During mitosis, a feedback mechanism called the spindle or mitotic checkpoint ensures that replicated chromosomes are segregated equally between the two daughter cells before division (reviewed in Shah and Cleveland, 2000 ). The mitotic checkpoint pathway is regulated by a group of evolutionarily conserved genes that include MAD1, MAD2, MAD3 (or BUBR1), BUB1, BUB3, and MPS1 (Burke, 2000
). These proteins preferentially bind to the kinetochores of unattached chromosomes where they are thought to generate the signal that suppresses anaphase onset (Burke, 2000
; Shah and Cleveland, 2000
). Currently, there are two ways in which the mitotic checkpoint can be activated experimentally. First, cells can be treated with chemicals that perturb microtubule dynamics such as nocodazole, taxol, vincristine, and vinblastine (Li and Benezra, 1996
; Sorger et al., 1997
). By either depolymerizing (nocodazole, vincristine, and vinblastine) or stabilizing (taxol) microtubules, these agents activate the mitotic checkpoint and induce a mitotic cell cycle arrest (Sorger et al., 1997
). Second, overexpression of the MAD (mitotic arrest deficient) checkpoint proteins has been reported to activate the spindle checkpoint and cause mitotic arrest (Fang et al., 1999
; Geley et al., 2001
). Regardless of the method of activation, a common feature of the checkpoint-arrested cells is a tendency to undergo apoptotic cell death. For example, many agents used to activate the mitotic checkpoint experimentally are also used clinically in the treatment of cancer (Rowinsky and Donehower, 1991
; Wilson and Jordan, 1995
) and induce apoptosis in the mitotically arrested cells (Woods et al., 1995
; Jordan et al., 1996
). Therefore, the available evidence links activation of the mitotic checkpoint with apoptosis, although the biochemical mechanisms and signaling pathways that underlie the toxicity of chemotherapeutic drugs are not clearly understood.
Studies in mammalian cells suggest that members of the mitogen-activated protein kinase (MAPK) family and the p21-activated kinase (PAK) family are involved in the regulation of cell survival and cell death (Xia et al., 1995 ; Verheij et al., 1996
; Kummer et al., 1997
; Potapova et al., 1997
; Aoshiba et al., 1999
; Bueno et al., 2000
; Communal et al., 2000
; Kurokawa et al., 2000
; Remacle-Bonnet et al., 2000
; Schurmann et al., 2000
; Tang et al., 2000
; Deschesnes et al., 2001
; Gnesutta et al., 2001
; Jakobi et al., 2001
). MAPKs comprise a family of serine/threonine protein kinases that function as critical mediators of signal transduction (Kyriakis and Avruch, 2001
) and include the extracellular signal-regulated kinases (ERKs), the c-Jun NH2-terminal kinases (JNKs), and the p38 MAPKs. The ERKs are activated in response to mitogen or growth factor stimulation, whereas the JNKs and p38 MAPKs are activated by proinflammatory cytokines and a variety of cellular stresses, including UV light, hyperosmolarity, heat shock, and microtubule disrupting drugs (Wang et al., 1998
, 2000
; Yujiri et al., 1999
; Kyriakis and Avruch, 2001
; Okano and Rustgi, 2001
; McDaid and Horwitz, 2001
; Stadheim et al., 2001
). The PAKs are a group of serine/threonine protein kinases that are directly activated by the GTPases Rac and Cdc42. Together with Ras, these GTPases also activate mitogen-activated protein (MAP) kinase pathways and regulate diverse cellular processes such as cell morphology, motility, transformation, and apoptosis (Bagrodia and Cerione, 1999
).
In the current study, we have examined the function of p38 MAPK in mitotically arrested HeLa cells. We show that antimicrotubule drugs cause concomitant activation of a p38 MAPK-mediated proapoptotic signaling pathway and a PAK-mediated prosurvival signaling pathway in the mitotically arrested cells. p38 MAPK stimulates apoptosis in mitotically arrested cells by inducing translocation of the proapoptotic protein Bax from the cytoplasm to the mitochondria, whereas PAK opposes p38 MAPK-induced cell death by phosphorylating the proapoptotic protein Bad.
