Acridine Derivatives as Inhibitors of the IRE1α–XBP1 Pathway Are Cytotoxic to Human Multiple Myeloma

The endoplasmic reticulum (ER) is the central organelle where newly synthesized proteins mature and are properly folded. A variety of stresses, including glucose deprivation, hypoxia, and chemotherapeutic treatment, cause accumulation of unfolded or misfolded proteins inside the ER, resulting in ER stress. In response to ER stress, the cell initiates the unfolded protein response (UPR) to restore protein folding homeostasis. The UPR actively reduces protein translation, increases expression of ER chaperones and protein-folding enzymes, and clears misfolded proteins for degradation (1). However, under prolonged ER stress, the UPR paradoxically also induces cell death by apoptosis (2).

In mammalian cells, the UPR consists of three primary signaling pathways, each of which includes an ER membrane-bound sensor protein that activates a b-ZIP (Basic Leucine Zipper domain) transcription factor. The three pathways are (i) inositol requiring kinase 1α (IRE1α) and X-box binding protein-1 (XBP1); (ii) protein kinase RNA-like ER kinase (PERK) and activating transcription factor 4 (ATF4); and (iii) activating transcription factor 6 (ATF6), which serves as both a sensor and transcriptional factor (1). The IRE1α–XBP1 branch is involved in lipid synthesis, ER-associated protein degradation (ERAD), protein folding, translocation to ER, and secretion (1). At the molecular level, IRE1α is a type I transmembrane protein consisting of a serine/threonine kinase domain and an RNase domain. In the presence of ER stress, IRE1α dimerizes and oligomerizes while stimulating trans-autophosphorylation, activating the RNase domain (3, 4). Activated IRE1α RNase excises a 26-nucleotide intron from the human XBP1 mRNA and causes a translational frame shift, generating the spliced and activated form of XBP1 (XBP1s; Fig. 1A; refs. 1, 5). In addition to its XBP1 splicing activity, activated IRE1α also preferentially degrades ER-associated mRNAs, a process known as regulated IRE1-dependent decay (RIDD; refs. 6, 7).

Studies in animal models have revealed that the UPR is implicated in various types of human cancer and targeting key components of the UPR has emerged as a promising therapeutic strategy (5). The IRE1α–XBP1 pathway plays an indispensable role in tumor growth, metastatic progression, and chemoresistance (8, 9). Expression and activation of XBP1 has been correlated with poor clinical outcome in breast cancer (10, 11) and angiogenesis in pancreatic cancer (12). As a mediator of survival, XBP1 has been extensively characterized in multiple myeloma, a plasma cell malignancy (13, 14). XBP1 is indispensable for plasma cell differentiation (15) and its expression is elevated in human multiple myeloma cells (16). XBP1 expression knockdown severely compromised RPMI 8226 multiple myeloma cell growth in vitro (17) and sensitized mouse myeloma cells for stress-induced apoptosis (18). Conversely, Eμ-myc–driven XBP1s expression promoted multiple myeloma pathogenesis in mice (14). Consistent with these results, XBP1s expression is associated with poor multiple myeloma patient survival (13, 19).

Several groups have identified small molecule inhibitors that selectively block IRE1α–XBP1 activation (5, 20), including those targeting the RNase domain [STF-083010 (21), salicylaldehydes (22), 4μ8C (23), MKC-3946 (17), and toyocamycin (24)] and directly interfering with ATP binding (“Compound 3”; ref. 25). Other members of the hydroxy–aryl–aldehydes (HAA) class, to which 4μ8C and MKC-3946 belong, were also extensively tested (26, 27). Given the complexity of the UPR, and the therapeutic potential of small molecule inhibitors targeting this pathway, combining conventional drug screening data with novel quantitative analysis may accelerate the drug development process.

To this end, we applied topological data analysis (TDA) to assist cell-based high-throughput screen (HTS). TDA is based on a branch of mathematics characterizing the geometric property of shapes with mathematical algorithms (28) and is a topologic framework of many machine-learning algorithms that quantify the shape of a large, but finite, number of data points. TDA has significant advantages over traditional hierarchical clustering in analyzing complex drug screening data. In contrast to hierarchical clustering that makes irreversible sequential choices about similarity and begins to lose utility when data grow in size, TDA maintains each data point individually throughout the process and performs extremely well in high-dimensional, large volume data and maintains its visual representations (29). Describing the results of an HTS is a multidimensional, complex signal challenge and TDA is particularly well suited to separate the signal from the background. To date, TDA has been applied to data analysis in cancer (29, 30), diabetes (31), neurological disorder (32), viral evolution (33), and the immune system (34).

