Adaptation to the spindle checkpoint is regulated by the interplay between Cdc28/Clbs and PP2A^Cdc55

During metaphase, chromosomes consist of pairs of sister chromatids held together by the cohesin complex. Cohesin cleavage by the protease separase is the essential step that starts the physical separation of sister chromatids and drives cells into anaphase (Nasmyth, 2002). For separase to cleave cohesin, cells need to degrade securin (Pds1 in Saccharomyces cerevisiae), the stoichiometric inhibitor of the protease. Degradation of Pds1 is mediated by the proteasome after ubiquitination by the essential E3 ubiquitin ligase anaphase promoting complex or cyclosome (APC/C) bound to its cofactor Cdc20 (Primorac and Musacchio, 2013). Degradation of Pds1 is insufficient, however, to guarantee exit from mitosis, which also requires the inactivation of the mitotic kinase cyclin-dependent kinase 1 (CDK1; Cdc28 in S. cerevisiae; Drapkin et al., 2009). Cdc28 inhibition occurs via the degradation of mitotic cyclins (Clbs, primarily Clb2 in budding yeast), an event that is initiated by APC/C^Cdc20 and is completed by another complex APC/C^Cdh1, where Cdh1 is a second cofactor that replaces Cdc20 as cells enter anaphase (Primorac and Musacchio, 2013).

A surveillance pathway known as the spindle checkpoint prevents APC/C^Cdc20 activation in the presence of as few as one kinetochore unattached to spindle microtubules (Rieder et al., 1995; Lara-Gonzalez et al., 2012). As a consequence of checkpoint activation, Cdc20 is sequestered in the mitotic checkpoint complex (MCC), which is composed of Cdc20 itself and three components of the checkpoint pathway: Mad2, Mad3 (BubR1 in mammals), and Bub3. Cdc20 in the MCC binds to APC/C, but APC/C^MCC is unable to recognize Pds1 and Clb2 (Lara-Gonzalez et al., 2012). After all kinetochores are properly attached to microtubules, the checkpoint is disengaged, MCC dissociates, APC/C^Cdc20 is activated, and cells can transit into anaphase.

Remarkably, the lifting of checkpoint-inducing stimuli is not strictly required for cells to enter anaphase. When the checkpoint is constantly induced, cells can adapt to the checkpoint and progress into the cell cycle after several hours of metaphase arrest (Rudner et al., 2000; Brito and Rieder, 2006; Rossio et al., 2010). How APC/C is activated in the presence of stimuli that induce Cdc20 sequestration, and thus APC/C inhibition, is still not completely understood.

In vertebrates, the mechanism of adaptation has been ascribed to a slow, APC/C-dependent degradation of cyclin B (thus the phenomenon is also known as slippage) taking place in cells arrested by the spindle checkpoint (Brito and Rieder, 2006). When cyclin B drops below a critical level, CDK1 is inactivated, and cells can no longer sustain the mitotic state leading to exit from mitosis.

The molecular mechanisms by which Cdc28/CDK1 inhibition could induce APC/C^Cdc20 activation, and transition into anaphase and adaptation, are not entirely clear. In mammals, CDK1 sustains the checkpoint (D’Angiolella et al., 2003; Morin et al., 2012) and thus indirectly inhibits APC/C^Cdc20. It has therefore been proposed that CDK1 inactivation relieves the inhibition of APC/C^Cdc20 exerted by the checkpoint and facilitates the transition to anaphase (Zeng et al., 2010; He et al., 2011).

A role for Cdc28 in maintenance of the spindle checkpoint, and thus in the inhibition of APC/C^Cdc20, has also been proposed in budding yeast (Li and Cai, 1997; Kitazono et al., 2003). However, in this organism Cdc28 has also a role in promoting APC/C^Cdc20 activity. Cdc28 activates APC/C by phosphorylating two of its subunits (Cdc16 and Cdc27), and thus favoring APC/C binding with Cdc20 (Rudner and Murray, 2000). Moreover, Cdc28 activity is required for the expression of Cdc20 during metaphase arrest (Liang et al., 2012). In summary, in yeast Cdc28 is reported to contribute to both APC/C^Cdc20 inhibition, indirectly via the checkpoint, and to APC/C^Cdc20 activation via APC phosphorylation and production of Cdc20 (Fig. 1 A). It is thus unclear whether Cdc28 activity promotes or inhibits metaphase arrest in cells subjected to a prolonged checkpoint stimulus.

Similarly, the relationship between APC/C^Cdc20 activation and phosphatases during adaptation to the spindle checkpoint is not fully understood. Based on the premise that Cdc28 is involved in spindle checkpoint maintenance, Cdc14, its main opposing protein phosphatase (Rock and Amon, 2009), should favor checkpoint silencing and APC/C^Cdc20 formation. Indeed, the overexpression of Cdc14 silences the checkpoint in a Cdc20-dependent manner, and Cdc14 prevents checkpoint reactivation in anaphase (Mirchenko and Uhlmann, 2010). It is unknown, however, whether physiological levels of Cdc14 have a role in promoting adaptation. This is not obvious, as Cdc14 is primarily required for events that follow anaphase and it is only fully activated well after APC/C^Cdc20 has been activated.

Phosphatase PP2A^Cdc55 (i.e., the type 2A phosphatase bound to its regulatory domain Cdc55) instead inhibits metaphase-to-anaphase transition: PP2A^Cdc55 activity drops when cells enter anaphase (Queralt et al., 2006), whereas cdc55Δ mutants are checkpoint deficient (Minshull et al., 1996). During an unperturbed cell cycle, PP2A^Cdc55 delays anaphase onset by inhibiting the release of Cdc14 from the nucleolus where it is bound to Net1 (Queralt et al., 2006; Wang and Ng, 2006; Yellman and Burke, 2006) and by inhibiting separase (Clift et al., 2009). Whether PP2A^Cdc55 has a similar role in inhibiting adaptation is unknown.

In this manuscript, we explore possible mechanisms of APC/C activation in cells arrested by the spindle checkpoint in budding yeast. We show that transition to anaphase in adaptation: (1) requires APC/C phosphorylation and Cdc28/Clbs activity; (2) is opposed by PP2A^Cdc55 via APC/C dephosphorylation; (3) is variable in length; and (4) does not require Cdc14. Finally, we suggest the presence of a positive feedback loop that allows a rapid transition from metaphase arrest to anaphase during adaptation to the spindle checkpoint.

