Parthenolide Inhibits Tubulin Carboxypeptidase Activity

Microtubules are key components of the cytoskeleton of eukaryotic cells. They are centrally involved in the control and mechanics of cell division, being the principal components of mitotic spindle. The building block of microtubules is the αβ-tubulin heterodimer. In most eukaryotic cells, the α chain of tubulin can be modified by enzymatic removal of the COOH-terminal tyrosine residue by an ill-defined tubulin carboxypeptidase (TCP) and by readdition of this tyrosine by a distinct enzyme, tubulin tyrosine ligase (TTL; refs. 1, 2). This tyrosination cycle is conserved among eukaryotes (1, 3) and generates two tubulin pools: intact tyrosinated α-tubulin (Tyr-tubulin) and detyrosinated α-tubulin (Glu-tubulin), which lacks the COOH-terminal tyrosine. Its physiologic importance in normal cells, tissues, or organisms has been shown with the generation of TTL-null mice. TTL-null mice die within hours after birth due to the disorganization of vital neuronal circuits (4), indicating that TTL is essential for neuronal organization.

TTL is also frequently suppressed during tumor progression (5), with resulting accumulation of Glu-tubulin in tumor cells. TTL suppression and resulting tubulin detyrosination in human cancers are associated with increased tumor aggressiveness (6–8). Experiments conducted in yeast (9), in TTL nude mice (4), and in TTL-null fibroblasts (10) show that tubulin tyrosination regulates microtubule interactions with CAP-Gly microtubule plus tip-tracking proteins, such as CLIP-170. In TTL-null fibroblasts, tubulin detyrosination and CAP-Gly protein mislocalization correlate with defects in spindle positioning during mitosis. Because impaired control of spindle positioning has previously been proposed as one factor favoring tumor invasiveness (11), these results partly shed light on the molecular basis for the apparent stimulatory effect of Glu-tubulin accumulation on tumor progression.

Inhibition of TCP could reverse abnormal Glu-tubulin accumulation in tumor cells and could lead to the restoration of normal Tyr-tubulin levels through tubulin synthesis (4). Such an inhibition could thus impair tumor progression (7, 8). Until now, there are no known specific TCP inhibitors (12). Moreover, the TCP gene has never been identified, although biochemical TCP activity has been reported to be present in some subcellular fractions (1, 3). Thus, the discovery of specific TCP inhibitors could also be useful for a ligand-based affinity purification of TCP. Finally, TCP inhibition is also necessary for a full understanding of the tyrosination cycle. Detyrosinated tubulin is very abundant in differentiated cells (13). Although tubulin tyrosination is essential for microtubule interaction with some proteins plus tip-tracking proteins, it is not known why eukaryotic cells developed a tyrosination cycle by introducing a detyrosination reaction.

Chemogenomic strategies have proven useful to isolate chemical compounds that could induce a specific phenotype (14, 15). In such approaches, also referred as chemical genetics, cells are tested against a library of small molecules using high-throughput screening and then examined for a desired phenotype (16–18). The subsequent identification of the intracellular protein target of these compounds, although often technically difficult (19), has obtained notable successes (20–23). We designed a phenotype assay of TCP activity and used this cell-based assay to screen a chemical library to isolate TCP inhibitors. In this report, we describe the assay, its application to a screen of 23,904 natural extracts, and the purification and characterization of two sesquiterpene lactones that were able to specifically inhibit TCP. We tested several structurally related compounds and found that parthenolide, a known nuclear factor-κB (NF-κB) inhibitor (24), can efficiently inhibit TCP activity. Parthenolide has also been shown to present some anticancer properties, but the underlying mechanisms are not fully understood. TCP inhibition could be one of the mechanisms underlying the known anticancer properties of parthenolide.

HeLa cells were purchased from the American Type Culture Collection (Rockville, MD). They were grown in RPMI 1640 with Glutamax supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin. Cells were maintained at 37°C in 5% CO₂ and 95% air.

Carboxypeptidase A (CPA), orthophenanthroline, bovine serum albumin (BSA), EGTA, DMSO, glycerol, Hoechst 33258, magnesium chloride solution, PIPES, potassium chloride, potassium phosphate monobasic, sodium phosphate dibasic, Triton X-100, and Tween 20 were purchased from Sigma-Aldrich (St. Louis, MO). RPMI 1640 with Glutamax I, trypsin-EDTA solution, and penicillin-streptomycin solution was from Life Technologies, Invitrogen (Carlsbad, CA). Sodium chloride was from Carlo Erba Reagents (Rodano, Milan, Italy). FBS was from Perbio (Brebières, France). SuperSignal ELISA Pico chemiluminescent substrate was from Pierce (Rockford, IL).