Using an HTS combined with TDA, we identified a cluster of acridine analogues as specific inhibitors of the IRE1α–XBP1 axis that are cytotoxic to multiple myeloma cell lines. We also revealed a previously unidentified mechanism of action (MOA) for any existing inhibitor of IRE1α. Therefore, this study provides mechanistic insight of the biology of IRE1α activation and demonstrates the potential for targeting this MOA for therapeutic gain.

For the chemical library screen, 113,500 compounds from the ChemDiv, SPECS, and Chembridge libraries were screened at the Stanford University High-Throughput Bioscience Center (HTBC; ref. 21). The detailed specifications and criteria for compound selection can be found at the HTBC website (http://htbc.stanford.edu/compounds.html). Individual screening compounds for the confirmation assay were ordered from ChemDiv. Acridine derivatives were synthesized or obtained from the compound library at Auckland Cancer Society Research Centre at University of Auckland. These compounds are further characterized in Supplementary Material. Thapsigargin (Tg) was obtained from Sigma Aldrich.

HT1080 and HEK293 cells overexpressing IRE1α-GFP were maintained in DMEM. RPMI 8226 and MM1.R cells were maintained in RPMI1640 medium. All media were supplemented with 10% FCS and 1% penicillin–streptomycin and cells were cultured at 37°C with 5% CO₂. Cell lines were authenticated using short tandem repeat analysis at ATCC and used within 6 months' culturing. The cells were obtained in 2001 (HT1080), 2009 (RPMI8226 and MM1.R), and 2011 (HEK293).

Analysis was performed using Iris software (Ayasdi Inc.). The description of the implementation of TDA in the software is described in ref. 29.

Western blotting was performed according to standard protocols. Antibodies used include: anti-XBP1s (1:1,000; BioLegend), anti-XBP1 (1:1,000; Abcam), anti-IRE1α, anti-phospho-eIF2α, anti-phospho-eIF2 (1:1,000; Cell Signaling Technology) and anti-β-actin (1:1,000; Santa Cruz Biotechnology).

For cell viability assay, 2 × 10⁴ cells per well were plated into 96-well plates and treatment started 0 to 12 hours after plating. After 24 hours of treatment, XTT reagent (ATCC) was added to the wells. Then cells were incubated for 2 hours and absorbance was measured at both 475 and 660 nm using a BioTek Synergy H1 plate reader.

In vitro nuclease assays were performed as described in ref. 35. Reactions were performed in nuclease reaction buffer (40 mmol/L HEPES 7.0, 10 mmol/L Mg(OAc)₂, 50 mmol/L KOAc, 5 mmol/L DTT) at 30°C for the indicated amount of time, with 2 mmol/L ADP, 1 μmol/L recombinant human IRE1α (hIRE1α), and labeled RNA at 0.15 nmol/L (10 fmol/reaction). Upon incubation for 30 minutes, reactions were stopped and RNA was extracted using phenol/chloroform, ethanol precipitated, analyzed on a denaturing 6% urea acrylamide gel and visualized by autoradiography.

IRE1α oligomerization assay was performed as described previously (36). Briefly, reactions were set up using the same nuclease reaction buffer, 2 μmol/L recombinant hIRE1α, 2 mmol/L ADP, and 60 μmol/L of the indicated compound. All reactions were performed in 20 μL with 10% DMSO to account for vehicle and incubated for 15 minutes at 30°C to allow for oligomerization. The optical density of the sample was measured at 500 nm using a NanoDrop 2000 (Thermo Scientific).

To visualize IRE1α foci, HEK293 cells stably expressing doxycycline-inducible hIRE1α-GFP (37) were grown on a cover slip and IRE1α-GFP was induced with 10 nmol/L doxycycline 24 hours before start of treatments. Cells were then pretreated with compound for 1 hour before addition of 300 nmol/L Tg for an additional 2 hours. For immunofluorescence, primary anti-GFP antibody (1:1,000; Roche) and Alexa Fluor 568 secondary antibody (1:2,000; Life Technologies) were used.

