Multisite phosphorylation of proteins has been proposed to transform a graded protein kinase signal into an ultrasensitive switch-like response 1–4. Although many multiphosphorylated targets have been identified, the dynamics and sequence of individual phosphorylation events within the multisite phosphorylation process have never been thoroughly studied. In budding yeast, the initiation of S phase is thought to be governed by complexes of Cdk1 and Cln cyclins that phosphorylate six or more sites on the Clb5-Cdk1 inhibitor Sic1, directing it to SCF-mediated destruction 1, 5, 6, 7, 8. The resulting Sic1-free Clb5-Cdk1 complex triggers S phase 9. Here, we demonstrate that Sic1 destruction depends on a more complex process in which both Cln2- and Clb5-Cdk1 act in processive multiphosphorylation cascades leading to the phosphorylation of a small number of specific phosphodegrons. The routes of these phosphorylation cascades are shaped by precisely oriented docking interactions mediated by cyclin-specific docking motifs in Sic1 and by Cks1, the phosphoadaptor subunit of Cdk1. Our results suggest that Clb5-Cdk1-dependent phosphorylation generates positive feedback that is required for switch-like Sic1 destruction. Our evidence for a docking network within clusters of phosphorylation sites uncovers a new level of complexity in Cdk1-dependent regulation of cell cycle transitions, and has general implications for the regulation of cellular processes by multisite phosphorylation.
To study the multiphosphorylation of Sic1, we used a non-inhibitory truncated version of Sic1 (Sic1ΔC) as a substrate for purified Cln2- and Clb5-Cdk1 complexes (Supplementary Fig. 1a). Intriguingly, both Cln2- and Clb5-Cdk1 generated phosphorylation patterns with abruptly accumulating hyperphosphorylated species (Fig. 1a, b, d, e, Supplementary Fig. 1b). This pattern depended on Cks1, the Cdk1 subunit that binds phosphate groups 10. Mutation of the phosphate-binding site of Cks1 reduced the accumulation of multiphosphorylated forms (Fig. 1a, Supplementary Fig. 2a). Similarly, a phosphorylated competitor phosphopeptide reduced phosphorylation (Fig. 1b). Cks1 mutation and the peptide had little effect on the phosphorylation of a Sic1ΔC version containing a single Cdk site (Fig 1c).
Cks1 is essential for Cdk1 function 11, 12, with roles at the G1/S and G2/M transitions 13, 14. We found that the Cks1:Cdk1 stoichiometry in vivo was about 1:1 for Cln2-Cdk1 and at least 0.5:1 for Clb5-Cdk1, confirming that Cks1-dependent multiphosphorylation is the prevalent mode of Cdk1 action in vivo (Supplementary Fig. 1a, b, c). An isothermal calorimetry binding assay of fully phosphorylated Sic1ΔC (pSic1ΔC) and Cks1 revealed a KD of 11 ± 2 μM, while the non-phosphorylated version showed no detectable binding (Supplementary Fig. 1d). Approximately 3–4 molecules of Cks1 bound each molecule of pSic1ΔC, suggesting that several phosphorylated sites can bind Cks1 independently. Finally, we found that the phospho-binding pocket of Cks1 is required for phosphorylation and degradation of Sic1 in vivo (Supplementary Fig. 1e, f).
To understand the Cks1-dependent mechanism, we analyzed Sic1ΔC multiphosphorylation over time (Fig. 1d, e). We did not observe significant accumulation of intermediate phosphorylated forms, suggesting that phosphorylation was processive. When we performed kinase reactions at Sic1ΔC concentrations higher than apparent KM, multiphosphorylation patterns remained constant despite the increase in the inhibition term 1+[S0P]/KM (Supplementary Fig. 2b–e). Thus, the enzyme displays processivity: that is, it is able to transfer two or more phosphates to the substrate during a single association event. This conclusion was additionally confirmed using different enzyme concentrations in the assay (Supplementary Fig. 2f) and in mathematical simulations (Supplementary Fig. 3). This processive pattern argues against the current model of ultrasensitivity in the Sic1 phosphorylation switch, which is based on the assumption of a distributive mechanism with equal specificity of different sites 1, 15.
