Ingenuity Network-Assisted Transcription Profiling: Identification of a New Pharmacologic Mechanism for MK886

The small molecular inhibitor MK886 is known to block 5-lipoxygenase-activating protein ALOX5AP and shows antitumor activity in multiple human cell lines. The broad antitumor therapeutic window reported in vivo for MK886 in rodents supports further consideration of this structural class. Better understanding of the mode of action of the drug is important for application in humans to take place. Affymetrix microarray study was conducted to explore MK886 pharmacologic mechanism. Ingenuity Pathway Analysis software was applied to validate the results at the transcriptional level by putting them in the context of an experimental proteomic network. Genes most affected by MK886 included actin B and focal adhesion components. A subsequent National Cancer Institute-60 panel study, RT-PCR validation followed by confocal microscopy, and Western blotting also pointed to actin B down-regulation, filamentous actin loss, and disorganization of the transcription machinery. In agreement with these observations, MK886 was found to enhance the effect of UV radiation in H720 lung cancer cell line. In light of the modification of cytoskeleton and cell motility by lipid phosphoinositide 3-kinase products, MK886 interaction with actin B might be biologically important. The low toxicity of MK886 in vivo was modeled and explained by binding and transport by dietary lipids. The rate of lipid absorbance is generally higher for tumors, suggesting a promise of a targeted liposome-based delivery system for this drug. These results suggest a novel antitumor pharmacologic mechanism.

5-Lipoxygenase is the upstream enzyme in the leukotriene pathway⁷

regulating chemotaxis (1), blood coagulation (2), inflammation (3), pain sensation (4), and blood pressure (5). It is also known to stimulate cancer growth (6–8). Interruption of the 5-lipoxygenase pathway is shown to exert a marked anticancer effect in a number of reports (9, 10). After activation by external stimuli, 5-lipoxygenase localizes on the inner leaflet of the nuclear membrane (11) depending on 5-lipoxygenase-activating protein ALOX5AP (FLAP) located in the same membrane. The drug compound MK886 (Biomol Research, Plymouth Meeting, PA) is known to bind FLAP at 100 nmol/L and was used in the pioneering work of FLAP isolation (12). In knockout mice studies, the ablation of FLAP led to the development of an atherosclerosis-resistant strain of mice (13). The role of the pathway in the inflammatory response leading to cardiovascular disease is being rapidly explored (14, 15).

In preclinical studies, the drug is well tolerated in rodents at high dosages of ∼800 μmol/L/kg (640 mg/kg; ref. 16). The drug level sufficient to prevent cancer in vivo is 10 mg/kg in rodents (17). A 64-fold difference between the in vivo tolerated and cancer-inhibiting dosages makes this structural class worthy of further exploration, with the goal of developing a human version of the treatment.

MK886 has been studied using small-scale experiments, and no consensus as to its prevailing mechanism of action has been reached. MK886 has been reported to have an antitumor effect caused by the mitochondrial permeability transition pore opening (18). The synergy between γ-linolenic acid effect and MK886 action against a refractory pancreatic cancer line PANC1 is described by Harris et al. (19). Protection against MK886-induced apoptosis is shown by the exogenous addition of a 5-lipoxygenase product, 5-hydroxyeicosatetraenoic acid and thiols (20). The fatty acid oxidation via alternative pathways is proposed as a cause of cell death (21). We have reported the relative redistribution of bioactive peroxide products between the 5- and 15-lipoxygenase pathways as well as the involvement of peroxisome proliferator-activated receptor γ (PPARγ) as the likely cause of MK886-mediated inhibitory effects (22). Another study suggests that MK886 drug effect seems to be independent of the presence of its target FLAP (23). These varied findings point to a complex MK886 mechanism.

