Definition of a Novel Feed-Forward Mechanism for Glycolysis-HIF1α Signaling in Hypoxic Tumors Highlights Aldolase A as a Therapeutic Target

Glycolysis is defined as the sequence of 10 enzymatic reactions converting glucose to pyruvate, which is accompanied by release of energy in the form of ATP. In normal cells, pyruvate then enters the mitochondrial carboxylic acid cycle in the presence of oxygen, or is converted to lactic acid in its absence. Glycolysis is critical for providing rapidly dividing normal and cancer cells with energy and metabolic intermediates to synthesize cellular biomass (1, 2).

Malignant transformation greatly increases aerobic glycolysis (the Warburg effect; ref. 3), which favors production of additional ATP and metabolites for biomass synthesis, and enables uncontrolled proliferation. Solid tumors also have an abnormal vasculature that leads to poor blood perfusion and hypoxia. A cancer cell's response to hypoxia is mediated by the hypoxia-inducible transcription factors HIF-1 and 2 (4), which increase the expression of numerous survival factors, including genes that encode VEGF (5, 6). HIF-1 also upregulates the transcription of several glycolytic enzymes (7). Both glycolysis and HIF activity are critical for cancer cell survival, and have been proposed as therapeutic targets for agents that inhibit tumor growth (8–11).

Although HIF-1 is critical for tumor growth, in part by inducing glycolytic enzymes, our findings suggest that glycolysis is necessary for maintaining HIF-1 activity. This constitutes a feed-forward loop that promotes increased HIF-1 activity and glycolysis, which we show is mediated by inhibition of the AMPK1/EA1-binding protein p300 pathway. HIF-1 is often studied in terms of its protein levels, but we show here that HIF-1 may not be functionally active if glycolysis is limiting. Inhibiting the glycolysis-HIF-1 feed-forward loop, therefore, offers a novel target for blocking tumor energy and biomass production and the HIF-1 survival response. Although HIF-1 and glycolysis have previously been proposed as targets for cancer treatment, efforts to develop inhibitors have been unsuccessful. Here, we used an unbiased approach to identify the glycolytic enzyme, fructose-bisphosphatate aldolase A (ALDOA) as a key target for inhibiting both glycolysis and HIF1 activity. By using an inhibitor that targets ALDOA, we found that inhibition of ALDOA does indeed break the feed-forward loop, blocking both glycolysis and HIF-1 activity in cells, with the prospect of inhibiting tumor growth in vivo.

MIA PaCa-2 and PANC-1 pancreatic, MDA-MB-231 metastatic breast, HT-29 colon and 786-O renal cell carcinoma cancer cell lines were obtained in 2012 from the ATCC. The identity of each line was authenticated on arrival, for each frozen stock and at two month intervals during culture by the Molecular Cytogenetics Facility, University of Texas MD Anderson Cancer Center. Each cell line was stably transfected with a pGL3 plasmid (Promega) containing a 5x repeat of the hypoxia transcriptional response element (HRE) flanking a luciferase reporter and a G418 selection marker (HRE-luc). The reporter plasmid was a gift of Dr. R. Gillies (Moffitt Cancer Center, Tampa, FL). Following selection, pools of stably transfected cells were generated and stored frozen for later use.

For conditional Aldolase A (ALDOA) knockdown in an in vivo murine model, four sequences predicted to target ALDOA gene expression were selected from the Thermo Scientific Dharmacon shRNA library and each was inserted in a TRIPZ lentiviral vector (Open Biosystems). The HRE luciferase MDA-MB-231 line described above was transduced with shALDOA-expressing lentivirus, and stable lines were selected in puromycin in 96-well plates with one cell per well to generate clonal populations. Sequence identification for use in both in vitro and in vivo experiments was determined by relative ALDOA by Western blot analysis. After puromycin- and G418-resistant clones were selected, shALDOA expression in cells was induced using 400 ng/mL doxycycline in both normoxia and hypoxia (1% O₂) and for in vivo tumors by feeding mice chow containing 625 mg/kg doxycycline (Harlan Laboratories) to achieve ALDOA knockdown.