APC/C is activated both by posttranslational modifications and by association with the cofactors Cdc20 and Cdh1. We asked which of these mechanisms is required for budding yeast cells to adapt to a prolonged checkpoint stimulus. To this aim, we compared the kinetics of the metaphase-to-anaphase transition in adapting cells (1) expressing nonphosphorylatable mutants for two subunits of the APC/C, Cdc27 (cdc27-5A) and Cdc16 (cdc16-6A); (2) deleted for CDH1; and (3) depleted of Cdc20.

In principle, we could have induced the spindle checkpoint by treating cells with tubulin-depolymerizing drugs like nocodazole. However, we observed that the effects of nocodazole begin to wear off after 5–6 h, and thus transition to anaphase after tubulin repolymerization and microtubule/kinetochore attachment can easily be mistaken for adaptation. We thus induced checkpoint activation by overexpressing one of its essential components, Mad2. We used yeast cells containing three copies of the MAD2 gene under the control of the inducible GAL1 promoter (GAL1-MAD2 (3X); Rossio et al., 2010). Addition of galactose to the medium of these cells switches on the GAL1 promoter and causes Mad2 to accumulate to a level that ectopically induces MCC formation, similarly to nocodazole (Mariani et al., 2012).

GAL1-MAD2 (3X) cells adapted to the checkpoint after several hours of growth in galactose. Adaptation was marked by the degradation of Clb2 and Pds1, and by the elongation of mitotic spindles, in agreement with previously published data (Rossio et al., 2010; Fig. 1, B and C). The presence of nonphosphorylatable subunits of APC/C prevented adaptation, with GAL1-MAD2 (3X) cdc16-7A cdc27-5A cells displaying persistently high levels of Clb2 and Pds1 (Fig. 1 D), and a large majority of metaphase spindles after several hours (Fig. 1 C). When grown in glucose, GAL1-MAD2 (3X) cdc16-7A cdc27-5A completed one cell cycle with kinetics that were only slightly delayed compared with GAL1-MAD2 (3X) cells (Fig. S1 A). These results are in agreement with data from Rudner and Murray (2000), who analyzed the growth of APC mutants in solid media under checkpoint-inducing conditions.

We next assessed the requirement of the cofactors Cdc20 and Cdh1 in adaptation to the spindle checkpoint. Given the essential role of Cdc20 in the metaphase-to-anaphase transition, it was not surprising that depletion of Cdc20 impaired adaptation to the checkpoint (Fig. S1 B). The role of CDH1, a nonessential gene in the metaphase-to-anaphase transition, was harder to predict. Because it was difficult to synchronize GAL1-MAD2 (3X) cdh1Δ cells in G1, we tested their ability to adapt to the spindle checkpoint using a cycling population. Our results show that GAL1-MAD2 (3X) cdh1Δ cells assembled and disassembled metaphase spindles very similarly to GAL-MAD2 (3X) cells, thus excluding a fundamental role for Cdh1 in adaptation (Fig. 1 E).

We conclude that although Cdh1 is dispensable for adaptation to the spindle checkpoint, both Cdc20 and phosphorylation of the APC/C subunits Cdc16 and Cdc27 are essential.

The observation that Cdc20 is required for cells to adapt is at odds with the fact that Cdc20 is sequestered in the MCC when the checkpoint is operational. We thus asked what mechanism underlies the release of Cdc20 in adapting cells. Our data showed that Clb2 is slowly, but steadily, degraded during adaptation (Fig. 1 B). Cdc28 requires Clbs for its activity in mitosis, and Cdc28 has been proposed to be required for checkpoint activity. It is thus plausible that the slow degradation of Clb2 by decreasing Cdc28 activity leads to checkpoint inactivation, Cdc20 release from the MCC, and ultimately to adaptation via the formation of APC/C^Cdc20.

To verify this hypothesis, we asked whether it is possible to overcome the spindle checkpoint arrest in yeast by inhibiting Cdc28. To modulate Cdc28 activity, we used a mutant of CDC28, cdc28-as1, that can be inhibited by the ATP analogue 1-NMPP1 (Bishop et al., 2000). APC/C^Cdc20 activity was monitored by the degradation of Pds1 and Clb2. Depending on the amount of 1-NMPP1, Cdc28-as1 can be fully or partially inactivated (Bishop et al., 2000; Liang et al., 2012). We reasoned that partial inactivation would better mimic the initial stages of Cdc28 inactivation after Clb2 degradation in adaptation. For this reason, we used 500 nM 1-NMPP1, a concentration that allows DNA replication but arrests cells in a premitotic state (Bishop et al., 2000; Liang et al., 2012). We confirmed this by adding 500 nM 1-NMPP1 to cdc28-as1 cells released from G1 arrest and measuring their DNA content by FACS and mitotic spindle assembly (Fig. S1, C and D).

cdc28-as1 cells were arrested in prometaphase by nocodazole for 3 h, and then treated with 500 nM 1-NMPP1. Having confirmed that nocodazole depolymerized spindles throughout the experiment (Fig. S1 E) and that cells arrested with a 2C DNA content (Fig. S1 F), we monitored cells for a further 2 h. We observed that APC/C was not reactivated toward Pds1 and Clb2, which were not destabilized, but Cdc20 levels rapidly decreased when compared with DMSO control (Fig. 2, A and B). The reduction in Cdc20 could be caused by its decreased synthesis or increased degradation. We favor the former hypothesis because Cdc28 inhibition arrests CDC20 transcription (Liang et al., 2012), whereas Cdc20 is quickly degraded during checkpoint arrest, its degradation being dependent on APC/C (Pan and Chen, 2004).

Our results suggest that the checkpoint is still active, and thus APC/C^Cdc20 is unable to target Pds1 and Clb2 after the partial inhibition of Cdc28. To confirm that this is the case, we repeated the experiment with a yeast strain carrying Mad2-GFP, which is recruited at kinetochores during a checkpoint-induced arrest (Gillett et al., 2004). Our results confirmed that the checkpoint pathway was still active at the kinetochores after Cdc28 inhibition, as Mad2-GFP colocalized with a component of the outer kinetochore, Nuf2, 2 h after treatment with 500 nM 1-NMPP1 (Fig. 2 E), like in the DMSO control. Cells released from nocodazole, which repolymerized mitotic spindles (Fig. S1 G), did not show kinetochore localization of Mad2 as expected.

