Activation of Aminoflavone (NSC 686288) by a Sulfotransferase Is Required for the Antiproliferative Effect of the Drug and for Induction of Histone γ-H2AX

Aminoflavone (AF) is entering clinical trials. We recently reported that AF induces DNA-protein cross-links (DPC) and γ-H2AX in MCF-7 human breast cancer cells. To elucidate the mechanism of action of AF and provide biomarkers indicative of AF activity, we correlated AF activity profile (GI₅₀) with gene expression patterns in the NCI-60 cell lines. Sulfotransferases (SULT) showed the highest positive correlation coefficients among ∼14,000 probe sets analyzed (r = 0.537, P < 0.001). Stable transfection of SULT1A1 into AF-resistant MDA-MB-231 cells sensitized these cells to AF. AF produced DPCs, γ-H2AX foci, and S-phase arrest in the SULT1A1-transfected but not in the parent MDA-MB-231 cells. Conversely, cells in which SULT1A1 was knocked down by small interfering RNA failed to induce γ-H2AX. Inhibition of SULTs and cytochrome P450 (CYP) enzymes by natural flavonoids blocked the antiproliferative activity of AF and the formation of AF-DNA adducts. AF also induces SULT1A1 and CYP expression in MCF-7 cells, suggesting the existence of an aryl hydrocarbon receptor–mediated positive feedback for AF activation by CYP and SULT1A1. Metabolism studies showed that AF can be oxidized by CYP at two amino groups to form N-hydroxyl metabolites that are substrates for bioactivation by SULTs. We propose that both N-sulfoxy-groups can be further converted to nitrenium ions that form adducts with DNA and proteins. The results reported here show the importance of SULT1A1 and CYP for AF activation and anticancer activity. They also suggest using SULT1A1 and γ-H2AX as biomarkers for prediction of AF activity during patient selection and monitoring of clinical trials. (Cancer Res 2006; 66(19): 9656-64)

Aminoflavone (AF; 5-amino-2,3-fluorophenyl-6,8-difluoro-7-methyl-4H-1-benzopyran-4-one; NSC 686288, Kyowa Hakko Kogyo Co., Ltd., Tokyo, Japan) is entering phase I clinical trials. It is one of a series of diaminoflavone analogues with potent growth-inhibitory activity against human breast and renal cancer cells in vitro and antitumor activity in mice bearing human tumor xenografts (1–4). AF displays a unique COMPARE pattern (5) of antiproliferative activity against the panel of 60 human cancer cell lines (the NCI-60) used to screen compounds for anticancer activity by the U.S. National Cancer Institute (NCI) Developmental Therapeutics Program. That is, AF shows a COMPARE pattern unlike that any other known classes of antitumor agents.³

A unique pattern of activity for a new agent in the NCI cell screen suggests a novel mechanism of drug action, transport, and/or metabolism. We recently reported that AF produces replication-dependent histone H2AX phosphorylation (γ-H2AX) and DNA-protein cross-links (DPC) in human breast cancer MCF-7 cells (6). AF also inhibits DNA synthesis, and it induces S-phase arrest accompanied by phosphorylation of RPA2 and p53, along with increased expression of p21^WAF1/CIP1 and MDM2 (6).

Previous studies showed that AF induces CYP1A1 and CYP1A2 in human breast cancer MCF-7 cells and that its cellular activity is accompanied by conversion to a series of metabolites by CYP1A1 and CYP1A2 (7). One of those metabolites was identified as a hydroxylamine; the others have yet to be identified (7). Some AF metabolites have been found to form covalent adducts with cellular proteins and DNA (7). Further studies showed that AF activates the aryl hydrocarbon receptor (AhR) signaling pathway, which, in turn, induces expression of CYP1A1 and CYP1B1 in MCF-7 cells (which are AF sensitive) but not in MDA-MB-435, PC-3, or AH^R100 cells (which are AF resistant; ref. 8).