To address the complexity of the proposed MK886 mechanism, we used a novel combinatorial research strategy. We used the data from expression profiling, reduced first by t test, to select for significantly affected genes and then reduced the output by an automatic mechanism-proposing software Ingenuity Pathway Analysis (Mountain View, CA). The National Cancer Institute (NCI)-60 type of information-rich data uses the similarity of cell growth inhibition profiles across multiple cell lines as the measure of the similarity between the compared drug mechanisms (24). The coclustering profiles are assumed to represent similar mechanism of action and we applied this tool as well.

By using microarrays followed by several layers of validating tools, we attempted to identify additional biological target(s) of MK886, thus suggesting a process to refine a lead compound for relevant pharmacologic applications. The lipid-processing pathways are prominent in the development of cancer and its therapy. The attempts to affect these pathways pharmacologically may involve side effects based on binding to unintended sites, and defining the actual antitumor mechanism can be helpful in the efforts to understand potential complications. The results of our current study points to a possible novel cytotoxic anticancer mechanism.

Cell lines. Cell lines [A549, non–small cell bronchoepithelial carcinoma, NCI-H720, atypical carcinoid, H157, Mus musculus (myeloma), H1299, non–small cell lung carcinoma], were procured from American Type Culture Collection (ATCC, Manassas, VA). NCI-N417 small cell lung carcinoma was derived from NCI. The A549, H157, N417, H1299, and H720 cells were grown in RPMI 1640 (Invitrogen, Carlsbad, CA), supplemented with 5% FCS, 2 mmol/L glutamine and penicillin/streptomycin (Invitrogen). Human osteosarcoma cells (143.98.2; ATCC) was grown in DMEM supplemented with 10% fetal bovine serum (FBS), penicillin (100 units/mL), and streptomycin (100 μg/mL). The malignant mouse astrocytoma cell line CT-2A was a generous gift from Prof. T.N. Seyfried (Boston College, Chestnut Hill, MA). Human umbilical vascular endothelial cells (HUVEC) and EBM2 medium supplemented with “bullet kit” were purchased at Cambrex (East Rutherford, NJ) and HUVECs were grown according to the supplied protocol. The immortalized mouse embryo fibroblast cell line MEF8 was obtained by transforming primary fibroblasts isolated from 12-day-old embryos with the SV40 major T antigen. MEF8 and CT2A cell lines were grown in a DMEM supplemented with 5% FCS (Invitrogen) at 37°C with 5% CO₂.

Real-time PCR. RNA isolation and quantification and quantitative real-time PCR (RT-PCR) were conducted as described in ref. (22). The gene sequences and the primer sequences (with melting temperatures) used for RT-PCR are given in the supplementary file “RT-PCR primers.” Normalization was conducted using the average of the panel of nine random genes, listed in the Supplementary Table S1.

Microarray data generation. Data sets in this report were generated by treating A549 and H720 cell line with MK886 (Calbiochem, San Diego, CA) at 1 μmol/L for 24 hours. Six Affymetrix Human Genome U133 Plus 2.0 microarray chips (Affymetrix, Santa Clara, CA) were used in triplicate for the treated and untreated classes according to the protocol of the manufacturer available online.⁸

Briefly, total RNA was isolated from the cancer cell lines using RNeasy mini-kit (Qiagen, Chatsworth, CA). The quality of the RNA isolation was assessed by A₂₆₀/A₂₈₀ (>1.9, <2.1) ratio as well as by 1% agarose gel electrophoresis (intact r-RNA bands). The first-strand cDNA, the double-strand cDNA, and cRNA were synthesized, and cRNA was fragmented using the protocol-recommended kits. All intermediate and final products were tested on agarose gel to comply with the molecular mass distribution as described in the protocol. The cRNA was mixed with internal controls for hybridization and B2 oligonucleotide for automatic grid alignment, the chip was prehybridized with 1× hybridization buffer for 10 minutes at 45°C and then hybridized with the cRNA cocktail for 16 hours at 60 rpm in a rotating exposure chamber. Upon completion, the arrays were washed, stained with streptavidin phycoerythrin, and scanned. The data were submitted to Gene Expression Omnibus public depository, according to the minimum information about a microarray experiment checklist, series entry GSE3202.