Parthenolide and costunolide were from Chromadex (Santa Ana, CA). 1N-02620, 1N-20266, 1N-21474, 1N-24000, 1N-24632, and 1N-37712 were from InterBioScreen (Moscow, Russia). Epoxyastaphiatone, alantolactone, and arglabine were from Specs (Delft, the Netherlands). Ergolide, deacetylmatricarin, and helenalin were from Apin Chemicals (Abingdon, United Kingdom). Santonin and sulindac were from Sigma-Aldrich (L'Isle d'Abeau, Chesnes, France). Cnicin was from Roth-Sochiel (Lauterbourg, France).

Dihydroparthenolide was obtained by hydrogenation of parthenolide as described by Kwok et al. (25).

The monoclonal ID3 (TTL blocking antibody) and the monoclonal Tyr-tubulin antibody (clone YL1/2) were generous gifts from Dr. J. Wehland (National Research Centre for Biotechnology, Braunschweig, Germany) and Dr. J.V. Kilmartin (Medical Research Council, Cambridge, United Kingdom), respectively. YL1/2 can also be purchased from Abcam (Cambridge, United Kingdom).⁵

Microplate preparation. HeLa cells were cultured as monolayers. The cells were seeded at 30,000 per well in 96-well polystyrene tissue culture plates (reference no. 655 083; Greiner Bio-One, Kremsmuenster, Austria) in 90 μL medium. They were allowed to grow for 24 h.

Drug addition. The day following microplate preparation, each well was robotically supplemented with 10 μL RPMI 1640 containing natural extracts and paclitaxel so that the final concentrations were 35 μg/mL for natural extracts and 5 μmol/L for paclitaxel. Bioinactive controls were obtained with paclitaxel alone (5 μmol/L). The final concentration of DMSO was 0.5%. After drugs dispense, cells were incubated for 2 h at 37°C, 5% CO₂, in the workstation incubator.

Cell immunostaining. After medium aspiration, cells were washed with TBS/0.1% Tween 20 and fixed for 1 h with 3.7% formaldehyde. Cells were permeabilized with 100 μL methanol for 5 min.

After two washes with 300 μL of TBS/0.03% Tween 20 [10 mmol/L Tris-HCl, 137 mmol/L NaCl, 0.03% (v/v) Tween 20 (pH 7.4)], cells were coincubated for 1.5 h at 37°C with Glu-tubulin and HRP-conjugated antibodies diluted, respectively, at 1:10,000 and 1:5,000. Cells were rinsed twice and then incubated for 10 min with SuperSignal ELISA Pico chemiluminescent substrate. Luminescence was measured using a BMG FLUOstar Optima plate reader (BMG Labtechnologies, Offenburg, Germany).

Result analysis. Raw data were exported from the software of the microplate reader to an Excel file and normalized according to the following procedure. Mean values of DMSO and paclitaxel were calculated for each individual microplate using intraplate control data points. Paclitaxel and DMSO mean values were considered as bioinactive control (0% inhibition) and bioactive control (100% inhibition) of TCP activity, respectively.

Assay performance. To evaluate the quality of the assay, the Z′ factor was calculated using the following equation according to Zhang et al. (26):

where c⁺ is the bioactive controls, c⁻ is the bioinactive controls, μ is the mean, and σ is the SD.

In our experimental conditions, the Z′ was comprised between 0.54 and 0.57, indicating that the assay is of good quality for high-throughput screening.

Plant samples were collected worldwide under the general framework of the International Convention on Biological Diversity (Rio de Janeiro, Argentina, 1992). Bases for the collection were the taxonomic identification of the species, botanical diversity, and availability for a second collection. No rare or protected plants belonged to the collection. When possible, the different organs of the plants were separated and treated as different samples (bark, leaves, root, etc.). Each plant sample was dried (55°C) and finely ground. An aliquot of the plant (20 g) was macerated overnight in 200 mL ethyl acetate in an Erlenmeyer flask; after filtration, the solvent was evaporated in vacuo to give the ethyl acetate crude extract. Part of the extract (50 mg) was adsorbed on 500 mg Sigel, which was placed on top of SiO₂ solid-phase extraction cartridge. Successive elutions with CHCl₃, CHCl₃/MeOH (17:3), and MeOH provided three “fractions,” which were conditioned in 96-well deep-well plates for the screening after evaporation of the solvents and replacement with 1 mL DMSO, along with the “crude extract” at a concentration of 50 mg/mL.