Luciferase activity was measured noninvasively using the IVIS imaging system (PerkinElmer). Mice were injected intraperitoneally with 300 mg/kg luciferin (Biosynth) and anesthetized by 3% isoflurane (Butler-Schein). Exposure time was set as 1 second for ventral view. All images were analyzed using Living Image software version 4.2 (PerkinElmer). In vivo bioluminescent signal was quantified by taking the average photon count per second per square centimeter.

A total of 5 × 10⁶ RPMI 8226 cells were implanted subcutaneously into the flanks of 4 to 6 weeks' old NOD SCID mice (Charles River Laboratories). Drug treatment started when the sizes of the tumors reached approximately 150 mm³. Four tumor-bearing mice per group were treated with 10 mg/kg 3,6-DMAD or vehicle (saline) intraperitoneally once every 2 days. Tumor volume was measured as previously described (38).

To identify inhibitors of XBP1 activation, we conducted a luciferase reporter-based high-throughput chemical library screen. In this reporter construct, firefly luciferase is fused in frame with the human XBP1s mRNA sequence, resulting in luciferase expression only when XBP1 mRNA is spliced by activated IRE1α into the XBP1s form. This construct, as well as a CMV promoter-driven control construct to normalize for potential off-target effects, were stably introduced into the HT1080 human fibrosarcoma cell line (HT1080-XBP1-luc and HT1080-CMV-luc, respectively; Fig. 1A; ref. 39). Using the HT1080-XBP1-luc cell line, we performed the primary screen with 113,500 unique chemical compounds at a single concentration of 10 μmol/L (Fig. 1B). ER stress was induced by 300 nmol/L Tg, an ER Ca²⁺ ATPase inhibitor, and luciferase activity was measured 24 hours later (21). We then prioritized 990 compounds that displayed >40 %inhibition (compared with vehicle control) without nonspecific toxicity or nonspecific inhibition of luciferase activity for a secondary screen, in which all the compounds were assessed at eight concentrations ranging from 0.156 to 20 μmol/L in both types of reporter cell lines to determine dose-dependent activity (Fig. 1B).

To evaluate the results in a comprehensive manner, inhibition values of all 990 compounds were visualized in a heatmap generated by hierarchical clustering (Fig. 1C). The six subclusters represent different types of compound effects, including activation of XBP1-luc activity at higher doses (I), nondose-dependent inhibition (II), dose-dependent inhibition (III and V), and toxicity at higher doses (IV and VI). To evaluate more robustly the relationships between compounds and increase the accuracy of the hit selection, we analyzed the results of the secondary screen using TDA. An unbiased compound network was constructed by simultaneously considering the inhibition values across all the conditions for all the compounds (Fig. 1D). By quantitatively color-scaling the individual nodes (groups of compounds) in the network based on their inhibition values, we identified two regions that displayed selective inhibition of HT1080-XBP1-luc but not HT1080-CMV-luc activity, thus highlighting a set of compounds that specifically inhibit XBP1 splicing with minimal nonspecific effects. Through more detailed analysis of the dose-inhibition curves for individual compounds in these regions, we identified four compounds with optimal curve fitting as well as magnitude of inhibition (Fig. 1D and E).

Interestingly, a previously identified IRE1α inhibitor through the same chemical library screen, STF083010, was also clustered in one of these regions, which served as an internal positive control for the validity of the TDA approach (21, 23). More significantly, the other compounds highlighted in these clusters would not necessarily be singled out as promising hits based upon conventional curve fitting–based selection criteria. To confirm the luciferase reporter-based assay results, we performed Western blotting to analyze the effect of these compounds on endogenous XBP1 splicing in HT1080 cells induced by Tg. Of the top four candidate compounds, STF089106 displayed the most prominent inhibition on XBP1s protein expression without affecting expression of unspliced XBP1 (XBP1u; Fig. 1F, data not shown). Phospho-eIF2α levels were not attenuated by STF089106 treatment, suggesting that STF089106 specifically targeted the IRE1α–XBP1 branch instead of inhibiting the UPR in general. On the basis of these data, we selected STF089106 (1, N-acridine-9-yl-N',N'-dimethylpropane-1,3-diamine, DAPA; refs. 40, 41) for further analysis.