To dissect the mechanism of the processive multiphosphorylation cascade, we first studied the impact of potential docking interactions between Sic1 and cyclins. In previous studies, we found that rapid Sic1 phosphorylation by Clb5-Cdk1 depends on an interaction between RXL motifs in Sic1 and the hydrophobic patch docking site (hp) in Clb5; a triple mutation in this site (Clb5hpm) decreases the net phosphorylation rate (Supplementary Fig. 4a–h) 16. We further found here that a version of Sic1ΔC with mutations at its four RXL motifs (Sic1ΔC-1234rxl) showed less abrupt production of multiphosphorylated species by Clb5-Cdk1, revealing that processive multiphosphorylation requires both Cks1-dependent and hp-dependent docking (Supplementary Fig. 4i). Cln2-Cdk1 exhibited only a mild RXL effect on the phosphorylation pattern, probably because Cln2 does not contain a conventional hp like that in the B-type cyclins. In recent studies, we also located in Sic1 a ten-amino-acid stretch, 136VLLPPSRPTS145, which confers Cln2 specificity16. Here we found that a five-alanine mutation of the first five hydrophobic residues in this stretch, or a synthetic competitor peptide containing the docking site, reduced the abrupt multiphosphorylation pattern for Cln2 (Supplementary Fig. 4j). In conclusion, both Clb5- and Cln2-Cdk1 use docking mechanisms, in addition to Cks1, to achieve processive multiphosphorylation of Sic1.
Using Sic1ΔC mutants carrying only one Cdk site (Fig. 2a, b), we found that Clb5-Cdk1 rapidly phosphorylated just four sites (T5, T33, S76, and S80), and this specificity depended on the RXL2 and RXL3 docking sites in Sic1 (Fig. 2b; note that in all figure labels, the indicated Cdk sites are those left unmutated, unless otherwise indicated). Cln2-Cdk1, on the other hand, exhibited a preference for the N-terminally located site T5 (Fig. 2b). Thus, docking interactions direct the associated kinase to a small number of primary phosphorylation sites. We speculate that these primary sites interact with Cks1 to drive processive phosphorylation of additional sites.
With these primary specificities in mind, we set out to map the pathways along which Cln2- and Clb5-Cdk1 catalyze the phosphorylation of the critical sites required for Sic1 degradation. The original model of Sic1 regulation proposed that six or more sites must be simultaneously phosphorylated in vivo to facilitate binding of phospho-Sic1 to the SCF subunit Cdc4 1. On the other hand, later binding studies revealed that closely positioned pairs of phosphorylation sites (pT5/pS9, pT45/pT48, or pS76/pS80; see Fig. 2a) each present separate entities with a strong affinity for Cdc4, suggesting that just two phosphorylation sites, in the right positions, might be sufficient for Sic1 degradation 17. Our results provided a way to reconcile these findings: we hypothesized that the requirement for six or more sites in vivo reflects a requirement for priming phosphorylation events that direct processive phosphorylation of critical phosphodegrons. To test this possibility, we first measured phosphorylation of a Sic1ΔC mutant with all Cdk sites changed to alanine except for the triple cluster S69/S76/S80, which contains two potential paired degrons (S69/S76 and S76/S80). There was no processive multiphosphorylation of the cluster S69/S76/S80 (Fig. 2c, lane 2 in each panel), but processivity could be induced by adding back single Cdk1 sites to the N-terminal side of the cluster. The rate of the appearance of multiphosphorylated species correlated with the site specificity data for Cln2 and Clb5 (Fig. 2b). Addition of the most Cln2-specific site, T5, caused a much greater effect in Cln2 reactions than the less Cln2-specific sites, T33 and T45. Addition of the Clb5-specific sites T5 and T33, but not the poor Clb5 site T45, greatly increased processivity in Clb5 reactions. These results suggest that sites T5, T33, and T45 are able to serve as priming sites for Cks1-dependent phosphorylation of the S69/S76/S80 degron cluster. The phosphopeptide-dependence was confirmed for these mutants as described for Sic1ΔC-wt in Figure 1b (data not shown).