Transient siRNA reverse transfections were carried out for global siRNA screening using X-tremeGene (Roche) according to the manufacturer's instructions with the genome-wide SmartPool siRNA library from Dharmacon using the MIA PaCa-2 HRE luciferase line. After identifying initial glycolysis genetic hits, follow-up work in each of the three additional cell lines listed used Lipofectamine RNAiMax (Qiagen) and Dharmacon SMARTpool siRNAs for HIF-1α, Aldolase A, AMPK, p300, PCAF, FIH, PLK-1 or the On-Target-Plus non-targeting pool #4 (OTP4). Total siRNA concentration was kept at 40 nmol/L for single or multiple siRNA combinations. Knockdown efficiency was determined by Western blotting of cell lysates 96 hours posttransfection.

Synthesis of naphthalene-2,6-diyl bis(dihydrogen phosphate) is described in Supplementary Materials and Methods S1.

Primary antibodies for Western blotting were: HIF-1α (BD Biosciences), Aldolase A (Thermo Scientific), β-actin, p300, phospho-p300 (all from Santa Cruz Biotechnology Inc.), and AMPK/phospho-AMPK (Cell Signaling Technology).

Viability of cell populations was quantified photometrically at 475 nm using the XTT Cell Viability Assay (Biotium), according to the manufacturer's instructions. HIF-1α activity was measured using a Dual-Glo Luciferase Assay System (Promega) according to the manufacturer's protocol. Relative Luciferase activity (% control) was calculated to correlate HIF-1 expression with cell viability data for each gene knockdown.

Cellular ATP was measured using an ATP Assay Kit (Abcam) according to the manufacturer's protocol and quantified 96 hours after siRNA transfection by both colorimetric (OD 570 nm) and fluorometric (Ex/Em = 535/587 nm) methods.

Glycolysis was measured as the rate of extracellular acidification (ECAR), using the Seahorse Bioscience XF96e platform (Seahorse Bioscience) and the XF Glycolysis Stress Test Assay according to the manufacturer's protocol. To measure glycolysis under hypoxia, a modified hanging drop tissue culture method was used to evaluate 3-dimensional spheroids of PANC-1 HRE cells transduced with shALDOA constructs. Three days after seeding cells and 24 hours before measuring glycolysis spheroid shALDOA expression was induced with 400 ng/mL doxycycline. A final volume of 175 μL of pre-conditioned assay medium containing 18 spheroids was added to each well of a test plate and incubated at 37°C in a CO₂-free incubator until the experiment was initiated. Spheroids exhibited a hypoxic core based on analysis with a fluorescent hypoxia probe LOX-1 (SCIVAX USA, Inc.) without the need for hypoxic gassing conditions.

ALDOA kinetic assays are described in Supplementary Materials and Methods S2.

Protein crystallization and structure solution are described in Supplementary Materials and Methods S3, and data collection and refinement statistics in Supplementary Tables S1 and S2.

Approximately 10⁷ MDA-MB-231 HRE cells, MDA-MB-231 cells harboring shALDOA clones 8.8 and 9.7, and MDA-MB-231 HRE empty vector cells, all in log cell growth, were suspended each in 0.2 mL PBS and injected subcutaneously into the mammary fat pads of female NOD-SCID mice. Groups contained five mice. When the tumors reached 250 mm³, chow containing doxycycline was substituted for control feed (Harlan Laboratories) in test groups. Mice were euthanized when they became clinically moribund, associated with the metastatic spread of the MDA-MB-231 tumor to liver and lung (12). Animal studies were approved by SBPMRI's Animal Care and Use Committee.

Data are shown as mean ± SD unless indicated otherwise. The Student t test assuming two-tailed distributions was used to calculate statistical significance between groups. Animal survival was determined by the Kaplan–Meier analysis with P < 0.05 considered statistically significant.

To identify genes that regulate HIF activity, we conducted a genome-wide siRNA screen using MIA PaCa-2 pancreatic cancer cells with a stably integrated 5 x HRE/promoter-luciferase (luc) HIF reporter to identify genes that when knocked down inhibited HIF activity under hypoxic conditions (1% O₂). It should be noted that the reporter cannot distinguish between HIF-1 and HIF-2 activity, although hypoxic MIA PaCa-2 cells express predominantly HIF-1, and HIF-2 has been reported not to upregulate glycolysis genes (5). Unexpectedly, the screen identified several glycolysis-related genes whose knockdown inhibited HIF-1 activity (Supplementary Table S3). The expression of glycolysis genes has been reported to be increased by HIF-1 during hypoxia (5, 13). We confirmed this using RNAseq in MIA PaCa-2 cells, and found that 16 glycolysis genes were upregulated in hypoxia, 15 of which showed HIF-1 dependence (Supplementary Table S4).