The reduction of Cdc20 did not allow us to come to a conclusion about the status of the checkpoint pathway downstream of Mad2 after Cdc28 inhibition. To confirm that the pathway was still active, we repeated the experiment in cells where CDC20 expression is not under the control of Cdc28. To this aim, we used cells deleted for YOX1, which produces an inhibitor of CDC20 expression whose activity is down-regulated by Cdc28 (Liang et al., 2012). We observed that Cdc20 was stable in these cells for up to 1.5 h after the addition of 500 nM 1-NMPP1, whereas Pds1 and Clb2 levels remained unchanged, indicating that the checkpoint was still active after Cdc28 inhibition (Fig. S2 A and Fig. S2, B and C, for the functionality of the drug).

In summary, our results show that the inhibition of Cdc28 with 500 nM 1-NMPP1 is sufficient to keep the upstream spindle checkpoint pathway active, but unable to support the production of Cdc20. As a consequence, APC/C^Cdc20 could not be activated and cells remained arrested in metaphase.

Previous results showed that Cdc28 inhibition during a spindle checkpoint–induced metaphase arrest leads to the degradation of Pds1 and Clb2, and separation of sister chromatids (Amon, 1997). We reasoned that the discrepancy between these published data and our results is most likely caused by the level of Cdc28 inhibition. Therefore, we increased the dose of 1-NMPP1 from 500 nM to 5 µM, a concentration that is known to also inhibit DNA replication (Bishop et al., 2000). To test the efficacy of the drug, we synchronized cells in G1 and released them in media supplemented with 5 µM 1-NMPP1. We confirmed that under these conditions, cells arrested with 1C DNA content for 2.5 h without forming metaphase spindles (Fig. S1, C and D). We then added 5 µM 1-NMPP1 to MAD2-GFP NUF2-mCherry cdc28as-1 cells arrested in prometaphase by nocodazole. Similar to the results obtained with 500 nM 1-NMPP1, we observed that Mad2-GFP was largely localized at the kinetochores up to 2 h after the addition of 5 µM 1-NMPP1 (Fig. 2 E), which suggests that the checkpoint was not dismantled. However, at variance with results obtained with 500 nM 1-NMPP1, Pds1 and Clb2 levels decreased markedly after 1.5 h of treatment (Fig. 2 D). Cdc20 levels decreased even more rapidly after treatment with 5 µM 1-NMPP1, which implies that the reduction in Pds1 and Clb2 was not caused by APC/C^Cdc20 (Fig. 2, C and D).

We conclude that although APC/C can be activated by strong inhibition of Cdc28, its activation does not require Cdc20 nor the inactivation of the upstream checkpoint pathway at the kinetochores.

We asked whether the degradation of Pds1 and Clb2 after strong inhibition of Cdc28 was caused by APC/C^Cdh1. This possibility is supported by evidence that Cdc28 inhibits Cdh1 (Zachariae et al., 1998). Unfortunately, we could not use cdh1Δ cells that carry the cdc28as-1 allele because the two mutations are synthetic lethal. We thus inhibited Cdc28 activity by overexpressing the stoichiometric inhibitor Sic1 (Schwob et al., 1994), expressed under the GAL1 promoter (Amon, 1997). We first confirmed that Sic1 overexpression efficiently inhibited Cdc28 activity in vivo (Fig. S2 D). We then arrested cells in prometaphase with nocodazole for 2.5 h, and afterward we induced Sic1 overexpression with galactose. Crucially, Clb2 and Pds1 were stable in cdh1Δ cells but were quickly degraded in cells carrying CDH1 wild type (Fig. 3 A). Notice that cells are under nocodazole and thus unable to form microtubules: that is why they cannot undergo cytokinesis and exit mitosis regardless of the degradation of Pds1 and Clb2 (Fig. 3 A). This result confirms the hypothesis that APC/C^Cdh1 is activated when Cdc28 is strongly inhibited. As expected, with physiological levels of Sic1, cells maintained the mitotic arrest with stable Pds1 and Clb2 independently of the presence of Cdh1 (Fig. 3 B). The degradation of Cdc20 in cdh1Δ cells is likely caused by the inhibition of its transcription after Cdc28 inactivation.

We thus conclude that when Cdc28 activity is greatly reduced, APC/C^Cdh1 is activated and degrades both Pds1 and Clb2. Although Pds1 and Clb2 were degraded, it would be wrong to conclude from these data that adaptation requires APC/C^Cdh1 activation mediated by a strong inactivation of Cdc28, as we have previously shown Cdh1 to be dispensable for adaptation (Fig. 1 E).

The experiments performed on cells arrested in prometaphase by nocodazole suggest that cells do not adapt to the checkpoint by inhibiting Cdc28 activity. We thus decided to test the opposite scenario, where Cdc28 activity is required for adaptation. In this scenario, it would be possible to inhibit adaptation by inhibiting Cdc28 with a minimal amount of ATP analogue sensitivity. To test this prediction, we assessed the effect of as little as 50 nM 1-NMPP1 on adaptation to a checkpoint arrest induced by Mad2 overexpression. GAL1-MAD2 (3X) cdc28-as1 cells were synchronized in G1 by α-factor and released in galactose to induce Mad2 overexpression. Under control conditions (DMSO), the majority of cells adapted within 5 h (Fig. 4 A, left; and Fig. 4 B); however, in the presence of 50 nM 1-NMPP1, adaptation was inhibited in the vast majority of cells (Fig. 4 A, right; and Fig. 4 B). At this low concentration of 1-NMPP1, no effects were observed on cells expressing wild-type CDC28 (Fig. S3, A and B). Moreover, when GAL1-MAD2 (3X) cdc28-as1 cells were grown in a medium lacking galactose and therefore were not checkpoint arrested, 50 nM 1-NMPP1 induced only a small delay in the transition to anaphase as detected by spindle dynamics and FACS (Fig. S3 C).

We conclude therefore that in budding yeast Cdc28 activity is required for adaptation to the spindle checkpoint.

The notion that active Cdc28 is required for adaptation prompted us to reconsider the observation that Clb2, which is required for Cdc28 activity, is slowly degraded as cells adapt (Fig. 1 B). We reasoned that Western blots might mask the real nature of adaptation taking place in individual cells. We thus resorted to single cell analysis and followed the dynamics of Clb2 in yeast strains in which the mitotic cyclin was tagged with GFP (Hood et al., 2001).