To elucidate the mechanism of action of AF and identify potential clinical biomarkers, we studied the correlations between AF activity and gene expression in the NCI-60 cell lines, using data from Affymetrix HG-U133A microarrays. Sulfotransferases (SULT) showed the highest positive correlation among ∼14,000 probe sets examined. Following-up that observation, we found that the MDA-MB-231 cells, which are highly resistant to AF, express very low levels of SULT1A1 but that stable transfection with SULT1A1 rendered those cells sensitive to AF. Conversely, AF-sensitive cells become resistant to AF after knockdown of SULT1A1 level by small interfering RNA (siRNA). Finally, new AF metabolites were characterized by liquid chromatography/tandem mass spectrometry (LC/MS/MS).

Statistical methods for analysis of AF activity and gene expression data. Gene expression patterns for the NCI-60 cell lines were assessed using data from studies with HG-U133A Affymetrix chip.⁴

^,5 The HG-U133A chip, which comprises ∼22,215 probe sets, is part of a two-chip set (HG-U133A and HG-U133B) used for the NCI-60 in a collaboration with U. Scherf, D. Dolginow, and colleagues at Gene Logic, Inc. (Gaithersburg, MD).⁶ The protocols for cell culture, cell harvests, and mRNA purification have been described previously (9). Gene expression data were normalized using the robust multiarray method in the R statistical package. The probe sets were filtered to delete those with very little pattern (i.e., SD <0.2 log₂ units) across the NCI-60 cell lines. A final list of 7,925 probe sets was obtained for analysis. AF drug activity against the NCI-60 panel was obtained from the Developmental Therapeutics Program screen with a selected set of 1,429 drugs that had been tested multiple times and whose screening data met previously established quality control criteria (9). For the present study, drug activity was expressed as −log₁₀GI₅₀, where GI₅₀ is the concentration of drug required to inhibit cell growth by 50% in a 48-hour sulforhodamine B assay.⁷ Statistical analyses were done by using the R package.⁸ To identify genes that show significant correlation with AF, we computed pairwise Pearson correlation coefficients (r). To limit the number of false positives, we used the multivariate permutation test to give a P value estimate of significance (10, 11). Briefly, sample labels of the gene expression data were permuted randomly, the Pearson correlation with drug activity was recomputed for each random permutation, and the fraction of the random permutation correlation coefficients that exceeded the correlation coefficient for the actual data was determined. The genes were then ordered by the resulting P values (genes with smallest P values being placed at the top of the list).

Cell culture. MCF-7 human breast cancer cells were obtained from American Type Culture Collection (Manassas, VA). MDA-MB-231 (MDA) and MDA-MB-231 cells stably transfected with a SULT1A1 cDNA under control of the cytomegalovirus promoter (MDA/SULT1A1; ref. 12) were kindly provided by Dr. David C. Spink (Wadsworth Center, Albany, NY). AhR-deficient MCF-7 cells (AH^R100; ref. 8) were kindly provided by Dr. David T. Vistica (Screening Technologies Branch, NCI, Frederick, MD). Cells were grown at 37°C in the presence of 5% CO₂ in DMEM medium supplemented with 10% fetal bovine serum (FBS; Life Technologies, Grand Island, NY), 100 units/mL penicillin, and 100 mg/mL streptomycin. SULT1A1-transfected cells were maintained in medium containing 600 μg/mL geneticin (Invitrogen, Carlsbad, CA).

Drugs and chemicals. AF (Kyowa Hakko Kogyo) was obtained from the Developmental Therapeutics Program. Apigenin, daidzein, and genistein were purchased from Sigma Chemical (St. Louis, MO). All the drugs were dissolved in DMSO to a concentration of 10 mmol/L and aliquots were stored at −20°C. Dilutions to desired concentrations were done in culture medium immediately before each experiment. Final DMSO concentrations did not exceed 0.1%.