To test the possibility that a glycolysis–HIF-1 feed-forward loop existed, we used a panel of siRNAs to knockdown 30 glycolysis genes and their isoforms (including pyruvate dehydrogenase, which is responsible for linking glycolysis to the citric acid cycle in the mitochondria), and then measured HRE-luciferase activity (a surrogate of HIF-1 activity) and cell proliferation. Using the same MIA PaCa-2 pancreatic cancer cells stably transfected with the HRE-luc reporter, we found compelling evidence that glycolytic enzyme activity is indeed critical for the normal functionality of HIF-1 (Fig. 1A). Similar results were obtained using HRE-luc PANC-1 pancreatic, MDA-MB-231 metastatic breast, and HT-29 colon cancer cell lines (Supplementary Fig. S1 and Supplementary Table S5). The greatest decrease in HRE-luc activity in all lines, normalized to cell number, was observed when ALDOA was knocked down. Similar inhibitory effects were observed in the 786-O renal adenocarcinoma cell line that expresses HIF-2 exclusively, suggesting that regulation of HIF activity by glycolytic enzymes is not limited to HIF-1 (Fig. 1B). Dual knockdown of PGK1 and PGK2 suggested they have additive activities on HRE-luc inhibition, although this was not accompanied by an additive decrease in cell proliferation (Fig. 1C, and Supplementary Fig. S2). Knockdown of ALDOA, or PGK1 or PGK2 in combination inhibited glycolysis in all cell lines (Fig. 2A and B), which was accompanied by lower cellular ATP levels (Fig. 2C and Supplementary Fig. S3).

We first established that, in all tumor cells tested, decreased HIF-1 activity caused by ALDOA or PGK1 or 2 knockdown occurred without changes in HIF-1 protein levels, and was similar in normoxia or hypoxia, although a greater overall effect was seen in hypoxia, where HIF-1 levels are elevated (Fig. 3A and Supplementary Fig. S4). In hypoxia, the hydroxylation of key HIF-1 proline residues by oxygen-sensitive dioxygenases is inhibited, thus preventing HIF-1 binding to the von Hippel-Lindau protein (pVHL), which normally leads to ubiquitination of HIF-1 and its proteasomal degradation (4). To evaluate the possible mechanisms underlying this change in HIF-1 activity, we co-transfected glycolysis-related siRNAs together with siRNAs targeting genes that are known to regulate HIF-1 activity. These included AMP-activated protein kinase (AMPK; refs. 14, 15); E1A-associated cellular p300 transcriptional coactivator (p300; refs. 16, 17); PCAF (p300/CBP-associated factor; refs. 18, 19); and FIH (factor inhibiting HIF-1), which interacts with HIF-1α and VHL to repress HIF-1 transcriptional activity (20, 21). We found that AMPK knockdown rescued HIF-1 inhibition caused by ALDOA or PGK2 knockdown, but observed little effect when either p300 or PCAF was knocked down (Fig. 3B). FIH knockdown also negated the effects of ALDOA and PGK1 knockdown on HIF-1 activity, likely due to loss of negative regulation of HIF-1. Western blotting showed that ALDOA knockdown significantly increased phosphorylation of AMPK on Thr¹⁷², a marker of AMPK activation in response to cellular stress such as ATP depletion (22, 23), in both normoxia and hypoxia (Fig. 4A). Further evidence that AMPK mediates the effects of ALDOA (or PGK1 or PGK2) knockdown on HIF-1 activity was the rescue of HIF-1 activity by the AMPK inhibitor, dorsomorphin (Fig. 4B). AMPK activation can lead to phosphorylation of p300 at Ser89 that attenuates the interaction of p300 with a variety of transcription factors in vitro and in vivo, including HIF-1 (24). We observed that knockdown of ALDOA or PGK2, although not PGK1, resulted in p300 Ser89 phosphorylation in HT-29 and MiaPaCa-2 cells (Fig. 4A). Taken together, the results suggest that inhibition of HIF-1 activity and a concomitant decrease in glycolysis are mediated by AMPK activation (possibly in response to low cellular ATP levels), which in turn promotes p300 phosphorylation, preventing p300 from coactivating HIF-1 transcriptional activity (Fig. 4C).