Asynchronous GAL1-MAD2 CLB2-GFP TUB2-Cherry cells were loaded in flow chambers in the presence of galactose, and after a short time they arrested as budded cells in metaphase. We then followed adaptation in individual cells (Fig. 5 A), and the results were doubly surprising. Not only did we observe that Clb2 increased constantly during metaphase arrest, but the mitotic cyclin was also degraded very rapidly when cells adapted and entered anaphase. More precisely, Clb2 nuclear accumulation was constant and biphasic. Initially, Clb2 increased with a similar rate as in dividing cells (Fig. S4 A, bottom; and Fig. 5 B, top right), then the rate of increase slowed down, although Clb2 kept on increasing. On average, adapting cells accumulated about twice as much nuclear Clb2 as cycling cells (Fig. 5 A and Fig. 5 B, bottom left). The phase of slow Clb2 increase lasted until cells adapted. Entry into anaphase, detected by the elongation of mitotic spindles, occurred together with the rapid degradation of Clb2, which resulted in complete loss of the total protein in 20–30 min. The rate of Clb2 degradation was comparable to the rate of degradation in dividing cells (Fig. 5 B, top left; and Fig. S4 A, top). Crucially, adaptation times (i.e., the time between the beginning of Clb2 accumulation and spindle elongation) were extremely variable within the cell population, spanning over a period of 500 min (Fig. 5 B, bottom right).

It is likely that the slow degradation of Clb2 observed in Western blots is caused by asynchrony in adaptation times, which masked the rapid nature of Clb2 degradation in individual cells. When we repeated the experiments inducing the spindle checkpoint with nocodazole, we observed a similar behavior (Fig. S4 C, top) in cells that underwent adaptation before nocodazole ceased to inhibit the polymerization of tubulin (Fig. S4 C, bottom). Thus, Clb2 dynamics observed during Mad2 overexpression are a bona fide representation of what happens when the spindle checkpoint is induced by impairing microtubule–kinetochore attachment.

In conclusion, Clb2 concentration increases, and does not decrease, before cells adapt in budding yeast. Moreover, the sudden and stochastic activation of APC/C, and not its residual activity, drives the transition to anaphase during adaptation.

We next asked why Cdc28 activity is necessary for adaptation (Fig. 4 A). We have already shown that phosphorylation of the APC/C subunits, Cdc16 and Cdc27, is required for adaptation (Fig. 1 D); therefore, it is possible that Cdc28 is needed to phosphorylate APC/C during adaptation. This notion is supported by published data showing that the two APC/C subunits are phosphorylated by Cdc28 when cells enter anaphase (Rudner and Murray, 2000). We confirmed that before adaptation Cdc16 is progressively phosphorylated, as detected by Western blotting using Phos-tag reagent (Fig. S5, A and B).

We reasoned that a phosphatase might oppose the increasing Cdc28/Clb2 activity during adaptation because cells took several hours to adapt despite accumulation of Clb2 in the nucleus. If Cdc28-mediated phosphorylation of APC/C is a key molecular event in overcoming the checkpoint arrest, then the phosphatase that dephosphorylates APC/C should be required to maintain arrest. A plausible candidate that could be responsible for APC/C dephosphorylation is the phosphatase PP2A^Cdc55, as cdc55Δ cells are checkpoint deficient (Minshull et al., 1996). Indeed, we confirmed that GAL1-MAD2 (3X) cdc55Δ cells do not undergo spindle checkpoint–induced arrest triggered by Mad2 overexpression (Fig. S5 C). We thus decided to test whether PP2A^Cdc55 controls the phosphorylation state of APC/C during nocodazole-induced arrest.

We compared the Western blot profiles of Cdc16 and Cdc27 in CDC55 wild-type and cdc55Δ cells after releasing them from a G1 arrest into nocodazole. Using acrylamide gels containing Phos-tag reagent (Kinoshita et al., 2008), we observed that Cdc16 is already more highly phosphorylated in cdc55Δ cells compared with CDC55 wild-type strains during G1 arrest (indicated by the arrows in Fig. 6 A, left). This difference became more evident in cells arrested for 2 and 3 h in nocodazole: in cdc55Δ cells we observed phosphorylation-dependent shifts that were completely missing in CDC55 cells (Fig. 6 A, left), and that were sensitive to phosphatase treatment (not depicted). A shift that was sensitive to phosphatase treatment was also detected in standard Western blots of lysates from cdc55Δ cells (Fig. S5 D), confirming that Cdc16 underwent enhanced phosphorylation in the absence of PP2A^Cdc55. The same analysis with Cdc27 did not reveal major differences in the total level of phosphorylation between CDC55 wild type and cdc55Δ cells (Fig. 6 A, right).

These data indicate that the protein phosphatase PP2A^Cdc55 directly or indirectly dephosphorylates Cdc16 and that these events could be critical to maintenance of checkpoint arrest and inhibition of adaptation.

Together, our results support a scenario where Cdc28/Clb phosphorylates and PP2A^Cdc55 dephosphorylates Cdc16. The final output of these opposing activities could be net phosphorylation, leading to adaptation, or net dephosphorylation, leading to mitotic arrest. If this hypothesis were correct, we would expect the loss of checkpoint arrest in cdc55Δ cells to be reversed in the presence of nonphosphorylatable Cdc16.

Although Cdc27 is not dephosphorylated by PP2A^Cdc55, the simultaneous presence of nonphosphorylatable Cdc27 and Cdc16 strongly decreases the ability of the single mutants to adapt (Rudner and Murray, 2000). For this reason, we tested the hypothesis by assessing spindle checkpoint activity in the triple mutant cdc16-7A cdc27-5A cdc55Δ. By following the kinetics of mitotic spindles and the stability of Pds1, we observed that, in contrast to GAL1-MAD2 (3X) cdc55Δ cells, GAL1-MAD2 (3X) cdc16-6A cdc27-5A cdc55Δ were arrested by the checkpoint and in fact were severely impaired in adaptation to the spindle checkpoint induced by Mad2 overexpression (Fig. 6, B–E). The fact that a minority of triple mutant cells eventually adapted, whereas no cdc16-6A cdc27-5A cells did, could be due to the presence of additional phosphorylation sites on the APC/C recognized by PP2A^Cdc55. Alternatively, PP2A^Cdc55 might also inhibit APC/C activation via other pathways, which are nevertheless less relevant than APC/C dephosphorylation. When Mad2 was not overexpressed, all three strains progressed through the cell cycle with similar kinetics (Fig. S5 E).

We conclude that increased phosphorylation of Cdc16 causes the checkpoint deficiency of cdc55Δ cells.