![]() | Figure 1. Nocodazole activates the mitotic checkpoint and causes mitotic cell cycle arrest. (A) HeLa cells were transfected with HA-MAD2 as described in MATERIALS AND METHODS. After 24 h, the cell were treated with nocodazole (3 μM) for 12 h and the (more ...) |
![]() | Figure 2. Activation of p38 MAPK, but not JNK or Erk, occurs during mitotic arrest but not during normal mitosis. (A) Exponentially growing HeLa cells were treated with either dimethyl sulfoxide alone (asynchronous population [Asyn]) or with nocodazole (3 μM) (more ...) |
We evaluated the activity of the MAPKs in the nocodazole-treated cells by using immunocomplex kinase assays. p38 MAPK was activated primarily in the mitotically arrested population (Figure 2A). Immunoblot analysis of p38α immunoprecipitates indicated that the activation of p38 MAPK was not due to variations in the level of immunoprecipitated p38 protein. In contrast to p38 MAPK, both JNK and Erk MAP kinases were activated primarily in the attached cells (Figure 2, A and B). The activation of MAP kinase family members was evaluated further using alternative drugs that perturb microtubule dynamics and cause mitotic arrest. As observed with nocodazole, treatment of HeLa cells with either taxol (1 μM), vincristine (1 μM), or vinblastine (1 μM) caused mitotic arrest (as shown by high cdk1 activity; Figure 2C) and activated p38 MAPK selectively in the mitotically arrested population. With the exception of taxol, which activated JNK equally in both the mitotic and the attached cells, vincristine, vinblastine, and nocodazole activated JNK preferentially in the attached cells (Figure 2C). All four drugs also activated the Erk MAP kinases preferentially in the attached cell population (Figure 2C). Importantly, treatment of HeLa cells with cytochalasin D (5 μM) or EDTA (0.5 mM) to induce cell rounding or detachment did not activate p38 but did stimulate JNK activity (our unpublished data). Furthermore, p38 MAPK was selectively activated in mitotic mouse NIH3T3 cells arrested with either nocodazole, taxol, vincristine, and vinblastine (our unpublished data). These results indicate that p38 MAPK is specifically activated in mammalian cells mitotically arrested by microtubule-interfering drugs.
We also evaluated whether p38 MAPK was activated in a synchronized population of mitotic HeLa cells. In normal mitotic cells (Mit), containing high cdk1 activity, p38 MAPK activity was not detected compared with cells arrested in mitosis with nocodazole (Noc; Figure 2D). To further confirm that p38 MAPK activation was not necessary for normal mitosis, we used the p38 MAPK inhibitors SB203580 and SB202190 (Lee et al., 1994 ). Addition of either SB203580 (20 μM) or SB202190 (10 μM) to HeLa cells 3 h after release from an aphidicolin-thymidine double block had no effect on either entry into or exit from mitosis compared with vehicle-treated control cells (our unpublished data). These results suggest that p38 MAPK is unlikely to be activated during normal mitosis.
Because p38 MAPK is activated only in the mitotically arrested cells, we performed a time-course analysis of p38 activation at intervals after the addition of nocodazole (3 μM) to asynchronous HeLa cells. JNK activity was also assayed in parallel. At 6 h after nocodazole addition, when the majority of the cells was still attached (>90%), p38 MAPK activity was low, whereas JNK activity was maximal (Figure 3, A and B). From 12 h onward, we were able to collect purely mitotic cells by shake-off as assessed by high cdk1 activity. The activation of p38 MAPK correlated closely with that of cdk1, whereas JNK activity had returned to near basal levels 12 h after nocodazole addition (Figure 3, A and B). Furthermore, removal of nocodazole from the mitotically arrested cells resulted in exit from mitosis, as shown by the inactivation of cdk1. This inactivation of cdk1 correlated with the inactivation of p38 (Figure 3, C and D), providing further evidence that p38 MAPK activation is specifically associated with mitotic arrest.
![]() | Figure 3. p38 MAPK activation and inactivation correlates with cdk1 activation and inactivation respectively. (A) Exponentially growing HeLa cells were treated with nocodazole (3 μM) and cell extracts prepared at the indicated time points after nocodazole (more ...) |
![]() | Figure 4. PAK activation correlates with the activation of p38 MAPK in mitotically arrested HeLa cells but PAK does not activate p38 MAPK. (A) Exponentially growing HeLa cells were treated with nocodazole (3 μM) and cell extracts prepared, after separation (more ...) |
To determine whether PAK is an upstream activator of p38 MAPK in HeLa cells we cotransfected either wild-type (WT) or dominant active (DA) PAK (L106F) with WT p38α. DAPAK, in comparison with WT PAK, was highly active as assessed by its ability to phosphorylate GST-Mekk2 but was unable to activate cotransfected p38α, as measured GST-ATF2 phosphorylation (Figure 4D). The transfected p38α was, however, activated by anisomycin as assessed by phosphorylation of GST-ATF2. These data suggest that DAPAK is not coupled to p38 MAPK activation in HeLa cells.