To further understand the basis for variability in the effects of ALDOA knockdown on proliferation among the 4 HRE luciferase lines, we next tested for possible compensatory mechanisms when ALDOA is eliminated, by immunoblotting to analyze expression of aldolase B and C isoforms in normoxia and hypoxia (Supplementary Fig. S5A and S5B). Interestingly, we found that HT-29 cells (a human colorectal adenocarcinoma cell line with epithelial morphology), which were the only cells found to be recalcitrant to ALDOA knockdown (Supplementary Fig. S1), were also the only cells to exhibit significantly higher expression of ALDOC, suggesting a compensatory effect in the absence of ALDOA at this step of glycolysis.

To validate ALDOA as a potential therapeutic target, and because we have shown that ALDOA knockdown is acutely toxic to cancer cells in vitro, we expressed a doxycycline-inducible ALDOA shRNA in MDA-MB-231 breast cancer cells, and then used those cells to establish an orthotopic model of metastatic breast cancer in female NOD-SCID mice (12). Two clonal lines that showed complete (clone 8.8) or partial (clone 9.7) ALDOA knockdown, glycolysis inhibition, and hypoxia response element (HRE) activity inhibition following doxycycline treatment, were used (Fig. 5A and Supplementary Fig. S6). Mice fed doxycycline starting either 1 week before implantation of cells or when the primary tumor reached approximately 250 mm³, showed an increased median lifespan. Mice fed doxycycline a week before implantation showed increases from 37 days in parental, and 41 days in empty vector transfected cells, to 50 and 56 days in two clonal cells lines (P < 0.001 in both cases); mice treated after tumors were established showed increases from 41 days (untreated) to 48 days (P < 0.001 compared to control; Fig. 5B). Transduction with clone 8.8, which showed complete ALDOA knockdown associated with glycolysis and HIF-1 inhibition, promoted longer median survival time than did the partial knockdown clone 9.7. Postmortem analysis of mice indicated marked tumor metastasis to the lung and liver. Although knockdown of ALDOA extended lifespan, doxycycline given either as pretreatment or when tumors were established had a similar effect. This suggests that ALDOA knockdown does not affect tumor implantation; rather, that ALDOA knockdown inhibits metastasis from the primary tumor to the lungs and liver, which occurs late in the tumor development process, and is the most likely cause of death.

To confirm that ALDOA is a target with potential for cancer therapy, we sought a small-molecule probe inhibitor of human ALDOA. We carried out a chemical library screen for inhibitors of ALDOA using a novel biochemical assay, which allowed us to identify small-molecule inhibitors that would have interfered with the classic ALDOA assay. We identified the compound TDZD-8 as an inhibitor with time-dependent inhibition (Fig. 6A and B). To determine the mechanism of inhibition, we first determined the crystal structure of native human ALDOA at a resolution of 2.4 Å (see Supplementary Materials and Methods S3 and Supplementary Table S1). In the native crystals, the C-terminal tail (“C-tail”; residues 345–363) lies across the active site, with the C-terminal Tyr363 inserted into the active site (Supplementary Fig. S7B). There is abundant evidence that the C-tail is highly mobile and that its conformation and dynamics are critical for catalysis (25). Indeed, we observed a 10-fold reduction in kcat in an ALDOA construct truncated before the C-tail (not shown). We next soaked a preformed native crystal in a solution containing 1 mmol/L TDZD-8, and determined its structure at 2.65 Å resolution (Supplementary Table S1). We found unambiguous evidence for TDZD-8 binding covalently to a single site, Cys239, which lies on an exposed loop distal to the catalytic site (Supplementary Fig. S7B). Computational docking studies support our observation that the thiadiazole ring of TDZD-8 can bind to the sulfhydryl group of Cys239 without the need for significant conformational changes in the protein.

In the crystalline form, however, lattice contacts reduce the mobility of the protein, which may obscure ligand-binding sites that are available in solution. This may be especially true for ALDOA, given the known mobility of the C-tail. In the course of our studies, we had cocrystallized ALDOA bound to the active site substrate-mimetic, ND1, and solved its structure at 2.2 Å resolution (Supplementary Table S2). The binding of ND1 sterically occludes the C-terminal tyrosine, causing the entire C-tail to be ejected from its groove and become disordered, a phenomenon that is typical of this class of inhibitor (Supplementary Fig. S7A). We therefore proceeded to soak preformed ALDOA-ND1 crystals with TDZD-8, and solved its crystal structure at 2.4 Å resolution (Supplementary Table S2). In this case, we observed strong ligand binding at C239 (as before) as well as at a second site, C289, which is also distal to the active site but proximal to the C-tail (Fig. 6C–E and Supplementary Fig. S7C and S7D). C289 is partly buried in the native structure, and the binding of TDZD-8 to C289 induces local conformational changes in loops that would contact the C-tail in native crystals.