Our data indicate that during adaptation, PP2A^Cdc55 inhibits the transition into anaphase by dephosphorylating Cdc16. Previous data showed that during a regular cycle PP2A^Cdc55 delays anaphase by preventing Cdc14 release from the nucleolus (Queralt et al., 2006; Wang and Ng, 2006; Yellman and Burke, 2006). It is thus possible that during mitotic arrest PP2A^Cdc55 not only prevents adaptation by dephosphorylating APC/C, but also by inhibiting the release of Cdc14, which could help cells to transit into anaphase by silencing the checkpoint (Mirchenko and Uhlmann, 2010).

To test this possibility, we determined whether constitutive Cdc14 activity is sufficient to inactivate a nocodazole-induced checkpoint. We used the net1Δ mutant, where Cdc14 is constitutively released throughout the cell cycle. Our results show that net1Δ cells are checkpoint proficient, as indicated by the stability of Pds1 (Fig. 7 A). Notably, cdc55Δ cells, which also release Cdc14 prematurely, are instead checkpoint deficient (Fig. 7 B). The metaphase-to-anaphase transition is similar in the two mutants during an unperturbed cell cycle (Fig. S5, F and G). These results indicate that Cdc14 release alone is not sufficient to inactivate the checkpoint, and imply that PP2A^Cdc55 activity does not arrest cells in metaphase by preventing Cdc14 release.

We then tested whether Cdc14 activity is essential for adapting cells to enter anaphase. If this were the case, we would expect that inactivation of Cdc14 after checkpoint arrest would cause a permanent metaphase block. Because CDC14 is an essential gene, we used the temperature-sensitive mutant cdc14-1, which produces a phosphatase that is inactivated after ∼30 min at 37°C (Akiyoshi and Biggins, 2010). In the absence of active Cdc14, cells arrest in anaphase with elongated spindles (Rock and Amon, 2009). To observe the effect of an incomplete activation of Cdc14, we also used a strain carrying a temperature-sensitive mutation of Cdc15, which causes partial release of Cdc14, but nevertheless arrests in anaphase (Rock and Amon, 2009). We activated the checkpoint in these temperature-sensitive mutants by overexpressing Mad2 from the GAL1 promoter.

Cells were synchronized in G1 at 23°C and released in galactose to activate the checkpoint. To inactivate cdc14-1 and cdc15-2, the temperature was increased to 37°C, 2 h after the release when most of the cells had large buds. The strains carrying cdc14-1 and cdc15-2 degraded Pds1 and Clb2 and disassembled metaphase spindles with only a small delay compared with GAL-MAD2 (3X), and in contrast to the adaptation-deficient GAL-MAD2 (3X) cdc27-5A cdc16-6A cells; Cdc14 and Cdc15 were clearly inactivated as cells accumulated in anaphase (Fig. 7, C and D). GAL1-MAD2 cdc15-2 cells were slightly delayed compared with GAL1-MAD2 cdc14-1 cells. As expected, when the same strains were released at 37°C in glucose (i.e., without switching on the GAL1 promoter and inducing Mad2 overexpression), cells carrying the mutations in CDC14 and CDC15 showed metaphase spindle kinetics similar to wild-type cells before arresting in anaphase, whereas the double APC/C cdc27-5A cdc16-6A mutant was delayed by ∼20 min in the exit from metaphase (Fig. S5 H).

We conclude that Cdc14 activation is not necessary for cells to transit into anaphase when the checkpoint is activated.

When cells are exposed to prolonged checkpoint-inducing stimuli, they are arrested in metaphase by the sequestration of the APC/C cofactor, Cdc20, in the MCC. After several hours, cells eventually adapt to the stimulus, resulting in segregation of their DNA material and exit from mitosis. Because Cdc20 is sequestered in the MCC during a checkpoint arrest, it is unclear how APC/C is activated during adaptation to drive cells into anaphase.

One possibility is that APC/C bypasses the requirement for Cdc20 during adaptation. Our data show that Cdh1, the other APC/C cofactor, is dispensable for adaptation to the spindle checkpoint, as also suggested by Rossio et al. (2010) and Rudner et al. (2000). A second option is that cells adapt because they switch off the checkpoint cascade. However, we did not obtain any evidence to support such a scenario. Previous studies in budding yeast suggest that the pathway upstream of MCC formation is switched off via a decrease of Mad1/Bub3 binding (Rossio et al., 2010), two components of the spindle checkpoint pathway that localize at the kinetochores. In our adaptation experiments, we have activated the checkpoint ectopically by overexpressing Mad2. We have shown previously that in this setting MCC formation occurs independently from Mad1 (Mariani et al., 2012) and thus the reduced binding of Mad1 with Bub3 is unlikely to lead to the release of Cdc20 from the MCC. It was also reported that Bub1 degradation helps adaptation (Goto et al., 2011), but we have previously shown that Bub1 does not localize to kinetochores in cells overexpressing Mad2 (Mariani et al., 2012). For this reason, we suspect that Bub1 inactivation does not underlie adaptation in our system. Another attractive candidate for checkpoint silencing is Cdc14, given its recently discovered role in preventing checkpoint reactivation in anaphase (Mirchenko and Uhlmann, 2010). However, we determined that Cdc14 was neither necessary nor sufficient for adapting cells to transit into anaphase. Cdc28 inactivation might also underlie checkpoint inactivation at the time of adaptation, as has been proposed to occur during the transition to anaphase in vertebrates (Zeng et al., 2010; He et al., 2011). However, we observed that the inhibition of Cdc28 does not delocalize Mad2 from the kinetochores in yeast cells arrested in prometaphase by the checkpoint, which suggests that in yeast there is not such a strong link between Cdc28 activity and checkpoint maintenance. Moreover, we demonstrated that inhibition of Cdc28 activity in checkpoint-arrested cells is incompatible with transition into anaphase. In conclusion, our data support the notion that adaptation takes place when the checkpoint is fully functional as originally proposed by Brito and Rieder (2006).