![]() | Figure 5. Nocodazole-mediated induction of apoptosis in HeLa cells requires p38 MAPK activity. (A) Immunoblot analysis (WB) of nocodazole-induced apoptosis, assessed by PARP cleavage, after treatment of exponentially growing HeLa cells with nocodazole (3 μM) (more ...) |
To determine whether p38 MAPK was involved in this apoptotic response, p38 MAPK inhibitors were used. SB203580 (20 μM) or SB202190 (10 μM) reduced the number of apoptotic cells in the mitotic population by 46 and 42%, respectively, as assessed by cytokeratin 18 cleavage (Figure 5D). The relationship between p38 MAPK activity and apoptosis was examined further by transfecting HeLa cells with dominant active (DA) MAP kinase kinase, MKK6 (S207D, T211D). Consistent with previous reports (Derijard et al., 1995 ; Jiang et al., 1997
) DAMKK6 activated all coexpressed p38 MAPK isoforms (Figure 6A). Transient expression of daMKK6 in HeLa cells caused apoptosis, as assessed by cytokeratin 18 cleavage and chromatin fragmentation (Figure 6B). The DAMKK6-induced apoptosis is likely to be mediated through activation of endogenous p38 MAPK, because treatment of the transfected cells with either SB303580 (20 μM) or SB202190 (20 μM) inhibited apoptosis by ~67% (Figure 6C).
![]() ![]() | Figure 6. DAMKK6 induces apoptosis in HeLa cells by activating p38 MAPK, whereas DAPAK inhibits daMKK6-induced apoptosis. (A) HeLa cells were transfected with either the FLAG epitope-tagged p38 isotypes or FLAG-DAMKK6 (S207D, T211D) alone or in combination with (more ...) |
To address the function of PAK in mitotically arrested cells, DAPAKα (L106F) was transfected into HeLa cells, either separately or together with DAMKK6 (Figure 6D), and their effect on survival assayed using the M30 antibody. In the absence of any treatment, 2% of the cells were found to be apoptotic (Figure 6E). Transfection with the empty vector alone increased the number of apoptotic cells to 12%. In contrast, 54% of DAMKK6-transfected cells were found to be apoptotic as assessed by the presence of both cleaved cytokeratin 18 and fragmented chromatin, whereas a signaling inactive mutant of MKK6 (S207A, T211A) did not induce apoptosis (14%). Coexpression of DAPAK with DAMKK6 suppressed DAMKK6-induced apoptosis to near basal levels (18%), whereas KDPAK did not affect the level of MKK6-induced apoptosis. We also determined whether PAK aids cell survival of nocodazole-arrested mitotic cells. Treatment of DAPAK-transfected HeLa cells with nocodazole (3 μM for 24 h) reduced the number of apoptotic cells in the mitotic population by 56% (nocodazole: percentage of M30 positive cells ± SEM; 34.2 ± 3.4, n = 3; nocodazole + DAPAK: 15 ± 2.2, n = 3). These results suggest that nocodazole activates two signaling pathways in mitotically arrested cells that have opposing effects on cell survival. Activation of p38 MAPK either by nocodazole or MKK6 stimulates cell death, whereas activation of PAK stimulates cell survival.