Finely ground aerial parts from the plant (97 g) were macerated overnight in 1 L ethyl acetate; the suspension was filtered and the organic filtrate was evaporated in vacuum to give 9.6 g of a gum. Part of this extract (4 g) was dissolved in 8 mL chloroform and adsorbed on a Sigel column (120 g) eluted in 200 mL fractions under 18 p.s.i. at a 30 mL/min flow rate with a chloroform/methanol gradient (10 min: CHCl₃; 50 min: CHCl₃/MeOH (17:3); 10 min: MeOH). The second fraction concentrated the activity and was further purified on a semiprep high-performance liquid chromatography Lichrospher column with a water/acetonitrile gradient. Compound V18 was eluted at 15.6 min and compound V19 was eluted at 16.6 min (gradient system: water/acetonitrile, from 6/4 to 1/19 in 20 min).

V18: crystals from diethyl ether (melting point 136°C), MS, UV, IR, and nuclear magnetic resonance (NMR) spectra identical with those published in the literature (27); V19: amorphous gum with MS, UV, IR, and NMR spectra identical with those published in the literature.

Some of the natural extract fractions were tested for their cytotoxic activity. This cytotoxic activity was estimated using 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) cell proliferation/cytotoxicity assay. This assay is a colorimetric assay, which measures the reduction of a tetrazolium component (MTT) into an insoluble formazan product by the mitochondria of viable cells. Briefly, cells were plated in 96-well flat-bottomed microplates at a density of 30,000 per well. Twenty-four hours after plating, the fractions were added and appropriately diluted in DMSO. After 2 h of incubation with the fractions, the medium was replaced with MTT (Sigma-Aldrich) dissolved at a final concentration of 1 mg/mL in serum-free, phenol red–free RPMI 1640 (Life Technologies). The cells were then let to incubate for 3 h to develop purple formazan precipitate. The MTT formazan was then solubilized in DMSO, and the absorbance was measured at a wavelength of 570 nm and a reference wavelength of 630 nm.

Cells grown for 2 days on glass coverslips were washed with warm PBS and fixed for 6 min in anhydrous methanol at −20°C. After an additional wash in PBS/0.1% (v/v) Tween 20, cells were sequentially incubated with primary antibodies and appropriate secondary antibodies. Antibodies were diluted in PBS/0.1% BSA. YL1/2 and L4 antibodies were used at a 1:4,000 dilution. Cy3-conjugated antibodies and Alexa Fluor 488–conjugated antibodies were used at a 1:1,000 and 1:500 dilution, respectively. Nuclei were stained with Hoechst 33258 (1 μg/mL).

Coverslips were mounted in FluorSave and observed using a Zeiss Axioskop microscope (Zeiss, Oberkochen, Germany).

Cells grown on glass coverslips were injected using a 5171 micromanipulator and a 5246 transjector (Eppendorf-Netheler-Hinz GmbH, Hamburg, Germany).

ID3 antibody used in microinjection experiments was diluted at a concentration of 4 mg/mL in PBS buffer. Before microinjection, the solution was clarified by centrifugation in a Beckman TLA-100 rotor (100,000 × g, 4°C, 10 min). The supernatant was then loaded into injection needles and injected into HeLa cells. After microinjection, cells were incubated for 4 h. They were then processed for immunofluorescence as described above.

Cells grown for 2 days on glass coverslips were permeabilized with warm OPT buffer [80 mmol/L PIPES, 1 mmol/L EGTA, 1 mmol/L MgCl₂, 0.5% Triton X-100, 10% glycerol (pH 6.8)]. OPT buffer permeabilizes cells while preserving intact microtubules (28). CPA (0.5 μg/mL) diluted in warm OPT buffer was added with or without inhibitory compounds (1 mmol/L). Cells were incubated for 15 min at 37°C and then fixed and processed for immunofluorescence as described above.