Real-time quantitative PCR. Cells were lysed and total RNA was extracted using RNAqueous-4PCR (Ambion, Austin, TX). Total RNA was reverse-transcribed using RETROscript (Ambion). Real-time quantitative PCR was done using ABsolute QPCR Mixes (Abgene, Rochester, NY) on an ABI 7900 real-time PCR instrument (AME Bioscience, Chicago, IL). Thermal cycling conditions were 50°C for 2 minutes, 95°C for 15 minutes, then 40 cycles of 95°C for 15 seconds and 60°C for 1 minute. Primer and probe sequences used were as follows: SULT1A1, GGCCTGATGACCTGCTCATC (sense), TCATGTCCAGAATCTGGCTTACC (antisense), FAM-CACCTACCCCAAGTCCGGCACTACCT-TAMRA (probe); p21^Waf1, GCAGACCAGCATGACAGATTTC (sense), GCGGATTAGGGCTTCCTCTT (antisense), FAM-CACTCCAAACGCCGGCTGATCTTC-TAMRA (probe); CYP1A1 (8), GATTGGGCACATGCTGACC (sense), TGCTGGCTCATCCTTGACAG (antisense), FAM-TGGGAAAGAACCCGCACCTGGC-TAMRA (probe); CYP1B1 (8), TTTCGGCTGCCGCTACA (sense), ACTCTTCGTTGTGGCTGAGCA (antisense), FAM-ACGACGACCCCGAGTTCCGTGAG-TAMRA (probe); and 18S RNA, GATTAAGTCCCTGCCCTTTGTACA (sense), GATCCGAGGGCCTCACTAAAC (antisense), FAM-CGCCCGTCGCTACTACCGATTGG-TAMRA (probe). Gene expression was analyzed using Sequence Detection Systems software, version 1.7 (ABI PRISM). mRNA levels of SULT1A1, p21^Waf1, CYP1A1, and CYP1B1 were normalized to the 18S RNA internal standard.

3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide assays. 3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assays were done as described previously (13). Briefly, MDA, MDA/SULT1A1, and MCF-7 cells were seeded on day 0 at a density of 3,000 per well in 96-well microtiter plates. On day 1, AF was added and incubation was continued for 72 hours. After 72 hours, 20 μL MTT (5 mg/mL; Sigma) were added to each well, and plates were maintained at 37°C for an additional 4 hours. One hundred microliters of DMSO were then added to each well to lyse the cells. Absorbance was measured at 570 nm using a multiwell spectrophotometer (E_max, Molecular Devices, Sunnyvale, CA).

Clonogenic assays. Cells were treated with AF for 6 hours. Following treatment, cells were washed in PBS and trypsinized. Cells were seeded in six-well plate in 3 mL medium at two densities (2,000 and 200 per well). Colonies were grown for 2 weeks. Plates were washed in PBS, and colonies were fixed with methanol. Staining was with methylene blue (0.04%).

Alkaline elution assays. Alkaline elutions were done to assess DNA damage by detecting DPC, as described previously (14, 15). Cellular DNA was radiolabeled with 1 μCi/mL [³H]thymidine (Perkin-Elmer Life Science Co., Boston, MA) for 48 hours at 37°C and then chased with nonradioactive medium overnight. After drug treatments, cells were scraped into HBSS and counted. Samples of the cells were then placed in drug-containing ice-cold HBSS. After alkaline elution, filters were incubated at 65°C with 1 mol/L HCl for 45 minutes, and 0.04 mol/L NaCl was then added for an additional 45 minutes of incubation. Radioactivity in each fraction was measured by liquid scintillation (Packard Instruments, Meriden, CT). DPCs were analyzed under nondeproteinizing, DNA-denaturing conditions using protein-adsorbing filters.

Laser scanning confocal microscopy. Laser scanning confocal microscopy was done as described (6). Cells were grown in culture medium on chamber slides. After drug treatment, the cells were fixed with paraformaldehyde and washed in PBS, then permeabilized in 100% methanol. Slides were blocked for 1 hour with PBS containing 1% bovine serum albumin and 5% goat serum (Jackson Immunolaboratories, West Grove, PA), incubated with anti-γ-H2AX antibody (Upstate, Chicago, IL), washed, incubated for 1 hour with AlexaFluor 488–conjugated goat anti-mouse IgG second antibody (Molecular Probes, Eugene, OR) at a 200-fold dilution, and washed in PBS. The slides were stained with propidium iodide, sealed with mounting medium (Vectashield; Vector Laboratories, Inc., Burlingame, CA), and viewed using a PCM2000 laser scanning confocal microscope (Nikon Co., Tokyo, Japan) with a ×40 objective. Images were saved as BMP files.

Flow cytometry analysis of DNA content. Cell cycle analyses were done using a FACScan flow cytometer (Becton Dickinson, Sunnyvale, CA) as previously described (13). Cell cycle distributions were calculated using ModFit LT software (Verity Software House, Inc., Topsham, ME).