Our data suggest that the reactivity of Cys239 toward TDZD-8 is not significantly influenced by the absence or presence of the C-tail; by contrast, the reactivity of Cys289 is strongly influenced, because binding is only observed when the C-tail has been removed. It should therefore follow that ligand binding to C289 in solution should perturb the conformation of the C-tail, and thereby modulate catalytic activity. Indeed, in solution, we observed a near-doubling of the half-life of inactivation by TDZD-8 (from 120 to 214 min) when Cys289 was mutated to Ala (Fig. 6B). Thus, TDZD-8 appears to act as an allosteric inhibitor, via modulation of the structure and/or dynamics of the C-tail, mediated principally through modification of Cys289, a residue that lies within a well-defined, three-dimensional pocket.

TDZD-8 treatment of MDA-MB-231 breast cancer cells inhibited glycolysis and HIF-1 activity as well as cancer cell proliferation in a dose-dependent manner at low μmol/L concentrations (Fig. 6F and Supplementary Fig. S8), and also decreased cellular [ATP] while increasing phospho-AMPK (Fig. 6G and Supplementary Fig. S9). Treatment of MDA-MB-231 cells with TDZD-8 under hypoxic conditions (1% O₂) for 6 hours resulted in an approximately 2-fold reduction of dihydroxy acetone phosphate (DHAP), the product of the cleavage of fructose-1,6-bisphosphate substrate by ALDOA, and pyruvate, whereas levels of the lower-abundance intermediates 3-phosphoglycerate and phosphoenolpyruvate were not affected (Supplementary Fig. S10).

We therefore used TDZD-8 as a pharmacologic probe to see whether we could show an antitumor effect. When administered intraperitoneally daily for 20 days at a dose of 12 mg/kg per day to mice with MDA-MB-231 orthotopic breast cancer tumors, TDZD-8 caused significant slowing of tumor growth, by about 60% by day 32 (Fig. 6H). Pharmacodynamic studies after a single dose of TDZD-8 showed an approximately 50% decrease in tumor lactate levels within 4 h, a 40% decrease in DHAP, and a slower decrease with daily dosing for 5 days in downstream phosphoglycerate (Fig. 6I). Thus, TDZD-8 itself at low levels, or an active metabolite appears to reach the tumor in sufficient amounts to inhibit tumor glycolysis, associated with antitumor activity.

We have shown that HIF-1–induced upregulation of glycolytic genes during anaerobic (hypoxic) glycolysis in cancer cells is itself stimulated by the product of glycolysis, ATP, thereby completing a glycolysis HIF-1 feed-forward loop that stimulates tumor growth. When anaerobic glycolysis is inhibited (e.g., via inhibition of ALDOA), ATP levels are reduced, and the feed-forward loop is broken via activation of the AMP-activated protein kinase-1 (AMPK1), which is sensitive to the ratio of AMP/ATP in the cell (22, 23). Activated AMPK1 inhibits the transcriptional activity of HIF-1 by phosphorylating its transcriptional coactivator, the E1A-associated, cellular p300 (p300), at Ser 89, which blocks formation of the p300–HIF-1 coactivation complex.

HIF-1 is critical for cancer cell survival and tumor growth in the stressed hypoxic tumor environment (4, 5). However, HIF-1 inhibitors used as single agents for the treatment of human cancer have not advanced in the clinic. Thus, we took an unbiased approach, using a high-throughput siRNA synthetic lethal screen, initially to identify genes that might be targets to inhibit tumor growth, which could also be used in combination with HIF-1 inhibition. Unexpectedly, several hits in the screen were glycolytic enzymes. Although it was known that HIF-1 increases glycolytic activity in tumors by inducing expression of glycolytic enzymes (5, 7, 26), it had not previously been reported that glycolysis increases HIF-1 activity. We found that the knockdown of 16 of 30 glycolytic enzymes and isoforms were associated with detectable inhibition of HIF-1 transcriptional activity. Among them, ALDOA and PGK1/2 knockdown resulted in robust inhibition of HIF-1 activity in all lines tested. Importantly, inhibitors that target these proteins should have the ability to block two processes critical for tumor growth: glycolysis, the source of energy and metabolic support (this could be tumor or stroma glycolysis); and HIF-1, which promotes cancer cell survival and tumor growth through increased angiogenesis.