Screen for chemicals that decrease paclitaxel-induced tubulin detyrosination. The rationale of the assay relies on the substrate properties of TCP and TTL. TTL, the enzyme that catalyzes the tubulin tyrosination reaction, adds tyrosine to soluble, detyrosinated tubulin and does not react with tubulin once it is assembled into microtubules (3, 29–31). TCP is likely to prefer polymerized substrate, and its binding to microtubules may be required for its function (30, 32, 33). TCP acts slowly on microtubules, whereas TTL recharges Glu-tubulin quickly on release from microtubules (30). As a consequence of the substrate specificity of TTL and of the kinetic differences of TTL and TCP, Tyr-tubulin is the main component of dynamic microtubules, being the dominant tubulin variant in cycling cells in vitro, whereas Glu-tubulin is a marker of long-lived stable microtubules (3, 30, 34–36). Depending on the cell type, these Glu-microtubules either are not detectable or represent a small subpopulation of the cellular microtubules. In HeLa cells, these Glu-microtubules are almost absent (28). However, they do accumulate on paclitaxel treatment, indicating that TCP is active in these cells. We set up an assay amenable to automation, in which TCP inhibitors were selected according to their ability to inhibit accumulation of Glu-tubulin in HeLa cells with paclitaxel-stabilized microtubules. Briefly, compounds were added to HeLa cells grown in 96-well microplates. Cells were treated with paclitaxel immediately after addition of the compounds. After 2 h of incubation, cells were rinsed, fixed, and permeabilized with methanol. Glu-tubulin was then quantified by immunoluminescence using a specific antibody (6, 28). The Glu-tubulin signal obtained after treatment of the cells with paclitaxel alone was defined as 0% of inhibition, whereas the basal Glu-tubulin signal obtained after treatment with only DMSO was taken as 100% of inhibition. The different steps of the assay are illustrated in Fig. 1.

Screening of partially purified natural extracts. A total of 23,904 partially purified extracts, corresponding to 3,252 natural species (81.5% plants, 2.5% insects, and 16% marine organisms), was screened for potential TCP-inhibiting activity using the above described assay (Fig. 1). Among these, 268 extracts (1.1% hit rate) were scored as positive, showing >50% of inhibition. However, in such a cell-based assay, a lowered Glu-tubulin signal could result from a global toxic effect of the extracts on the cells. We thus assayed the extract cytotoxicity using a MTT assay. A subset of 127 extracts showed some cytotoxicity after 2 h of incubation on the cells and was discarded. Among the 141 remaining extracts, 49 extracts showed an apparent inhibitory activity superior to 70%. These most active extracts were set aside for further characterization.

Bioguided fractionation and identification of TCP inhibitors. The 49 extracts were submitted to fractionation by combining chromatographies (normal and reverse C18 phases), and each fraction was tested again using several different biological complementary assays (bioguided fractionation). The aim of these other assays was to discriminate between false- and true-positive compounds.

All fractions were first screened using tubulin immunofluorescence microscopy to examine their effects on microtubule structure (data not shown). Fractions able to depolymerize the microtubule network were classed as potential false positive. Indeed, such compounds could have competed with paclitaxel during the high-throughput screening assay and thus led to a diminished Glu-tubulin signal that does not result to TCP inhibition. Such depolymerizing fractions were discarded, and only fractions showing no detectable effect on the microtubule network were kept. These last fractions were checked for their inhibitory effect on paclitaxel-induced tubulin detyrosination using tubulin immunofluorescence microscopy (Fig. 2). When cells were only treated with DMSO (Fig. 2A), the Glu-microtubules represented a small subpopulation of cellular microtubules as expected. After paclitaxel treatment (Fig. 2B), Glu-microtubules did accumulate, reflecting TCP action on stabilized microtubules. This accumulation was prevented when cells were treated with the fractions selected by the screening strategy, together with paclitaxel (Fig. 2C), although paclitaxel still exerted its stabilizing effect as indicated by the presence of tyrosinated microtubule bundles.

Finally, the inhibitory effect of the remaining fractions was investigated with another assay of TCP activity using microinjection experiments (Fig. 3). In such an assay, TCP activity is assessed after blocking TTL activity in cells by microinjection of ID3, a monoclonal antibody directed against the active site of TTL (29, 37). The activity of TCP is revealed in injected cells by the resulting accumulation of Glu-tubulin (Fig. 3A). In the presence of a TCP-inhibiting fraction, such an accumulation does not occur. In that case, the level of detyrosinated tubulin (Fig. 3B) is as low in microinjected cells as in non-microinjected cells, which can be localized thanks to Hoechst 33258 staining of the nuclei.

