Pharmacologic Abrogation of the Mitotic Spindle Checkpoint by an Indolocarbazole Discovered by Cellular Screening Efficiently Kills Cancer Cells

Antimitotic drugs, including various taxanes and Vinca alkaloids, are frequently used for anticancer treatment (1). It is well established that these drugs activate the spindle assembly checkpoint, a signal transduction pathway that monitors the proper alignment of chromosomes during mitosis (2, 3). The drug-induced and spindle checkpoint–mediated mitotic arrest is followed by the induction of apoptosis, which occurs either in mitosis or in the postmitotic G₁ phase on escape from a prolonged mitotic arrest (4). The latter decision might depend on two separate pathways, which control the activation of caspases and a slow but continuous degradation of cyclin B during mitosis, respectively (5, 6). The drug-induced activation of caspases during mitosis might be a key event required for efficient tumor cell killing (6, 7), but the mechanisms leading to caspase activation remain elusive. Interestingly, the mitotic spindle checkpoint is required for the efficient induction of apoptosis on antimitotic drug treatment, regardless of whether a cell dies in mitosis or in the subsequent G₁ phase. Thus, compromising the mitotic spindle checkpoint results in a significant decrease in sensitivity of tumor cells toward antimitotic drugs (8–12). These findings were recently supported by an unbiased small interfering RNA (siRNA) screen that identified spindle checkpoint proteins as major determinants for the sensitivity to paclitaxel (13). The molecular basis for a requirement of the spindle checkpoint in the activation of the apoptotic machinery is unclear, but it might be related to inhibition of Akt during a mitotic arrest (13) or to a dephosphorylation and activation of caspase-9 that requires a prolonged duration of mitosis (14).

The spindle checkpoint pathway requires the function of various proteins, including the kinases Bub1, BubR1, Mps1, Mad1, Mad2, Bub3, and several others, which are recruited to kinetochores on activation of the checkpoint (2, 3). Furthermore, inhibition of the centromeric Aurora-B kinase or overexpression of the centrosomal Aurora-A kinase led to spindle checkpoint malfunction (15–17). Activation of the spindle checkpoint in response to misaligned chromosomes either in the early phases of a regular mitosis or chronically on antimitotic drug treatment causes the inhibition of the ubiquitin ligase activity within the anaphase-promoting complex/cyclosome (APC/C), which is required for the degradation of various mitotic substrates, including securin and cyclin B. Accordingly, failure of the spindle checkpoint results in a premature separation of sister chromatids and an unscheduled exit from mitosis, which leads to the generation of aneuploid progenitors (2, 3).

In human cancer, a partial dysfunction of the spindle checkpoint (15, 18–22) is a frequent event and might contribute to tumorigenesis (18, 23–28). In fact, mice heterozygous and hypomorphic for spindle checkpoint genes show gross aneuploidy and carcinogen-induced or spontaneous tumorigenesis (20, 29). Importantly, the analyses of homozygous knockout mice as well as siRNA studies in human cells revealed that the spindle checkpoint is essential for cell viability (30–33). The essential nature of the spindle checkpoint raises the interesting idea to use the spindle checkpoint as a novel target for anticancer therapy.

To test this concept, we performed a cell-based screen and identified the indolocarbazole Gö6976 that inhibits the mitotic spindle checkpoint by restraining the phosphorylation and kinetochore localization of several spindle checkpoint proteins. Interestingly, we identified the mitotic Aurora-A and Aurora-B kinases as targets for Gö6976 in vitro and in vivo and show that the pharmacologic inactivation of the spindle checkpoint efficiently kills human cancer cells.

Cell culture and treatments. All cell lines were maintained under standard culture conditions (11) and treated with 300 nmol/L nocodazole (Sigma), 100 nmol/L Taxol (Sigma), or 68 μmol/L monastrol (Biomol), 200 μmol/L ALLN (Calbiochem), 20 μmol/L MG132 (Calbiochem), 5 to 100 μmol/L roscovitine (Sigma), 0.2 to 20 μmol/L Gö6976 (Calbiochem), 0.05 to 2 μmol/L ZM447439 (Tocris), 50 μmol/L H89 (Calbiochem), 100 μmol/L protein kinase A (PKA) inhibitor (Calbiochem), 1 μmol/L Ro31-8220 (Calbiochem), 1 μmol/L Ro32-0432 (Calbiochem), 40 μmol/L rottlerin (Calbiochem), 50 μmol/L PD98059 (Calbiochem), 10 μmol/L U0126 (Calbiochem), 30 μmol/L SB202190 (Calbiochem), 30 μmol/L SB203580 (Calbiochem), 50 nmol/L staurosporine (Sigma), 20 μmol/L LY294002 (Sigma), 0.1 to 2 μmol/L UCN-01 (a gift from the Developmental Therapeutics Program of the National Cancer Institute), and 10 μmol/L Raf inhibitor (BAY 43-9006). Cells were synchronized in G₁-S by a double-thymidine block using 2 mmol/L thymidine (Sigma).