![]() ![]() | Figure 7. DAMKK6 induces p38 MAPK-dependent redistribution of the proapoptotic protein Bax in HeLa cells. HeLa cells were transfected with either GFP-Bax or FLAG-DAMKK6 alone or cotransfected with either GFP-Bax and FLAG-DAMKK6, GFP-Bax, and FLAG-KDMKK6 or EGFP (more ...) |
We determined whether DAMKK6 was inducing redistribution of GFP-Bax to the mitochondria. In cells cotransfected with GFP-Bax and DAMKK6 (Figure 8, A–E), the punctate distribution of GFP-Bax (Figure 8A) coincided with mitochondrial staining by MitoTracker red (Figure 8B). These results demonstrate that the DAMKK6-induced activation of p38 MAPK is sufficient to stimulate Bax translocation from the cytosol to the mitochondria. Next, we determined whether Bax also translocates to the mitochondria in nocodazole-arrested mitotic cells. HeLa cells transfected with GFP-Bax were treated with nocodazole (3 μM) for 24 h and subsequently stained with an antibody to cytochrome c to detect the mitochondria (Figure 8, F–I). After 24 h of exposure to nocodazole between 20 and 30% of the transfected, mitotic cells displayed a punctate, perinuclear distribution of Bax (Figure 8F), which overlapped with a mitochondrial-rich region of the cell (Figure 8G). These cells were undergoing apoptosis as assessed by the presence of fragmented chromatin (Figure 8H). In nocodazole-arrested, nonapoptotic cells GFP-Bax was distributed homogeneously throughout the cytoplasm and did not colocalize with the mitochondria, which were distributed concentrically around the condensed chromatin (our unpublished data). Control nocodazole-arrested mitotic cells expressing GFP alone (Figure 8, J–M) also did not show a punctate, perinuclear pattern of Bax fluorescence despite being clearly apoptotic, based on membrane blebbing (Figure 8J) and DNA fragmentation (Figure 8L). These experiments indicated that Bax translocates from the cytoplasm to the mitochondria in nocodazole-arrested mitotic cells as they undergo apoptosis.
![]() | Figure 8. Both DAMKK6 and nocodazole induce Bax translocation to the mitochondria. Exponentially growing HeLa cells were cotransfected with GFP-Bax and FLAG-DAMKK6 (A–E) or with either GFP-Bax (F–I) or EGFP (J–M) alone. Sixteen hours (more ...) |
Consistent with previous data regarding Bad phosphorylation by PAK (Schurmann et al., 2000 ), our studies have indicated that recombinant PAKα and native PAKα, immunoprecipitated from nocodazole-arrested mitotic cells, both efficiently phosphorylated recombinant GST-mBad (our unpublished data). Because our current data have shown that PAKα is activated in mitotically arrested cells, we evaluated whether Bad is also phosphorylated in this cell population. Western blots of native Bad indicated that the mobility of the Bad protein was retarded preferentially in the mitotically arrested cells, suggesting that Bad undergoes posttranslational modification in this population (Figure 9A). To assess the phosphorylation state of Bad in the mitotically arrested cells HeLa cells were transfected with a mammalian expression vector encoding GST-Bad and then treated with either nocodazole, taxol, vincristine, or vinblastine. The phosphorylation state of Bad in the mitotic and attached cell populations was assessed by immunoblotting with phospo-specific Bad antibodies, which specifically detect Bad phosphorylation on serine 112 (S112), serine 136 (S136), or serine 155 (S155). All four drug treatments induced an increase in the level of Bad phosphorylation on residues S116, S136, and S155 in the mitotically arrested cells in comparison with either the asynchronous or attached populations, although we did not detect a mobility shift with GST-Bad (Figure 9B). To determine whether one or more of the serine residues was a specific target for PAK, we coexpressed GST-mBAD with either DAPAK (L107F) or KDPAK (K298A) and then assessed the phosphorylation state of GST-mBad by immunoblotting. DAPAK did not seem to affect the phosphorylation state of Bad on either S112 or S136. However, DAPAK increased phosphorylation of Bad at S155 compared with basal Bad phosphorylation at this site or when coexpressed with KDPAK (Figure 9C). Furthermore, this phosphorylation was abolished when S155 was replaced with alanine by site-directed mutagenesis, confirming that PAK phosphorylates mBad at this site. However, basal phosphorylation of Bad (S155A) at S112 and S136 was still observed although its level was slightly reduced compared with wild-type Bad. To suppress PAK activation we used GST-PAK (amino acids 83–149 of PAK1) an inhibitor of PAK1, PAK2, and PAK3 (our unpublished data). Coexpression of GST-PAK (83–149) with GST-Bad reduced phosphorylation of Bad at S112, S136, and S155 (Figure 9D) in nocodazole-arrested mitotic cells. Expression of GST-PAK (83–149) also sensitized nocodazole-arrested mitotic cells to cell death over the time course examined compared with cells transfected with vector alone (Figure 9E). Together, these results suggest that PAK phosphorylation of Bad may be a mechanism by which PAK contributes to cell survival in mitotically arrested cells.