Acridines are DNA-binding agents and topoisomerase poisons (42) that have been previously identified as antitumor and antibacterial therapeutics (43). DAPA is an analogue of mAMSA, an anilinoacridine with activity against leukemia (44), and the acridine carboxamide (2, N-[2-(dimethylamino)ethyl]acridine-4-carboxamide, DACA). DACA has been tested in phase II trials against multiple types of cancers (45–48).

Upon identification of DAPA as a potential screening hit, we independently synthesized this molecule to confirm its activity (designated Compound 5) and designed a small set of analogues to explore the structure–activity relationships (SAR) of this compound (Fig. 2). 9-Aminoacridines bearing basic amine side chains have been characterized as DNA-affinic cytotoxins (49) and we sought to determine whether the SAR for inhibiting XBP1 splicing is mapped to that required for DNA binding. DAPA contains a DNA intercalating 9-aminoacridine chromophore and we explored variations in the chromophore with analogues (Compound 5, 6, 10, and 11). Variations in side chain amine pK_a (acid dissociation constant) were explored with analogues Compound 5, 7, 8, and 9. We further analyzed the role of the side chain using a series of neutral analogues of varying lipophilicity with analogues Compound 1 to 3 (Fig. 2A).

We then compared the effect of these analogues (5 to 30 μmol/L) on endogenous UPR marker proteins in HT1080 cells treated with Tg. All compounds containing the entire acridine chromophore demonstrated inhibition of XBP1 splicing, whereas saturation (Compound 11) or removal of one ring (Compound 6) resulted in loss of inhibition, suggesting that an acridine chromophore was important for inhibiting XBP1s. In contrast, phospho-eIF2α levels remained relatively unchanged, indicating that these compounds could preferentially block IRE1α–XBP1 with minimal impact on the other branches of UPR.

Next, we expanded our analysis of the more potent analogues (Compound 1, 3 to 5, 7 to 10) at lower concentrations (0.5 to 3 μmol/L; Fig. 2B). Both 9-aminoacridine (Compound 5) and acridine-4-carboxamide (Compound 4) displayed XBP1s inhibitory activity with the 9-aminoacridine demonstrating the greatest XBP1s inhibition at 1 μmol/L. Importantly, the 3,6-dimethylamino substitution on the acridine ring increased activity with Compound 10 showing XBP1s inhibition at as low as 0.5 μmol/L. These SAR data indicate that the acridine ring is critical for the activity of the drug to inhibit XBP1s. Variation in the basicity of the amine side chain was tolerated, although it appears that less basic amines, for example, Compound 9, were weaker inhibitors. A range of neutral substituents were also tolerated, albeit with reduced potency compared to Compound 5 (Fig. 2B). This targeted SAR dataset validated the activity of DAPA as an inhibitor of XBP1 splicing and confirmed the utility of 9-aminoacridines as the basis for future drug development.

We explored the mechanism of inhibition of IRE1α–XBP1 by analyzing a selection of representative analogues, including Compound 10 (most potent), 5 (highly potent), 4 (limited potency), and 11 (inactive). To test whether inhibition occurs by directly disrupting the IRE1α–XBP1 interaction, we performed an in vitro IRE1α RNase assay with recombinant human IRE1α and ³²P-labeled human XBP1 mRNA substrate (Fig. 3, top). Although Compound 10 blocked in vitro XBP1 splicing almost completely at 10 μmol/L and Compound 11 was inactive as expected, Compounds 4 and 5 did not show activity in this assay, even at concentrations up to 500 μmol/L. Furthermore, we examined human BLOS1 cleavage as an assay for RIDD. We determined that Compound 10 inhibited RIDD with similar efficacy as XBP1 splicing (Fig. 3, bottom), suggesting that 3,6-dimethylamino substitution of the acridine ring is critical for global inhibition of IRE1α endonuclease activity. However, as both Compounds 4 and 5 displayed inhibitory activity against XBP1s in the cell-based assay, there could be an alternative mechanism by which these compounds interfere with IRE1α activity in intact cells that is not evident in the cell-free system.