Transfection with siRNA for SULT1A1. Synthetic siRNA targeting SULT1A1 (sequence 5′-GUUCAUGGUCGGAGAAGUGTT-3′) was purchased from Ambion. Negative control siRNA was purchased from Qiagen (Valencia, CA). siRNA was transfected into MDA/SULT1A1 cells using OligofectAMINE 2000 (Invitrogen) according to the instructions of the manufacturer. After transfection with siRNA for 48 hours, cells were seeded in six-well plates or four-chamber slides. Cells were treated with AF or collected for reverse transcription-PCR (RT-PCR) on the 2nd day.

Analysis of AF metabolism with human liver microsomes. Microsomal incubations were carried out for 30 minutes at 37°C in 20 mmol/L PBS (pH 7.4) containing 0.5 mg/mL pooled human liver microsomes (BD Gentest, San Jose, CA), 2 mmol/L MgCl₂, 20 μmol/L AF, and 1 mg/mL freshly prepared β-NADPH in a final volume of 0.2 mL. Reactions were terminated by adding an equal volume of acetonitrile. After centrifugation for 10 minutes at 14,000 rpm, the supernatant was transferred into a sample vial for LC-MS analysis on a Waters UPLC-Q-TOF Premier system (Milford, MA). An Acquity UPLC BEH C₁₈ column (Waters) was used to separate AF and its metabolites at 30°C. Mobile phase was flowed through the column at the rate of 0.6 mL/min with a gradient from 5% aqueous acetonitrile to 95% aqueous acetonitrile containing 0.1% formic acid in a 10-minute run. The Q-TOF mass spectrometer was operated in the turbo ion spray mode with positive ionization. Capillary voltage was maintained at 3 kV, and the desolvation temperature was 300°C. Nitrogen was applied as the cone and desolvation gas. Chromatograms and spectra of AF and its metabolites were analyzed using MassLynx software.

Covalent binding of AF to DNA in tumor cells. MCF-7, MDA, and MDA/SULT1A1 cells were prelabeled with 0.005 μCi/mL [¹⁴C]thymidine (53.6 mCi/mmol) for 24 hours at 37°C. The medium was removed and replaced with fresh DMEM containing 10% FBS. AF [1 μmol/L final concentration consisting of 0.2 μmol/L ³H labeled drug (7.09 Ci/mmol) and 0.8 μmol/L unlabeled drug] was added on the 2nd day. Whole DNA was extracted using the DNeasy Tissue kit from Qiagen after a 6-hour treatment. DNA was eluted with the DNeasy Tissue kit buffer. The concentration of DNA in the eluate was determined based on the absorbance at 260 nm. Purity of the DNA samples was verified with the A₂₆₀/A₂₈₀ ratio of the eluate (always between 1.8 and 2.0, indicative of quite pure DNA). The eluate (40 μL) was used to determine the amount of bound radiolabel by liquid scintillation counting.

SULT expression is highly correlated with AF antiproliferative activity in the NCI-60 cell lines. To understand the unique antiproliferative activity pattern of AF in the NCI-60 cell line screen, we computed pairwise Pearson correlation coefficients (r) between AF activity and gene expression levels (16) for each cell line. Analysis of 7,952 probe sets on the Affymetrix HG-U133A microarray chip showed SULT1A1, SULT1A2, and SULT1A3 to be among the most highly correlated genes (Fig. 1A). The correlations were statistically significant (P < 0.001). In contrast, basal levels of cytochrome P450 (CYP) family members (CYP1A1, CYP1A2, etc), which are involved in AF metabolism (4, 7), showed no significant correlation with AF activity (Fig. 1A). AhR, a transcriptional regulator of CYP1A1, also showed low correlation with AF activity.

We next plotted SULT1A1 expression levels and AF activity for each of the NCI-60 cell lines (Fig. 1B and C). MCF-7 cells, which express the highest levels of SULT1A1 mRNA, are among the most sensitive to AF, whereas MDA-MB-231 cells, which show very low expression of SULT1A1, are resistant to AF. The activity of AF and expression of SULT1A1 tend to be in the same direction, indicating positive correlation. The higher the SULT1A1 expression, the greater the sensitivity of a given cell line.