Microarray data analysis. Microarray probe data were obtained with Microarray Suit 5.0 (Mas 5.0; Affymetrix, Santa Clara, CA). The robust multiarray average method and quantile normalization were used to produce normalized probe set summary measures. The calculation was implemented in the R package Affy of the Bioconductor software project.⁹

Identification of probe sets that were differentially expressed between the treatment and control group was done using two-sided univariate t tests. Specifically, differentially expressed genes were identified as those genes that were significant at the 0.001 level (P < 0.001) and that were at least 1.5-fold different in the mean expressions.

Pathway identification. The differentially expressed probe sets were overlaid on a cellular pathway map in the Ingenuity Pathway Analysis using resource database Knowledge Base (Winter 04 Release containing 20,000 genes). The resulting networks were represented in table and graphic format.

Anoikis treatment. Poly-hydroxyethyl methacrylate (poly-HEMA, Sigma, Saint Louis, MO) was solubilized in methanol (50 mg/mL) and diluted in ethanol to a final concentration of 10 mg/mL. To prepare poly-HEMA-coated dishes, 4 mL poly-HEMA solution were placed onto 100 mm dishes and dried in a tissue culture hood. The poly-HEMA coating was repeated twice, followed by three washes with PBS. One million five hundred thousand trypsinized A549 cells were plated onto poly-HEMA-coated dishes for 12 hours in RPMI with 5% FBS. The cells were resuspended after 12 hours and RNA extracted for RT-PCR analysis.

Filamentous actin labeling for flow cytometry and confocal microscopy study. The cells were trypsinized, washed in PBS and fixed in 4% immunohistochemistry grade formalin for 10 minutes at room temperature. Cells were permeabilized with 0.5% Triton X-100 for 5 minutes at room temperature and washed with PBS. The rhodamine phalloidin conjugate (Cytoskeleton, Inc., Denver, CO) was diluted in methanol and used according to the protocol of the manufacturer. For confocal microscopy, cells were seeded onto glass slides and exposed to a regular medium (control) or to 10 μmol/L MK886 for 20 hours. After treatment, cells were fixed in 4% paraformaldehyde for 10 minutes, permeabilized in 1% Triton X-100 in PBS for 10 minutes, and exposed to Bodipy-phallacidin (Molecular Probes, Eugene, OR) at a dilution 1:200 in PBS for 1 hour. Slides were observed with a confocal microscope (Leitz DM IRB, Berlin, Germany).

Spin-down assay. Rabbit skeletal muscle actin and pyrene muscle actin were procured from Cytoskeleton. The actin solution (0.25 mg/mL) was redistributed into the Eppendorf tubes, 120 μL in each, mixed with MK886 and other drugs, and left to incubate on ice for 15 minutes. After incubation, the 20× polymerization buffer (KMEI), prepared according to the protocol of the manufacturer, was added to initiate polymerization. After 15 minutes, the tubes were transferred to a cooled table top centrifuge and spun at 14,000 rpm (25,000 × g). The resulting molecular weight gradient of different filamentous actin (F-actin) polymerized products was sampled (10 μL of total 120 μL) from the top and the protein concentration analyzed by the Lowry assay (25) in triplicates using a 1 mL cuvette.

Viscosity shift assay. Cannon-Manning Semi-Micro Viscosimeter (Cannon Instrument, Co., State College, PA) was used to assess the viscosity shift upon actin polymerization in the presence and in the absence of the compounds of interest. Briefly, the time interval elapsing while the fixed volume of a liquid passes between the upper and lower marks is directly proportional to viscosity (in quadruplicates).