![]() ![]() | Figure 9. PAK phosphorylates the proapoptotic protein Bad. (A) Immunoblot analysis of native Bad after treatment of HeLa cells with nocodazole (Noc, 3 μM), taxol (Tax), vincristine (Vinc), or vinblastine (Vinb) (all 1 μM) for 12 h. The cell extracts (more ...) |
We have identified the signaling events associated with activation of the spindle checkpoint by drugs that perturb microtubule dynamics. The results of this study demonstrate that 1) chemotherapeutic drugs cause activation of both p38 MAPK and PAK preferentially in the mitotically arrested cells; 2) chemotherapeutic drug-induced cell death is mediated through p38 MAPK, whereas chemotherapeutic drug-induced cell survival is mediated through PAK; 3) p38 MAPK activity induces the translocation of the cell death activator Bax, from the cytosol to the mitochondria; and 4) PAK phosphorylates Bad, thereby providing a mechanism to inhibit its proapoptotic function.
Our data indicate that the mitotic checkpoint is activated in nocodazole-arrested cells as assessed by the presence of the checkpoint protein MAD2 on the condensed chromatin. In these mitotic checkpoint-arrested cells, we demonstrate for the first time the concomitant activation of both p38 MAPK and PAK. Activation of either p38 MAPK or PAK was not observed in the attached cell population where both JNK and the Erk MAP kinases were found to be activated. The function of either JNK or Erk in the attached population is currently unknown. However, our studies indicate that Erk may suppress JNK-mediated cell death in this cell population (our unpublished data). Previous studies that have examined chemotherapeutic drug-activated signaling pathways have reported activation of JNK, p38 MAPK and Erk MAP kinases either alone or in combination (Wang et al., 1998 ; Shtil et al., 1999
; Subbaramaiah et al., 2000
; McDaid and Horwitz, 2001
; Okano and Rustgi, 2001
; Seidman et al., 2001
; Stadheim et al., 2001
). Discrepancies between the signaling pathways that are activated by chemotherapeutic agents may relate to the fact that these earlier studies did not separate the mitotic and the nonmitotic cell populations. The data presented in this study suggest that there is a remarkable divergence in the signaling pathways activated by antimicrotubule drugs in these two cell populations.
The upstream signaling events that lead to the activation of p38 MAPK and PAK during mitotic arrest remain to be identified. Although it has been reported that PAKs can activate p38 MAPK (Bagrodia et al., 1995 ), our data indicate that p38 MAPK and PAK activation are likely to be independent events. First, cotransfection of a DAPAK with p38α did not activate the p38 MAPK. Second, p38 MAPK was inactivated before PAK, after release of cells from a mitotic block, suggesting that PAK does not regulate the activation of p38 MAPK in mitotically arrested cells. We are currently examining the upstream components of the p38 MAPK and PAK signaling pathways in mitotically arrested cells.
The downstream target(s) of p38 MAPK that cause cell death in the mitotically arrested cells are largely unknown. A recent observation (Ghatan et al., 2000 ) that p38 MAPK induces apoptosis in neuronal cells by regulating the translocation of Bax from the cytoplasm to the mitochondria led us to examine the effect of p38 MAPK on Bax in the mitotically arrested cells. The translocation of Bax from the cytoplasm to the mitochondria is both necessary and sufficient to induce apoptotic cell death (Hsu et al., 1997
; Wolter et al., 1997
; Nechustan et al., 1999
). Bax reduces mitochondrial membrane potential, causes the release of cytochrome c from the mitochondria and activates caspases (Xiang et al., 1996
; Eskes et al., 1998
; Jurgensmeier et al., 1998
; Desagher et al., 1999
; Finucane et al., 1999
). In this study we demonstrate, first, that DAMKK6 causes the translocation of Bax to mitochondria, an effect that is reversed by the p38 MAPK inhibitors. Second, we show that Bax also translocates to mitochondria in the mitotically arrested cells and that the p38 MAPK inhibitors are able to suppress nocodazole-induced apoptosis. Therefore, our data support the conclusion that Bax translocation to the mitochondria, and subsequent cell death, are regulated by p38 MAPK in the mitotically blocked cells. The mechanism by which p38 MAPK may affect the cellular distribution of Bax is currently under investigation.