Next, we determined the effect of these compounds on IRE1α dimerization and oligomerization, necessary for activation of its RNase activity. Initially, we performed an in vitro IRE1α oligomerization assay, in which addition of ADP to purified IRE1α stimulates its oligomerization (Fig. 4A; ref. 36). Although treatment with each compound alone did not affect the optical readings, Compounds 4, 5, and 10 significantly attenuated ADP-induced IRE1α oligomerization. Interestingly, STF083010 did not show any activity in this assay, highlighting the unique mechanistic specificity of the acridine analogues on IRE1α oligomerization. The lack of inhibitory activity for Compound 11 demonstrates that an acridine ring is essential for inhibiting IRE1α activity by disrupting its oligomerization. In summary, the basic acridine ring in addition to an aminoalkyl side chain at the 4- or 9-position was sufficient for disrupting IRE1α oligomerization, but not adequate for inhibiting the established RNase activity. In contrast, the 3,6-dimethylamino substitution enabled both disruption of IRE1α oligomerization and inhibition of XBP1 splicing activity.

To verify the results from the in vitro oligomerization assay in intact cells, we utilized HEK293 cells overexpressing GFP-tagged IRE1α in which IRE1α-GFP foci formation can be detected through fluorescence imaging after Tg treatment (Fig. 4B; ref. 37). We used immunofluorescence staining to detect IRE1α-GFP foci as higher concentrations of Compound 10 caused autofluorescence (Fig. 4C). We found that higher than 1 μmol/L of Compound 10 completely inhibited Tg-induced foci formation in living cells. These data are consistent with the in vitro oligomerization assay results and indicate that Compound 10 blocks XBP1 splicing by both inhibiting IRE1α RNase activity directly and disrupting IRE1α oligomerization.

We performed XTT viability assays on both RPMI 8226 and MM1.R (dexamethasone-resistant) human multiple myeloma cells treated with selected analogues to characterize drug cytotoxicity. Consistent with their high potency in inhibiting XBP1s, Compounds 5 and 10 displayed significant cytotoxicity in these cells within the concentration range (0 to 6 μmol/L; Fig. 5A) and the weaker Compound 4 demonstrated only limited cytotoxicity (Fig. 5A). As a negative control, Compound 11, which was inactive in blocking XBP1s, failed to show any cytotoxicity on the multiple myeloma cells (Fig. 5A).

We also assessed the effect of these compounds on a panel of UPR marker proteins in multiple myeloma cells treated between 1 and 3 μmol/L and found a consistent correlation between the cytotoxicity of these compounds with the degree of inhibition of XBP1s (Fig. 5B). In summary, these findings demonstrate that 9-aminoacridines, particularly Compound 10, display cytotoxicity in multiple myeloma cells in vitro, which correlates with inhibition of the IRE1α–XBP1 pathway.

On the basis of the cell-based and cell-free assay results, we chose to evaluate the activity of 3,6-DMAD (Compound 10) in vivo. We utilized transgenic mice constitutively expressing the XBP1-luciferase reporter, which expresses luciferase activity under basal and inducible ER stress in vivo (39). After three intraperitoneal administrations of 3,6-DMAD at a dose of 10 mg/kg every 12 hours, this compound significantly inhibited in vivo luciferase activity assessed 3.5 days after the initial treatment (Fig. 6A), whereas treatment with vehicle did not cause any significant change (Fig. 6B).

To assess the effect of 3,6-DMAD on multiple myeloma tumor growth in vivo, we treated NOD SCID mice transplanted with RPMI 8226 cells subcutaneously with 10 mg/kg of 3,6-DMAD administered intraperitoneally every 48 hours. Compared with vehicle group, 3,6-DMAD-treatment significantly inhibited tumor xenograft growth (Fig. 6C and D).

We previously performed a luciferase reporter–based HTS targeting the IRE1α–XBP1 pathway of the UPR. Using TDA on the HTS dataset, we identified DAPA as a potent inhibitor of XBP1 splicing. We synthesized a set of DAPA analogues to explore the SAR and conclude that the acridine chromophore is critical for inhibition of XBP1 splicing. Furthermore, a 3,6-dimethylamino substitution on the chromophore, 3,6-DMAD, showed the most potent and comprehensive IRE1α inhibitory activity in all cell-based and cell-free assays. These compounds also revealed a novel MOA through its disruption of IRE1α oligomerization. Finally, 3,6-DMAD demonstrated significant cytotoxicity against multiple myeloma cell lines, blocked XBP1 splicing in XBP1-luc transgenic reporter mice, and inhibited multiple myeloma tumor xenograft growth.

