whereas rapamycin attenuated it. However, PGJ2, like rapamycin, seemed to slightly decrease the monosome fraction (Fig. 2B).
It has been shown before that mTOR inhibits HIF translation as part of a global effect on translation (9, 19–21). We therefore examined whether the effect of PGJ2 on HIF2a translation could be attributed to inhibition of mTOR activity. 786-O cells were treated with medium only, DMSO (vehicle only), rapamycin, or PGJ2 and the effect of these treatments on mTOR activity was monitored by detecting phospho-S6, p70S6K, and phospho-p70S6K levels, as measured by Western blot using total or phospho-specific antibodies. HIF2a protein expression and HRE activity were decreased by both PGJ2 and rapamycin. However, only rapamycin decreased phosphorylation of the downstream mTOR targets p70S6K and rpS6 (Fig. 2C).
The 5′-UTR of HIF2a mRNA is necessary and sufficient for the inhibitory effect of PGJ2
Because PGJ2 inhibits translation of the HIF2a message in a mTOR-independent manner, we next examined whether the 5′-UTR of the HIF2a message might be involved in mediating its effect. We therefore generated two luciferase reporter constructs, the first driven by the endogenous HIF2a promoter alone (HIF) and the second by the HIF2a promoter containing the 5′-UTR (HIF-UTR). These constructs were stably transfected into 786-O cells. Treatment with PGJ2 decreased the luciferase activity derived from the UTR-containing plasmid but not that from the plasmid driven by the HIF promoter alone. Moreover, this effect was equal in magnitude to the effect on HRE-driven reporter activity. The ratio of normalized HRE-luciferase over SV40-luciferase reporter activities is shown compared with the ratio of normalized HIF-UTR-luciferase over HIF-luciferase activities in vehicle-only DMSO– versus PGJ2-treated cells (Fig. 2D, top). These experiments indicate that HIF2a 5′-UTR is necessary for the inhibitory effect of PGJ2 on HIF2a translation and corroborate the quantitative RT-PCR findings that the effect of PGJ2 on HIF is not transcriptional.
We next wanted to test if the HIF2a 5′-UTR is sufficient for conferring sensitivity to PGJ2. To test this, we generated a luciferase reporter construct (SV-UTR) in which the entire 488-bp HIF2a 5′-UTR was cloned between the SV40 promoter and the luciferase gene of the same SV40-luciferase reporter (SV) used as a HIF/hypoxia-independent control in the previously discussed experiments. We compared this set of luciferase reporters to a second matched set consisting of a cytomegalovirus (CMV)- luciferase reporter alone and the same CMV-luciferase into which a synthetic stem loop was cloned between the CMV promoter and the start of the luciferase gene (CMV-SL; kind gifts of Michele Pagano, Department of Pathology and NYU Cancer Institute, New York University School of Medicine, New York, NY). This luciferase reporter set serves as a random 5′-UTR control that might capture possible effects on RNA helicase activity (22). Shown are the ratios of normalized CMV-SL over CMV and SV-UTR over SV luciferase reporter activities from vehicle-only DMSO– versus PGJ2-treated cells. The results indicate that the effect of PGJ2 that is mediated through the HIF2a 5′-UTR element is indeed heterologously transferable to a different promoter element, whereas PGJ2 had no effect on the synthetic 5′-UTR element (Fig. 2D, bottom).
The IRE within the HIF2a 5′-UTR is responsible for the effect of PGJ2 on HIF translation
To map the domain responsible for mediating the effect of PGJ2 on translation of the HIF2a message, we created several reporter constructs in which different segments of the HIF2a 5′-UTR were cloned between the HIF2a promoter and the start of the luciferase gene and used these constructs to generate stable 786-O–derived cell lines. These lines were then treated with vehicle-only DMSO or PGJ2. We found that the effect of PGJ2 maps to a 50-bp segment of the 5′-UTR that contains a recently reported IRE (9). A mutation in the IRE consensus loop completely abolished PGJ2 sensitivity. These results are summarized in Fig. 3A, where the normalized ratio of uciferase activities from PGJ2- over vehicle-only DMSO–treated cells is shown.