The activation of PAKα by chemotherapeutic drugs in a mitotically arrested cell population has not been described previously. We have shown in this study that DAPAKα suppresses MKK6-induced apoptosis, suggesting that one function of native PAKα may be to suppress cell death in the mitotically blocked cells. Indeed, our data have shown that expression of DAPAKα confers a survival advantage during nocodazole-induced mitotic arrest, whereas inhibition of PAK activation enhances nocodazole-induced cell death. In mammalian cells PAK isoforms have been shown to play distinct roles in apoptosis. Both PAK1 (PAKα) and PAK4 have been shown to protect cells from apoptosis induced by serum withdrawal, UV irradiation, or tumor necrosis factor-α in fibroblasts, or by growth factor withdrawal in lymphoid cells (Schurmann et al., 2000 ; Tang et al., 2000
; Gnesutta et al., 2001
), whereas a Xenopus PAK (X-PAK1) is required to suppress apoptosis during prophase arrest in frog oocytes (Faure et al., 1997
). However, γPAK (PAK2) seems to have both an antiapoptotic and a proapoptotic function (Jakobi et al., 2001
). The proapoptotic function of γPAK has been attributed to the generation of constitutively active fragment after caspase-mediated cleavage of γPAK (Rudel and Bokoch, 1997
). Although the activation of both γPAK and PAK4 in mitotically arrested cells awaits investigation the results from the present study are consistent with reports demonstrating that PAKα mediates cell survival in response to diverse apoptotic stimuli.
The target(s) of PAKα that regulate cell survival in mitotically blocked cells are also unknown. One mechanism through which PAK may protect lymphoid progenitor cells from apoptosis, after growth factor withdrawal, is by phosphorylation of the BH3-only proapoptotic protein Bad (Schurmann et al., 2000 ). PAK is reported to phosphorylate Bad on serine residues 112 and 136, thereby facilitating binding to 14-3-3τ and its sequestration in the cytoplasm. However, other prosurvival signals are also reported to phosphorylated Bad (Downward, 1999
). For example, protein kinase B phosphorylates Bad on serine residue 136, whereas protein kinase A is reported to phosphorylate Bad at serine 112. Both serine residues 112 and 136 occur within a protein kinase A phosphorylation motif (RRXS) in the BH3 domain of Bad that mediates its death-promoting activity through heterodimerization to the Bcl-XL family members (Tan et al., 2000
). In this study, we report that Bad is phosphorylated on serine residue 155 in addition to serine residues 112 and 136 in the mitotically arrested cells. Serine residue 155 also lies within the BH3 domain of Bad and is found in the sequence motif RRXS (Tan et al., 2000
) that is used by PAK to phosphorylate both serine residues 112 and 136. In overexpression experiments, we have shown that DAPAK primarily phosphorylates S155 of Bad. Phosphorylation of Bad at S155 is reported to induce cell survival by preventing the BH3-dependent dimerization of Bad with Bcl-XL rather than promoting binding to 14-3-3 (Tan et al., 2000
). Therefore, one mechanism by which cell death is inhibited in the mitotically arrested cells may be through phosphorylation of Bad by PAK at S155, thereby suppressing the proapoptotic Bad-Bcl-XL dimerization. We have shown that inhibition of PAK in mitotically arrested cells suppresses phosphorylation of Bad not only at S155 but also at S112 and S136 implying that native PAK is able to phosphorylate all three sites in vivo.
In summary, the present study provides evidence that both cell survival and cell death pathways are activated in mitotically arrested cells. A p38 MAPK-activated and Bax-dependent pathway contributes to cell death, whereas a PAK-activated and Bad-dependent pathway contributes to cell survival. The results of our study have implications for the design of future chemotherapeutic drug therapies that target the mitotic checkpoint. In particular, the identification of other survival pathways that are activated in response to the chemotherapeutic agents, and a search for agents that suppress them may considerably increase the effectiveness of the current anticancer therapies.
We thank Ed Manser, Roger Davis, Jiahuai Han, Gary Bokoch, and Richard Youle for generously providing various expression plasmids. Roger Snowden for assistance with flow cytometry and Kulvinder Sikand for assistance with confocal microscopy. This work was supported by grants from the Wellcome Trust and the Biotechnology and Biological Sciences Research Council (to R.P. and J.L.B.).