To investigate whether the positive correlation between AF activity and SULT1A1 expression is tissue type dependent, we plotted the cell lines according to their tissues of origin (Fig. 1C). AF activity and SULT1A1 expression were tightly correlated for the breast, central nervous system, and non–small lung cancer cell lines. They were moderately well-correlated for the melanomas, ovarian cancers, and renal cancers but poorly correlated for the colon cancer and leukemia cell lines. Some colon cell lines, such as HCC-2998 and COLO-205, were relatively resistant to AF although their SULT1A1 levels are high. However, some renal lines (TK10 and CAKI-1) were sensitive to AF although their SULT1A1 levels are about average. Overall, these data suggest the importance of SULT expression for AF activity against at least some of the tissue types. It should be noted, however, that the numbers of cell types in each tissue category are small.

Transfection with SULT1A1 renders MDA-MB-231 cells sensitive to AF. The mRNA levels of AF-resistant cells (MDA-MB-231), AF-sensitive cells (MCF-7), and MDA-MB-231 cells transfected with SULT1A1 (MDA/SULT1A1) were determined by real-time PCR. As expected from the microarray analysis (Fig. 1), MDA-MB-231 (MDA) cells express very low levels of SULT1A1, whereas MCF-7 cells express ∼8-fold higher levels (Fig. 2A). Cells stably transfected with SULT1A1 (MDA/SULT1A1; ref. 12) express >20-fold higher levels than the parental MDA-MB-231 cells (Fig. 2A). Transfection of SULT1A1 restored SULT activity in the MDA-MB-231 cells used in our experiments (12).

To detect whether cells with forced expression of SULT1A1 gain sensitivity to AF, MTT assays and clonogenic assays were done after treatment with AF. As shown in Fig. 2B and C, cells transfected with SULT1A1 were clearly more sensitive to AF than parental MDA-MB-231 cells, which are resistant to AF. Their sensitivity was comparable with that of MCF-7 cells. These results show that transfection of SULT1A1 into AF-resistant cells sensitizes the cells to AF.

Transfection of SULT1A1 into MDA-MB-231 cells restores formation of DPC and histone γ-H2AX foci in response to AF. We previously showed that AF induces DPC in MCF-7 cells (6). The DPC, in turn, produce S-phase DNA double-stand breaks, which arrest cells in S phase (6). To determine whether forced expression of SULT1A1 in MDA-MB-231 cells can restore the sensitivity of cells to AF, we measured DPC, γ-H2AX generation, and cell cycle distribution in AF-treated MDA/SULT1A1 cells. As shown in Fig. 3A, elution curves of parental MDA-MB-231 treated with AF were similar to those of untreated samples. DNA eluted quickly through the protein-adsorbing filters, suggesting that very few DPC had formed in the cells in response to AF. By contrast, AF produced profuse DPC in SULT1A1-transfected cells, as indicated by reduced elution rates compared with those for untreated cells (Fig. 3A).

Because γ-H2AX focus formation is a sensitive and selective indicator of double-stand break (17, 18), and double-stand break formation is correlated with the antiproliferative activity of AF, we used γ-H2AX as a marker for monitoring the activity of AF. We looked for γ-H2AX focus formation in AF-treated parental and MDA/SULT1A1 cells. As shown in Fig. 3B, γ-H2AX foci were not induced and remained at basal levels (8.4% of cells positive) after treatment of the parental cells with 1 μmol/L AF for 6 hours. In contrast, γ-H2AX foci were strongly induced (59.6% of cells positive) in the SULT1A1-transfected cells treated with AF. Without AF treatment, only 3% of the transfected cells were positive. These results indicate that SULT is required for induction of γ-H2AX foci by AF.

To investigate whether introduction of SULT1A1 also restored the cell cycle effects of AF, MDA and MDA/SULT1A1 cells were treated with 1 μmol/L AF for 18 hours, and their cell cycle distribution was analyzed. Whereas AF had no effect in parental MDA cells, AF arrested the transfected cells in S phase and also induced apoptosis indicated by a sub-G₁ peak (Fig. 3C).