Flow cytometry and determination of the mitotic index. The DNA content and the mitotic index were determined as described (11).

Western blotting. Cell lysates were prepared, and SDS-PAGE, semidry Western blotting, and detection were performed as described (11) using the following antibodies: Bub1 (Bethyl; Stephen Taylor, University of Manchester, Manchester, United Kingdom), BubR1 (Chemicon; Stephen Taylor), Bub3 (BD Biosciences), Mps1 (Santa Cruz Biotechnology), Plk1 (Santa Cruz Biotechnology), Aurora-A (Santa Cruz Biotechnology; pT288: Cell Signaling), Aurora-B (BD Biosciences; pT232: Cell Signaling), cyclin B (Santa Cruz Biotechnology), tubulin (Sigma), actin (Sigma), securin (NeoMarkers), histone H3 (pS-10; Cell Signaling), and poly(ADP-ribose) polymerase (PARP; Pharmingen). Signals were quantitated using the ImageJ software (NIH).

Chromosome spread analysis. Premature sister chromatid separation and karyotype analysis were performed as described (11).

Microscopy. Cells were fixed/permeabilized in 2% paraformaldehyde/methanol and incubation with antibodies was carried out for 2 h. Images were taken with a Leica DM6000B microscope and a charge-coupled device camera (Orca-ER, Hamamatsu) with a Z-optical spacing of 0.2 μm. Images were deconvolved using the Leica LAS-AF software and maximum projection images are shown. Pixel quantitations of CREST and Bub1/BubR1 signals on kinetochores were performed using the Leica LAS-AF software.

Kinase assays. In vitro kinase assays using recombinant Aurora kinases (34) were performed for 30 min at 30°C in kinase buffer [50 mmol/L Tris-HCl (pH 7.7), 100 mmol/L KCl, 5 mmol/L MgCl₂, 1 mmol/L DTT] in the presence of 50 to 2,000 μmol/L of ATP (Sigma), 1 μCi [³²P]ATP (6,000 Ci/mmol; Amersham), and 7 μg MBP (Sigma). Substrate phosphorylation was visualized by autoradiography and quantitated using a phosphoimager (Fuji). For immunoprecipitation kinase assays, Aurora-A or Aurora-B was immunoprecipitated from lysates and subjected to in vitro kinase assays.

Apoptosis assays. Caspase activity assays, detection of PARP cleavage, and chromosomal DNA fragmentation assays as well as determination of the sub-G₁ DNA content were performed as described previously (11). For caspase inhibition, the pan-caspase inhibitor Boc-D-FMK (100 μmol/L; Calbiochem) was used.

Identification of the indolocarbazole compound Gö6976 as a potent abrogator of a spindle damage–induced mitotic arrest. We performed a cell-based screen, in which HCT116 cells were first treated with nocodazole to arrest cells in mitosis in a spindle checkpoint–dependent manner. Subsequently, cells were treated with small-molecule inhibitors followed by determination of the mitotic index to monitor the mitotic arrest (Fig. 1A). Because several kinases are involved in spindle checkpoint signaling, we screened commercially available protein kinase 1inhibitors, including established inhibitors for PKA, protein kinase C (PKC), cyclin-dependent kinase (CDK), Chk1, mitogen-activated protein/extracellular signal-regulated kinase kinase, p38, Raf, and phosphatidylinositol 3-kinase (Fig. 1B). Among the small-molecule inhibitors tested, we identified the indolocarbazole Gö6976 (Supplementary Fig. S1A) as a compound that potently abrogated the induced mitotic arrest and forced cells to exit mitosis with a 4N DNA content (Fig. 1B and C). In HCT116 and HeLa cells, Gö6976 showed a half-maximal effect at 0.8 to 1.2 μmol/L in the presence of nocodazole and Taxol, which is ∼25 times more potent than the CDK1 inhibitor roscovitine in abrogating the mitotic arrest (Fig. 1D). Very similar results were obtained after treatment with the KSP/Eg5 inhibitor monastrol and in various human cell lines (Supplementary Fig. S1B and C). We also tested established structurally related and similarly potent PKC inhibitors (Ro31-8220 and Ro32-0432) as well as the structurally unrelated PKC inhibitor rottlerin and found no effect on the spindle damage–induced mitotic arrest, suggesting that Gö6976 acts independently of PKC inhibition. In addition, other indolocarbazole compounds, including UCN-01 or staurosporine, had no or little effects on the spindle damage–induced mitotic arrest (Fig. 1B). No significant inhibition of CDK1 by Gö6976 was observed (Supplementary Fig. S1D).