In this scenario, adaptation would be driven by a mechanism that competes with the checkpoint and eventually overrides it through the formation of a critical amount of APC/C^Cdc20 that drives cells into anaphase. Here, we propose that the phosphorylation of APC/C, favored by Cdc28 and opposed by PP2A^Cdc55, plays a critical role in the adaptation process. The observation that the mitotic cyclin Clb2, a key regulator of Cdc28, accumulates steadily before cells adapt, supports the idea that a threshold level of active APC/C phosphorylation needs to be reached before cells can overcome checkpoint arrest. Possibly, APC/C phosphorylation drives the formation of APC/C^Cdc20 because it increases the affinity of APC/C for Cdc20 (Rudner and Murray, 2000). Although we demonstrate that APC/C phosphorylation is necessary for adaptation, we cannot conclude that it is the ultimate event that causes it. Further work is needed to clarify this point, and particularly to study the regulation of PP2A^Cdc55 in adapting cells. Whatever the mechanism that triggers adaptation, the high variability in adaptation times complicates the analysis of this process.

It is worthwhile to note that in mammals the molecular mechanism that drives adaptation is likely to differ from that in yeast. Indeed, it has been shown that during adaptation in single mammalian cells cyclin B is slowly degraded (Brito and Rieder, 2006), which is at variance with our results in single budding yeast cells.

Our data provide a new explanation for the old observation that the regulatory subunit of the PP2A phosphatase, Cdc55, is required for the spindle checkpoint. We suggest that PP2A^Cdc55 contributes to checkpoint arrest by inhibiting the formation of APC/C^Cdc20 through the dephosphorylation of the APC/C subunit, Cdc16. In this scenario, cells lacking Cdc55 would “always be adapted” because the inability to dephosphorylate APC/C would render cells unable to mount a checkpoint arrest. Accordingly, cdc55Δ cells become checkpoint proficient again in a strain where the phosphorylation sites of Cdc16 and Cdc27 recognized by Cdc28/Clb2 are mutated to alanine. This result sheds light on a previous observation that a CDC28 mutant with decreased kinase activity (CDC28-VF) rescues the checkpoint defect of cdc55Δ (Minshull et al., 1996). In our interpretation, the rescue is caused by a balancing out of APC/C phosphorylation and dephosphorylation, achieved by the simultaneous decrease of Cdc28 kinase and PP2A phosphatase activity.

Our results showing that PP2A^Cdc55 dephosphorylates Cdc16, which are in agreement with recent findings (Lianga et al., 2013), do not exclude the existence of phosphatases other than PP2A^Cdc55 that could dephosphorylate APC/C. The phosphorylation state of Cdc27 is not affected by the deletion of CDC55, and yet the alanine substitution of the phosphorylation sites on Cdc27 decreases the capability of cells to adapt (Rudner and Murray, 2000). Moreover, the previous observation that the deletion of CDC55 is fully balanced by the partial inhibition of Cdc28 (Minshull et al., 1996) could be explained by the presence of overlapping phosphatases.

During adaptation, activation of APC/C^Cdc20 requires Cdc28/Clbs. Consistently, Clb2 accumulates in the nucleus steadily for several hours until cells adapt. However, as soon as APC/C^Cdc20 is activated it rapidly degrades Clb2: the high level of Cdc28/Clbs that cells accumulated in several hours is lost in a few minutes (Fig. 5 A). The later stages of Clb2 degradation are likely due to APC/C^Cdh1, but the initial stages are APC/C^Cdc20 dependent, and are aided by APC/C phosphorylation. It is therefore surprising that the rapidly diminishing levels of Cdc28/Clbs are able to sustain efficient phosphorylation of APC/C, in the initial stages after APC/C^Cdc20 activation, in the presence of the counteracting phosphatase PP2A^Cdc55.

This conundrum does not concern only adaptation in budding yeast. In cycling extracts of Xenopus laevis, it was recently shown that this paradox is resolved by a time delay between CDK1 and APC/C^Cdc20 activation, which guarantees high levels of APC/C^Cdc20 after it starts to degrade Cyclin B (Yang et al., 2013). Our data suggest a possible molecular mechanism to explain the establishment of the time delay. During a regular transition to anaphase, PP2A^Cdc55 is inhibited by separase (Queralt et al., 2006), which is in turn activated by APC/C^Cdc20. Together, these proteins form a positive feedback loop (APC/C^Cdc20 ⊣ Securin ⊣ separase ⊣ PP2A^Cdc55 ⊣ APC/C^Cdc20) that is coupled to the negative feedback loop, whereby APC/C^Cdc20 inactivates Cdc28/Clbs that in turn activates APC/C (Fig. 8 A). A simple model based on these ideas (see Materials and methods) shows that Clb2 increases steadily until APC/C is activated, and then the negative feedback loop comes into action (Fig. 8 B). After this point, APC/C^Cdc20 locks APC/C in the phosphorylated state by inhibiting the protein phosphatase responsible for APC/C dephosphorylation. Hereafter, the decrease of Cdc28/Clb2 activity becomes irrelevant, as the phosphatase that could oppose Cdc28 activity has been inactivated. In fact, APC/C^Cdc20 activity stays high after Clb2 is long gone (in reality it will likely be removed, together with the remaining Clb2, by APC/C^Cdh1, which is missing in the model).

Our model is based on the idea that the positive feedback loop creates two states (i.e., it is bistable): a metaphase state with APC/C^Cdc20 inactive and PP2A^Cdc55 active, and an anaphase state that follows adaptation where the activation states are reversed. The idea that a bistable switch controls transition into anaphase is not new. It was proposed for the first time after the observation that APC/C^Cdc20 degrades Mps1, a component of the checkpoint (Palframan et al., 2006). In that case, inactive APC/C^Cdc20 and active Mps1 characterized metaphase, whereas the activation states were swapped in anaphase. Other positive feedback loops have been proposed more recently to govern the transition (Zeng et al., 2010; He et al., 2011). Currently, it is not known how these switches cooperate with each other, and in fact more data are needed to confirm the relevance of a bistable switch controlling the transition to anaphase in budding yeast.

Strains used in this study are listed in Table S1. All yeast strains were derivatives of or were backcrossed at least three times with W303 (ade2-1, trp1-1, leu2-3, 112, his3-11, 15, ura3, and ssd1). cdc27-5A and cdc16-6A mutants were obtained from A. Murray (Department of Molecular and Cellular Biology and Center for Systems Biology, Harvard University, Cambridge, MA); cdc28-as1 and yox1Δ were received from U. Surana (Institute of Molecular and Cell Biology, Agency for Science, Technology and Research [A*STAR], Singapore); cdc14-1, cdc15-2, and GAL1-SIC1 were from R. Visintin (IEO, Milan, Italy); and the original construct for GAL1-MAD2 was developed in the lab of S. Piatti (Centre de Recherche en Biochimie Macromoléculaire, Montpellier, France).