We chose ALDOA as a proof-of-principle target for inhibitor development, in part because its knockdown in cancer cells was associated with greater inhibition of cancer cell proliferation than PGK1/2 knockdown. This could be because ALDOA has other roles, including “moonlighting” as a nuclear protein (27), or that ALDOA is more important as a driver of glycolysis, whereas PGK1/2 is a driver of tumor angiogenesis, which would not be apparent from our cell-based studies. ALDOA expression has been reported to be significantly elevated relative to other glycolytic enzymes in a number of human tumor types (28, 29). Another consideration is that ALDOA is the major ALDO isoform driving glycolysis in cancer cells, sometimes aided by ALDOC, whereas PGK1 and PGK2 appear to have redundant activities.

To demonstrate the potential of small-molecule ALDOA inhibitors for cancer therapy, we turned to a probe inhibitor of human ALDOA that we discovered through a chemical library screen, using a novel biochemical assay. The compound, TDZD-8, a 1,2,4-thiadiazole, showed time-dependent inhibition of ALDOA that suggested a covalent interaction with the protein. Our crystallographic studies showed that TDZD-8 bound to 2 Cys residues (Cys289 and Cys239) on the surface of each ALDOA monomer. Although both residues lie distal to the active site, one of them Cys289, lies in a well-defined pocket, and we found its reactivity toward TDZD-8 to be strongly influenced by the presence or absence of the C-tail. Thus, we propose that TDZD-8 binding to Cys289 in solution should allosterically perturb the conformation or flexibility of the C-tail, thereby inhibiting catalytic activity. The electron density derived by crystallography is consistent with disulfide bond formation and ring-opening of TDZD-8. Studies using Cys-directed reagents to inhibit ALDO Cys residues have previously suggested that they are involved in enzyme activity (30). Of ALDO's 8 Cys residues, only 4 are accessible in the absence of denaturing agents, and these include Cys-289 and Cys239. TDZ8-8 has allowed us, for the first time, to demonstrate crystallographically an allosteric interaction between ALDOA Cys289 and the catalytic site. Most importantly, although TDZD-8 is a simple chemical probe without optimized drug-like properties, it has nonetheless allowed us to demonstrate an association between inhibition of glycolysis, HIF-1 activity and the proliferation of cancer cell lines at low μmol/L concentrations. The compound also exhibited in vivo antitumor against MDA-MB-231 xenografts in mice and was associated with decreased levels of the glycolytic products of ALDOA activity in tumors.

In summary, we have shown that a feed-forward loop in tumors, simultaneously promoting increased HIF-1 activity and increased glycolysis, offers a target, ALDOA, with which to block tumor energy/metabolite production pathways and the HIF-1α survival response. Our HIF-1 activity-oriented RNAi screen and subsequent mechanism-based analysis expand our understanding of known and novel regulators of the HIF-1 transcription factor, and point to a previously uncharacterized regulation of HIF-1 activity by increased glycolytic enzyme activity.

No potential conflicts of interest were disclosed.

Conception and design: G. Grandjean, M.Y. Koh, J. Kingston, L.A. Bankston, A. Devkota, R.C. Liddington, K.N. Dalby, G. Powis

Development of methodology: G. Grandjean, P.R. de Jong, B.P. James, M.Y. Koh, J. Kingston, A. Aleshin, A. Devkota, G. Stancu, K.N. Dalby, G. Powis

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Grandjean, P.R. de Jong, R. Lemos, J. Kingston, A. Aleshin, L.A. Bankston, C.P. Miller, G. Stancu, R.C. Liddington, K.N. Dalby, G. Powis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Grandjean, P.R. de Jong, B.P. James, M.Y. Koh, R. Lemos, J. Kingston, A. Aleshin, L.A. Bankston, C.P. Miller, R. Edupuganti, A. Devkota, G. Stancu, R.C. Liddington, K.N. Dalby, G. Powis

Writing, review, and/or revision of the manuscript: G. Grandjean, L.A. Bankston, E.J. Cho, R. Edupuganti, R.C. Liddington, K.N. Dalby, G. Powis

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Grandjean, R. Lemos, E.J. Cho, R. Edupuganti, G. Powis

Study supervision: G. Grandjean, K.N. Dalby, G. Powis

Supported by NIH grants CA163541, CA188260 (to G. Powis) and CCSG grant P30CA030199. The help of SBPMDI Cancer Center Animal and Genomic Services is gratefully acknowledged.

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.