Gö6976 inhibits the mitotic spindle checkpoint. In the presence of nocodazole or Taxol, treatment with 2 μmol/L Gö6976 induced a rapid exit from mitosis, which was associated with the rapid degradation of cyclin B and securin that could be blocked by the proteasome inhibitor ALLN, indicating that Gö6976 relieves the inhibition of the APC/C (Fig. 2A). Gö6976 allowed the entry into mitosis but suppressed the accumulation of mitotic cells in the presence of nocodazole, whereas roscovitine treatment prevented the entry into mitosis (Supplementary Fig. S2A). Furthermore, premature exit from mitosis led to endoreduplication (Supplementary Fig. S2B), which is associated with spindle checkpoint impairment as shown previously (35). Gö6976 treatment forced cells out of mitosis and a single nucleus with a 4N DNA content was reformed, whereas CDK inhibitors caused a high rate of multinucleation (Supplementary Fig. S2C), supporting the notion that Gö6976 acts independently from CDK1.

Consistent with a spindle checkpoint inhibition by Gö6976, we found a high rate of premature chromatid separation on Gö6976 treatment (Fig. 2B). Prolonged treatment led to a significant increase in aneuploid cells, whereas the overall cell cycle progression was not significantly disturbed (Supplementary Fig. S2D and E), although dead cells became increasingly apparent.

Gö6976 inhibits the proper phosphorylation and localization of spindle checkpoint proteins. We investigated whether Gö6976 interferes with the localization or amount and phosphorylation of spindle checkpoint proteins. Significantly, Gö6976 treatment reduced the kinetochore localization of Bub1 and BubR1 by ∼80% (Fig. 2C). Aurora-B was also found to be mislocalized from the centromere region to chromosome arms on Gö6976 treatment (Supplementary Fig. S3), whereas Aurora-A retained its centrosomal localization (data not shown). Furthermore, on Western blots, we found that the hyperphosphorylation of Bub1, Mps1, Aurora-A, and Aurora-B, which is closely associated with spindle checkpoint activation, was abolished on Gö6976 treatment both on forced exit from mitosis and in mitotic cells (Fig. 2D).

Gö6976 causes aberrant spindle morphology and chromosome alignment defects. Next, we investigated the mitotic progression in the presence of Gö6976. Treatment of asynchronously growing HCT116 cells with 2 μmol/L Gö6976 resulted in a transient accumulation of mitotic cells (Fig. 3A) and this was associated with profound alterations in the spindle structure and severely misaligned chromosomes (Fig. 3B). The spindles often seemed cylindrical and flattened and groups of chromosomes were frequently found near the spindle poles, suggesting that those chromosomes are not properly attached to microtubules. Similar chromosome alignment defects were observed when cells were released from a monastrol block into Gö6976 (Supplementary Fig. S4A).

Remarkably, the severe alignment defects induced by Gö6976 did not cause a robust mitotic arrest as would be expected in the presence of a functional spindle checkpoint. Instead, the accumulation of mitotic cells was only weak and transient (Fig. 3A) but dependent on the spindle checkpoint (Supplementary Fig. S4B), suggesting that the mitotic arrest cannot be maintained, possibly mediated by Gö6976-induced spindle checkpoint inhibition.