Pyrene actin fluorescence shift. The pyrene-labeled actin (Cytoskeleton) was mixed at a ratio of 1:5 with unlabeled actin, spun at 10,000 × g to remove polymerized actin, and polymerized in the cuvette of a fluorimeter (ISS PC1 spectrofluorometer, Champaign, IL) according to the protocol of the manufacturer. The time course of the reaction was monitored at the excitation wavelength 360 nm and emission at 400 nm.

Combined MK886 and UV radiation exposure. The A549 and H720 cell lines were seeded in the Costar 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) plates at the density of 2 × 10⁴ cells per well (H720 that forms clumps was seeded based on total protein, equal to the A549 protein content). The cells were exposed to 0.75 μmol/L MK886 serum-free for 24 hours. After the exposure to MK886, the cells were irradiated by UV at 302 nm in Multilimage Light Cabinet (Imgen Technologies, Alexandria, VA) at 960 Wt/m² irradiation density for the indicated periods of time. The formal analysis of synergy was conducted as in ref. (26).

Bioavailability studies. Liposin II (10%) is a soluble, nonpyrogenic triglyceride emulsion supplied as a 10% solution containing 5% safflower oil and 5% soybean oil (neutral glycolipids), 2.5% glycerin in H₂O for injections, and 1.2% egg phosphatides as an emulsifier (Abbott Laboratories, Chicago, IL). Liposin II was added to the medium at 0.01% and cell culture survival was assessed by MTT. The experiment was conducted in triplicate.

Supplementary Data. Supplementary Data are available on the AACR website under the name “MK886 data” and contain the supporting information describing microarray data, RT-PCR primers and results, and NCI-60 panel profiling.

Defining the optimal exposure. To study the MK886 mechanism at a level of exposure that minimizes nonspecific effects, we evaluated a range of drug concentrations and intervals of exposure. The optimal time point likely took place at 1 mmol/L MK886 at 24 hours in serum-free conditions, because the effect of the drug was still reversible, as opposed to longer exposures, and enabled us to focus on relatively specific upstream events (see Supplementary Data)

Microarray and pathway analysis results. Six Affymetrix U133 Plus 2.0 human arrays were hybridized, scanned, and 25 probe sets were identified as differentially expressed genes between MK886 and control at the 0.001 level and with at least 1.5-fold mean difference in H720 cells. Overall, 97 probes showed the P values <0.001. Weakness of microarray result called for additional ways to ensure validity of the biological information extracted in these experiments. The differentially expressed probes were subjected to verification by RT-PCR and yielded 21 probes showing the change in the same direction corresponding to false detection rate of 16% (Supplementary Table S2). The results of Gene Ontology analysis are given in Table 1, and the results of the network reconstruction by Ingenuity in Fig. 1 (see also Supplementary Figure S2, annotation to the network).

Participation of genes in Ingenuity clusters further decreased the P value of detection due to lower probability of random detection if the responses formed proteomic connections. Even weak responses, but relating to confirmed protein-protein interactions, thus become more meaningful. The Gene Ontology analysis inquired into the functions of all significantly affected genes, regardless of their mutual relationships. Both analyses (by Ontology and by Ingenuity) were concordant in the finding that DNA and RNA binding, recombination, and repair was the dominant category among the known affected functions. Assuming that the choice of dosage and time point led to detection of relatively upstream events, the study focused first on the highest Ingenuity score cluster containing the most statistically robust candidates for hypothesis building. Actin B was among the genes constituting network 1 (Supplementary Table S1) and was significantly up-regulated in this treatment (3- to 4-fold) at the level of transcription. Among the other fundamental cytoskeleton components, villin 2¹⁰

is thought to be the interface component between cytoskeleton, plasma membrane, and focal adhesion kinase–associated cytoskeleton components.