Knocking down SULT1A1 reduces induction of histone γ-H2AX by AF. To investigate whether knocking down SULT1A1 in AF-sensitive cells would influence the effects of AF, we transfected MDA/SULT1A1 cells with siRNA targeting SULT1A1. After transfection with 10 nmol/L siRNA, the SULT1A1 mRNA level decreased by ∼83% (relative to the level in cells transfected with 100 nmol/L negative control siRNA; Fig. 3D,, top). Similar reduction of SULT1A1 was obtained with 50 and 100 nmol/L SULT1A1 siRNA. Detection of γ-H2AX induction by AF showed a 20- to 50-fold reduction in the cells transfected with the siRNA targeting SULT1A1 (Fig. 3D , bottom).

Natural flavonoids antagonize the cytotoxicity of AF. Natural flavonoids, such as apigenin, daidzein, and genistein (see structures in Fig. 4), have been reported to inhibit both SULTs (19, 20) and CYP enzymes (21). They also activate AhR although much less than TCDD, a known potent AhR activator (22). To test whether flavonoids would interfere with the activity of AF, we treated MCF-7 cells with AF alone or with different concentration of flavonoids. As shown in Fig. 4, the three flavonoid derivatives blocked the antiproliferative activity of AF in a dose-dependent manner. Cotreatment with 30 μmol/L flavonoids completely blocked the cytotoxicity of AF. These results suggest that natural flavonoids antagonize AF activity by inhibiting both SULT1A1 and CYP.

Induction of SULT1A1, CYP1A1, and CYP1B1 gene expression by AF. Because AF has been reported to induce CYP1A1 expression (7, 8), we investigated whether AF also affects SULT1A1 expression. MCF-7 cells and MDA cells were treated with 1 μmol/L AF, and SULT1A1 expression was measured at different times of exposure. AF treatment increased SULT1A1 mRNA in a time-dependent manner in MCF-7 cells (Fig. 5A). SULT1A1 mRNA was also induced in a dose-dependent manner with significant induction at submicromolar concentrations of AF (Fig. 5B). Induction of SULT1A1 by AF in MCF-7 cells is mild compared with the induction of CYP1A1 and CYP1B1 (Fig. 5D; refs. 7, 8). In contrast, AF failed to induce SULT1A1 mRNA expression in MDA cells (Fig. 5A), which have low basal mRNA levels of SULT1A1 and are resistant to AF (see Fig. 2).

To explore whether SULT1A1 induction by AF is activated by AhR, we measured SULT1A1 mRNA levels after treatment with 1 μmol/L AF for 6 hours in MCF-7 cells and AhR-deficient MCF-7 cells (AH^R100). As shown in Fig. 5C, AF failed to induce SULT1A1 expression in AH^R100 cells (Fig. 5C). Examination of the SULT1A1 gene revealed the presence of two AhR-responsive elements in the promoter region of SULT1A1 gene. These observations suggest that the transcriptional activation of SULT1A1 induced by AF is mediated by AhR. We reported recently that AF also induces p21^waf1 expression in MCF-7 cells as a result of DNA damage (6). To investigate whether silencing AhR pathway in AH^R100 cells would also affect the cellular response by AF, we measured the p21^waf1 induction at the same time. Figure 5C shows that induction of p21^waf1 by AF in AH^R100 cells is half of that in MCF-7 cells. These results suggest that induction of CYP and SUL1A1 by AF may further enhance activation and its cellular activity as well.

To detect whether AF induces the expression of CYP1A1 and CYP1B1 in the AF-sensitive MDA/SULT1A1 cells (see Fig. 2), cells were treated with 0.1 μmol/L AF for 6 hours and mRNA levels were measured. As shown in Fig. 5D, AF failed to induce CYP1A1 and CYP1B1 in MDA-MB-231 parental and SULT1A1-transfected cells (Fig. 5D). By contrast, AF greatly induces expression of CYP1A1 (∼120-fold) and CYP1B1 (∼20-fold) in MCF-7 cells, which is consistent with previous results (7, 8). The lack of CYP induction in MDA cells is consistent with a previous study showing that the AhR activator, TCDD, failed to induce CYP1A1 and CYP1B1 in MDA-MB-231 cells (23).

Together, these results indicate that induction of SULT1A1 by AF is mediated by the AhR pathway in MCF-7 cells.