Gö6976 accelerates exit from an aberrant mitosis. To test whether Gö6976 treatment indeed allows an exit from mitosis in the presence of unaligned chromosomes, we released cells from a nocodazole block in the presence or absence of Gö6976 and followed the progression through mitosis qualitatively by immunofluorescence (Fig. 3C) and quantitatively by monitoring the mitotic index (Fig. 3D). Whereas control cells progressed normally through mitosis and exited mitosis properly, Gö6976-treated cells exhibited a significant acceleration of mitotic progression and exited from mitosis in the presence of misaligned chromosomes. Most Gö6976-treated cells underwent cytokinesis despite the presence of unaligned and missegregated chromosomes, producing daughter cells with highly unequal DNA content (Fig. 3D,, top). By the end of mitosis, apoptotic cells became readily apparent, suggesting that the severe missegregation observed causes cell death (Fig. 3C).

The mitotic Aurora-A and Aurora-B kinases are inhibited by Gö6976 in vitro and in vivo. To gain insights into the mitotic target kinase of Gö6976, we evaluated candidate mitotic kinases known to be involved in spindle checkpoint function. Because we observed a lack of phosphorylation of Bub1, Mps1, Aurora-A, and Aurora-B (Fig. 2D), which might relate to their functional inactivation, we evaluated these kinases in particular. In vitro, no inhibition of Bub1, Mps1, Plk1, and p38 by Gö6976 was found. However, Aurora-B kinase activity was significantly inhibited by Gö6976 in these assays (Fig. 4A). We determined the IC₅₀ values for Gö6976 and ZM447439, which is a potent and well-characterized inhibitor for Aurora-A and Aurora-B kinases in vitro (36), and found that Gö6976 inhibited Aurora-B with similar potency as ZM447439 (100 versus 80 nmol/L at 50 μmol/L ATP; Fig. 4B). However, our analyses revealed that Gö6976 was nearly five times more potent than ZM447439 in inhibiting the Aurora-A kinase (120 versus 580 nmol/L at 50 μmol/L ATP; Fig. 4B). As expected for an ATP-competitive inhibitor, the inhibitory activity of Gö6976 decreased with higher ATP concentrations (Supplementary Fig. S5A; Supplementary Text). To test whether Gö6976 also inhibits the Aurora kinases in vivo, we evaluated the activation loop–specific phosphorylation of Aurora-A and Aurora-B kinases (pT288 for Aurora-A and pT232 for Aurora-B), which is indicative of their activation. In agreement with previous data (36), treatment of mitotic cells with ZM447439 abolished selectively the activation of cellular Aurora-B (75% inhibition) but left the activity of Aurora-A unchanged. However, in agreement with our in vitro data, Gö6976 treatment inhibited the activity of both Aurora-A and Aurora-B, with similar potency in vivo (60% and 58% inhibition, respectively; Fig. 4C; Supplementary Fig. S5B).

We also determined the activity of Aurora-A and Aurora-B immunoprecipitated from drug-treated cells. Again, Gö6976 treatment caused significant inhibition of Aurora-A and Aurora-B (Fig. 4D), whereas immunoprecipitated CDK1-cyclin B complexes were not inhibited by Gö6976 (Supplementary Fig. S5C). These results provide strong evidence that Gö6976 is a potent inhibitor of Aurora-A and Aurora-B in vitro and in vivo, although we cannot exclude the possibility that Gö6976 targets additional kinases during mitosis.

In addition to its crucial function in chromosome alignment and spindle checkpoint function, Aurora-B is also required for cytokinesis (16). Indeed, treatment of asynchronously growing cells with ZM447439 prevents cytokinesis and results in profound polyploidization. Surprisingly, although Gö6976 inhibits Aurora-B in vivo, neither profound defects in cytokinesis nor polyploidization was detected (Supplementary Fig. S6).

The aberrant progression through mitosis induced by Gö6976 causes apoptosis in tumor cells. To test whether the Gö6976-induced aberrant progression of mitosis causes apoptosis, we synchronized HCT116 cells at G₁-S, treated the cells with Gö6976 before entry into mitosis, and evaluated the activation of caspase-3 in intact cells. Indeed, Gö6976 treatment induced caspase-3 activity, PARP cleavage, and DNA fragmentation after progression through a single mitosis and this was significantly suppressed when entry into mitosis was prevented by inhibiting CDK1 (Fig. 5A; Supplementary Fig. S7A). Further, Gö6976-induced apoptosis was caspase dependent (Fig. 5B).