IRP1 is necessary for the effect of PGJ2 on HIF2a
The IREs are the mRNA target sequences for the binding of iron regulatory proteins IRP1 and IRP2 (23). Nevertheless, slight variations in the IRE sequences can result in higher affinity for one of the two proteins (24). We showed before that it is mainly IRP1 and not IRP2 that can measurably bind to the HIF2a IRE element, under normoxic conditions, when expressed in endogenous levels (9). We therefore examined whether IRP1 and/or IRP2 is required for the effect of PGJ2 on HIF2a by assessing the expression of the HIF2a target gene EGLN3 in 786-O–derived lines infected with shRNAs targeting IRP1, IRP2, or both IRP1 and IRP2 concomitantly (Fig. 3B). In keeping with our previous observation that HIF2a IRE translation is primarily repressed through IRP1, knocking down the expression of IRP1 resulted in increased activity of HIF2a, consistent with the observation that IRP1 represses HIF2a translation (Fig. 3B). More importantly, downregulation of IRP1 abolished the ability of PGJ2 to repress HIF2a activity, whereas inactivation of IRP2 did not (Fig. 3B). These findings are consistent with the model by which PGJ2 represses HIF2a translation by enhancing endogenous IRP1 mRNA binding activity.
PGJ2 directly enhances the binding of IRP1 to the HIF2a IRE
We therefore next sought to determine if PGJ2 affects the expression of IRP1 and/or IRP2, and we found that, as is the case for the three small-molecule HIF inhibitors that linked to PGJ2 in the Connectivity Map, IRP1 expression is not affected by PGJ2 treatment whereas IRP2 expression is minimally increased (data not shown). However, IRP2 contributes very little, if at all, to the total IRP bound to the HIF2a IRE (9). We therefore hypothesized that the effect of PGJ2 on HIF2a IRE is due to the enhanced binding of IRP1.
To examine this latter possibility, we performed an EMSA using a radiolabeled wild-type or mutant IRE probe to determine if IRP1 binding is increased in PGJ2-treated cells (Fig. 4A). We found that pretreatment of cells with PGJ2 increased the specific IRE binding activity of cell lysates under normoxic conditions. This shifted band can be supershifted with IRP1, but not IRP2, antibodies (data not shown). The effect of PGJ2 on IRP1 binding to the HIF2a IRE probe under 1% oxygen was very subtle but seemed to be consistently present. Given the scant effect of PGJ2 on enhancing IRP1 binding in hypoxia, as measured by EMSA, we decided to further examine the ability of PGJ2 to decrease HIF2a activity at various concentrations of ambient oxygen that correspond to physiologic or extreme conditions of tissue oxygenation. The results indicate that PGJ2 clearly has the ability to repress HIF2a activity at a wide range of oxygen concentrations spanning the range of physiologic tissue oxygenation (Fig. 4B). Rapamycin, which decreases HIF2a activity and expression (Fig. 5A and B), does not affect IRP1 binding to the HIF2a IRE (Fig. 5C). Instead, it decreases HIF2a transcription (Fig. 5D), as reported before (25).
Connectivity mapping linked the gene expression signature of small-molecule HIF2a inhibitors to the molecular signature of PGJ2. Here we experimentally validate this hypothesis and we show that this function of PGJ2 is mediated through enhanced IRP1 binding to the HIF2a IRE. This is the first report of an endogenous cellular metabolite that regulates HIF2a translation through the IRE mechanism.
Activation of the arachidonic acid cascade generates prostaglandin H2 (PGH2) through the activity of cyclooxygenase synthases. PGH2 can be further metabolized to prostaglandin E2 (PGE2) or PGD2 through corresponding synthases. PGJ2 is a nonenzymatic degradation product of PGD2. PGE2, in contrast to PGJ2, has the ability to promote HIF1a activity (26, 27) through activation of the epidermal growth factor receptor/Src/extracellular signal–regulated kinase signaling pathways (28, 29). In addition, PGE2 promotes colonic adenomatous polyposis through PPAR-δ activation (30, 31). It therefore seems that PGE2 promotes the “proinflammatory” and “pro-proliferative” effects of the arachidonic acid cascade. In contrast, PGJ2 inhibits HIF2a translation and activity. Although PGJ2 has recently been reported to stabilize HIF1a through inhibition of a lysosomal degradation pathway (32), we see no such effect in PGJ2-treated UMRC2 cells. It seems that PGJ2 may “oppose” the effects of PGE2 and therefore may be a key determinant of the final outcome of arachidonic acid signaling.