The overall process allowed the selection of an extract from Vernonia perrottetii, and this led to the isolation of two compounds (V18 and V19). Their structures were established by analysis of their NMR data and comparison with published values. These compounds were structurally related sesquiterpene lactones (V18 and V19; Fig. 4). They showed a potent TCP inhibitory activity, as assessed by ID3 microinjection experiment, for a concentration of 10 μmol/L.

Preliminary structure-activity relationships. We tested the effect of 17 available structural analogues of the active compounds (Fig. 4), at a final concentration of 20 μmol/L, on TCP activity evaluated by ID3 microinjection experiments. Among these compounds, we found that only the known anti-inflammatory sesquiterpene lactone parthenolide, originally extracted from feverfew Tanacetum parthenium, and, to a much lesser extent, cnicin displayed a TCP inhibitory activity (Fig. 5D).

Parthenolide showed a similar TCP inhibitory activity than the two compounds isolated from V. perrottetii (Fig. 5A). This indicates that the C8 lateral chain, present on V18 and V19, is not necessary for TCP inhibitory activity.

There are strong indications that alkylation of cysteine(s) is a general mechanism for sesquiterpene lactones, which possess α,β-unsaturated carbonyl structures, such as α-methylene lactones (24). These structural elements preferably react with nucleophiles, especially sulfhydryl groups, by a Michael-type addition. To test the importance of such an unsaturated α, β-lactone, we synthesized dihydroparthenolide and assayed it for TCP inhibition. Dihydroparthenolide did not show any activity (Fig. 5B), indicating that the unsaturated α, β-lactone is in fact necessary for TCP inhibitory activity. However, we found that the presence of the unsaturated α, β-lactone, even if it is necessary, is not sufficient for TCP inhibition. For instance, costunolide showed no TCP inhibitory activity (Fig. 5C). Costunolide possesses an unsaturated α, β-lactone and differs from parthenolide by the lost of an epoxide function. This underlies the importance of the epoxide function in TCP inhibitory activity. The replacement of this epoxide function by an alcohol (cnicin) leads to a partial loss of activity (Fig. 5D).

We tested the cytotoxicity of parthenolide and of the compounds V18 and V19. HeLa cells were exposed to increasing concentrations of these molecules, and cytotoxicity was measured after 48 h using MTT assay. We found that parthenolide, V18, and V19 had similar cytotoxic properties with an IC₅₀ of 8, 3.5, and 7.1 μmol/L, respectively.

Does the observed parthenolide inhibition of TCP activity result from NF-κB inhibition? One well-described activity of parthenolide is its potent anti-inflammatory effect, which is mainly achieved through its inhibitory effects on NF-κB activation. Thus, the observed TCP inhibition by parthenolide could actually be an indirect result of NF-κB inhibition. Costunolide, an also known NF-κB inhibitor, has no effect on TCP activity, indicating that TCP inhibition does not result from NF-κB inhibition. However, this conclusion remains uncertain due to the different NF-κB inhibitory potencies of parthenolide and costunolide (24). To get insight into that question, we tested the effect of sulindac, an inhibitor of NF-κB activation (38, 39) that is not structurally related to parthenolide or costunolide, on TCP activity. We found that 1 mmol/L sulindac has no inhibitory effect on the generation of Glu-tubulin induced by ID3 microinjection (data not shown). This result strongly shows that parthenolide inhibition of TCP activity is independent of NF-κB inhibition.

Specific action of parthenolide on TCP. In vitro, CPA is able to excise the COOH-terminal tyrosine of tubulin as well as does TCP in cells (40). We tested if parthenolide could also inhibit CPA using a cell-based assay of CPA activity. In such an assay, the effect of the application of exogenous CPA on the cellular microtubule network is observed by the generation of Glu-tubulin in permeabilized cells. The buffer used for the permeabilization preserves microtubule network (28). In the absence of CPA, the microtubule network is predominantly composed of Tyr-tubulin, whereas Glu-tubulin is almost absent (Fig. 6A).