The Ingenuity Pathway analysis¹¹

suggested putative downstream targets responding to the potential cytoskeleton disruption by MK886. The binding partners of actin B that responded to MK886 treatment included SWI/SNF-related matrix-associated actin-dependent regulator of chromatin a4 (SMARCA4¹²). According to the connectivity diagram in Fig. 1, SMARCA4 interacts with E2F transcription factor 1, playing multiple crucial roles in DNA-damage response, apoptosis, and the cell cycle.¹³ The up-regulation of SMARCA4 observed in these experiments may be related to E2F1-mediated down-regulation of its nearest neighbor partners, defined by the Ingenuity-derived connections (Fig. 1), including a number of vital DNA processing proteins. The link was supported by finding of another E2F transcription factor family tumor-suppressor member E2F transcription factor 4 in the network 2 (Supplementary Table S1).

As shown in Fig. 1, these downstream partner genes included mutS2 homologue (MSH2) gene involved in mismatch repair. Another partner, Asp-Glu-Ala-Asp/His box polypeptide 11 (DDX11), is a member of the DEAD box protein family of RNA helicases.¹⁴

Still, another SMARCA4 down-regulated “client” was high-mobility group box 1, implicated in chromatin structural modulation.¹⁵ Other downstream SMARCA4 targets included suppressor of RNA polymerase B homologue SURB7.¹⁶ All these findings were pointing to a novel link between cytoskeleton modulation, chromatin remodeling, and regulation of expression.

NCI-60 panel analysis. The microarray results were validated by the NCI-60 panel of activity profiles (24).¹⁷

The primary biological activity data for MK886 (code number S736463) and the results of the analysis are in the corresponding Supplementary Data. The data modestly pointed to coclustering of MK886 with genotoxic agents despite chemical stability of MK886 structure. This finding corroborated the results suggested by Ingenuity clustering.

RT-PCR validation of microarray data. The gene identities, their network affiliations, and the values of differential expression by microarray and RT-PCR are presented in the Supplementary Table S2. The similarity of transcription profiles was previously shown to indicate the similarity of the mechanism of action for the compared agents (27) and thus gave the rationale to our approach. Figure 2A presents a comparison of H720-MK886-treated transcription profiles generated, respectively, by RT-PCR and microarray data, indicating similarity for the entire profile. Comparison of anoikis-treated A549 and MK886-treated H720 transcription profiles indicated an even greater similarity than in the above case (Fig. 2A) and suggesting that MK886 mechanism is also related to cytoskeleton alteration because its signature showed similarity to transcription profiling signature of anoikis.

Figure 2B shows RT-PCR validation of the prediction based on the sum of high-throughput methods. The changes in the transcript levels of DNA processing enzymes TOPO2A and TBP were expected, in accordance with the cytoskeleton-transcription linkage hypothesis, and were observed. The transcript levels were affected 3- to 4-fold and in paradoxically opposite directions, being increased in A549 and decreased in H720 cell lines. The 20-fold increase in DDX11 transcript level in H720 and 7-fold increase of the same transcript in A549 treated by 100 nmol/L Taxol provides more evidence of the link between the cytoskeleton and DNA/RNA processing machinery.

Involvement of TBP in the studied mechanism required the use of random gene panel for RT-PCR normalization. Although cumbersome, this approach allowed to avoid the bias caused by participation of the designated housekeeping genes in the studied mechanism.

Validation of microarray results at protein level. MK886 treatment of A549 caused a dramatic decrease in F-actin level in a certain fraction of cell populations, detected by confocal microscopy. Interestingly (Fig. 3), immortalized but nonmalignant murine cells (MEF-8) were more resistant to the drug compared with the purely malignant (CT-2A). Western blotting also confirmed actin down-regulation in multiple cell lines (in Supplementary Data).

Flow cytometry. The flow cytometry measurement of the extent of F-actin depolymerization suggested that at 1 μmol/L MK886 and 15 hours postexposure, the depolymerization already developed, whereas proteolytic activation and cell inhibition were still reversible (in Supplementary Data). This result also suggested the importance of cytoskeleton alteration.