We treated 15 human cell lines with Gö6976 and found a significant but varying induction of caspase activity in most cancer cell lines (Fig. 5C), whereas nontransformed human cells (human BJ-hTert fibroblasts and primary human umbilical vein endothelial cells) exhibited low caspase-3 activity. The reason for the different sensitivities observed is not known at present but could be recapitulated in colony formation assays comparing the sensitivity of HeLa and BJ-hTert fibroblasts toward Gö6976 (Supplementary Fig. S7B; Supplementary Text). Accordingly, we found that HeLa and HCT116 cancer cells showed a higher rate of cell killing than BJ-hTert fibroblasts after treatment with Gö6976 (Supplementary Fig. S7C and D). We also compared the efficacy of Gö6976 with ZM447439 and Taxol and found a higher efficacy of Taxol but a lower efficacy of ZM447439 toward HeLa and HCT116 cells (Supplementary Fig. S7D and E).

Gö6976 acts synergistically with Taxol. Recent data showed that the efficacy of antimitotic drugs could be greatly enhanced by subsequent inactivation of the spindle checkpoint (7, 37). To investigate whether Taxol and a Gö6976-mediated inactivation of the spindle checkpoint act synergistically, cells were treated for 14 hours with Taxol to arrest cells in mitosis followed by treatment with Gö6976 to induce exit from mitosis. In fact, a synergistic effect of Gö6976 and Taxol was clearly observed (Fig. 5D and E). It is important to note that a reversed order of drug treatment caused an inhibition of the spindle checkpoint before Taxol treatment and reduced sensitivity toward Taxol (data not shown; Supplementary Fig. S2B; refs. 11, 35, 38).

A spindle checkpoint override is required for efficacy of Gö6976. To investigate if the efficacy of Gö6976 requires the override of the mitotic spindle checkpoint, we took advantage of a human colon carcinoma cell line, HT29, which exhibits a very strong and prolonged mitotic arrest in response to spindle damage compared with most cancer cell lines, including HCT116 cells (Supplementary Fig. S8). Accordingly, Gö6976 could only weakly override the spindle checkpoint in HT29 cells (Fig. 6A), which was directly associated with resistance toward Gö6976 (Fig. 6B), suggesting that the spindle checkpoint override is directly associated with the sensitivity toward Gö6976. It is important to note that HT29 cells are not resistant per se because they are capable of inducing apoptosis on chemotherapeutic treatment (data not shown).

Targeting of spindle checkpoint–compromised cancer cells by Gö6976. Several reports have indicated that the spindle checkpoint is required for the efficient induction of cell death after treatment with antimitotic drugs (8, 9, 11, 12, 39). Accordingly, spindle checkpoint–compromised HCT116 cells with lowered expression of MAD1 or MAD2 showed a lowered mitotic arrest and reduced caspase-3 activity in response to nocodazole treatment (Fig. 6C; ref. 11). In both spindle checkpoint–impaired cell lines, Gö6976 treatment further lowered the mitotic arrest in response to spindle damage (Fig. 6D) and induced caspase-3 activity similar to parental cells (Fig. 6E), suggesting that spindle checkpoint–compromised cancer cells are sensitive toward Gö6976 treatment.

Due to its essential nature, the spindle checkpoint has been suggested as a potential novel anticancer drug target (32, 33). We wished to prove this conceptual idea and have identified the indolocarbazole Gö6976 as an inhibitor of the spindle checkpoint. Importantly, we show that this inhibitor induces a highly aberrant and accelerated mitosis that results in efficient killing of human cancer cells. Thus, we propose to add the spindle checkpoint to the growing list of attractive mitotic drug targets.

The “classic” mitotic drug target is the mitotic spindle, whose function is inhibited by microtubule-targeting drugs including various taxanes or Vinca alkaloids (40). However, due to the ability to inhibit microtubule function in interphase or even in differentiated cells, a plethora of unwanted side effects, including neutropenia and neuropathy, is observed after treatment with these drugs. Therefore, novel mitotic drug targets that function exclusively in mitosis, including the Plk1 and the Aurora kinases, but also mitotic Eg5/KSP or Cenp-E kinesins, have gained considerable interest (1, 41). Similarly, no function for the mitotic spindle checkpoint pathway outside of mitosis has been described thus far. Thus, inhibition of the spindle checkpoint is expected to have no effect on interphase or differentiated cells. However, it is important to note that targeting mitosis per se might harbor the risk of chromosome missegregation in surviving cells that could contribute to de novo tumorigenesis.