The population experiment in Fig. S1 B was performed in synthetic medium lacking methionine. All other population experiments were performed using YEP medium (1% yeast extract, 2% Bacto Peptone, and 50 mg/liter adenine) supplemented with 2% glucose (YEPD), 2% raffinose (YEPR), or 2% raffinose and 2% galactose (YEPRG). Single cell adaptation experiments under Mad2 overexpression were performed using synthetic complete medium supplemented with ammonium sulfate.

α-Factor and nocodazole were used at 5 µg/ml and 15 µg/ml, respectively. α-Factor readdition was at 20 µg/ml. Methionine was added at a final concentration of 20 mM to repress the MET3 promoter. Unless otherwise specified, all experiments were performed at 30°C. In population experiments of adaptation, galactose was added 1 h before the release from α-factor (20 min for Fig. S1 B), while α-factor was re-added 1.5–2 h after the release from G1, unless otherwise stated.

In all Western blot experiments, samples were collected at the indicated time points, and cells were pelleted by centrifugation for 2 min at room temperatures. Then 100 µl TCA was added to the pellets to precipitate proteins. TCA protein extracts were prepared according to Fraschini et al. (1999). Protein samples were loaded and separated in 10% or 12.5% polyacrylamide gels (with a bis-acrylamide/acrylamide ratio of 1:29) with the voltage of separation apparatuses set to 140 V. For detection of Cdc16 and Cdc27 phosphorylation-specific bands, the Phos-tag system (50 µM Phos-tag reagent) was used (Kinoshita et al., 2008). To visualize phosphorylation shifts of Cdc16 on normal gels, we prepared 7.5% polyacrylamide gels with a bis-acrylamide/acrylamide ratio of 1:80. Proteins were transferred from gels to Protran membranes for 1 h, with transfer apparatuses set at 100 V. For the detection of Myc-tagged proteins (Pds1-Myc18, Cdc16-Myc6, and Cdc27-Myc9; the last two strains were a gift from J.-M. Peters, Research Institute of Molecular Pathology [IMP], Vienna, Austria) and of Mad2, membranes were probed, respectively, with 9E10 anti-Myc and anti-ScMad2 antibodies produced at the Monoclonal Antibodies Facility at the IFOM-IEO Campus (1:1,000 dilution). Commercial antibodies were used as follows: Cdc20 (yC-20; Santa Cruz Biotechnology, Inc.) at 1:1,000, Clb2 (y180; Santa Cruz Biotechnology, Inc.) at 1:1,000, and Pgk1 (D660; Invitrogen) at 1:5,000. Secondary antibodies were from Bio-Rad Laboratories and proteins were detected by an enhanced chemiluminescence system (Pierce ECL; Thermo Fischer Scientific) according to the manufacturer’s instructions. Blots were acquired as digitalized images by a Chemidoc XRS+System (Bio-Rad Laboratories) and the software Imagelab was used to quantify the signals. Molecular weights in the Western blot panels refer to Prestained Protein Marker (New England Biolabs, Inc.) bands run on the same blot.

To quantify Cdc16 and Cdc27 phospho-specific bands on P-Tag gels, the signal corresponding to the phosphorylation shift (above the second band visible during the G1-arrest) was normalized over the total amount of protein in the same gel at the same time point.

Flow cytometric DNA quantitation was determined using a flow cytometer (FACScan or FACScalibur; BD) and analyzed with FlowJO Software. For each sample, we scored 10,000 events.

Metaphase and anaphase spindles (i.e., immunofluorescence [IF] analysis) were detected by α-tubulin immunostaining with the YOL34 monoclonal antibody (AbD Serotec) followed by indirect IF using FITC-conjugated anti–rat antibody (Jackson ImmunoResearch Laboratories, Inc.).

Single cell analysis was performed using microfluidic chambers (CELLASIC), with cells growing at 30°C in synthetic medium containing raffinose. Time-lapse movies were recorded using a DeltaVision Elite imaging system (Applied Precision) based on an inverted microscope (IX71; Olympus) with a camera (CoolSNAP HQ2; Photometrics) and a UPlan-Apochromat 60× (1.4 NA) oil immersion objective lens (Olympus). Cells expressing Mad2-3GFP (gift from T. Tanaka, Centre for Gene Regulation and Expression, College of Life Sciences, University of Dundee, Scotland, UK) and Nuf2-mCherry (a gift from S. Westermann, Research Institute of Molecular Pathology [IMP], Vienna, Austria) were fixed in cold 100% ethanol, and images were acquired on the DeltaVision Elite imaging system using a UPlan-Apochromat 100× (1.4 NA) oil immersion objective lens (Olympus). We counted cells as colocalized where the maximum of Mad2-GFP signal was localized with one Nuf2-mCherry dot.

To perform the phosphatase assay on protein samples extracted with TCA, Laemmli buffer was exchanged with Buffer3 (10 mM MgCl₂, 100 mM NaCl, 50 mM TRIS, and 1 mM DTT, pH 7.9) using Amicon Ultra centrifugal filters (Millipore). The extracts were then incubated for 30 min at 37°C with buffer alone, calf intestinal phosphates (CIP; New England Biolabs, Inc.), or CIP plus 10 mM sodium orthovanadate (to inhibit the CIP).

Image analysis was performed with software written in MATLAB (MathWorks). For the segmentation and tracking of yeast cells we used the program phyloCell, written by G. Charvin (Institut de Génétique et de Biologie Moléculaire et Cellulaire, Illkirch, France). From the fluorescent signal of the whole cell we subtracted a background signal that we obtained from cells not carrying the fluorescent markers grown in the same conditions. We observed a slight increase in autofluorescence during adaptation and corrected our measurements accordingly.

The calculation of the concentration of fluorescence in the nucleus required us to know the area of the nucleus. Because our cells do not carry a nuclear marker, we used the Clb2-GFP signal itself, as Clb2 accumulates in the nucleus during mitosis. The identification of the nucleus was thus successful only when the intensity of the Clb2 nuclear signal was significantly higher than in the rest of the cell. The size of the nucleus could then be identified by k-means clustering, using k = 3 (the clusters corresponding to nucleus, cytoplasm, and extracellular region; see Video 1). In contrast, when the intensities in the nucleus and in the cytoplasm were similar, the clustering gave unrealistically high values for the size of the nucleus. Therefore, the moment in which the measurement becomes reliable is marked by a drop in the ratio of the size of the nuclear cluster to the size of the whole cell (Fig. S4 B, top). We considered the clustering to be reliable when this ratio was below a value of 30%. We used the nuclear signal for our calculations only in this region (Fig. S4 B, bottom, solid line), after smoothing using a moving average. For the sake of representation, in Fig. 5 A and Fig. S4 C we also plotted the nuclear concentrations when the definition of the nucleus was unreliable.