MK886 affects actin polymerization in vitro. To further evaluate the possibility of cytoskeleton involvement, we explored if actin directly interacts with MK886. Figure 4A presents the results of the spin-down assay in the presence of several pharmacologic agents. It can be noted that all tested agents affected the residual concentration of actin monomer after polymerization. The biggest effect was achieved in the cytocholasin D control (Alexis, San Diego, CA) that is used normally as actin destabilization control. MK886 showed a significant effect as well. The effects of ciglitazone and AA861 [2,3,5-trimethyl-6(12-hydroxy-5,10-dodecadynyl)-1,4-benzochinone] from Biomol (Plymouth Meeting, PA) were modest but detectable. Figure 4B presents the results of a time course following pyrene-actin fluorescence enhancement. In this experiment, no significant effect of MK886 was determined at 2 and 6 μmol/L concentrations. However, at a higher level (18 μmol/L), MK886 showed interference, suggesting an effect on the fraction of the buried pyrene groups. In a control experiment, the actin and KMEI buffer were diluted to the same extent that was expected after an 18 μmol/L MK886 addition. The difference observed between this control and the actual addition of 18 μmol/L MK886 was significant. Figure 4C presents the shifts in viscosity after polymerization buffer was added to actin preincubated with drugs or with a solvent control. It is known that shifting viscosity indicates the change in the extent of actin polymerization and formation of smaller aggregates in vitro following the interaction with MK886. The strongest effect upon the viscosity increment was observed with the cytocholasin D. However, as in the spin-down assay, the presence of cytocholasin D was unable to completely block the polymerization. The cytocholasin D effect was followed by the effects of ciglitazon, MK886, and AA861.

Combined MK886 and UV radiation exposure. The MK886-induced down-regulation of the components of UV damage repair machinery prompted us to test its combination with UV in search of possible therapeutic applications, e.g., sensitization for radiotherapy. Figure 5 shows the UV dose responses in the presence and the absence of a relatively small MK886 dosage, exerting a small effect by itself. Our data indicate a synergistic effect for H720 cell line (FICI = 0.25). However, the effect for A549 was only slightly superadditive according to the formal analysis (FICI = 0.95). This experiment suggested that for some cell lines, cytoskeleton alteration may enhance DNA damage.

Bioavailability studies. What makes MK886 cancer selective? Bioavailability of MK886 was modeled in vitro by creating a system containing an emulsified lipid mixture (liposin II) and serum components, mimicking the conditions in the blood stream. Figure 6 presents the survival data for A549, suggesting a very significant protection against MK886 in the combined presence of serum albumin and lipid emulsion, exceeding the protection shown by the single components.

In our previous works (22, 26), we showed biochemically that MK886 causes nuclear translocation of PPARγ. This transcription factor is thought to be a tumor suppressor, and this translocation occurs within the first 2 hours postexposure. As a result, a potential tumor suppressor might accumulate in the nucleus, initiating the first step of an apoptotic cascade. In addition, according to Fig. 1, MK886 may act via cytoskeleton-induced alteration of chromatin structure leading to repression of vital pathways that maintain DNA integrity and enable replication. Another possibility stems from the known affinity purification technique for actin B, using immobilized DNaseI as a bait (28). The sequestering of the potentially dangerous enzyme depends on the state of actin polymerization and proteolytic activation, converting DNaseI to its catalytically active form, able to cleave chromatin. This indirect genotoxicity is potentially supported by NCI-60 profiling data. Similar report connecting cytoskeleton and genotoxicity exists for ocadaic acid, a marine-derived product of lipid nature (29). On the other hand, actin depolymerization by cytocholasin D does not lead to the same consequences. This outcome indicates that cytoskeleton alteration has to be a corroborating factor in a more complex mechanism.

A broader picture in literature points to the role of cytoskeleton alteration in the synergistic interactions with the agents directly or indirectly targeting DNA. In our study, MK886 have shown a synergistic interaction with UV damage in H720 cell line. The comparison with A549 is difficult due to a very different initial sensitivity of these cell lines to UV damage. Even greater synergistic effects are described for cytocholasin D combined with actinomycin D or etoposide or mitomycin C (30). MK886 is not unique in this aspect and our study partially explains these effects.