Cells overexpressing Sic1 containing only the triple cluster S69/S76/S80 were inviable (Fig. 2d). Addition of T5, T33, or T45 did not prevent this lethal effect, suggesting that phosphorylation of the S69/S76/S80 cluster alone is not sufficient for degradation of Sic1. However, viability improved when both T33 and T45 were added. Notably, the addition of T45, together with a nonconsensus Cdk site T48 (Fig. 2a), creates a double degron, T45/T48, as predicted previously 17. T48 is known to be phosphorylated in vivo 7. We hypothesized that T33 serves as a docking site for both the T45/T48 and S69/S76/S80 clusters, and that T45 serves both as a constituent site of the T45/T48 degron as well as a Cks1-dependent docking site for the degron cluster S69/S76/S80. Indeed, T48 alone (the Sic1-9A mutant with all Cdk sites changed to alanine) was a very poor substrate for Clb5 and Cln2 (data not shown), but the addition of T33 or the T33/T45 pair made it a specific site (Fig. 2e, Supplementary Fig. 5, Supplementary Table 1), implying that T33 phosphorylation allows bypass of the consensus motif requirement of a +1 proline at T48. Strikingly, we found that the alanine mutation in T48 had a strong growth-suppressing effect in the galactose assay within the context of sites T33/T45/S69/S76/S80 and also had a weaker effect in the background containing all the Cdk sites (Fig. 2f, Supplementary Fig. 4k). Our results indicate that the benefit of multisite phosphorylation of Sic1, compared to a system with a single phosphorylated site with high affinity for Cdc4 (e.g. based on an optimal degron site of Cyclin E 1), is likely to be the ability of this mechanism to provide docking-dependent kinase specificity for paired degrons. Thus, phosphorylation sites in Sic1 can be divided into three categories: (i) paired degron sites that are spaced 3–7 amino acids from each another; (ii) N-terminally positioned priming sites for each paired degron; and (iii) sites that serve as both priming and degron sites (e.g. T45).
It is unclear why a single paired degron with a priming site is insufficient for degradation of Sic1, as suggested by our observation that viability in the galactose plate assay requires both T45/T48 and either S69/S76 or S76/S80 (Fig. 2f). To test if the simultaneous presence of both paired degrons is required for degradation of Sic1, we used western blotting to measure phosphorylation and degradation of mutated versions of Sic1ΔC. Remarkably, a Sic1ΔC construct (equivalent to the Sic1-5p mutant of Nash et al. 1) containing only the degron T45/T48, but missing the intact degron around S76, was rapidly degraded after the release of cells from G1 (Fig. 2g). This degradation was abolished by mutation of the single nonconsensus Cdk site at T48. A construct also containing the S69/S76/S80 degron (Sic1-7p of Nash et al. 1) was more rapidly degraded, and mutation of T48 in this background did not influence the degradation rate. We conclude that the T45/T48 degron is sufficient to promote some degradation of Sic1 in vivo, but this rate of degradation is not sufficient to prevent the lethality of overexpressed Sic1.
Our model assumes differential roles of Cln2 and Clb5 in the order of Sic1 phosphorylation events. To explore this possibility, we developed methods for determining the apparent rate constants, which we termed kdock, for individual Cks1-enhanced phosphorylation steps (Fig. 3a, b; Supplementary Table 1). The results revealed considerable differences between Clb5 and Cln2. Clb5-Cdk1 was much more effective than Cln2-Cdk1 in taking shortcuts to the critical degron pair of S76/S80, using T5 and T33 as priming sites for Cks1, and with assistance from RXL-mediated docking (Supplementary Tables 1 and 2; Fig. 3a–c; Supplementary Fig. 6). Notably, in the case of Clb5, different RXL motifs supported different Cks1-dependent docking events (Supplementary Table 2).