Potential pathways of AF bioactivation. The results described above reveal the importance of SULTs for the activity of AF. To identify the AF metabolites, LC/MS/MS analyses were done after incubation of AF with human liver microsomes (Fig. 6). N⁴′-hydroxy-AF (metabolite III) was identified in agreement with an earlier report (7). We also found that AF is hydroxylated at position 3 (metabolite IV) and N-hydroxylated on the amino group at position 5 (metabolite II). Two dihydroxylated products (metabolites V and VI) were also detected (Fig. 6). Similar results were obtained in medium of mouse hepatoma cells (Hepa1) incubated in AF containing medium (data not shown) and in urine samples from mice administrated with AF by p.o. gavage (24). These results suggest that AF is metabolized by one or more CYP.

N-hydroxy metabolites are known substrates for SULTs. N-sulfoxy metabolites are unstable and rearrange to nitrenium ions by spontaneous heterolytic cleavage to a very reactive derivative that can covalently bind to DNA and proteins (25). To confirm that SULTs are required for AF to form AF-DNA adducts, MCF-7, MDA, and MDA/SULT1A1 cells were treated with ³H-radiolabeled AF, and AF-DNA adducts were measured. Covalent binding of AF to DNA was observed in MCF-7 cells (Fig. 6B). By contrast, almost no covalent AF-DNA adducts could be detected in AF-treated MDA cells. AF-DNA adducts increased greatly in the MDA/SULT1A1 cells (Fig. 6B). Moreover, DNA-AF adducts decreased dramatically in MCF-7 cells treated with AF in the presence of apigenin, daidzein, or genistein (Fig. 6B), which are inhibitors of SULTs (19, 20) and CYP (21). These results indicate the importance of SULT1A1 and CYP in the bioactivation of AF and in the generation of AF-DNA adducts.

AF, an “investigational new drug,” shows a unique COMPARE pattern of antiproliferative activity against the NCI-60 in the NCI tumor cell line screen. In this study, we looked for correlations between gene expression patterns and AF activity in the NCI-60 cell lines. SULTs showed the highest positive correlations among ∼7,925 probe sets present on the Affymetrix HG-U133A microarray chip (r = 0.537, P < 0.001). Transfection with SULT1A1 rendered MDA-MB-231 cells sensitive to AF and restored DPC and histone γ-H2AX focus formation in response to AF. siRNA knockdown of SULT1A1 in AF-sensitive cells or inhibition of SULTs by natural flavonoids reduced AF activity. Moreover, we found that AF induces AhR-dependent SULT1A1 gene expression. Taken together, those results indicate that SULTs play an important role in the cellular activity of AF.

Cytosolic SULTs are important phase II enzymes. They transfer the sulfate (SO^3-) from the cofactor 5′-phosphoadenosine-3′-phosphosulfate to nucleophilic groups of a wide array of xenobiotic substrates and small endogenous compounds, including steroids (26–28). For most xenobiotics and small endogenous substrates, sulfation provides a detoxification pathway leading to soluble products that are excreted in the urine or bile (28). However, for xenobiotics, such as N-hydroxy-arylamines, N-hydroxy heterocyclic amines, and hydroxymethyl polycyclic aromatic hydrocarbons, sulfation leads to metabolic activation to highly reactive nitrenium electrophiles capable of reacting with DNA and other cellular nucleophilic macromolecules (25, 26, 28). Indeed, SULTs are known to activate promutagens and procarcinogens, such as PhIP (29), benzylic alcohol (30), 3-nitrobenzanthrone (31), and N-hydroxy-2-acetylaminofluorene (32).