At present, is not clear how antimitotic drugs induce tumor cell death. Most recent results have indicated that cells activate caspase-dependent pathways during mitosis and die either in mitosis or on an unscheduled exit from mitosis (5–7). However, it is unknown how caspases are activated on drug treatment, but the activation of the spindle checkpoint associated with a transient mitotic delay seems to be necessary for the efficient induction of cell death (8–13). Interestingly, during mitosis, caspase-9 is inhibited by CDK1/cyclin B–mediated phosphorylation. Prolonged mitotic arrest, which requires the spindle checkpoint, may lead to a progressive dephosphorylation, activation of caspase-9, and the induction of apoptosis, thereby providing a timer mechanism monitoring the length of mitosis (14). In contrast to this mechanism, we show here when targeting the spindle checkpoint pathway that the induction of apoptosis is not dependent on a functional spindle checkpoint. Similarly, because Aurora-B is required for the spindle checkpoint, ZM447439 also induces spindle checkpoint–independent apoptosis.³

We show that Gö6976 targets the mitotic Aurora kinases, which might explain the phenotypes observed after treatment with Gö6976. Gö6976 treatment causes spindle formation defects and severe chromosome misalignment, phenotypes attributable to Aurora-A inhibition (42). In addition, inhibition of Aurora-B might explain the checkpoint override in the presence of misaligned chromosomes because Aurora-B has been implicated in spindle checkpoint function (36, 43). Thus, the efficacy of Gö6976 might depend on a two-step mechanism: first, the induction of mitotic damage associated with a transient mitotic delay, and second, the override of the spindle checkpoint allowing exit from an aberrant mitosis. The latter might be particularly supportive for the induction of apoptosis. In fact, when tumor cells are first treated with Taxol followed by the inactivation of the spindle checkpoint, the induction of cell death is greatly enhanced (Fig. 5; refs. 7, 37). Interestingly, the reverse order of events, Taxol treatment of cells in which the spindle checkpoint is already inhibited, led to endoreduplication and a reduced sensitivity toward Taxol³ (13, 35, 38), suggesting that a therapeutic combination of Taxol and spindle checkpoint inhibitors strictly depends on the order of drug treatment.

Although we show that Gö6976 inhibits Aurora-B in vitro and in vivo, we found some remarkable differences to the inhibition by selective Aurora-B inhibitors, including ZM447439 or hesperadin (36, 43). Whereas ZM447439 and hesperadin inhibit selectively a Taxol-induced spindle checkpoint, Gö6976 overrides the spindle checkpoint on Taxol or nocodazole treatment. Moreover, in contrast to ZM447439 or hesperadin, Gö6976 did not cause defects in cytokinesis or induce polyploidization. At present, we cannot explain these interesting differences, but they might be related to the concomitant inhibition of Aurora-A by Gö6976 or due to different inhibitor selectivity for distinct Aurora-B–containing protein complexes.

Previous work has identified the c-Jun NH₂-terminal kinase inhibitor SP600125 as an inhibitor of the spindle checkpoint kinase Mps1 (44). Remarkably, SP600125-mediated inhibition of Mps1 led to an inactivation of the spindle checkpoint in cancer cells but not in nontransformed fibroblasts. Although we have not investigated the spindle checkpoint activity in nontransformed cells, we found that Gö6976 killed many cancer cells more efficiently than nontransformed human cells. The molecular basis for this selectivity is not clear at present but deserves further detailed investigation in future studies.

No potential conflicts of interest were disclosed.

Grant support: University Medical Center Giessen and Marburg and Deutsche Forschungsgemeinschaft Heisenberg fellowship (H. Bastians) and Behring-Röntgen-Stiftung.

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 Stephen Taylor, Paolo Sassone-Corsi, Bert Vogelstein, Robert Benezra, Qi Min Zhan, Stefan Dimitrov, Stefan Gaubatz, Martin Eilers, Matthias Dobbelstein, and Mathias Schmidt for providing antibodies, cell lines, and plasmids; Rolf Müller for support; Katja Scheffler for experimental help; the Developmental Therapeutics Program of the National Cancer Institute for the gift of UCN-01; Gary Gorbski for discussions; and Heike Krebber and all members of the Bastians lab for comments on the manuscript.