We propose that in late G1, Clb5-Cdk1 is inhibited by Sic1, and the cascade of phosphorylation events begins with T5 phosphorylation by Cln2-Cdk1. This priming event is followed by docking-enhanced phosphorylations leading to a phosphorylated chain of sites pT5/pT33/pT45/pS76 but no fully phosphorylated paired degrons, as phosphorylation by Cln2-Cdk1 of suboptimal sites in the degrons (T48 and S69, or S80) is slow (Fig. 3a, b). However, the phosphorylated cluster pT5/pT33/pT45/pS76 serves as a powerful Cln2-Cdk1-dependent docking platform for emerging Clb5-Cdk1. As Cln2 levels rise, such priming forces would create a synergistic effect between Cln2 and Clb5, greatly amplifying the impact of low emerging levels of free Clb5-Cdk1 complexes and defining the point of no return for Clb5-dependent positive feedback. A prediction of this model is that changing the limiting suboptimal degron sites to optimal Cdk sites will rescue the lethality of Sic1-1234rxl (Supplementary Fig. 4f), as the degradation in this case should be driven primarily by Cln2. Indeed, changing T48 to a Cdk1 site by introducing a proline at position +1 partially rescued the lethal phenotype of Sic1-1234rxl (Fig. 3d). A similar rescue was attained by introducing a positive determinant for Cln2-Cdk1, an arginine at position +2 16, to the site S80. Optimization of the S69 site had no effect. Almost complete rescue was gained by a triple mutation with all three limiting degron sites (T48, S69, and S80) changed to optimal Cdk sites. Importantly, these effects are unlikely to be due to improved binding of phosphodegrons to Cdc4, as the basic residues on the C-terminal side of pS/pT are known to disrupt the Cdc4 interaction 1. Finally, in order to confirm that degradation of these Sic1 mutants is driven by Cln-Cdk1, instead of Clb5-Cdk1, we additionally mutated the Cln2-specific docking site VLLPP in the triple mutant background (Fig. 3d, lower panel). The vllpp mutation abolished the rescue effect of the triple mutant. These data indicate that Cln2 alone does have the potential to drive Sic1 degradation, but the Cln2-driven phosphorylation cascade is terminated at the rate-limiting final steps. However, this mechanism allows creation of the Clb5 docking platform containing the chain of optimal sites pT5/pT33/pT45/pS76.
Finally, to compare further the functions of Clb5 and Cln2, we analyzed the degradation of endogeneous Sic1. We found that mutation of either the Cln-specific vllpp docking motif or Clb5-specific RXL docking sites delayed Sic1 degradation (Fig. 3e), confirming that both Cln2 and Clb5 have a role in the timing of Sic1 degradation. However, when all Clb-Cdk1 activity in the cell was specifically inhibited by overexpression of nondegradable Sic1, endogenous Sic1 was completely stabilized (Fig. 3f, Supplementary Fig. 7), arguing that the key trigger for Sic1 degradation and the G1/S transition is the emerging free Clb5-Cdk1, after its levels exceed those of the inhibitory complex.
The inability of Cln alone to cause Sic1 degradation could be attributed to the slow phosphorylation rate of sites in the S76/S80 degron. Indeed, by introducing optimal Cln consensus motifs into the slow degron sites, analogously to the experiment described in Figure 3d, the Sic1 degradation pattern was restored to normal despite the absence of Clb-Cdk1 activity (Fig. 3g). The processive multiphosphorylation cascades, composed of a set of fast and slow steps and different docking specificities, enable this discrimination between the signal outputs of different cyclin-Cdk1 complexes. Furthermore, the Cln output state acts as a primer state for the second complex, creating a potential ‘AND gate’ in which Cln2 is not allowed to trigger the G1/S transition until sufficient levels of Clb5 activity accumulate.
In conclusion, our model provides novel insights into the multisite phosphorylation mechanism of Sic1 and, potentially, of Cdk1 targets in general. The multiple sites create a network of docking connections that exploit Cks1-dependent and cyclin-specific docking interactions to process Cdk1 signals to achieve proper tuning of the timing of the G1/S transition (Supplementary Fig. 8). As most Cdk1 targets in the cell contain clusters of multiple sites 18, the regulation of cell cycle switchpoints by Cdk1-dependent multiphosphorylation might prove to be far more complex than generally anticipated, and it is possible that beneath the seemingly random constellations of phosphorylation sites, an intricate signal processing logic may be hidden.