After an exposition of 15 min to CPA, the Glu-tubulin staining is largely enhanced, extending to the entire microtubule network (Fig. 6B). This augmentation is prevented when orthophenanthroline is added to cells in combination with CPA (Fig. 6C). Orthophenanthroline is a known CPA inhibitor with no effect on TCP (12). When cells are exposed with CPA in combination with parthenolide, the Glu-tubulin staining is the same as the Glu-tubulin staining observed with CPA alone (Fig. 6D). Although parthenolide is a reactive compound because of its unsaturated α, β-lactone, this result shows that it has no effect on CPA even at the highest concentration (1 mmol/L) tested.

By screening a natural extract library on a cell-based assay of TCP activity, we have isolated two structurally related sesquiterpene lactones that produce a strong inhibition on cellular TCP activity.

When searching for therapeutic agents, cell-based assays are particularly valuable because they select not only for activity against a particular molecular target but also for other desirable properties, such as the ability to permeate cells and to retain activity in tissue culture medium and in cells (23). Moreover, by screening for inhibitors on the entire cell, multiple potential targets, direct or indirect, are tested, allowing the cell to dictate the best target.

Although the screen that we set up could lead to the selection of some false positives, the overall strategy was discriminative and led to the isolation of only two compounds. This is noteworthy as regards the high number of compounds tested. We have screened extracts, which could be considered as mixtures of natural substances. Thus, the true number of different compounds tested is much higher than 23,904 (i.e., the number of extracts tested).

We tested several commercially available analogues of the compounds that were isolated by the screening procedure and found that parthenolide, but not the closely related compound costunolide, has also a potent inhibitory effect on TCP. Parthenolide, as well as other sesquiterpene lactones, contains an α-methylene-β-lactone moiety, which is highly reactive with cellular thiols, resulting in alkylation of sulfydryl residues through Michael-type addition. We found that this α-methylene-β-lactone moiety was indispensable for the inhibitory effect of parthenolide on TCP. If the observed inhibitory effect of parthenolide is the consequence of a direct reaction of parthenolide with TCP, this result indicates that TCP could belong to the class of cysteine proteases. Moreover, as parthenolide has no effect on CPA and as known metalloproteinases inhibitors have no effect on TCP (12), it seems that TCP is not a metalloproteinase.

Numerous bioactivities, probably as a consequence of its high reactivity, have been described for parthenolide and other related sesquiterpene lactones. These bioactivities include, for example, anti-inflammatory effect through inhibition of NF-κB activation (41) and signal transducers and activators of transcription signaling pathway (42) or induction of apoptosis through several different mechanisms, such as caspase activation and mitochondrial dysfunction (43, 44). Contrary to the panel of actions described for sesquiterpene lactones, TCP inhibition looks to be a specific action of parthenolide and of the two Vernonia compounds selected.

We have shown that the inhibitory effect of parthenolide on TCP does not seem to result from the parthenolide action on NF-κB pathway. However, we cannot exclude that TCP inhibition is an indirect consequence of an action of parthenolide on another known or unknown target. For example, it cannot be excluded that parthenolide inhibits regulating enzymes, such as kinases or phosphatases. It has been shown that inhibitors of serine/threonine phosphatases seem to dissociate TCP from microtubules (45). Such an inhibition could thus represent an indirect form of TCP inhibition.

Our initial aim was to use the inhibitor selected by the screening procedure as a hook, in an attempt to isolate TCP, as it has been successfully done in some chemical genetic approaches. However, because of its high reactivity and its numerous targets, parthenolide is no suitable for such a strategy. We are currently testing other chemical libraries in an attempt to isolate a compound more convenient for TCP purification.

The anticancer potential of parthenolide has been described both in vitro (43, 46–49) and on animal models (44, 50). Moreover, it has been shown that parthenolide could reduce metastasis (50). We believe that TCP inhibition, which could antagonize the abnormal accumulation of detyrosinated tubulin in tumor cells, could be one of the mechanisms underlying the anticancer and antimetastatic properties of parthenolide. Because parthenolide has shown safety in phase I clinical trials, our results have direct clinical relevance.

Grant support: Alliance des Recherches sur le Cancer, an Association pour la Recherche sur le Cancer initiative, and Rhône-Alpes Region. X. Fonrose is a recipient of a fellowship from Association pour la Recherche sur le Cancer.

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.

We thank Dr. Jacques Fahy (Institut de Recherche Pierre Fabre, Boulogne, France) for parthenolide hydrogenation experiment, Delphine Scokaert for her contribution in the initial step of this project, Catherine Pillet for technical assistance, and Muriel Batut and Stephanie Brillant for the isolation of the compounds from Vernonia.