AF can be hydroxylated at two positions (5′ and 4′) occupied by amino groups (Fig. 6; ref. 7). The resulting hydroxylamines can be readily conjugated by SULTs, thereby forming N-sulfoxy-AF metabolites that can be converted into nitrenium ions by spontaneous heterolytic cleavage. The nitrenium ions can then form adducts with DNA or proteins (Fig. 6C). Thus, the active metabolites of N-sulfoxy-AF can act as bifunctional linkers (using the 5′ and 4′ positions) that cross-link DNA and proteins to form DPC. Although CYP show low correlation with AF activity in NCI-60 cell lines (Fig. 1A), CYP, which are phase I enzymes, are critical and upstream from SULT1A1 for AF activation by SULT1A1. CYP might be active even at low levels and can be greatly induced after AF treatment (Fig. 5D; refs. 7, 8), which may explain the low correlation between AF activity and CYP expression in the NCI-60 cell lines (Fig. 1A). On the other hand, in spite of the fact that MDA cells have ∼3-fold higher CYP1A1 and 15-fold higher CYP1B1 than MCF-7 cells (data not shown), AF does not induce DNA damage in MDA-MB-231 cells, whereas it does in MDA/SULT1A1 cells that have similar levels of CYP1A1 and CYB1B1. Moreover, relative down-regulation of SULT1A1 with siRNA shows the critical role of SULT1A1 for AF activation and cellular response. Therefore, CYP1A1/CYP1B1 are not sufficient to fully activate AF and produce cellular damage. Our study suggests that both CYP and SULTs are required for the generation of the AF metabolites that damage DNA (Fig. 6). Nevertheless, SULT1A1 gene expression seems to be a better correlate than CYP for AF activity.

It is noteworthy that tamoxifen (TAM) is another anticancer drug metabolized by SULTs (33). Tamoxifen is used as the standard adjuvant endocrine therapy for breast cancer patients and as a chemopreventive agent for women at high risk for the disease (34). The increased incidence of endometrial cancers in patients treated with tamoxifen has been attributed to genotoxic damage induced by tamoxifen-DNA adducts after O-sulfation of tamoxifen by SULT (35). To our knowledge, AF is the only reported anticancer agent for which activation by SULTs is required for its therapeutic activity. However, activation of AF by sulfation can both kill cancer cells and injure normal tissues. In that context, AF is known to elicit lung toxicity,⁹

an effect that might be related to high SULT activity present in lung tissues (36). In the present study, we find that expression of SULT1A1 is not especially high in non–small cell lung cancers of the NCI-60 panel, but that AF activity is tightly correlated with SULT1A1 expression in those cells (r_mean = 0.837; P < 0.001; Fig. 1B).

AF was found to induce CYP1A1, CYP1B1, and SULT1A1 in MCF-7 cells, indicating that AF can induce its own metabolism. Earlier studies (8) revealed that induction of CYP1A1 and CYP1B1 expression by AF was mediated by AhR, suggesting that AF or one of its metabolites might be an activator of the AhR. AF failed to induce SULT1A1 in AhR-deficient cells (Fig. 5C) and two AhR-responsive elements were revealed in the promoter region of SULT1A1 gene, suggesting that SULT1A1 might also be a target gene for AhR. Thus, activation of AhR pathway may provide a positive feedback for the activation of AF, which may further enhance the cellular responses to AF. It is also likely that the 60 cell lines from the NCI panel have distinctive AhR signaling. For instance, as shown in Fig. 5D, there is no CYP1A1 and CYP1B1 induction in MDA-MB-231 parental and SULT1A1-transfected cells. Different AhR signaling in NCI-60 cell lines may contribute to the differential sensitivities to AF.

The antitumor activity of AF in animal tumor models can be rationalized in light of the present study. First, SULT levels are relatively high in some tumor cells (Fig. 1) and also are inducible after AF treatment (Fig. 5). Second, tumor cells tend to proliferate more rapidly than do normal cells, and AF preferentially damages DNA in proliferating cells (6). Third, natural flavonoids antagonized the activity of AF, which might be used to protect normal tissues. Fourth, it is likely that SULT levels are not the only factor that determines the activity of AF. For example, some colon cell lines from the NCI-60 panel are resistant to AF despite relatively high expression of SULTs (see Fig. 1), suggesting that cell growth inhibition by AF is a multifactorial coordinated process. Further studies are warranted to elucidate additional factors that determine the cellular response to AF.

In summary, both CYP and SULTs are important for metabolic activation of AF to electrophilic metabolites capable of killing cells by binding cellular macromolecules, including DNA and proteins. The potential bioactivation pathway of AF is depicted in Fig. 6C. Our study suggests that SULT levels should be assayed in tumor samples from patients as part of the selection process before treatment with AF. They also suggest that natural flavonoids are pharmacologic antagonist of AF, as they block AF activation. Another potential pharmacodynamic biomarker for AF is γ-H2AX induction, which might be used to monitor the pharmacodynamic activity of AF in clinical trials.

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.