The signal before interpolation is visible in Video 1.

We observed that the increase in nuclear Clb2, both in cycling cells and in the initial phase of adaptation, was approximately linear. Therefore, to quantify the expression rates (Fig. 5 B, top right), we performed a linear fit and compared the resulting slopes for the two conditions. We only used the part of the curve where the increase was sufficiently linear and for which the measurement of the nuclear region was reliable (see “Nuclear signal”).

It was not possible to measure the rate of Clb2 degradation in adapting cells directly in the nucleus because the measurement of the nuclear region was not reliable during the degradation phase. For this reason we analyzed Clb2 degradation using the mean fluorescence in the whole cell (Fig. 5 B, top left). We assumed that this degradation can be described by a simple exponential decay (Fig. S4 A, top) and that at the end of the process Clb2 is fully degraded.

The maxima of fluorescence (Fig. 5 B, bottom left) were measured by taking the local maxima of nuclear fluorescence in the case of cycling cells. In the case of the adapting cells, we took the highest fluorescence level after smoothing that was reached during the adaptation process.

Adaptation time (Fig. 5 B, bottom right) was measured in cells adapting to Mad2 overexpression as the time that elapses between the moment Clb2 starts to accumulate to the time the mitotic spindle elongates. If the time of spindle elongation was uncertain, we chose the beginning of Clb2 degradation. When we analyzed adaptation of cells treated with nocodazole (Fig. S4 C, top), we obviously chose only cells that had not formed clusters of tubulin by the time of Clb2 degradation, and thus we did not take into account cells that degraded Clb2 after tubulin reappeared (Fig. S4 C, bottom).

Because Cdc28 is present in excess over Clb2 and their binding is spontaneous, in the model we follow the dynamics of Clb2 as a readout for the complex Cdc28/Clb2. As for APC/C, we did not explicitly introduce Cdc20 but we assume that Cdc20 binds efficiently to phosphorylated APC/C (i.e., APCP comes always with Cdc20 bound to it and is a readout for APC/C^Cdc20). As for PP2A^Cdc55, given that the mechanism whereby separase degradation induces the inhibition of the phosphatase is not known, we simply introduce two species, active and inactive. Simulations were performed with XPPAUT. To produce the behavior shown in Fig. 8, we required the model to be in a limit cycle solution. The parameter values have been chosen accordingly (i.e., any other choice of parameter that produces a limit cycle solution would equally prove the point). Similarly, initial conditions were chosen from the limit cycle solution with kph = 1.2 (i.e., checkpoint OFF) where Clb2 concentration was minimal. In the checkpoint ON condition, kph = 0.6, which implies reduced APC/C^Cdc20 formation. In the following sections, we describe the reactions of the model, and the corresponding mathematical formalism (i.e., algebraic and differential equations).

The model includes the following reactions.

APC + Clb2 → APCP + Clb2 (kph = phosphorylation; Japc = Michaelis constant for APCP phosphorylation)

APCP + Cdc55a → APC + Cdc55a (kdeph = dephosphorylation; Japci = Michaelis constant for APCP dephosphorylation)

Cdc55i → Cdc55a (ka = Cdc55 activation)

Cdc55a + APCP → Cdc55i + APCP (ki = Cdc55 inactivation)

→ Cdc20 (ksyn = Cdc20 synthesis)

Cdc20 + APCP → APCP (kdeg = Cdc20 degradation)

When translated in mathematical formalism, the reactions produce the following algebraic-differential equations.

The following equations were used.

APC = APCtot − APCP

Cdc55i = Cdc55tot − Cdc55a

Molecular species that could not be identified with algebraic equations were numerically computed using the following differential equations.

dAPCP/dt = (APC × kph × Clb2)/(Japc+APC) − [APCP × (kdeph × Cdc55a)]/(Japci + APCP)

dCdc55a/dt = ka × Cdc55i-Cdc55a × (ki × APCP)

dClb2/dt = ksyn − Clb2 × (kdeg × APCP)

APC/C phosphorylation: kph = 0.6 (checkpoint ON), kph = 1.2 (checkpoint OFF)

APC/C dephosphorylation: kdeph = 2.4

PP2A^Cdc55 activation: ka = 1.2

PP2A^Cdc55 inactivation: ki = 1.2

Cdc20 synthesis: ksyn = 0.02

Cdc20 degradation: kdeg = 0.055

Michaelis constant for APC/C activation: Japc = 0.01

Michaelis constant for APC/C inactivation: Japci = 0.01

Total APC/C: APCtot = 1

Total Cdc55: Cdc55tot = 0.9

Initial conditions were as follows.

APCP = 0.012, Cdc55A = 0.875, Clb2 = 0.965

Fig. S1 shows that Cdc28 inhibition arrests cells in mitosis with an active checkpoint. Fig. S2 shows that partial inhibition of Cdc28 does not destabilize Pds1 and Clb2 in the presence of Cdc20. Fig. S3 shows that 50 nM of 1-NMPP1 only mildly delays cell cycle progression. Fig. S4 shows single cell analysis of Clb2 accumulation during a checkpoint-arrest. Fig. S5 shows that Cdc16 phosphorylation induces a gel shift during a prolonged spindle checkpoint arrest, and that Cdc14 is not required for adaptation. Table S1 is a strain list. Video 1 shows the signal of Clb2-GFP and TUB2-Cherry in one adapting cell and the definition of the nuclear volume after clustering.

We thank Luca Mariani and Emanuele Martini for helping at different stages of this project; Andrew Murray, Jan Michael Peters, Simonetta Piatti, Pamela Silver, Uttam Surana, Tomo Tanaka, Rosella Visintin, and Stefan Westermann for reagents and strains; Rosalind Gunby, Andrea Musacchio, Simonetta Piatti, and Rosella Visintin for discussions and comments on the manuscript; Fred Cross for introducing us to single cell analysis; and the imaging facility of the IFOM-IEO campus.

The group of Andrea Ciliberto is financed by the Associazione Italiana Ricerca sul Cancro (AIRC) and Regione Lombardia (DIVA). Elena Chiroli benefits of a fellowship from Fondazione Veronesi.