The complexity of the data obtained from HTS results in large signals dominating the analysis whereas subtle and possibly more potent hits may remain undetected. In this study, we have shown that topological methods may be superior in detecting subtle but real signals from HTS data (29, 30, 33). Based upon a conventional analysis in which compounds are ranked by %inhibition of XBP1-luciferase signal, the acridine compounds were not obvious candidate for further investigation (21). The methodology from this study utilizes a network-based organization of the data so that the relatedness of promising lead compounds to other compounds in the chemical library is readily available and this information can be used for rapidly developing SAR data, resulting in a more efficient drug discovery process.

Our group first reported the feasibility of targeting IRE1α with a small molecule inhibitor, STF-083010 (21). This compound inhibited in vitro RNase activity of IRE1α, displayed cytotoxicity against multiple myeloma cell lines, and delayed RPMI 8226 xenograft growth. The mechanism of inhibition of STF083010 was further elucidated by a later study (23), showing that both STF-083010 and 4μ8C, a novel IRE1α RNase inhibitor, binds to K907 in the RNase domain of hIRE1α. Another XBP1s inhibitor, MKC-3946, delayed RPMI 8226 xenograft growth following a 100 mg/kg daily dose (17). Its MOA may be similar to STF083010 and 4μ8C, as MKC-3946 has the same HAA motif common to the released forms of STF083010 and 4μ8C (26). In addition, toyocamycin, an adenosine analogue, inhibits XBP1s and significantly suppressed RPMI 8226 xenograft growth with unknown MOA (24).

The mechanism of IRE1α activation during ER stress was extensively characterized. Accumulation of unfolded and misfolded proteins in the ER stimulates IRE1α trans-autophosphorylation and dimerization of its kinase and RNase domain to form the active catalytic site (3). Further analysis revealed that upon activation, IRE1α assembles into high-order oligomers, which is essential to its function (4, 37). This process was successfully visualized by detecting GFP-tagged IRE1α foci formation through live cell imaging (37). Our in vitro and cell-based assays revealed that a set of acridine derivatives block XBP1 splicing with variable potency through interfering with IRE1α oligomerization as a novel MOA. 3,6-DMAD, the most potent of this class of drugs, inhibited both oligomerization and in vitro RNase activity of IRE1α, suggesting that substitution of the acridine chromophore induces additional activity targeting the RNase activity. Further SAR analysis to clarify the critical structural features of this chemical prototype for inhibiting XBP1 splicing will be the subject of future studies.

In conclusion, by utilizing TDA analysis on HTS, we identified a class of acridine derivatives as potent and specific inhibitors of the IRE1α–XBP1 branch of UPR with prominent cytotoxicity on multiple myeloma cells as well as in vivo multiple myeloma tumor growth. Therefore, our findings have defined a class of compounds with a novel MOA and provide a strong preclinical rationale for further development of this class of compounds as a therapeutic strategy for multiple myeloma patients.

M. Alagappan is a consultant/advisory board member for Ayasdi Inc. P.Y. Lum has ownership interest in Ayasdi Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: D. Jiang, N. Denko, A.J. Giaccia, Q.-T. Le, M. Niwa, A.C. Koong

Development of methodology: D. Jiang, A. Tam, M. Alagappan, A. Gupta, M. Niwa, A.C. Koong

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Jiang, A. Tam, M.P. Hay, M. Kozak, D.E. Solow-Cordero, M. Niwa

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Jiang, A. Tam, M. Alagappan, D.E. Solow-Cordero, P.Y. Lum, N. Denko, A.J. Giaccia, Q.-T. Le, M. Niwa, A.C. Koong

Writing, review, and/or revision of the manuscript: D. Jiang, M. Alagappan, M.P. Hay, P.Y. Lum, N. Denko, A.J. Giaccia, Q.-T. Le, M. Niwa, A.C. Koong

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Jiang, D.E. Solow-Cordero, A.C. Koong

Study supervision: M. Niwa, A.C. Koong

This work was supported by grants from the NIH (P01CA067166; to Q.T. Le, N. Denko, A.J. Giaccia, A.C. Koong).

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.