It seems that a broad array of sites being more sensitive in cancer than in normal cells is targeted in MK886 mechanism of action. The products of 5-lipoxygenase inhibited by MK886 are known stimulators of cancer growth by autocrine and paracrine pathways. The loss of this stimulation is relatively disadvantageous for cancer (9, 10). The close structural analogue of FLAP is microsomal glutathione transferase and MK886 is likely to suppress the protective glutathione conjugation in the cancer cells that show the increased free radical production and dependence on such a protection (20). The MK886 effect on mitochondria is also cancer specific due to a lower level of pH that destabilizes mitochondria (31). It is to this list of damaging factors that cytoskeleton alteration is added in case of MK886. In such a context, the additional component makes the drug effect self-synergistic, possibly explaining a cooperative dose response curve of MK886 (serum-free; Fig. 6). By contrast, a low dose of cytocholasin D affecting cytoskeleton alone, in the absence of corroborative factors, may not suffice to trigger the massive apoptosis observed for MK886.

To understand why the rodents display such a tolerance to a drug with such a broad mechanism, we conducted a bioavailability study modeling the major factors influencing MK886 biodistribution. Figure 6 shows a very significant protection of A549 against MK886 in the presence of both serum albumin and emulsified fat, simulating a natural concentration buffer. Absorption of the emulsified lipids together with bound MK886 may lead to the conditions when tumor is at a selective disadvantage. Multiple studies suggest increased lipid absorption by tumors compared with normal cells (32, 33). This difference might arise due to a more permeable vasculature in tumor sites, overexpression of scavenger receptors in cancer, more intense phagocytosis in tumor cells, and greater reliance on anaerobic sources of energy. As a result, the influx of the drug in cancer may exceed the influx in normal cells, selectively affecting the latter under equal other conditions. The novel experimental drug delivery systems use specifically designed liposomes for that purpose (34). In case of MK886, its hydrophobic character might enable its tumor-specific delivery via a targeted liposome delivery system. Figure 6 shows a steep dose response to MK886, indicating that even a small relative dose accumulation in a tumor site may cause a disproportional therapeutic effect. We find that these features of MK886 mechanism and biodistribution recommend it for future exploration in murine models of cancer.

The known fact of cytoskeleton modification by phosphoinositide-3-kinase products (35) points to a biologically intended role of lipid binding to actin or its protein partners, exploited in MK886 and ciglitazone antitumor mechanism. Catalano et al. (8) described inhibition of p53 nuclear translocation by 5-lipoxygenase products. It would be interesting to explore the role of natural 5-lipoxygenase and cyclooxygenase-2 products along the lines in the present report.

This study found that cytoskeleton and focal adhesion functions responded to MK886 treatment with the Ingenuity score higher than other responses, and relation of this effect to anoikis was suggested by the RT-PCR experiment. This result was confirmed more directly by interference of the drug with actin polymerization. The link between cytoskeleton disruption and DNA processing deregulation was modestly confirmed by NCI-60 data and more reliably by RT-PCR and Western blotting. A novel antitumor mechanism was proposed based on these findings, addressing previously unknown aspects of the action and the safety of known pharmacologic agents. A mechanistic explanation was proposed for the augmentation of DNA-directed toxicity by cytoskeleton alteration, observed in a number of reports. Application of Inegnuity software in this project enabled us to formulate a mechanistic hypothesis based on relatively small-scale microarray study. Our work suggested utility of this algorithm for other similar studies, aiming to translate microarray results into a mechanistic hypothesis, leading to the insights into therapy.

We thank Dr. W. Tellford (NCI Flow Cytometry core facility) and Dr. R.H. Shoemaker (Screening Technologies Branch, NCI) for technical assistance, and the Fellows Editorial Board of NIH for their help with the review of the manuscript.