One possibility to explain the link between HIF2a inhibitors and PGJ2 is that the former increase the endogenous expression of PGJ2, a nonenzymatic degradation product of PGD2. The latter is made from PGH2 by lipocalin or hematopoietic PGD2 synthases. Neither the addition of the cyclooxygenase-1/2 inhibitors indomethacin and sulindac nor the hematopoietic PGD2 synthase inhibitor HQL-79 had any effect on the activity of the compounds (Supplementary Fig. S4A). Similarly, these pharmacologic manipulations had no effect on baseline HIF2a expression in 786-O cells (data not shown). We were unable to detect PGD2 in compound-treated 786-O cell supernatants directly, using a commercially available ELISA kit (data not shown). However, we found that PGJ2 and all compounds increased the expression of RhoB, an endogenous PPAR-γ target gene (Supplementary Fig. S4B and C; ref. 33). Taken together, these data suggest, but clearly do not prove, that the compounds may serve, directly or indirectly, as PGJ2 mimetics. Future experiments, involving direct measurement of PGD2 and PGJ2 by mass spectrometry as well as cells generated from PGD2 synthase–knockout mice, should answer definitely the relation between HIF2a inhibitors and endogenous PGJ2.
An analysis of the Oncomine database shows that the expression of IRP1 and lipocalin-type PGD2 synthase is downregulated in kidney cancer. Specifically, 3 of 17 analyses (2 of 7 data sets) showed decreased IRP1 expression in tumor versus normal with a P value of <0.0001. Similarly, 3 of 14 analyses (2 of 6 data sets) showed decreased lipocalin PGDS expression, also with a P value of <0.0001. These data suggest that there may be selective pressure in RCC to reduce the activity of IRP1. Of relevant interest also is the observation that constitutional deletion of hematopoietic PGD synthase promotes colonic polyposis in APC+/− mice (34).
PGJ2 has been reported to inhibit global translation of cellular proteins through binding to and inactivation of eIF4A and sequestration of TRAF2 into stress granules (35). In our work, we used lower concentrations of PGJ2 that did not affect global translation. It is likely that the first response of cells to moderate levels of PGJ2 concentrations is to selectively downregulate HIF2a translation, whereas higher doses may be associated with a global reduction of protein translation.
Upregulation of IRP1 binding activity by PGJ2 is likely to promote iron uptake and availability in sites of inflammation. Iron is essential for the oxidative burst of polymorphonuclear cells and macrophages and is therefore important for the maintenance of local inflammatory and immune responses (36, 37). The availability of PGJ2 may be a mechanism by which the end of inflammatory phase (by inhibiting translation of the HIF2a message) is linked to restoration of iron stores in resident inflammatory and immune cells.
In cells, the majority of IRP1 functions as a cytosolic aconitase. Reduction of cellular iron stores causes a conformational change in the iron-sulfur cluster (ISC) of IRP1 protein that concomitantly decreases its aconitase activity and promotes its binding to IREs (38). Our data indicate that changes in the intracellular levels of PGJ2 promote the RNA binding activity of IRP1. The exact mechanism by which PGJ2 signals to IRP1 will be the object of further studies. Currently, we have shown that PGJ2 does not act as an iron chelator, and therefore it is likely to affect IRP1 through a novel, ISC-dependent or ISC-independent mechanism (Supplementary Fig. S5).
In summary, we provide evidence for a novel connection between cancer and inflammation in which programs involved in the resolution of inflammation may modify tumor vascularization. This connection may contribute to new strategies of cancer chemoprevention.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
We thank M. Pagano for the pCMV and pCMV-SL constructs and N. Dyson and J. Settleman for their comments on the manuscript.
Grant Support: NIH grant 5R01CA104574 and the MGH Bertucci Center for Genitourinary Oncology Award (O. Iliopoulos), the VHL Family Alliance, the DF/HCC Kidney Cancer Program Career Development Award (M. Zimmer), and NIH Genomics Based Drug Discovery-Target ID Project grant RL1HG004671, administratively linked to NIH grants RL1CA133834, RL1GM084437, and UL1RR024924 (J. Lamb and T.R. Golub).
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.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
- Received August 3, 2009.
- Revision received January 6, 2010.
- Accepted January 25, 2010.
- ©2010 American Association for Cancer Research.